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"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
33 #include "xfs_log_priv.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_alloc.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_quota.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_bmap_btree.h"
45 #include "xfs_dinode.h"
46 #include "xfs_error.h"
49 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
56 xlog_clear_stale_blocks(
61 xlog_recover_check_summary(
64 #define xlog_recover_check_summary(log)
68 * This structure is used during recovery to record the buf log items which
69 * have been canceled and should not be replayed.
71 struct xfs_buf_cancel
{
75 struct list_head bc_list
;
79 * Sector aligned buffer routines for buffer create/read/write/access
83 * Verify the given count of basic blocks is valid number of blocks
84 * to specify for an operation involving the given XFS log buffer.
85 * Returns nonzero if the count is valid, 0 otherwise.
89 xlog_buf_bbcount_valid(
93 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
97 * Allocate a buffer to hold log data. The buffer needs to be able
98 * to map to a range of nbblks basic blocks at any valid (basic
99 * block) offset within the log.
108 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
109 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
111 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
116 * We do log I/O in units of log sectors (a power-of-2
117 * multiple of the basic block size), so we round up the
118 * requested size to accommodate the basic blocks required
119 * for complete log sectors.
121 * In addition, the buffer may be used for a non-sector-
122 * aligned block offset, in which case an I/O of the
123 * requested size could extend beyond the end of the
124 * buffer. If the requested size is only 1 basic block it
125 * will never straddle a sector boundary, so this won't be
126 * an issue. Nor will this be a problem if the log I/O is
127 * done in basic blocks (sector size 1). But otherwise we
128 * extend the buffer by one extra log sector to ensure
129 * there's space to accommodate this possibility.
131 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
132 nbblks
+= log
->l_sectBBsize
;
133 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
135 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
149 * Return the address of the start of the given block number's data
150 * in a log buffer. The buffer covers a log sector-aligned region.
159 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
161 ASSERT(offset
+ nbblks
<= bp
->b_length
);
162 return bp
->b_addr
+ BBTOB(offset
);
167 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
178 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
179 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
181 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
185 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
186 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
189 ASSERT(nbblks
<= bp
->b_length
);
191 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
193 bp
->b_io_length
= nbblks
;
196 xfsbdstrat(log
->l_mp
, bp
);
197 error
= xfs_buf_iowait(bp
);
199 xfs_buf_ioerror_alert(bp
, __func__
);
213 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
217 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
222 * Read at an offset into the buffer. Returns with the buffer in it's original
223 * state regardless of the result of the read.
228 xfs_daddr_t blk_no
, /* block to read from */
229 int nbblks
, /* blocks to read */
233 xfs_caddr_t orig_offset
= bp
->b_addr
;
234 int orig_len
= BBTOB(bp
->b_length
);
237 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
241 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
243 /* must reset buffer pointer even on error */
244 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
251 * Write out the buffer at the given block for the given number of blocks.
252 * The buffer is kept locked across the write and is returned locked.
253 * This can only be used for synchronous log writes.
264 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
265 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
267 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
271 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
272 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
275 ASSERT(nbblks
<= bp
->b_length
);
277 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
278 XFS_BUF_ZEROFLAGS(bp
);
281 bp
->b_io_length
= nbblks
;
284 error
= xfs_bwrite(bp
);
286 xfs_buf_ioerror_alert(bp
, __func__
);
293 * dump debug superblock and log record information
296 xlog_header_check_dump(
298 xlog_rec_header_t
*head
)
300 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d",
301 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
302 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
303 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
306 #define xlog_header_check_dump(mp, head)
310 * check log record header for recovery
313 xlog_header_check_recover(
315 xlog_rec_header_t
*head
)
317 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
320 * IRIX doesn't write the h_fmt field and leaves it zeroed
321 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
322 * a dirty log created in IRIX.
324 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
326 "dirty log written in incompatible format - can't recover");
327 xlog_header_check_dump(mp
, head
);
328 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
329 XFS_ERRLEVEL_HIGH
, mp
);
330 return XFS_ERROR(EFSCORRUPTED
);
331 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
333 "dirty log entry has mismatched uuid - can't recover");
334 xlog_header_check_dump(mp
, head
);
335 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
336 XFS_ERRLEVEL_HIGH
, mp
);
337 return XFS_ERROR(EFSCORRUPTED
);
343 * read the head block of the log and check the header
346 xlog_header_check_mount(
348 xlog_rec_header_t
*head
)
350 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
352 if (uuid_is_nil(&head
->h_fs_uuid
)) {
354 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
355 * h_fs_uuid is nil, we assume this log was last mounted
356 * by IRIX and continue.
358 xfs_warn(mp
, "nil uuid in log - IRIX style log");
359 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
360 xfs_warn(mp
, "log has mismatched uuid - can't recover");
361 xlog_header_check_dump(mp
, head
);
362 XFS_ERROR_REPORT("xlog_header_check_mount",
363 XFS_ERRLEVEL_HIGH
, mp
);
364 return XFS_ERROR(EFSCORRUPTED
);
375 * We're not going to bother about retrying
376 * this during recovery. One strike!
378 xfs_buf_ioerror_alert(bp
, __func__
);
379 xfs_force_shutdown(bp
->b_target
->bt_mount
,
380 SHUTDOWN_META_IO_ERROR
);
383 xfs_buf_ioend(bp
, 0);
387 * This routine finds (to an approximation) the first block in the physical
388 * log which contains the given cycle. It uses a binary search algorithm.
389 * Note that the algorithm can not be perfect because the disk will not
390 * necessarily be perfect.
393 xlog_find_cycle_start(
396 xfs_daddr_t first_blk
,
397 xfs_daddr_t
*last_blk
,
407 mid_blk
= BLK_AVG(first_blk
, end_blk
);
408 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
409 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
412 mid_cycle
= xlog_get_cycle(offset
);
413 if (mid_cycle
== cycle
)
414 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
416 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
417 mid_blk
= BLK_AVG(first_blk
, end_blk
);
419 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
420 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
428 * Check that a range of blocks does not contain stop_on_cycle_no.
429 * Fill in *new_blk with the block offset where such a block is
430 * found, or with -1 (an invalid block number) if there is no such
431 * block in the range. The scan needs to occur from front to back
432 * and the pointer into the region must be updated since a later
433 * routine will need to perform another test.
436 xlog_find_verify_cycle(
438 xfs_daddr_t start_blk
,
440 uint stop_on_cycle_no
,
441 xfs_daddr_t
*new_blk
)
447 xfs_caddr_t buf
= NULL
;
451 * Greedily allocate a buffer big enough to handle the full
452 * range of basic blocks we'll be examining. If that fails,
453 * try a smaller size. We need to be able to read at least
454 * a log sector, or we're out of luck.
456 bufblks
= 1 << ffs(nbblks
);
457 while (bufblks
> log
->l_logBBsize
)
459 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
461 if (bufblks
< log
->l_sectBBsize
)
465 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
468 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
470 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
474 for (j
= 0; j
< bcount
; j
++) {
475 cycle
= xlog_get_cycle(buf
);
476 if (cycle
== stop_on_cycle_no
) {
493 * Potentially backup over partial log record write.
495 * In the typical case, last_blk is the number of the block directly after
496 * a good log record. Therefore, we subtract one to get the block number
497 * of the last block in the given buffer. extra_bblks contains the number
498 * of blocks we would have read on a previous read. This happens when the
499 * last log record is split over the end of the physical log.
501 * extra_bblks is the number of blocks potentially verified on a previous
502 * call to this routine.
505 xlog_find_verify_log_record(
507 xfs_daddr_t start_blk
,
508 xfs_daddr_t
*last_blk
,
513 xfs_caddr_t offset
= NULL
;
514 xlog_rec_header_t
*head
= NULL
;
517 int num_blks
= *last_blk
- start_blk
;
520 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
522 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
523 if (!(bp
= xlog_get_bp(log
, 1)))
527 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
530 offset
+= ((num_blks
- 1) << BBSHIFT
);
533 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
535 /* valid log record not found */
537 "Log inconsistent (didn't find previous header)");
539 error
= XFS_ERROR(EIO
);
544 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
549 head
= (xlog_rec_header_t
*)offset
;
551 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
559 * We hit the beginning of the physical log & still no header. Return
560 * to caller. If caller can handle a return of -1, then this routine
561 * will be called again for the end of the physical log.
569 * We have the final block of the good log (the first block
570 * of the log record _before_ the head. So we check the uuid.
572 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
576 * We may have found a log record header before we expected one.
577 * last_blk will be the 1st block # with a given cycle #. We may end
578 * up reading an entire log record. In this case, we don't want to
579 * reset last_blk. Only when last_blk points in the middle of a log
580 * record do we update last_blk.
582 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
583 uint h_size
= be32_to_cpu(head
->h_size
);
585 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
586 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
592 if (*last_blk
- i
+ extra_bblks
!=
593 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
602 * Head is defined to be the point of the log where the next log write
603 * could go. This means that incomplete LR writes at the end are
604 * eliminated when calculating the head. We aren't guaranteed that previous
605 * LR have complete transactions. We only know that a cycle number of
606 * current cycle number -1 won't be present in the log if we start writing
607 * from our current block number.
609 * last_blk contains the block number of the first block with a given
612 * Return: zero if normal, non-zero if error.
617 xfs_daddr_t
*return_head_blk
)
621 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
623 uint first_half_cycle
, last_half_cycle
;
625 int error
, log_bbnum
= log
->l_logBBsize
;
627 /* Is the end of the log device zeroed? */
628 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
629 *return_head_blk
= first_blk
;
631 /* Is the whole lot zeroed? */
633 /* Linux XFS shouldn't generate totally zeroed logs -
634 * mkfs etc write a dummy unmount record to a fresh
635 * log so we can store the uuid in there
637 xfs_warn(log
->l_mp
, "totally zeroed log");
642 xfs_warn(log
->l_mp
, "empty log check failed");
646 first_blk
= 0; /* get cycle # of 1st block */
647 bp
= xlog_get_bp(log
, 1);
651 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
655 first_half_cycle
= xlog_get_cycle(offset
);
657 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
658 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
662 last_half_cycle
= xlog_get_cycle(offset
);
663 ASSERT(last_half_cycle
!= 0);
666 * If the 1st half cycle number is equal to the last half cycle number,
667 * then the entire log is stamped with the same cycle number. In this
668 * case, head_blk can't be set to zero (which makes sense). The below
669 * math doesn't work out properly with head_blk equal to zero. Instead,
670 * we set it to log_bbnum which is an invalid block number, but this
671 * value makes the math correct. If head_blk doesn't changed through
672 * all the tests below, *head_blk is set to zero at the very end rather
673 * than log_bbnum. In a sense, log_bbnum and zero are the same block
674 * in a circular file.
676 if (first_half_cycle
== last_half_cycle
) {
678 * In this case we believe that the entire log should have
679 * cycle number last_half_cycle. We need to scan backwards
680 * from the end verifying that there are no holes still
681 * containing last_half_cycle - 1. If we find such a hole,
682 * then the start of that hole will be the new head. The
683 * simple case looks like
684 * x | x ... | x - 1 | x
685 * Another case that fits this picture would be
686 * x | x + 1 | x ... | x
687 * In this case the head really is somewhere at the end of the
688 * log, as one of the latest writes at the beginning was
691 * x | x + 1 | x ... | x - 1 | x
692 * This is really the combination of the above two cases, and
693 * the head has to end up at the start of the x-1 hole at the
696 * In the 256k log case, we will read from the beginning to the
697 * end of the log and search for cycle numbers equal to x-1.
698 * We don't worry about the x+1 blocks that we encounter,
699 * because we know that they cannot be the head since the log
702 head_blk
= log_bbnum
;
703 stop_on_cycle
= last_half_cycle
- 1;
706 * In this case we want to find the first block with cycle
707 * number matching last_half_cycle. We expect the log to be
709 * x + 1 ... | x ... | x
710 * The first block with cycle number x (last_half_cycle) will
711 * be where the new head belongs. First we do a binary search
712 * for the first occurrence of last_half_cycle. The binary
713 * search may not be totally accurate, so then we scan back
714 * from there looking for occurrences of last_half_cycle before
715 * us. If that backwards scan wraps around the beginning of
716 * the log, then we look for occurrences of last_half_cycle - 1
717 * at the end of the log. The cases we're looking for look
719 * v binary search stopped here
720 * x + 1 ... | x | x + 1 | x ... | x
721 * ^ but we want to locate this spot
723 * <---------> less than scan distance
724 * x + 1 ... | x ... | x - 1 | x
725 * ^ we want to locate this spot
727 stop_on_cycle
= last_half_cycle
;
728 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
729 &head_blk
, last_half_cycle
)))
734 * Now validate the answer. Scan back some number of maximum possible
735 * blocks and make sure each one has the expected cycle number. The
736 * maximum is determined by the total possible amount of buffering
737 * in the in-core log. The following number can be made tighter if
738 * we actually look at the block size of the filesystem.
740 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
741 if (head_blk
>= num_scan_bblks
) {
743 * We are guaranteed that the entire check can be performed
746 start_blk
= head_blk
- num_scan_bblks
;
747 if ((error
= xlog_find_verify_cycle(log
,
748 start_blk
, num_scan_bblks
,
749 stop_on_cycle
, &new_blk
)))
753 } else { /* need to read 2 parts of log */
755 * We are going to scan backwards in the log in two parts.
756 * First we scan the physical end of the log. In this part
757 * of the log, we are looking for blocks with cycle number
758 * last_half_cycle - 1.
759 * If we find one, then we know that the log starts there, as
760 * we've found a hole that didn't get written in going around
761 * the end of the physical log. The simple case for this is
762 * x + 1 ... | x ... | x - 1 | x
763 * <---------> less than scan distance
764 * If all of the blocks at the end of the log have cycle number
765 * last_half_cycle, then we check the blocks at the start of
766 * the log looking for occurrences of last_half_cycle. If we
767 * find one, then our current estimate for the location of the
768 * first occurrence of last_half_cycle is wrong and we move
769 * back to the hole we've found. This case looks like
770 * x + 1 ... | x | x + 1 | x ...
771 * ^ binary search stopped here
772 * Another case we need to handle that only occurs in 256k
774 * x + 1 ... | x ... | x+1 | x ...
775 * ^ binary search stops here
776 * In a 256k log, the scan at the end of the log will see the
777 * x + 1 blocks. We need to skip past those since that is
778 * certainly not the head of the log. By searching for
779 * last_half_cycle-1 we accomplish that.
781 ASSERT(head_blk
<= INT_MAX
&&
782 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
783 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
784 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
785 num_scan_bblks
- (int)head_blk
,
786 (stop_on_cycle
- 1), &new_blk
)))
794 * Scan beginning of log now. The last part of the physical
795 * log is good. This scan needs to verify that it doesn't find
796 * the last_half_cycle.
799 ASSERT(head_blk
<= INT_MAX
);
800 if ((error
= xlog_find_verify_cycle(log
,
801 start_blk
, (int)head_blk
,
802 stop_on_cycle
, &new_blk
)))
810 * Now we need to make sure head_blk is not pointing to a block in
811 * the middle of a log record.
813 num_scan_bblks
= XLOG_REC_SHIFT(log
);
814 if (head_blk
>= num_scan_bblks
) {
815 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
817 /* start ptr at last block ptr before head_blk */
818 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
819 &head_blk
, 0)) == -1) {
820 error
= XFS_ERROR(EIO
);
826 ASSERT(head_blk
<= INT_MAX
);
827 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
828 &head_blk
, 0)) == -1) {
829 /* We hit the beginning of the log during our search */
830 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
832 ASSERT(start_blk
<= INT_MAX
&&
833 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
834 ASSERT(head_blk
<= INT_MAX
);
835 if ((error
= xlog_find_verify_log_record(log
,
837 (int)head_blk
)) == -1) {
838 error
= XFS_ERROR(EIO
);
842 if (new_blk
!= log_bbnum
)
849 if (head_blk
== log_bbnum
)
850 *return_head_blk
= 0;
852 *return_head_blk
= head_blk
;
854 * When returning here, we have a good block number. Bad block
855 * means that during a previous crash, we didn't have a clean break
856 * from cycle number N to cycle number N-1. In this case, we need
857 * to find the first block with cycle number N-1.
865 xfs_warn(log
->l_mp
, "failed to find log head");
870 * Find the sync block number or the tail of the log.
872 * This will be the block number of the last record to have its
873 * associated buffers synced to disk. Every log record header has
874 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
875 * to get a sync block number. The only concern is to figure out which
876 * log record header to believe.
878 * The following algorithm uses the log record header with the largest
879 * lsn. The entire log record does not need to be valid. We only care
880 * that the header is valid.
882 * We could speed up search by using current head_blk buffer, but it is not
888 xfs_daddr_t
*head_blk
,
889 xfs_daddr_t
*tail_blk
)
891 xlog_rec_header_t
*rhead
;
892 xlog_op_header_t
*op_head
;
893 xfs_caddr_t offset
= NULL
;
896 xfs_daddr_t umount_data_blk
;
897 xfs_daddr_t after_umount_blk
;
904 * Find previous log record
906 if ((error
= xlog_find_head(log
, head_blk
)))
909 bp
= xlog_get_bp(log
, 1);
912 if (*head_blk
== 0) { /* special case */
913 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
917 if (xlog_get_cycle(offset
) == 0) {
919 /* leave all other log inited values alone */
925 * Search backwards looking for log record header block
927 ASSERT(*head_blk
< INT_MAX
);
928 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
929 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
933 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
939 * If we haven't found the log record header block, start looking
940 * again from the end of the physical log. XXXmiken: There should be
941 * a check here to make sure we didn't search more than N blocks in
945 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
946 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
950 if (*(__be32
*)offset
==
951 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
958 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
961 return XFS_ERROR(EIO
);
964 /* find blk_no of tail of log */
965 rhead
= (xlog_rec_header_t
*)offset
;
966 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
969 * Reset log values according to the state of the log when we
970 * crashed. In the case where head_blk == 0, we bump curr_cycle
971 * one because the next write starts a new cycle rather than
972 * continuing the cycle of the last good log record. At this
973 * point we have guaranteed that all partial log records have been
974 * accounted for. Therefore, we know that the last good log record
975 * written was complete and ended exactly on the end boundary
976 * of the physical log.
978 log
->l_prev_block
= i
;
979 log
->l_curr_block
= (int)*head_blk
;
980 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
983 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
984 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
985 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
986 BBTOB(log
->l_curr_block
));
987 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
988 BBTOB(log
->l_curr_block
));
991 * Look for unmount record. If we find it, then we know there
992 * was a clean unmount. Since 'i' could be the last block in
993 * the physical log, we convert to a log block before comparing
996 * Save the current tail lsn to use to pass to
997 * xlog_clear_stale_blocks() below. We won't want to clear the
998 * unmount record if there is one, so we pass the lsn of the
999 * unmount record rather than the block after it.
1001 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1002 int h_size
= be32_to_cpu(rhead
->h_size
);
1003 int h_version
= be32_to_cpu(rhead
->h_version
);
1005 if ((h_version
& XLOG_VERSION_2
) &&
1006 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1007 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1008 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1016 after_umount_blk
= (i
+ hblks
+ (int)
1017 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1018 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1019 if (*head_blk
== after_umount_blk
&&
1020 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1021 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1022 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1026 op_head
= (xlog_op_header_t
*)offset
;
1027 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1029 * Set tail and last sync so that newly written
1030 * log records will point recovery to after the
1031 * current unmount record.
1033 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1034 log
->l_curr_cycle
, after_umount_blk
);
1035 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1036 log
->l_curr_cycle
, after_umount_blk
);
1037 *tail_blk
= after_umount_blk
;
1040 * Note that the unmount was clean. If the unmount
1041 * was not clean, we need to know this to rebuild the
1042 * superblock counters from the perag headers if we
1043 * have a filesystem using non-persistent counters.
1045 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1050 * Make sure that there are no blocks in front of the head
1051 * with the same cycle number as the head. This can happen
1052 * because we allow multiple outstanding log writes concurrently,
1053 * and the later writes might make it out before earlier ones.
1055 * We use the lsn from before modifying it so that we'll never
1056 * overwrite the unmount record after a clean unmount.
1058 * Do this only if we are going to recover the filesystem
1060 * NOTE: This used to say "if (!readonly)"
1061 * However on Linux, we can & do recover a read-only filesystem.
1062 * We only skip recovery if NORECOVERY is specified on mount,
1063 * in which case we would not be here.
1065 * But... if the -device- itself is readonly, just skip this.
1066 * We can't recover this device anyway, so it won't matter.
1068 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1069 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1075 xfs_warn(log
->l_mp
, "failed to locate log tail");
1080 * Is the log zeroed at all?
1082 * The last binary search should be changed to perform an X block read
1083 * once X becomes small enough. You can then search linearly through
1084 * the X blocks. This will cut down on the number of reads we need to do.
1086 * If the log is partially zeroed, this routine will pass back the blkno
1087 * of the first block with cycle number 0. It won't have a complete LR
1091 * 0 => the log is completely written to
1092 * -1 => use *blk_no as the first block of the log
1093 * >0 => error has occurred
1098 xfs_daddr_t
*blk_no
)
1102 uint first_cycle
, last_cycle
;
1103 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1104 xfs_daddr_t num_scan_bblks
;
1105 int error
, log_bbnum
= log
->l_logBBsize
;
1109 /* check totally zeroed log */
1110 bp
= xlog_get_bp(log
, 1);
1113 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1117 first_cycle
= xlog_get_cycle(offset
);
1118 if (first_cycle
== 0) { /* completely zeroed log */
1124 /* check partially zeroed log */
1125 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1129 last_cycle
= xlog_get_cycle(offset
);
1130 if (last_cycle
!= 0) { /* log completely written to */
1133 } else if (first_cycle
!= 1) {
1135 * If the cycle of the last block is zero, the cycle of
1136 * the first block must be 1. If it's not, maybe we're
1137 * not looking at a log... Bail out.
1140 "Log inconsistent or not a log (last==0, first!=1)");
1141 error
= XFS_ERROR(EINVAL
);
1145 /* we have a partially zeroed log */
1146 last_blk
= log_bbnum
-1;
1147 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1151 * Validate the answer. Because there is no way to guarantee that
1152 * the entire log is made up of log records which are the same size,
1153 * we scan over the defined maximum blocks. At this point, the maximum
1154 * is not chosen to mean anything special. XXXmiken
1156 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1157 ASSERT(num_scan_bblks
<= INT_MAX
);
1159 if (last_blk
< num_scan_bblks
)
1160 num_scan_bblks
= last_blk
;
1161 start_blk
= last_blk
- num_scan_bblks
;
1164 * We search for any instances of cycle number 0 that occur before
1165 * our current estimate of the head. What we're trying to detect is
1166 * 1 ... | 0 | 1 | 0...
1167 * ^ binary search ends here
1169 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1170 (int)num_scan_bblks
, 0, &new_blk
)))
1176 * Potentially backup over partial log record write. We don't need
1177 * to search the end of the log because we know it is zero.
1179 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1180 &last_blk
, 0)) == -1) {
1181 error
= XFS_ERROR(EIO
);
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 XFS_ERROR(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 XFS_ERROR(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 ******************************************************************************
1444 STATIC xlog_recover_t
*
1445 xlog_recover_find_tid(
1446 struct hlist_head
*head
,
1449 xlog_recover_t
*trans
;
1451 hlist_for_each_entry(trans
, head
, r_list
) {
1452 if (trans
->r_log_tid
== tid
)
1459 xlog_recover_new_tid(
1460 struct hlist_head
*head
,
1464 xlog_recover_t
*trans
;
1466 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1467 trans
->r_log_tid
= tid
;
1469 INIT_LIST_HEAD(&trans
->r_itemq
);
1471 INIT_HLIST_NODE(&trans
->r_list
);
1472 hlist_add_head(&trans
->r_list
, head
);
1476 xlog_recover_add_item(
1477 struct list_head
*head
)
1479 xlog_recover_item_t
*item
;
1481 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1482 INIT_LIST_HEAD(&item
->ri_list
);
1483 list_add_tail(&item
->ri_list
, head
);
1487 xlog_recover_add_to_cont_trans(
1489 struct xlog_recover
*trans
,
1493 xlog_recover_item_t
*item
;
1494 xfs_caddr_t ptr
, old_ptr
;
1497 if (list_empty(&trans
->r_itemq
)) {
1498 /* finish copying rest of trans header */
1499 xlog_recover_add_item(&trans
->r_itemq
);
1500 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1501 sizeof(xfs_trans_header_t
) - len
;
1502 memcpy(ptr
, dp
, len
); /* d, s, l */
1505 /* take the tail entry */
1506 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1508 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1509 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1511 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
1512 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1513 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1514 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1515 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1520 * The next region to add is the start of a new region. It could be
1521 * a whole region or it could be the first part of a new region. Because
1522 * of this, the assumption here is that the type and size fields of all
1523 * format structures fit into the first 32 bits of the structure.
1525 * This works because all regions must be 32 bit aligned. Therefore, we
1526 * either have both fields or we have neither field. In the case we have
1527 * neither field, the data part of the region is zero length. We only have
1528 * a log_op_header and can throw away the header since a new one will appear
1529 * later. If we have at least 4 bytes, then we can determine how many regions
1530 * will appear in the current log item.
1533 xlog_recover_add_to_trans(
1535 struct xlog_recover
*trans
,
1539 xfs_inode_log_format_t
*in_f
; /* any will do */
1540 xlog_recover_item_t
*item
;
1545 if (list_empty(&trans
->r_itemq
)) {
1546 /* we need to catch log corruptions here */
1547 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1548 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1551 return XFS_ERROR(EIO
);
1553 if (len
== sizeof(xfs_trans_header_t
))
1554 xlog_recover_add_item(&trans
->r_itemq
);
1555 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1559 ptr
= kmem_alloc(len
, KM_SLEEP
);
1560 memcpy(ptr
, dp
, len
);
1561 in_f
= (xfs_inode_log_format_t
*)ptr
;
1563 /* take the tail entry */
1564 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1565 if (item
->ri_total
!= 0 &&
1566 item
->ri_total
== item
->ri_cnt
) {
1567 /* tail item is in use, get a new one */
1568 xlog_recover_add_item(&trans
->r_itemq
);
1569 item
= list_entry(trans
->r_itemq
.prev
,
1570 xlog_recover_item_t
, ri_list
);
1573 if (item
->ri_total
== 0) { /* first region to be added */
1574 if (in_f
->ilf_size
== 0 ||
1575 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1577 "bad number of regions (%d) in inode log format",
1581 return XFS_ERROR(EIO
);
1584 item
->ri_total
= in_f
->ilf_size
;
1586 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1589 ASSERT(item
->ri_total
> item
->ri_cnt
);
1590 /* Description region is ri_buf[0] */
1591 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1592 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1594 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1599 * Sort the log items in the transaction.
1601 * The ordering constraints are defined by the inode allocation and unlink
1602 * behaviour. The rules are:
1604 * 1. Every item is only logged once in a given transaction. Hence it
1605 * represents the last logged state of the item. Hence ordering is
1606 * dependent on the order in which operations need to be performed so
1607 * required initial conditions are always met.
1609 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1610 * there's nothing to replay from them so we can simply cull them
1611 * from the transaction. However, we can't do that until after we've
1612 * replayed all the other items because they may be dependent on the
1613 * cancelled buffer and replaying the cancelled buffer can remove it
1614 * form the cancelled buffer table. Hence they have tobe done last.
1616 * 3. Inode allocation buffers must be replayed before inode items that
1617 * read the buffer and replay changes into it. For filesystems using the
1618 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1619 * treated the same as inode allocation buffers as they create and
1620 * initialise the buffers directly.
1622 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1623 * This ensures that inodes are completely flushed to the inode buffer
1624 * in a "free" state before we remove the unlinked inode list pointer.
1626 * Hence the ordering needs to be inode allocation buffers first, inode items
1627 * second, inode unlink buffers third and cancelled buffers last.
1629 * But there's a problem with that - we can't tell an inode allocation buffer
1630 * apart from a regular buffer, so we can't separate them. We can, however,
1631 * tell an inode unlink buffer from the others, and so we can separate them out
1632 * from all the other buffers and move them to last.
1634 * Hence, 4 lists, in order from head to tail:
1635 * - buffer_list for all buffers except cancelled/inode unlink buffers
1636 * - item_list for all non-buffer items
1637 * - inode_buffer_list for inode unlink buffers
1638 * - cancel_list for the cancelled buffers
1640 * Note that we add objects to the tail of the lists so that first-to-last
1641 * ordering is preserved within the lists. Adding objects to the head of the
1642 * list means when we traverse from the head we walk them in last-to-first
1643 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1644 * but for all other items there may be specific ordering that we need to
1648 xlog_recover_reorder_trans(
1650 struct xlog_recover
*trans
,
1653 xlog_recover_item_t
*item
, *n
;
1655 LIST_HEAD(sort_list
);
1656 LIST_HEAD(cancel_list
);
1657 LIST_HEAD(buffer_list
);
1658 LIST_HEAD(inode_buffer_list
);
1659 LIST_HEAD(inode_list
);
1661 list_splice_init(&trans
->r_itemq
, &sort_list
);
1662 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1663 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1665 switch (ITEM_TYPE(item
)) {
1666 case XFS_LI_ICREATE
:
1667 list_move_tail(&item
->ri_list
, &buffer_list
);
1670 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1671 trace_xfs_log_recover_item_reorder_head(log
,
1673 list_move(&item
->ri_list
, &cancel_list
);
1676 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1677 list_move(&item
->ri_list
, &inode_buffer_list
);
1680 list_move_tail(&item
->ri_list
, &buffer_list
);
1684 case XFS_LI_QUOTAOFF
:
1687 trace_xfs_log_recover_item_reorder_tail(log
,
1689 list_move_tail(&item
->ri_list
, &inode_list
);
1693 "%s: unrecognized type of log operation",
1697 * return the remaining items back to the transaction
1698 * item list so they can be freed in caller.
1700 if (!list_empty(&sort_list
))
1701 list_splice_init(&sort_list
, &trans
->r_itemq
);
1702 error
= XFS_ERROR(EIO
);
1707 ASSERT(list_empty(&sort_list
));
1708 if (!list_empty(&buffer_list
))
1709 list_splice(&buffer_list
, &trans
->r_itemq
);
1710 if (!list_empty(&inode_list
))
1711 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1712 if (!list_empty(&inode_buffer_list
))
1713 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1714 if (!list_empty(&cancel_list
))
1715 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1720 * Build up the table of buf cancel records so that we don't replay
1721 * cancelled data in the second pass. For buffer records that are
1722 * not cancel records, there is nothing to do here so we just return.
1724 * If we get a cancel record which is already in the table, this indicates
1725 * that the buffer was cancelled multiple times. In order to ensure
1726 * that during pass 2 we keep the record in the table until we reach its
1727 * last occurrence in the log, we keep a reference count in the cancel
1728 * record in the table to tell us how many times we expect to see this
1729 * record during the second pass.
1732 xlog_recover_buffer_pass1(
1734 struct xlog_recover_item
*item
)
1736 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1737 struct list_head
*bucket
;
1738 struct xfs_buf_cancel
*bcp
;
1741 * If this isn't a cancel buffer item, then just return.
1743 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1744 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1749 * Insert an xfs_buf_cancel record into the hash table of them.
1750 * If there is already an identical record, bump its reference count.
1752 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1753 list_for_each_entry(bcp
, bucket
, bc_list
) {
1754 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1755 bcp
->bc_len
== buf_f
->blf_len
) {
1757 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1762 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1763 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1764 bcp
->bc_len
= buf_f
->blf_len
;
1765 bcp
->bc_refcount
= 1;
1766 list_add_tail(&bcp
->bc_list
, bucket
);
1768 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1773 * Check to see whether the buffer being recovered has a corresponding
1774 * entry in the buffer cancel record table. If it is, return the cancel
1775 * buffer structure to the caller.
1777 STATIC
struct xfs_buf_cancel
*
1778 xlog_peek_buffer_cancelled(
1784 struct list_head
*bucket
;
1785 struct xfs_buf_cancel
*bcp
;
1787 if (!log
->l_buf_cancel_table
) {
1788 /* empty table means no cancelled buffers in the log */
1789 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1793 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1794 list_for_each_entry(bcp
, bucket
, bc_list
) {
1795 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1800 * We didn't find a corresponding entry in the table, so return 0 so
1801 * that the buffer is NOT cancelled.
1803 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1808 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1809 * otherwise return 0. If the buffer is actually a buffer cancel item
1810 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1811 * table and remove it from the table if this is the last reference.
1813 * We remove the cancel record from the table when we encounter its last
1814 * occurrence in the log so that if the same buffer is re-used again after its
1815 * last cancellation we actually replay the changes made at that point.
1818 xlog_check_buffer_cancelled(
1824 struct xfs_buf_cancel
*bcp
;
1826 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
1831 * We've go a match, so return 1 so that the recovery of this buffer
1832 * is cancelled. If this buffer is actually a buffer cancel log
1833 * item, then decrement the refcount on the one in the table and
1834 * remove it if this is the last reference.
1836 if (flags
& XFS_BLF_CANCEL
) {
1837 if (--bcp
->bc_refcount
== 0) {
1838 list_del(&bcp
->bc_list
);
1846 * Perform recovery for a buffer full of inodes. In these buffers, the only
1847 * data which should be recovered is that which corresponds to the
1848 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1849 * data for the inodes is always logged through the inodes themselves rather
1850 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1852 * The only time when buffers full of inodes are fully recovered is when the
1853 * buffer is full of newly allocated inodes. In this case the buffer will
1854 * not be marked as an inode buffer and so will be sent to
1855 * xlog_recover_do_reg_buffer() below during recovery.
1858 xlog_recover_do_inode_buffer(
1859 struct xfs_mount
*mp
,
1860 xlog_recover_item_t
*item
,
1862 xfs_buf_log_format_t
*buf_f
)
1868 int reg_buf_offset
= 0;
1869 int reg_buf_bytes
= 0;
1870 int next_unlinked_offset
;
1872 xfs_agino_t
*logged_nextp
;
1873 xfs_agino_t
*buffer_nextp
;
1875 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1878 * Post recovery validation only works properly on CRC enabled
1881 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1882 bp
->b_ops
= &xfs_inode_buf_ops
;
1884 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1885 for (i
= 0; i
< inodes_per_buf
; i
++) {
1886 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1887 offsetof(xfs_dinode_t
, di_next_unlinked
);
1889 while (next_unlinked_offset
>=
1890 (reg_buf_offset
+ reg_buf_bytes
)) {
1892 * The next di_next_unlinked field is beyond
1893 * the current logged region. Find the next
1894 * logged region that contains or is beyond
1895 * the current di_next_unlinked field.
1898 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1899 buf_f
->blf_map_size
, bit
);
1902 * If there are no more logged regions in the
1903 * buffer, then we're done.
1908 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1909 buf_f
->blf_map_size
, bit
);
1911 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1912 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1917 * If the current logged region starts after the current
1918 * di_next_unlinked field, then move on to the next
1919 * di_next_unlinked field.
1921 if (next_unlinked_offset
< reg_buf_offset
)
1924 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1925 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1926 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1927 BBTOB(bp
->b_io_length
));
1930 * The current logged region contains a copy of the
1931 * current di_next_unlinked field. Extract its value
1932 * and copy it to the buffer copy.
1934 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1935 next_unlinked_offset
- reg_buf_offset
;
1936 if (unlikely(*logged_nextp
== 0)) {
1938 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1939 "Trying to replay bad (0) inode di_next_unlinked field.",
1941 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1942 XFS_ERRLEVEL_LOW
, mp
);
1943 return XFS_ERROR(EFSCORRUPTED
);
1946 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1947 next_unlinked_offset
);
1948 *buffer_nextp
= *logged_nextp
;
1951 * If necessary, recalculate the CRC in the on-disk inode. We
1952 * have to leave the inode in a consistent state for whoever
1955 xfs_dinode_calc_crc(mp
, (struct xfs_dinode
*)
1956 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
1964 * V5 filesystems know the age of the buffer on disk being recovered. We can
1965 * have newer objects on disk than we are replaying, and so for these cases we
1966 * don't want to replay the current change as that will make the buffer contents
1967 * temporarily invalid on disk.
1969 * The magic number might not match the buffer type we are going to recover
1970 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1971 * extract the LSN of the existing object in the buffer based on it's current
1972 * magic number. If we don't recognise the magic number in the buffer, then
1973 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1974 * so can recover the buffer.
1976 * Note: we cannot rely solely on magic number matches to determine that the
1977 * buffer has a valid LSN - we also need to verify that it belongs to this
1978 * filesystem, so we need to extract the object's LSN and compare it to that
1979 * which we read from the superblock. If the UUIDs don't match, then we've got a
1980 * stale metadata block from an old filesystem instance that we need to recover
1984 xlog_recover_get_buf_lsn(
1985 struct xfs_mount
*mp
,
1991 void *blk
= bp
->b_addr
;
1995 /* v4 filesystems always recover immediately */
1996 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1997 goto recover_immediately
;
1999 magic32
= be32_to_cpu(*(__be32
*)blk
);
2001 case XFS_ABTB_CRC_MAGIC
:
2002 case XFS_ABTC_CRC_MAGIC
:
2003 case XFS_ABTB_MAGIC
:
2004 case XFS_ABTC_MAGIC
:
2005 case XFS_IBT_CRC_MAGIC
:
2006 case XFS_IBT_MAGIC
: {
2007 struct xfs_btree_block
*btb
= blk
;
2009 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
2010 uuid
= &btb
->bb_u
.s
.bb_uuid
;
2013 case XFS_BMAP_CRC_MAGIC
:
2014 case XFS_BMAP_MAGIC
: {
2015 struct xfs_btree_block
*btb
= blk
;
2017 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
2018 uuid
= &btb
->bb_u
.l
.bb_uuid
;
2022 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
2023 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
2025 case XFS_AGFL_MAGIC
:
2026 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
2027 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
2030 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
2031 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
2033 case XFS_SYMLINK_MAGIC
:
2034 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
2035 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
2037 case XFS_DIR3_BLOCK_MAGIC
:
2038 case XFS_DIR3_DATA_MAGIC
:
2039 case XFS_DIR3_FREE_MAGIC
:
2040 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
2041 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
2043 case XFS_ATTR3_RMT_MAGIC
:
2044 lsn
= be64_to_cpu(((struct xfs_attr3_rmt_hdr
*)blk
)->rm_lsn
);
2045 uuid
= &((struct xfs_attr3_rmt_hdr
*)blk
)->rm_uuid
;
2048 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
2049 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
2055 if (lsn
!= (xfs_lsn_t
)-1) {
2056 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2057 goto recover_immediately
;
2061 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
2063 case XFS_DIR3_LEAF1_MAGIC
:
2064 case XFS_DIR3_LEAFN_MAGIC
:
2065 case XFS_DA3_NODE_MAGIC
:
2066 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
2067 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
2073 if (lsn
!= (xfs_lsn_t
)-1) {
2074 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2075 goto recover_immediately
;
2080 * We do individual object checks on dquot and inode buffers as they
2081 * have their own individual LSN records. Also, we could have a stale
2082 * buffer here, so we have to at least recognise these buffer types.
2084 * A notd complexity here is inode unlinked list processing - it logs
2085 * the inode directly in the buffer, but we don't know which inodes have
2086 * been modified, and there is no global buffer LSN. Hence we need to
2087 * recover all inode buffer types immediately. This problem will be
2088 * fixed by logical logging of the unlinked list modifications.
2090 magic16
= be16_to_cpu(*(__be16
*)blk
);
2092 case XFS_DQUOT_MAGIC
:
2093 case XFS_DINODE_MAGIC
:
2094 goto recover_immediately
;
2099 /* unknown buffer contents, recover immediately */
2101 recover_immediately
:
2102 return (xfs_lsn_t
)-1;
2107 * Validate the recovered buffer is of the correct type and attach the
2108 * appropriate buffer operations to them for writeback. Magic numbers are in a
2110 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2111 * the first 32 bits of the buffer (most blocks),
2112 * inside a struct xfs_da_blkinfo at the start of the buffer.
2115 xlog_recover_validate_buf_type(
2116 struct xfs_mount
*mp
,
2118 xfs_buf_log_format_t
*buf_f
)
2120 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
2125 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
2126 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
2127 magicda
= be16_to_cpu(info
->magic
);
2128 switch (xfs_blft_from_flags(buf_f
)) {
2129 case XFS_BLFT_BTREE_BUF
:
2131 case XFS_ABTB_CRC_MAGIC
:
2132 case XFS_ABTC_CRC_MAGIC
:
2133 case XFS_ABTB_MAGIC
:
2134 case XFS_ABTC_MAGIC
:
2135 bp
->b_ops
= &xfs_allocbt_buf_ops
;
2137 case XFS_IBT_CRC_MAGIC
:
2139 bp
->b_ops
= &xfs_inobt_buf_ops
;
2141 case XFS_BMAP_CRC_MAGIC
:
2142 case XFS_BMAP_MAGIC
:
2143 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2146 xfs_warn(mp
, "Bad btree block magic!");
2151 case XFS_BLFT_AGF_BUF
:
2152 if (magic32
!= XFS_AGF_MAGIC
) {
2153 xfs_warn(mp
, "Bad AGF block magic!");
2157 bp
->b_ops
= &xfs_agf_buf_ops
;
2159 case XFS_BLFT_AGFL_BUF
:
2160 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2162 if (magic32
!= XFS_AGFL_MAGIC
) {
2163 xfs_warn(mp
, "Bad AGFL block magic!");
2167 bp
->b_ops
= &xfs_agfl_buf_ops
;
2169 case XFS_BLFT_AGI_BUF
:
2170 if (magic32
!= XFS_AGI_MAGIC
) {
2171 xfs_warn(mp
, "Bad AGI block magic!");
2175 bp
->b_ops
= &xfs_agi_buf_ops
;
2177 case XFS_BLFT_UDQUOT_BUF
:
2178 case XFS_BLFT_PDQUOT_BUF
:
2179 case XFS_BLFT_GDQUOT_BUF
:
2180 #ifdef CONFIG_XFS_QUOTA
2181 if (magic16
!= XFS_DQUOT_MAGIC
) {
2182 xfs_warn(mp
, "Bad DQUOT block magic!");
2186 bp
->b_ops
= &xfs_dquot_buf_ops
;
2189 "Trying to recover dquots without QUOTA support built in!");
2193 case XFS_BLFT_DINO_BUF
:
2195 * we get here with inode allocation buffers, not buffers that
2196 * track unlinked list changes.
2198 if (magic16
!= XFS_DINODE_MAGIC
) {
2199 xfs_warn(mp
, "Bad INODE block magic!");
2203 bp
->b_ops
= &xfs_inode_buf_ops
;
2205 case XFS_BLFT_SYMLINK_BUF
:
2206 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2207 xfs_warn(mp
, "Bad symlink block magic!");
2211 bp
->b_ops
= &xfs_symlink_buf_ops
;
2213 case XFS_BLFT_DIR_BLOCK_BUF
:
2214 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2215 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2216 xfs_warn(mp
, "Bad dir block magic!");
2220 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2222 case XFS_BLFT_DIR_DATA_BUF
:
2223 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2224 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2225 xfs_warn(mp
, "Bad dir data magic!");
2229 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2231 case XFS_BLFT_DIR_FREE_BUF
:
2232 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2233 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2234 xfs_warn(mp
, "Bad dir3 free magic!");
2238 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2240 case XFS_BLFT_DIR_LEAF1_BUF
:
2241 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2242 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2243 xfs_warn(mp
, "Bad dir leaf1 magic!");
2247 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2249 case XFS_BLFT_DIR_LEAFN_BUF
:
2250 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2251 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2252 xfs_warn(mp
, "Bad dir leafn magic!");
2256 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2258 case XFS_BLFT_DA_NODE_BUF
:
2259 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2260 magicda
!= XFS_DA3_NODE_MAGIC
) {
2261 xfs_warn(mp
, "Bad da node magic!");
2265 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2267 case XFS_BLFT_ATTR_LEAF_BUF
:
2268 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2269 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2270 xfs_warn(mp
, "Bad attr leaf magic!");
2274 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2276 case XFS_BLFT_ATTR_RMT_BUF
:
2277 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2279 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2280 xfs_warn(mp
, "Bad attr remote magic!");
2284 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2286 case XFS_BLFT_SB_BUF
:
2287 if (magic32
!= XFS_SB_MAGIC
) {
2288 xfs_warn(mp
, "Bad SB block magic!");
2292 bp
->b_ops
= &xfs_sb_buf_ops
;
2295 xfs_warn(mp
, "Unknown buffer type %d!",
2296 xfs_blft_from_flags(buf_f
));
2302 * Perform a 'normal' buffer recovery. Each logged region of the
2303 * buffer should be copied over the corresponding region in the
2304 * given buffer. The bitmap in the buf log format structure indicates
2305 * where to place the logged data.
2308 xlog_recover_do_reg_buffer(
2309 struct xfs_mount
*mp
,
2310 xlog_recover_item_t
*item
,
2312 xfs_buf_log_format_t
*buf_f
)
2319 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2322 i
= 1; /* 0 is the buf format structure */
2324 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2325 buf_f
->blf_map_size
, bit
);
2328 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2329 buf_f
->blf_map_size
, bit
);
2331 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2332 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2333 ASSERT(BBTOB(bp
->b_io_length
) >=
2334 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2337 * The dirty regions logged in the buffer, even though
2338 * contiguous, may span multiple chunks. This is because the
2339 * dirty region may span a physical page boundary in a buffer
2340 * and hence be split into two separate vectors for writing into
2341 * the log. Hence we need to trim nbits back to the length of
2342 * the current region being copied out of the log.
2344 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2345 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2348 * Do a sanity check if this is a dquot buffer. Just checking
2349 * the first dquot in the buffer should do. XXXThis is
2350 * probably a good thing to do for other buf types also.
2353 if (buf_f
->blf_flags
&
2354 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2355 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2357 "XFS: NULL dquot in %s.", __func__
);
2360 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2362 "XFS: dquot too small (%d) in %s.",
2363 item
->ri_buf
[i
].i_len
, __func__
);
2366 error
= xfs_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2367 -1, 0, XFS_QMOPT_DOWARN
,
2368 "dquot_buf_recover");
2373 memcpy(xfs_buf_offset(bp
,
2374 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2375 item
->ri_buf
[i
].i_addr
, /* source */
2376 nbits
<<XFS_BLF_SHIFT
); /* length */
2382 /* Shouldn't be any more regions */
2383 ASSERT(i
== item
->ri_total
);
2386 * We can only do post recovery validation on items on CRC enabled
2387 * fielsystems as we need to know when the buffer was written to be able
2388 * to determine if we should have replayed the item. If we replay old
2389 * metadata over a newer buffer, then it will enter a temporarily
2390 * inconsistent state resulting in verification failures. Hence for now
2391 * just avoid the verification stage for non-crc filesystems
2393 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2394 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2398 * Perform a dquot buffer recovery.
2399 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2400 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2401 * Else, treat it as a regular buffer and do recovery.
2404 xlog_recover_do_dquot_buffer(
2405 struct xfs_mount
*mp
,
2407 struct xlog_recover_item
*item
,
2409 struct xfs_buf_log_format
*buf_f
)
2413 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2416 * Filesystems are required to send in quota flags at mount time.
2418 if (mp
->m_qflags
== 0) {
2423 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2424 type
|= XFS_DQ_USER
;
2425 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2426 type
|= XFS_DQ_PROJ
;
2427 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2428 type
|= XFS_DQ_GROUP
;
2430 * This type of quotas was turned off, so ignore this buffer
2432 if (log
->l_quotaoffs_flag
& type
)
2435 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2439 * This routine replays a modification made to a buffer at runtime.
2440 * There are actually two types of buffer, regular and inode, which
2441 * are handled differently. Inode buffers are handled differently
2442 * in that we only recover a specific set of data from them, namely
2443 * the inode di_next_unlinked fields. This is because all other inode
2444 * data is actually logged via inode records and any data we replay
2445 * here which overlaps that may be stale.
2447 * When meta-data buffers are freed at run time we log a buffer item
2448 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2449 * of the buffer in the log should not be replayed at recovery time.
2450 * This is so that if the blocks covered by the buffer are reused for
2451 * file data before we crash we don't end up replaying old, freed
2452 * meta-data into a user's file.
2454 * To handle the cancellation of buffer log items, we make two passes
2455 * over the log during recovery. During the first we build a table of
2456 * those buffers which have been cancelled, and during the second we
2457 * only replay those buffers which do not have corresponding cancel
2458 * records in the table. See xlog_recover_buffer_pass[1,2] above
2459 * for more details on the implementation of the table of cancel records.
2462 xlog_recover_buffer_pass2(
2464 struct list_head
*buffer_list
,
2465 struct xlog_recover_item
*item
,
2466 xfs_lsn_t current_lsn
)
2468 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2469 xfs_mount_t
*mp
= log
->l_mp
;
2476 * In this pass we only want to recover all the buffers which have
2477 * not been cancelled and are not cancellation buffers themselves.
2479 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2480 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2481 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2485 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2488 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2489 buf_flags
|= XBF_UNMAPPED
;
2491 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2494 return XFS_ERROR(ENOMEM
);
2495 error
= bp
->b_error
;
2497 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2502 * recover the buffer only if we get an LSN from it and it's less than
2503 * the lsn of the transaction we are replaying.
2505 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2506 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0)
2509 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2510 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2511 } else if (buf_f
->blf_flags
&
2512 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2513 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2515 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2521 * Perform delayed write on the buffer. Asynchronous writes will be
2522 * slower when taking into account all the buffers to be flushed.
2524 * Also make sure that only inode buffers with good sizes stay in
2525 * the buffer cache. The kernel moves inodes in buffers of 1 block
2526 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2527 * buffers in the log can be a different size if the log was generated
2528 * by an older kernel using unclustered inode buffers or a newer kernel
2529 * running with a different inode cluster size. Regardless, if the
2530 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2531 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2532 * the buffer out of the buffer cache so that the buffer won't
2533 * overlap with future reads of those inodes.
2535 if (XFS_DINODE_MAGIC
==
2536 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2537 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2538 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2540 error
= xfs_bwrite(bp
);
2542 ASSERT(bp
->b_target
->bt_mount
== mp
);
2543 bp
->b_iodone
= xlog_recover_iodone
;
2544 xfs_buf_delwri_queue(bp
, buffer_list
);
2553 * Inode fork owner changes
2555 * If we have been told that we have to reparent the inode fork, it's because an
2556 * extent swap operation on a CRC enabled filesystem has been done and we are
2557 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2560 * The complexity here is that we don't have an inode context to work with, so
2561 * after we've replayed the inode we need to instantiate one. This is where the
2564 * We are in the middle of log recovery, so we can't run transactions. That
2565 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2566 * that will result in the corresponding iput() running the inode through
2567 * xfs_inactive(). If we've just replayed an inode core that changes the link
2568 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2569 * transactions (bad!).
2571 * So, to avoid this, we instantiate an inode directly from the inode core we've
2572 * just recovered. We have the buffer still locked, and all we really need to
2573 * instantiate is the inode core and the forks being modified. We can do this
2574 * manually, then run the inode btree owner change, and then tear down the
2575 * xfs_inode without having to run any transactions at all.
2577 * Also, because we don't have a transaction context available here but need to
2578 * gather all the buffers we modify for writeback so we pass the buffer_list
2579 * instead for the operation to use.
2583 xfs_recover_inode_owner_change(
2584 struct xfs_mount
*mp
,
2585 struct xfs_dinode
*dip
,
2586 struct xfs_inode_log_format
*in_f
,
2587 struct list_head
*buffer_list
)
2589 struct xfs_inode
*ip
;
2592 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2594 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2598 /* instantiate the inode */
2599 xfs_dinode_from_disk(&ip
->i_d
, dip
);
2600 ASSERT(ip
->i_d
.di_version
>= 3);
2602 error
= xfs_iformat_fork(ip
, dip
);
2607 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2608 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2609 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2610 ip
->i_ino
, buffer_list
);
2615 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2616 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2617 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2618 ip
->i_ino
, buffer_list
);
2629 xlog_recover_inode_pass2(
2631 struct list_head
*buffer_list
,
2632 struct xlog_recover_item
*item
,
2633 xfs_lsn_t current_lsn
)
2635 xfs_inode_log_format_t
*in_f
;
2636 xfs_mount_t
*mp
= log
->l_mp
;
2645 xfs_icdinode_t
*dicp
;
2649 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2650 in_f
= item
->ri_buf
[0].i_addr
;
2652 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2654 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2660 * Inode buffers can be freed, look out for it,
2661 * and do not replay the inode.
2663 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2664 in_f
->ilf_len
, 0)) {
2666 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2669 trace_xfs_log_recover_inode_recover(log
, in_f
);
2671 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2672 &xfs_inode_buf_ops
);
2677 error
= bp
->b_error
;
2679 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2682 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2683 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2686 * Make sure the place we're flushing out to really looks
2689 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2691 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2692 __func__
, dip
, bp
, in_f
->ilf_ino
);
2693 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2694 XFS_ERRLEVEL_LOW
, mp
);
2695 error
= EFSCORRUPTED
;
2698 dicp
= item
->ri_buf
[1].i_addr
;
2699 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2701 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2702 __func__
, item
, in_f
->ilf_ino
);
2703 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2704 XFS_ERRLEVEL_LOW
, mp
);
2705 error
= EFSCORRUPTED
;
2710 * If the inode has an LSN in it, recover the inode only if it's less
2711 * than the lsn of the transaction we are replaying. Note: we still
2712 * need to replay an owner change even though the inode is more recent
2713 * than the transaction as there is no guarantee that all the btree
2714 * blocks are more recent than this transaction, too.
2716 if (dip
->di_version
>= 3) {
2717 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
2719 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2720 trace_xfs_log_recover_inode_skip(log
, in_f
);
2722 goto out_owner_change
;
2727 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2728 * are transactional and if ordering is necessary we can determine that
2729 * more accurately by the LSN field in the V3 inode core. Don't trust
2730 * the inode versions we might be changing them here - use the
2731 * superblock flag to determine whether we need to look at di_flushiter
2732 * to skip replay when the on disk inode is newer than the log one
2734 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2735 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2737 * Deal with the wrap case, DI_MAX_FLUSH is less
2738 * than smaller numbers
2740 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2741 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2744 trace_xfs_log_recover_inode_skip(log
, in_f
);
2750 /* Take the opportunity to reset the flush iteration count */
2751 dicp
->di_flushiter
= 0;
2753 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2754 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2755 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2756 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2757 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2759 "%s: Bad regular inode log record, rec ptr 0x%p, "
2760 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2761 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2762 error
= EFSCORRUPTED
;
2765 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2766 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2767 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2768 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2769 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2770 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2772 "%s: Bad dir inode log record, rec ptr 0x%p, "
2773 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2774 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2775 error
= EFSCORRUPTED
;
2779 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2780 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2781 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2783 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2784 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2785 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2786 dicp
->di_nextents
+ dicp
->di_anextents
,
2788 error
= EFSCORRUPTED
;
2791 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2792 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2793 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2795 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2796 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2797 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2798 error
= EFSCORRUPTED
;
2801 isize
= xfs_icdinode_size(dicp
->di_version
);
2802 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2803 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2804 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2806 "%s: Bad inode log record length %d, rec ptr 0x%p",
2807 __func__
, item
->ri_buf
[1].i_len
, item
);
2808 error
= EFSCORRUPTED
;
2812 /* The core is in in-core format */
2813 xfs_dinode_to_disk(dip
, dicp
);
2815 /* the rest is in on-disk format */
2816 if (item
->ri_buf
[1].i_len
> isize
) {
2817 memcpy((char *)dip
+ isize
,
2818 item
->ri_buf
[1].i_addr
+ isize
,
2819 item
->ri_buf
[1].i_len
- isize
);
2822 fields
= in_f
->ilf_fields
;
2823 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2825 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2828 memcpy(XFS_DFORK_DPTR(dip
),
2829 &in_f
->ilf_u
.ilfu_uuid
,
2834 if (in_f
->ilf_size
== 2)
2835 goto out_owner_change
;
2836 len
= item
->ri_buf
[2].i_len
;
2837 src
= item
->ri_buf
[2].i_addr
;
2838 ASSERT(in_f
->ilf_size
<= 4);
2839 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2840 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2841 (len
== in_f
->ilf_dsize
));
2843 switch (fields
& XFS_ILOG_DFORK
) {
2844 case XFS_ILOG_DDATA
:
2846 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2849 case XFS_ILOG_DBROOT
:
2850 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2851 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2852 XFS_DFORK_DSIZE(dip
, mp
));
2857 * There are no data fork flags set.
2859 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2864 * If we logged any attribute data, recover it. There may or
2865 * may not have been any other non-core data logged in this
2868 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2869 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2874 len
= item
->ri_buf
[attr_index
].i_len
;
2875 src
= item
->ri_buf
[attr_index
].i_addr
;
2876 ASSERT(len
== in_f
->ilf_asize
);
2878 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2879 case XFS_ILOG_ADATA
:
2881 dest
= XFS_DFORK_APTR(dip
);
2882 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2883 memcpy(dest
, src
, len
);
2886 case XFS_ILOG_ABROOT
:
2887 dest
= XFS_DFORK_APTR(dip
);
2888 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2889 len
, (xfs_bmdr_block_t
*)dest
,
2890 XFS_DFORK_ASIZE(dip
, mp
));
2894 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2902 if (in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
))
2903 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
2905 /* re-generate the checksum. */
2906 xfs_dinode_calc_crc(log
->l_mp
, dip
);
2908 ASSERT(bp
->b_target
->bt_mount
== mp
);
2909 bp
->b_iodone
= xlog_recover_iodone
;
2910 xfs_buf_delwri_queue(bp
, buffer_list
);
2917 return XFS_ERROR(error
);
2921 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2922 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2926 xlog_recover_quotaoff_pass1(
2928 struct xlog_recover_item
*item
)
2930 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2934 * The logitem format's flag tells us if this was user quotaoff,
2935 * group/project quotaoff or both.
2937 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2938 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2939 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2940 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2941 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2942 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2948 * Recover a dquot record
2951 xlog_recover_dquot_pass2(
2953 struct list_head
*buffer_list
,
2954 struct xlog_recover_item
*item
,
2955 xfs_lsn_t current_lsn
)
2957 xfs_mount_t
*mp
= log
->l_mp
;
2959 struct xfs_disk_dquot
*ddq
, *recddq
;
2961 xfs_dq_logformat_t
*dq_f
;
2966 * Filesystems are required to send in quota flags at mount time.
2968 if (mp
->m_qflags
== 0)
2971 recddq
= item
->ri_buf
[1].i_addr
;
2972 if (recddq
== NULL
) {
2973 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2974 return XFS_ERROR(EIO
);
2976 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2977 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2978 item
->ri_buf
[1].i_len
, __func__
);
2979 return XFS_ERROR(EIO
);
2983 * This type of quotas was turned off, so ignore this record.
2985 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2987 if (log
->l_quotaoffs_flag
& type
)
2991 * At this point we know that quota was _not_ turned off.
2992 * Since the mount flags are not indicating to us otherwise, this
2993 * must mean that quota is on, and the dquot needs to be replayed.
2994 * Remember that we may not have fully recovered the superblock yet,
2995 * so we can't do the usual trick of looking at the SB quota bits.
2997 * The other possibility, of course, is that the quota subsystem was
2998 * removed since the last mount - ENOSYS.
3000 dq_f
= item
->ri_buf
[0].i_addr
;
3002 error
= xfs_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3003 "xlog_recover_dquot_pass2 (log copy)");
3005 return XFS_ERROR(EIO
);
3006 ASSERT(dq_f
->qlf_len
== 1);
3008 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3009 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
3015 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
3018 * At least the magic num portion should be on disk because this
3019 * was among a chunk of dquots created earlier, and we did some
3020 * minimal initialization then.
3022 error
= xfs_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3023 "xlog_recover_dquot_pass2");
3026 return XFS_ERROR(EIO
);
3030 * If the dquot has an LSN in it, recover the dquot only if it's less
3031 * than the lsn of the transaction we are replaying.
3033 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3034 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
3035 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
3037 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3042 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
3043 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3044 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
3048 ASSERT(dq_f
->qlf_size
== 2);
3049 ASSERT(bp
->b_target
->bt_mount
== mp
);
3050 bp
->b_iodone
= xlog_recover_iodone
;
3051 xfs_buf_delwri_queue(bp
, buffer_list
);
3059 * This routine is called to create an in-core extent free intent
3060 * item from the efi format structure which was logged on disk.
3061 * It allocates an in-core efi, copies the extents from the format
3062 * structure into it, and adds the efi to the AIL with the given
3066 xlog_recover_efi_pass2(
3068 struct xlog_recover_item
*item
,
3072 xfs_mount_t
*mp
= log
->l_mp
;
3073 xfs_efi_log_item_t
*efip
;
3074 xfs_efi_log_format_t
*efi_formatp
;
3076 efi_formatp
= item
->ri_buf
[0].i_addr
;
3078 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
3079 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
3080 &(efip
->efi_format
)))) {
3081 xfs_efi_item_free(efip
);
3084 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
3086 spin_lock(&log
->l_ailp
->xa_lock
);
3088 * xfs_trans_ail_update() drops the AIL lock.
3090 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
3096 * This routine is called when an efd format structure is found in
3097 * a committed transaction in the log. It's purpose is to cancel
3098 * the corresponding efi if it was still in the log. To do this
3099 * it searches the AIL for the efi with an id equal to that in the
3100 * efd format structure. If we find it, we remove the efi from the
3104 xlog_recover_efd_pass2(
3106 struct xlog_recover_item
*item
)
3108 xfs_efd_log_format_t
*efd_formatp
;
3109 xfs_efi_log_item_t
*efip
= NULL
;
3110 xfs_log_item_t
*lip
;
3112 struct xfs_ail_cursor cur
;
3113 struct xfs_ail
*ailp
= log
->l_ailp
;
3115 efd_formatp
= item
->ri_buf
[0].i_addr
;
3116 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
3117 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
3118 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
3119 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
3120 efi_id
= efd_formatp
->efd_efi_id
;
3123 * Search for the efi with the id in the efd format structure
3126 spin_lock(&ailp
->xa_lock
);
3127 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3128 while (lip
!= NULL
) {
3129 if (lip
->li_type
== XFS_LI_EFI
) {
3130 efip
= (xfs_efi_log_item_t
*)lip
;
3131 if (efip
->efi_format
.efi_id
== efi_id
) {
3133 * xfs_trans_ail_delete() drops the
3136 xfs_trans_ail_delete(ailp
, lip
,
3137 SHUTDOWN_CORRUPT_INCORE
);
3138 xfs_efi_item_free(efip
);
3139 spin_lock(&ailp
->xa_lock
);
3143 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3145 xfs_trans_ail_cursor_done(ailp
, &cur
);
3146 spin_unlock(&ailp
->xa_lock
);
3152 * This routine is called when an inode create format structure is found in a
3153 * committed transaction in the log. It's purpose is to initialise the inodes
3154 * being allocated on disk. This requires us to get inode cluster buffers that
3155 * match the range to be intialised, stamped with inode templates and written
3156 * by delayed write so that subsequent modifications will hit the cached buffer
3157 * and only need writing out at the end of recovery.
3160 xlog_recover_do_icreate_pass2(
3162 struct list_head
*buffer_list
,
3163 xlog_recover_item_t
*item
)
3165 struct xfs_mount
*mp
= log
->l_mp
;
3166 struct xfs_icreate_log
*icl
;
3167 xfs_agnumber_t agno
;
3168 xfs_agblock_t agbno
;
3171 xfs_agblock_t length
;
3173 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3174 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3175 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3179 if (icl
->icl_size
!= 1) {
3180 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3184 agno
= be32_to_cpu(icl
->icl_ag
);
3185 if (agno
>= mp
->m_sb
.sb_agcount
) {
3186 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3189 agbno
= be32_to_cpu(icl
->icl_agbno
);
3190 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3191 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3194 isize
= be32_to_cpu(icl
->icl_isize
);
3195 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3196 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3199 count
= be32_to_cpu(icl
->icl_count
);
3201 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3204 length
= be32_to_cpu(icl
->icl_length
);
3205 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3206 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3210 /* existing allocation is fixed value */
3211 ASSERT(count
== XFS_IALLOC_INODES(mp
));
3212 ASSERT(length
== XFS_IALLOC_BLOCKS(mp
));
3213 if (count
!= XFS_IALLOC_INODES(mp
) ||
3214 length
!= XFS_IALLOC_BLOCKS(mp
)) {
3215 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count 2");
3220 * Inode buffers can be freed. Do not replay the inode initialisation as
3221 * we could be overwriting something written after this inode buffer was
3224 * XXX: we need to iterate all buffers and only init those that are not
3225 * cancelled. I think that a more fine grained factoring of
3226 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3229 if (xlog_check_buffer_cancelled(log
,
3230 XFS_AGB_TO_DADDR(mp
, agno
, agbno
), length
, 0))
3233 xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, agno
, agbno
, length
,
3234 be32_to_cpu(icl
->icl_gen
));
3239 * Free up any resources allocated by the transaction
3241 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3244 xlog_recover_free_trans(
3245 struct xlog_recover
*trans
)
3247 xlog_recover_item_t
*item
, *n
;
3250 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3251 /* Free the regions in the item. */
3252 list_del(&item
->ri_list
);
3253 for (i
= 0; i
< item
->ri_cnt
; i
++)
3254 kmem_free(item
->ri_buf
[i
].i_addr
);
3255 /* Free the item itself */
3256 kmem_free(item
->ri_buf
);
3259 /* Free the transaction recover structure */
3264 xlog_recover_buffer_ra_pass2(
3266 struct xlog_recover_item
*item
)
3268 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3269 struct xfs_mount
*mp
= log
->l_mp
;
3271 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3272 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3276 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3277 buf_f
->blf_len
, NULL
);
3281 xlog_recover_inode_ra_pass2(
3283 struct xlog_recover_item
*item
)
3285 struct xfs_inode_log_format ilf_buf
;
3286 struct xfs_inode_log_format
*ilfp
;
3287 struct xfs_mount
*mp
= log
->l_mp
;
3290 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3291 ilfp
= item
->ri_buf
[0].i_addr
;
3294 memset(ilfp
, 0, sizeof(*ilfp
));
3295 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3300 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3303 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3304 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3308 xlog_recover_dquot_ra_pass2(
3310 struct xlog_recover_item
*item
)
3312 struct xfs_mount
*mp
= log
->l_mp
;
3313 struct xfs_disk_dquot
*recddq
;
3314 struct xfs_dq_logformat
*dq_f
;
3318 if (mp
->m_qflags
== 0)
3321 recddq
= item
->ri_buf
[1].i_addr
;
3324 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
3327 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3329 if (log
->l_quotaoffs_flag
& type
)
3332 dq_f
= item
->ri_buf
[0].i_addr
;
3334 ASSERT(dq_f
->qlf_len
== 1);
3336 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3337 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), NULL
);
3341 xlog_recover_ra_pass2(
3343 struct xlog_recover_item
*item
)
3345 switch (ITEM_TYPE(item
)) {
3347 xlog_recover_buffer_ra_pass2(log
, item
);
3350 xlog_recover_inode_ra_pass2(log
, item
);
3353 xlog_recover_dquot_ra_pass2(log
, item
);
3357 case XFS_LI_QUOTAOFF
:
3364 xlog_recover_commit_pass1(
3366 struct xlog_recover
*trans
,
3367 struct xlog_recover_item
*item
)
3369 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3371 switch (ITEM_TYPE(item
)) {
3373 return xlog_recover_buffer_pass1(log
, item
);
3374 case XFS_LI_QUOTAOFF
:
3375 return xlog_recover_quotaoff_pass1(log
, item
);
3380 case XFS_LI_ICREATE
:
3381 /* nothing to do in pass 1 */
3384 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3385 __func__
, ITEM_TYPE(item
));
3387 return XFS_ERROR(EIO
);
3392 xlog_recover_commit_pass2(
3394 struct xlog_recover
*trans
,
3395 struct list_head
*buffer_list
,
3396 struct xlog_recover_item
*item
)
3398 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3400 switch (ITEM_TYPE(item
)) {
3402 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
3405 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
3408 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3410 return xlog_recover_efd_pass2(log
, item
);
3412 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
3414 case XFS_LI_ICREATE
:
3415 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3416 case XFS_LI_QUOTAOFF
:
3417 /* nothing to do in pass2 */
3420 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3421 __func__
, ITEM_TYPE(item
));
3423 return XFS_ERROR(EIO
);
3428 xlog_recover_items_pass2(
3430 struct xlog_recover
*trans
,
3431 struct list_head
*buffer_list
,
3432 struct list_head
*item_list
)
3434 struct xlog_recover_item
*item
;
3437 list_for_each_entry(item
, item_list
, ri_list
) {
3438 error
= xlog_recover_commit_pass2(log
, trans
,
3448 * Perform the transaction.
3450 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3451 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3454 xlog_recover_commit_trans(
3456 struct xlog_recover
*trans
,
3461 int items_queued
= 0;
3462 struct xlog_recover_item
*item
;
3463 struct xlog_recover_item
*next
;
3464 LIST_HEAD (buffer_list
);
3465 LIST_HEAD (ra_list
);
3466 LIST_HEAD (done_list
);
3468 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3470 hlist_del(&trans
->r_list
);
3472 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3476 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
3478 case XLOG_RECOVER_PASS1
:
3479 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3481 case XLOG_RECOVER_PASS2
:
3482 xlog_recover_ra_pass2(log
, item
);
3483 list_move_tail(&item
->ri_list
, &ra_list
);
3485 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
3486 error
= xlog_recover_items_pass2(log
, trans
,
3487 &buffer_list
, &ra_list
);
3488 list_splice_tail_init(&ra_list
, &done_list
);
3502 if (!list_empty(&ra_list
)) {
3504 error
= xlog_recover_items_pass2(log
, trans
,
3505 &buffer_list
, &ra_list
);
3506 list_splice_tail_init(&ra_list
, &done_list
);
3509 if (!list_empty(&done_list
))
3510 list_splice_init(&done_list
, &trans
->r_itemq
);
3512 xlog_recover_free_trans(trans
);
3514 error2
= xfs_buf_delwri_submit(&buffer_list
);
3515 return error
? error
: error2
;
3519 xlog_recover_unmount_trans(
3521 struct xlog_recover
*trans
)
3523 /* Do nothing now */
3524 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3529 * There are two valid states of the r_state field. 0 indicates that the
3530 * transaction structure is in a normal state. We have either seen the
3531 * start of the transaction or the last operation we added was not a partial
3532 * operation. If the last operation we added to the transaction was a
3533 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3535 * NOTE: skip LRs with 0 data length.
3538 xlog_recover_process_data(
3540 struct hlist_head rhash
[],
3541 struct xlog_rec_header
*rhead
,
3547 xlog_op_header_t
*ohead
;
3548 xlog_recover_t
*trans
;
3554 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
3555 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
3557 /* check the log format matches our own - else we can't recover */
3558 if (xlog_header_check_recover(log
->l_mp
, rhead
))
3559 return (XFS_ERROR(EIO
));
3561 while ((dp
< lp
) && num_logops
) {
3562 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
3563 ohead
= (xlog_op_header_t
*)dp
;
3564 dp
+= sizeof(xlog_op_header_t
);
3565 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
3566 ohead
->oh_clientid
!= XFS_LOG
) {
3567 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
3568 __func__
, ohead
->oh_clientid
);
3570 return (XFS_ERROR(EIO
));
3572 tid
= be32_to_cpu(ohead
->oh_tid
);
3573 hash
= XLOG_RHASH(tid
);
3574 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
3575 if (trans
== NULL
) { /* not found; add new tid */
3576 if (ohead
->oh_flags
& XLOG_START_TRANS
)
3577 xlog_recover_new_tid(&rhash
[hash
], tid
,
3578 be64_to_cpu(rhead
->h_lsn
));
3580 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
3581 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
3582 __func__
, be32_to_cpu(ohead
->oh_len
));
3584 return (XFS_ERROR(EIO
));
3586 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
3587 if (flags
& XLOG_WAS_CONT_TRANS
)
3588 flags
&= ~XLOG_CONTINUE_TRANS
;
3590 case XLOG_COMMIT_TRANS
:
3591 error
= xlog_recover_commit_trans(log
,
3594 case XLOG_UNMOUNT_TRANS
:
3595 error
= xlog_recover_unmount_trans(log
, trans
);
3597 case XLOG_WAS_CONT_TRANS
:
3598 error
= xlog_recover_add_to_cont_trans(log
,
3600 be32_to_cpu(ohead
->oh_len
));
3602 case XLOG_START_TRANS
:
3603 xfs_warn(log
->l_mp
, "%s: bad transaction",
3606 error
= XFS_ERROR(EIO
);
3609 case XLOG_CONTINUE_TRANS
:
3610 error
= xlog_recover_add_to_trans(log
, trans
,
3611 dp
, be32_to_cpu(ohead
->oh_len
));
3614 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
3617 error
= XFS_ERROR(EIO
);
3621 xlog_recover_free_trans(trans
);
3625 dp
+= be32_to_cpu(ohead
->oh_len
);
3632 * Process an extent free intent item that was recovered from
3633 * the log. We need to free the extents that it describes.
3636 xlog_recover_process_efi(
3638 xfs_efi_log_item_t
*efip
)
3640 xfs_efd_log_item_t
*efdp
;
3645 xfs_fsblock_t startblock_fsb
;
3647 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
3650 * First check the validity of the extents described by the
3651 * EFI. If any are bad, then assume that all are bad and
3652 * just toss the EFI.
3654 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3655 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3656 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3657 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3658 if ((startblock_fsb
== 0) ||
3659 (extp
->ext_len
== 0) ||
3660 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3661 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3663 * This will pull the EFI from the AIL and
3664 * free the memory associated with it.
3666 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3667 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3668 return XFS_ERROR(EIO
);
3672 tp
= xfs_trans_alloc(mp
, 0);
3673 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
3676 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3678 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3679 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3680 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3683 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3687 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3688 error
= xfs_trans_commit(tp
, 0);
3692 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3697 * When this is called, all of the EFIs which did not have
3698 * corresponding EFDs should be in the AIL. What we do now
3699 * is free the extents associated with each one.
3701 * Since we process the EFIs in normal transactions, they
3702 * will be removed at some point after the commit. This prevents
3703 * us from just walking down the list processing each one.
3704 * We'll use a flag in the EFI to skip those that we've already
3705 * processed and use the AIL iteration mechanism's generation
3706 * count to try to speed this up at least a bit.
3708 * When we start, we know that the EFIs are the only things in
3709 * the AIL. As we process them, however, other items are added
3710 * to the AIL. Since everything added to the AIL must come after
3711 * everything already in the AIL, we stop processing as soon as
3712 * we see something other than an EFI in the AIL.
3715 xlog_recover_process_efis(
3718 xfs_log_item_t
*lip
;
3719 xfs_efi_log_item_t
*efip
;
3721 struct xfs_ail_cursor cur
;
3722 struct xfs_ail
*ailp
;
3725 spin_lock(&ailp
->xa_lock
);
3726 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3727 while (lip
!= NULL
) {
3729 * We're done when we see something other than an EFI.
3730 * There should be no EFIs left in the AIL now.
3732 if (lip
->li_type
!= XFS_LI_EFI
) {
3734 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3735 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3741 * Skip EFIs that we've already processed.
3743 efip
= (xfs_efi_log_item_t
*)lip
;
3744 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3745 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3749 spin_unlock(&ailp
->xa_lock
);
3750 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3751 spin_lock(&ailp
->xa_lock
);
3754 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3757 xfs_trans_ail_cursor_done(ailp
, &cur
);
3758 spin_unlock(&ailp
->xa_lock
);
3763 * This routine performs a transaction to null out a bad inode pointer
3764 * in an agi unlinked inode hash bucket.
3767 xlog_recover_clear_agi_bucket(
3769 xfs_agnumber_t agno
,
3778 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3779 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
3783 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3787 agi
= XFS_BUF_TO_AGI(agibp
);
3788 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3789 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3790 (sizeof(xfs_agino_t
) * bucket
);
3791 xfs_trans_log_buf(tp
, agibp
, offset
,
3792 (offset
+ sizeof(xfs_agino_t
) - 1));
3794 error
= xfs_trans_commit(tp
, 0);
3800 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3802 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3807 xlog_recover_process_one_iunlink(
3808 struct xfs_mount
*mp
,
3809 xfs_agnumber_t agno
,
3813 struct xfs_buf
*ibp
;
3814 struct xfs_dinode
*dip
;
3815 struct xfs_inode
*ip
;
3819 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3820 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3825 * Get the on disk inode to find the next inode in the bucket.
3827 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3831 ASSERT(ip
->i_d
.di_nlink
== 0);
3832 ASSERT(ip
->i_d
.di_mode
!= 0);
3834 /* setup for the next pass */
3835 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3839 * Prevent any DMAPI event from being sent when the reference on
3840 * the inode is dropped.
3842 ip
->i_d
.di_dmevmask
= 0;
3851 * We can't read in the inode this bucket points to, or this inode
3852 * is messed up. Just ditch this bucket of inodes. We will lose
3853 * some inodes and space, but at least we won't hang.
3855 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3856 * clear the inode pointer in the bucket.
3858 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3863 * xlog_iunlink_recover
3865 * This is called during recovery to process any inodes which
3866 * we unlinked but not freed when the system crashed. These
3867 * inodes will be on the lists in the AGI blocks. What we do
3868 * here is scan all the AGIs and fully truncate and free any
3869 * inodes found on the lists. Each inode is removed from the
3870 * lists when it has been fully truncated and is freed. The
3871 * freeing of the inode and its removal from the list must be
3875 xlog_recover_process_iunlinks(
3879 xfs_agnumber_t agno
;
3890 * Prevent any DMAPI event from being sent while in this function.
3892 mp_dmevmask
= mp
->m_dmevmask
;
3895 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3897 * Find the agi for this ag.
3899 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3902 * AGI is b0rked. Don't process it.
3904 * We should probably mark the filesystem as corrupt
3905 * after we've recovered all the ag's we can....
3910 * Unlock the buffer so that it can be acquired in the normal
3911 * course of the transaction to truncate and free each inode.
3912 * Because we are not racing with anyone else here for the AGI
3913 * buffer, we don't even need to hold it locked to read the
3914 * initial unlinked bucket entries out of the buffer. We keep
3915 * buffer reference though, so that it stays pinned in memory
3916 * while we need the buffer.
3918 agi
= XFS_BUF_TO_AGI(agibp
);
3919 xfs_buf_unlock(agibp
);
3921 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3922 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3923 while (agino
!= NULLAGINO
) {
3924 agino
= xlog_recover_process_one_iunlink(mp
,
3925 agno
, agino
, bucket
);
3928 xfs_buf_rele(agibp
);
3931 mp
->m_dmevmask
= mp_dmevmask
;
3935 * Upack the log buffer data and crc check it. If the check fails, issue a
3936 * warning if and only if the CRC in the header is non-zero. This makes the
3937 * check an advisory warning, and the zero CRC check will prevent failure
3938 * warnings from being emitted when upgrading the kernel from one that does not
3939 * add CRCs by default.
3941 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3942 * corruption failure
3945 xlog_unpack_data_crc(
3946 struct xlog_rec_header
*rhead
,
3952 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
3953 if (crc
!= rhead
->h_crc
) {
3954 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
3955 xfs_alert(log
->l_mp
,
3956 "log record CRC mismatch: found 0x%x, expected 0x%x.",
3957 le32_to_cpu(rhead
->h_crc
),
3959 xfs_hex_dump(dp
, 32);
3963 * If we've detected a log record corruption, then we can't
3964 * recover past this point. Abort recovery if we are enforcing
3965 * CRC protection by punting an error back up the stack.
3967 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
3968 return EFSCORRUPTED
;
3976 struct xlog_rec_header
*rhead
,
3983 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
3987 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3988 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3989 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3993 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3994 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3995 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3996 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3997 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3998 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
4007 xlog_valid_rec_header(
4009 struct xlog_rec_header
*rhead
,
4014 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
4015 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4016 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4017 return XFS_ERROR(EFSCORRUPTED
);
4020 (!rhead
->h_version
||
4021 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
4022 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
4023 __func__
, be32_to_cpu(rhead
->h_version
));
4024 return XFS_ERROR(EIO
);
4027 /* LR body must have data or it wouldn't have been written */
4028 hlen
= be32_to_cpu(rhead
->h_len
);
4029 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
4030 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4031 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4032 return XFS_ERROR(EFSCORRUPTED
);
4034 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
4035 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4036 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4037 return XFS_ERROR(EFSCORRUPTED
);
4043 * Read the log from tail to head and process the log records found.
4044 * Handle the two cases where the tail and head are in the same cycle
4045 * and where the active portion of the log wraps around the end of
4046 * the physical log separately. The pass parameter is passed through
4047 * to the routines called to process the data and is not looked at
4051 xlog_do_recovery_pass(
4053 xfs_daddr_t head_blk
,
4054 xfs_daddr_t tail_blk
,
4057 xlog_rec_header_t
*rhead
;
4060 xfs_buf_t
*hbp
, *dbp
;
4061 int error
= 0, h_size
;
4062 int bblks
, split_bblks
;
4063 int hblks
, split_hblks
, wrapped_hblks
;
4064 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
4066 ASSERT(head_blk
!= tail_blk
);
4069 * Read the header of the tail block and get the iclog buffer size from
4070 * h_size. Use this to tell how many sectors make up the log header.
4072 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4074 * When using variable length iclogs, read first sector of
4075 * iclog header and extract the header size from it. Get a
4076 * new hbp that is the correct size.
4078 hbp
= xlog_get_bp(log
, 1);
4082 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
4086 rhead
= (xlog_rec_header_t
*)offset
;
4087 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
4090 h_size
= be32_to_cpu(rhead
->h_size
);
4091 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
4092 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
4093 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
4094 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
4097 hbp
= xlog_get_bp(log
, hblks
);
4102 ASSERT(log
->l_sectBBsize
== 1);
4104 hbp
= xlog_get_bp(log
, 1);
4105 h_size
= XLOG_BIG_RECORD_BSIZE
;
4110 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
4116 memset(rhash
, 0, sizeof(rhash
));
4117 if (tail_blk
<= head_blk
) {
4118 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
4119 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4123 rhead
= (xlog_rec_header_t
*)offset
;
4124 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4128 /* blocks in data section */
4129 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4130 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
4135 error
= xlog_unpack_data(rhead
, offset
, log
);
4139 error
= xlog_recover_process_data(log
,
4140 rhash
, rhead
, offset
, pass
);
4143 blk_no
+= bblks
+ hblks
;
4147 * Perform recovery around the end of the physical log.
4148 * When the head is not on the same cycle number as the tail,
4149 * we can't do a sequential recovery as above.
4152 while (blk_no
< log
->l_logBBsize
) {
4154 * Check for header wrapping around physical end-of-log
4156 offset
= hbp
->b_addr
;
4159 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
4160 /* Read header in one read */
4161 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
4166 /* This LR is split across physical log end */
4167 if (blk_no
!= log
->l_logBBsize
) {
4168 /* some data before physical log end */
4169 ASSERT(blk_no
<= INT_MAX
);
4170 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
4171 ASSERT(split_hblks
> 0);
4172 error
= xlog_bread(log
, blk_no
,
4180 * Note: this black magic still works with
4181 * large sector sizes (non-512) only because:
4182 * - we increased the buffer size originally
4183 * by 1 sector giving us enough extra space
4184 * for the second read;
4185 * - the log start is guaranteed to be sector
4187 * - we read the log end (LR header start)
4188 * _first_, then the log start (LR header end)
4189 * - order is important.
4191 wrapped_hblks
= hblks
- split_hblks
;
4192 error
= xlog_bread_offset(log
, 0,
4194 offset
+ BBTOB(split_hblks
));
4198 rhead
= (xlog_rec_header_t
*)offset
;
4199 error
= xlog_valid_rec_header(log
, rhead
,
4200 split_hblks
? blk_no
: 0);
4204 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4207 /* Read in data for log record */
4208 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
4209 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
4214 /* This log record is split across the
4215 * physical end of log */
4216 offset
= dbp
->b_addr
;
4218 if (blk_no
!= log
->l_logBBsize
) {
4219 /* some data is before the physical
4221 ASSERT(!wrapped_hblks
);
4222 ASSERT(blk_no
<= INT_MAX
);
4224 log
->l_logBBsize
- (int)blk_no
;
4225 ASSERT(split_bblks
> 0);
4226 error
= xlog_bread(log
, blk_no
,
4234 * Note: this black magic still works with
4235 * large sector sizes (non-512) only because:
4236 * - we increased the buffer size originally
4237 * by 1 sector giving us enough extra space
4238 * for the second read;
4239 * - the log start is guaranteed to be sector
4241 * - we read the log end (LR header start)
4242 * _first_, then the log start (LR header end)
4243 * - order is important.
4245 error
= xlog_bread_offset(log
, 0,
4246 bblks
- split_bblks
, dbp
,
4247 offset
+ BBTOB(split_bblks
));
4252 error
= xlog_unpack_data(rhead
, offset
, log
);
4256 error
= xlog_recover_process_data(log
, rhash
,
4257 rhead
, offset
, pass
);
4263 ASSERT(blk_no
>= log
->l_logBBsize
);
4264 blk_no
-= log
->l_logBBsize
;
4266 /* read first part of physical log */
4267 while (blk_no
< head_blk
) {
4268 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4272 rhead
= (xlog_rec_header_t
*)offset
;
4273 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4277 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4278 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
4283 error
= xlog_unpack_data(rhead
, offset
, log
);
4287 error
= xlog_recover_process_data(log
, rhash
,
4288 rhead
, offset
, pass
);
4291 blk_no
+= bblks
+ hblks
;
4303 * Do the recovery of the log. We actually do this in two phases.
4304 * The two passes are necessary in order to implement the function
4305 * of cancelling a record written into the log. The first pass
4306 * determines those things which have been cancelled, and the
4307 * second pass replays log items normally except for those which
4308 * have been cancelled. The handling of the replay and cancellations
4309 * takes place in the log item type specific routines.
4311 * The table of items which have cancel records in the log is allocated
4312 * and freed at this level, since only here do we know when all of
4313 * the log recovery has been completed.
4316 xlog_do_log_recovery(
4318 xfs_daddr_t head_blk
,
4319 xfs_daddr_t tail_blk
)
4323 ASSERT(head_blk
!= tail_blk
);
4326 * First do a pass to find all of the cancelled buf log items.
4327 * Store them in the buf_cancel_table for use in the second pass.
4329 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4330 sizeof(struct list_head
),
4332 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4333 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4335 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4336 XLOG_RECOVER_PASS1
);
4338 kmem_free(log
->l_buf_cancel_table
);
4339 log
->l_buf_cancel_table
= NULL
;
4343 * Then do a second pass to actually recover the items in the log.
4344 * When it is complete free the table of buf cancel items.
4346 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4347 XLOG_RECOVER_PASS2
);
4352 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4353 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4357 kmem_free(log
->l_buf_cancel_table
);
4358 log
->l_buf_cancel_table
= NULL
;
4364 * Do the actual recovery
4369 xfs_daddr_t head_blk
,
4370 xfs_daddr_t tail_blk
)
4377 * First replay the images in the log.
4379 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4384 * If IO errors happened during recovery, bail out.
4386 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4391 * We now update the tail_lsn since much of the recovery has completed
4392 * and there may be space available to use. If there were no extent
4393 * or iunlinks, we can free up the entire log and set the tail_lsn to
4394 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4395 * lsn of the last known good LR on disk. If there are extent frees
4396 * or iunlinks they will have some entries in the AIL; so we look at
4397 * the AIL to determine how to set the tail_lsn.
4399 xlog_assign_tail_lsn(log
->l_mp
);
4402 * Now that we've finished replaying all buffer and inode
4403 * updates, re-read in the superblock and reverify it.
4405 bp
= xfs_getsb(log
->l_mp
, 0);
4407 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
4409 XFS_BUF_UNASYNC(bp
);
4410 bp
->b_ops
= &xfs_sb_buf_ops
;
4411 xfsbdstrat(log
->l_mp
, bp
);
4412 error
= xfs_buf_iowait(bp
);
4414 xfs_buf_ioerror_alert(bp
, __func__
);
4420 /* Convert superblock from on-disk format */
4421 sbp
= &log
->l_mp
->m_sb
;
4422 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4423 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4424 ASSERT(xfs_sb_good_version(sbp
));
4427 /* We've re-read the superblock so re-initialize per-cpu counters */
4428 xfs_icsb_reinit_counters(log
->l_mp
);
4430 xlog_recover_check_summary(log
);
4432 /* Normal transactions can now occur */
4433 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4438 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4440 * Return error or zero.
4446 xfs_daddr_t head_blk
, tail_blk
;
4449 /* find the tail of the log */
4450 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
4453 if (tail_blk
!= head_blk
) {
4454 /* There used to be a comment here:
4456 * disallow recovery on read-only mounts. note -- mount
4457 * checks for ENOSPC and turns it into an intelligent
4459 * ...but this is no longer true. Now, unless you specify
4460 * NORECOVERY (in which case this function would never be
4461 * called), we just go ahead and recover. We do this all
4462 * under the vfs layer, so we can get away with it unless
4463 * the device itself is read-only, in which case we fail.
4465 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
4470 * Version 5 superblock log feature mask validation. We know the
4471 * log is dirty so check if there are any unknown log features
4472 * in what we need to recover. If there are unknown features
4473 * (e.g. unsupported transactions, then simply reject the
4474 * attempt at recovery before touching anything.
4476 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
4477 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
4478 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
4480 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4481 "The log can not be fully and/or safely recovered by this kernel.\n"
4482 "Please recover the log on a kernel that supports the unknown features.",
4483 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
4484 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
4488 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
4489 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4492 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
4493 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
4499 * In the first part of recovery we replay inodes and buffers and build
4500 * up the list of extent free items which need to be processed. Here
4501 * we process the extent free items and clean up the on disk unlinked
4502 * inode lists. This is separated from the first part of recovery so
4503 * that the root and real-time bitmap inodes can be read in from disk in
4504 * between the two stages. This is necessary so that we can free space
4505 * in the real-time portion of the file system.
4508 xlog_recover_finish(
4512 * Now we're ready to do the transactions needed for the
4513 * rest of recovery. Start with completing all the extent
4514 * free intent records and then process the unlinked inode
4515 * lists. At this point, we essentially run in normal mode
4516 * except that we're still performing recovery actions
4517 * rather than accepting new requests.
4519 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
4521 error
= xlog_recover_process_efis(log
);
4523 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
4527 * Sync the log to get all the EFIs out of the AIL.
4528 * This isn't absolutely necessary, but it helps in
4529 * case the unlink transactions would have problems
4530 * pushing the EFIs out of the way.
4532 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
4534 xlog_recover_process_iunlinks(log
);
4536 xlog_recover_check_summary(log
);
4538 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
4539 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4541 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4543 xfs_info(log
->l_mp
, "Ending clean mount");
4551 * Read all of the agf and agi counters and check that they
4552 * are consistent with the superblock counters.
4555 xlog_recover_check_summary(
4562 xfs_agnumber_t agno
;
4563 __uint64_t freeblks
;
4573 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4574 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4576 xfs_alert(mp
, "%s agf read failed agno %d error %d",
4577 __func__
, agno
, error
);
4579 agfp
= XFS_BUF_TO_AGF(agfbp
);
4580 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4581 be32_to_cpu(agfp
->agf_flcount
);
4582 xfs_buf_relse(agfbp
);
4585 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4587 xfs_alert(mp
, "%s agi read failed agno %d error %d",
4588 __func__
, agno
, error
);
4590 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4592 itotal
+= be32_to_cpu(agi
->agi_count
);
4593 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4594 xfs_buf_relse(agibp
);