2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_log_recover.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_alloc.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_cksum.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_bmap_btree.h"
44 #include "xfs_error.h"
47 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
54 xlog_clear_stale_blocks(
59 xlog_recover_check_summary(
62 #define xlog_recover_check_summary(log)
66 * This structure is used during recovery to record the buf log items which
67 * have been canceled and should not be replayed.
69 struct xfs_buf_cancel
{
73 struct list_head bc_list
;
77 * Sector aligned buffer routines for buffer create/read/write/access
81 * Verify the given count of basic blocks is valid number of blocks
82 * to specify for an operation involving the given XFS log buffer.
83 * Returns nonzero if the count is valid, 0 otherwise.
87 xlog_buf_bbcount_valid(
91 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
106 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
107 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
109 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
114 * We do log I/O in units of log sectors (a power-of-2
115 * multiple of the basic block size), so we round up the
116 * requested size to accommodate the basic blocks required
117 * for complete log sectors.
119 * In addition, the buffer may be used for a non-sector-
120 * aligned block offset, in which case an I/O of the
121 * requested size could extend beyond the end of the
122 * buffer. If the requested size is only 1 basic block it
123 * will never straddle a sector boundary, so this won't be
124 * an issue. Nor will this be a problem if the log I/O is
125 * done in basic blocks (sector size 1). But otherwise we
126 * extend the buffer by one extra log sector to ensure
127 * there's space to accommodate this possibility.
129 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
130 nbblks
+= log
->l_sectBBsize
;
131 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
133 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
147 * Return the address of the start of the given block number's data
148 * in a log buffer. The buffer covers a log sector-aligned region.
157 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
159 ASSERT(offset
+ nbblks
<= bp
->b_length
);
160 return bp
->b_addr
+ BBTOB(offset
);
165 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
176 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
177 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
179 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
180 return -EFSCORRUPTED
;
183 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
184 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
187 ASSERT(nbblks
<= bp
->b_length
);
189 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
191 bp
->b_io_length
= nbblks
;
194 error
= xfs_buf_submit_wait(bp
);
195 if (error
&& !XFS_FORCED_SHUTDOWN(log
->l_mp
))
196 xfs_buf_ioerror_alert(bp
, __func__
);
210 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
214 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
219 * Read at an offset into the buffer. Returns with the buffer in it's original
220 * state regardless of the result of the read.
225 xfs_daddr_t blk_no
, /* block to read from */
226 int nbblks
, /* blocks to read */
230 char *orig_offset
= bp
->b_addr
;
231 int orig_len
= BBTOB(bp
->b_length
);
234 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
238 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
240 /* must reset buffer pointer even on error */
241 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
248 * Write out the buffer at the given block for the given number of blocks.
249 * The buffer is kept locked across the write and is returned locked.
250 * This can only be used for synchronous log writes.
261 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
262 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
264 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
265 return -EFSCORRUPTED
;
268 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
269 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
272 ASSERT(nbblks
<= bp
->b_length
);
274 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
275 XFS_BUF_ZEROFLAGS(bp
);
278 bp
->b_io_length
= nbblks
;
281 error
= xfs_bwrite(bp
);
283 xfs_buf_ioerror_alert(bp
, __func__
);
290 * dump debug superblock and log record information
293 xlog_header_check_dump(
295 xlog_rec_header_t
*head
)
297 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d",
298 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
299 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
300 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
303 #define xlog_header_check_dump(mp, head)
307 * check log record header for recovery
310 xlog_header_check_recover(
312 xlog_rec_header_t
*head
)
314 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
317 * IRIX doesn't write the h_fmt field and leaves it zeroed
318 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
319 * a dirty log created in IRIX.
321 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
323 "dirty log written in incompatible format - can't recover");
324 xlog_header_check_dump(mp
, head
);
325 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
326 XFS_ERRLEVEL_HIGH
, mp
);
327 return -EFSCORRUPTED
;
328 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
330 "dirty log entry has mismatched uuid - can't recover");
331 xlog_header_check_dump(mp
, head
);
332 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
333 XFS_ERRLEVEL_HIGH
, mp
);
334 return -EFSCORRUPTED
;
340 * read the head block of the log and check the header
343 xlog_header_check_mount(
345 xlog_rec_header_t
*head
)
347 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
349 if (uuid_is_nil(&head
->h_fs_uuid
)) {
351 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
352 * h_fs_uuid is nil, we assume this log was last mounted
353 * by IRIX and continue.
355 xfs_warn(mp
, "nil uuid in log - IRIX style log");
356 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
357 xfs_warn(mp
, "log has mismatched uuid - can't recover");
358 xlog_header_check_dump(mp
, head
);
359 XFS_ERROR_REPORT("xlog_header_check_mount",
360 XFS_ERRLEVEL_HIGH
, mp
);
361 return -EFSCORRUPTED
;
372 * We're not going to bother about retrying
373 * this during recovery. One strike!
375 if (!XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
376 xfs_buf_ioerror_alert(bp
, __func__
);
377 xfs_force_shutdown(bp
->b_target
->bt_mount
,
378 SHUTDOWN_META_IO_ERROR
);
386 * This routine finds (to an approximation) the first block in the physical
387 * log which contains the given cycle. It uses a binary search algorithm.
388 * Note that the algorithm can not be perfect because the disk will not
389 * necessarily be perfect.
392 xlog_find_cycle_start(
395 xfs_daddr_t first_blk
,
396 xfs_daddr_t
*last_blk
,
406 mid_blk
= BLK_AVG(first_blk
, end_blk
);
407 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
408 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
411 mid_cycle
= xlog_get_cycle(offset
);
412 if (mid_cycle
== cycle
)
413 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
415 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
416 mid_blk
= BLK_AVG(first_blk
, end_blk
);
418 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
419 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
427 * Check that a range of blocks does not contain stop_on_cycle_no.
428 * Fill in *new_blk with the block offset where such a block is
429 * found, or with -1 (an invalid block number) if there is no such
430 * block in the range. The scan needs to occur from front to back
431 * and the pointer into the region must be updated since a later
432 * routine will need to perform another test.
435 xlog_find_verify_cycle(
437 xfs_daddr_t start_blk
,
439 uint stop_on_cycle_no
,
440 xfs_daddr_t
*new_blk
)
450 * Greedily allocate a buffer big enough to handle the full
451 * range of basic blocks we'll be examining. If that fails,
452 * try a smaller size. We need to be able to read at least
453 * a log sector, or we're out of luck.
455 bufblks
= 1 << ffs(nbblks
);
456 while (bufblks
> log
->l_logBBsize
)
458 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
460 if (bufblks
< log
->l_sectBBsize
)
464 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
467 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
469 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
473 for (j
= 0; j
< bcount
; j
++) {
474 cycle
= xlog_get_cycle(buf
);
475 if (cycle
== stop_on_cycle_no
) {
492 * Potentially backup over partial log record write.
494 * In the typical case, last_blk is the number of the block directly after
495 * a good log record. Therefore, we subtract one to get the block number
496 * of the last block in the given buffer. extra_bblks contains the number
497 * of blocks we would have read on a previous read. This happens when the
498 * last log record is split over the end of the physical log.
500 * extra_bblks is the number of blocks potentially verified on a previous
501 * call to this routine.
504 xlog_find_verify_log_record(
506 xfs_daddr_t start_blk
,
507 xfs_daddr_t
*last_blk
,
513 xlog_rec_header_t
*head
= NULL
;
516 int num_blks
= *last_blk
- start_blk
;
519 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
521 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
522 if (!(bp
= xlog_get_bp(log
, 1)))
526 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
529 offset
+= ((num_blks
- 1) << BBSHIFT
);
532 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
534 /* valid log record not found */
536 "Log inconsistent (didn't find previous header)");
543 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
548 head
= (xlog_rec_header_t
*)offset
;
550 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
558 * We hit the beginning of the physical log & still no header. Return
559 * to caller. If caller can handle a return of -1, then this routine
560 * will be called again for the end of the physical log.
568 * We have the final block of the good log (the first block
569 * of the log record _before_ the head. So we check the uuid.
571 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
575 * We may have found a log record header before we expected one.
576 * last_blk will be the 1st block # with a given cycle #. We may end
577 * up reading an entire log record. In this case, we don't want to
578 * reset last_blk. Only when last_blk points in the middle of a log
579 * record do we update last_blk.
581 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
582 uint h_size
= be32_to_cpu(head
->h_size
);
584 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
585 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
591 if (*last_blk
- i
+ extra_bblks
!=
592 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
601 * Head is defined to be the point of the log where the next log write
602 * could go. This means that incomplete LR writes at the end are
603 * eliminated when calculating the head. We aren't guaranteed that previous
604 * LR have complete transactions. We only know that a cycle number of
605 * current cycle number -1 won't be present in the log if we start writing
606 * from our current block number.
608 * last_blk contains the block number of the first block with a given
611 * Return: zero if normal, non-zero if error.
616 xfs_daddr_t
*return_head_blk
)
620 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
622 uint first_half_cycle
, last_half_cycle
;
624 int error
, log_bbnum
= log
->l_logBBsize
;
626 /* Is the end of the log device zeroed? */
627 error
= xlog_find_zeroed(log
, &first_blk
);
629 xfs_warn(log
->l_mp
, "empty log check failed");
633 *return_head_blk
= first_blk
;
635 /* Is the whole lot zeroed? */
637 /* Linux XFS shouldn't generate totally zeroed logs -
638 * mkfs etc write a dummy unmount record to a fresh
639 * log so we can store the uuid in there
641 xfs_warn(log
->l_mp
, "totally zeroed log");
647 first_blk
= 0; /* get cycle # of 1st block */
648 bp
= xlog_get_bp(log
, 1);
652 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
656 first_half_cycle
= xlog_get_cycle(offset
);
658 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
659 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
663 last_half_cycle
= xlog_get_cycle(offset
);
664 ASSERT(last_half_cycle
!= 0);
667 * If the 1st half cycle number is equal to the last half cycle number,
668 * then the entire log is stamped with the same cycle number. In this
669 * case, head_blk can't be set to zero (which makes sense). The below
670 * math doesn't work out properly with head_blk equal to zero. Instead,
671 * we set it to log_bbnum which is an invalid block number, but this
672 * value makes the math correct. If head_blk doesn't changed through
673 * all the tests below, *head_blk is set to zero at the very end rather
674 * than log_bbnum. In a sense, log_bbnum and zero are the same block
675 * in a circular file.
677 if (first_half_cycle
== last_half_cycle
) {
679 * In this case we believe that the entire log should have
680 * cycle number last_half_cycle. We need to scan backwards
681 * from the end verifying that there are no holes still
682 * containing last_half_cycle - 1. If we find such a hole,
683 * then the start of that hole will be the new head. The
684 * simple case looks like
685 * x | x ... | x - 1 | x
686 * Another case that fits this picture would be
687 * x | x + 1 | x ... | x
688 * In this case the head really is somewhere at the end of the
689 * log, as one of the latest writes at the beginning was
692 * x | x + 1 | x ... | x - 1 | x
693 * This is really the combination of the above two cases, and
694 * the head has to end up at the start of the x-1 hole at the
697 * In the 256k log case, we will read from the beginning to the
698 * end of the log and search for cycle numbers equal to x-1.
699 * We don't worry about the x+1 blocks that we encounter,
700 * because we know that they cannot be the head since the log
703 head_blk
= log_bbnum
;
704 stop_on_cycle
= last_half_cycle
- 1;
707 * In this case we want to find the first block with cycle
708 * number matching last_half_cycle. We expect the log to be
710 * x + 1 ... | x ... | x
711 * The first block with cycle number x (last_half_cycle) will
712 * be where the new head belongs. First we do a binary search
713 * for the first occurrence of last_half_cycle. The binary
714 * search may not be totally accurate, so then we scan back
715 * from there looking for occurrences of last_half_cycle before
716 * us. If that backwards scan wraps around the beginning of
717 * the log, then we look for occurrences of last_half_cycle - 1
718 * at the end of the log. The cases we're looking for look
720 * v binary search stopped here
721 * x + 1 ... | x | x + 1 | x ... | x
722 * ^ but we want to locate this spot
724 * <---------> less than scan distance
725 * x + 1 ... | x ... | x - 1 | x
726 * ^ we want to locate this spot
728 stop_on_cycle
= last_half_cycle
;
729 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
730 &head_blk
, last_half_cycle
)))
735 * Now validate the answer. Scan back some number of maximum possible
736 * blocks and make sure each one has the expected cycle number. The
737 * maximum is determined by the total possible amount of buffering
738 * in the in-core log. The following number can be made tighter if
739 * we actually look at the block size of the filesystem.
741 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
742 if (head_blk
>= num_scan_bblks
) {
744 * We are guaranteed that the entire check can be performed
747 start_blk
= head_blk
- num_scan_bblks
;
748 if ((error
= xlog_find_verify_cycle(log
,
749 start_blk
, num_scan_bblks
,
750 stop_on_cycle
, &new_blk
)))
754 } else { /* need to read 2 parts of log */
756 * We are going to scan backwards in the log in two parts.
757 * First we scan the physical end of the log. In this part
758 * of the log, we are looking for blocks with cycle number
759 * last_half_cycle - 1.
760 * If we find one, then we know that the log starts there, as
761 * we've found a hole that didn't get written in going around
762 * the end of the physical log. The simple case for this is
763 * x + 1 ... | x ... | x - 1 | x
764 * <---------> less than scan distance
765 * If all of the blocks at the end of the log have cycle number
766 * last_half_cycle, then we check the blocks at the start of
767 * the log looking for occurrences of last_half_cycle. If we
768 * find one, then our current estimate for the location of the
769 * first occurrence of last_half_cycle is wrong and we move
770 * back to the hole we've found. This case looks like
771 * x + 1 ... | x | x + 1 | x ...
772 * ^ binary search stopped here
773 * Another case we need to handle that only occurs in 256k
775 * x + 1 ... | x ... | x+1 | x ...
776 * ^ binary search stops here
777 * In a 256k log, the scan at the end of the log will see the
778 * x + 1 blocks. We need to skip past those since that is
779 * certainly not the head of the log. By searching for
780 * last_half_cycle-1 we accomplish that.
782 ASSERT(head_blk
<= INT_MAX
&&
783 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
784 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
785 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
786 num_scan_bblks
- (int)head_blk
,
787 (stop_on_cycle
- 1), &new_blk
)))
795 * Scan beginning of log now. The last part of the physical
796 * log is good. This scan needs to verify that it doesn't find
797 * the last_half_cycle.
800 ASSERT(head_blk
<= INT_MAX
);
801 if ((error
= xlog_find_verify_cycle(log
,
802 start_blk
, (int)head_blk
,
803 stop_on_cycle
, &new_blk
)))
811 * Now we need to make sure head_blk is not pointing to a block in
812 * the middle of a log record.
814 num_scan_bblks
= XLOG_REC_SHIFT(log
);
815 if (head_blk
>= num_scan_bblks
) {
816 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
818 /* start ptr at last block ptr before head_blk */
819 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
826 ASSERT(head_blk
<= INT_MAX
);
827 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
831 /* We hit the beginning of the log during our search */
832 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
834 ASSERT(start_blk
<= INT_MAX
&&
835 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
836 ASSERT(head_blk
<= INT_MAX
);
837 error
= xlog_find_verify_log_record(log
, start_blk
,
838 &new_blk
, (int)head_blk
);
843 if (new_blk
!= log_bbnum
)
850 if (head_blk
== log_bbnum
)
851 *return_head_blk
= 0;
853 *return_head_blk
= head_blk
;
855 * When returning here, we have a good block number. Bad block
856 * means that during a previous crash, we didn't have a clean break
857 * from cycle number N to cycle number N-1. In this case, we need
858 * to find the first block with cycle number N-1.
866 xfs_warn(log
->l_mp
, "failed to find log head");
871 * Find the sync block number or the tail of the log.
873 * This will be the block number of the last record to have its
874 * associated buffers synced to disk. Every log record header has
875 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
876 * to get a sync block number. The only concern is to figure out which
877 * log record header to believe.
879 * The following algorithm uses the log record header with the largest
880 * lsn. The entire log record does not need to be valid. We only care
881 * that the header is valid.
883 * We could speed up search by using current head_blk buffer, but it is not
889 xfs_daddr_t
*head_blk
,
890 xfs_daddr_t
*tail_blk
)
892 xlog_rec_header_t
*rhead
;
893 xlog_op_header_t
*op_head
;
897 xfs_daddr_t umount_data_blk
;
898 xfs_daddr_t after_umount_blk
;
905 * Find previous log record
907 if ((error
= xlog_find_head(log
, head_blk
)))
910 bp
= xlog_get_bp(log
, 1);
913 if (*head_blk
== 0) { /* special case */
914 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
918 if (xlog_get_cycle(offset
) == 0) {
920 /* leave all other log inited values alone */
926 * Search backwards looking for log record header block
928 ASSERT(*head_blk
< INT_MAX
);
929 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
930 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
934 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
940 * If we haven't found the log record header block, start looking
941 * again from the end of the physical log. XXXmiken: There should be
942 * a check here to make sure we didn't search more than N blocks in
946 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
947 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
951 if (*(__be32
*)offset
==
952 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
959 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
965 /* find blk_no of tail of log */
966 rhead
= (xlog_rec_header_t
*)offset
;
967 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
970 * Reset log values according to the state of the log when we
971 * crashed. In the case where head_blk == 0, we bump curr_cycle
972 * one because the next write starts a new cycle rather than
973 * continuing the cycle of the last good log record. At this
974 * point we have guaranteed that all partial log records have been
975 * accounted for. Therefore, we know that the last good log record
976 * written was complete and ended exactly on the end boundary
977 * of the physical log.
979 log
->l_prev_block
= i
;
980 log
->l_curr_block
= (int)*head_blk
;
981 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
984 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
985 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
986 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
987 BBTOB(log
->l_curr_block
));
988 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
989 BBTOB(log
->l_curr_block
));
992 * Look for unmount record. If we find it, then we know there
993 * was a clean unmount. Since 'i' could be the last block in
994 * the physical log, we convert to a log block before comparing
997 * Save the current tail lsn to use to pass to
998 * xlog_clear_stale_blocks() below. We won't want to clear the
999 * unmount record if there is one, so we pass the lsn of the
1000 * unmount record rather than the block after it.
1002 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1003 int h_size
= be32_to_cpu(rhead
->h_size
);
1004 int h_version
= be32_to_cpu(rhead
->h_version
);
1006 if ((h_version
& XLOG_VERSION_2
) &&
1007 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1008 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1009 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1017 after_umount_blk
= (i
+ hblks
+ (int)
1018 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1019 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1020 if (*head_blk
== after_umount_blk
&&
1021 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1022 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1023 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1027 op_head
= (xlog_op_header_t
*)offset
;
1028 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1030 * Set tail and last sync so that newly written
1031 * log records will point recovery to after the
1032 * current unmount record.
1034 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1035 log
->l_curr_cycle
, after_umount_blk
);
1036 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1037 log
->l_curr_cycle
, after_umount_blk
);
1038 *tail_blk
= after_umount_blk
;
1041 * Note that the unmount was clean. If the unmount
1042 * was not clean, we need to know this to rebuild the
1043 * superblock counters from the perag headers if we
1044 * have a filesystem using non-persistent counters.
1046 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1051 * Make sure that there are no blocks in front of the head
1052 * with the same cycle number as the head. This can happen
1053 * because we allow multiple outstanding log writes concurrently,
1054 * and the later writes might make it out before earlier ones.
1056 * We use the lsn from before modifying it so that we'll never
1057 * overwrite the unmount record after a clean unmount.
1059 * Do this only if we are going to recover the filesystem
1061 * NOTE: This used to say "if (!readonly)"
1062 * However on Linux, we can & do recover a read-only filesystem.
1063 * We only skip recovery if NORECOVERY is specified on mount,
1064 * in which case we would not be here.
1066 * But... if the -device- itself is readonly, just skip this.
1067 * We can't recover this device anyway, so it won't matter.
1069 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1070 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1076 xfs_warn(log
->l_mp
, "failed to locate log tail");
1081 * Is the log zeroed at all?
1083 * The last binary search should be changed to perform an X block read
1084 * once X becomes small enough. You can then search linearly through
1085 * the X blocks. This will cut down on the number of reads we need to do.
1087 * If the log is partially zeroed, this routine will pass back the blkno
1088 * of the first block with cycle number 0. It won't have a complete LR
1092 * 0 => the log is completely written to
1093 * 1 => use *blk_no as the first block of the log
1094 * <0 => error has occurred
1099 xfs_daddr_t
*blk_no
)
1103 uint first_cycle
, last_cycle
;
1104 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1105 xfs_daddr_t num_scan_bblks
;
1106 int error
, log_bbnum
= log
->l_logBBsize
;
1110 /* check totally zeroed log */
1111 bp
= xlog_get_bp(log
, 1);
1114 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1118 first_cycle
= xlog_get_cycle(offset
);
1119 if (first_cycle
== 0) { /* completely zeroed log */
1125 /* check partially zeroed log */
1126 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1130 last_cycle
= xlog_get_cycle(offset
);
1131 if (last_cycle
!= 0) { /* log completely written to */
1134 } else if (first_cycle
!= 1) {
1136 * If the cycle of the last block is zero, the cycle of
1137 * the first block must be 1. If it's not, maybe we're
1138 * not looking at a log... Bail out.
1141 "Log inconsistent or not a log (last==0, first!=1)");
1146 /* we have a partially zeroed log */
1147 last_blk
= log_bbnum
-1;
1148 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1152 * Validate the answer. Because there is no way to guarantee that
1153 * the entire log is made up of log records which are the same size,
1154 * we scan over the defined maximum blocks. At this point, the maximum
1155 * is not chosen to mean anything special. XXXmiken
1157 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1158 ASSERT(num_scan_bblks
<= INT_MAX
);
1160 if (last_blk
< num_scan_bblks
)
1161 num_scan_bblks
= last_blk
;
1162 start_blk
= last_blk
- num_scan_bblks
;
1165 * We search for any instances of cycle number 0 that occur before
1166 * our current estimate of the head. What we're trying to detect is
1167 * 1 ... | 0 | 1 | 0...
1168 * ^ binary search ends here
1170 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1171 (int)num_scan_bblks
, 0, &new_blk
)))
1177 * Potentially backup over partial log record write. We don't need
1178 * to search the end of the log because we know it is zero.
1180 error
= xlog_find_verify_log_record(log
, start_blk
, &last_blk
, 0);
1195 * These are simple subroutines used by xlog_clear_stale_blocks() below
1196 * to initialize a buffer full of empty log record headers and write
1197 * them into the log.
1208 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1210 memset(buf
, 0, BBSIZE
);
1211 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1212 recp
->h_cycle
= cpu_to_be32(cycle
);
1213 recp
->h_version
= cpu_to_be32(
1214 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1215 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1216 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1217 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1218 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1222 xlog_write_log_records(
1233 int sectbb
= log
->l_sectBBsize
;
1234 int end_block
= start_block
+ blocks
;
1240 * Greedily allocate a buffer big enough to handle the full
1241 * range of basic blocks to be written. If that fails, try
1242 * a smaller size. We need to be able to write at least a
1243 * log sector, or we're out of luck.
1245 bufblks
= 1 << ffs(blocks
);
1246 while (bufblks
> log
->l_logBBsize
)
1248 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1250 if (bufblks
< sectbb
)
1254 /* We may need to do a read at the start to fill in part of
1255 * the buffer in the starting sector not covered by the first
1258 balign
= round_down(start_block
, sectbb
);
1259 if (balign
!= start_block
) {
1260 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1264 j
= start_block
- balign
;
1267 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1268 int bcount
, endcount
;
1270 bcount
= min(bufblks
, end_block
- start_block
);
1271 endcount
= bcount
- j
;
1273 /* We may need to do a read at the end to fill in part of
1274 * the buffer in the final sector not covered by the write.
1275 * If this is the same sector as the above read, skip it.
1277 ealign
= round_down(end_block
, sectbb
);
1278 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1279 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1280 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1287 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1288 for (; j
< endcount
; j
++) {
1289 xlog_add_record(log
, offset
, cycle
, i
+j
,
1290 tail_cycle
, tail_block
);
1293 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1296 start_block
+= endcount
;
1306 * This routine is called to blow away any incomplete log writes out
1307 * in front of the log head. We do this so that we won't become confused
1308 * if we come up, write only a little bit more, and then crash again.
1309 * If we leave the partial log records out there, this situation could
1310 * cause us to think those partial writes are valid blocks since they
1311 * have the current cycle number. We get rid of them by overwriting them
1312 * with empty log records with the old cycle number rather than the
1315 * The tail lsn is passed in rather than taken from
1316 * the log so that we will not write over the unmount record after a
1317 * clean unmount in a 512 block log. Doing so would leave the log without
1318 * any valid log records in it until a new one was written. If we crashed
1319 * during that time we would not be able to recover.
1322 xlog_clear_stale_blocks(
1326 int tail_cycle
, head_cycle
;
1327 int tail_block
, head_block
;
1328 int tail_distance
, max_distance
;
1332 tail_cycle
= CYCLE_LSN(tail_lsn
);
1333 tail_block
= BLOCK_LSN(tail_lsn
);
1334 head_cycle
= log
->l_curr_cycle
;
1335 head_block
= log
->l_curr_block
;
1338 * Figure out the distance between the new head of the log
1339 * and the tail. We want to write over any blocks beyond the
1340 * head that we may have written just before the crash, but
1341 * we don't want to overwrite the tail of the log.
1343 if (head_cycle
== tail_cycle
) {
1345 * The tail is behind the head in the physical log,
1346 * so the distance from the head to the tail is the
1347 * distance from the head to the end of the log plus
1348 * the distance from the beginning of the log to the
1351 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1352 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1353 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1354 return -EFSCORRUPTED
;
1356 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1359 * The head is behind the tail in the physical log,
1360 * so the distance from the head to the tail is just
1361 * the tail block minus the head block.
1363 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1364 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1365 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1366 return -EFSCORRUPTED
;
1368 tail_distance
= tail_block
- head_block
;
1372 * If the head is right up against the tail, we can't clear
1375 if (tail_distance
<= 0) {
1376 ASSERT(tail_distance
== 0);
1380 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1382 * Take the smaller of the maximum amount of outstanding I/O
1383 * we could have and the distance to the tail to clear out.
1384 * We take the smaller so that we don't overwrite the tail and
1385 * we don't waste all day writing from the head to the tail
1388 max_distance
= MIN(max_distance
, tail_distance
);
1390 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1392 * We can stomp all the blocks we need to without
1393 * wrapping around the end of the log. Just do it
1394 * in a single write. Use the cycle number of the
1395 * current cycle minus one so that the log will look like:
1398 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1399 head_block
, max_distance
, tail_cycle
,
1405 * We need to wrap around the end of the physical log in
1406 * order to clear all the blocks. Do it in two separate
1407 * I/Os. The first write should be from the head to the
1408 * end of the physical log, and it should use the current
1409 * cycle number minus one just like above.
1411 distance
= log
->l_logBBsize
- head_block
;
1412 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1413 head_block
, distance
, tail_cycle
,
1420 * Now write the blocks at the start of the physical log.
1421 * This writes the remainder of the blocks we want to clear.
1422 * It uses the current cycle number since we're now on the
1423 * same cycle as the head so that we get:
1424 * n ... n ... | n - 1 ...
1425 * ^^^^^ blocks we're writing
1427 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1428 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1429 tail_cycle
, tail_block
);
1437 /******************************************************************************
1439 * Log recover routines
1441 ******************************************************************************
1445 * Sort the log items in the transaction.
1447 * The ordering constraints are defined by the inode allocation and unlink
1448 * behaviour. The rules are:
1450 * 1. Every item is only logged once in a given transaction. Hence it
1451 * represents the last logged state of the item. Hence ordering is
1452 * dependent on the order in which operations need to be performed so
1453 * required initial conditions are always met.
1455 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1456 * there's nothing to replay from them so we can simply cull them
1457 * from the transaction. However, we can't do that until after we've
1458 * replayed all the other items because they may be dependent on the
1459 * cancelled buffer and replaying the cancelled buffer can remove it
1460 * form the cancelled buffer table. Hence they have tobe done last.
1462 * 3. Inode allocation buffers must be replayed before inode items that
1463 * read the buffer and replay changes into it. For filesystems using the
1464 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1465 * treated the same as inode allocation buffers as they create and
1466 * initialise the buffers directly.
1468 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1469 * This ensures that inodes are completely flushed to the inode buffer
1470 * in a "free" state before we remove the unlinked inode list pointer.
1472 * Hence the ordering needs to be inode allocation buffers first, inode items
1473 * second, inode unlink buffers third and cancelled buffers last.
1475 * But there's a problem with that - we can't tell an inode allocation buffer
1476 * apart from a regular buffer, so we can't separate them. We can, however,
1477 * tell an inode unlink buffer from the others, and so we can separate them out
1478 * from all the other buffers and move them to last.
1480 * Hence, 4 lists, in order from head to tail:
1481 * - buffer_list for all buffers except cancelled/inode unlink buffers
1482 * - item_list for all non-buffer items
1483 * - inode_buffer_list for inode unlink buffers
1484 * - cancel_list for the cancelled buffers
1486 * Note that we add objects to the tail of the lists so that first-to-last
1487 * ordering is preserved within the lists. Adding objects to the head of the
1488 * list means when we traverse from the head we walk them in last-to-first
1489 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1490 * but for all other items there may be specific ordering that we need to
1494 xlog_recover_reorder_trans(
1496 struct xlog_recover
*trans
,
1499 xlog_recover_item_t
*item
, *n
;
1501 LIST_HEAD(sort_list
);
1502 LIST_HEAD(cancel_list
);
1503 LIST_HEAD(buffer_list
);
1504 LIST_HEAD(inode_buffer_list
);
1505 LIST_HEAD(inode_list
);
1507 list_splice_init(&trans
->r_itemq
, &sort_list
);
1508 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1509 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1511 switch (ITEM_TYPE(item
)) {
1512 case XFS_LI_ICREATE
:
1513 list_move_tail(&item
->ri_list
, &buffer_list
);
1516 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1517 trace_xfs_log_recover_item_reorder_head(log
,
1519 list_move(&item
->ri_list
, &cancel_list
);
1522 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1523 list_move(&item
->ri_list
, &inode_buffer_list
);
1526 list_move_tail(&item
->ri_list
, &buffer_list
);
1530 case XFS_LI_QUOTAOFF
:
1533 trace_xfs_log_recover_item_reorder_tail(log
,
1535 list_move_tail(&item
->ri_list
, &inode_list
);
1539 "%s: unrecognized type of log operation",
1543 * return the remaining items back to the transaction
1544 * item list so they can be freed in caller.
1546 if (!list_empty(&sort_list
))
1547 list_splice_init(&sort_list
, &trans
->r_itemq
);
1553 ASSERT(list_empty(&sort_list
));
1554 if (!list_empty(&buffer_list
))
1555 list_splice(&buffer_list
, &trans
->r_itemq
);
1556 if (!list_empty(&inode_list
))
1557 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1558 if (!list_empty(&inode_buffer_list
))
1559 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1560 if (!list_empty(&cancel_list
))
1561 list_splice_tail(&cancel_list
, &trans
->r_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_buffer_pass1(
1580 struct xlog_recover_item
*item
)
1582 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1583 struct list_head
*bucket
;
1584 struct xfs_buf_cancel
*bcp
;
1587 * If this isn't a cancel buffer item, then just return.
1589 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1590 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1595 * Insert an xfs_buf_cancel record into the hash table of them.
1596 * If there is already an identical record, bump its reference count.
1598 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1599 list_for_each_entry(bcp
, bucket
, bc_list
) {
1600 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1601 bcp
->bc_len
== buf_f
->blf_len
) {
1603 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1608 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1609 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1610 bcp
->bc_len
= buf_f
->blf_len
;
1611 bcp
->bc_refcount
= 1;
1612 list_add_tail(&bcp
->bc_list
, bucket
);
1614 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1619 * Check to see whether the buffer being recovered has a corresponding
1620 * entry in the buffer cancel record table. If it is, return the cancel
1621 * buffer structure to the caller.
1623 STATIC
struct xfs_buf_cancel
*
1624 xlog_peek_buffer_cancelled(
1630 struct list_head
*bucket
;
1631 struct xfs_buf_cancel
*bcp
;
1633 if (!log
->l_buf_cancel_table
) {
1634 /* empty table means no cancelled buffers in the log */
1635 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1639 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1640 list_for_each_entry(bcp
, bucket
, bc_list
) {
1641 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1646 * We didn't find a corresponding entry in the table, so return 0 so
1647 * that the buffer is NOT cancelled.
1649 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1654 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1655 * otherwise return 0. If the buffer is actually a buffer cancel item
1656 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1657 * table and remove it from the table if this is the last reference.
1659 * We remove the cancel record from the table when we encounter its last
1660 * occurrence in the log so that if the same buffer is re-used again after its
1661 * last cancellation we actually replay the changes made at that point.
1664 xlog_check_buffer_cancelled(
1670 struct xfs_buf_cancel
*bcp
;
1672 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
1677 * We've go a match, so return 1 so that the recovery of this buffer
1678 * is cancelled. If this buffer is actually a buffer cancel log
1679 * item, then decrement the refcount on the one in the table and
1680 * remove it if this is the last reference.
1682 if (flags
& XFS_BLF_CANCEL
) {
1683 if (--bcp
->bc_refcount
== 0) {
1684 list_del(&bcp
->bc_list
);
1692 * Perform recovery for a buffer full of inodes. In these buffers, the only
1693 * data which should be recovered is that which corresponds to the
1694 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1695 * data for the inodes is always logged through the inodes themselves rather
1696 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1698 * The only time when buffers full of inodes are fully recovered is when the
1699 * buffer is full of newly allocated inodes. In this case the buffer will
1700 * not be marked as an inode buffer and so will be sent to
1701 * xlog_recover_do_reg_buffer() below during recovery.
1704 xlog_recover_do_inode_buffer(
1705 struct xfs_mount
*mp
,
1706 xlog_recover_item_t
*item
,
1708 xfs_buf_log_format_t
*buf_f
)
1714 int reg_buf_offset
= 0;
1715 int reg_buf_bytes
= 0;
1716 int next_unlinked_offset
;
1718 xfs_agino_t
*logged_nextp
;
1719 xfs_agino_t
*buffer_nextp
;
1721 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1724 * Post recovery validation only works properly on CRC enabled
1727 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1728 bp
->b_ops
= &xfs_inode_buf_ops
;
1730 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1731 for (i
= 0; i
< inodes_per_buf
; i
++) {
1732 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1733 offsetof(xfs_dinode_t
, di_next_unlinked
);
1735 while (next_unlinked_offset
>=
1736 (reg_buf_offset
+ reg_buf_bytes
)) {
1738 * The next di_next_unlinked field is beyond
1739 * the current logged region. Find the next
1740 * logged region that contains or is beyond
1741 * the current di_next_unlinked field.
1744 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1745 buf_f
->blf_map_size
, bit
);
1748 * If there are no more logged regions in the
1749 * buffer, then we're done.
1754 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1755 buf_f
->blf_map_size
, bit
);
1757 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1758 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1763 * If the current logged region starts after the current
1764 * di_next_unlinked field, then move on to the next
1765 * di_next_unlinked field.
1767 if (next_unlinked_offset
< reg_buf_offset
)
1770 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1771 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1772 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1773 BBTOB(bp
->b_io_length
));
1776 * The current logged region contains a copy of the
1777 * current di_next_unlinked field. Extract its value
1778 * and copy it to the buffer copy.
1780 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1781 next_unlinked_offset
- reg_buf_offset
;
1782 if (unlikely(*logged_nextp
== 0)) {
1784 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1785 "Trying to replay bad (0) inode di_next_unlinked field.",
1787 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1788 XFS_ERRLEVEL_LOW
, mp
);
1789 return -EFSCORRUPTED
;
1792 buffer_nextp
= xfs_buf_offset(bp
, next_unlinked_offset
);
1793 *buffer_nextp
= *logged_nextp
;
1796 * If necessary, recalculate the CRC in the on-disk inode. We
1797 * have to leave the inode in a consistent state for whoever
1800 xfs_dinode_calc_crc(mp
,
1801 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
1809 * V5 filesystems know the age of the buffer on disk being recovered. We can
1810 * have newer objects on disk than we are replaying, and so for these cases we
1811 * don't want to replay the current change as that will make the buffer contents
1812 * temporarily invalid on disk.
1814 * The magic number might not match the buffer type we are going to recover
1815 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1816 * extract the LSN of the existing object in the buffer based on it's current
1817 * magic number. If we don't recognise the magic number in the buffer, then
1818 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1819 * so can recover the buffer.
1821 * Note: we cannot rely solely on magic number matches to determine that the
1822 * buffer has a valid LSN - we also need to verify that it belongs to this
1823 * filesystem, so we need to extract the object's LSN and compare it to that
1824 * which we read from the superblock. If the UUIDs don't match, then we've got a
1825 * stale metadata block from an old filesystem instance that we need to recover
1829 xlog_recover_get_buf_lsn(
1830 struct xfs_mount
*mp
,
1836 void *blk
= bp
->b_addr
;
1840 /* v4 filesystems always recover immediately */
1841 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1842 goto recover_immediately
;
1844 magic32
= be32_to_cpu(*(__be32
*)blk
);
1846 case XFS_ABTB_CRC_MAGIC
:
1847 case XFS_ABTC_CRC_MAGIC
:
1848 case XFS_ABTB_MAGIC
:
1849 case XFS_ABTC_MAGIC
:
1850 case XFS_IBT_CRC_MAGIC
:
1851 case XFS_IBT_MAGIC
: {
1852 struct xfs_btree_block
*btb
= blk
;
1854 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
1855 uuid
= &btb
->bb_u
.s
.bb_uuid
;
1858 case XFS_BMAP_CRC_MAGIC
:
1859 case XFS_BMAP_MAGIC
: {
1860 struct xfs_btree_block
*btb
= blk
;
1862 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
1863 uuid
= &btb
->bb_u
.l
.bb_uuid
;
1867 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
1868 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
1870 case XFS_AGFL_MAGIC
:
1871 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
1872 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
1875 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
1876 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
1878 case XFS_SYMLINK_MAGIC
:
1879 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
1880 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
1882 case XFS_DIR3_BLOCK_MAGIC
:
1883 case XFS_DIR3_DATA_MAGIC
:
1884 case XFS_DIR3_FREE_MAGIC
:
1885 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
1886 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
1888 case XFS_ATTR3_RMT_MAGIC
:
1889 lsn
= be64_to_cpu(((struct xfs_attr3_rmt_hdr
*)blk
)->rm_lsn
);
1890 uuid
= &((struct xfs_attr3_rmt_hdr
*)blk
)->rm_uuid
;
1893 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
1894 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
1900 if (lsn
!= (xfs_lsn_t
)-1) {
1901 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
1902 goto recover_immediately
;
1906 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
1908 case XFS_DIR3_LEAF1_MAGIC
:
1909 case XFS_DIR3_LEAFN_MAGIC
:
1910 case XFS_DA3_NODE_MAGIC
:
1911 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
1912 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
1918 if (lsn
!= (xfs_lsn_t
)-1) {
1919 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
1920 goto recover_immediately
;
1925 * We do individual object checks on dquot and inode buffers as they
1926 * have their own individual LSN records. Also, we could have a stale
1927 * buffer here, so we have to at least recognise these buffer types.
1929 * A notd complexity here is inode unlinked list processing - it logs
1930 * the inode directly in the buffer, but we don't know which inodes have
1931 * been modified, and there is no global buffer LSN. Hence we need to
1932 * recover all inode buffer types immediately. This problem will be
1933 * fixed by logical logging of the unlinked list modifications.
1935 magic16
= be16_to_cpu(*(__be16
*)blk
);
1937 case XFS_DQUOT_MAGIC
:
1938 case XFS_DINODE_MAGIC
:
1939 goto recover_immediately
;
1944 /* unknown buffer contents, recover immediately */
1946 recover_immediately
:
1947 return (xfs_lsn_t
)-1;
1952 * Validate the recovered buffer is of the correct type and attach the
1953 * appropriate buffer operations to them for writeback. Magic numbers are in a
1955 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1956 * the first 32 bits of the buffer (most blocks),
1957 * inside a struct xfs_da_blkinfo at the start of the buffer.
1960 xlog_recover_validate_buf_type(
1961 struct xfs_mount
*mp
,
1963 xfs_buf_log_format_t
*buf_f
)
1965 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
1971 * We can only do post recovery validation on items on CRC enabled
1972 * fielsystems as we need to know when the buffer was written to be able
1973 * to determine if we should have replayed the item. If we replay old
1974 * metadata over a newer buffer, then it will enter a temporarily
1975 * inconsistent state resulting in verification failures. Hence for now
1976 * just avoid the verification stage for non-crc filesystems
1978 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1981 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
1982 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
1983 magicda
= be16_to_cpu(info
->magic
);
1984 switch (xfs_blft_from_flags(buf_f
)) {
1985 case XFS_BLFT_BTREE_BUF
:
1987 case XFS_ABTB_CRC_MAGIC
:
1988 case XFS_ABTC_CRC_MAGIC
:
1989 case XFS_ABTB_MAGIC
:
1990 case XFS_ABTC_MAGIC
:
1991 bp
->b_ops
= &xfs_allocbt_buf_ops
;
1993 case XFS_IBT_CRC_MAGIC
:
1994 case XFS_FIBT_CRC_MAGIC
:
1996 case XFS_FIBT_MAGIC
:
1997 bp
->b_ops
= &xfs_inobt_buf_ops
;
1999 case XFS_BMAP_CRC_MAGIC
:
2000 case XFS_BMAP_MAGIC
:
2001 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2004 xfs_warn(mp
, "Bad btree block magic!");
2009 case XFS_BLFT_AGF_BUF
:
2010 if (magic32
!= XFS_AGF_MAGIC
) {
2011 xfs_warn(mp
, "Bad AGF block magic!");
2015 bp
->b_ops
= &xfs_agf_buf_ops
;
2017 case XFS_BLFT_AGFL_BUF
:
2018 if (magic32
!= XFS_AGFL_MAGIC
) {
2019 xfs_warn(mp
, "Bad AGFL block magic!");
2023 bp
->b_ops
= &xfs_agfl_buf_ops
;
2025 case XFS_BLFT_AGI_BUF
:
2026 if (magic32
!= XFS_AGI_MAGIC
) {
2027 xfs_warn(mp
, "Bad AGI block magic!");
2031 bp
->b_ops
= &xfs_agi_buf_ops
;
2033 case XFS_BLFT_UDQUOT_BUF
:
2034 case XFS_BLFT_PDQUOT_BUF
:
2035 case XFS_BLFT_GDQUOT_BUF
:
2036 #ifdef CONFIG_XFS_QUOTA
2037 if (magic16
!= XFS_DQUOT_MAGIC
) {
2038 xfs_warn(mp
, "Bad DQUOT block magic!");
2042 bp
->b_ops
= &xfs_dquot_buf_ops
;
2045 "Trying to recover dquots without QUOTA support built in!");
2049 case XFS_BLFT_DINO_BUF
:
2050 if (magic16
!= XFS_DINODE_MAGIC
) {
2051 xfs_warn(mp
, "Bad INODE block magic!");
2055 bp
->b_ops
= &xfs_inode_buf_ops
;
2057 case XFS_BLFT_SYMLINK_BUF
:
2058 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2059 xfs_warn(mp
, "Bad symlink block magic!");
2063 bp
->b_ops
= &xfs_symlink_buf_ops
;
2065 case XFS_BLFT_DIR_BLOCK_BUF
:
2066 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2067 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2068 xfs_warn(mp
, "Bad dir block magic!");
2072 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2074 case XFS_BLFT_DIR_DATA_BUF
:
2075 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2076 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2077 xfs_warn(mp
, "Bad dir data magic!");
2081 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2083 case XFS_BLFT_DIR_FREE_BUF
:
2084 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2085 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2086 xfs_warn(mp
, "Bad dir3 free magic!");
2090 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2092 case XFS_BLFT_DIR_LEAF1_BUF
:
2093 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2094 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2095 xfs_warn(mp
, "Bad dir leaf1 magic!");
2099 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2101 case XFS_BLFT_DIR_LEAFN_BUF
:
2102 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2103 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2104 xfs_warn(mp
, "Bad dir leafn magic!");
2108 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2110 case XFS_BLFT_DA_NODE_BUF
:
2111 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2112 magicda
!= XFS_DA3_NODE_MAGIC
) {
2113 xfs_warn(mp
, "Bad da node magic!");
2117 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2119 case XFS_BLFT_ATTR_LEAF_BUF
:
2120 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2121 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2122 xfs_warn(mp
, "Bad attr leaf magic!");
2126 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2128 case XFS_BLFT_ATTR_RMT_BUF
:
2129 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2130 xfs_warn(mp
, "Bad attr remote magic!");
2134 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2136 case XFS_BLFT_SB_BUF
:
2137 if (magic32
!= XFS_SB_MAGIC
) {
2138 xfs_warn(mp
, "Bad SB block magic!");
2142 bp
->b_ops
= &xfs_sb_buf_ops
;
2145 xfs_warn(mp
, "Unknown buffer type %d!",
2146 xfs_blft_from_flags(buf_f
));
2152 * Perform a 'normal' buffer recovery. Each logged region of the
2153 * buffer should be copied over the corresponding region in the
2154 * given buffer. The bitmap in the buf log format structure indicates
2155 * where to place the logged data.
2158 xlog_recover_do_reg_buffer(
2159 struct xfs_mount
*mp
,
2160 xlog_recover_item_t
*item
,
2162 xfs_buf_log_format_t
*buf_f
)
2169 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2172 i
= 1; /* 0 is the buf format structure */
2174 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2175 buf_f
->blf_map_size
, bit
);
2178 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2179 buf_f
->blf_map_size
, bit
);
2181 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2182 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2183 ASSERT(BBTOB(bp
->b_io_length
) >=
2184 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2187 * The dirty regions logged in the buffer, even though
2188 * contiguous, may span multiple chunks. This is because the
2189 * dirty region may span a physical page boundary in a buffer
2190 * and hence be split into two separate vectors for writing into
2191 * the log. Hence we need to trim nbits back to the length of
2192 * the current region being copied out of the log.
2194 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2195 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2198 * Do a sanity check if this is a dquot buffer. Just checking
2199 * the first dquot in the buffer should do. XXXThis is
2200 * probably a good thing to do for other buf types also.
2203 if (buf_f
->blf_flags
&
2204 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2205 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2207 "XFS: NULL dquot in %s.", __func__
);
2210 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2212 "XFS: dquot too small (%d) in %s.",
2213 item
->ri_buf
[i
].i_len
, __func__
);
2216 error
= xfs_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2217 -1, 0, XFS_QMOPT_DOWARN
,
2218 "dquot_buf_recover");
2223 memcpy(xfs_buf_offset(bp
,
2224 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2225 item
->ri_buf
[i
].i_addr
, /* source */
2226 nbits
<<XFS_BLF_SHIFT
); /* length */
2232 /* Shouldn't be any more regions */
2233 ASSERT(i
== item
->ri_total
);
2235 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2239 * Perform a dquot buffer recovery.
2240 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2241 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2242 * Else, treat it as a regular buffer and do recovery.
2244 * Return false if the buffer was tossed and true if we recovered the buffer to
2245 * indicate to the caller if the buffer needs writing.
2248 xlog_recover_do_dquot_buffer(
2249 struct xfs_mount
*mp
,
2251 struct xlog_recover_item
*item
,
2253 struct xfs_buf_log_format
*buf_f
)
2257 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2260 * Filesystems are required to send in quota flags at mount time.
2266 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2267 type
|= XFS_DQ_USER
;
2268 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2269 type
|= XFS_DQ_PROJ
;
2270 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2271 type
|= XFS_DQ_GROUP
;
2273 * This type of quotas was turned off, so ignore this buffer
2275 if (log
->l_quotaoffs_flag
& type
)
2278 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2283 * This routine replays a modification made to a buffer at runtime.
2284 * There are actually two types of buffer, regular and inode, which
2285 * are handled differently. Inode buffers are handled differently
2286 * in that we only recover a specific set of data from them, namely
2287 * the inode di_next_unlinked fields. This is because all other inode
2288 * data is actually logged via inode records and any data we replay
2289 * here which overlaps that may be stale.
2291 * When meta-data buffers are freed at run time we log a buffer item
2292 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2293 * of the buffer in the log should not be replayed at recovery time.
2294 * This is so that if the blocks covered by the buffer are reused for
2295 * file data before we crash we don't end up replaying old, freed
2296 * meta-data into a user's file.
2298 * To handle the cancellation of buffer log items, we make two passes
2299 * over the log during recovery. During the first we build a table of
2300 * those buffers which have been cancelled, and during the second we
2301 * only replay those buffers which do not have corresponding cancel
2302 * records in the table. See xlog_recover_buffer_pass[1,2] above
2303 * for more details on the implementation of the table of cancel records.
2306 xlog_recover_buffer_pass2(
2308 struct list_head
*buffer_list
,
2309 struct xlog_recover_item
*item
,
2310 xfs_lsn_t current_lsn
)
2312 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2313 xfs_mount_t
*mp
= log
->l_mp
;
2320 * In this pass we only want to recover all the buffers which have
2321 * not been cancelled and are not cancellation buffers themselves.
2323 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2324 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2325 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2329 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2332 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2333 buf_flags
|= XBF_UNMAPPED
;
2335 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2339 error
= bp
->b_error
;
2341 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2346 * Recover the buffer only if we get an LSN from it and it's less than
2347 * the lsn of the transaction we are replaying.
2349 * Note that we have to be extremely careful of readahead here.
2350 * Readahead does not attach verfiers to the buffers so if we don't
2351 * actually do any replay after readahead because of the LSN we found
2352 * in the buffer if more recent than that current transaction then we
2353 * need to attach the verifier directly. Failure to do so can lead to
2354 * future recovery actions (e.g. EFI and unlinked list recovery) can
2355 * operate on the buffers and they won't get the verifier attached. This
2356 * can lead to blocks on disk having the correct content but a stale
2359 * It is safe to assume these clean buffers are currently up to date.
2360 * If the buffer is dirtied by a later transaction being replayed, then
2361 * the verifier will be reset to match whatever recover turns that
2364 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2365 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2366 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2370 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2371 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2374 } else if (buf_f
->blf_flags
&
2375 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2378 dirty
= xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2382 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2386 * Perform delayed write on the buffer. Asynchronous writes will be
2387 * slower when taking into account all the buffers to be flushed.
2389 * Also make sure that only inode buffers with good sizes stay in
2390 * the buffer cache. The kernel moves inodes in buffers of 1 block
2391 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2392 * buffers in the log can be a different size if the log was generated
2393 * by an older kernel using unclustered inode buffers or a newer kernel
2394 * running with a different inode cluster size. Regardless, if the
2395 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2396 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2397 * the buffer out of the buffer cache so that the buffer won't
2398 * overlap with future reads of those inodes.
2400 if (XFS_DINODE_MAGIC
==
2401 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2402 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2403 (__uint32_t
)log
->l_mp
->m_inode_cluster_size
))) {
2405 error
= xfs_bwrite(bp
);
2407 ASSERT(bp
->b_target
->bt_mount
== mp
);
2408 bp
->b_iodone
= xlog_recover_iodone
;
2409 xfs_buf_delwri_queue(bp
, buffer_list
);
2418 * Inode fork owner changes
2420 * If we have been told that we have to reparent the inode fork, it's because an
2421 * extent swap operation on a CRC enabled filesystem has been done and we are
2422 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2425 * The complexity here is that we don't have an inode context to work with, so
2426 * after we've replayed the inode we need to instantiate one. This is where the
2429 * We are in the middle of log recovery, so we can't run transactions. That
2430 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2431 * that will result in the corresponding iput() running the inode through
2432 * xfs_inactive(). If we've just replayed an inode core that changes the link
2433 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2434 * transactions (bad!).
2436 * So, to avoid this, we instantiate an inode directly from the inode core we've
2437 * just recovered. We have the buffer still locked, and all we really need to
2438 * instantiate is the inode core and the forks being modified. We can do this
2439 * manually, then run the inode btree owner change, and then tear down the
2440 * xfs_inode without having to run any transactions at all.
2442 * Also, because we don't have a transaction context available here but need to
2443 * gather all the buffers we modify for writeback so we pass the buffer_list
2444 * instead for the operation to use.
2448 xfs_recover_inode_owner_change(
2449 struct xfs_mount
*mp
,
2450 struct xfs_dinode
*dip
,
2451 struct xfs_inode_log_format
*in_f
,
2452 struct list_head
*buffer_list
)
2454 struct xfs_inode
*ip
;
2457 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2459 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2463 /* instantiate the inode */
2464 xfs_dinode_from_disk(&ip
->i_d
, dip
);
2465 ASSERT(ip
->i_d
.di_version
>= 3);
2467 error
= xfs_iformat_fork(ip
, dip
);
2472 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2473 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2474 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2475 ip
->i_ino
, buffer_list
);
2480 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2481 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2482 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2483 ip
->i_ino
, buffer_list
);
2494 xlog_recover_inode_pass2(
2496 struct list_head
*buffer_list
,
2497 struct xlog_recover_item
*item
,
2498 xfs_lsn_t current_lsn
)
2500 xfs_inode_log_format_t
*in_f
;
2501 xfs_mount_t
*mp
= log
->l_mp
;
2510 xfs_icdinode_t
*dicp
;
2514 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2515 in_f
= item
->ri_buf
[0].i_addr
;
2517 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2519 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2525 * Inode buffers can be freed, look out for it,
2526 * and do not replay the inode.
2528 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2529 in_f
->ilf_len
, 0)) {
2531 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2534 trace_xfs_log_recover_inode_recover(log
, in_f
);
2536 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2537 &xfs_inode_buf_ops
);
2542 error
= bp
->b_error
;
2544 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2547 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2548 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2551 * Make sure the place we're flushing out to really looks
2554 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2556 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2557 __func__
, dip
, bp
, in_f
->ilf_ino
);
2558 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2559 XFS_ERRLEVEL_LOW
, mp
);
2560 error
= -EFSCORRUPTED
;
2563 dicp
= item
->ri_buf
[1].i_addr
;
2564 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2566 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2567 __func__
, item
, in_f
->ilf_ino
);
2568 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2569 XFS_ERRLEVEL_LOW
, mp
);
2570 error
= -EFSCORRUPTED
;
2575 * If the inode has an LSN in it, recover the inode only if it's less
2576 * than the lsn of the transaction we are replaying. Note: we still
2577 * need to replay an owner change even though the inode is more recent
2578 * than the transaction as there is no guarantee that all the btree
2579 * blocks are more recent than this transaction, too.
2581 if (dip
->di_version
>= 3) {
2582 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
2584 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2585 trace_xfs_log_recover_inode_skip(log
, in_f
);
2587 goto out_owner_change
;
2592 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2593 * are transactional and if ordering is necessary we can determine that
2594 * more accurately by the LSN field in the V3 inode core. Don't trust
2595 * the inode versions we might be changing them here - use the
2596 * superblock flag to determine whether we need to look at di_flushiter
2597 * to skip replay when the on disk inode is newer than the log one
2599 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2600 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2602 * Deal with the wrap case, DI_MAX_FLUSH is less
2603 * than smaller numbers
2605 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2606 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2609 trace_xfs_log_recover_inode_skip(log
, in_f
);
2615 /* Take the opportunity to reset the flush iteration count */
2616 dicp
->di_flushiter
= 0;
2618 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2619 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2620 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2621 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2622 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2624 "%s: Bad regular inode log record, rec ptr 0x%p, "
2625 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2626 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2627 error
= -EFSCORRUPTED
;
2630 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2631 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2632 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2633 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2634 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2635 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2637 "%s: Bad dir inode log record, rec ptr 0x%p, "
2638 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2639 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2640 error
= -EFSCORRUPTED
;
2644 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2645 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2646 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2648 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2649 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2650 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2651 dicp
->di_nextents
+ dicp
->di_anextents
,
2653 error
= -EFSCORRUPTED
;
2656 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2657 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2658 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2660 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2661 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2662 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2663 error
= -EFSCORRUPTED
;
2666 isize
= xfs_icdinode_size(dicp
->di_version
);
2667 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2668 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2669 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2671 "%s: Bad inode log record length %d, rec ptr 0x%p",
2672 __func__
, item
->ri_buf
[1].i_len
, item
);
2673 error
= -EFSCORRUPTED
;
2677 /* The core is in in-core format */
2678 xfs_dinode_to_disk(dip
, dicp
);
2680 /* the rest is in on-disk format */
2681 if (item
->ri_buf
[1].i_len
> isize
) {
2682 memcpy((char *)dip
+ isize
,
2683 item
->ri_buf
[1].i_addr
+ isize
,
2684 item
->ri_buf
[1].i_len
- isize
);
2687 fields
= in_f
->ilf_fields
;
2688 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2690 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2693 memcpy(XFS_DFORK_DPTR(dip
),
2694 &in_f
->ilf_u
.ilfu_uuid
,
2699 if (in_f
->ilf_size
== 2)
2700 goto out_owner_change
;
2701 len
= item
->ri_buf
[2].i_len
;
2702 src
= item
->ri_buf
[2].i_addr
;
2703 ASSERT(in_f
->ilf_size
<= 4);
2704 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2705 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2706 (len
== in_f
->ilf_dsize
));
2708 switch (fields
& XFS_ILOG_DFORK
) {
2709 case XFS_ILOG_DDATA
:
2711 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2714 case XFS_ILOG_DBROOT
:
2715 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2716 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2717 XFS_DFORK_DSIZE(dip
, mp
));
2722 * There are no data fork flags set.
2724 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2729 * If we logged any attribute data, recover it. There may or
2730 * may not have been any other non-core data logged in this
2733 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2734 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2739 len
= item
->ri_buf
[attr_index
].i_len
;
2740 src
= item
->ri_buf
[attr_index
].i_addr
;
2741 ASSERT(len
== in_f
->ilf_asize
);
2743 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2744 case XFS_ILOG_ADATA
:
2746 dest
= XFS_DFORK_APTR(dip
);
2747 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2748 memcpy(dest
, src
, len
);
2751 case XFS_ILOG_ABROOT
:
2752 dest
= XFS_DFORK_APTR(dip
);
2753 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2754 len
, (xfs_bmdr_block_t
*)dest
,
2755 XFS_DFORK_ASIZE(dip
, mp
));
2759 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2767 if (in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
))
2768 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
2770 /* re-generate the checksum. */
2771 xfs_dinode_calc_crc(log
->l_mp
, dip
);
2773 ASSERT(bp
->b_target
->bt_mount
== mp
);
2774 bp
->b_iodone
= xlog_recover_iodone
;
2775 xfs_buf_delwri_queue(bp
, buffer_list
);
2786 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2787 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2791 xlog_recover_quotaoff_pass1(
2793 struct xlog_recover_item
*item
)
2795 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2799 * The logitem format's flag tells us if this was user quotaoff,
2800 * group/project quotaoff or both.
2802 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2803 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2804 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2805 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2806 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2807 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2813 * Recover a dquot record
2816 xlog_recover_dquot_pass2(
2818 struct list_head
*buffer_list
,
2819 struct xlog_recover_item
*item
,
2820 xfs_lsn_t current_lsn
)
2822 xfs_mount_t
*mp
= log
->l_mp
;
2824 struct xfs_disk_dquot
*ddq
, *recddq
;
2826 xfs_dq_logformat_t
*dq_f
;
2831 * Filesystems are required to send in quota flags at mount time.
2833 if (mp
->m_qflags
== 0)
2836 recddq
= item
->ri_buf
[1].i_addr
;
2837 if (recddq
== NULL
) {
2838 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2841 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2842 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2843 item
->ri_buf
[1].i_len
, __func__
);
2848 * This type of quotas was turned off, so ignore this record.
2850 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2852 if (log
->l_quotaoffs_flag
& type
)
2856 * At this point we know that quota was _not_ turned off.
2857 * Since the mount flags are not indicating to us otherwise, this
2858 * must mean that quota is on, and the dquot needs to be replayed.
2859 * Remember that we may not have fully recovered the superblock yet,
2860 * so we can't do the usual trick of looking at the SB quota bits.
2862 * The other possibility, of course, is that the quota subsystem was
2863 * removed since the last mount - ENOSYS.
2865 dq_f
= item
->ri_buf
[0].i_addr
;
2867 error
= xfs_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2868 "xlog_recover_dquot_pass2 (log copy)");
2871 ASSERT(dq_f
->qlf_len
== 1);
2874 * At this point we are assuming that the dquots have been allocated
2875 * and hence the buffer has valid dquots stamped in it. It should,
2876 * therefore, pass verifier validation. If the dquot is bad, then the
2877 * we'll return an error here, so we don't need to specifically check
2878 * the dquot in the buffer after the verifier has run.
2880 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
2881 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
2882 &xfs_dquot_buf_ops
);
2887 ddq
= xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2890 * If the dquot has an LSN in it, recover the dquot only if it's less
2891 * than the lsn of the transaction we are replaying.
2893 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2894 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
2895 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
2897 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2902 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2903 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2904 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
2908 ASSERT(dq_f
->qlf_size
== 2);
2909 ASSERT(bp
->b_target
->bt_mount
== mp
);
2910 bp
->b_iodone
= xlog_recover_iodone
;
2911 xfs_buf_delwri_queue(bp
, buffer_list
);
2919 * This routine is called to create an in-core extent free intent
2920 * item from the efi format structure which was logged on disk.
2921 * It allocates an in-core efi, copies the extents from the format
2922 * structure into it, and adds the efi to the AIL with the given
2926 xlog_recover_efi_pass2(
2928 struct xlog_recover_item
*item
,
2932 xfs_mount_t
*mp
= log
->l_mp
;
2933 xfs_efi_log_item_t
*efip
;
2934 xfs_efi_log_format_t
*efi_formatp
;
2936 efi_formatp
= item
->ri_buf
[0].i_addr
;
2938 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2939 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2940 &(efip
->efi_format
)))) {
2941 xfs_efi_item_free(efip
);
2944 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
2946 spin_lock(&log
->l_ailp
->xa_lock
);
2948 * xfs_trans_ail_update() drops the AIL lock.
2950 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
2956 * This routine is called when an efd format structure is found in
2957 * a committed transaction in the log. It's purpose is to cancel
2958 * the corresponding efi if it was still in the log. To do this
2959 * it searches the AIL for the efi with an id equal to that in the
2960 * efd format structure. If we find it, we remove the efi from the
2964 xlog_recover_efd_pass2(
2966 struct xlog_recover_item
*item
)
2968 xfs_efd_log_format_t
*efd_formatp
;
2969 xfs_efi_log_item_t
*efip
= NULL
;
2970 xfs_log_item_t
*lip
;
2972 struct xfs_ail_cursor cur
;
2973 struct xfs_ail
*ailp
= log
->l_ailp
;
2975 efd_formatp
= item
->ri_buf
[0].i_addr
;
2976 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2977 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2978 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2979 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2980 efi_id
= efd_formatp
->efd_efi_id
;
2983 * Search for the efi with the id in the efd format structure
2986 spin_lock(&ailp
->xa_lock
);
2987 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2988 while (lip
!= NULL
) {
2989 if (lip
->li_type
== XFS_LI_EFI
) {
2990 efip
= (xfs_efi_log_item_t
*)lip
;
2991 if (efip
->efi_format
.efi_id
== efi_id
) {
2993 * xfs_trans_ail_delete() drops the
2996 xfs_trans_ail_delete(ailp
, lip
,
2997 SHUTDOWN_CORRUPT_INCORE
);
2998 xfs_efi_item_free(efip
);
2999 spin_lock(&ailp
->xa_lock
);
3003 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3005 xfs_trans_ail_cursor_done(&cur
);
3006 spin_unlock(&ailp
->xa_lock
);
3012 * This routine is called when an inode create format structure is found in a
3013 * committed transaction in the log. It's purpose is to initialise the inodes
3014 * being allocated on disk. This requires us to get inode cluster buffers that
3015 * match the range to be intialised, stamped with inode templates and written
3016 * by delayed write so that subsequent modifications will hit the cached buffer
3017 * and only need writing out at the end of recovery.
3020 xlog_recover_do_icreate_pass2(
3022 struct list_head
*buffer_list
,
3023 xlog_recover_item_t
*item
)
3025 struct xfs_mount
*mp
= log
->l_mp
;
3026 struct xfs_icreate_log
*icl
;
3027 xfs_agnumber_t agno
;
3028 xfs_agblock_t agbno
;
3031 xfs_agblock_t length
;
3033 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3034 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3035 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3039 if (icl
->icl_size
!= 1) {
3040 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3044 agno
= be32_to_cpu(icl
->icl_ag
);
3045 if (agno
>= mp
->m_sb
.sb_agcount
) {
3046 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3049 agbno
= be32_to_cpu(icl
->icl_agbno
);
3050 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3051 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3054 isize
= be32_to_cpu(icl
->icl_isize
);
3055 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3056 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3059 count
= be32_to_cpu(icl
->icl_count
);
3061 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3064 length
= be32_to_cpu(icl
->icl_length
);
3065 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3066 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3071 * The inode chunk is either full or sparse and we only support
3072 * m_ialloc_min_blks sized sparse allocations at this time.
3074 if (length
!= mp
->m_ialloc_blks
&&
3075 length
!= mp
->m_ialloc_min_blks
) {
3077 "%s: unsupported chunk length", __FUNCTION__
);
3081 /* verify inode count is consistent with extent length */
3082 if ((count
>> mp
->m_sb
.sb_inopblog
) != length
) {
3084 "%s: inconsistent inode count and chunk length",
3090 * Inode buffers can be freed. Do not replay the inode initialisation as
3091 * we could be overwriting something written after this inode buffer was
3094 * XXX: we need to iterate all buffers and only init those that are not
3095 * cancelled. I think that a more fine grained factoring of
3096 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3099 if (xlog_check_buffer_cancelled(log
,
3100 XFS_AGB_TO_DADDR(mp
, agno
, agbno
), length
, 0))
3103 xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, count
, agno
, agbno
, length
,
3104 be32_to_cpu(icl
->icl_gen
));
3109 xlog_recover_buffer_ra_pass2(
3111 struct xlog_recover_item
*item
)
3113 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3114 struct xfs_mount
*mp
= log
->l_mp
;
3116 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3117 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3121 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3122 buf_f
->blf_len
, NULL
);
3126 xlog_recover_inode_ra_pass2(
3128 struct xlog_recover_item
*item
)
3130 struct xfs_inode_log_format ilf_buf
;
3131 struct xfs_inode_log_format
*ilfp
;
3132 struct xfs_mount
*mp
= log
->l_mp
;
3135 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3136 ilfp
= item
->ri_buf
[0].i_addr
;
3139 memset(ilfp
, 0, sizeof(*ilfp
));
3140 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3145 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3148 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3149 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3153 xlog_recover_dquot_ra_pass2(
3155 struct xlog_recover_item
*item
)
3157 struct xfs_mount
*mp
= log
->l_mp
;
3158 struct xfs_disk_dquot
*recddq
;
3159 struct xfs_dq_logformat
*dq_f
;
3163 if (mp
->m_qflags
== 0)
3166 recddq
= item
->ri_buf
[1].i_addr
;
3169 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
3172 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3174 if (log
->l_quotaoffs_flag
& type
)
3177 dq_f
= item
->ri_buf
[0].i_addr
;
3179 ASSERT(dq_f
->qlf_len
== 1);
3181 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3182 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), NULL
);
3186 xlog_recover_ra_pass2(
3188 struct xlog_recover_item
*item
)
3190 switch (ITEM_TYPE(item
)) {
3192 xlog_recover_buffer_ra_pass2(log
, item
);
3195 xlog_recover_inode_ra_pass2(log
, item
);
3198 xlog_recover_dquot_ra_pass2(log
, item
);
3202 case XFS_LI_QUOTAOFF
:
3209 xlog_recover_commit_pass1(
3211 struct xlog_recover
*trans
,
3212 struct xlog_recover_item
*item
)
3214 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3216 switch (ITEM_TYPE(item
)) {
3218 return xlog_recover_buffer_pass1(log
, item
);
3219 case XFS_LI_QUOTAOFF
:
3220 return xlog_recover_quotaoff_pass1(log
, item
);
3225 case XFS_LI_ICREATE
:
3226 /* nothing to do in pass 1 */
3229 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3230 __func__
, ITEM_TYPE(item
));
3237 xlog_recover_commit_pass2(
3239 struct xlog_recover
*trans
,
3240 struct list_head
*buffer_list
,
3241 struct xlog_recover_item
*item
)
3243 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3245 switch (ITEM_TYPE(item
)) {
3247 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
3250 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
3253 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3255 return xlog_recover_efd_pass2(log
, item
);
3257 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
3259 case XFS_LI_ICREATE
:
3260 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3261 case XFS_LI_QUOTAOFF
:
3262 /* nothing to do in pass2 */
3265 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3266 __func__
, ITEM_TYPE(item
));
3273 xlog_recover_items_pass2(
3275 struct xlog_recover
*trans
,
3276 struct list_head
*buffer_list
,
3277 struct list_head
*item_list
)
3279 struct xlog_recover_item
*item
;
3282 list_for_each_entry(item
, item_list
, ri_list
) {
3283 error
= xlog_recover_commit_pass2(log
, trans
,
3293 * Perform the transaction.
3295 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3296 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3299 xlog_recover_commit_trans(
3301 struct xlog_recover
*trans
,
3306 int items_queued
= 0;
3307 struct xlog_recover_item
*item
;
3308 struct xlog_recover_item
*next
;
3309 LIST_HEAD (buffer_list
);
3310 LIST_HEAD (ra_list
);
3311 LIST_HEAD (done_list
);
3313 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3315 hlist_del(&trans
->r_list
);
3317 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3321 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
3323 case XLOG_RECOVER_PASS1
:
3324 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3326 case XLOG_RECOVER_PASS2
:
3327 xlog_recover_ra_pass2(log
, item
);
3328 list_move_tail(&item
->ri_list
, &ra_list
);
3330 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
3331 error
= xlog_recover_items_pass2(log
, trans
,
3332 &buffer_list
, &ra_list
);
3333 list_splice_tail_init(&ra_list
, &done_list
);
3347 if (!list_empty(&ra_list
)) {
3349 error
= xlog_recover_items_pass2(log
, trans
,
3350 &buffer_list
, &ra_list
);
3351 list_splice_tail_init(&ra_list
, &done_list
);
3354 if (!list_empty(&done_list
))
3355 list_splice_init(&done_list
, &trans
->r_itemq
);
3357 error2
= xfs_buf_delwri_submit(&buffer_list
);
3358 return error
? error
: error2
;
3362 xlog_recover_add_item(
3363 struct list_head
*head
)
3365 xlog_recover_item_t
*item
;
3367 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
3368 INIT_LIST_HEAD(&item
->ri_list
);
3369 list_add_tail(&item
->ri_list
, head
);
3373 xlog_recover_add_to_cont_trans(
3375 struct xlog_recover
*trans
,
3379 xlog_recover_item_t
*item
;
3380 char *ptr
, *old_ptr
;
3383 if (list_empty(&trans
->r_itemq
)) {
3384 /* finish copying rest of trans header */
3385 xlog_recover_add_item(&trans
->r_itemq
);
3386 ptr
= (char *)&trans
->r_theader
+
3387 sizeof(xfs_trans_header_t
) - len
;
3388 memcpy(ptr
, dp
, len
);
3391 /* take the tail entry */
3392 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3394 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
3395 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
3397 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
3398 memcpy(&ptr
[old_len
], dp
, len
);
3399 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
3400 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
3401 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
3406 * The next region to add is the start of a new region. It could be
3407 * a whole region or it could be the first part of a new region. Because
3408 * of this, the assumption here is that the type and size fields of all
3409 * format structures fit into the first 32 bits of the structure.
3411 * This works because all regions must be 32 bit aligned. Therefore, we
3412 * either have both fields or we have neither field. In the case we have
3413 * neither field, the data part of the region is zero length. We only have
3414 * a log_op_header and can throw away the header since a new one will appear
3415 * later. If we have at least 4 bytes, then we can determine how many regions
3416 * will appear in the current log item.
3419 xlog_recover_add_to_trans(
3421 struct xlog_recover
*trans
,
3425 xfs_inode_log_format_t
*in_f
; /* any will do */
3426 xlog_recover_item_t
*item
;
3431 if (list_empty(&trans
->r_itemq
)) {
3432 /* we need to catch log corruptions here */
3433 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
3434 xfs_warn(log
->l_mp
, "%s: bad header magic number",
3439 if (len
== sizeof(xfs_trans_header_t
))
3440 xlog_recover_add_item(&trans
->r_itemq
);
3441 memcpy(&trans
->r_theader
, dp
, len
);
3445 ptr
= kmem_alloc(len
, KM_SLEEP
);
3446 memcpy(ptr
, dp
, len
);
3447 in_f
= (xfs_inode_log_format_t
*)ptr
;
3449 /* take the tail entry */
3450 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
3451 if (item
->ri_total
!= 0 &&
3452 item
->ri_total
== item
->ri_cnt
) {
3453 /* tail item is in use, get a new one */
3454 xlog_recover_add_item(&trans
->r_itemq
);
3455 item
= list_entry(trans
->r_itemq
.prev
,
3456 xlog_recover_item_t
, ri_list
);
3459 if (item
->ri_total
== 0) { /* first region to be added */
3460 if (in_f
->ilf_size
== 0 ||
3461 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
3463 "bad number of regions (%d) in inode log format",
3470 item
->ri_total
= in_f
->ilf_size
;
3472 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
3475 ASSERT(item
->ri_total
> item
->ri_cnt
);
3476 /* Description region is ri_buf[0] */
3477 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
3478 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
3480 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
3485 * Free up any resources allocated by the transaction
3487 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3490 xlog_recover_free_trans(
3491 struct xlog_recover
*trans
)
3493 xlog_recover_item_t
*item
, *n
;
3496 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3497 /* Free the regions in the item. */
3498 list_del(&item
->ri_list
);
3499 for (i
= 0; i
< item
->ri_cnt
; i
++)
3500 kmem_free(item
->ri_buf
[i
].i_addr
);
3501 /* Free the item itself */
3502 kmem_free(item
->ri_buf
);
3505 /* Free the transaction recover structure */
3510 * On error or completion, trans is freed.
3513 xlog_recovery_process_trans(
3515 struct xlog_recover
*trans
,
3522 bool freeit
= false;
3524 /* mask off ophdr transaction container flags */
3525 flags
&= ~XLOG_END_TRANS
;
3526 if (flags
& XLOG_WAS_CONT_TRANS
)
3527 flags
&= ~XLOG_CONTINUE_TRANS
;
3530 * Callees must not free the trans structure. We'll decide if we need to
3531 * free it or not based on the operation being done and it's result.
3534 /* expected flag values */
3536 case XLOG_CONTINUE_TRANS
:
3537 error
= xlog_recover_add_to_trans(log
, trans
, dp
, len
);
3539 case XLOG_WAS_CONT_TRANS
:
3540 error
= xlog_recover_add_to_cont_trans(log
, trans
, dp
, len
);
3542 case XLOG_COMMIT_TRANS
:
3543 error
= xlog_recover_commit_trans(log
, trans
, pass
);
3544 /* success or fail, we are now done with this transaction. */
3548 /* unexpected flag values */
3549 case XLOG_UNMOUNT_TRANS
:
3550 /* just skip trans */
3551 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3554 case XLOG_START_TRANS
:
3556 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x", __func__
, flags
);
3561 if (error
|| freeit
)
3562 xlog_recover_free_trans(trans
);
3567 * Lookup the transaction recovery structure associated with the ID in the
3568 * current ophdr. If the transaction doesn't exist and the start flag is set in
3569 * the ophdr, then allocate a new transaction for future ID matches to find.
3570 * Either way, return what we found during the lookup - an existing transaction
3573 STATIC
struct xlog_recover
*
3574 xlog_recover_ophdr_to_trans(
3575 struct hlist_head rhash
[],
3576 struct xlog_rec_header
*rhead
,
3577 struct xlog_op_header
*ohead
)
3579 struct xlog_recover
*trans
;
3581 struct hlist_head
*rhp
;
3583 tid
= be32_to_cpu(ohead
->oh_tid
);
3584 rhp
= &rhash
[XLOG_RHASH(tid
)];
3585 hlist_for_each_entry(trans
, rhp
, r_list
) {
3586 if (trans
->r_log_tid
== tid
)
3591 * skip over non-start transaction headers - we could be
3592 * processing slack space before the next transaction starts
3594 if (!(ohead
->oh_flags
& XLOG_START_TRANS
))
3597 ASSERT(be32_to_cpu(ohead
->oh_len
) == 0);
3600 * This is a new transaction so allocate a new recovery container to
3601 * hold the recovery ops that will follow.
3603 trans
= kmem_zalloc(sizeof(struct xlog_recover
), KM_SLEEP
);
3604 trans
->r_log_tid
= tid
;
3605 trans
->r_lsn
= be64_to_cpu(rhead
->h_lsn
);
3606 INIT_LIST_HEAD(&trans
->r_itemq
);
3607 INIT_HLIST_NODE(&trans
->r_list
);
3608 hlist_add_head(&trans
->r_list
, rhp
);
3611 * Nothing more to do for this ophdr. Items to be added to this new
3612 * transaction will be in subsequent ophdr containers.
3618 xlog_recover_process_ophdr(
3620 struct hlist_head rhash
[],
3621 struct xlog_rec_header
*rhead
,
3622 struct xlog_op_header
*ohead
,
3627 struct xlog_recover
*trans
;
3630 /* Do we understand who wrote this op? */
3631 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
3632 ohead
->oh_clientid
!= XFS_LOG
) {
3633 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
3634 __func__
, ohead
->oh_clientid
);
3640 * Check the ophdr contains all the data it is supposed to contain.
3642 len
= be32_to_cpu(ohead
->oh_len
);
3643 if (dp
+ len
> end
) {
3644 xfs_warn(log
->l_mp
, "%s: bad length 0x%x", __func__
, len
);
3649 trans
= xlog_recover_ophdr_to_trans(rhash
, rhead
, ohead
);
3651 /* nothing to do, so skip over this ophdr */
3655 return xlog_recovery_process_trans(log
, trans
, dp
, len
,
3656 ohead
->oh_flags
, pass
);
3660 * There are two valid states of the r_state field. 0 indicates that the
3661 * transaction structure is in a normal state. We have either seen the
3662 * start of the transaction or the last operation we added was not a partial
3663 * operation. If the last operation we added to the transaction was a
3664 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3666 * NOTE: skip LRs with 0 data length.
3669 xlog_recover_process_data(
3671 struct hlist_head rhash
[],
3672 struct xlog_rec_header
*rhead
,
3676 struct xlog_op_header
*ohead
;
3681 end
= dp
+ be32_to_cpu(rhead
->h_len
);
3682 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
3684 /* check the log format matches our own - else we can't recover */
3685 if (xlog_header_check_recover(log
->l_mp
, rhead
))
3688 while ((dp
< end
) && num_logops
) {
3690 ohead
= (struct xlog_op_header
*)dp
;
3691 dp
+= sizeof(*ohead
);
3694 /* errors will abort recovery */
3695 error
= xlog_recover_process_ophdr(log
, rhash
, rhead
, ohead
,
3700 dp
+= be32_to_cpu(ohead
->oh_len
);
3707 * Process an extent free intent item that was recovered from
3708 * the log. We need to free the extents that it describes.
3711 xlog_recover_process_efi(
3713 xfs_efi_log_item_t
*efip
)
3715 xfs_efd_log_item_t
*efdp
;
3720 xfs_fsblock_t startblock_fsb
;
3722 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
3725 * First check the validity of the extents described by the
3726 * EFI. If any are bad, then assume that all are bad and
3727 * just toss the EFI.
3729 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3730 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3731 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3732 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3733 if ((startblock_fsb
== 0) ||
3734 (extp
->ext_len
== 0) ||
3735 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3736 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3738 * This will pull the EFI from the AIL and
3739 * free the memory associated with it.
3741 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3742 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3747 tp
= xfs_trans_alloc(mp
, 0);
3748 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
3751 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3753 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3754 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3755 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3758 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3762 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3763 error
= xfs_trans_commit(tp
);
3767 xfs_trans_cancel(tp
);
3772 * When this is called, all of the EFIs which did not have
3773 * corresponding EFDs should be in the AIL. What we do now
3774 * is free the extents associated with each one.
3776 * Since we process the EFIs in normal transactions, they
3777 * will be removed at some point after the commit. This prevents
3778 * us from just walking down the list processing each one.
3779 * We'll use a flag in the EFI to skip those that we've already
3780 * processed and use the AIL iteration mechanism's generation
3781 * count to try to speed this up at least a bit.
3783 * When we start, we know that the EFIs are the only things in
3784 * the AIL. As we process them, however, other items are added
3785 * to the AIL. Since everything added to the AIL must come after
3786 * everything already in the AIL, we stop processing as soon as
3787 * we see something other than an EFI in the AIL.
3790 xlog_recover_process_efis(
3793 xfs_log_item_t
*lip
;
3794 xfs_efi_log_item_t
*efip
;
3796 struct xfs_ail_cursor cur
;
3797 struct xfs_ail
*ailp
;
3800 spin_lock(&ailp
->xa_lock
);
3801 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3802 while (lip
!= NULL
) {
3804 * We're done when we see something other than an EFI.
3805 * There should be no EFIs left in the AIL now.
3807 if (lip
->li_type
!= XFS_LI_EFI
) {
3809 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3810 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3816 * Skip EFIs that we've already processed.
3818 efip
= (xfs_efi_log_item_t
*)lip
;
3819 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3820 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3824 spin_unlock(&ailp
->xa_lock
);
3825 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3826 spin_lock(&ailp
->xa_lock
);
3829 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3832 xfs_trans_ail_cursor_done(&cur
);
3833 spin_unlock(&ailp
->xa_lock
);
3838 * This routine performs a transaction to null out a bad inode pointer
3839 * in an agi unlinked inode hash bucket.
3842 xlog_recover_clear_agi_bucket(
3844 xfs_agnumber_t agno
,
3853 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3854 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
3858 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3862 agi
= XFS_BUF_TO_AGI(agibp
);
3863 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3864 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3865 (sizeof(xfs_agino_t
) * bucket
);
3866 xfs_trans_log_buf(tp
, agibp
, offset
,
3867 (offset
+ sizeof(xfs_agino_t
) - 1));
3869 error
= xfs_trans_commit(tp
);
3875 xfs_trans_cancel(tp
);
3877 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3882 xlog_recover_process_one_iunlink(
3883 struct xfs_mount
*mp
,
3884 xfs_agnumber_t agno
,
3888 struct xfs_buf
*ibp
;
3889 struct xfs_dinode
*dip
;
3890 struct xfs_inode
*ip
;
3894 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3895 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3900 * Get the on disk inode to find the next inode in the bucket.
3902 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3906 ASSERT(ip
->i_d
.di_nlink
== 0);
3907 ASSERT(ip
->i_d
.di_mode
!= 0);
3909 /* setup for the next pass */
3910 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3914 * Prevent any DMAPI event from being sent when the reference on
3915 * the inode is dropped.
3917 ip
->i_d
.di_dmevmask
= 0;
3926 * We can't read in the inode this bucket points to, or this inode
3927 * is messed up. Just ditch this bucket of inodes. We will lose
3928 * some inodes and space, but at least we won't hang.
3930 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3931 * clear the inode pointer in the bucket.
3933 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3938 * xlog_iunlink_recover
3940 * This is called during recovery to process any inodes which
3941 * we unlinked but not freed when the system crashed. These
3942 * inodes will be on the lists in the AGI blocks. What we do
3943 * here is scan all the AGIs and fully truncate and free any
3944 * inodes found on the lists. Each inode is removed from the
3945 * lists when it has been fully truncated and is freed. The
3946 * freeing of the inode and its removal from the list must be
3950 xlog_recover_process_iunlinks(
3954 xfs_agnumber_t agno
;
3965 * Prevent any DMAPI event from being sent while in this function.
3967 mp_dmevmask
= mp
->m_dmevmask
;
3970 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3972 * Find the agi for this ag.
3974 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3977 * AGI is b0rked. Don't process it.
3979 * We should probably mark the filesystem as corrupt
3980 * after we've recovered all the ag's we can....
3985 * Unlock the buffer so that it can be acquired in the normal
3986 * course of the transaction to truncate and free each inode.
3987 * Because we are not racing with anyone else here for the AGI
3988 * buffer, we don't even need to hold it locked to read the
3989 * initial unlinked bucket entries out of the buffer. We keep
3990 * buffer reference though, so that it stays pinned in memory
3991 * while we need the buffer.
3993 agi
= XFS_BUF_TO_AGI(agibp
);
3994 xfs_buf_unlock(agibp
);
3996 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3997 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3998 while (agino
!= NULLAGINO
) {
3999 agino
= xlog_recover_process_one_iunlink(mp
,
4000 agno
, agino
, bucket
);
4003 xfs_buf_rele(agibp
);
4006 mp
->m_dmevmask
= mp_dmevmask
;
4010 * Upack the log buffer data and crc check it. If the check fails, issue a
4011 * warning if and only if the CRC in the header is non-zero. This makes the
4012 * check an advisory warning, and the zero CRC check will prevent failure
4013 * warnings from being emitted when upgrading the kernel from one that does not
4014 * add CRCs by default.
4016 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
4017 * corruption failure
4020 xlog_unpack_data_crc(
4021 struct xlog_rec_header
*rhead
,
4027 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
4028 if (crc
!= rhead
->h_crc
) {
4029 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
4030 xfs_alert(log
->l_mp
,
4031 "log record CRC mismatch: found 0x%x, expected 0x%x.",
4032 le32_to_cpu(rhead
->h_crc
),
4034 xfs_hex_dump(dp
, 32);
4038 * If we've detected a log record corruption, then we can't
4039 * recover past this point. Abort recovery if we are enforcing
4040 * CRC protection by punting an error back up the stack.
4042 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
4043 return -EFSCORRUPTED
;
4051 struct xlog_rec_header
*rhead
,
4058 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
4062 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
4063 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
4064 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
4068 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4069 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
4070 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
4071 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4072 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
4073 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
4082 xlog_valid_rec_header(
4084 struct xlog_rec_header
*rhead
,
4089 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
4090 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4091 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4092 return -EFSCORRUPTED
;
4095 (!rhead
->h_version
||
4096 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
4097 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
4098 __func__
, be32_to_cpu(rhead
->h_version
));
4102 /* LR body must have data or it wouldn't have been written */
4103 hlen
= be32_to_cpu(rhead
->h_len
);
4104 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
4105 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4106 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4107 return -EFSCORRUPTED
;
4109 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
4110 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4111 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4112 return -EFSCORRUPTED
;
4118 * Read the log from tail to head and process the log records found.
4119 * Handle the two cases where the tail and head are in the same cycle
4120 * and where the active portion of the log wraps around the end of
4121 * the physical log separately. The pass parameter is passed through
4122 * to the routines called to process the data and is not looked at
4126 xlog_do_recovery_pass(
4128 xfs_daddr_t head_blk
,
4129 xfs_daddr_t tail_blk
,
4132 xlog_rec_header_t
*rhead
;
4135 xfs_buf_t
*hbp
, *dbp
;
4136 int error
= 0, h_size
;
4137 int bblks
, split_bblks
;
4138 int hblks
, split_hblks
, wrapped_hblks
;
4139 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
4141 ASSERT(head_blk
!= tail_blk
);
4144 * Read the header of the tail block and get the iclog buffer size from
4145 * h_size. Use this to tell how many sectors make up the log header.
4147 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4149 * When using variable length iclogs, read first sector of
4150 * iclog header and extract the header size from it. Get a
4151 * new hbp that is the correct size.
4153 hbp
= xlog_get_bp(log
, 1);
4157 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
4161 rhead
= (xlog_rec_header_t
*)offset
;
4162 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
4165 h_size
= be32_to_cpu(rhead
->h_size
);
4166 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
4167 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
4168 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
4169 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
4172 hbp
= xlog_get_bp(log
, hblks
);
4177 ASSERT(log
->l_sectBBsize
== 1);
4179 hbp
= xlog_get_bp(log
, 1);
4180 h_size
= XLOG_BIG_RECORD_BSIZE
;
4185 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
4191 memset(rhash
, 0, sizeof(rhash
));
4193 if (tail_blk
> head_blk
) {
4195 * Perform recovery around the end of the physical log.
4196 * When the head is not on the same cycle number as the tail,
4197 * we can't do a sequential recovery.
4199 while (blk_no
< log
->l_logBBsize
) {
4201 * Check for header wrapping around physical end-of-log
4203 offset
= hbp
->b_addr
;
4206 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
4207 /* Read header in one read */
4208 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
4213 /* This LR is split across physical log end */
4214 if (blk_no
!= log
->l_logBBsize
) {
4215 /* some data before physical log end */
4216 ASSERT(blk_no
<= INT_MAX
);
4217 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
4218 ASSERT(split_hblks
> 0);
4219 error
= xlog_bread(log
, blk_no
,
4227 * Note: this black magic still works with
4228 * large sector sizes (non-512) only because:
4229 * - we increased the buffer size originally
4230 * by 1 sector giving us enough extra space
4231 * for the second read;
4232 * - the log start is guaranteed to be sector
4234 * - we read the log end (LR header start)
4235 * _first_, then the log start (LR header end)
4236 * - order is important.
4238 wrapped_hblks
= hblks
- split_hblks
;
4239 error
= xlog_bread_offset(log
, 0,
4241 offset
+ BBTOB(split_hblks
));
4245 rhead
= (xlog_rec_header_t
*)offset
;
4246 error
= xlog_valid_rec_header(log
, rhead
,
4247 split_hblks
? blk_no
: 0);
4251 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4254 /* Read in data for log record */
4255 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
4256 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
4261 /* This log record is split across the
4262 * physical end of log */
4263 offset
= dbp
->b_addr
;
4265 if (blk_no
!= log
->l_logBBsize
) {
4266 /* some data is before the physical
4268 ASSERT(!wrapped_hblks
);
4269 ASSERT(blk_no
<= INT_MAX
);
4271 log
->l_logBBsize
- (int)blk_no
;
4272 ASSERT(split_bblks
> 0);
4273 error
= xlog_bread(log
, blk_no
,
4281 * Note: this black magic still works with
4282 * large sector sizes (non-512) only because:
4283 * - we increased the buffer size originally
4284 * by 1 sector giving us enough extra space
4285 * for the second read;
4286 * - the log start is guaranteed to be sector
4288 * - we read the log end (LR header start)
4289 * _first_, then the log start (LR header end)
4290 * - order is important.
4292 error
= xlog_bread_offset(log
, 0,
4293 bblks
- split_bblks
, dbp
,
4294 offset
+ BBTOB(split_bblks
));
4299 error
= xlog_unpack_data(rhead
, offset
, log
);
4303 error
= xlog_recover_process_data(log
, rhash
,
4304 rhead
, offset
, pass
);
4310 ASSERT(blk_no
>= log
->l_logBBsize
);
4311 blk_no
-= log
->l_logBBsize
;
4314 /* read first part of physical log */
4315 while (blk_no
< head_blk
) {
4316 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4320 rhead
= (xlog_rec_header_t
*)offset
;
4321 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4325 /* blocks in data section */
4326 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4327 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
4332 error
= xlog_unpack_data(rhead
, offset
, log
);
4336 error
= xlog_recover_process_data(log
, rhash
,
4337 rhead
, offset
, pass
);
4340 blk_no
+= bblks
+ hblks
;
4351 * Do the recovery of the log. We actually do this in two phases.
4352 * The two passes are necessary in order to implement the function
4353 * of cancelling a record written into the log. The first pass
4354 * determines those things which have been cancelled, and the
4355 * second pass replays log items normally except for those which
4356 * have been cancelled. The handling of the replay and cancellations
4357 * takes place in the log item type specific routines.
4359 * The table of items which have cancel records in the log is allocated
4360 * and freed at this level, since only here do we know when all of
4361 * the log recovery has been completed.
4364 xlog_do_log_recovery(
4366 xfs_daddr_t head_blk
,
4367 xfs_daddr_t tail_blk
)
4371 ASSERT(head_blk
!= tail_blk
);
4374 * First do a pass to find all of the cancelled buf log items.
4375 * Store them in the buf_cancel_table for use in the second pass.
4377 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4378 sizeof(struct list_head
),
4380 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4381 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4383 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4384 XLOG_RECOVER_PASS1
);
4386 kmem_free(log
->l_buf_cancel_table
);
4387 log
->l_buf_cancel_table
= NULL
;
4391 * Then do a second pass to actually recover the items in the log.
4392 * When it is complete free the table of buf cancel items.
4394 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4395 XLOG_RECOVER_PASS2
);
4400 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4401 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4405 kmem_free(log
->l_buf_cancel_table
);
4406 log
->l_buf_cancel_table
= NULL
;
4412 * Do the actual recovery
4417 xfs_daddr_t head_blk
,
4418 xfs_daddr_t tail_blk
)
4425 * First replay the images in the log.
4427 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4432 * If IO errors happened during recovery, bail out.
4434 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4439 * We now update the tail_lsn since much of the recovery has completed
4440 * and there may be space available to use. If there were no extent
4441 * or iunlinks, we can free up the entire log and set the tail_lsn to
4442 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4443 * lsn of the last known good LR on disk. If there are extent frees
4444 * or iunlinks they will have some entries in the AIL; so we look at
4445 * the AIL to determine how to set the tail_lsn.
4447 xlog_assign_tail_lsn(log
->l_mp
);
4450 * Now that we've finished replaying all buffer and inode
4451 * updates, re-read in the superblock and reverify it.
4453 bp
= xfs_getsb(log
->l_mp
, 0);
4455 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
4457 XFS_BUF_UNASYNC(bp
);
4458 bp
->b_ops
= &xfs_sb_buf_ops
;
4460 error
= xfs_buf_submit_wait(bp
);
4462 if (!XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4463 xfs_buf_ioerror_alert(bp
, __func__
);
4470 /* Convert superblock from on-disk format */
4471 sbp
= &log
->l_mp
->m_sb
;
4472 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4473 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4474 ASSERT(xfs_sb_good_version(sbp
));
4475 xfs_reinit_percpu_counters(log
->l_mp
);
4480 xlog_recover_check_summary(log
);
4482 /* Normal transactions can now occur */
4483 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4488 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4490 * Return error or zero.
4496 xfs_daddr_t head_blk
, tail_blk
;
4499 /* find the tail of the log */
4500 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
4503 if (tail_blk
!= head_blk
) {
4504 /* There used to be a comment here:
4506 * disallow recovery on read-only mounts. note -- mount
4507 * checks for ENOSPC and turns it into an intelligent
4509 * ...but this is no longer true. Now, unless you specify
4510 * NORECOVERY (in which case this function would never be
4511 * called), we just go ahead and recover. We do this all
4512 * under the vfs layer, so we can get away with it unless
4513 * the device itself is read-only, in which case we fail.
4515 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
4520 * Version 5 superblock log feature mask validation. We know the
4521 * log is dirty so check if there are any unknown log features
4522 * in what we need to recover. If there are unknown features
4523 * (e.g. unsupported transactions, then simply reject the
4524 * attempt at recovery before touching anything.
4526 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
4527 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
4528 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
4530 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4531 "The log can not be fully and/or safely recovered by this kernel.\n"
4532 "Please recover the log on a kernel that supports the unknown features.",
4533 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
4534 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
4539 * Delay log recovery if the debug hook is set. This is debug
4540 * instrumention to coordinate simulation of I/O failures with
4543 if (xfs_globals
.log_recovery_delay
) {
4544 xfs_notice(log
->l_mp
,
4545 "Delaying log recovery for %d seconds.",
4546 xfs_globals
.log_recovery_delay
);
4547 msleep(xfs_globals
.log_recovery_delay
* 1000);
4550 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
4551 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4554 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
4555 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
4561 * In the first part of recovery we replay inodes and buffers and build
4562 * up the list of extent free items which need to be processed. Here
4563 * we process the extent free items and clean up the on disk unlinked
4564 * inode lists. This is separated from the first part of recovery so
4565 * that the root and real-time bitmap inodes can be read in from disk in
4566 * between the two stages. This is necessary so that we can free space
4567 * in the real-time portion of the file system.
4570 xlog_recover_finish(
4574 * Now we're ready to do the transactions needed for the
4575 * rest of recovery. Start with completing all the extent
4576 * free intent records and then process the unlinked inode
4577 * lists. At this point, we essentially run in normal mode
4578 * except that we're still performing recovery actions
4579 * rather than accepting new requests.
4581 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
4583 error
= xlog_recover_process_efis(log
);
4585 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
4589 * Sync the log to get all the EFIs out of the AIL.
4590 * This isn't absolutely necessary, but it helps in
4591 * case the unlink transactions would have problems
4592 * pushing the EFIs out of the way.
4594 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
4596 xlog_recover_process_iunlinks(log
);
4598 xlog_recover_check_summary(log
);
4600 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
4601 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4603 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4605 xfs_info(log
->l_mp
, "Ending clean mount");
4613 * Read all of the agf and agi counters and check that they
4614 * are consistent with the superblock counters.
4617 xlog_recover_check_summary(
4624 xfs_agnumber_t agno
;
4625 __uint64_t freeblks
;
4635 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4636 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4638 xfs_alert(mp
, "%s agf read failed agno %d error %d",
4639 __func__
, agno
, error
);
4641 agfp
= XFS_BUF_TO_AGF(agfbp
);
4642 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4643 be32_to_cpu(agfp
->agf_flcount
);
4644 xfs_buf_relse(agfbp
);
4647 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4649 xfs_alert(mp
, "%s agi read failed agno %d error %d",
4650 __func__
, agno
, error
);
4652 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4654 itotal
+= be32_to_cpu(agi
->agi_count
);
4655 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4656 xfs_buf_relse(agibp
);