2 * Copyright (c) 2000-2003,2005 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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_error.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
44 #include "xfs_log_priv.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_log_recover.h"
47 #include "xfs_extfree_item.h"
48 #include "xfs_trans_priv.h"
49 #include "xfs_quota.h"
52 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
53 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
54 STATIC
void xlog_recover_insert_item_backq(xlog_recover_item_t
**q
,
55 xlog_recover_item_t
*item
);
57 STATIC
void xlog_recover_check_summary(xlog_t
*);
58 STATIC
void xlog_recover_check_ail(xfs_mount_t
*, xfs_log_item_t
*, int);
60 #define xlog_recover_check_summary(log)
61 #define xlog_recover_check_ail(mp, lip, gen)
66 * Sector aligned buffer routines for buffer create/read/write/access
69 #define XLOG_SECTOR_ROUNDUP_BBCOUNT(log, bbs) \
70 ( ((log)->l_sectbb_mask && (bbs & (log)->l_sectbb_mask)) ? \
71 ((bbs + (log)->l_sectbb_mask + 1) & ~(log)->l_sectbb_mask) : (bbs) )
72 #define XLOG_SECTOR_ROUNDDOWN_BLKNO(log, bno) ((bno) & ~(log)->l_sectbb_mask)
79 ASSERT(num_bblks
> 0);
81 if (log
->l_sectbb_log
) {
83 num_bblks
+= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
84 num_bblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, num_bblks
);
86 return xfs_buf_get_noaddr(BBTOB(num_bblks
), log
->l_mp
->m_logdev_targp
);
98 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
109 if (log
->l_sectbb_log
) {
110 blk_no
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, blk_no
);
111 nbblks
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, nbblks
);
115 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
118 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
121 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
122 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
124 xfsbdstrat(log
->l_mp
, bp
);
125 if ((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 printk("%s: SB : uuid = ", __FUNCTION__
);
197 for (b
= 0; b
< 16; b
++)
198 printk("%02x",((unsigned char *)&mp
->m_sb
.sb_uuid
)[b
]);
199 printk(", fmt = %d\n", XLOG_FMT
);
200 printk(" log : uuid = ");
201 for (b
= 0; b
< 16; b
++)
202 printk("%02x",((unsigned char *)&head
->h_fs_uuid
)[b
]);
203 printk(", fmt = %d\n", INT_GET(head
->h_fmt
, ARCH_CONVERT
));
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(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == 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(INT_GET(head
->h_fmt
, ARCH_CONVERT
) != 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(INT_GET(head
->h_magicno
, ARCH_CONVERT
) == 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
, XFS_METADATA_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
= GET_CYCLE(offset
, ARCH_CONVERT
);
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
= GET_CYCLE(buf
, ARCH_CONVERT
);
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
==
453 INT_GET(head
->h_magicno
, ARCH_CONVERT
))
461 * We hit the beginning of the physical log & still no header. Return
462 * to caller. If caller can handle a return of -1, then this routine
463 * will be called again for the end of the physical log.
471 * We have the final block of the good log (the first block
472 * of the log record _before_ the head. So we check the uuid.
474 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
478 * We may have found a log record header before we expected one.
479 * last_blk will be the 1st block # with a given cycle #. We may end
480 * up reading an entire log record. In this case, we don't want to
481 * reset last_blk. Only when last_blk points in the middle of a log
482 * record do we update last_blk.
484 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
485 uint h_size
= INT_GET(head
->h_size
, ARCH_CONVERT
);
487 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
488 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
494 if (*last_blk
- i
+ extra_bblks
495 != BTOBB(INT_GET(head
->h_len
, ARCH_CONVERT
)) + xhdrs
)
504 * Head is defined to be the point of the log where the next log write
505 * write could go. This means that incomplete LR writes at the end are
506 * eliminated when calculating the head. We aren't guaranteed that previous
507 * LR have complete transactions. We only know that a cycle number of
508 * current cycle number -1 won't be present in the log if we start writing
509 * from our current block number.
511 * last_blk contains the block number of the first block with a given
514 * Return: zero if normal, non-zero if error.
519 xfs_daddr_t
*return_head_blk
)
523 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
525 uint first_half_cycle
, last_half_cycle
;
527 int error
, log_bbnum
= log
->l_logBBsize
;
529 /* Is the end of the log device zeroed? */
530 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
531 *return_head_blk
= first_blk
;
533 /* Is the whole lot zeroed? */
535 /* Linux XFS shouldn't generate totally zeroed logs -
536 * mkfs etc write a dummy unmount record to a fresh
537 * log so we can store the uuid in there
539 xlog_warn("XFS: totally zeroed log");
544 xlog_warn("XFS: empty log check failed");
548 first_blk
= 0; /* get cycle # of 1st block */
549 bp
= xlog_get_bp(log
, 1);
552 if ((error
= xlog_bread(log
, 0, 1, bp
)))
554 offset
= xlog_align(log
, 0, 1, bp
);
555 first_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
557 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
558 if ((error
= xlog_bread(log
, last_blk
, 1, bp
)))
560 offset
= xlog_align(log
, last_blk
, 1, bp
);
561 last_half_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
562 ASSERT(last_half_cycle
!= 0);
565 * If the 1st half cycle number is equal to the last half cycle number,
566 * then the entire log is stamped with the same cycle number. In this
567 * case, head_blk can't be set to zero (which makes sense). The below
568 * math doesn't work out properly with head_blk equal to zero. Instead,
569 * we set it to log_bbnum which is an invalid block number, but this
570 * value makes the math correct. If head_blk doesn't changed through
571 * all the tests below, *head_blk is set to zero at the very end rather
572 * than log_bbnum. In a sense, log_bbnum and zero are the same block
573 * in a circular file.
575 if (first_half_cycle
== last_half_cycle
) {
577 * In this case we believe that the entire log should have
578 * cycle number last_half_cycle. We need to scan backwards
579 * from the end verifying that there are no holes still
580 * containing last_half_cycle - 1. If we find such a hole,
581 * then the start of that hole will be the new head. The
582 * simple case looks like
583 * x | x ... | x - 1 | x
584 * Another case that fits this picture would be
585 * x | x + 1 | x ... | x
586 * In this case the head really is somwhere at the end of the
587 * log, as one of the latest writes at the beginning was
590 * x | x + 1 | x ... | x - 1 | x
591 * This is really the combination of the above two cases, and
592 * the head has to end up at the start of the x-1 hole at the
595 * In the 256k log case, we will read from the beginning to the
596 * end of the log and search for cycle numbers equal to x-1.
597 * We don't worry about the x+1 blocks that we encounter,
598 * because we know that they cannot be the head since the log
601 head_blk
= log_bbnum
;
602 stop_on_cycle
= last_half_cycle
- 1;
605 * In this case we want to find the first block with cycle
606 * number matching last_half_cycle. We expect the log to be
609 * The first block with cycle number x (last_half_cycle) will
610 * be where the new head belongs. First we do a binary search
611 * for the first occurrence of last_half_cycle. The binary
612 * search may not be totally accurate, so then we scan back
613 * from there looking for occurrences of last_half_cycle before
614 * us. If that backwards scan wraps around the beginning of
615 * the log, then we look for occurrences of last_half_cycle - 1
616 * at the end of the log. The cases we're looking for look
618 * x + 1 ... | x | x + 1 | x ...
619 * ^ binary search stopped here
621 * x + 1 ... | x ... | x - 1 | x
622 * <---------> less than scan distance
624 stop_on_cycle
= last_half_cycle
;
625 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
626 &head_blk
, last_half_cycle
)))
631 * Now validate the answer. Scan back some number of maximum possible
632 * blocks and make sure each one has the expected cycle number. The
633 * maximum is determined by the total possible amount of buffering
634 * in the in-core log. The following number can be made tighter if
635 * we actually look at the block size of the filesystem.
637 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
638 if (head_blk
>= num_scan_bblks
) {
640 * We are guaranteed that the entire check can be performed
643 start_blk
= head_blk
- num_scan_bblks
;
644 if ((error
= xlog_find_verify_cycle(log
,
645 start_blk
, num_scan_bblks
,
646 stop_on_cycle
, &new_blk
)))
650 } else { /* need to read 2 parts of log */
652 * We are going to scan backwards in the log in two parts.
653 * First we scan the physical end of the log. In this part
654 * of the log, we are looking for blocks with cycle number
655 * last_half_cycle - 1.
656 * If we find one, then we know that the log starts there, as
657 * we've found a hole that didn't get written in going around
658 * the end of the physical log. The simple case for this is
659 * x + 1 ... | x ... | x - 1 | x
660 * <---------> less than scan distance
661 * If all of the blocks at the end of the log have cycle number
662 * last_half_cycle, then we check the blocks at the start of
663 * the log looking for occurrences of last_half_cycle. If we
664 * find one, then our current estimate for the location of the
665 * first occurrence of last_half_cycle is wrong and we move
666 * back to the hole we've found. This case looks like
667 * x + 1 ... | x | x + 1 | x ...
668 * ^ binary search stopped here
669 * Another case we need to handle that only occurs in 256k
671 * x + 1 ... | x ... | x+1 | x ...
672 * ^ binary search stops here
673 * In a 256k log, the scan at the end of the log will see the
674 * x + 1 blocks. We need to skip past those since that is
675 * certainly not the head of the log. By searching for
676 * last_half_cycle-1 we accomplish that.
678 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
679 ASSERT(head_blk
<= INT_MAX
&&
680 (xfs_daddr_t
) num_scan_bblks
- head_blk
>= 0);
681 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
682 num_scan_bblks
- (int)head_blk
,
683 (stop_on_cycle
- 1), &new_blk
)))
691 * Scan beginning of log now. The last part of the physical
692 * log is good. This scan needs to verify that it doesn't find
693 * the last_half_cycle.
696 ASSERT(head_blk
<= INT_MAX
);
697 if ((error
= xlog_find_verify_cycle(log
,
698 start_blk
, (int)head_blk
,
699 stop_on_cycle
, &new_blk
)))
707 * Now we need to make sure head_blk is not pointing to a block in
708 * the middle of a log record.
710 num_scan_bblks
= XLOG_REC_SHIFT(log
);
711 if (head_blk
>= num_scan_bblks
) {
712 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
714 /* start ptr at last block ptr before head_blk */
715 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
716 &head_blk
, 0)) == -1) {
717 error
= XFS_ERROR(EIO
);
723 ASSERT(head_blk
<= INT_MAX
);
724 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
725 &head_blk
, 0)) == -1) {
726 /* We hit the beginning of the log during our search */
727 start_blk
= log_bbnum
- num_scan_bblks
+ head_blk
;
729 ASSERT(start_blk
<= INT_MAX
&&
730 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
731 ASSERT(head_blk
<= INT_MAX
);
732 if ((error
= xlog_find_verify_log_record(log
,
734 (int)head_blk
)) == -1) {
735 error
= XFS_ERROR(EIO
);
739 if (new_blk
!= log_bbnum
)
746 if (head_blk
== log_bbnum
)
747 *return_head_blk
= 0;
749 *return_head_blk
= head_blk
;
751 * When returning here, we have a good block number. Bad block
752 * means that during a previous crash, we didn't have a clean break
753 * from cycle number N to cycle number N-1. In this case, we need
754 * to find the first block with cycle number N-1.
762 xlog_warn("XFS: failed to find log head");
767 * Find the sync block number or the tail of the log.
769 * This will be the block number of the last record to have its
770 * associated buffers synced to disk. Every log record header has
771 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
772 * to get a sync block number. The only concern is to figure out which
773 * log record header to believe.
775 * The following algorithm uses the log record header with the largest
776 * lsn. The entire log record does not need to be valid. We only care
777 * that the header is valid.
779 * We could speed up search by using current head_blk buffer, but it is not
785 xfs_daddr_t
*head_blk
,
786 xfs_daddr_t
*tail_blk
)
788 xlog_rec_header_t
*rhead
;
789 xlog_op_header_t
*op_head
;
790 xfs_caddr_t offset
= NULL
;
793 xfs_daddr_t umount_data_blk
;
794 xfs_daddr_t after_umount_blk
;
801 * Find previous log record
803 if ((error
= xlog_find_head(log
, head_blk
)))
806 bp
= xlog_get_bp(log
, 1);
809 if (*head_blk
== 0) { /* special case */
810 if ((error
= xlog_bread(log
, 0, 1, bp
)))
812 offset
= xlog_align(log
, 0, 1, bp
);
813 if (GET_CYCLE(offset
, ARCH_CONVERT
) == 0) {
815 /* leave all other log inited values alone */
821 * Search backwards looking for log record header block
823 ASSERT(*head_blk
< INT_MAX
);
824 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
825 if ((error
= xlog_bread(log
, i
, 1, bp
)))
827 offset
= xlog_align(log
, i
, 1, bp
);
828 if (XLOG_HEADER_MAGIC_NUM
==
829 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
835 * If we haven't found the log record header block, start looking
836 * again from the end of the physical log. XXXmiken: There should be
837 * a check here to make sure we didn't search more than N blocks in
841 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
842 if ((error
= xlog_bread(log
, i
, 1, bp
)))
844 offset
= xlog_align(log
, i
, 1, bp
);
845 if (XLOG_HEADER_MAGIC_NUM
==
846 INT_GET(*(uint
*)offset
, ARCH_CONVERT
)) {
853 xlog_warn("XFS: xlog_find_tail: couldn't find sync record");
855 return XFS_ERROR(EIO
);
858 /* find blk_no of tail of log */
859 rhead
= (xlog_rec_header_t
*)offset
;
860 *tail_blk
= BLOCK_LSN(INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
));
863 * Reset log values according to the state of the log when we
864 * crashed. In the case where head_blk == 0, we bump curr_cycle
865 * one because the next write starts a new cycle rather than
866 * continuing the cycle of the last good log record. At this
867 * point we have guaranteed that all partial log records have been
868 * accounted for. Therefore, we know that the last good log record
869 * written was complete and ended exactly on the end boundary
870 * of the physical log.
872 log
->l_prev_block
= i
;
873 log
->l_curr_block
= (int)*head_blk
;
874 log
->l_curr_cycle
= INT_GET(rhead
->h_cycle
, ARCH_CONVERT
);
877 log
->l_tail_lsn
= INT_GET(rhead
->h_tail_lsn
, ARCH_CONVERT
);
878 log
->l_last_sync_lsn
= INT_GET(rhead
->h_lsn
, ARCH_CONVERT
);
879 log
->l_grant_reserve_cycle
= log
->l_curr_cycle
;
880 log
->l_grant_reserve_bytes
= BBTOB(log
->l_curr_block
);
881 log
->l_grant_write_cycle
= log
->l_curr_cycle
;
882 log
->l_grant_write_bytes
= BBTOB(log
->l_curr_block
);
885 * Look for unmount record. If we find it, then we know there
886 * was a clean unmount. Since 'i' could be the last block in
887 * the physical log, we convert to a log block before comparing
890 * Save the current tail lsn to use to pass to
891 * xlog_clear_stale_blocks() below. We won't want to clear the
892 * unmount record if there is one, so we pass the lsn of the
893 * unmount record rather than the block after it.
895 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
896 int h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
897 int h_version
= INT_GET(rhead
->h_version
, ARCH_CONVERT
);
899 if ((h_version
& XLOG_VERSION_2
) &&
900 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
901 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
902 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
910 after_umount_blk
= (i
+ hblks
+ (int)
911 BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
))) % log
->l_logBBsize
;
912 tail_lsn
= log
->l_tail_lsn
;
913 if (*head_blk
== after_umount_blk
&&
914 INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
) == 1) {
915 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
916 if ((error
= xlog_bread(log
, umount_data_blk
, 1, bp
))) {
919 offset
= xlog_align(log
, umount_data_blk
, 1, bp
);
920 op_head
= (xlog_op_header_t
*)offset
;
921 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
923 * Set tail and last sync so that newly written
924 * log records will point recovery to after the
925 * current unmount record.
927 ASSIGN_ANY_LSN_HOST(log
->l_tail_lsn
, log
->l_curr_cycle
,
929 ASSIGN_ANY_LSN_HOST(log
->l_last_sync_lsn
, log
->l_curr_cycle
,
931 *tail_blk
= after_umount_blk
;
936 * Make sure that there are no blocks in front of the head
937 * with the same cycle number as the head. This can happen
938 * because we allow multiple outstanding log writes concurrently,
939 * and the later writes might make it out before earlier ones.
941 * We use the lsn from before modifying it so that we'll never
942 * overwrite the unmount record after a clean unmount.
944 * Do this only if we are going to recover the filesystem
946 * NOTE: This used to say "if (!readonly)"
947 * However on Linux, we can & do recover a read-only filesystem.
948 * We only skip recovery if NORECOVERY is specified on mount,
949 * in which case we would not be here.
951 * But... if the -device- itself is readonly, just skip this.
952 * We can't recover this device anyway, so it won't matter.
954 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
)) {
955 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
963 xlog_warn("XFS: failed to locate log tail");
968 * Is the log zeroed at all?
970 * The last binary search should be changed to perform an X block read
971 * once X becomes small enough. You can then search linearly through
972 * the X blocks. This will cut down on the number of reads we need to do.
974 * If the log is partially zeroed, this routine will pass back the blkno
975 * of the first block with cycle number 0. It won't have a complete LR
979 * 0 => the log is completely written to
980 * -1 => use *blk_no as the first block of the log
981 * >0 => error has occurred
990 uint first_cycle
, last_cycle
;
991 xfs_daddr_t new_blk
, last_blk
, start_blk
;
992 xfs_daddr_t num_scan_bblks
;
993 int error
, log_bbnum
= log
->l_logBBsize
;
995 /* check totally zeroed log */
996 bp
= xlog_get_bp(log
, 1);
999 if ((error
= xlog_bread(log
, 0, 1, bp
)))
1001 offset
= xlog_align(log
, 0, 1, bp
);
1002 first_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1003 if (first_cycle
== 0) { /* completely zeroed log */
1009 /* check partially zeroed log */
1010 if ((error
= xlog_bread(log
, log_bbnum
-1, 1, bp
)))
1012 offset
= xlog_align(log
, log_bbnum
-1, 1, bp
);
1013 last_cycle
= GET_CYCLE(offset
, ARCH_CONVERT
);
1014 if (last_cycle
!= 0) { /* log completely written to */
1017 } else if (first_cycle
!= 1) {
1019 * If the cycle of the last block is zero, the cycle of
1020 * the first block must be 1. If it's not, maybe we're
1021 * not looking at a log... Bail out.
1023 xlog_warn("XFS: Log inconsistent or not a log (last==0, first!=1)");
1024 return XFS_ERROR(EINVAL
);
1027 /* we have a partially zeroed log */
1028 last_blk
= log_bbnum
-1;
1029 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1033 * Validate the answer. Because there is no way to guarantee that
1034 * the entire log is made up of log records which are the same size,
1035 * we scan over the defined maximum blocks. At this point, the maximum
1036 * is not chosen to mean anything special. XXXmiken
1038 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1039 ASSERT(num_scan_bblks
<= INT_MAX
);
1041 if (last_blk
< num_scan_bblks
)
1042 num_scan_bblks
= last_blk
;
1043 start_blk
= last_blk
- num_scan_bblks
;
1046 * We search for any instances of cycle number 0 that occur before
1047 * our current estimate of the head. What we're trying to detect is
1048 * 1 ... | 0 | 1 | 0...
1049 * ^ binary search ends here
1051 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1052 (int)num_scan_bblks
, 0, &new_blk
)))
1058 * Potentially backup over partial log record write. We don't need
1059 * to search the end of the log because we know it is zero.
1061 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1062 &last_blk
, 0)) == -1) {
1063 error
= XFS_ERROR(EIO
);
1077 * These are simple subroutines used by xlog_clear_stale_blocks() below
1078 * to initialize a buffer full of empty log record headers and write
1079 * them into the log.
1090 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1092 memset(buf
, 0, BBSIZE
);
1093 INT_SET(recp
->h_magicno
, ARCH_CONVERT
, XLOG_HEADER_MAGIC_NUM
);
1094 INT_SET(recp
->h_cycle
, ARCH_CONVERT
, cycle
);
1095 INT_SET(recp
->h_version
, ARCH_CONVERT
,
1096 XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) ? 2 : 1);
1097 ASSIGN_ANY_LSN_DISK(recp
->h_lsn
, cycle
, block
);
1098 ASSIGN_ANY_LSN_DISK(recp
->h_tail_lsn
, tail_cycle
, tail_block
);
1099 INT_SET(recp
->h_fmt
, ARCH_CONVERT
, XLOG_FMT
);
1100 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1104 xlog_write_log_records(
1115 int sectbb
= XLOG_SECTOR_ROUNDUP_BBCOUNT(log
, 1);
1116 int end_block
= start_block
+ blocks
;
1121 bufblks
= 1 << ffs(blocks
);
1122 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1124 if (bufblks
<= log
->l_sectbb_log
)
1128 /* We may need to do a read at the start to fill in part of
1129 * the buffer in the starting sector not covered by the first
1132 balign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, start_block
);
1133 if (balign
!= start_block
) {
1134 if ((error
= xlog_bread(log
, start_block
, 1, bp
))) {
1138 j
= start_block
- balign
;
1141 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1142 int bcount
, endcount
;
1144 bcount
= min(bufblks
, end_block
- start_block
);
1145 endcount
= bcount
- j
;
1147 /* We may need to do a read at the end to fill in part of
1148 * the buffer in the final sector not covered by the write.
1149 * If this is the same sector as the above read, skip it.
1151 ealign
= XLOG_SECTOR_ROUNDDOWN_BLKNO(log
, end_block
);
1152 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1153 offset
= XFS_BUF_PTR(bp
);
1154 balign
= BBTOB(ealign
- start_block
);
1155 XFS_BUF_SET_PTR(bp
, offset
+ balign
, BBTOB(sectbb
));
1156 if ((error
= xlog_bread(log
, ealign
, sectbb
, bp
)))
1158 XFS_BUF_SET_PTR(bp
, offset
, bufblks
);
1161 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1162 for (; j
< endcount
; j
++) {
1163 xlog_add_record(log
, offset
, cycle
, i
+j
,
1164 tail_cycle
, tail_block
);
1167 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1170 start_block
+= endcount
;
1178 * This routine is called to blow away any incomplete log writes out
1179 * in front of the log head. We do this so that we won't become confused
1180 * if we come up, write only a little bit more, and then crash again.
1181 * If we leave the partial log records out there, this situation could
1182 * cause us to think those partial writes are valid blocks since they
1183 * have the current cycle number. We get rid of them by overwriting them
1184 * with empty log records with the old cycle number rather than the
1187 * The tail lsn is passed in rather than taken from
1188 * the log so that we will not write over the unmount record after a
1189 * clean unmount in a 512 block log. Doing so would leave the log without
1190 * any valid log records in it until a new one was written. If we crashed
1191 * during that time we would not be able to recover.
1194 xlog_clear_stale_blocks(
1198 int tail_cycle
, head_cycle
;
1199 int tail_block
, head_block
;
1200 int tail_distance
, max_distance
;
1204 tail_cycle
= CYCLE_LSN(tail_lsn
);
1205 tail_block
= BLOCK_LSN(tail_lsn
);
1206 head_cycle
= log
->l_curr_cycle
;
1207 head_block
= log
->l_curr_block
;
1210 * Figure out the distance between the new head of the log
1211 * and the tail. We want to write over any blocks beyond the
1212 * head that we may have written just before the crash, but
1213 * we don't want to overwrite the tail of the log.
1215 if (head_cycle
== tail_cycle
) {
1217 * The tail is behind the head in the physical log,
1218 * so the distance from the head to the tail is the
1219 * distance from the head to the end of the log plus
1220 * the distance from the beginning of the log to the
1223 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1224 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1225 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1226 return XFS_ERROR(EFSCORRUPTED
);
1228 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1231 * The head is behind the tail in the physical log,
1232 * so the distance from the head to the tail is just
1233 * the tail block minus the head block.
1235 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1236 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1237 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1238 return XFS_ERROR(EFSCORRUPTED
);
1240 tail_distance
= tail_block
- head_block
;
1244 * If the head is right up against the tail, we can't clear
1247 if (tail_distance
<= 0) {
1248 ASSERT(tail_distance
== 0);
1252 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1254 * Take the smaller of the maximum amount of outstanding I/O
1255 * we could have and the distance to the tail to clear out.
1256 * We take the smaller so that we don't overwrite the tail and
1257 * we don't waste all day writing from the head to the tail
1260 max_distance
= MIN(max_distance
, tail_distance
);
1262 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1264 * We can stomp all the blocks we need to without
1265 * wrapping around the end of the log. Just do it
1266 * in a single write. Use the cycle number of the
1267 * current cycle minus one so that the log will look like:
1270 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1271 head_block
, max_distance
, tail_cycle
,
1277 * We need to wrap around the end of the physical log in
1278 * order to clear all the blocks. Do it in two separate
1279 * I/Os. The first write should be from the head to the
1280 * end of the physical log, and it should use the current
1281 * cycle number minus one just like above.
1283 distance
= log
->l_logBBsize
- head_block
;
1284 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1285 head_block
, distance
, tail_cycle
,
1292 * Now write the blocks at the start of the physical log.
1293 * This writes the remainder of the blocks we want to clear.
1294 * It uses the current cycle number since we're now on the
1295 * same cycle as the head so that we get:
1296 * n ... n ... | n - 1 ...
1297 * ^^^^^ blocks we're writing
1299 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1300 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1301 tail_cycle
, tail_block
);
1309 /******************************************************************************
1311 * Log recover routines
1313 ******************************************************************************
1316 STATIC xlog_recover_t
*
1317 xlog_recover_find_tid(
1321 xlog_recover_t
*p
= q
;
1324 if (p
->r_log_tid
== tid
)
1332 xlog_recover_put_hashq(
1334 xlog_recover_t
*trans
)
1341 xlog_recover_add_item(
1342 xlog_recover_item_t
**itemq
)
1344 xlog_recover_item_t
*item
;
1346 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1347 xlog_recover_insert_item_backq(itemq
, item
);
1351 xlog_recover_add_to_cont_trans(
1352 xlog_recover_t
*trans
,
1356 xlog_recover_item_t
*item
;
1357 xfs_caddr_t ptr
, old_ptr
;
1360 item
= trans
->r_itemq
;
1362 /* finish copying rest of trans header */
1363 xlog_recover_add_item(&trans
->r_itemq
);
1364 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1365 sizeof(xfs_trans_header_t
) - len
;
1366 memcpy(ptr
, dp
, len
); /* d, s, l */
1369 item
= item
->ri_prev
;
1371 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1372 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1374 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1375 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1376 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1377 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1382 * The next region to add is the start of a new region. It could be
1383 * a whole region or it could be the first part of a new region. Because
1384 * of this, the assumption here is that the type and size fields of all
1385 * format structures fit into the first 32 bits of the structure.
1387 * This works because all regions must be 32 bit aligned. Therefore, we
1388 * either have both fields or we have neither field. In the case we have
1389 * neither field, the data part of the region is zero length. We only have
1390 * a log_op_header and can throw away the header since a new one will appear
1391 * later. If we have at least 4 bytes, then we can determine how many regions
1392 * will appear in the current log item.
1395 xlog_recover_add_to_trans(
1396 xlog_recover_t
*trans
,
1400 xfs_inode_log_format_t
*in_f
; /* any will do */
1401 xlog_recover_item_t
*item
;
1406 item
= trans
->r_itemq
;
1408 ASSERT(*(uint
*)dp
== XFS_TRANS_HEADER_MAGIC
);
1409 if (len
== sizeof(xfs_trans_header_t
))
1410 xlog_recover_add_item(&trans
->r_itemq
);
1411 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1415 ptr
= kmem_alloc(len
, KM_SLEEP
);
1416 memcpy(ptr
, dp
, len
);
1417 in_f
= (xfs_inode_log_format_t
*)ptr
;
1419 if (item
->ri_prev
->ri_total
!= 0 &&
1420 item
->ri_prev
->ri_total
== item
->ri_prev
->ri_cnt
) {
1421 xlog_recover_add_item(&trans
->r_itemq
);
1423 item
= trans
->r_itemq
;
1424 item
= item
->ri_prev
;
1426 if (item
->ri_total
== 0) { /* first region to be added */
1427 item
->ri_total
= in_f
->ilf_size
;
1428 ASSERT(item
->ri_total
<= XLOG_MAX_REGIONS_IN_ITEM
);
1429 item
->ri_buf
= kmem_zalloc((item
->ri_total
*
1430 sizeof(xfs_log_iovec_t
)), KM_SLEEP
);
1432 ASSERT(item
->ri_total
> item
->ri_cnt
);
1433 /* Description region is ri_buf[0] */
1434 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1435 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1441 xlog_recover_new_tid(
1446 xlog_recover_t
*trans
;
1448 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1449 trans
->r_log_tid
= tid
;
1451 xlog_recover_put_hashq(q
, trans
);
1455 xlog_recover_unlink_tid(
1457 xlog_recover_t
*trans
)
1468 if (tp
->r_next
== trans
) {
1476 "XFS: xlog_recover_unlink_tid: trans not found");
1478 return XFS_ERROR(EIO
);
1480 tp
->r_next
= tp
->r_next
->r_next
;
1486 xlog_recover_insert_item_backq(
1487 xlog_recover_item_t
**q
,
1488 xlog_recover_item_t
*item
)
1491 item
->ri_prev
= item
->ri_next
= item
;
1495 item
->ri_prev
= (*q
)->ri_prev
;
1496 (*q
)->ri_prev
= item
;
1497 item
->ri_prev
->ri_next
= item
;
1502 xlog_recover_insert_item_frontq(
1503 xlog_recover_item_t
**q
,
1504 xlog_recover_item_t
*item
)
1506 xlog_recover_insert_item_backq(q
, item
);
1511 xlog_recover_reorder_trans(
1513 xlog_recover_t
*trans
)
1515 xlog_recover_item_t
*first_item
, *itemq
, *itemq_next
;
1516 xfs_buf_log_format_t
*buf_f
;
1517 xfs_buf_log_format_v1_t
*obuf_f
;
1520 first_item
= itemq
= trans
->r_itemq
;
1521 trans
->r_itemq
= NULL
;
1523 itemq_next
= itemq
->ri_next
;
1524 buf_f
= (xfs_buf_log_format_t
*)itemq
->ri_buf
[0].i_addr
;
1525 switch (ITEM_TYPE(itemq
)) {
1527 flags
= buf_f
->blf_flags
;
1529 case XFS_LI_6_1_BUF
:
1530 case XFS_LI_5_3_BUF
:
1531 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1532 flags
= obuf_f
->blf_flags
;
1536 switch (ITEM_TYPE(itemq
)) {
1538 case XFS_LI_6_1_BUF
:
1539 case XFS_LI_5_3_BUF
:
1540 if (!(flags
& XFS_BLI_CANCEL
)) {
1541 xlog_recover_insert_item_frontq(&trans
->r_itemq
,
1546 case XFS_LI_6_1_INODE
:
1547 case XFS_LI_5_3_INODE
:
1549 case XFS_LI_QUOTAOFF
:
1552 xlog_recover_insert_item_backq(&trans
->r_itemq
, itemq
);
1556 "XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
1558 return XFS_ERROR(EIO
);
1561 } while (first_item
!= itemq
);
1566 * Build up the table of buf cancel records so that we don't replay
1567 * cancelled data in the second pass. For buffer records that are
1568 * not cancel records, there is nothing to do here so we just return.
1570 * If we get a cancel record which is already in the table, this indicates
1571 * that the buffer was cancelled multiple times. In order to ensure
1572 * that during pass 2 we keep the record in the table until we reach its
1573 * last occurrence in the log, we keep a reference count in the cancel
1574 * record in the table to tell us how many times we expect to see this
1575 * record during the second pass.
1578 xlog_recover_do_buffer_pass1(
1580 xfs_buf_log_format_t
*buf_f
)
1582 xfs_buf_cancel_t
*bcp
;
1583 xfs_buf_cancel_t
*nextp
;
1584 xfs_buf_cancel_t
*prevp
;
1585 xfs_buf_cancel_t
**bucket
;
1586 xfs_buf_log_format_v1_t
*obuf_f
;
1587 xfs_daddr_t blkno
= 0;
1591 switch (buf_f
->blf_type
) {
1593 blkno
= buf_f
->blf_blkno
;
1594 len
= buf_f
->blf_len
;
1595 flags
= buf_f
->blf_flags
;
1597 case XFS_LI_6_1_BUF
:
1598 case XFS_LI_5_3_BUF
:
1599 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1600 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1601 len
= obuf_f
->blf_len
;
1602 flags
= obuf_f
->blf_flags
;
1607 * If this isn't a cancel buffer item, then just return.
1609 if (!(flags
& XFS_BLI_CANCEL
))
1613 * Insert an xfs_buf_cancel record into the hash table of
1614 * them. If there is already an identical record, bump
1615 * its reference count.
1617 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1618 XLOG_BC_TABLE_SIZE
];
1620 * If the hash bucket is empty then just insert a new record into
1623 if (*bucket
== NULL
) {
1624 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1626 bcp
->bc_blkno
= blkno
;
1628 bcp
->bc_refcount
= 1;
1629 bcp
->bc_next
= NULL
;
1635 * The hash bucket is not empty, so search for duplicates of our
1636 * record. If we find one them just bump its refcount. If not
1637 * then add us at the end of the list.
1641 while (nextp
!= NULL
) {
1642 if (nextp
->bc_blkno
== blkno
&& nextp
->bc_len
== len
) {
1643 nextp
->bc_refcount
++;
1647 nextp
= nextp
->bc_next
;
1649 ASSERT(prevp
!= NULL
);
1650 bcp
= (xfs_buf_cancel_t
*)kmem_alloc(sizeof(xfs_buf_cancel_t
),
1652 bcp
->bc_blkno
= blkno
;
1654 bcp
->bc_refcount
= 1;
1655 bcp
->bc_next
= NULL
;
1656 prevp
->bc_next
= bcp
;
1660 * Check to see whether the buffer being recovered has a corresponding
1661 * entry in the buffer cancel record table. If it does then return 1
1662 * so that it will be cancelled, otherwise return 0. If the buffer is
1663 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
1664 * the refcount on the entry in the table and remove it from the table
1665 * if this is the last reference.
1667 * We remove the cancel record from the table when we encounter its
1668 * last occurrence in the log so that if the same buffer is re-used
1669 * again after its last cancellation we actually replay the changes
1670 * made at that point.
1673 xlog_check_buffer_cancelled(
1679 xfs_buf_cancel_t
*bcp
;
1680 xfs_buf_cancel_t
*prevp
;
1681 xfs_buf_cancel_t
**bucket
;
1683 if (log
->l_buf_cancel_table
== NULL
) {
1685 * There is nothing in the table built in pass one,
1686 * so this buffer must not be cancelled.
1688 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1692 bucket
= &log
->l_buf_cancel_table
[(__uint64_t
)blkno
%
1693 XLOG_BC_TABLE_SIZE
];
1697 * There is no corresponding entry in the table built
1698 * in pass one, so this buffer has not been cancelled.
1700 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1705 * Search for an entry in the buffer cancel table that
1706 * matches our buffer.
1709 while (bcp
!= NULL
) {
1710 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
) {
1712 * We've go a match, so return 1 so that the
1713 * recovery of this buffer is cancelled.
1714 * If this buffer is actually a buffer cancel
1715 * log item, then decrement the refcount on the
1716 * one in the table and remove it if this is the
1719 if (flags
& XFS_BLI_CANCEL
) {
1721 if (bcp
->bc_refcount
== 0) {
1722 if (prevp
== NULL
) {
1723 *bucket
= bcp
->bc_next
;
1725 prevp
->bc_next
= bcp
->bc_next
;
1728 sizeof(xfs_buf_cancel_t
));
1737 * We didn't find a corresponding entry in the table, so
1738 * return 0 so that the buffer is NOT cancelled.
1740 ASSERT(!(flags
& XFS_BLI_CANCEL
));
1745 xlog_recover_do_buffer_pass2(
1747 xfs_buf_log_format_t
*buf_f
)
1749 xfs_buf_log_format_v1_t
*obuf_f
;
1750 xfs_daddr_t blkno
= 0;
1754 switch (buf_f
->blf_type
) {
1756 blkno
= buf_f
->blf_blkno
;
1757 flags
= buf_f
->blf_flags
;
1758 len
= buf_f
->blf_len
;
1760 case XFS_LI_6_1_BUF
:
1761 case XFS_LI_5_3_BUF
:
1762 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1763 blkno
= (xfs_daddr_t
) obuf_f
->blf_blkno
;
1764 flags
= obuf_f
->blf_flags
;
1765 len
= (xfs_daddr_t
) obuf_f
->blf_len
;
1769 return xlog_check_buffer_cancelled(log
, blkno
, len
, flags
);
1773 * Perform recovery for a buffer full of inodes. In these buffers,
1774 * the only data which should be recovered is that which corresponds
1775 * to the di_next_unlinked pointers in the on disk inode structures.
1776 * The rest of the data for the inodes is always logged through the
1777 * inodes themselves rather than the inode buffer and is recovered
1778 * in xlog_recover_do_inode_trans().
1780 * The only time when buffers full of inodes are fully recovered is
1781 * when the buffer is full of newly allocated inodes. In this case
1782 * the buffer will not be marked as an inode buffer and so will be
1783 * sent to xlog_recover_do_reg_buffer() below during recovery.
1786 xlog_recover_do_inode_buffer(
1788 xlog_recover_item_t
*item
,
1790 xfs_buf_log_format_t
*buf_f
)
1798 int next_unlinked_offset
;
1800 xfs_agino_t
*logged_nextp
;
1801 xfs_agino_t
*buffer_nextp
;
1802 xfs_buf_log_format_v1_t
*obuf_f
;
1803 unsigned int *data_map
= NULL
;
1804 unsigned int map_size
= 0;
1806 switch (buf_f
->blf_type
) {
1808 data_map
= buf_f
->blf_data_map
;
1809 map_size
= buf_f
->blf_map_size
;
1811 case XFS_LI_6_1_BUF
:
1812 case XFS_LI_5_3_BUF
:
1813 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1814 data_map
= obuf_f
->blf_data_map
;
1815 map_size
= obuf_f
->blf_map_size
;
1819 * Set the variables corresponding to the current region to
1820 * 0 so that we'll initialize them on the first pass through
1828 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1829 for (i
= 0; i
< inodes_per_buf
; i
++) {
1830 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1831 offsetof(xfs_dinode_t
, di_next_unlinked
);
1833 while (next_unlinked_offset
>=
1834 (reg_buf_offset
+ reg_buf_bytes
)) {
1836 * The next di_next_unlinked field is beyond
1837 * the current logged region. Find the next
1838 * logged region that contains or is beyond
1839 * the current di_next_unlinked field.
1842 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1845 * If there are no more logged regions in the
1846 * buffer, then we're done.
1852 nbits
= xfs_contig_bits(data_map
, map_size
,
1855 reg_buf_offset
= bit
<< XFS_BLI_SHIFT
;
1856 reg_buf_bytes
= nbits
<< XFS_BLI_SHIFT
;
1861 * If the current logged region starts after the current
1862 * di_next_unlinked field, then move on to the next
1863 * di_next_unlinked field.
1865 if (next_unlinked_offset
< reg_buf_offset
) {
1869 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1870 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLI_CHUNK
) == 0);
1871 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1874 * The current logged region contains a copy of the
1875 * current di_next_unlinked field. Extract its value
1876 * and copy it to the buffer copy.
1878 logged_nextp
= (xfs_agino_t
*)
1879 ((char *)(item
->ri_buf
[item_index
].i_addr
) +
1880 (next_unlinked_offset
- reg_buf_offset
));
1881 if (unlikely(*logged_nextp
== 0)) {
1882 xfs_fs_cmn_err(CE_ALERT
, mp
,
1883 "bad inode buffer log record (ptr = 0x%p, bp = 0x%p). XFS trying to replay bad (0) inode di_next_unlinked field",
1885 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1886 XFS_ERRLEVEL_LOW
, mp
);
1887 return XFS_ERROR(EFSCORRUPTED
);
1890 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1891 next_unlinked_offset
);
1892 INT_SET(*buffer_nextp
, ARCH_CONVERT
, *logged_nextp
);
1899 * Perform a 'normal' buffer recovery. Each logged region of the
1900 * buffer should be copied over the corresponding region in the
1901 * given buffer. The bitmap in the buf log format structure indicates
1902 * where to place the logged data.
1906 xlog_recover_do_reg_buffer(
1908 xlog_recover_item_t
*item
,
1910 xfs_buf_log_format_t
*buf_f
)
1915 xfs_buf_log_format_v1_t
*obuf_f
;
1916 unsigned int *data_map
= NULL
;
1917 unsigned int map_size
= 0;
1920 switch (buf_f
->blf_type
) {
1922 data_map
= buf_f
->blf_data_map
;
1923 map_size
= buf_f
->blf_map_size
;
1925 case XFS_LI_6_1_BUF
:
1926 case XFS_LI_5_3_BUF
:
1927 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
1928 data_map
= obuf_f
->blf_data_map
;
1929 map_size
= obuf_f
->blf_map_size
;
1933 i
= 1; /* 0 is the buf format structure */
1935 bit
= xfs_next_bit(data_map
, map_size
, bit
);
1938 nbits
= xfs_contig_bits(data_map
, map_size
, bit
);
1940 ASSERT(item
->ri_buf
[i
].i_addr
!= 0);
1941 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLI_CHUNK
== 0);
1942 ASSERT(XFS_BUF_COUNT(bp
) >=
1943 ((uint
)bit
<< XFS_BLI_SHIFT
)+(nbits
<<XFS_BLI_SHIFT
));
1946 * Do a sanity check if this is a dquot buffer. Just checking
1947 * the first dquot in the buffer should do. XXXThis is
1948 * probably a good thing to do for other buf types also.
1951 if (buf_f
->blf_flags
&
1952 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
1953 error
= xfs_qm_dqcheck((xfs_disk_dquot_t
*)
1954 item
->ri_buf
[i
].i_addr
,
1955 -1, 0, XFS_QMOPT_DOWARN
,
1956 "dquot_buf_recover");
1959 memcpy(xfs_buf_offset(bp
,
1960 (uint
)bit
<< XFS_BLI_SHIFT
), /* dest */
1961 item
->ri_buf
[i
].i_addr
, /* source */
1962 nbits
<<XFS_BLI_SHIFT
); /* length */
1967 /* Shouldn't be any more regions */
1968 ASSERT(i
== item
->ri_total
);
1972 * Do some primitive error checking on ondisk dquot data structures.
1976 xfs_disk_dquot_t
*ddq
,
1978 uint type
, /* used only when IO_dorepair is true */
1982 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1986 * We can encounter an uninitialized dquot buffer for 2 reasons:
1987 * 1. If we crash while deleting the quotainode(s), and those blks got
1988 * used for user data. This is because we take the path of regular
1989 * file deletion; however, the size field of quotainodes is never
1990 * updated, so all the tricks that we play in itruncate_finish
1991 * don't quite matter.
1993 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1994 * But the allocation will be replayed so we'll end up with an
1995 * uninitialized quota block.
1997 * This is all fine; things are still consistent, and we haven't lost
1998 * any quota information. Just don't complain about bad dquot blks.
2000 if (be16_to_cpu(ddq
->d_magic
) != XFS_DQUOT_MAGIC
) {
2001 if (flags
& XFS_QMOPT_DOWARN
)
2003 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2004 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2007 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2008 if (flags
& XFS_QMOPT_DOWARN
)
2010 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2011 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2015 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2016 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2017 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2018 if (flags
& XFS_QMOPT_DOWARN
)
2020 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2021 str
, id
, ddq
->d_flags
);
2025 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2026 if (flags
& XFS_QMOPT_DOWARN
)
2028 "%s : ondisk-dquot 0x%p, ID mismatch: "
2029 "0x%x expected, found id 0x%x",
2030 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2034 if (!errs
&& ddq
->d_id
) {
2035 if (ddq
->d_blk_softlimit
&&
2036 be64_to_cpu(ddq
->d_bcount
) >=
2037 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2038 if (!ddq
->d_btimer
) {
2039 if (flags
& XFS_QMOPT_DOWARN
)
2041 "%s : Dquot ID 0x%x (0x%p) "
2042 "BLK TIMER NOT STARTED",
2043 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2047 if (ddq
->d_ino_softlimit
&&
2048 be64_to_cpu(ddq
->d_icount
) >=
2049 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2050 if (!ddq
->d_itimer
) {
2051 if (flags
& XFS_QMOPT_DOWARN
)
2053 "%s : Dquot ID 0x%x (0x%p) "
2054 "INODE TIMER NOT STARTED",
2055 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2059 if (ddq
->d_rtb_softlimit
&&
2060 be64_to_cpu(ddq
->d_rtbcount
) >=
2061 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2062 if (!ddq
->d_rtbtimer
) {
2063 if (flags
& XFS_QMOPT_DOWARN
)
2065 "%s : Dquot ID 0x%x (0x%p) "
2066 "RTBLK TIMER NOT STARTED",
2067 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2073 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2076 if (flags
& XFS_QMOPT_DOWARN
)
2077 cmn_err(CE_NOTE
, "Re-initializing dquot ID 0x%x", id
);
2080 * Typically, a repair is only requested by quotacheck.
2083 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2084 memset(d
, 0, sizeof(xfs_dqblk_t
));
2086 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2087 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2088 d
->dd_diskdq
.d_flags
= type
;
2089 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2095 * Perform a dquot buffer recovery.
2096 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2097 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2098 * Else, treat it as a regular buffer and do recovery.
2101 xlog_recover_do_dquot_buffer(
2104 xlog_recover_item_t
*item
,
2106 xfs_buf_log_format_t
*buf_f
)
2111 * Filesystems are required to send in quota flags at mount time.
2113 if (mp
->m_qflags
== 0) {
2118 if (buf_f
->blf_flags
& XFS_BLI_UDQUOT_BUF
)
2119 type
|= XFS_DQ_USER
;
2120 if (buf_f
->blf_flags
& XFS_BLI_PDQUOT_BUF
)
2121 type
|= XFS_DQ_PROJ
;
2122 if (buf_f
->blf_flags
& XFS_BLI_GDQUOT_BUF
)
2123 type
|= XFS_DQ_GROUP
;
2125 * This type of quotas was turned off, so ignore this buffer
2127 if (log
->l_quotaoffs_flag
& type
)
2130 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2134 * This routine replays a modification made to a buffer at runtime.
2135 * There are actually two types of buffer, regular and inode, which
2136 * are handled differently. Inode buffers are handled differently
2137 * in that we only recover a specific set of data from them, namely
2138 * the inode di_next_unlinked fields. This is because all other inode
2139 * data is actually logged via inode records and any data we replay
2140 * here which overlaps that may be stale.
2142 * When meta-data buffers are freed at run time we log a buffer item
2143 * with the XFS_BLI_CANCEL bit set to indicate that previous copies
2144 * of the buffer in the log should not be replayed at recovery time.
2145 * This is so that if the blocks covered by the buffer are reused for
2146 * file data before we crash we don't end up replaying old, freed
2147 * meta-data into a user's file.
2149 * To handle the cancellation of buffer log items, we make two passes
2150 * over the log during recovery. During the first we build a table of
2151 * those buffers which have been cancelled, and during the second we
2152 * only replay those buffers which do not have corresponding cancel
2153 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2154 * for more details on the implementation of the table of cancel records.
2157 xlog_recover_do_buffer_trans(
2159 xlog_recover_item_t
*item
,
2162 xfs_buf_log_format_t
*buf_f
;
2163 xfs_buf_log_format_v1_t
*obuf_f
;
2172 buf_f
= (xfs_buf_log_format_t
*)item
->ri_buf
[0].i_addr
;
2174 if (pass
== XLOG_RECOVER_PASS1
) {
2176 * In this pass we're only looking for buf items
2177 * with the XFS_BLI_CANCEL bit set.
2179 xlog_recover_do_buffer_pass1(log
, buf_f
);
2183 * In this pass we want to recover all the buffers
2184 * which have not been cancelled and are not
2185 * cancellation buffers themselves. The routine
2186 * we call here will tell us whether or not to
2187 * continue with the replay of this buffer.
2189 cancel
= xlog_recover_do_buffer_pass2(log
, buf_f
);
2194 switch (buf_f
->blf_type
) {
2196 blkno
= buf_f
->blf_blkno
;
2197 len
= buf_f
->blf_len
;
2198 flags
= buf_f
->blf_flags
;
2200 case XFS_LI_6_1_BUF
:
2201 case XFS_LI_5_3_BUF
:
2202 obuf_f
= (xfs_buf_log_format_v1_t
*)buf_f
;
2203 blkno
= obuf_f
->blf_blkno
;
2204 len
= obuf_f
->blf_len
;
2205 flags
= obuf_f
->blf_flags
;
2208 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
2209 "xfs_log_recover: unknown buffer type 0x%x, logdev %s",
2210 buf_f
->blf_type
, log
->l_mp
->m_logname
?
2211 log
->l_mp
->m_logname
: "internal");
2212 XFS_ERROR_REPORT("xlog_recover_do_buffer_trans",
2213 XFS_ERRLEVEL_LOW
, log
->l_mp
);
2214 return XFS_ERROR(EFSCORRUPTED
);
2218 if (flags
& XFS_BLI_INODE_BUF
) {
2219 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, blkno
, len
,
2222 bp
= xfs_buf_read(mp
->m_ddev_targp
, blkno
, len
, 0);
2224 if (XFS_BUF_ISERROR(bp
)) {
2225 xfs_ioerror_alert("xlog_recover_do..(read#1)", log
->l_mp
,
2227 error
= XFS_BUF_GETERROR(bp
);
2233 if (flags
& XFS_BLI_INODE_BUF
) {
2234 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2236 (XFS_BLI_UDQUOT_BUF
|XFS_BLI_PDQUOT_BUF
|XFS_BLI_GDQUOT_BUF
)) {
2237 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2239 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2242 return XFS_ERROR(error
);
2245 * Perform delayed write on the buffer. Asynchronous writes will be
2246 * slower when taking into account all the buffers to be flushed.
2248 * Also make sure that only inode buffers with good sizes stay in
2249 * the buffer cache. The kernel moves inodes in buffers of 1 block
2250 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2251 * buffers in the log can be a different size if the log was generated
2252 * by an older kernel using unclustered inode buffers or a newer kernel
2253 * running with a different inode cluster size. Regardless, if the
2254 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2255 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2256 * the buffer out of the buffer cache so that the buffer won't
2257 * overlap with future reads of those inodes.
2259 if (XFS_DINODE_MAGIC
==
2260 INT_GET(*((__uint16_t
*)(xfs_buf_offset(bp
, 0))), ARCH_CONVERT
) &&
2261 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2262 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2264 error
= xfs_bwrite(mp
, bp
);
2266 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2267 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2268 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2269 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2270 xfs_bdwrite(mp
, bp
);
2277 xlog_recover_do_inode_trans(
2279 xlog_recover_item_t
*item
,
2282 xfs_inode_log_format_t
*in_f
;
2294 xfs_dinode_core_t
*dicp
;
2296 if (pass
== XLOG_RECOVER_PASS1
) {
2300 in_f
= (xfs_inode_log_format_t
*)item
->ri_buf
[0].i_addr
;
2301 ino
= in_f
->ilf_ino
;
2303 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2304 imap
.im_blkno
= (xfs_daddr_t
)in_f
->ilf_blkno
;
2305 imap
.im_len
= in_f
->ilf_len
;
2306 imap
.im_boffset
= in_f
->ilf_boffset
;
2309 * It's an old inode format record. We don't know where
2310 * its cluster is located on disk, and we can't allow
2311 * xfs_imap() to figure it out because the inode btrees
2312 * are not ready to be used. Therefore do not pass the
2313 * XFS_IMAP_LOOKUP flag to xfs_imap(). This will give
2314 * us only the single block in which the inode lives
2315 * rather than its cluster, so we must make sure to
2316 * invalidate the buffer when we write it out below.
2319 xfs_imap(log
->l_mp
, NULL
, ino
, &imap
, 0);
2323 * Inode buffers can be freed, look out for it,
2324 * and do not replay the inode.
2326 if (xlog_check_buffer_cancelled(log
, imap
.im_blkno
, imap
.im_len
, 0))
2329 bp
= xfs_buf_read_flags(mp
->m_ddev_targp
, imap
.im_blkno
, imap
.im_len
,
2331 if (XFS_BUF_ISERROR(bp
)) {
2332 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2334 error
= XFS_BUF_GETERROR(bp
);
2339 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2340 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, imap
.im_boffset
);
2343 * Make sure the place we're flushing out to really looks
2346 if (unlikely(INT_GET(dip
->di_core
.di_magic
, ARCH_CONVERT
) != XFS_DINODE_MAGIC
)) {
2348 xfs_fs_cmn_err(CE_ALERT
, mp
,
2349 "xfs_inode_recover: Bad inode magic number, dino ptr = 0x%p, dino bp = 0x%p, ino = %Ld",
2351 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(1)",
2352 XFS_ERRLEVEL_LOW
, mp
);
2353 return XFS_ERROR(EFSCORRUPTED
);
2355 dicp
= (xfs_dinode_core_t
*)(item
->ri_buf
[1].i_addr
);
2356 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2358 xfs_fs_cmn_err(CE_ALERT
, mp
,
2359 "xfs_inode_recover: Bad inode log record, rec ptr 0x%p, ino %Ld",
2361 XFS_ERROR_REPORT("xlog_recover_do_inode_trans(2)",
2362 XFS_ERRLEVEL_LOW
, mp
);
2363 return XFS_ERROR(EFSCORRUPTED
);
2366 /* Skip replay when the on disk inode is newer than the log one */
2367 if (dicp
->di_flushiter
<
2368 INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)) {
2370 * Deal with the wrap case, DI_MAX_FLUSH is less
2371 * than smaller numbers
2373 if ((INT_GET(dip
->di_core
.di_flushiter
, ARCH_CONVERT
)
2375 (dicp
->di_flushiter
< (DI_MAX_FLUSH
>>1))) {
2382 /* Take the opportunity to reset the flush iteration count */
2383 dicp
->di_flushiter
= 0;
2385 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2386 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2387 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2388 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(3)",
2389 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2391 xfs_fs_cmn_err(CE_ALERT
, mp
,
2392 "xfs_inode_recover: Bad regular inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2393 item
, dip
, bp
, ino
);
2394 return XFS_ERROR(EFSCORRUPTED
);
2396 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2397 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2398 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2399 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2400 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(4)",
2401 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2403 xfs_fs_cmn_err(CE_ALERT
, mp
,
2404 "xfs_inode_recover: Bad dir inode log record, rec ptr 0x%p, ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2405 item
, dip
, bp
, ino
);
2406 return XFS_ERROR(EFSCORRUPTED
);
2409 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2410 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(5)",
2411 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2413 xfs_fs_cmn_err(CE_ALERT
, mp
,
2414 "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",
2416 dicp
->di_nextents
+ dicp
->di_anextents
,
2418 return XFS_ERROR(EFSCORRUPTED
);
2420 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2421 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(6)",
2422 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2424 xfs_fs_cmn_err(CE_ALERT
, mp
,
2425 "xfs_inode_recover: Bad inode log rec ptr 0x%p, dino ptr 0x%p, dino bp 0x%p, ino %Ld, forkoff 0x%x",
2426 item
, dip
, bp
, ino
, dicp
->di_forkoff
);
2427 return XFS_ERROR(EFSCORRUPTED
);
2429 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
))) {
2430 XFS_CORRUPTION_ERROR("xlog_recover_do_inode_trans(7)",
2431 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2433 xfs_fs_cmn_err(CE_ALERT
, mp
,
2434 "xfs_inode_recover: Bad inode log record length %d, rec ptr 0x%p",
2435 item
->ri_buf
[1].i_len
, item
);
2436 return XFS_ERROR(EFSCORRUPTED
);
2439 /* The core is in in-core format */
2440 xfs_xlate_dinode_core((xfs_caddr_t
)&dip
->di_core
,
2441 (xfs_dinode_core_t
*)item
->ri_buf
[1].i_addr
, -1);
2443 /* the rest is in on-disk format */
2444 if (item
->ri_buf
[1].i_len
> sizeof(xfs_dinode_core_t
)) {
2445 memcpy((xfs_caddr_t
) dip
+ sizeof(xfs_dinode_core_t
),
2446 item
->ri_buf
[1].i_addr
+ sizeof(xfs_dinode_core_t
),
2447 item
->ri_buf
[1].i_len
- sizeof(xfs_dinode_core_t
));
2450 fields
= in_f
->ilf_fields
;
2451 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2453 INT_SET(dip
->di_u
.di_dev
, ARCH_CONVERT
, in_f
->ilf_u
.ilfu_rdev
);
2457 dip
->di_u
.di_muuid
= in_f
->ilf_u
.ilfu_uuid
;
2461 if (in_f
->ilf_size
== 2)
2462 goto write_inode_buffer
;
2463 len
= item
->ri_buf
[2].i_len
;
2464 src
= item
->ri_buf
[2].i_addr
;
2465 ASSERT(in_f
->ilf_size
<= 4);
2466 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2467 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2468 (len
== in_f
->ilf_dsize
));
2470 switch (fields
& XFS_ILOG_DFORK
) {
2471 case XFS_ILOG_DDATA
:
2473 memcpy(&dip
->di_u
, src
, len
);
2476 case XFS_ILOG_DBROOT
:
2477 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2478 &(dip
->di_u
.di_bmbt
),
2479 XFS_DFORK_DSIZE(dip
, mp
));
2484 * There are no data fork flags set.
2486 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2491 * If we logged any attribute data, recover it. There may or
2492 * may not have been any other non-core data logged in this
2495 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2496 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2501 len
= item
->ri_buf
[attr_index
].i_len
;
2502 src
= item
->ri_buf
[attr_index
].i_addr
;
2503 ASSERT(len
== in_f
->ilf_asize
);
2505 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2506 case XFS_ILOG_ADATA
:
2508 dest
= XFS_DFORK_APTR(dip
);
2509 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2510 memcpy(dest
, src
, len
);
2513 case XFS_ILOG_ABROOT
:
2514 dest
= XFS_DFORK_APTR(dip
);
2515 xfs_bmbt_to_bmdr((xfs_bmbt_block_t
*)src
, len
,
2516 (xfs_bmdr_block_t
*)dest
,
2517 XFS_DFORK_ASIZE(dip
, mp
));
2521 xlog_warn("XFS: xlog_recover_do_inode_trans: Invalid flag");
2524 return XFS_ERROR(EIO
);
2529 if (ITEM_TYPE(item
) == XFS_LI_INODE
) {
2530 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2531 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2532 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2533 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2534 xfs_bdwrite(mp
, bp
);
2537 error
= xfs_bwrite(mp
, bp
);
2544 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2545 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2549 xlog_recover_do_quotaoff_trans(
2551 xlog_recover_item_t
*item
,
2554 xfs_qoff_logformat_t
*qoff_f
;
2556 if (pass
== XLOG_RECOVER_PASS2
) {
2560 qoff_f
= (xfs_qoff_logformat_t
*)item
->ri_buf
[0].i_addr
;
2564 * The logitem format's flag tells us if this was user quotaoff,
2565 * group quotaoff or both.
2567 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2568 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2569 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2570 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2576 * Recover a dquot record
2579 xlog_recover_do_dquot_trans(
2581 xlog_recover_item_t
*item
,
2586 struct xfs_disk_dquot
*ddq
, *recddq
;
2588 xfs_dq_logformat_t
*dq_f
;
2591 if (pass
== XLOG_RECOVER_PASS1
) {
2597 * Filesystems are required to send in quota flags at mount time.
2599 if (mp
->m_qflags
== 0)
2602 recddq
= (xfs_disk_dquot_t
*)item
->ri_buf
[1].i_addr
;
2605 * This type of quotas was turned off, so ignore this record.
2607 type
= INT_GET(recddq
->d_flags
, ARCH_CONVERT
) &
2608 (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2610 if (log
->l_quotaoffs_flag
& type
)
2614 * At this point we know that quota was _not_ turned off.
2615 * Since the mount flags are not indicating to us otherwise, this
2616 * must mean that quota is on, and the dquot needs to be replayed.
2617 * Remember that we may not have fully recovered the superblock yet,
2618 * so we can't do the usual trick of looking at the SB quota bits.
2620 * The other possibility, of course, is that the quota subsystem was
2621 * removed since the last mount - ENOSYS.
2623 dq_f
= (xfs_dq_logformat_t
*)item
->ri_buf
[0].i_addr
;
2625 if ((error
= xfs_qm_dqcheck(recddq
,
2627 0, XFS_QMOPT_DOWARN
,
2628 "xlog_recover_do_dquot_trans (log copy)"))) {
2629 return XFS_ERROR(EIO
);
2631 ASSERT(dq_f
->qlf_len
== 1);
2633 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2635 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2638 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2639 bp
, dq_f
->qlf_blkno
);
2643 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2646 * At least the magic num portion should be on disk because this
2647 * was among a chunk of dquots created earlier, and we did some
2648 * minimal initialization then.
2650 if (xfs_qm_dqcheck(ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2651 "xlog_recover_do_dquot_trans")) {
2653 return XFS_ERROR(EIO
);
2656 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2658 ASSERT(dq_f
->qlf_size
== 2);
2659 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
||
2660 XFS_BUF_FSPRIVATE(bp
, xfs_mount_t
*) == mp
);
2661 XFS_BUF_SET_FSPRIVATE(bp
, mp
);
2662 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_recover_iodone
);
2663 xfs_bdwrite(mp
, bp
);
2669 * This routine is called to create an in-core extent free intent
2670 * item from the efi format structure which was logged on disk.
2671 * It allocates an in-core efi, copies the extents from the format
2672 * structure into it, and adds the efi to the AIL with the given
2676 xlog_recover_do_efi_trans(
2678 xlog_recover_item_t
*item
,
2683 xfs_efi_log_item_t
*efip
;
2684 xfs_efi_log_format_t
*efi_formatp
;
2687 if (pass
== XLOG_RECOVER_PASS1
) {
2691 efi_formatp
= (xfs_efi_log_format_t
*)item
->ri_buf
[0].i_addr
;
2692 ASSERT(item
->ri_buf
[0].i_len
==
2693 (sizeof(xfs_efi_log_format_t
) +
2694 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
))));
2697 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2698 memcpy((char *)&(efip
->efi_format
), (char *)efi_formatp
,
2699 sizeof(xfs_efi_log_format_t
) +
2700 ((efi_formatp
->efi_nextents
- 1) * sizeof(xfs_extent_t
)));
2701 efip
->efi_next_extent
= efi_formatp
->efi_nextents
;
2702 efip
->efi_flags
|= XFS_EFI_COMMITTED
;
2706 * xfs_trans_update_ail() drops the AIL lock.
2708 xfs_trans_update_ail(mp
, (xfs_log_item_t
*)efip
, lsn
, s
);
2713 * This routine is called when an efd format structure is found in
2714 * a committed transaction in the log. It's purpose is to cancel
2715 * the corresponding efi if it was still in the log. To do this
2716 * it searches the AIL for the efi with an id equal to that in the
2717 * efd format structure. If we find it, we remove the efi from the
2721 xlog_recover_do_efd_trans(
2723 xlog_recover_item_t
*item
,
2727 xfs_efd_log_format_t
*efd_formatp
;
2728 xfs_efi_log_item_t
*efip
= NULL
;
2729 xfs_log_item_t
*lip
;
2734 if (pass
== XLOG_RECOVER_PASS1
) {
2738 efd_formatp
= (xfs_efd_log_format_t
*)item
->ri_buf
[0].i_addr
;
2739 ASSERT(item
->ri_buf
[0].i_len
==
2740 (sizeof(xfs_efd_log_format_t
) +
2741 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_t
))));
2742 efi_id
= efd_formatp
->efd_efi_id
;
2745 * Search for the efi with the id in the efd format structure
2750 lip
= xfs_trans_first_ail(mp
, &gen
);
2751 while (lip
!= NULL
) {
2752 if (lip
->li_type
== XFS_LI_EFI
) {
2753 efip
= (xfs_efi_log_item_t
*)lip
;
2754 if (efip
->efi_format
.efi_id
== efi_id
) {
2756 * xfs_trans_delete_ail() drops the
2759 xfs_trans_delete_ail(mp
, lip
, s
);
2763 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
2767 * If we found it, then free it up. If it wasn't there, it
2768 * must have been overwritten in the log. Oh well.
2771 xfs_efi_item_free(efip
);
2778 * Perform the transaction
2780 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2781 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2784 xlog_recover_do_trans(
2786 xlog_recover_t
*trans
,
2790 xlog_recover_item_t
*item
, *first_item
;
2792 if ((error
= xlog_recover_reorder_trans(log
, trans
)))
2794 first_item
= item
= trans
->r_itemq
;
2797 * we don't need to worry about the block number being
2798 * truncated in > 1 TB buffers because in user-land,
2799 * we're now n32 or 64-bit so xfs_daddr_t is 64-bits so
2800 * the blkno's will get through the user-mode buffer
2801 * cache properly. The only bad case is o32 kernels
2802 * where xfs_daddr_t is 32-bits but mount will warn us
2803 * off a > 1 TB filesystem before we get here.
2805 if ((ITEM_TYPE(item
) == XFS_LI_BUF
) ||
2806 (ITEM_TYPE(item
) == XFS_LI_6_1_BUF
) ||
2807 (ITEM_TYPE(item
) == XFS_LI_5_3_BUF
)) {
2808 if ((error
= xlog_recover_do_buffer_trans(log
, item
,
2811 } else if ((ITEM_TYPE(item
) == XFS_LI_INODE
) ||
2812 (ITEM_TYPE(item
) == XFS_LI_6_1_INODE
) ||
2813 (ITEM_TYPE(item
) == XFS_LI_5_3_INODE
)) {
2814 if ((error
= xlog_recover_do_inode_trans(log
, item
,
2817 } else if (ITEM_TYPE(item
) == XFS_LI_EFI
) {
2818 xlog_recover_do_efi_trans(log
, item
, trans
->r_lsn
,
2820 } else if (ITEM_TYPE(item
) == XFS_LI_EFD
) {
2821 xlog_recover_do_efd_trans(log
, item
, pass
);
2822 } else if (ITEM_TYPE(item
) == XFS_LI_DQUOT
) {
2823 if ((error
= xlog_recover_do_dquot_trans(log
, item
,
2826 } else if ((ITEM_TYPE(item
) == XFS_LI_QUOTAOFF
)) {
2827 if ((error
= xlog_recover_do_quotaoff_trans(log
, item
,
2831 xlog_warn("XFS: xlog_recover_do_trans");
2833 error
= XFS_ERROR(EIO
);
2836 item
= item
->ri_next
;
2837 } while (first_item
!= item
);
2843 * Free up any resources allocated by the transaction
2845 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2848 xlog_recover_free_trans(
2849 xlog_recover_t
*trans
)
2851 xlog_recover_item_t
*first_item
, *item
, *free_item
;
2854 item
= first_item
= trans
->r_itemq
;
2857 item
= item
->ri_next
;
2858 /* Free the regions in the item. */
2859 for (i
= 0; i
< free_item
->ri_cnt
; i
++) {
2860 kmem_free(free_item
->ri_buf
[i
].i_addr
,
2861 free_item
->ri_buf
[i
].i_len
);
2863 /* Free the item itself */
2864 kmem_free(free_item
->ri_buf
,
2865 (free_item
->ri_total
* sizeof(xfs_log_iovec_t
)));
2866 kmem_free(free_item
, sizeof(xlog_recover_item_t
));
2867 } while (first_item
!= item
);
2868 /* Free the transaction recover structure */
2869 kmem_free(trans
, sizeof(xlog_recover_t
));
2873 xlog_recover_commit_trans(
2876 xlog_recover_t
*trans
,
2881 if ((error
= xlog_recover_unlink_tid(q
, trans
)))
2883 if ((error
= xlog_recover_do_trans(log
, trans
, pass
)))
2885 xlog_recover_free_trans(trans
); /* no error */
2890 xlog_recover_unmount_trans(
2891 xlog_recover_t
*trans
)
2893 /* Do nothing now */
2894 xlog_warn("XFS: xlog_recover_unmount_trans: Unmount LR");
2899 * There are two valid states of the r_state field. 0 indicates that the
2900 * transaction structure is in a normal state. We have either seen the
2901 * start of the transaction or the last operation we added was not a partial
2902 * operation. If the last operation we added to the transaction was a
2903 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2905 * NOTE: skip LRs with 0 data length.
2908 xlog_recover_process_data(
2910 xlog_recover_t
*rhash
[],
2911 xlog_rec_header_t
*rhead
,
2917 xlog_op_header_t
*ohead
;
2918 xlog_recover_t
*trans
;
2924 lp
= dp
+ INT_GET(rhead
->h_len
, ARCH_CONVERT
);
2925 num_logops
= INT_GET(rhead
->h_num_logops
, ARCH_CONVERT
);
2927 /* check the log format matches our own - else we can't recover */
2928 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2929 return (XFS_ERROR(EIO
));
2931 while ((dp
< lp
) && num_logops
) {
2932 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2933 ohead
= (xlog_op_header_t
*)dp
;
2934 dp
+= sizeof(xlog_op_header_t
);
2935 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2936 ohead
->oh_clientid
!= XFS_LOG
) {
2938 "XFS: xlog_recover_process_data: bad clientid");
2940 return (XFS_ERROR(EIO
));
2942 tid
= INT_GET(ohead
->oh_tid
, ARCH_CONVERT
);
2943 hash
= XLOG_RHASH(tid
);
2944 trans
= xlog_recover_find_tid(rhash
[hash
], tid
);
2945 if (trans
== NULL
) { /* not found; add new tid */
2946 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2947 xlog_recover_new_tid(&rhash
[hash
], tid
,
2948 INT_GET(rhead
->h_lsn
, ARCH_CONVERT
));
2950 ASSERT(dp
+INT_GET(ohead
->oh_len
, ARCH_CONVERT
) <= lp
);
2951 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2952 if (flags
& XLOG_WAS_CONT_TRANS
)
2953 flags
&= ~XLOG_CONTINUE_TRANS
;
2955 case XLOG_COMMIT_TRANS
:
2956 error
= xlog_recover_commit_trans(log
,
2957 &rhash
[hash
], trans
, pass
);
2959 case XLOG_UNMOUNT_TRANS
:
2960 error
= xlog_recover_unmount_trans(trans
);
2962 case XLOG_WAS_CONT_TRANS
:
2963 error
= xlog_recover_add_to_cont_trans(trans
,
2964 dp
, INT_GET(ohead
->oh_len
,
2967 case XLOG_START_TRANS
:
2969 "XFS: xlog_recover_process_data: bad transaction");
2971 error
= XFS_ERROR(EIO
);
2974 case XLOG_CONTINUE_TRANS
:
2975 error
= xlog_recover_add_to_trans(trans
,
2976 dp
, INT_GET(ohead
->oh_len
,
2981 "XFS: xlog_recover_process_data: bad flag");
2983 error
= XFS_ERROR(EIO
);
2989 dp
+= INT_GET(ohead
->oh_len
, ARCH_CONVERT
);
2996 * Process an extent free intent item that was recovered from
2997 * the log. We need to free the extents that it describes.
3000 xlog_recover_process_efi(
3002 xfs_efi_log_item_t
*efip
)
3004 xfs_efd_log_item_t
*efdp
;
3008 xfs_fsblock_t startblock_fsb
;
3010 ASSERT(!(efip
->efi_flags
& XFS_EFI_RECOVERED
));
3013 * First check the validity of the extents described by the
3014 * EFI. If any are bad, then assume that all are bad and
3015 * just toss the EFI.
3017 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3018 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3019 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3020 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3021 if ((startblock_fsb
== 0) ||
3022 (extp
->ext_len
== 0) ||
3023 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3024 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3026 * This will pull the EFI from the AIL and
3027 * free the memory associated with it.
3029 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3034 tp
= xfs_trans_alloc(mp
, 0);
3035 xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3036 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3038 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3039 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3040 xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3041 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3045 efip
->efi_flags
|= XFS_EFI_RECOVERED
;
3046 xfs_trans_commit(tp
, 0, NULL
);
3050 * Verify that once we've encountered something other than an EFI
3051 * in the AIL that there are no more EFIs in the AIL.
3055 xlog_recover_check_ail(
3057 xfs_log_item_t
*lip
,
3063 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3064 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3066 * The check will be bogus if we restart from the
3067 * beginning of the AIL, so ASSERT that we don't.
3068 * We never should since we're holding the AIL lock
3071 ASSERT(gen
== orig_gen
);
3072 } while (lip
!= NULL
);
3077 * When this is called, all of the EFIs which did not have
3078 * corresponding EFDs should be in the AIL. What we do now
3079 * is free the extents associated with each one.
3081 * Since we process the EFIs in normal transactions, they
3082 * will be removed at some point after the commit. This prevents
3083 * us from just walking down the list processing each one.
3084 * We'll use a flag in the EFI to skip those that we've already
3085 * processed and use the AIL iteration mechanism's generation
3086 * count to try to speed this up at least a bit.
3088 * When we start, we know that the EFIs are the only things in
3089 * the AIL. As we process them, however, other items are added
3090 * to the AIL. Since everything added to the AIL must come after
3091 * everything already in the AIL, we stop processing as soon as
3092 * we see something other than an EFI in the AIL.
3095 xlog_recover_process_efis(
3098 xfs_log_item_t
*lip
;
3099 xfs_efi_log_item_t
*efip
;
3107 lip
= xfs_trans_first_ail(mp
, &gen
);
3108 while (lip
!= NULL
) {
3110 * We're done when we see something other than an EFI.
3112 if (lip
->li_type
!= XFS_LI_EFI
) {
3113 xlog_recover_check_ail(mp
, lip
, gen
);
3118 * Skip EFIs that we've already processed.
3120 efip
= (xfs_efi_log_item_t
*)lip
;
3121 if (efip
->efi_flags
& XFS_EFI_RECOVERED
) {
3122 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3127 xlog_recover_process_efi(mp
, efip
);
3129 lip
= xfs_trans_next_ail(mp
, lip
, &gen
, NULL
);
3135 * This routine performs a transaction to null out a bad inode pointer
3136 * in an agi unlinked inode hash bucket.
3139 xlog_recover_clear_agi_bucket(
3141 xfs_agnumber_t agno
,
3150 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3151 xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
), 0, 0, 0);
3153 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
3154 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3155 XFS_FSS_TO_BB(mp
, 1), 0, &agibp
);
3157 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3161 agi
= XFS_BUF_TO_AGI(agibp
);
3162 if (be32_to_cpu(agi
->agi_magicnum
) != XFS_AGI_MAGIC
) {
3163 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3167 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3168 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3169 (sizeof(xfs_agino_t
) * bucket
);
3170 xfs_trans_log_buf(tp
, agibp
, offset
,
3171 (offset
+ sizeof(xfs_agino_t
) - 1));
3173 (void) xfs_trans_commit(tp
, 0, NULL
);
3177 * xlog_iunlink_recover
3179 * This is called during recovery to process any inodes which
3180 * we unlinked but not freed when the system crashed. These
3181 * inodes will be on the lists in the AGI blocks. What we do
3182 * here is scan all the AGIs and fully truncate and free any
3183 * inodes found on the lists. Each inode is removed from the
3184 * lists when it has been fully truncated and is freed. The
3185 * freeing of the inode and its removal from the list must be
3189 xlog_recover_process_iunlinks(
3193 xfs_agnumber_t agno
;
3208 * Prevent any DMAPI event from being sent while in this function.
3210 mp_dmevmask
= mp
->m_dmevmask
;
3213 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3215 * Find the agi for this ag.
3217 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3218 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)),
3219 XFS_FSS_TO_BB(mp
, 1), 0);
3220 if (XFS_BUF_ISERROR(agibp
)) {
3221 xfs_ioerror_alert("xlog_recover_process_iunlinks(#1)",
3223 XFS_AG_DADDR(mp
, agno
, XFS_AGI_DADDR(mp
)));
3225 agi
= XFS_BUF_TO_AGI(agibp
);
3226 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(agi
->agi_magicnum
));
3228 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3230 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3231 while (agino
!= NULLAGINO
) {
3234 * Release the agi buffer so that it can
3235 * be acquired in the normal course of the
3236 * transaction to truncate and free the inode.
3238 xfs_buf_relse(agibp
);
3240 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3241 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
, 0);
3242 ASSERT(error
|| (ip
!= NULL
));
3246 * Get the on disk inode to find the
3247 * next inode in the bucket.
3249 error
= xfs_itobp(mp
, NULL
, ip
, &dip
,
3251 ASSERT(error
|| (dip
!= NULL
));
3255 ASSERT(ip
->i_d
.di_nlink
== 0);
3257 /* setup for the next pass */
3258 agino
= INT_GET(dip
->di_next_unlinked
,
3262 * Prevent any DMAPI event from
3263 * being sent when the
3264 * reference on the inode is
3267 ip
->i_d
.di_dmevmask
= 0;
3270 * If this is a new inode, handle
3271 * it specially. Otherwise,
3272 * just drop our reference to the
3273 * inode. If there are no
3274 * other references, this will
3276 * xfs_inactive() which will
3277 * truncate the file and free
3280 if (ip
->i_d
.di_mode
== 0)
3281 xfs_iput_new(ip
, 0);
3283 VN_RELE(XFS_ITOV(ip
));
3286 * We can't read in the inode
3287 * this bucket points to, or
3288 * this inode is messed up. Just
3289 * ditch this bucket of inodes. We
3290 * will lose some inodes and space,
3291 * but at least we won't hang. Call
3292 * xlog_recover_clear_agi_bucket()
3293 * to perform a transaction to clear
3294 * the inode pointer in the bucket.
3296 xlog_recover_clear_agi_bucket(mp
, agno
,
3303 * Reacquire the agibuffer and continue around
3306 agibp
= xfs_buf_read(mp
->m_ddev_targp
,
3307 XFS_AG_DADDR(mp
, agno
,
3309 XFS_FSS_TO_BB(mp
, 1), 0);
3310 if (XFS_BUF_ISERROR(agibp
)) {
3312 "xlog_recover_process_iunlinks(#2)",
3314 XFS_AG_DADDR(mp
, agno
,
3315 XFS_AGI_DADDR(mp
)));
3317 agi
= XFS_BUF_TO_AGI(agibp
);
3318 ASSERT(XFS_AGI_MAGIC
== be32_to_cpu(
3319 agi
->agi_magicnum
));
3324 * Release the buffer for the current agi so we can
3325 * go on to the next one.
3327 xfs_buf_relse(agibp
);
3330 mp
->m_dmevmask
= mp_dmevmask
;
3336 xlog_pack_data_checksum(
3338 xlog_in_core_t
*iclog
,
3345 up
= (uint
*)iclog
->ic_datap
;
3346 /* divide length by 4 to get # words */
3347 for (i
= 0; i
< (size
>> 2); i
++) {
3348 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3351 INT_SET(iclog
->ic_header
.h_chksum
, ARCH_CONVERT
, chksum
);
3354 #define xlog_pack_data_checksum(log, iclog, size)
3358 * Stamp cycle number in every block
3363 xlog_in_core_t
*iclog
,
3367 int size
= iclog
->ic_offset
+ roundoff
;
3370 xlog_in_core_2_t
*xhdr
;
3372 xlog_pack_data_checksum(log
, iclog
, size
);
3374 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3376 dp
= iclog
->ic_datap
;
3377 for (i
= 0; i
< BTOBB(size
) &&
3378 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3379 iclog
->ic_header
.h_cycle_data
[i
] = *(uint
*)dp
;
3380 *(uint
*)dp
= cycle_lsn
;
3384 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3385 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3386 for ( ; i
< BTOBB(size
); i
++) {
3387 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3388 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3389 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(uint
*)dp
;
3390 *(uint
*)dp
= cycle_lsn
;
3394 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3395 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3400 #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY)
3402 xlog_unpack_data_checksum(
3403 xlog_rec_header_t
*rhead
,
3407 uint
*up
= (uint
*)dp
;
3411 /* divide length by 4 to get # words */
3412 for (i
=0; i
< INT_GET(rhead
->h_len
, ARCH_CONVERT
) >> 2; i
++) {
3413 chksum
^= INT_GET(*up
, ARCH_CONVERT
);
3416 if (chksum
!= INT_GET(rhead
->h_chksum
, ARCH_CONVERT
)) {
3417 if (rhead
->h_chksum
||
3418 ((log
->l_flags
& XLOG_CHKSUM_MISMATCH
) == 0)) {
3420 "XFS: LogR chksum mismatch: was (0x%x) is (0x%x)",
3421 INT_GET(rhead
->h_chksum
, ARCH_CONVERT
), chksum
);
3423 "XFS: Disregard message if filesystem was created with non-DEBUG kernel");
3424 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3426 "XFS: LogR this is a LogV2 filesystem");
3428 log
->l_flags
|= XLOG_CHKSUM_MISMATCH
;
3433 #define xlog_unpack_data_checksum(rhead, dp, log)
3438 xlog_rec_header_t
*rhead
,
3443 xlog_in_core_2_t
*xhdr
;
3445 for (i
= 0; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)) &&
3446 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3447 *(uint
*)dp
= *(uint
*)&rhead
->h_cycle_data
[i
];
3451 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3452 xhdr
= (xlog_in_core_2_t
*)rhead
;
3453 for ( ; i
< BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
)); i
++) {
3454 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3455 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3456 *(uint
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3461 xlog_unpack_data_checksum(rhead
, dp
, log
);
3465 xlog_valid_rec_header(
3467 xlog_rec_header_t
*rhead
,
3473 (INT_GET(rhead
->h_magicno
, ARCH_CONVERT
) !=
3474 XLOG_HEADER_MAGIC_NUM
))) {
3475 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3476 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3477 return XFS_ERROR(EFSCORRUPTED
);
3480 (!rhead
->h_version
||
3481 (INT_GET(rhead
->h_version
, ARCH_CONVERT
) &
3482 (~XLOG_VERSION_OKBITS
)) != 0))) {
3483 xlog_warn("XFS: %s: unrecognised log version (%d).",
3484 __FUNCTION__
, INT_GET(rhead
->h_version
, ARCH_CONVERT
));
3485 return XFS_ERROR(EIO
);
3488 /* LR body must have data or it wouldn't have been written */
3489 hlen
= INT_GET(rhead
->h_len
, ARCH_CONVERT
);
3490 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3491 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3492 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3493 return XFS_ERROR(EFSCORRUPTED
);
3495 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3496 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3497 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3498 return XFS_ERROR(EFSCORRUPTED
);
3504 * Read the log from tail to head and process the log records found.
3505 * Handle the two cases where the tail and head are in the same cycle
3506 * and where the active portion of the log wraps around the end of
3507 * the physical log separately. The pass parameter is passed through
3508 * to the routines called to process the data and is not looked at
3512 xlog_do_recovery_pass(
3514 xfs_daddr_t head_blk
,
3515 xfs_daddr_t tail_blk
,
3518 xlog_rec_header_t
*rhead
;
3520 xfs_caddr_t bufaddr
, offset
;
3521 xfs_buf_t
*hbp
, *dbp
;
3522 int error
= 0, h_size
;
3523 int bblks
, split_bblks
;
3524 int hblks
, split_hblks
, wrapped_hblks
;
3525 xlog_recover_t
*rhash
[XLOG_RHASH_SIZE
];
3527 ASSERT(head_blk
!= tail_blk
);
3530 * Read the header of the tail block and get the iclog buffer size from
3531 * h_size. Use this to tell how many sectors make up the log header.
3533 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
)) {
3535 * When using variable length iclogs, read first sector of
3536 * iclog header and extract the header size from it. Get a
3537 * new hbp that is the correct size.
3539 hbp
= xlog_get_bp(log
, 1);
3542 if ((error
= xlog_bread(log
, tail_blk
, 1, hbp
)))
3544 offset
= xlog_align(log
, tail_blk
, 1, hbp
);
3545 rhead
= (xlog_rec_header_t
*)offset
;
3546 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3549 h_size
= INT_GET(rhead
->h_size
, ARCH_CONVERT
);
3550 if ((INT_GET(rhead
->h_version
, ARCH_CONVERT
)
3551 & XLOG_VERSION_2
) &&
3552 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3553 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3554 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3557 hbp
= xlog_get_bp(log
, hblks
);
3562 ASSERT(log
->l_sectbb_log
== 0);
3564 hbp
= xlog_get_bp(log
, 1);
3565 h_size
= XLOG_BIG_RECORD_BSIZE
;
3570 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3576 memset(rhash
, 0, sizeof(rhash
));
3577 if (tail_blk
<= head_blk
) {
3578 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3579 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3581 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3582 rhead
= (xlog_rec_header_t
*)offset
;
3583 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3587 /* blocks in data section */
3588 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3589 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
);
3592 offset
= xlog_align(log
, blk_no
+ hblks
, bblks
, dbp
);
3593 xlog_unpack_data(rhead
, offset
, log
);
3594 if ((error
= xlog_recover_process_data(log
,
3595 rhash
, rhead
, offset
, pass
)))
3597 blk_no
+= bblks
+ hblks
;
3601 * Perform recovery around the end of the physical log.
3602 * When the head is not on the same cycle number as the tail,
3603 * we can't do a sequential recovery as above.
3606 while (blk_no
< log
->l_logBBsize
) {
3608 * Check for header wrapping around physical end-of-log
3613 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3614 /* Read header in one read */
3615 error
= xlog_bread(log
, blk_no
, hblks
, hbp
);
3618 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3620 /* This LR is split across physical log end */
3621 if (blk_no
!= log
->l_logBBsize
) {
3622 /* some data before physical log end */
3623 ASSERT(blk_no
<= INT_MAX
);
3624 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3625 ASSERT(split_hblks
> 0);
3626 if ((error
= xlog_bread(log
, blk_no
,
3629 offset
= xlog_align(log
, blk_no
,
3633 * Note: this black magic still works with
3634 * large sector sizes (non-512) only because:
3635 * - we increased the buffer size originally
3636 * by 1 sector giving us enough extra space
3637 * for the second read;
3638 * - the log start is guaranteed to be sector
3640 * - we read the log end (LR header start)
3641 * _first_, then the log start (LR header end)
3642 * - order is important.
3644 bufaddr
= XFS_BUF_PTR(hbp
);
3645 XFS_BUF_SET_PTR(hbp
,
3646 bufaddr
+ BBTOB(split_hblks
),
3647 BBTOB(hblks
- split_hblks
));
3648 wrapped_hblks
= hblks
- split_hblks
;
3649 error
= xlog_bread(log
, 0, wrapped_hblks
, hbp
);
3652 XFS_BUF_SET_PTR(hbp
, bufaddr
, BBTOB(hblks
));
3654 offset
= xlog_align(log
, 0,
3655 wrapped_hblks
, hbp
);
3657 rhead
= (xlog_rec_header_t
*)offset
;
3658 error
= xlog_valid_rec_header(log
, rhead
,
3659 split_hblks
? blk_no
: 0);
3663 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3666 /* Read in data for log record */
3667 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3668 error
= xlog_bread(log
, blk_no
, bblks
, dbp
);
3671 offset
= xlog_align(log
, blk_no
, bblks
, dbp
);
3673 /* This log record is split across the
3674 * physical end of log */
3677 if (blk_no
!= log
->l_logBBsize
) {
3678 /* some data is before the physical
3680 ASSERT(!wrapped_hblks
);
3681 ASSERT(blk_no
<= INT_MAX
);
3683 log
->l_logBBsize
- (int)blk_no
;
3684 ASSERT(split_bblks
> 0);
3685 if ((error
= xlog_bread(log
, blk_no
,
3688 offset
= xlog_align(log
, blk_no
,
3692 * Note: this black magic still works with
3693 * large sector sizes (non-512) only because:
3694 * - we increased the buffer size originally
3695 * by 1 sector giving us enough extra space
3696 * for the second read;
3697 * - the log start is guaranteed to be sector
3699 * - we read the log end (LR header start)
3700 * _first_, then the log start (LR header end)
3701 * - order is important.
3703 bufaddr
= XFS_BUF_PTR(dbp
);
3704 XFS_BUF_SET_PTR(dbp
,
3705 bufaddr
+ BBTOB(split_bblks
),
3706 BBTOB(bblks
- split_bblks
));
3707 if ((error
= xlog_bread(log
, wrapped_hblks
,
3708 bblks
- split_bblks
, dbp
)))
3710 XFS_BUF_SET_PTR(dbp
, bufaddr
, h_size
);
3712 offset
= xlog_align(log
, wrapped_hblks
,
3713 bblks
- split_bblks
, dbp
);
3715 xlog_unpack_data(rhead
, offset
, log
);
3716 if ((error
= xlog_recover_process_data(log
, rhash
,
3717 rhead
, offset
, pass
)))
3722 ASSERT(blk_no
>= log
->l_logBBsize
);
3723 blk_no
-= log
->l_logBBsize
;
3725 /* read first part of physical log */
3726 while (blk_no
< head_blk
) {
3727 if ((error
= xlog_bread(log
, blk_no
, hblks
, hbp
)))
3729 offset
= xlog_align(log
, blk_no
, hblks
, hbp
);
3730 rhead
= (xlog_rec_header_t
*)offset
;
3731 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3734 bblks
= (int)BTOBB(INT_GET(rhead
->h_len
, ARCH_CONVERT
));
3735 if ((error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
)))
3737 offset
= xlog_align(log
, blk_no
+hblks
, bblks
, dbp
);
3738 xlog_unpack_data(rhead
, offset
, log
);
3739 if ((error
= xlog_recover_process_data(log
, rhash
,
3740 rhead
, offset
, pass
)))
3742 blk_no
+= bblks
+ hblks
;
3754 * Do the recovery of the log. We actually do this in two phases.
3755 * The two passes are necessary in order to implement the function
3756 * of cancelling a record written into the log. The first pass
3757 * determines those things which have been cancelled, and the
3758 * second pass replays log items normally except for those which
3759 * have been cancelled. The handling of the replay and cancellations
3760 * takes place in the log item type specific routines.
3762 * The table of items which have cancel records in the log is allocated
3763 * and freed at this level, since only here do we know when all of
3764 * the log recovery has been completed.
3767 xlog_do_log_recovery(
3769 xfs_daddr_t head_blk
,
3770 xfs_daddr_t tail_blk
)
3774 ASSERT(head_blk
!= tail_blk
);
3777 * First do a pass to find all of the cancelled buf log items.
3778 * Store them in the buf_cancel_table for use in the second pass.
3780 log
->l_buf_cancel_table
=
3781 (xfs_buf_cancel_t
**)kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3782 sizeof(xfs_buf_cancel_t
*),
3784 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3785 XLOG_RECOVER_PASS1
);
3787 kmem_free(log
->l_buf_cancel_table
,
3788 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3789 log
->l_buf_cancel_table
= NULL
;
3793 * Then do a second pass to actually recover the items in the log.
3794 * When it is complete free the table of buf cancel items.
3796 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3797 XLOG_RECOVER_PASS2
);
3802 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3803 ASSERT(log
->l_buf_cancel_table
[i
] == NULL
);
3807 kmem_free(log
->l_buf_cancel_table
,
3808 XLOG_BC_TABLE_SIZE
* sizeof(xfs_buf_cancel_t
*));
3809 log
->l_buf_cancel_table
= NULL
;
3815 * Do the actual recovery
3820 xfs_daddr_t head_blk
,
3821 xfs_daddr_t tail_blk
)
3828 * First replay the images in the log.
3830 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3835 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3838 * If IO errors happened during recovery, bail out.
3840 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3845 * We now update the tail_lsn since much of the recovery has completed
3846 * and there may be space available to use. If there were no extent
3847 * or iunlinks, we can free up the entire log and set the tail_lsn to
3848 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3849 * lsn of the last known good LR on disk. If there are extent frees
3850 * or iunlinks they will have some entries in the AIL; so we look at
3851 * the AIL to determine how to set the tail_lsn.
3853 xlog_assign_tail_lsn(log
->l_mp
);
3856 * Now that we've finished replaying all buffer and inode
3857 * updates, re-read in the superblock.
3859 bp
= xfs_getsb(log
->l_mp
, 0);
3862 xfsbdstrat(log
->l_mp
, bp
);
3863 if ((error
= xfs_iowait(bp
))) {
3864 xfs_ioerror_alert("xlog_do_recover",
3865 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3871 /* Convert superblock from on-disk format */
3872 sbp
= &log
->l_mp
->m_sb
;
3873 xfs_xlatesb(XFS_BUF_TO_SBP(bp
), sbp
, 1, XFS_SB_ALL_BITS
);
3874 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3875 ASSERT(XFS_SB_GOOD_VERSION(sbp
));
3878 xlog_recover_check_summary(log
);
3880 /* Normal transactions can now occur */
3881 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3886 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3888 * Return error or zero.
3894 xfs_daddr_t head_blk
, tail_blk
;
3897 /* find the tail of the log */
3898 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3901 if (tail_blk
!= head_blk
) {
3902 /* There used to be a comment here:
3904 * disallow recovery on read-only mounts. note -- mount
3905 * checks for ENOSPC and turns it into an intelligent
3907 * ...but this is no longer true. Now, unless you specify
3908 * NORECOVERY (in which case this function would never be
3909 * called), we just go ahead and recover. We do this all
3910 * under the vfs layer, so we can get away with it unless
3911 * the device itself is read-only, in which case we fail.
3913 if ((error
= xfs_dev_is_read_only(log
->l_mp
,
3914 "recovery required"))) {
3919 "Starting XFS recovery on filesystem: %s (logdev: %s)",
3920 log
->l_mp
->m_fsname
, log
->l_mp
->m_logname
?
3921 log
->l_mp
->m_logname
: "internal");
3923 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3924 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3930 * In the first part of recovery we replay inodes and buffers and build
3931 * up the list of extent free items which need to be processed. Here
3932 * we process the extent free items and clean up the on disk unlinked
3933 * inode lists. This is separated from the first part of recovery so
3934 * that the root and real-time bitmap inodes can be read in from disk in
3935 * between the two stages. This is necessary so that we can free space
3936 * in the real-time portion of the file system.
3939 xlog_recover_finish(
3944 * Now we're ready to do the transactions needed for the
3945 * rest of recovery. Start with completing all the extent
3946 * free intent records and then process the unlinked inode
3947 * lists. At this point, we essentially run in normal mode
3948 * except that we're still performing recovery actions
3949 * rather than accepting new requests.
3951 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3952 xlog_recover_process_efis(log
);
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",
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_xlatesb(XFS_BUF_TO_SBP(sbbp
), sbp
, 1, XFS_SB_ALL_BITS
);
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
);