2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_log_priv.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_log_recover.h"
41 #include "xfs_extfree_item.h"
42 #include "xfs_trans_priv.h"
43 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
54 xlog_clear_stale_blocks(
59 xlog_recover_check_summary(
62 #define xlog_recover_check_summary(log)
66 * This structure is used during recovery to record the buf log items which
67 * have been canceled and should not be replayed.
69 struct xfs_buf_cancel
{
73 struct list_head bc_list
;
77 * Sector aligned buffer routines for buffer create/read/write/access
81 * Verify the given count of basic blocks is valid number of blocks
82 * to specify for an operation involving the given XFS log buffer.
83 * Returns nonzero if the count is valid, 0 otherwise.
87 xlog_buf_bbcount_valid(
91 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
106 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
107 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
109 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
114 * We do log I/O in units of log sectors (a power-of-2
115 * multiple of the basic block size), so we round up the
116 * requested size to accommodate the basic blocks required
117 * for complete log sectors.
119 * In addition, the buffer may be used for a non-sector-
120 * aligned block offset, in which case an I/O of the
121 * requested size could extend beyond the end of the
122 * buffer. If the requested size is only 1 basic block it
123 * will never straddle a sector boundary, so this won't be
124 * an issue. Nor will this be a problem if the log I/O is
125 * done in basic blocks (sector size 1). But otherwise we
126 * extend the buffer by one extra log sector to ensure
127 * there's space to accommodate this possibility.
129 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
130 nbblks
+= log
->l_sectBBsize
;
131 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
133 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
147 * Return the address of the start of the given block number's data
148 * in a log buffer. The buffer covers a log sector-aligned region.
157 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
159 ASSERT(offset
+ nbblks
<= bp
->b_length
);
160 return bp
->b_addr
+ BBTOB(offset
);
165 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
176 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
177 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
179 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
183 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
184 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
187 ASSERT(nbblks
<= bp
->b_length
);
189 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
191 bp
->b_io_length
= nbblks
;
194 xfsbdstrat(log
->l_mp
, bp
);
195 error
= xfs_buf_iowait(bp
);
197 xfs_buf_ioerror_alert(bp
, __func__
);
211 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
215 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
220 * Read at an offset into the buffer. Returns with the buffer in it's original
221 * state regardless of the result of the read.
226 xfs_daddr_t blk_no
, /* block to read from */
227 int nbblks
, /* blocks to read */
231 xfs_caddr_t orig_offset
= bp
->b_addr
;
232 int orig_len
= BBTOB(bp
->b_length
);
235 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
239 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
241 /* must reset buffer pointer even on error */
242 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
249 * Write out the buffer at the given block for the given number of blocks.
250 * The buffer is kept locked across the write and is returned locked.
251 * This can only be used for synchronous log writes.
262 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
263 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
265 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
269 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
270 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
273 ASSERT(nbblks
<= bp
->b_length
);
275 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
276 XFS_BUF_ZEROFLAGS(bp
);
279 bp
->b_io_length
= nbblks
;
282 error
= xfs_bwrite(bp
);
284 xfs_buf_ioerror_alert(bp
, __func__
);
291 * dump debug superblock and log record information
294 xlog_header_check_dump(
296 xlog_rec_header_t
*head
)
298 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d\n",
299 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
300 xfs_debug(mp
, " log : uuid = %pU, fmt = %d\n",
301 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
304 #define xlog_header_check_dump(mp, head)
308 * check log record header for recovery
311 xlog_header_check_recover(
313 xlog_rec_header_t
*head
)
315 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
318 * IRIX doesn't write the h_fmt field and leaves it zeroed
319 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
320 * a dirty log created in IRIX.
322 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
324 "dirty log written in incompatible format - can't recover");
325 xlog_header_check_dump(mp
, head
);
326 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
327 XFS_ERRLEVEL_HIGH
, mp
);
328 return XFS_ERROR(EFSCORRUPTED
);
329 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
331 "dirty log entry has mismatched uuid - can't recover");
332 xlog_header_check_dump(mp
, head
);
333 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
334 XFS_ERRLEVEL_HIGH
, mp
);
335 return XFS_ERROR(EFSCORRUPTED
);
341 * read the head block of the log and check the header
344 xlog_header_check_mount(
346 xlog_rec_header_t
*head
)
348 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
350 if (uuid_is_nil(&head
->h_fs_uuid
)) {
352 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
353 * h_fs_uuid is nil, we assume this log was last mounted
354 * by IRIX and continue.
356 xfs_warn(mp
, "nil uuid in log - IRIX style log");
357 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
358 xfs_warn(mp
, "log has mismatched uuid - can't recover");
359 xlog_header_check_dump(mp
, head
);
360 XFS_ERROR_REPORT("xlog_header_check_mount",
361 XFS_ERRLEVEL_HIGH
, mp
);
362 return XFS_ERROR(EFSCORRUPTED
);
373 * We're not going to bother about retrying
374 * this during recovery. One strike!
376 xfs_buf_ioerror_alert(bp
, __func__
);
377 xfs_force_shutdown(bp
->b_target
->bt_mount
,
378 SHUTDOWN_META_IO_ERROR
);
381 xfs_buf_ioend(bp
, 0);
385 * This routine finds (to an approximation) the first block in the physical
386 * log which contains the given cycle. It uses a binary search algorithm.
387 * Note that the algorithm can not be perfect because the disk will not
388 * necessarily be perfect.
391 xlog_find_cycle_start(
394 xfs_daddr_t first_blk
,
395 xfs_daddr_t
*last_blk
,
405 mid_blk
= BLK_AVG(first_blk
, end_blk
);
406 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
407 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
410 mid_cycle
= xlog_get_cycle(offset
);
411 if (mid_cycle
== cycle
)
412 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
414 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
415 mid_blk
= BLK_AVG(first_blk
, end_blk
);
417 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
418 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
426 * Check that a range of blocks does not contain stop_on_cycle_no.
427 * Fill in *new_blk with the block offset where such a block is
428 * found, or with -1 (an invalid block number) if there is no such
429 * block in the range. The scan needs to occur from front to back
430 * and the pointer into the region must be updated since a later
431 * routine will need to perform another test.
434 xlog_find_verify_cycle(
436 xfs_daddr_t start_blk
,
438 uint stop_on_cycle_no
,
439 xfs_daddr_t
*new_blk
)
445 xfs_caddr_t buf
= NULL
;
449 * Greedily allocate a buffer big enough to handle the full
450 * range of basic blocks we'll be examining. If that fails,
451 * try a smaller size. We need to be able to read at least
452 * a log sector, or we're out of luck.
454 bufblks
= 1 << ffs(nbblks
);
455 while (bufblks
> log
->l_logBBsize
)
457 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
459 if (bufblks
< log
->l_sectBBsize
)
463 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
466 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
468 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
472 for (j
= 0; j
< bcount
; j
++) {
473 cycle
= xlog_get_cycle(buf
);
474 if (cycle
== stop_on_cycle_no
) {
491 * Potentially backup over partial log record write.
493 * In the typical case, last_blk is the number of the block directly after
494 * a good log record. Therefore, we subtract one to get the block number
495 * of the last block in the given buffer. extra_bblks contains the number
496 * of blocks we would have read on a previous read. This happens when the
497 * last log record is split over the end of the physical log.
499 * extra_bblks is the number of blocks potentially verified on a previous
500 * call to this routine.
503 xlog_find_verify_log_record(
505 xfs_daddr_t start_blk
,
506 xfs_daddr_t
*last_blk
,
511 xfs_caddr_t offset
= NULL
;
512 xlog_rec_header_t
*head
= NULL
;
515 int num_blks
= *last_blk
- start_blk
;
518 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
520 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
521 if (!(bp
= xlog_get_bp(log
, 1)))
525 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
528 offset
+= ((num_blks
- 1) << BBSHIFT
);
531 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
533 /* valid log record not found */
535 "Log inconsistent (didn't find previous header)");
537 error
= XFS_ERROR(EIO
);
542 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
547 head
= (xlog_rec_header_t
*)offset
;
549 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
557 * We hit the beginning of the physical log & still no header. Return
558 * to caller. If caller can handle a return of -1, then this routine
559 * will be called again for the end of the physical log.
567 * We have the final block of the good log (the first block
568 * of the log record _before_ the head. So we check the uuid.
570 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
574 * We may have found a log record header before we expected one.
575 * last_blk will be the 1st block # with a given cycle #. We may end
576 * up reading an entire log record. In this case, we don't want to
577 * reset last_blk. Only when last_blk points in the middle of a log
578 * record do we update last_blk.
580 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
581 uint h_size
= be32_to_cpu(head
->h_size
);
583 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
584 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
590 if (*last_blk
- i
+ extra_bblks
!=
591 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
600 * Head is defined to be the point of the log where the next log write
601 * write could go. This means that incomplete LR writes at the end are
602 * eliminated when calculating the head. We aren't guaranteed that previous
603 * LR have complete transactions. We only know that a cycle number of
604 * current cycle number -1 won't be present in the log if we start writing
605 * from our current block number.
607 * last_blk contains the block number of the first block with a given
610 * Return: zero if normal, non-zero if error.
615 xfs_daddr_t
*return_head_blk
)
619 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
621 uint first_half_cycle
, last_half_cycle
;
623 int error
, log_bbnum
= log
->l_logBBsize
;
625 /* Is the end of the log device zeroed? */
626 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
627 *return_head_blk
= first_blk
;
629 /* Is the whole lot zeroed? */
631 /* Linux XFS shouldn't generate totally zeroed logs -
632 * mkfs etc write a dummy unmount record to a fresh
633 * log so we can store the uuid in there
635 xfs_warn(log
->l_mp
, "totally zeroed log");
640 xfs_warn(log
->l_mp
, "empty log check failed");
644 first_blk
= 0; /* get cycle # of 1st block */
645 bp
= xlog_get_bp(log
, 1);
649 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
653 first_half_cycle
= xlog_get_cycle(offset
);
655 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
656 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
660 last_half_cycle
= xlog_get_cycle(offset
);
661 ASSERT(last_half_cycle
!= 0);
664 * If the 1st half cycle number is equal to the last half cycle number,
665 * then the entire log is stamped with the same cycle number. In this
666 * case, head_blk can't be set to zero (which makes sense). The below
667 * math doesn't work out properly with head_blk equal to zero. Instead,
668 * we set it to log_bbnum which is an invalid block number, but this
669 * value makes the math correct. If head_blk doesn't changed through
670 * all the tests below, *head_blk is set to zero at the very end rather
671 * than log_bbnum. In a sense, log_bbnum and zero are the same block
672 * in a circular file.
674 if (first_half_cycle
== last_half_cycle
) {
676 * In this case we believe that the entire log should have
677 * cycle number last_half_cycle. We need to scan backwards
678 * from the end verifying that there are no holes still
679 * containing last_half_cycle - 1. If we find such a hole,
680 * then the start of that hole will be the new head. The
681 * simple case looks like
682 * x | x ... | x - 1 | x
683 * Another case that fits this picture would be
684 * x | x + 1 | x ... | x
685 * In this case the head really is somewhere at the end of the
686 * log, as one of the latest writes at the beginning was
689 * x | x + 1 | x ... | x - 1 | x
690 * This is really the combination of the above two cases, and
691 * the head has to end up at the start of the x-1 hole at the
694 * In the 256k log case, we will read from the beginning to the
695 * end of the log and search for cycle numbers equal to x-1.
696 * We don't worry about the x+1 blocks that we encounter,
697 * because we know that they cannot be the head since the log
700 head_blk
= log_bbnum
;
701 stop_on_cycle
= last_half_cycle
- 1;
704 * In this case we want to find the first block with cycle
705 * number matching last_half_cycle. We expect the log to be
707 * x + 1 ... | x ... | x
708 * The first block with cycle number x (last_half_cycle) will
709 * be where the new head belongs. First we do a binary search
710 * for the first occurrence of last_half_cycle. The binary
711 * search may not be totally accurate, so then we scan back
712 * from there looking for occurrences of last_half_cycle before
713 * us. If that backwards scan wraps around the beginning of
714 * the log, then we look for occurrences of last_half_cycle - 1
715 * at the end of the log. The cases we're looking for look
717 * v binary search stopped here
718 * x + 1 ... | x | x + 1 | x ... | x
719 * ^ but we want to locate this spot
721 * <---------> less than scan distance
722 * x + 1 ... | x ... | x - 1 | x
723 * ^ we want to locate this spot
725 stop_on_cycle
= last_half_cycle
;
726 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
727 &head_blk
, last_half_cycle
)))
732 * Now validate the answer. Scan back some number of maximum possible
733 * blocks and make sure each one has the expected cycle number. The
734 * maximum is determined by the total possible amount of buffering
735 * in the in-core log. The following number can be made tighter if
736 * we actually look at the block size of the filesystem.
738 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
739 if (head_blk
>= num_scan_bblks
) {
741 * We are guaranteed that the entire check can be performed
744 start_blk
= head_blk
- num_scan_bblks
;
745 if ((error
= xlog_find_verify_cycle(log
,
746 start_blk
, num_scan_bblks
,
747 stop_on_cycle
, &new_blk
)))
751 } else { /* need to read 2 parts of log */
753 * We are going to scan backwards in the log in two parts.
754 * First we scan the physical end of the log. In this part
755 * of the log, we are looking for blocks with cycle number
756 * last_half_cycle - 1.
757 * If we find one, then we know that the log starts there, as
758 * we've found a hole that didn't get written in going around
759 * the end of the physical log. The simple case for this is
760 * x + 1 ... | x ... | x - 1 | x
761 * <---------> less than scan distance
762 * If all of the blocks at the end of the log have cycle number
763 * last_half_cycle, then we check the blocks at the start of
764 * the log looking for occurrences of last_half_cycle. If we
765 * find one, then our current estimate for the location of the
766 * first occurrence of last_half_cycle is wrong and we move
767 * back to the hole we've found. This case looks like
768 * x + 1 ... | x | x + 1 | x ...
769 * ^ binary search stopped here
770 * Another case we need to handle that only occurs in 256k
772 * x + 1 ... | x ... | x+1 | x ...
773 * ^ binary search stops here
774 * In a 256k log, the scan at the end of the log will see the
775 * x + 1 blocks. We need to skip past those since that is
776 * certainly not the head of the log. By searching for
777 * last_half_cycle-1 we accomplish that.
779 ASSERT(head_blk
<= INT_MAX
&&
780 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
781 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
782 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
783 num_scan_bblks
- (int)head_blk
,
784 (stop_on_cycle
- 1), &new_blk
)))
792 * Scan beginning of log now. The last part of the physical
793 * log is good. This scan needs to verify that it doesn't find
794 * the last_half_cycle.
797 ASSERT(head_blk
<= INT_MAX
);
798 if ((error
= xlog_find_verify_cycle(log
,
799 start_blk
, (int)head_blk
,
800 stop_on_cycle
, &new_blk
)))
808 * Now we need to make sure head_blk is not pointing to a block in
809 * the middle of a log record.
811 num_scan_bblks
= XLOG_REC_SHIFT(log
);
812 if (head_blk
>= num_scan_bblks
) {
813 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
815 /* start ptr at last block ptr before head_blk */
816 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
817 &head_blk
, 0)) == -1) {
818 error
= XFS_ERROR(EIO
);
824 ASSERT(head_blk
<= INT_MAX
);
825 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
826 &head_blk
, 0)) == -1) {
827 /* We hit the beginning of the log during our search */
828 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
830 ASSERT(start_blk
<= INT_MAX
&&
831 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
832 ASSERT(head_blk
<= INT_MAX
);
833 if ((error
= xlog_find_verify_log_record(log
,
835 (int)head_blk
)) == -1) {
836 error
= XFS_ERROR(EIO
);
840 if (new_blk
!= log_bbnum
)
847 if (head_blk
== log_bbnum
)
848 *return_head_blk
= 0;
850 *return_head_blk
= head_blk
;
852 * When returning here, we have a good block number. Bad block
853 * means that during a previous crash, we didn't have a clean break
854 * from cycle number N to cycle number N-1. In this case, we need
855 * to find the first block with cycle number N-1.
863 xfs_warn(log
->l_mp
, "failed to find log head");
868 * Find the sync block number or the tail of the log.
870 * This will be the block number of the last record to have its
871 * associated buffers synced to disk. Every log record header has
872 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
873 * to get a sync block number. The only concern is to figure out which
874 * log record header to believe.
876 * The following algorithm uses the log record header with the largest
877 * lsn. The entire log record does not need to be valid. We only care
878 * that the header is valid.
880 * We could speed up search by using current head_blk buffer, but it is not
886 xfs_daddr_t
*head_blk
,
887 xfs_daddr_t
*tail_blk
)
889 xlog_rec_header_t
*rhead
;
890 xlog_op_header_t
*op_head
;
891 xfs_caddr_t offset
= NULL
;
894 xfs_daddr_t umount_data_blk
;
895 xfs_daddr_t after_umount_blk
;
902 * Find previous log record
904 if ((error
= xlog_find_head(log
, head_blk
)))
907 bp
= xlog_get_bp(log
, 1);
910 if (*head_blk
== 0) { /* special case */
911 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
915 if (xlog_get_cycle(offset
) == 0) {
917 /* leave all other log inited values alone */
923 * Search backwards looking for log record header block
925 ASSERT(*head_blk
< INT_MAX
);
926 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
927 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
931 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
937 * If we haven't found the log record header block, start looking
938 * again from the end of the physical log. XXXmiken: There should be
939 * a check here to make sure we didn't search more than N blocks in
943 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
944 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
948 if (*(__be32
*)offset
==
949 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
956 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
958 return XFS_ERROR(EIO
);
961 /* find blk_no of tail of log */
962 rhead
= (xlog_rec_header_t
*)offset
;
963 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
966 * Reset log values according to the state of the log when we
967 * crashed. In the case where head_blk == 0, we bump curr_cycle
968 * one because the next write starts a new cycle rather than
969 * continuing the cycle of the last good log record. At this
970 * point we have guaranteed that all partial log records have been
971 * accounted for. Therefore, we know that the last good log record
972 * written was complete and ended exactly on the end boundary
973 * of the physical log.
975 log
->l_prev_block
= i
;
976 log
->l_curr_block
= (int)*head_blk
;
977 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
980 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
981 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
982 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
983 BBTOB(log
->l_curr_block
));
984 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
985 BBTOB(log
->l_curr_block
));
988 * Look for unmount record. If we find it, then we know there
989 * was a clean unmount. Since 'i' could be the last block in
990 * the physical log, we convert to a log block before comparing
993 * Save the current tail lsn to use to pass to
994 * xlog_clear_stale_blocks() below. We won't want to clear the
995 * unmount record if there is one, so we pass the lsn of the
996 * unmount record rather than the block after it.
998 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
999 int h_size
= be32_to_cpu(rhead
->h_size
);
1000 int h_version
= be32_to_cpu(rhead
->h_version
);
1002 if ((h_version
& XLOG_VERSION_2
) &&
1003 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1004 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1005 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1013 after_umount_blk
= (i
+ hblks
+ (int)
1014 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1015 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1016 if (*head_blk
== after_umount_blk
&&
1017 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1018 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1019 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1023 op_head
= (xlog_op_header_t
*)offset
;
1024 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1026 * Set tail and last sync so that newly written
1027 * log records will point recovery to after the
1028 * current unmount record.
1030 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1031 log
->l_curr_cycle
, after_umount_blk
);
1032 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1033 log
->l_curr_cycle
, after_umount_blk
);
1034 *tail_blk
= after_umount_blk
;
1037 * Note that the unmount was clean. If the unmount
1038 * was not clean, we need to know this to rebuild the
1039 * superblock counters from the perag headers if we
1040 * have a filesystem using non-persistent counters.
1042 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1047 * Make sure that there are no blocks in front of the head
1048 * with the same cycle number as the head. This can happen
1049 * because we allow multiple outstanding log writes concurrently,
1050 * and the later writes might make it out before earlier ones.
1052 * We use the lsn from before modifying it so that we'll never
1053 * overwrite the unmount record after a clean unmount.
1055 * Do this only if we are going to recover the filesystem
1057 * NOTE: This used to say "if (!readonly)"
1058 * However on Linux, we can & do recover a read-only filesystem.
1059 * We only skip recovery if NORECOVERY is specified on mount,
1060 * in which case we would not be here.
1062 * But... if the -device- itself is readonly, just skip this.
1063 * We can't recover this device anyway, so it won't matter.
1065 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1066 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1072 xfs_warn(log
->l_mp
, "failed to locate log tail");
1077 * Is the log zeroed at all?
1079 * The last binary search should be changed to perform an X block read
1080 * once X becomes small enough. You can then search linearly through
1081 * the X blocks. This will cut down on the number of reads we need to do.
1083 * If the log is partially zeroed, this routine will pass back the blkno
1084 * of the first block with cycle number 0. It won't have a complete LR
1088 * 0 => the log is completely written to
1089 * -1 => use *blk_no as the first block of the log
1090 * >0 => error has occurred
1095 xfs_daddr_t
*blk_no
)
1099 uint first_cycle
, last_cycle
;
1100 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1101 xfs_daddr_t num_scan_bblks
;
1102 int error
, log_bbnum
= log
->l_logBBsize
;
1106 /* check totally zeroed log */
1107 bp
= xlog_get_bp(log
, 1);
1110 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1114 first_cycle
= xlog_get_cycle(offset
);
1115 if (first_cycle
== 0) { /* completely zeroed log */
1121 /* check partially zeroed log */
1122 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1126 last_cycle
= xlog_get_cycle(offset
);
1127 if (last_cycle
!= 0) { /* log completely written to */
1130 } else if (first_cycle
!= 1) {
1132 * If the cycle of the last block is zero, the cycle of
1133 * the first block must be 1. If it's not, maybe we're
1134 * not looking at a log... Bail out.
1137 "Log inconsistent or not a log (last==0, first!=1)");
1138 return XFS_ERROR(EINVAL
);
1141 /* we have a partially zeroed log */
1142 last_blk
= log_bbnum
-1;
1143 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1147 * Validate the answer. Because there is no way to guarantee that
1148 * the entire log is made up of log records which are the same size,
1149 * we scan over the defined maximum blocks. At this point, the maximum
1150 * is not chosen to mean anything special. XXXmiken
1152 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1153 ASSERT(num_scan_bblks
<= INT_MAX
);
1155 if (last_blk
< num_scan_bblks
)
1156 num_scan_bblks
= last_blk
;
1157 start_blk
= last_blk
- num_scan_bblks
;
1160 * We search for any instances of cycle number 0 that occur before
1161 * our current estimate of the head. What we're trying to detect is
1162 * 1 ... | 0 | 1 | 0...
1163 * ^ binary search ends here
1165 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1166 (int)num_scan_bblks
, 0, &new_blk
)))
1172 * Potentially backup over partial log record write. We don't need
1173 * to search the end of the log because we know it is zero.
1175 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1176 &last_blk
, 0)) == -1) {
1177 error
= XFS_ERROR(EIO
);
1191 * These are simple subroutines used by xlog_clear_stale_blocks() below
1192 * to initialize a buffer full of empty log record headers and write
1193 * them into the log.
1204 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1206 memset(buf
, 0, BBSIZE
);
1207 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1208 recp
->h_cycle
= cpu_to_be32(cycle
);
1209 recp
->h_version
= cpu_to_be32(
1210 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1211 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1212 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1213 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1214 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1218 xlog_write_log_records(
1229 int sectbb
= log
->l_sectBBsize
;
1230 int end_block
= start_block
+ blocks
;
1236 * Greedily allocate a buffer big enough to handle the full
1237 * range of basic blocks to be written. If that fails, try
1238 * a smaller size. We need to be able to write at least a
1239 * log sector, or we're out of luck.
1241 bufblks
= 1 << ffs(blocks
);
1242 while (bufblks
> log
->l_logBBsize
)
1244 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1246 if (bufblks
< sectbb
)
1250 /* We may need to do a read at the start to fill in part of
1251 * the buffer in the starting sector not covered by the first
1254 balign
= round_down(start_block
, sectbb
);
1255 if (balign
!= start_block
) {
1256 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1260 j
= start_block
- balign
;
1263 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1264 int bcount
, endcount
;
1266 bcount
= min(bufblks
, end_block
- start_block
);
1267 endcount
= bcount
- j
;
1269 /* We may need to do a read at the end to fill in part of
1270 * the buffer in the final sector not covered by the write.
1271 * If this is the same sector as the above read, skip it.
1273 ealign
= round_down(end_block
, sectbb
);
1274 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1275 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1276 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1283 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1284 for (; j
< endcount
; j
++) {
1285 xlog_add_record(log
, offset
, cycle
, i
+j
,
1286 tail_cycle
, tail_block
);
1289 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1292 start_block
+= endcount
;
1302 * This routine is called to blow away any incomplete log writes out
1303 * in front of the log head. We do this so that we won't become confused
1304 * if we come up, write only a little bit more, and then crash again.
1305 * If we leave the partial log records out there, this situation could
1306 * cause us to think those partial writes are valid blocks since they
1307 * have the current cycle number. We get rid of them by overwriting them
1308 * with empty log records with the old cycle number rather than the
1311 * The tail lsn is passed in rather than taken from
1312 * the log so that we will not write over the unmount record after a
1313 * clean unmount in a 512 block log. Doing so would leave the log without
1314 * any valid log records in it until a new one was written. If we crashed
1315 * during that time we would not be able to recover.
1318 xlog_clear_stale_blocks(
1322 int tail_cycle
, head_cycle
;
1323 int tail_block
, head_block
;
1324 int tail_distance
, max_distance
;
1328 tail_cycle
= CYCLE_LSN(tail_lsn
);
1329 tail_block
= BLOCK_LSN(tail_lsn
);
1330 head_cycle
= log
->l_curr_cycle
;
1331 head_block
= log
->l_curr_block
;
1334 * Figure out the distance between the new head of the log
1335 * and the tail. We want to write over any blocks beyond the
1336 * head that we may have written just before the crash, but
1337 * we don't want to overwrite the tail of the log.
1339 if (head_cycle
== tail_cycle
) {
1341 * The tail is behind the head in the physical log,
1342 * so the distance from the head to the tail is the
1343 * distance from the head to the end of the log plus
1344 * the distance from the beginning of the log to the
1347 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1348 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1349 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1350 return XFS_ERROR(EFSCORRUPTED
);
1352 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1355 * The head is behind the tail in the physical log,
1356 * so the distance from the head to the tail is just
1357 * the tail block minus the head block.
1359 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1360 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1361 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1362 return XFS_ERROR(EFSCORRUPTED
);
1364 tail_distance
= tail_block
- head_block
;
1368 * If the head is right up against the tail, we can't clear
1371 if (tail_distance
<= 0) {
1372 ASSERT(tail_distance
== 0);
1376 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1378 * Take the smaller of the maximum amount of outstanding I/O
1379 * we could have and the distance to the tail to clear out.
1380 * We take the smaller so that we don't overwrite the tail and
1381 * we don't waste all day writing from the head to the tail
1384 max_distance
= MIN(max_distance
, tail_distance
);
1386 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1388 * We can stomp all the blocks we need to without
1389 * wrapping around the end of the log. Just do it
1390 * in a single write. Use the cycle number of the
1391 * current cycle minus one so that the log will look like:
1394 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1395 head_block
, max_distance
, tail_cycle
,
1401 * We need to wrap around the end of the physical log in
1402 * order to clear all the blocks. Do it in two separate
1403 * I/Os. The first write should be from the head to the
1404 * end of the physical log, and it should use the current
1405 * cycle number minus one just like above.
1407 distance
= log
->l_logBBsize
- head_block
;
1408 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1409 head_block
, distance
, tail_cycle
,
1416 * Now write the blocks at the start of the physical log.
1417 * This writes the remainder of the blocks we want to clear.
1418 * It uses the current cycle number since we're now on the
1419 * same cycle as the head so that we get:
1420 * n ... n ... | n - 1 ...
1421 * ^^^^^ blocks we're writing
1423 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1424 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1425 tail_cycle
, tail_block
);
1433 /******************************************************************************
1435 * Log recover routines
1437 ******************************************************************************
1440 STATIC xlog_recover_t
*
1441 xlog_recover_find_tid(
1442 struct hlist_head
*head
,
1445 xlog_recover_t
*trans
;
1447 hlist_for_each_entry(trans
, head
, r_list
) {
1448 if (trans
->r_log_tid
== tid
)
1455 xlog_recover_new_tid(
1456 struct hlist_head
*head
,
1460 xlog_recover_t
*trans
;
1462 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1463 trans
->r_log_tid
= tid
;
1465 INIT_LIST_HEAD(&trans
->r_itemq
);
1467 INIT_HLIST_NODE(&trans
->r_list
);
1468 hlist_add_head(&trans
->r_list
, head
);
1472 xlog_recover_add_item(
1473 struct list_head
*head
)
1475 xlog_recover_item_t
*item
;
1477 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1478 INIT_LIST_HEAD(&item
->ri_list
);
1479 list_add_tail(&item
->ri_list
, head
);
1483 xlog_recover_add_to_cont_trans(
1485 struct xlog_recover
*trans
,
1489 xlog_recover_item_t
*item
;
1490 xfs_caddr_t ptr
, old_ptr
;
1493 if (list_empty(&trans
->r_itemq
)) {
1494 /* finish copying rest of trans header */
1495 xlog_recover_add_item(&trans
->r_itemq
);
1496 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1497 sizeof(xfs_trans_header_t
) - len
;
1498 memcpy(ptr
, dp
, len
); /* d, s, l */
1501 /* take the tail entry */
1502 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1504 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1505 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1507 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
1508 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1509 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1510 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1511 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1516 * The next region to add is the start of a new region. It could be
1517 * a whole region or it could be the first part of a new region. Because
1518 * of this, the assumption here is that the type and size fields of all
1519 * format structures fit into the first 32 bits of the structure.
1521 * This works because all regions must be 32 bit aligned. Therefore, we
1522 * either have both fields or we have neither field. In the case we have
1523 * neither field, the data part of the region is zero length. We only have
1524 * a log_op_header and can throw away the header since a new one will appear
1525 * later. If we have at least 4 bytes, then we can determine how many regions
1526 * will appear in the current log item.
1529 xlog_recover_add_to_trans(
1531 struct xlog_recover
*trans
,
1535 xfs_inode_log_format_t
*in_f
; /* any will do */
1536 xlog_recover_item_t
*item
;
1541 if (list_empty(&trans
->r_itemq
)) {
1542 /* we need to catch log corruptions here */
1543 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1544 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1547 return XFS_ERROR(EIO
);
1549 if (len
== sizeof(xfs_trans_header_t
))
1550 xlog_recover_add_item(&trans
->r_itemq
);
1551 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1555 ptr
= kmem_alloc(len
, KM_SLEEP
);
1556 memcpy(ptr
, dp
, len
);
1557 in_f
= (xfs_inode_log_format_t
*)ptr
;
1559 /* take the tail entry */
1560 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1561 if (item
->ri_total
!= 0 &&
1562 item
->ri_total
== item
->ri_cnt
) {
1563 /* tail item is in use, get a new one */
1564 xlog_recover_add_item(&trans
->r_itemq
);
1565 item
= list_entry(trans
->r_itemq
.prev
,
1566 xlog_recover_item_t
, ri_list
);
1569 if (item
->ri_total
== 0) { /* first region to be added */
1570 if (in_f
->ilf_size
== 0 ||
1571 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1573 "bad number of regions (%d) in inode log format",
1576 return XFS_ERROR(EIO
);
1579 item
->ri_total
= in_f
->ilf_size
;
1581 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1584 ASSERT(item
->ri_total
> item
->ri_cnt
);
1585 /* Description region is ri_buf[0] */
1586 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1587 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1589 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1594 * Sort the log items in the transaction. Cancelled buffers need
1595 * to be put first so they are processed before any items that might
1596 * modify the buffers. If they are cancelled, then the modifications
1597 * don't need to be replayed.
1600 xlog_recover_reorder_trans(
1602 struct xlog_recover
*trans
,
1605 xlog_recover_item_t
*item
, *n
;
1606 LIST_HEAD(sort_list
);
1608 list_splice_init(&trans
->r_itemq
, &sort_list
);
1609 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1610 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1612 switch (ITEM_TYPE(item
)) {
1614 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1615 trace_xfs_log_recover_item_reorder_head(log
,
1617 list_move(&item
->ri_list
, &trans
->r_itemq
);
1622 case XFS_LI_QUOTAOFF
:
1625 trace_xfs_log_recover_item_reorder_tail(log
,
1627 list_move_tail(&item
->ri_list
, &trans
->r_itemq
);
1631 "%s: unrecognized type of log operation",
1634 return XFS_ERROR(EIO
);
1637 ASSERT(list_empty(&sort_list
));
1642 * Build up the table of buf cancel records so that we don't replay
1643 * cancelled data in the second pass. For buffer records that are
1644 * not cancel records, there is nothing to do here so we just return.
1646 * If we get a cancel record which is already in the table, this indicates
1647 * that the buffer was cancelled multiple times. In order to ensure
1648 * that during pass 2 we keep the record in the table until we reach its
1649 * last occurrence in the log, we keep a reference count in the cancel
1650 * record in the table to tell us how many times we expect to see this
1651 * record during the second pass.
1654 xlog_recover_buffer_pass1(
1656 struct xlog_recover_item
*item
)
1658 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1659 struct list_head
*bucket
;
1660 struct xfs_buf_cancel
*bcp
;
1663 * If this isn't a cancel buffer item, then just return.
1665 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1666 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1671 * Insert an xfs_buf_cancel record into the hash table of them.
1672 * If there is already an identical record, bump its reference count.
1674 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1675 list_for_each_entry(bcp
, bucket
, bc_list
) {
1676 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1677 bcp
->bc_len
== buf_f
->blf_len
) {
1679 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1684 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1685 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1686 bcp
->bc_len
= buf_f
->blf_len
;
1687 bcp
->bc_refcount
= 1;
1688 list_add_tail(&bcp
->bc_list
, bucket
);
1690 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1695 * Check to see whether the buffer being recovered has a corresponding
1696 * entry in the buffer cancel record table. If it does then return 1
1697 * so that it will be cancelled, otherwise return 0. If the buffer is
1698 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1699 * the refcount on the entry in the table and remove it from the table
1700 * if this is the last reference.
1702 * We remove the cancel record from the table when we encounter its
1703 * last occurrence in the log so that if the same buffer is re-used
1704 * again after its last cancellation we actually replay the changes
1705 * made at that point.
1708 xlog_check_buffer_cancelled(
1714 struct list_head
*bucket
;
1715 struct xfs_buf_cancel
*bcp
;
1717 if (log
->l_buf_cancel_table
== NULL
) {
1719 * There is nothing in the table built in pass one,
1720 * so this buffer must not be cancelled.
1722 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1727 * Search for an entry in the cancel table that matches our buffer.
1729 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1730 list_for_each_entry(bcp
, bucket
, bc_list
) {
1731 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1736 * We didn't find a corresponding entry in the table, so return 0 so
1737 * that the buffer is NOT cancelled.
1739 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1744 * We've go a match, so return 1 so that the recovery of this buffer
1745 * is cancelled. If this buffer is actually a buffer cancel log
1746 * item, then decrement the refcount on the one in the table and
1747 * remove it if this is the last reference.
1749 if (flags
& XFS_BLF_CANCEL
) {
1750 if (--bcp
->bc_refcount
== 0) {
1751 list_del(&bcp
->bc_list
);
1759 * Perform recovery for a buffer full of inodes. In these buffers, the only
1760 * data which should be recovered is that which corresponds to the
1761 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1762 * data for the inodes is always logged through the inodes themselves rather
1763 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1765 * The only time when buffers full of inodes are fully recovered is when the
1766 * buffer is full of newly allocated inodes. In this case the buffer will
1767 * not be marked as an inode buffer and so will be sent to
1768 * xlog_recover_do_reg_buffer() below during recovery.
1771 xlog_recover_do_inode_buffer(
1772 struct xfs_mount
*mp
,
1773 xlog_recover_item_t
*item
,
1775 xfs_buf_log_format_t
*buf_f
)
1781 int reg_buf_offset
= 0;
1782 int reg_buf_bytes
= 0;
1783 int next_unlinked_offset
;
1785 xfs_agino_t
*logged_nextp
;
1786 xfs_agino_t
*buffer_nextp
;
1788 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1790 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1791 for (i
= 0; i
< inodes_per_buf
; i
++) {
1792 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1793 offsetof(xfs_dinode_t
, di_next_unlinked
);
1795 while (next_unlinked_offset
>=
1796 (reg_buf_offset
+ reg_buf_bytes
)) {
1798 * The next di_next_unlinked field is beyond
1799 * the current logged region. Find the next
1800 * logged region that contains or is beyond
1801 * the current di_next_unlinked field.
1804 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1805 buf_f
->blf_map_size
, bit
);
1808 * If there are no more logged regions in the
1809 * buffer, then we're done.
1814 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1815 buf_f
->blf_map_size
, bit
);
1817 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1818 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1823 * If the current logged region starts after the current
1824 * di_next_unlinked field, then move on to the next
1825 * di_next_unlinked field.
1827 if (next_unlinked_offset
< reg_buf_offset
)
1830 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1831 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1832 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1833 BBTOB(bp
->b_io_length
));
1836 * The current logged region contains a copy of the
1837 * current di_next_unlinked field. Extract its value
1838 * and copy it to the buffer copy.
1840 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1841 next_unlinked_offset
- reg_buf_offset
;
1842 if (unlikely(*logged_nextp
== 0)) {
1844 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1845 "Trying to replay bad (0) inode di_next_unlinked field.",
1847 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1848 XFS_ERRLEVEL_LOW
, mp
);
1849 return XFS_ERROR(EFSCORRUPTED
);
1852 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1853 next_unlinked_offset
);
1854 *buffer_nextp
= *logged_nextp
;
1861 * Perform a 'normal' buffer recovery. Each logged region of the
1862 * buffer should be copied over the corresponding region in the
1863 * given buffer. The bitmap in the buf log format structure indicates
1864 * where to place the logged data.
1867 xlog_recover_do_reg_buffer(
1868 struct xfs_mount
*mp
,
1869 xlog_recover_item_t
*item
,
1871 xfs_buf_log_format_t
*buf_f
)
1878 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
1881 i
= 1; /* 0 is the buf format structure */
1883 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1884 buf_f
->blf_map_size
, bit
);
1887 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1888 buf_f
->blf_map_size
, bit
);
1890 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1891 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
1892 ASSERT(BBTOB(bp
->b_io_length
) >=
1893 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
1896 * Do a sanity check if this is a dquot buffer. Just checking
1897 * the first dquot in the buffer should do. XXXThis is
1898 * probably a good thing to do for other buf types also.
1901 if (buf_f
->blf_flags
&
1902 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
1903 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1905 "XFS: NULL dquot in %s.", __func__
);
1908 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
1910 "XFS: dquot too small (%d) in %s.",
1911 item
->ri_buf
[i
].i_len
, __func__
);
1914 error
= xfs_qm_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
1915 -1, 0, XFS_QMOPT_DOWARN
,
1916 "dquot_buf_recover");
1921 memcpy(xfs_buf_offset(bp
,
1922 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
1923 item
->ri_buf
[i
].i_addr
, /* source */
1924 nbits
<<XFS_BLF_SHIFT
); /* length */
1930 /* Shouldn't be any more regions */
1931 ASSERT(i
== item
->ri_total
);
1933 switch (buf_f
->blf_flags
& XFS_BLF_TYPE_MASK
) {
1934 case XFS_BLF_BTREE_BUF
:
1935 switch (be32_to_cpu(*(__be32
*)bp
->b_addr
)) {
1936 case XFS_ABTB_CRC_MAGIC
:
1937 case XFS_ABTC_CRC_MAGIC
:
1938 case XFS_ABTB_MAGIC
:
1939 case XFS_ABTC_MAGIC
:
1940 bp
->b_ops
= &xfs_allocbt_buf_ops
;
1942 case XFS_IBT_CRC_MAGIC
:
1944 bp
->b_ops
= &xfs_inobt_buf_ops
;
1946 case XFS_BMAP_CRC_MAGIC
:
1947 case XFS_BMAP_MAGIC
:
1948 bp
->b_ops
= &xfs_bmbt_buf_ops
;
1951 xfs_warn(mp
, "Bad btree block magic!");
1962 * Do some primitive error checking on ondisk dquot data structures.
1966 struct xfs_mount
*mp
,
1967 xfs_disk_dquot_t
*ddq
,
1969 uint type
, /* used only when IO_dorepair is true */
1973 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1977 * We can encounter an uninitialized dquot buffer for 2 reasons:
1978 * 1. If we crash while deleting the quotainode(s), and those blks got
1979 * used for user data. This is because we take the path of regular
1980 * file deletion; however, the size field of quotainodes is never
1981 * updated, so all the tricks that we play in itruncate_finish
1982 * don't quite matter.
1984 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1985 * But the allocation will be replayed so we'll end up with an
1986 * uninitialized quota block.
1988 * This is all fine; things are still consistent, and we haven't lost
1989 * any quota information. Just don't complain about bad dquot blks.
1991 if (ddq
->d_magic
!= cpu_to_be16(XFS_DQUOT_MAGIC
)) {
1992 if (flags
& XFS_QMOPT_DOWARN
)
1994 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1995 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
1998 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
1999 if (flags
& XFS_QMOPT_DOWARN
)
2001 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2002 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2006 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2007 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2008 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2009 if (flags
& XFS_QMOPT_DOWARN
)
2011 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2012 str
, id
, ddq
->d_flags
);
2016 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2017 if (flags
& XFS_QMOPT_DOWARN
)
2019 "%s : ondisk-dquot 0x%p, ID mismatch: "
2020 "0x%x expected, found id 0x%x",
2021 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2025 if (!errs
&& ddq
->d_id
) {
2026 if (ddq
->d_blk_softlimit
&&
2027 be64_to_cpu(ddq
->d_bcount
) >
2028 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2029 if (!ddq
->d_btimer
) {
2030 if (flags
& XFS_QMOPT_DOWARN
)
2032 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
2033 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2037 if (ddq
->d_ino_softlimit
&&
2038 be64_to_cpu(ddq
->d_icount
) >
2039 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2040 if (!ddq
->d_itimer
) {
2041 if (flags
& XFS_QMOPT_DOWARN
)
2043 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2044 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2048 if (ddq
->d_rtb_softlimit
&&
2049 be64_to_cpu(ddq
->d_rtbcount
) >
2050 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2051 if (!ddq
->d_rtbtimer
) {
2052 if (flags
& XFS_QMOPT_DOWARN
)
2054 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2055 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2061 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2064 if (flags
& XFS_QMOPT_DOWARN
)
2065 xfs_notice(mp
, "Re-initializing dquot ID 0x%x", id
);
2068 * Typically, a repair is only requested by quotacheck.
2071 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2072 memset(d
, 0, sizeof(xfs_dqblk_t
));
2074 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2075 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2076 d
->dd_diskdq
.d_flags
= type
;
2077 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2083 * Perform a dquot buffer recovery.
2084 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2085 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2086 * Else, treat it as a regular buffer and do recovery.
2089 xlog_recover_do_dquot_buffer(
2090 struct xfs_mount
*mp
,
2092 struct xlog_recover_item
*item
,
2094 struct xfs_buf_log_format
*buf_f
)
2098 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2101 * Filesystems are required to send in quota flags at mount time.
2103 if (mp
->m_qflags
== 0) {
2108 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2109 type
|= XFS_DQ_USER
;
2110 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2111 type
|= XFS_DQ_PROJ
;
2112 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2113 type
|= XFS_DQ_GROUP
;
2115 * This type of quotas was turned off, so ignore this buffer
2117 if (log
->l_quotaoffs_flag
& type
)
2120 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2124 * This routine replays a modification made to a buffer at runtime.
2125 * There are actually two types of buffer, regular and inode, which
2126 * are handled differently. Inode buffers are handled differently
2127 * in that we only recover a specific set of data from them, namely
2128 * the inode di_next_unlinked fields. This is because all other inode
2129 * data is actually logged via inode records and any data we replay
2130 * here which overlaps that may be stale.
2132 * When meta-data buffers are freed at run time we log a buffer item
2133 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2134 * of the buffer in the log should not be replayed at recovery time.
2135 * This is so that if the blocks covered by the buffer are reused for
2136 * file data before we crash we don't end up replaying old, freed
2137 * meta-data into a user's file.
2139 * To handle the cancellation of buffer log items, we make two passes
2140 * over the log during recovery. During the first we build a table of
2141 * those buffers which have been cancelled, and during the second we
2142 * only replay those buffers which do not have corresponding cancel
2143 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2144 * for more details on the implementation of the table of cancel records.
2147 xlog_recover_buffer_pass2(
2149 struct list_head
*buffer_list
,
2150 struct xlog_recover_item
*item
)
2152 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2153 xfs_mount_t
*mp
= log
->l_mp
;
2159 * In this pass we only want to recover all the buffers which have
2160 * not been cancelled and are not cancellation buffers themselves.
2162 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2163 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2164 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2168 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2171 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2172 buf_flags
|= XBF_UNMAPPED
;
2174 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2177 return XFS_ERROR(ENOMEM
);
2178 error
= bp
->b_error
;
2180 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2185 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2186 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2187 } else if (buf_f
->blf_flags
&
2188 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2189 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2191 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2194 return XFS_ERROR(error
);
2197 * Perform delayed write on the buffer. Asynchronous writes will be
2198 * slower when taking into account all the buffers to be flushed.
2200 * Also make sure that only inode buffers with good sizes stay in
2201 * the buffer cache. The kernel moves inodes in buffers of 1 block
2202 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2203 * buffers in the log can be a different size if the log was generated
2204 * by an older kernel using unclustered inode buffers or a newer kernel
2205 * running with a different inode cluster size. Regardless, if the
2206 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2207 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2208 * the buffer out of the buffer cache so that the buffer won't
2209 * overlap with future reads of those inodes.
2211 if (XFS_DINODE_MAGIC
==
2212 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2213 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2214 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2216 error
= xfs_bwrite(bp
);
2218 ASSERT(bp
->b_target
->bt_mount
== mp
);
2219 bp
->b_iodone
= xlog_recover_iodone
;
2220 xfs_buf_delwri_queue(bp
, buffer_list
);
2228 xlog_recover_inode_pass2(
2230 struct list_head
*buffer_list
,
2231 struct xlog_recover_item
*item
)
2233 xfs_inode_log_format_t
*in_f
;
2234 xfs_mount_t
*mp
= log
->l_mp
;
2243 xfs_icdinode_t
*dicp
;
2246 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2247 in_f
= item
->ri_buf
[0].i_addr
;
2249 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2251 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2257 * Inode buffers can be freed, look out for it,
2258 * and do not replay the inode.
2260 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2261 in_f
->ilf_len
, 0)) {
2263 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2266 trace_xfs_log_recover_inode_recover(log
, in_f
);
2268 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2274 error
= bp
->b_error
;
2276 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2280 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2281 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2284 * Make sure the place we're flushing out to really looks
2287 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2290 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2291 __func__
, dip
, bp
, in_f
->ilf_ino
);
2292 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2293 XFS_ERRLEVEL_LOW
, mp
);
2294 error
= EFSCORRUPTED
;
2297 dicp
= item
->ri_buf
[1].i_addr
;
2298 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2301 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2302 __func__
, item
, in_f
->ilf_ino
);
2303 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2304 XFS_ERRLEVEL_LOW
, mp
);
2305 error
= EFSCORRUPTED
;
2309 /* Skip replay when the on disk inode is newer than the log one */
2310 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2312 * Deal with the wrap case, DI_MAX_FLUSH is less
2313 * than smaller numbers
2315 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2316 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2320 trace_xfs_log_recover_inode_skip(log
, in_f
);
2325 /* Take the opportunity to reset the flush iteration count */
2326 dicp
->di_flushiter
= 0;
2328 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2329 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2330 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2331 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2332 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2335 "%s: Bad regular inode log record, rec ptr 0x%p, "
2336 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2337 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2338 error
= EFSCORRUPTED
;
2341 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2342 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2343 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2344 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2345 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2346 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2349 "%s: Bad dir inode log record, rec ptr 0x%p, "
2350 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2351 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2352 error
= EFSCORRUPTED
;
2356 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2357 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2358 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2361 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2362 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2363 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2364 dicp
->di_nextents
+ dicp
->di_anextents
,
2366 error
= EFSCORRUPTED
;
2369 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2370 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2371 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2374 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2375 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2376 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2377 error
= EFSCORRUPTED
;
2380 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
))) {
2381 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2382 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2385 "%s: Bad inode log record length %d, rec ptr 0x%p",
2386 __func__
, item
->ri_buf
[1].i_len
, item
);
2387 error
= EFSCORRUPTED
;
2391 /* The core is in in-core format */
2392 xfs_dinode_to_disk(dip
, item
->ri_buf
[1].i_addr
);
2394 /* the rest is in on-disk format */
2395 if (item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
)) {
2396 memcpy((xfs_caddr_t
) dip
+ sizeof(struct xfs_icdinode
),
2397 item
->ri_buf
[1].i_addr
+ sizeof(struct xfs_icdinode
),
2398 item
->ri_buf
[1].i_len
- sizeof(struct xfs_icdinode
));
2401 fields
= in_f
->ilf_fields
;
2402 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2404 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2407 memcpy(XFS_DFORK_DPTR(dip
),
2408 &in_f
->ilf_u
.ilfu_uuid
,
2413 if (in_f
->ilf_size
== 2)
2414 goto write_inode_buffer
;
2415 len
= item
->ri_buf
[2].i_len
;
2416 src
= item
->ri_buf
[2].i_addr
;
2417 ASSERT(in_f
->ilf_size
<= 4);
2418 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2419 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2420 (len
== in_f
->ilf_dsize
));
2422 switch (fields
& XFS_ILOG_DFORK
) {
2423 case XFS_ILOG_DDATA
:
2425 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2428 case XFS_ILOG_DBROOT
:
2429 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2430 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2431 XFS_DFORK_DSIZE(dip
, mp
));
2436 * There are no data fork flags set.
2438 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2443 * If we logged any attribute data, recover it. There may or
2444 * may not have been any other non-core data logged in this
2447 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2448 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2453 len
= item
->ri_buf
[attr_index
].i_len
;
2454 src
= item
->ri_buf
[attr_index
].i_addr
;
2455 ASSERT(len
== in_f
->ilf_asize
);
2457 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2458 case XFS_ILOG_ADATA
:
2460 dest
= XFS_DFORK_APTR(dip
);
2461 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2462 memcpy(dest
, src
, len
);
2465 case XFS_ILOG_ABROOT
:
2466 dest
= XFS_DFORK_APTR(dip
);
2467 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2468 len
, (xfs_bmdr_block_t
*)dest
,
2469 XFS_DFORK_ASIZE(dip
, mp
));
2473 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2482 ASSERT(bp
->b_target
->bt_mount
== mp
);
2483 bp
->b_iodone
= xlog_recover_iodone
;
2484 xfs_buf_delwri_queue(bp
, buffer_list
);
2489 return XFS_ERROR(error
);
2493 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2494 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2498 xlog_recover_quotaoff_pass1(
2500 struct xlog_recover_item
*item
)
2502 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2506 * The logitem format's flag tells us if this was user quotaoff,
2507 * group/project quotaoff or both.
2509 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2510 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2511 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2512 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2513 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2514 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2520 * Recover a dquot record
2523 xlog_recover_dquot_pass2(
2525 struct list_head
*buffer_list
,
2526 struct xlog_recover_item
*item
)
2528 xfs_mount_t
*mp
= log
->l_mp
;
2530 struct xfs_disk_dquot
*ddq
, *recddq
;
2532 xfs_dq_logformat_t
*dq_f
;
2537 * Filesystems are required to send in quota flags at mount time.
2539 if (mp
->m_qflags
== 0)
2542 recddq
= item
->ri_buf
[1].i_addr
;
2543 if (recddq
== NULL
) {
2544 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2545 return XFS_ERROR(EIO
);
2547 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2548 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2549 item
->ri_buf
[1].i_len
, __func__
);
2550 return XFS_ERROR(EIO
);
2554 * This type of quotas was turned off, so ignore this record.
2556 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2558 if (log
->l_quotaoffs_flag
& type
)
2562 * At this point we know that quota was _not_ turned off.
2563 * Since the mount flags are not indicating to us otherwise, this
2564 * must mean that quota is on, and the dquot needs to be replayed.
2565 * Remember that we may not have fully recovered the superblock yet,
2566 * so we can't do the usual trick of looking at the SB quota bits.
2568 * The other possibility, of course, is that the quota subsystem was
2569 * removed since the last mount - ENOSYS.
2571 dq_f
= item
->ri_buf
[0].i_addr
;
2573 error
= xfs_qm_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2574 "xlog_recover_dquot_pass2 (log copy)");
2576 return XFS_ERROR(EIO
);
2577 ASSERT(dq_f
->qlf_len
== 1);
2579 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
2580 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
2586 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2589 * At least the magic num portion should be on disk because this
2590 * was among a chunk of dquots created earlier, and we did some
2591 * minimal initialization then.
2593 error
= xfs_qm_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2594 "xlog_recover_dquot_pass2");
2597 return XFS_ERROR(EIO
);
2600 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2602 ASSERT(dq_f
->qlf_size
== 2);
2603 ASSERT(bp
->b_target
->bt_mount
== mp
);
2604 bp
->b_iodone
= xlog_recover_iodone
;
2605 xfs_buf_delwri_queue(bp
, buffer_list
);
2612 * This routine is called to create an in-core extent free intent
2613 * item from the efi format structure which was logged on disk.
2614 * It allocates an in-core efi, copies the extents from the format
2615 * structure into it, and adds the efi to the AIL with the given
2619 xlog_recover_efi_pass2(
2621 struct xlog_recover_item
*item
,
2625 xfs_mount_t
*mp
= log
->l_mp
;
2626 xfs_efi_log_item_t
*efip
;
2627 xfs_efi_log_format_t
*efi_formatp
;
2629 efi_formatp
= item
->ri_buf
[0].i_addr
;
2631 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2632 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2633 &(efip
->efi_format
)))) {
2634 xfs_efi_item_free(efip
);
2637 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
2639 spin_lock(&log
->l_ailp
->xa_lock
);
2641 * xfs_trans_ail_update() drops the AIL lock.
2643 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
2649 * This routine is called when an efd format structure is found in
2650 * a committed transaction in the log. It's purpose is to cancel
2651 * the corresponding efi if it was still in the log. To do this
2652 * it searches the AIL for the efi with an id equal to that in the
2653 * efd format structure. If we find it, we remove the efi from the
2657 xlog_recover_efd_pass2(
2659 struct xlog_recover_item
*item
)
2661 xfs_efd_log_format_t
*efd_formatp
;
2662 xfs_efi_log_item_t
*efip
= NULL
;
2663 xfs_log_item_t
*lip
;
2665 struct xfs_ail_cursor cur
;
2666 struct xfs_ail
*ailp
= log
->l_ailp
;
2668 efd_formatp
= item
->ri_buf
[0].i_addr
;
2669 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2670 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2671 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2672 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2673 efi_id
= efd_formatp
->efd_efi_id
;
2676 * Search for the efi with the id in the efd format structure
2679 spin_lock(&ailp
->xa_lock
);
2680 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2681 while (lip
!= NULL
) {
2682 if (lip
->li_type
== XFS_LI_EFI
) {
2683 efip
= (xfs_efi_log_item_t
*)lip
;
2684 if (efip
->efi_format
.efi_id
== efi_id
) {
2686 * xfs_trans_ail_delete() drops the
2689 xfs_trans_ail_delete(ailp
, lip
,
2690 SHUTDOWN_CORRUPT_INCORE
);
2691 xfs_efi_item_free(efip
);
2692 spin_lock(&ailp
->xa_lock
);
2696 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2698 xfs_trans_ail_cursor_done(ailp
, &cur
);
2699 spin_unlock(&ailp
->xa_lock
);
2705 * Free up any resources allocated by the transaction
2707 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2710 xlog_recover_free_trans(
2711 struct xlog_recover
*trans
)
2713 xlog_recover_item_t
*item
, *n
;
2716 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2717 /* Free the regions in the item. */
2718 list_del(&item
->ri_list
);
2719 for (i
= 0; i
< item
->ri_cnt
; i
++)
2720 kmem_free(item
->ri_buf
[i
].i_addr
);
2721 /* Free the item itself */
2722 kmem_free(item
->ri_buf
);
2725 /* Free the transaction recover structure */
2730 xlog_recover_commit_pass1(
2732 struct xlog_recover
*trans
,
2733 struct xlog_recover_item
*item
)
2735 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
2737 switch (ITEM_TYPE(item
)) {
2739 return xlog_recover_buffer_pass1(log
, item
);
2740 case XFS_LI_QUOTAOFF
:
2741 return xlog_recover_quotaoff_pass1(log
, item
);
2746 /* nothing to do in pass 1 */
2749 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
2750 __func__
, ITEM_TYPE(item
));
2752 return XFS_ERROR(EIO
);
2757 xlog_recover_commit_pass2(
2759 struct xlog_recover
*trans
,
2760 struct list_head
*buffer_list
,
2761 struct xlog_recover_item
*item
)
2763 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
2765 switch (ITEM_TYPE(item
)) {
2767 return xlog_recover_buffer_pass2(log
, buffer_list
, item
);
2769 return xlog_recover_inode_pass2(log
, buffer_list
, item
);
2771 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
2773 return xlog_recover_efd_pass2(log
, item
);
2775 return xlog_recover_dquot_pass2(log
, buffer_list
, item
);
2776 case XFS_LI_QUOTAOFF
:
2777 /* nothing to do in pass2 */
2780 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
2781 __func__
, ITEM_TYPE(item
));
2783 return XFS_ERROR(EIO
);
2788 * Perform the transaction.
2790 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2791 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2794 xlog_recover_commit_trans(
2796 struct xlog_recover
*trans
,
2799 int error
= 0, error2
;
2800 xlog_recover_item_t
*item
;
2801 LIST_HEAD (buffer_list
);
2803 hlist_del(&trans
->r_list
);
2805 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
2809 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
2811 case XLOG_RECOVER_PASS1
:
2812 error
= xlog_recover_commit_pass1(log
, trans
, item
);
2814 case XLOG_RECOVER_PASS2
:
2815 error
= xlog_recover_commit_pass2(log
, trans
,
2816 &buffer_list
, item
);
2826 xlog_recover_free_trans(trans
);
2829 error2
= xfs_buf_delwri_submit(&buffer_list
);
2830 return error
? error
: error2
;
2834 xlog_recover_unmount_trans(
2836 struct xlog_recover
*trans
)
2838 /* Do nothing now */
2839 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
2844 * There are two valid states of the r_state field. 0 indicates that the
2845 * transaction structure is in a normal state. We have either seen the
2846 * start of the transaction or the last operation we added was not a partial
2847 * operation. If the last operation we added to the transaction was a
2848 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2850 * NOTE: skip LRs with 0 data length.
2853 xlog_recover_process_data(
2855 struct hlist_head rhash
[],
2856 struct xlog_rec_header
*rhead
,
2862 xlog_op_header_t
*ohead
;
2863 xlog_recover_t
*trans
;
2869 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2870 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2872 /* check the log format matches our own - else we can't recover */
2873 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2874 return (XFS_ERROR(EIO
));
2876 while ((dp
< lp
) && num_logops
) {
2877 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2878 ohead
= (xlog_op_header_t
*)dp
;
2879 dp
+= sizeof(xlog_op_header_t
);
2880 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2881 ohead
->oh_clientid
!= XFS_LOG
) {
2882 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
2883 __func__
, ohead
->oh_clientid
);
2885 return (XFS_ERROR(EIO
));
2887 tid
= be32_to_cpu(ohead
->oh_tid
);
2888 hash
= XLOG_RHASH(tid
);
2889 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
2890 if (trans
== NULL
) { /* not found; add new tid */
2891 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2892 xlog_recover_new_tid(&rhash
[hash
], tid
,
2893 be64_to_cpu(rhead
->h_lsn
));
2895 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2896 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
2897 __func__
, be32_to_cpu(ohead
->oh_len
));
2899 return (XFS_ERROR(EIO
));
2901 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2902 if (flags
& XLOG_WAS_CONT_TRANS
)
2903 flags
&= ~XLOG_CONTINUE_TRANS
;
2905 case XLOG_COMMIT_TRANS
:
2906 error
= xlog_recover_commit_trans(log
,
2909 case XLOG_UNMOUNT_TRANS
:
2910 error
= xlog_recover_unmount_trans(log
, trans
);
2912 case XLOG_WAS_CONT_TRANS
:
2913 error
= xlog_recover_add_to_cont_trans(log
,
2915 be32_to_cpu(ohead
->oh_len
));
2917 case XLOG_START_TRANS
:
2918 xfs_warn(log
->l_mp
, "%s: bad transaction",
2921 error
= XFS_ERROR(EIO
);
2924 case XLOG_CONTINUE_TRANS
:
2925 error
= xlog_recover_add_to_trans(log
, trans
,
2926 dp
, be32_to_cpu(ohead
->oh_len
));
2929 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
2932 error
= XFS_ERROR(EIO
);
2938 dp
+= be32_to_cpu(ohead
->oh_len
);
2945 * Process an extent free intent item that was recovered from
2946 * the log. We need to free the extents that it describes.
2949 xlog_recover_process_efi(
2951 xfs_efi_log_item_t
*efip
)
2953 xfs_efd_log_item_t
*efdp
;
2958 xfs_fsblock_t startblock_fsb
;
2960 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
2963 * First check the validity of the extents described by the
2964 * EFI. If any are bad, then assume that all are bad and
2965 * just toss the EFI.
2967 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2968 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2969 startblock_fsb
= XFS_BB_TO_FSB(mp
,
2970 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
2971 if ((startblock_fsb
== 0) ||
2972 (extp
->ext_len
== 0) ||
2973 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
2974 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
2976 * This will pull the EFI from the AIL and
2977 * free the memory associated with it.
2979 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
2980 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
2981 return XFS_ERROR(EIO
);
2985 tp
= xfs_trans_alloc(mp
, 0);
2986 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
2989 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
2991 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2992 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2993 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
2996 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3000 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3001 error
= xfs_trans_commit(tp
, 0);
3005 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3010 * When this is called, all of the EFIs which did not have
3011 * corresponding EFDs should be in the AIL. What we do now
3012 * is free the extents associated with each one.
3014 * Since we process the EFIs in normal transactions, they
3015 * will be removed at some point after the commit. This prevents
3016 * us from just walking down the list processing each one.
3017 * We'll use a flag in the EFI to skip those that we've already
3018 * processed and use the AIL iteration mechanism's generation
3019 * count to try to speed this up at least a bit.
3021 * When we start, we know that the EFIs are the only things in
3022 * the AIL. As we process them, however, other items are added
3023 * to the AIL. Since everything added to the AIL must come after
3024 * everything already in the AIL, we stop processing as soon as
3025 * we see something other than an EFI in the AIL.
3028 xlog_recover_process_efis(
3031 xfs_log_item_t
*lip
;
3032 xfs_efi_log_item_t
*efip
;
3034 struct xfs_ail_cursor cur
;
3035 struct xfs_ail
*ailp
;
3038 spin_lock(&ailp
->xa_lock
);
3039 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3040 while (lip
!= NULL
) {
3042 * We're done when we see something other than an EFI.
3043 * There should be no EFIs left in the AIL now.
3045 if (lip
->li_type
!= XFS_LI_EFI
) {
3047 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3048 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3054 * Skip EFIs that we've already processed.
3056 efip
= (xfs_efi_log_item_t
*)lip
;
3057 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3058 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3062 spin_unlock(&ailp
->xa_lock
);
3063 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3064 spin_lock(&ailp
->xa_lock
);
3067 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3070 xfs_trans_ail_cursor_done(ailp
, &cur
);
3071 spin_unlock(&ailp
->xa_lock
);
3076 * This routine performs a transaction to null out a bad inode pointer
3077 * in an agi unlinked inode hash bucket.
3080 xlog_recover_clear_agi_bucket(
3082 xfs_agnumber_t agno
,
3091 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3092 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3097 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3101 agi
= XFS_BUF_TO_AGI(agibp
);
3102 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3103 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3104 (sizeof(xfs_agino_t
) * bucket
);
3105 xfs_trans_log_buf(tp
, agibp
, offset
,
3106 (offset
+ sizeof(xfs_agino_t
) - 1));
3108 error
= xfs_trans_commit(tp
, 0);
3114 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3116 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3121 xlog_recover_process_one_iunlink(
3122 struct xfs_mount
*mp
,
3123 xfs_agnumber_t agno
,
3127 struct xfs_buf
*ibp
;
3128 struct xfs_dinode
*dip
;
3129 struct xfs_inode
*ip
;
3133 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3134 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3139 * Get the on disk inode to find the next inode in the bucket.
3141 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3145 ASSERT(ip
->i_d
.di_nlink
== 0);
3146 ASSERT(ip
->i_d
.di_mode
!= 0);
3148 /* setup for the next pass */
3149 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3153 * Prevent any DMAPI event from being sent when the reference on
3154 * the inode is dropped.
3156 ip
->i_d
.di_dmevmask
= 0;
3165 * We can't read in the inode this bucket points to, or this inode
3166 * is messed up. Just ditch this bucket of inodes. We will lose
3167 * some inodes and space, but at least we won't hang.
3169 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3170 * clear the inode pointer in the bucket.
3172 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3177 * xlog_iunlink_recover
3179 * This is called during recovery to process any inodes which
3180 * we unlinked but not freed when the system crashed. These
3181 * inodes will be on the lists in the AGI blocks. What we do
3182 * here is scan all the AGIs and fully truncate and free any
3183 * inodes found on the lists. Each inode is removed from the
3184 * lists when it has been fully truncated and is freed. The
3185 * freeing of the inode and its removal from the list must be
3189 xlog_recover_process_iunlinks(
3193 xfs_agnumber_t agno
;
3204 * Prevent any DMAPI event from being sent while in this function.
3206 mp_dmevmask
= mp
->m_dmevmask
;
3209 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3211 * Find the agi for this ag.
3213 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3216 * AGI is b0rked. Don't process it.
3218 * We should probably mark the filesystem as corrupt
3219 * after we've recovered all the ag's we can....
3224 * Unlock the buffer so that it can be acquired in the normal
3225 * course of the transaction to truncate and free each inode.
3226 * Because we are not racing with anyone else here for the AGI
3227 * buffer, we don't even need to hold it locked to read the
3228 * initial unlinked bucket entries out of the buffer. We keep
3229 * buffer reference though, so that it stays pinned in memory
3230 * while we need the buffer.
3232 agi
= XFS_BUF_TO_AGI(agibp
);
3233 xfs_buf_unlock(agibp
);
3235 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3236 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3237 while (agino
!= NULLAGINO
) {
3238 agino
= xlog_recover_process_one_iunlink(mp
,
3239 agno
, agino
, bucket
);
3242 xfs_buf_rele(agibp
);
3245 mp
->m_dmevmask
= mp_dmevmask
;
3249 * Upack the log buffer data and crc check it. If the check fails, issue a
3250 * warning if and only if the CRC in the header is non-zero. This makes the
3251 * check an advisory warning, and the zero CRC check will prevent failure
3252 * warnings from being emitted when upgrading the kernel from one that does not
3253 * add CRCs by default.
3255 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3256 * corruption failure
3259 xlog_unpack_data_crc(
3260 struct xlog_rec_header
*rhead
,
3266 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
3267 if (crc
!= rhead
->h_crc
) {
3268 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
3269 xfs_alert(log
->l_mp
,
3270 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
3271 le32_to_cpu(rhead
->h_crc
),
3273 xfs_hex_dump(dp
, 32);
3277 * If we've detected a log record corruption, then we can't
3278 * recover past this point. Abort recovery if we are enforcing
3279 * CRC protection by punting an error back up the stack.
3281 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
3282 return EFSCORRUPTED
;
3290 struct xlog_rec_header
*rhead
,
3297 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
3301 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3302 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3303 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3307 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3308 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3309 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3310 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3311 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3312 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3321 xlog_valid_rec_header(
3323 struct xlog_rec_header
*rhead
,
3328 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
3329 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3330 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3331 return XFS_ERROR(EFSCORRUPTED
);
3334 (!rhead
->h_version
||
3335 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3336 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
3337 __func__
, be32_to_cpu(rhead
->h_version
));
3338 return XFS_ERROR(EIO
);
3341 /* LR body must have data or it wouldn't have been written */
3342 hlen
= be32_to_cpu(rhead
->h_len
);
3343 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3344 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3345 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3346 return XFS_ERROR(EFSCORRUPTED
);
3348 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3349 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3350 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3351 return XFS_ERROR(EFSCORRUPTED
);
3357 * Read the log from tail to head and process the log records found.
3358 * Handle the two cases where the tail and head are in the same cycle
3359 * and where the active portion of the log wraps around the end of
3360 * the physical log separately. The pass parameter is passed through
3361 * to the routines called to process the data and is not looked at
3365 xlog_do_recovery_pass(
3367 xfs_daddr_t head_blk
,
3368 xfs_daddr_t tail_blk
,
3371 xlog_rec_header_t
*rhead
;
3374 xfs_buf_t
*hbp
, *dbp
;
3375 int error
= 0, h_size
;
3376 int bblks
, split_bblks
;
3377 int hblks
, split_hblks
, wrapped_hblks
;
3378 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3380 ASSERT(head_blk
!= tail_blk
);
3383 * Read the header of the tail block and get the iclog buffer size from
3384 * h_size. Use this to tell how many sectors make up the log header.
3386 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3388 * When using variable length iclogs, read first sector of
3389 * iclog header and extract the header size from it. Get a
3390 * new hbp that is the correct size.
3392 hbp
= xlog_get_bp(log
, 1);
3396 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3400 rhead
= (xlog_rec_header_t
*)offset
;
3401 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3404 h_size
= be32_to_cpu(rhead
->h_size
);
3405 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3406 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3407 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3408 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3411 hbp
= xlog_get_bp(log
, hblks
);
3416 ASSERT(log
->l_sectBBsize
== 1);
3418 hbp
= xlog_get_bp(log
, 1);
3419 h_size
= XLOG_BIG_RECORD_BSIZE
;
3424 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3430 memset(rhash
, 0, sizeof(rhash
));
3431 if (tail_blk
<= head_blk
) {
3432 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3433 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3437 rhead
= (xlog_rec_header_t
*)offset
;
3438 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3442 /* blocks in data section */
3443 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3444 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3449 error
= xlog_unpack_data(rhead
, offset
, log
);
3453 error
= xlog_recover_process_data(log
,
3454 rhash
, rhead
, offset
, pass
);
3457 blk_no
+= bblks
+ hblks
;
3461 * Perform recovery around the end of the physical log.
3462 * When the head is not on the same cycle number as the tail,
3463 * we can't do a sequential recovery as above.
3466 while (blk_no
< log
->l_logBBsize
) {
3468 * Check for header wrapping around physical end-of-log
3470 offset
= hbp
->b_addr
;
3473 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3474 /* Read header in one read */
3475 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3480 /* This LR is split across physical log end */
3481 if (blk_no
!= log
->l_logBBsize
) {
3482 /* some data before physical log end */
3483 ASSERT(blk_no
<= INT_MAX
);
3484 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3485 ASSERT(split_hblks
> 0);
3486 error
= xlog_bread(log
, blk_no
,
3494 * Note: this black magic still works with
3495 * large sector sizes (non-512) only because:
3496 * - we increased the buffer size originally
3497 * by 1 sector giving us enough extra space
3498 * for the second read;
3499 * - the log start is guaranteed to be sector
3501 * - we read the log end (LR header start)
3502 * _first_, then the log start (LR header end)
3503 * - order is important.
3505 wrapped_hblks
= hblks
- split_hblks
;
3506 error
= xlog_bread_offset(log
, 0,
3508 offset
+ BBTOB(split_hblks
));
3512 rhead
= (xlog_rec_header_t
*)offset
;
3513 error
= xlog_valid_rec_header(log
, rhead
,
3514 split_hblks
? blk_no
: 0);
3518 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3521 /* Read in data for log record */
3522 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3523 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3528 /* This log record is split across the
3529 * physical end of log */
3530 offset
= dbp
->b_addr
;
3532 if (blk_no
!= log
->l_logBBsize
) {
3533 /* some data is before the physical
3535 ASSERT(!wrapped_hblks
);
3536 ASSERT(blk_no
<= INT_MAX
);
3538 log
->l_logBBsize
- (int)blk_no
;
3539 ASSERT(split_bblks
> 0);
3540 error
= xlog_bread(log
, blk_no
,
3548 * Note: this black magic still works with
3549 * large sector sizes (non-512) only because:
3550 * - we increased the buffer size originally
3551 * by 1 sector giving us enough extra space
3552 * for the second read;
3553 * - the log start is guaranteed to be sector
3555 * - we read the log end (LR header start)
3556 * _first_, then the log start (LR header end)
3557 * - order is important.
3559 error
= xlog_bread_offset(log
, 0,
3560 bblks
- split_bblks
, dbp
,
3561 offset
+ BBTOB(split_bblks
));
3566 error
= xlog_unpack_data(rhead
, offset
, log
);
3570 error
= xlog_recover_process_data(log
, rhash
,
3571 rhead
, offset
, pass
);
3577 ASSERT(blk_no
>= log
->l_logBBsize
);
3578 blk_no
-= log
->l_logBBsize
;
3580 /* read first part of physical log */
3581 while (blk_no
< head_blk
) {
3582 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3586 rhead
= (xlog_rec_header_t
*)offset
;
3587 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3591 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3592 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3597 error
= xlog_unpack_data(rhead
, offset
, log
);
3601 error
= xlog_recover_process_data(log
, rhash
,
3602 rhead
, offset
, pass
);
3605 blk_no
+= bblks
+ hblks
;
3617 * Do the recovery of the log. We actually do this in two phases.
3618 * The two passes are necessary in order to implement the function
3619 * of cancelling a record written into the log. The first pass
3620 * determines those things which have been cancelled, and the
3621 * second pass replays log items normally except for those which
3622 * have been cancelled. The handling of the replay and cancellations
3623 * takes place in the log item type specific routines.
3625 * The table of items which have cancel records in the log is allocated
3626 * and freed at this level, since only here do we know when all of
3627 * the log recovery has been completed.
3630 xlog_do_log_recovery(
3632 xfs_daddr_t head_blk
,
3633 xfs_daddr_t tail_blk
)
3637 ASSERT(head_blk
!= tail_blk
);
3640 * First do a pass to find all of the cancelled buf log items.
3641 * Store them in the buf_cancel_table for use in the second pass.
3643 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3644 sizeof(struct list_head
),
3646 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3647 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
3649 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3650 XLOG_RECOVER_PASS1
);
3652 kmem_free(log
->l_buf_cancel_table
);
3653 log
->l_buf_cancel_table
= NULL
;
3657 * Then do a second pass to actually recover the items in the log.
3658 * When it is complete free the table of buf cancel items.
3660 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3661 XLOG_RECOVER_PASS2
);
3666 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3667 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
3671 kmem_free(log
->l_buf_cancel_table
);
3672 log
->l_buf_cancel_table
= NULL
;
3678 * Do the actual recovery
3683 xfs_daddr_t head_blk
,
3684 xfs_daddr_t tail_blk
)
3691 * First replay the images in the log.
3693 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3698 * If IO errors happened during recovery, bail out.
3700 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3705 * We now update the tail_lsn since much of the recovery has completed
3706 * and there may be space available to use. If there were no extent
3707 * or iunlinks, we can free up the entire log and set the tail_lsn to
3708 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3709 * lsn of the last known good LR on disk. If there are extent frees
3710 * or iunlinks they will have some entries in the AIL; so we look at
3711 * the AIL to determine how to set the tail_lsn.
3713 xlog_assign_tail_lsn(log
->l_mp
);
3716 * Now that we've finished replaying all buffer and inode
3717 * updates, re-read in the superblock and reverify it.
3719 bp
= xfs_getsb(log
->l_mp
, 0);
3721 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3723 XFS_BUF_UNASYNC(bp
);
3724 bp
->b_ops
= &xfs_sb_buf_ops
;
3725 xfsbdstrat(log
->l_mp
, bp
);
3726 error
= xfs_buf_iowait(bp
);
3728 xfs_buf_ioerror_alert(bp
, __func__
);
3734 /* Convert superblock from on-disk format */
3735 sbp
= &log
->l_mp
->m_sb
;
3736 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3737 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3738 ASSERT(xfs_sb_good_version(sbp
));
3741 /* We've re-read the superblock so re-initialize per-cpu counters */
3742 xfs_icsb_reinit_counters(log
->l_mp
);
3744 xlog_recover_check_summary(log
);
3746 /* Normal transactions can now occur */
3747 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3752 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3754 * Return error or zero.
3760 xfs_daddr_t head_blk
, tail_blk
;
3763 /* find the tail of the log */
3764 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3767 if (tail_blk
!= head_blk
) {
3768 /* There used to be a comment here:
3770 * disallow recovery on read-only mounts. note -- mount
3771 * checks for ENOSPC and turns it into an intelligent
3773 * ...but this is no longer true. Now, unless you specify
3774 * NORECOVERY (in which case this function would never be
3775 * called), we just go ahead and recover. We do this all
3776 * under the vfs layer, so we can get away with it unless
3777 * the device itself is read-only, in which case we fail.
3779 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3783 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
3784 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
3787 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3788 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3794 * In the first part of recovery we replay inodes and buffers and build
3795 * up the list of extent free items which need to be processed. Here
3796 * we process the extent free items and clean up the on disk unlinked
3797 * inode lists. This is separated from the first part of recovery so
3798 * that the root and real-time bitmap inodes can be read in from disk in
3799 * between the two stages. This is necessary so that we can free space
3800 * in the real-time portion of the file system.
3803 xlog_recover_finish(
3807 * Now we're ready to do the transactions needed for the
3808 * rest of recovery. Start with completing all the extent
3809 * free intent records and then process the unlinked inode
3810 * lists. At this point, we essentially run in normal mode
3811 * except that we're still performing recovery actions
3812 * rather than accepting new requests.
3814 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3816 error
= xlog_recover_process_efis(log
);
3818 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
3822 * Sync the log to get all the EFIs out of the AIL.
3823 * This isn't absolutely necessary, but it helps in
3824 * case the unlink transactions would have problems
3825 * pushing the EFIs out of the way.
3827 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
3829 xlog_recover_process_iunlinks(log
);
3831 xlog_recover_check_summary(log
);
3833 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
3834 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
3836 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3838 xfs_info(log
->l_mp
, "Ending clean mount");
3846 * Read all of the agf and agi counters and check that they
3847 * are consistent with the superblock counters.
3850 xlog_recover_check_summary(
3857 xfs_agnumber_t agno
;
3858 __uint64_t freeblks
;
3868 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3869 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
3871 xfs_alert(mp
, "%s agf read failed agno %d error %d",
3872 __func__
, agno
, error
);
3874 agfp
= XFS_BUF_TO_AGF(agfbp
);
3875 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
3876 be32_to_cpu(agfp
->agf_flcount
);
3877 xfs_buf_relse(agfbp
);
3880 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3882 xfs_alert(mp
, "%s agi read failed agno %d error %d",
3883 __func__
, agno
, error
);
3885 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
3887 itotal
+= be32_to_cpu(agi
->agi_count
);
3888 ifree
+= be32_to_cpu(agi
->agi_freecount
);
3889 xfs_buf_relse(agibp
);