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_format.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_cksum.h"
45 #include "xfs_trace.h"
46 #include "xfs_icache.h"
47 #include "xfs_icreate_item.h"
49 /* Need all the magic numbers and buffer ops structures from these headers */
50 #include "xfs_symlink.h"
51 #include "xfs_da_btree.h"
52 #include "xfs_dir2_format.h"
54 #include "xfs_attr_leaf.h"
55 #include "xfs_attr_remote.h"
57 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
64 xlog_clear_stale_blocks(
69 xlog_recover_check_summary(
72 #define xlog_recover_check_summary(log)
76 * This structure is used during recovery to record the buf log items which
77 * have been canceled and should not be replayed.
79 struct xfs_buf_cancel
{
83 struct list_head bc_list
;
87 * Sector aligned buffer routines for buffer create/read/write/access
91 * Verify the given count of basic blocks is valid number of blocks
92 * to specify for an operation involving the given XFS log buffer.
93 * Returns nonzero if the count is valid, 0 otherwise.
97 xlog_buf_bbcount_valid(
101 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
105 * Allocate a buffer to hold log data. The buffer needs to be able
106 * to map to a range of nbblks basic blocks at any valid (basic
107 * block) offset within the log.
116 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
117 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
119 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
124 * We do log I/O in units of log sectors (a power-of-2
125 * multiple of the basic block size), so we round up the
126 * requested size to accommodate the basic blocks required
127 * for complete log sectors.
129 * In addition, the buffer may be used for a non-sector-
130 * aligned block offset, in which case an I/O of the
131 * requested size could extend beyond the end of the
132 * buffer. If the requested size is only 1 basic block it
133 * will never straddle a sector boundary, so this won't be
134 * an issue. Nor will this be a problem if the log I/O is
135 * done in basic blocks (sector size 1). But otherwise we
136 * extend the buffer by one extra log sector to ensure
137 * there's space to accommodate this possibility.
139 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
140 nbblks
+= log
->l_sectBBsize
;
141 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
143 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
157 * Return the address of the start of the given block number's data
158 * in a log buffer. The buffer covers a log sector-aligned region.
167 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
169 ASSERT(offset
+ nbblks
<= bp
->b_length
);
170 return bp
->b_addr
+ BBTOB(offset
);
175 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
186 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
187 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
189 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
193 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
194 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
197 ASSERT(nbblks
<= bp
->b_length
);
199 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
201 bp
->b_io_length
= nbblks
;
204 xfsbdstrat(log
->l_mp
, bp
);
205 error
= xfs_buf_iowait(bp
);
207 xfs_buf_ioerror_alert(bp
, __func__
);
221 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
225 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
230 * Read at an offset into the buffer. Returns with the buffer in it's original
231 * state regardless of the result of the read.
236 xfs_daddr_t blk_no
, /* block to read from */
237 int nbblks
, /* blocks to read */
241 xfs_caddr_t orig_offset
= bp
->b_addr
;
242 int orig_len
= BBTOB(bp
->b_length
);
245 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
249 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
251 /* must reset buffer pointer even on error */
252 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
259 * Write out the buffer at the given block for the given number of blocks.
260 * The buffer is kept locked across the write and is returned locked.
261 * This can only be used for synchronous log writes.
272 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
273 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
275 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
279 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
280 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
283 ASSERT(nbblks
<= bp
->b_length
);
285 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
286 XFS_BUF_ZEROFLAGS(bp
);
289 bp
->b_io_length
= nbblks
;
292 error
= xfs_bwrite(bp
);
294 xfs_buf_ioerror_alert(bp
, __func__
);
301 * dump debug superblock and log record information
304 xlog_header_check_dump(
306 xlog_rec_header_t
*head
)
308 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d\n",
309 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
310 xfs_debug(mp
, " log : uuid = %pU, fmt = %d\n",
311 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
314 #define xlog_header_check_dump(mp, head)
318 * check log record header for recovery
321 xlog_header_check_recover(
323 xlog_rec_header_t
*head
)
325 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
328 * IRIX doesn't write the h_fmt field and leaves it zeroed
329 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
330 * a dirty log created in IRIX.
332 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
334 "dirty log written in incompatible format - can't recover");
335 xlog_header_check_dump(mp
, head
);
336 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
337 XFS_ERRLEVEL_HIGH
, mp
);
338 return XFS_ERROR(EFSCORRUPTED
);
339 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
341 "dirty log entry has mismatched uuid - can't recover");
342 xlog_header_check_dump(mp
, head
);
343 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
344 XFS_ERRLEVEL_HIGH
, mp
);
345 return XFS_ERROR(EFSCORRUPTED
);
351 * read the head block of the log and check the header
354 xlog_header_check_mount(
356 xlog_rec_header_t
*head
)
358 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
360 if (uuid_is_nil(&head
->h_fs_uuid
)) {
362 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
363 * h_fs_uuid is nil, we assume this log was last mounted
364 * by IRIX and continue.
366 xfs_warn(mp
, "nil uuid in log - IRIX style log");
367 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
368 xfs_warn(mp
, "log has mismatched uuid - can't recover");
369 xlog_header_check_dump(mp
, head
);
370 XFS_ERROR_REPORT("xlog_header_check_mount",
371 XFS_ERRLEVEL_HIGH
, mp
);
372 return XFS_ERROR(EFSCORRUPTED
);
383 * We're not going to bother about retrying
384 * this during recovery. One strike!
386 xfs_buf_ioerror_alert(bp
, __func__
);
387 xfs_force_shutdown(bp
->b_target
->bt_mount
,
388 SHUTDOWN_META_IO_ERROR
);
391 xfs_buf_ioend(bp
, 0);
395 * This routine finds (to an approximation) the first block in the physical
396 * log which contains the given cycle. It uses a binary search algorithm.
397 * Note that the algorithm can not be perfect because the disk will not
398 * necessarily be perfect.
401 xlog_find_cycle_start(
404 xfs_daddr_t first_blk
,
405 xfs_daddr_t
*last_blk
,
415 mid_blk
= BLK_AVG(first_blk
, end_blk
);
416 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
417 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
420 mid_cycle
= xlog_get_cycle(offset
);
421 if (mid_cycle
== cycle
)
422 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
424 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
425 mid_blk
= BLK_AVG(first_blk
, end_blk
);
427 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
428 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
436 * Check that a range of blocks does not contain stop_on_cycle_no.
437 * Fill in *new_blk with the block offset where such a block is
438 * found, or with -1 (an invalid block number) if there is no such
439 * block in the range. The scan needs to occur from front to back
440 * and the pointer into the region must be updated since a later
441 * routine will need to perform another test.
444 xlog_find_verify_cycle(
446 xfs_daddr_t start_blk
,
448 uint stop_on_cycle_no
,
449 xfs_daddr_t
*new_blk
)
455 xfs_caddr_t buf
= NULL
;
459 * Greedily allocate a buffer big enough to handle the full
460 * range of basic blocks we'll be examining. If that fails,
461 * try a smaller size. We need to be able to read at least
462 * a log sector, or we're out of luck.
464 bufblks
= 1 << ffs(nbblks
);
465 while (bufblks
> log
->l_logBBsize
)
467 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
469 if (bufblks
< log
->l_sectBBsize
)
473 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
476 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
478 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
482 for (j
= 0; j
< bcount
; j
++) {
483 cycle
= xlog_get_cycle(buf
);
484 if (cycle
== stop_on_cycle_no
) {
501 * Potentially backup over partial log record write.
503 * In the typical case, last_blk is the number of the block directly after
504 * a good log record. Therefore, we subtract one to get the block number
505 * of the last block in the given buffer. extra_bblks contains the number
506 * of blocks we would have read on a previous read. This happens when the
507 * last log record is split over the end of the physical log.
509 * extra_bblks is the number of blocks potentially verified on a previous
510 * call to this routine.
513 xlog_find_verify_log_record(
515 xfs_daddr_t start_blk
,
516 xfs_daddr_t
*last_blk
,
521 xfs_caddr_t offset
= NULL
;
522 xlog_rec_header_t
*head
= NULL
;
525 int num_blks
= *last_blk
- start_blk
;
528 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
530 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
531 if (!(bp
= xlog_get_bp(log
, 1)))
535 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
538 offset
+= ((num_blks
- 1) << BBSHIFT
);
541 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
543 /* valid log record not found */
545 "Log inconsistent (didn't find previous header)");
547 error
= XFS_ERROR(EIO
);
552 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
557 head
= (xlog_rec_header_t
*)offset
;
559 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
567 * We hit the beginning of the physical log & still no header. Return
568 * to caller. If caller can handle a return of -1, then this routine
569 * will be called again for the end of the physical log.
577 * We have the final block of the good log (the first block
578 * of the log record _before_ the head. So we check the uuid.
580 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
584 * We may have found a log record header before we expected one.
585 * last_blk will be the 1st block # with a given cycle #. We may end
586 * up reading an entire log record. In this case, we don't want to
587 * reset last_blk. Only when last_blk points in the middle of a log
588 * record do we update last_blk.
590 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
591 uint h_size
= be32_to_cpu(head
->h_size
);
593 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
594 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
600 if (*last_blk
- i
+ extra_bblks
!=
601 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
610 * Head is defined to be the point of the log where the next log write
611 * write could go. This means that incomplete LR writes at the end are
612 * eliminated when calculating the head. We aren't guaranteed that previous
613 * LR have complete transactions. We only know that a cycle number of
614 * current cycle number -1 won't be present in the log if we start writing
615 * from our current block number.
617 * last_blk contains the block number of the first block with a given
620 * Return: zero if normal, non-zero if error.
625 xfs_daddr_t
*return_head_blk
)
629 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
631 uint first_half_cycle
, last_half_cycle
;
633 int error
, log_bbnum
= log
->l_logBBsize
;
635 /* Is the end of the log device zeroed? */
636 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
637 *return_head_blk
= first_blk
;
639 /* Is the whole lot zeroed? */
641 /* Linux XFS shouldn't generate totally zeroed logs -
642 * mkfs etc write a dummy unmount record to a fresh
643 * log so we can store the uuid in there
645 xfs_warn(log
->l_mp
, "totally zeroed log");
650 xfs_warn(log
->l_mp
, "empty log check failed");
654 first_blk
= 0; /* get cycle # of 1st block */
655 bp
= xlog_get_bp(log
, 1);
659 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
663 first_half_cycle
= xlog_get_cycle(offset
);
665 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
666 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
670 last_half_cycle
= xlog_get_cycle(offset
);
671 ASSERT(last_half_cycle
!= 0);
674 * If the 1st half cycle number is equal to the last half cycle number,
675 * then the entire log is stamped with the same cycle number. In this
676 * case, head_blk can't be set to zero (which makes sense). The below
677 * math doesn't work out properly with head_blk equal to zero. Instead,
678 * we set it to log_bbnum which is an invalid block number, but this
679 * value makes the math correct. If head_blk doesn't changed through
680 * all the tests below, *head_blk is set to zero at the very end rather
681 * than log_bbnum. In a sense, log_bbnum and zero are the same block
682 * in a circular file.
684 if (first_half_cycle
== last_half_cycle
) {
686 * In this case we believe that the entire log should have
687 * cycle number last_half_cycle. We need to scan backwards
688 * from the end verifying that there are no holes still
689 * containing last_half_cycle - 1. If we find such a hole,
690 * then the start of that hole will be the new head. The
691 * simple case looks like
692 * x | x ... | x - 1 | x
693 * Another case that fits this picture would be
694 * x | x + 1 | x ... | x
695 * In this case the head really is somewhere at the end of the
696 * log, as one of the latest writes at the beginning was
699 * x | x + 1 | x ... | x - 1 | x
700 * This is really the combination of the above two cases, and
701 * the head has to end up at the start of the x-1 hole at the
704 * In the 256k log case, we will read from the beginning to the
705 * end of the log and search for cycle numbers equal to x-1.
706 * We don't worry about the x+1 blocks that we encounter,
707 * because we know that they cannot be the head since the log
710 head_blk
= log_bbnum
;
711 stop_on_cycle
= last_half_cycle
- 1;
714 * In this case we want to find the first block with cycle
715 * number matching last_half_cycle. We expect the log to be
717 * x + 1 ... | x ... | x
718 * The first block with cycle number x (last_half_cycle) will
719 * be where the new head belongs. First we do a binary search
720 * for the first occurrence of last_half_cycle. The binary
721 * search may not be totally accurate, so then we scan back
722 * from there looking for occurrences of last_half_cycle before
723 * us. If that backwards scan wraps around the beginning of
724 * the log, then we look for occurrences of last_half_cycle - 1
725 * at the end of the log. The cases we're looking for look
727 * v binary search stopped here
728 * x + 1 ... | x | x + 1 | x ... | x
729 * ^ but we want to locate this spot
731 * <---------> less than scan distance
732 * x + 1 ... | x ... | x - 1 | x
733 * ^ we want to locate this spot
735 stop_on_cycle
= last_half_cycle
;
736 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
737 &head_blk
, last_half_cycle
)))
742 * Now validate the answer. Scan back some number of maximum possible
743 * blocks and make sure each one has the expected cycle number. The
744 * maximum is determined by the total possible amount of buffering
745 * in the in-core log. The following number can be made tighter if
746 * we actually look at the block size of the filesystem.
748 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
749 if (head_blk
>= num_scan_bblks
) {
751 * We are guaranteed that the entire check can be performed
754 start_blk
= head_blk
- num_scan_bblks
;
755 if ((error
= xlog_find_verify_cycle(log
,
756 start_blk
, num_scan_bblks
,
757 stop_on_cycle
, &new_blk
)))
761 } else { /* need to read 2 parts of log */
763 * We are going to scan backwards in the log in two parts.
764 * First we scan the physical end of the log. In this part
765 * of the log, we are looking for blocks with cycle number
766 * last_half_cycle - 1.
767 * If we find one, then we know that the log starts there, as
768 * we've found a hole that didn't get written in going around
769 * the end of the physical log. The simple case for this is
770 * x + 1 ... | x ... | x - 1 | x
771 * <---------> less than scan distance
772 * If all of the blocks at the end of the log have cycle number
773 * last_half_cycle, then we check the blocks at the start of
774 * the log looking for occurrences of last_half_cycle. If we
775 * find one, then our current estimate for the location of the
776 * first occurrence of last_half_cycle is wrong and we move
777 * back to the hole we've found. This case looks like
778 * x + 1 ... | x | x + 1 | x ...
779 * ^ binary search stopped here
780 * Another case we need to handle that only occurs in 256k
782 * x + 1 ... | x ... | x+1 | x ...
783 * ^ binary search stops here
784 * In a 256k log, the scan at the end of the log will see the
785 * x + 1 blocks. We need to skip past those since that is
786 * certainly not the head of the log. By searching for
787 * last_half_cycle-1 we accomplish that.
789 ASSERT(head_blk
<= INT_MAX
&&
790 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
791 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
792 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
793 num_scan_bblks
- (int)head_blk
,
794 (stop_on_cycle
- 1), &new_blk
)))
802 * Scan beginning of log now. The last part of the physical
803 * log is good. This scan needs to verify that it doesn't find
804 * the last_half_cycle.
807 ASSERT(head_blk
<= INT_MAX
);
808 if ((error
= xlog_find_verify_cycle(log
,
809 start_blk
, (int)head_blk
,
810 stop_on_cycle
, &new_blk
)))
818 * Now we need to make sure head_blk is not pointing to a block in
819 * the middle of a log record.
821 num_scan_bblks
= XLOG_REC_SHIFT(log
);
822 if (head_blk
>= num_scan_bblks
) {
823 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
825 /* start ptr at last block ptr before head_blk */
826 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
827 &head_blk
, 0)) == -1) {
828 error
= XFS_ERROR(EIO
);
834 ASSERT(head_blk
<= INT_MAX
);
835 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
836 &head_blk
, 0)) == -1) {
837 /* We hit the beginning of the log during our search */
838 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
840 ASSERT(start_blk
<= INT_MAX
&&
841 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
842 ASSERT(head_blk
<= INT_MAX
);
843 if ((error
= xlog_find_verify_log_record(log
,
845 (int)head_blk
)) == -1) {
846 error
= XFS_ERROR(EIO
);
850 if (new_blk
!= log_bbnum
)
857 if (head_blk
== log_bbnum
)
858 *return_head_blk
= 0;
860 *return_head_blk
= head_blk
;
862 * When returning here, we have a good block number. Bad block
863 * means that during a previous crash, we didn't have a clean break
864 * from cycle number N to cycle number N-1. In this case, we need
865 * to find the first block with cycle number N-1.
873 xfs_warn(log
->l_mp
, "failed to find log head");
878 * Find the sync block number or the tail of the log.
880 * This will be the block number of the last record to have its
881 * associated buffers synced to disk. Every log record header has
882 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
883 * to get a sync block number. The only concern is to figure out which
884 * log record header to believe.
886 * The following algorithm uses the log record header with the largest
887 * lsn. The entire log record does not need to be valid. We only care
888 * that the header is valid.
890 * We could speed up search by using current head_blk buffer, but it is not
896 xfs_daddr_t
*head_blk
,
897 xfs_daddr_t
*tail_blk
)
899 xlog_rec_header_t
*rhead
;
900 xlog_op_header_t
*op_head
;
901 xfs_caddr_t offset
= NULL
;
904 xfs_daddr_t umount_data_blk
;
905 xfs_daddr_t after_umount_blk
;
912 * Find previous log record
914 if ((error
= xlog_find_head(log
, head_blk
)))
917 bp
= xlog_get_bp(log
, 1);
920 if (*head_blk
== 0) { /* special case */
921 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
925 if (xlog_get_cycle(offset
) == 0) {
927 /* leave all other log inited values alone */
933 * Search backwards looking for log record header block
935 ASSERT(*head_blk
< INT_MAX
);
936 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
937 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
941 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
947 * If we haven't found the log record header block, start looking
948 * again from the end of the physical log. XXXmiken: There should be
949 * a check here to make sure we didn't search more than N blocks in
953 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
954 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
958 if (*(__be32
*)offset
==
959 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
966 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
968 return XFS_ERROR(EIO
);
971 /* find blk_no of tail of log */
972 rhead
= (xlog_rec_header_t
*)offset
;
973 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
976 * Reset log values according to the state of the log when we
977 * crashed. In the case where head_blk == 0, we bump curr_cycle
978 * one because the next write starts a new cycle rather than
979 * continuing the cycle of the last good log record. At this
980 * point we have guaranteed that all partial log records have been
981 * accounted for. Therefore, we know that the last good log record
982 * written was complete and ended exactly on the end boundary
983 * of the physical log.
985 log
->l_prev_block
= i
;
986 log
->l_curr_block
= (int)*head_blk
;
987 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
990 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
991 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
992 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
993 BBTOB(log
->l_curr_block
));
994 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
995 BBTOB(log
->l_curr_block
));
998 * Look for unmount record. If we find it, then we know there
999 * was a clean unmount. Since 'i' could be the last block in
1000 * the physical log, we convert to a log block before comparing
1003 * Save the current tail lsn to use to pass to
1004 * xlog_clear_stale_blocks() below. We won't want to clear the
1005 * unmount record if there is one, so we pass the lsn of the
1006 * unmount record rather than the block after it.
1008 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1009 int h_size
= be32_to_cpu(rhead
->h_size
);
1010 int h_version
= be32_to_cpu(rhead
->h_version
);
1012 if ((h_version
& XLOG_VERSION_2
) &&
1013 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1014 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1015 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1023 after_umount_blk
= (i
+ hblks
+ (int)
1024 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1025 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1026 if (*head_blk
== after_umount_blk
&&
1027 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1028 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1029 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1033 op_head
= (xlog_op_header_t
*)offset
;
1034 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1036 * Set tail and last sync so that newly written
1037 * log records will point recovery to after the
1038 * current unmount record.
1040 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1041 log
->l_curr_cycle
, after_umount_blk
);
1042 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1043 log
->l_curr_cycle
, after_umount_blk
);
1044 *tail_blk
= after_umount_blk
;
1047 * Note that the unmount was clean. If the unmount
1048 * was not clean, we need to know this to rebuild the
1049 * superblock counters from the perag headers if we
1050 * have a filesystem using non-persistent counters.
1052 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1057 * Make sure that there are no blocks in front of the head
1058 * with the same cycle number as the head. This can happen
1059 * because we allow multiple outstanding log writes concurrently,
1060 * and the later writes might make it out before earlier ones.
1062 * We use the lsn from before modifying it so that we'll never
1063 * overwrite the unmount record after a clean unmount.
1065 * Do this only if we are going to recover the filesystem
1067 * NOTE: This used to say "if (!readonly)"
1068 * However on Linux, we can & do recover a read-only filesystem.
1069 * We only skip recovery if NORECOVERY is specified on mount,
1070 * in which case we would not be here.
1072 * But... if the -device- itself is readonly, just skip this.
1073 * We can't recover this device anyway, so it won't matter.
1075 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1076 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1082 xfs_warn(log
->l_mp
, "failed to locate log tail");
1087 * Is the log zeroed at all?
1089 * The last binary search should be changed to perform an X block read
1090 * once X becomes small enough. You can then search linearly through
1091 * the X blocks. This will cut down on the number of reads we need to do.
1093 * If the log is partially zeroed, this routine will pass back the blkno
1094 * of the first block with cycle number 0. It won't have a complete LR
1098 * 0 => the log is completely written to
1099 * -1 => use *blk_no as the first block of the log
1100 * >0 => error has occurred
1105 xfs_daddr_t
*blk_no
)
1109 uint first_cycle
, last_cycle
;
1110 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1111 xfs_daddr_t num_scan_bblks
;
1112 int error
, log_bbnum
= log
->l_logBBsize
;
1116 /* check totally zeroed log */
1117 bp
= xlog_get_bp(log
, 1);
1120 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1124 first_cycle
= xlog_get_cycle(offset
);
1125 if (first_cycle
== 0) { /* completely zeroed log */
1131 /* check partially zeroed log */
1132 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1136 last_cycle
= xlog_get_cycle(offset
);
1137 if (last_cycle
!= 0) { /* log completely written to */
1140 } else if (first_cycle
!= 1) {
1142 * If the cycle of the last block is zero, the cycle of
1143 * the first block must be 1. If it's not, maybe we're
1144 * not looking at a log... Bail out.
1147 "Log inconsistent or not a log (last==0, first!=1)");
1148 return XFS_ERROR(EINVAL
);
1151 /* we have a partially zeroed log */
1152 last_blk
= log_bbnum
-1;
1153 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1157 * Validate the answer. Because there is no way to guarantee that
1158 * the entire log is made up of log records which are the same size,
1159 * we scan over the defined maximum blocks. At this point, the maximum
1160 * is not chosen to mean anything special. XXXmiken
1162 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1163 ASSERT(num_scan_bblks
<= INT_MAX
);
1165 if (last_blk
< num_scan_bblks
)
1166 num_scan_bblks
= last_blk
;
1167 start_blk
= last_blk
- num_scan_bblks
;
1170 * We search for any instances of cycle number 0 that occur before
1171 * our current estimate of the head. What we're trying to detect is
1172 * 1 ... | 0 | 1 | 0...
1173 * ^ binary search ends here
1175 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1176 (int)num_scan_bblks
, 0, &new_blk
)))
1182 * Potentially backup over partial log record write. We don't need
1183 * to search the end of the log because we know it is zero.
1185 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1186 &last_blk
, 0)) == -1) {
1187 error
= XFS_ERROR(EIO
);
1201 * These are simple subroutines used by xlog_clear_stale_blocks() below
1202 * to initialize a buffer full of empty log record headers and write
1203 * them into the log.
1214 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1216 memset(buf
, 0, BBSIZE
);
1217 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1218 recp
->h_cycle
= cpu_to_be32(cycle
);
1219 recp
->h_version
= cpu_to_be32(
1220 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1221 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1222 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1223 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1224 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1228 xlog_write_log_records(
1239 int sectbb
= log
->l_sectBBsize
;
1240 int end_block
= start_block
+ blocks
;
1246 * Greedily allocate a buffer big enough to handle the full
1247 * range of basic blocks to be written. If that fails, try
1248 * a smaller size. We need to be able to write at least a
1249 * log sector, or we're out of luck.
1251 bufblks
= 1 << ffs(blocks
);
1252 while (bufblks
> log
->l_logBBsize
)
1254 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1256 if (bufblks
< sectbb
)
1260 /* We may need to do a read at the start to fill in part of
1261 * the buffer in the starting sector not covered by the first
1264 balign
= round_down(start_block
, sectbb
);
1265 if (balign
!= start_block
) {
1266 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1270 j
= start_block
- balign
;
1273 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1274 int bcount
, endcount
;
1276 bcount
= min(bufblks
, end_block
- start_block
);
1277 endcount
= bcount
- j
;
1279 /* We may need to do a read at the end to fill in part of
1280 * the buffer in the final sector not covered by the write.
1281 * If this is the same sector as the above read, skip it.
1283 ealign
= round_down(end_block
, sectbb
);
1284 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1285 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1286 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1293 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1294 for (; j
< endcount
; j
++) {
1295 xlog_add_record(log
, offset
, cycle
, i
+j
,
1296 tail_cycle
, tail_block
);
1299 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1302 start_block
+= endcount
;
1312 * This routine is called to blow away any incomplete log writes out
1313 * in front of the log head. We do this so that we won't become confused
1314 * if we come up, write only a little bit more, and then crash again.
1315 * If we leave the partial log records out there, this situation could
1316 * cause us to think those partial writes are valid blocks since they
1317 * have the current cycle number. We get rid of them by overwriting them
1318 * with empty log records with the old cycle number rather than the
1321 * The tail lsn is passed in rather than taken from
1322 * the log so that we will not write over the unmount record after a
1323 * clean unmount in a 512 block log. Doing so would leave the log without
1324 * any valid log records in it until a new one was written. If we crashed
1325 * during that time we would not be able to recover.
1328 xlog_clear_stale_blocks(
1332 int tail_cycle
, head_cycle
;
1333 int tail_block
, head_block
;
1334 int tail_distance
, max_distance
;
1338 tail_cycle
= CYCLE_LSN(tail_lsn
);
1339 tail_block
= BLOCK_LSN(tail_lsn
);
1340 head_cycle
= log
->l_curr_cycle
;
1341 head_block
= log
->l_curr_block
;
1344 * Figure out the distance between the new head of the log
1345 * and the tail. We want to write over any blocks beyond the
1346 * head that we may have written just before the crash, but
1347 * we don't want to overwrite the tail of the log.
1349 if (head_cycle
== tail_cycle
) {
1351 * The tail is behind the head in the physical log,
1352 * so the distance from the head to the tail is the
1353 * distance from the head to the end of the log plus
1354 * the distance from the beginning of the log to the
1357 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1358 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1359 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1360 return XFS_ERROR(EFSCORRUPTED
);
1362 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1365 * The head is behind the tail in the physical log,
1366 * so the distance from the head to the tail is just
1367 * the tail block minus the head block.
1369 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1370 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1371 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1372 return XFS_ERROR(EFSCORRUPTED
);
1374 tail_distance
= tail_block
- head_block
;
1378 * If the head is right up against the tail, we can't clear
1381 if (tail_distance
<= 0) {
1382 ASSERT(tail_distance
== 0);
1386 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1388 * Take the smaller of the maximum amount of outstanding I/O
1389 * we could have and the distance to the tail to clear out.
1390 * We take the smaller so that we don't overwrite the tail and
1391 * we don't waste all day writing from the head to the tail
1394 max_distance
= MIN(max_distance
, tail_distance
);
1396 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1398 * We can stomp all the blocks we need to without
1399 * wrapping around the end of the log. Just do it
1400 * in a single write. Use the cycle number of the
1401 * current cycle minus one so that the log will look like:
1404 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1405 head_block
, max_distance
, tail_cycle
,
1411 * We need to wrap around the end of the physical log in
1412 * order to clear all the blocks. Do it in two separate
1413 * I/Os. The first write should be from the head to the
1414 * end of the physical log, and it should use the current
1415 * cycle number minus one just like above.
1417 distance
= log
->l_logBBsize
- head_block
;
1418 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1419 head_block
, distance
, tail_cycle
,
1426 * Now write the blocks at the start of the physical log.
1427 * This writes the remainder of the blocks we want to clear.
1428 * It uses the current cycle number since we're now on the
1429 * same cycle as the head so that we get:
1430 * n ... n ... | n - 1 ...
1431 * ^^^^^ blocks we're writing
1433 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1434 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1435 tail_cycle
, tail_block
);
1443 /******************************************************************************
1445 * Log recover routines
1447 ******************************************************************************
1450 STATIC xlog_recover_t
*
1451 xlog_recover_find_tid(
1452 struct hlist_head
*head
,
1455 xlog_recover_t
*trans
;
1457 hlist_for_each_entry(trans
, head
, r_list
) {
1458 if (trans
->r_log_tid
== tid
)
1465 xlog_recover_new_tid(
1466 struct hlist_head
*head
,
1470 xlog_recover_t
*trans
;
1472 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1473 trans
->r_log_tid
= tid
;
1475 INIT_LIST_HEAD(&trans
->r_itemq
);
1477 INIT_HLIST_NODE(&trans
->r_list
);
1478 hlist_add_head(&trans
->r_list
, head
);
1482 xlog_recover_add_item(
1483 struct list_head
*head
)
1485 xlog_recover_item_t
*item
;
1487 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1488 INIT_LIST_HEAD(&item
->ri_list
);
1489 list_add_tail(&item
->ri_list
, head
);
1493 xlog_recover_add_to_cont_trans(
1495 struct xlog_recover
*trans
,
1499 xlog_recover_item_t
*item
;
1500 xfs_caddr_t ptr
, old_ptr
;
1503 if (list_empty(&trans
->r_itemq
)) {
1504 /* finish copying rest of trans header */
1505 xlog_recover_add_item(&trans
->r_itemq
);
1506 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1507 sizeof(xfs_trans_header_t
) - len
;
1508 memcpy(ptr
, dp
, len
); /* d, s, l */
1511 /* take the tail entry */
1512 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1514 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1515 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1517 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
1518 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1519 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1520 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1521 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1526 * The next region to add is the start of a new region. It could be
1527 * a whole region or it could be the first part of a new region. Because
1528 * of this, the assumption here is that the type and size fields of all
1529 * format structures fit into the first 32 bits of the structure.
1531 * This works because all regions must be 32 bit aligned. Therefore, we
1532 * either have both fields or we have neither field. In the case we have
1533 * neither field, the data part of the region is zero length. We only have
1534 * a log_op_header and can throw away the header since a new one will appear
1535 * later. If we have at least 4 bytes, then we can determine how many regions
1536 * will appear in the current log item.
1539 xlog_recover_add_to_trans(
1541 struct xlog_recover
*trans
,
1545 xfs_inode_log_format_t
*in_f
; /* any will do */
1546 xlog_recover_item_t
*item
;
1551 if (list_empty(&trans
->r_itemq
)) {
1552 /* we need to catch log corruptions here */
1553 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1554 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1557 return XFS_ERROR(EIO
);
1559 if (len
== sizeof(xfs_trans_header_t
))
1560 xlog_recover_add_item(&trans
->r_itemq
);
1561 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1565 ptr
= kmem_alloc(len
, KM_SLEEP
);
1566 memcpy(ptr
, dp
, len
);
1567 in_f
= (xfs_inode_log_format_t
*)ptr
;
1569 /* take the tail entry */
1570 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1571 if (item
->ri_total
!= 0 &&
1572 item
->ri_total
== item
->ri_cnt
) {
1573 /* tail item is in use, get a new one */
1574 xlog_recover_add_item(&trans
->r_itemq
);
1575 item
= list_entry(trans
->r_itemq
.prev
,
1576 xlog_recover_item_t
, ri_list
);
1579 if (item
->ri_total
== 0) { /* first region to be added */
1580 if (in_f
->ilf_size
== 0 ||
1581 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1583 "bad number of regions (%d) in inode log format",
1586 return XFS_ERROR(EIO
);
1589 item
->ri_total
= in_f
->ilf_size
;
1591 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1594 ASSERT(item
->ri_total
> item
->ri_cnt
);
1595 /* Description region is ri_buf[0] */
1596 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1597 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1599 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1604 * Sort the log items in the transaction.
1606 * The ordering constraints are defined by the inode allocation and unlink
1607 * behaviour. The rules are:
1609 * 1. Every item is only logged once in a given transaction. Hence it
1610 * represents the last logged state of the item. Hence ordering is
1611 * dependent on the order in which operations need to be performed so
1612 * required initial conditions are always met.
1614 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1615 * there's nothing to replay from them so we can simply cull them
1616 * from the transaction. However, we can't do that until after we've
1617 * replayed all the other items because they may be dependent on the
1618 * cancelled buffer and replaying the cancelled buffer can remove it
1619 * form the cancelled buffer table. Hence they have tobe done last.
1621 * 3. Inode allocation buffers must be replayed before inode items that
1622 * read the buffer and replay changes into it. For filesystems using the
1623 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1624 * treated the same as inode allocation buffers as they create and
1625 * initialise the buffers directly.
1627 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1628 * This ensures that inodes are completely flushed to the inode buffer
1629 * in a "free" state before we remove the unlinked inode list pointer.
1631 * Hence the ordering needs to be inode allocation buffers first, inode items
1632 * second, inode unlink buffers third and cancelled buffers last.
1634 * But there's a problem with that - we can't tell an inode allocation buffer
1635 * apart from a regular buffer, so we can't separate them. We can, however,
1636 * tell an inode unlink buffer from the others, and so we can separate them out
1637 * from all the other buffers and move them to last.
1639 * Hence, 4 lists, in order from head to tail:
1640 * - buffer_list for all buffers except cancelled/inode unlink buffers
1641 * - item_list for all non-buffer items
1642 * - inode_buffer_list for inode unlink buffers
1643 * - cancel_list for the cancelled buffers
1645 * Note that we add objects to the tail of the lists so that first-to-last
1646 * ordering is preserved within the lists. Adding objects to the head of the
1647 * list means when we traverse from the head we walk them in last-to-first
1648 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1649 * but for all other items there may be specific ordering that we need to
1653 xlog_recover_reorder_trans(
1655 struct xlog_recover
*trans
,
1658 xlog_recover_item_t
*item
, *n
;
1659 LIST_HEAD(sort_list
);
1660 LIST_HEAD(cancel_list
);
1661 LIST_HEAD(buffer_list
);
1662 LIST_HEAD(inode_buffer_list
);
1663 LIST_HEAD(inode_list
);
1665 list_splice_init(&trans
->r_itemq
, &sort_list
);
1666 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1667 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1669 switch (ITEM_TYPE(item
)) {
1670 case XFS_LI_ICREATE
:
1671 list_move_tail(&item
->ri_list
, &buffer_list
);
1674 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1675 trace_xfs_log_recover_item_reorder_head(log
,
1677 list_move(&item
->ri_list
, &cancel_list
);
1680 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1681 list_move(&item
->ri_list
, &inode_buffer_list
);
1684 list_move_tail(&item
->ri_list
, &buffer_list
);
1688 case XFS_LI_QUOTAOFF
:
1691 trace_xfs_log_recover_item_reorder_tail(log
,
1693 list_move_tail(&item
->ri_list
, &inode_list
);
1697 "%s: unrecognized type of log operation",
1700 return XFS_ERROR(EIO
);
1703 ASSERT(list_empty(&sort_list
));
1704 if (!list_empty(&buffer_list
))
1705 list_splice(&buffer_list
, &trans
->r_itemq
);
1706 if (!list_empty(&inode_list
))
1707 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1708 if (!list_empty(&inode_buffer_list
))
1709 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1710 if (!list_empty(&cancel_list
))
1711 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1716 * Build up the table of buf cancel records so that we don't replay
1717 * cancelled data in the second pass. For buffer records that are
1718 * not cancel records, there is nothing to do here so we just return.
1720 * If we get a cancel record which is already in the table, this indicates
1721 * that the buffer was cancelled multiple times. In order to ensure
1722 * that during pass 2 we keep the record in the table until we reach its
1723 * last occurrence in the log, we keep a reference count in the cancel
1724 * record in the table to tell us how many times we expect to see this
1725 * record during the second pass.
1728 xlog_recover_buffer_pass1(
1730 struct xlog_recover_item
*item
)
1732 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1733 struct list_head
*bucket
;
1734 struct xfs_buf_cancel
*bcp
;
1737 * If this isn't a cancel buffer item, then just return.
1739 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1740 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1745 * Insert an xfs_buf_cancel record into the hash table of them.
1746 * If there is already an identical record, bump its reference count.
1748 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1749 list_for_each_entry(bcp
, bucket
, bc_list
) {
1750 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1751 bcp
->bc_len
== buf_f
->blf_len
) {
1753 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1758 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1759 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1760 bcp
->bc_len
= buf_f
->blf_len
;
1761 bcp
->bc_refcount
= 1;
1762 list_add_tail(&bcp
->bc_list
, bucket
);
1764 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1769 * Check to see whether the buffer being recovered has a corresponding
1770 * entry in the buffer cancel record table. If it does then return 1
1771 * so that it will be cancelled, otherwise return 0. If the buffer is
1772 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1773 * the refcount on the entry in the table and remove it from the table
1774 * if this is the last reference.
1776 * We remove the cancel record from the table when we encounter its
1777 * last occurrence in the log so that if the same buffer is re-used
1778 * again after its last cancellation we actually replay the changes
1779 * made at that point.
1782 xlog_check_buffer_cancelled(
1788 struct list_head
*bucket
;
1789 struct xfs_buf_cancel
*bcp
;
1791 if (log
->l_buf_cancel_table
== NULL
) {
1793 * There is nothing in the table built in pass one,
1794 * so this buffer must not be cancelled.
1796 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1801 * Search for an entry in the cancel table that matches our buffer.
1803 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1804 list_for_each_entry(bcp
, bucket
, bc_list
) {
1805 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1810 * We didn't find a corresponding entry in the table, so return 0 so
1811 * that the buffer is NOT cancelled.
1813 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1818 * We've go a match, so return 1 so that the recovery of this buffer
1819 * is cancelled. If this buffer is actually a buffer cancel log
1820 * item, then decrement the refcount on the one in the table and
1821 * remove it if this is the last reference.
1823 if (flags
& XFS_BLF_CANCEL
) {
1824 if (--bcp
->bc_refcount
== 0) {
1825 list_del(&bcp
->bc_list
);
1833 * Perform recovery for a buffer full of inodes. In these buffers, the only
1834 * data which should be recovered is that which corresponds to the
1835 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1836 * data for the inodes is always logged through the inodes themselves rather
1837 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1839 * The only time when buffers full of inodes are fully recovered is when the
1840 * buffer is full of newly allocated inodes. In this case the buffer will
1841 * not be marked as an inode buffer and so will be sent to
1842 * xlog_recover_do_reg_buffer() below during recovery.
1845 xlog_recover_do_inode_buffer(
1846 struct xfs_mount
*mp
,
1847 xlog_recover_item_t
*item
,
1849 xfs_buf_log_format_t
*buf_f
)
1855 int reg_buf_offset
= 0;
1856 int reg_buf_bytes
= 0;
1857 int next_unlinked_offset
;
1859 xfs_agino_t
*logged_nextp
;
1860 xfs_agino_t
*buffer_nextp
;
1862 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1865 * Post recovery validation only works properly on CRC enabled
1868 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1869 bp
->b_ops
= &xfs_inode_buf_ops
;
1871 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1872 for (i
= 0; i
< inodes_per_buf
; i
++) {
1873 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1874 offsetof(xfs_dinode_t
, di_next_unlinked
);
1876 while (next_unlinked_offset
>=
1877 (reg_buf_offset
+ reg_buf_bytes
)) {
1879 * The next di_next_unlinked field is beyond
1880 * the current logged region. Find the next
1881 * logged region that contains or is beyond
1882 * the current di_next_unlinked field.
1885 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1886 buf_f
->blf_map_size
, bit
);
1889 * If there are no more logged regions in the
1890 * buffer, then we're done.
1895 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1896 buf_f
->blf_map_size
, bit
);
1898 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1899 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1904 * If the current logged region starts after the current
1905 * di_next_unlinked field, then move on to the next
1906 * di_next_unlinked field.
1908 if (next_unlinked_offset
< reg_buf_offset
)
1911 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1912 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1913 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1914 BBTOB(bp
->b_io_length
));
1917 * The current logged region contains a copy of the
1918 * current di_next_unlinked field. Extract its value
1919 * and copy it to the buffer copy.
1921 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1922 next_unlinked_offset
- reg_buf_offset
;
1923 if (unlikely(*logged_nextp
== 0)) {
1925 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1926 "Trying to replay bad (0) inode di_next_unlinked field.",
1928 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1929 XFS_ERRLEVEL_LOW
, mp
);
1930 return XFS_ERROR(EFSCORRUPTED
);
1933 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1934 next_unlinked_offset
);
1935 *buffer_nextp
= *logged_nextp
;
1938 * If necessary, recalculate the CRC in the on-disk inode. We
1939 * have to leave the inode in a consistent state for whoever
1942 xfs_dinode_calc_crc(mp
, (struct xfs_dinode
*)
1943 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
1951 * Validate the recovered buffer is of the correct type and attach the
1952 * appropriate buffer operations to them for writeback. Magic numbers are in a
1954 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1955 * the first 32 bits of the buffer (most blocks),
1956 * inside a struct xfs_da_blkinfo at the start of the buffer.
1959 xlog_recovery_validate_buf_type(
1960 struct xfs_mount
*mp
,
1962 xfs_buf_log_format_t
*buf_f
)
1964 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
1969 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
1970 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
1971 magicda
= be16_to_cpu(info
->magic
);
1972 switch (xfs_blft_from_flags(buf_f
)) {
1973 case XFS_BLFT_BTREE_BUF
:
1975 case XFS_ABTB_CRC_MAGIC
:
1976 case XFS_ABTC_CRC_MAGIC
:
1977 case XFS_ABTB_MAGIC
:
1978 case XFS_ABTC_MAGIC
:
1979 bp
->b_ops
= &xfs_allocbt_buf_ops
;
1981 case XFS_IBT_CRC_MAGIC
:
1983 bp
->b_ops
= &xfs_inobt_buf_ops
;
1985 case XFS_BMAP_CRC_MAGIC
:
1986 case XFS_BMAP_MAGIC
:
1987 bp
->b_ops
= &xfs_bmbt_buf_ops
;
1990 xfs_warn(mp
, "Bad btree block magic!");
1995 case XFS_BLFT_AGF_BUF
:
1996 if (magic32
!= XFS_AGF_MAGIC
) {
1997 xfs_warn(mp
, "Bad AGF block magic!");
2001 bp
->b_ops
= &xfs_agf_buf_ops
;
2003 case XFS_BLFT_AGFL_BUF
:
2004 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2006 if (magic32
!= XFS_AGFL_MAGIC
) {
2007 xfs_warn(mp
, "Bad AGFL block magic!");
2011 bp
->b_ops
= &xfs_agfl_buf_ops
;
2013 case XFS_BLFT_AGI_BUF
:
2014 if (magic32
!= XFS_AGI_MAGIC
) {
2015 xfs_warn(mp
, "Bad AGI block magic!");
2019 bp
->b_ops
= &xfs_agi_buf_ops
;
2021 case XFS_BLFT_UDQUOT_BUF
:
2022 case XFS_BLFT_PDQUOT_BUF
:
2023 case XFS_BLFT_GDQUOT_BUF
:
2024 #ifdef CONFIG_XFS_QUOTA
2025 if (magic16
!= XFS_DQUOT_MAGIC
) {
2026 xfs_warn(mp
, "Bad DQUOT block magic!");
2030 bp
->b_ops
= &xfs_dquot_buf_ops
;
2033 "Trying to recover dquots without QUOTA support built in!");
2037 case XFS_BLFT_DINO_BUF
:
2039 * we get here with inode allocation buffers, not buffers that
2040 * track unlinked list changes.
2042 if (magic16
!= XFS_DINODE_MAGIC
) {
2043 xfs_warn(mp
, "Bad INODE block magic!");
2047 bp
->b_ops
= &xfs_inode_buf_ops
;
2049 case XFS_BLFT_SYMLINK_BUF
:
2050 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2051 xfs_warn(mp
, "Bad symlink block magic!");
2055 bp
->b_ops
= &xfs_symlink_buf_ops
;
2057 case XFS_BLFT_DIR_BLOCK_BUF
:
2058 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2059 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2060 xfs_warn(mp
, "Bad dir block magic!");
2064 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2066 case XFS_BLFT_DIR_DATA_BUF
:
2067 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2068 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2069 xfs_warn(mp
, "Bad dir data magic!");
2073 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2075 case XFS_BLFT_DIR_FREE_BUF
:
2076 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2077 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2078 xfs_warn(mp
, "Bad dir3 free magic!");
2082 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2084 case XFS_BLFT_DIR_LEAF1_BUF
:
2085 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2086 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2087 xfs_warn(mp
, "Bad dir leaf1 magic!");
2091 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2093 case XFS_BLFT_DIR_LEAFN_BUF
:
2094 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2095 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2096 xfs_warn(mp
, "Bad dir leafn magic!");
2100 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2102 case XFS_BLFT_DA_NODE_BUF
:
2103 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2104 magicda
!= XFS_DA3_NODE_MAGIC
) {
2105 xfs_warn(mp
, "Bad da node magic!");
2109 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2111 case XFS_BLFT_ATTR_LEAF_BUF
:
2112 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2113 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2114 xfs_warn(mp
, "Bad attr leaf magic!");
2118 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2120 case XFS_BLFT_ATTR_RMT_BUF
:
2121 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2123 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2124 xfs_warn(mp
, "Bad attr remote magic!");
2128 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2130 case XFS_BLFT_SB_BUF
:
2131 if (magic32
!= XFS_SB_MAGIC
) {
2132 xfs_warn(mp
, "Bad SB block magic!");
2136 bp
->b_ops
= &xfs_sb_buf_ops
;
2139 xfs_warn(mp
, "Unknown buffer type %d!",
2140 xfs_blft_from_flags(buf_f
));
2146 * Perform a 'normal' buffer recovery. Each logged region of the
2147 * buffer should be copied over the corresponding region in the
2148 * given buffer. The bitmap in the buf log format structure indicates
2149 * where to place the logged data.
2152 xlog_recover_do_reg_buffer(
2153 struct xfs_mount
*mp
,
2154 xlog_recover_item_t
*item
,
2156 xfs_buf_log_format_t
*buf_f
)
2163 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2166 i
= 1; /* 0 is the buf format structure */
2168 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2169 buf_f
->blf_map_size
, bit
);
2172 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2173 buf_f
->blf_map_size
, bit
);
2175 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2176 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2177 ASSERT(BBTOB(bp
->b_io_length
) >=
2178 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2181 * The dirty regions logged in the buffer, even though
2182 * contiguous, may span multiple chunks. This is because the
2183 * dirty region may span a physical page boundary in a buffer
2184 * and hence be split into two separate vectors for writing into
2185 * the log. Hence we need to trim nbits back to the length of
2186 * the current region being copied out of the log.
2188 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2189 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2192 * Do a sanity check if this is a dquot buffer. Just checking
2193 * the first dquot in the buffer should do. XXXThis is
2194 * probably a good thing to do for other buf types also.
2197 if (buf_f
->blf_flags
&
2198 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2199 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2201 "XFS: NULL dquot in %s.", __func__
);
2204 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2206 "XFS: dquot too small (%d) in %s.",
2207 item
->ri_buf
[i
].i_len
, __func__
);
2210 error
= xfs_qm_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2211 -1, 0, XFS_QMOPT_DOWARN
,
2212 "dquot_buf_recover");
2217 memcpy(xfs_buf_offset(bp
,
2218 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2219 item
->ri_buf
[i
].i_addr
, /* source */
2220 nbits
<<XFS_BLF_SHIFT
); /* length */
2226 /* Shouldn't be any more regions */
2227 ASSERT(i
== item
->ri_total
);
2230 * We can only do post recovery validation on items on CRC enabled
2231 * fielsystems as we need to know when the buffer was written to be able
2232 * to determine if we should have replayed the item. If we replay old
2233 * metadata over a newer buffer, then it will enter a temporarily
2234 * inconsistent state resulting in verification failures. Hence for now
2235 * just avoid the verification stage for non-crc filesystems
2237 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2238 xlog_recovery_validate_buf_type(mp
, bp
, buf_f
);
2242 * Do some primitive error checking on ondisk dquot data structures.
2246 struct xfs_mount
*mp
,
2247 xfs_disk_dquot_t
*ddq
,
2249 uint type
, /* used only when IO_dorepair is true */
2253 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2257 * We can encounter an uninitialized dquot buffer for 2 reasons:
2258 * 1. If we crash while deleting the quotainode(s), and those blks got
2259 * used for user data. This is because we take the path of regular
2260 * file deletion; however, the size field of quotainodes is never
2261 * updated, so all the tricks that we play in itruncate_finish
2262 * don't quite matter.
2264 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2265 * But the allocation will be replayed so we'll end up with an
2266 * uninitialized quota block.
2268 * This is all fine; things are still consistent, and we haven't lost
2269 * any quota information. Just don't complain about bad dquot blks.
2271 if (ddq
->d_magic
!= cpu_to_be16(XFS_DQUOT_MAGIC
)) {
2272 if (flags
& XFS_QMOPT_DOWARN
)
2274 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2275 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2278 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2279 if (flags
& XFS_QMOPT_DOWARN
)
2281 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2282 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2286 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2287 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2288 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2289 if (flags
& XFS_QMOPT_DOWARN
)
2291 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2292 str
, id
, ddq
->d_flags
);
2296 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2297 if (flags
& XFS_QMOPT_DOWARN
)
2299 "%s : ondisk-dquot 0x%p, ID mismatch: "
2300 "0x%x expected, found id 0x%x",
2301 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2305 if (!errs
&& ddq
->d_id
) {
2306 if (ddq
->d_blk_softlimit
&&
2307 be64_to_cpu(ddq
->d_bcount
) >
2308 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2309 if (!ddq
->d_btimer
) {
2310 if (flags
& XFS_QMOPT_DOWARN
)
2312 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
2313 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2317 if (ddq
->d_ino_softlimit
&&
2318 be64_to_cpu(ddq
->d_icount
) >
2319 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2320 if (!ddq
->d_itimer
) {
2321 if (flags
& XFS_QMOPT_DOWARN
)
2323 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2324 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2328 if (ddq
->d_rtb_softlimit
&&
2329 be64_to_cpu(ddq
->d_rtbcount
) >
2330 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2331 if (!ddq
->d_rtbtimer
) {
2332 if (flags
& XFS_QMOPT_DOWARN
)
2334 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2335 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2341 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2344 if (flags
& XFS_QMOPT_DOWARN
)
2345 xfs_notice(mp
, "Re-initializing dquot ID 0x%x", id
);
2348 * Typically, a repair is only requested by quotacheck.
2351 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2352 memset(d
, 0, sizeof(xfs_dqblk_t
));
2354 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2355 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2356 d
->dd_diskdq
.d_flags
= type
;
2357 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2359 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2360 uuid_copy(&d
->dd_uuid
, &mp
->m_sb
.sb_uuid
);
2361 xfs_update_cksum((char *)d
, sizeof(struct xfs_dqblk
),
2369 * Perform a dquot buffer recovery.
2370 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2371 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2372 * Else, treat it as a regular buffer and do recovery.
2375 xlog_recover_do_dquot_buffer(
2376 struct xfs_mount
*mp
,
2378 struct xlog_recover_item
*item
,
2380 struct xfs_buf_log_format
*buf_f
)
2384 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2387 * Filesystems are required to send in quota flags at mount time.
2389 if (mp
->m_qflags
== 0) {
2394 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2395 type
|= XFS_DQ_USER
;
2396 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2397 type
|= XFS_DQ_PROJ
;
2398 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2399 type
|= XFS_DQ_GROUP
;
2401 * This type of quotas was turned off, so ignore this buffer
2403 if (log
->l_quotaoffs_flag
& type
)
2406 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2410 * This routine replays a modification made to a buffer at runtime.
2411 * There are actually two types of buffer, regular and inode, which
2412 * are handled differently. Inode buffers are handled differently
2413 * in that we only recover a specific set of data from them, namely
2414 * the inode di_next_unlinked fields. This is because all other inode
2415 * data is actually logged via inode records and any data we replay
2416 * here which overlaps that may be stale.
2418 * When meta-data buffers are freed at run time we log a buffer item
2419 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2420 * of the buffer in the log should not be replayed at recovery time.
2421 * This is so that if the blocks covered by the buffer are reused for
2422 * file data before we crash we don't end up replaying old, freed
2423 * meta-data into a user's file.
2425 * To handle the cancellation of buffer log items, we make two passes
2426 * over the log during recovery. During the first we build a table of
2427 * those buffers which have been cancelled, and during the second we
2428 * only replay those buffers which do not have corresponding cancel
2429 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2430 * for more details on the implementation of the table of cancel records.
2433 xlog_recover_buffer_pass2(
2435 struct list_head
*buffer_list
,
2436 struct xlog_recover_item
*item
)
2438 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2439 xfs_mount_t
*mp
= log
->l_mp
;
2445 * In this pass we only want to recover all the buffers which have
2446 * not been cancelled and are not cancellation buffers themselves.
2448 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2449 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2450 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2454 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2457 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2458 buf_flags
|= XBF_UNMAPPED
;
2460 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2463 return XFS_ERROR(ENOMEM
);
2464 error
= bp
->b_error
;
2466 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2471 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2472 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2473 } else if (buf_f
->blf_flags
&
2474 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2475 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2477 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2480 return XFS_ERROR(error
);
2483 * Perform delayed write on the buffer. Asynchronous writes will be
2484 * slower when taking into account all the buffers to be flushed.
2486 * Also make sure that only inode buffers with good sizes stay in
2487 * the buffer cache. The kernel moves inodes in buffers of 1 block
2488 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2489 * buffers in the log can be a different size if the log was generated
2490 * by an older kernel using unclustered inode buffers or a newer kernel
2491 * running with a different inode cluster size. Regardless, if the
2492 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2493 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2494 * the buffer out of the buffer cache so that the buffer won't
2495 * overlap with future reads of those inodes.
2497 if (XFS_DINODE_MAGIC
==
2498 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2499 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2500 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2502 error
= xfs_bwrite(bp
);
2504 ASSERT(bp
->b_target
->bt_mount
== mp
);
2505 bp
->b_iodone
= xlog_recover_iodone
;
2506 xfs_buf_delwri_queue(bp
, buffer_list
);
2514 xlog_recover_inode_pass2(
2516 struct list_head
*buffer_list
,
2517 struct xlog_recover_item
*item
)
2519 xfs_inode_log_format_t
*in_f
;
2520 xfs_mount_t
*mp
= log
->l_mp
;
2529 xfs_icdinode_t
*dicp
;
2533 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2534 in_f
= item
->ri_buf
[0].i_addr
;
2536 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2538 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2544 * Inode buffers can be freed, look out for it,
2545 * and do not replay the inode.
2547 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2548 in_f
->ilf_len
, 0)) {
2550 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2553 trace_xfs_log_recover_inode_recover(log
, in_f
);
2555 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2556 &xfs_inode_buf_ops
);
2561 error
= bp
->b_error
;
2563 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2567 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2568 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2571 * Make sure the place we're flushing out to really looks
2574 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2577 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2578 __func__
, dip
, bp
, in_f
->ilf_ino
);
2579 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2580 XFS_ERRLEVEL_LOW
, mp
);
2581 error
= EFSCORRUPTED
;
2584 dicp
= item
->ri_buf
[1].i_addr
;
2585 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2588 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2589 __func__
, item
, in_f
->ilf_ino
);
2590 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2591 XFS_ERRLEVEL_LOW
, mp
);
2592 error
= EFSCORRUPTED
;
2597 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2598 * are transactional and if ordering is necessary we can determine that
2599 * more accurately by the LSN field in the V3 inode core. Don't trust
2600 * the inode versions we might be changing them here - use the
2601 * superblock flag to determine whether we need to look at di_flushiter
2602 * to skip replay when the on disk inode is newer than the log one
2604 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2605 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2607 * Deal with the wrap case, DI_MAX_FLUSH is less
2608 * than smaller numbers
2610 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2611 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2615 trace_xfs_log_recover_inode_skip(log
, in_f
);
2621 /* Take the opportunity to reset the flush iteration count */
2622 dicp
->di_flushiter
= 0;
2624 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2625 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2626 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2627 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2628 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2631 "%s: Bad regular inode log record, rec ptr 0x%p, "
2632 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2633 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2634 error
= EFSCORRUPTED
;
2637 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2638 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2639 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2640 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2641 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2642 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2645 "%s: Bad dir inode log record, rec ptr 0x%p, "
2646 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2647 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2648 error
= EFSCORRUPTED
;
2652 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2653 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2654 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2657 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2658 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2659 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2660 dicp
->di_nextents
+ dicp
->di_anextents
,
2662 error
= EFSCORRUPTED
;
2665 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2666 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2667 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2670 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2671 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2672 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2673 error
= EFSCORRUPTED
;
2676 isize
= xfs_icdinode_size(dicp
->di_version
);
2677 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2678 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2679 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2682 "%s: Bad inode log record length %d, rec ptr 0x%p",
2683 __func__
, item
->ri_buf
[1].i_len
, item
);
2684 error
= EFSCORRUPTED
;
2688 /* The core is in in-core format */
2689 xfs_dinode_to_disk(dip
, dicp
);
2691 /* the rest is in on-disk format */
2692 if (item
->ri_buf
[1].i_len
> isize
) {
2693 memcpy((char *)dip
+ isize
,
2694 item
->ri_buf
[1].i_addr
+ isize
,
2695 item
->ri_buf
[1].i_len
- isize
);
2698 fields
= in_f
->ilf_fields
;
2699 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2701 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2704 memcpy(XFS_DFORK_DPTR(dip
),
2705 &in_f
->ilf_u
.ilfu_uuid
,
2710 if (in_f
->ilf_size
== 2)
2711 goto write_inode_buffer
;
2712 len
= item
->ri_buf
[2].i_len
;
2713 src
= item
->ri_buf
[2].i_addr
;
2714 ASSERT(in_f
->ilf_size
<= 4);
2715 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2716 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2717 (len
== in_f
->ilf_dsize
));
2719 switch (fields
& XFS_ILOG_DFORK
) {
2720 case XFS_ILOG_DDATA
:
2722 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2725 case XFS_ILOG_DBROOT
:
2726 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2727 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2728 XFS_DFORK_DSIZE(dip
, mp
));
2733 * There are no data fork flags set.
2735 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2740 * If we logged any attribute data, recover it. There may or
2741 * may not have been any other non-core data logged in this
2744 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2745 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2750 len
= item
->ri_buf
[attr_index
].i_len
;
2751 src
= item
->ri_buf
[attr_index
].i_addr
;
2752 ASSERT(len
== in_f
->ilf_asize
);
2754 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2755 case XFS_ILOG_ADATA
:
2757 dest
= XFS_DFORK_APTR(dip
);
2758 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2759 memcpy(dest
, src
, len
);
2762 case XFS_ILOG_ABROOT
:
2763 dest
= XFS_DFORK_APTR(dip
);
2764 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2765 len
, (xfs_bmdr_block_t
*)dest
,
2766 XFS_DFORK_ASIZE(dip
, mp
));
2770 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2779 /* re-generate the checksum. */
2780 xfs_dinode_calc_crc(log
->l_mp
, dip
);
2782 ASSERT(bp
->b_target
->bt_mount
== mp
);
2783 bp
->b_iodone
= xlog_recover_iodone
;
2784 xfs_buf_delwri_queue(bp
, buffer_list
);
2789 return XFS_ERROR(error
);
2793 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2794 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2798 xlog_recover_quotaoff_pass1(
2800 struct xlog_recover_item
*item
)
2802 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2806 * The logitem format's flag tells us if this was user quotaoff,
2807 * group/project quotaoff or both.
2809 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2810 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2811 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2812 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2813 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2814 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2820 * Recover a dquot record
2823 xlog_recover_dquot_pass2(
2825 struct list_head
*buffer_list
,
2826 struct xlog_recover_item
*item
)
2828 xfs_mount_t
*mp
= log
->l_mp
;
2830 struct xfs_disk_dquot
*ddq
, *recddq
;
2832 xfs_dq_logformat_t
*dq_f
;
2837 * Filesystems are required to send in quota flags at mount time.
2839 if (mp
->m_qflags
== 0)
2842 recddq
= item
->ri_buf
[1].i_addr
;
2843 if (recddq
== NULL
) {
2844 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2845 return XFS_ERROR(EIO
);
2847 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2848 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2849 item
->ri_buf
[1].i_len
, __func__
);
2850 return XFS_ERROR(EIO
);
2854 * This type of quotas was turned off, so ignore this record.
2856 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2858 if (log
->l_quotaoffs_flag
& type
)
2862 * At this point we know that quota was _not_ turned off.
2863 * Since the mount flags are not indicating to us otherwise, this
2864 * must mean that quota is on, and the dquot needs to be replayed.
2865 * Remember that we may not have fully recovered the superblock yet,
2866 * so we can't do the usual trick of looking at the SB quota bits.
2868 * The other possibility, of course, is that the quota subsystem was
2869 * removed since the last mount - ENOSYS.
2871 dq_f
= item
->ri_buf
[0].i_addr
;
2873 error
= xfs_qm_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2874 "xlog_recover_dquot_pass2 (log copy)");
2876 return XFS_ERROR(EIO
);
2877 ASSERT(dq_f
->qlf_len
== 1);
2879 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
2880 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
2886 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2889 * At least the magic num portion should be on disk because this
2890 * was among a chunk of dquots created earlier, and we did some
2891 * minimal initialization then.
2893 error
= xfs_qm_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2894 "xlog_recover_dquot_pass2");
2897 return XFS_ERROR(EIO
);
2900 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2901 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2902 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
2906 ASSERT(dq_f
->qlf_size
== 2);
2907 ASSERT(bp
->b_target
->bt_mount
== mp
);
2908 bp
->b_iodone
= xlog_recover_iodone
;
2909 xfs_buf_delwri_queue(bp
, buffer_list
);
2916 * This routine is called to create an in-core extent free intent
2917 * item from the efi format structure which was logged on disk.
2918 * It allocates an in-core efi, copies the extents from the format
2919 * structure into it, and adds the efi to the AIL with the given
2923 xlog_recover_efi_pass2(
2925 struct xlog_recover_item
*item
,
2929 xfs_mount_t
*mp
= log
->l_mp
;
2930 xfs_efi_log_item_t
*efip
;
2931 xfs_efi_log_format_t
*efi_formatp
;
2933 efi_formatp
= item
->ri_buf
[0].i_addr
;
2935 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2936 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2937 &(efip
->efi_format
)))) {
2938 xfs_efi_item_free(efip
);
2941 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
2943 spin_lock(&log
->l_ailp
->xa_lock
);
2945 * xfs_trans_ail_update() drops the AIL lock.
2947 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
2953 * This routine is called when an efd format structure is found in
2954 * a committed transaction in the log. It's purpose is to cancel
2955 * the corresponding efi if it was still in the log. To do this
2956 * it searches the AIL for the efi with an id equal to that in the
2957 * efd format structure. If we find it, we remove the efi from the
2961 xlog_recover_efd_pass2(
2963 struct xlog_recover_item
*item
)
2965 xfs_efd_log_format_t
*efd_formatp
;
2966 xfs_efi_log_item_t
*efip
= NULL
;
2967 xfs_log_item_t
*lip
;
2969 struct xfs_ail_cursor cur
;
2970 struct xfs_ail
*ailp
= log
->l_ailp
;
2972 efd_formatp
= item
->ri_buf
[0].i_addr
;
2973 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2974 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2975 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2976 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2977 efi_id
= efd_formatp
->efd_efi_id
;
2980 * Search for the efi with the id in the efd format structure
2983 spin_lock(&ailp
->xa_lock
);
2984 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2985 while (lip
!= NULL
) {
2986 if (lip
->li_type
== XFS_LI_EFI
) {
2987 efip
= (xfs_efi_log_item_t
*)lip
;
2988 if (efip
->efi_format
.efi_id
== efi_id
) {
2990 * xfs_trans_ail_delete() drops the
2993 xfs_trans_ail_delete(ailp
, lip
,
2994 SHUTDOWN_CORRUPT_INCORE
);
2995 xfs_efi_item_free(efip
);
2996 spin_lock(&ailp
->xa_lock
);
3000 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3002 xfs_trans_ail_cursor_done(ailp
, &cur
);
3003 spin_unlock(&ailp
->xa_lock
);
3009 * This routine is called when an inode create format structure is found in a
3010 * committed transaction in the log. It's purpose is to initialise the inodes
3011 * being allocated on disk. This requires us to get inode cluster buffers that
3012 * match the range to be intialised, stamped with inode templates and written
3013 * by delayed write so that subsequent modifications will hit the cached buffer
3014 * and only need writing out at the end of recovery.
3017 xlog_recover_do_icreate_pass2(
3019 struct list_head
*buffer_list
,
3020 xlog_recover_item_t
*item
)
3022 struct xfs_mount
*mp
= log
->l_mp
;
3023 struct xfs_icreate_log
*icl
;
3024 xfs_agnumber_t agno
;
3025 xfs_agblock_t agbno
;
3028 xfs_agblock_t length
;
3030 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3031 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3032 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3036 if (icl
->icl_size
!= 1) {
3037 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3041 agno
= be32_to_cpu(icl
->icl_ag
);
3042 if (agno
>= mp
->m_sb
.sb_agcount
) {
3043 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3046 agbno
= be32_to_cpu(icl
->icl_agbno
);
3047 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3048 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3051 isize
= be32_to_cpu(icl
->icl_isize
);
3052 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3053 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3056 count
= be32_to_cpu(icl
->icl_count
);
3058 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3061 length
= be32_to_cpu(icl
->icl_length
);
3062 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3063 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3067 /* existing allocation is fixed value */
3068 ASSERT(count
== XFS_IALLOC_INODES(mp
));
3069 ASSERT(length
== XFS_IALLOC_BLOCKS(mp
));
3070 if (count
!= XFS_IALLOC_INODES(mp
) ||
3071 length
!= XFS_IALLOC_BLOCKS(mp
)) {
3072 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count 2");
3077 * Inode buffers can be freed. Do not replay the inode initialisation as
3078 * we could be overwriting something written after this inode buffer was
3081 * XXX: we need to iterate all buffers and only init those that are not
3082 * cancelled. I think that a more fine grained factoring of
3083 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3086 if (xlog_check_buffer_cancelled(log
,
3087 XFS_AGB_TO_DADDR(mp
, agno
, agbno
), length
, 0))
3090 xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, agno
, agbno
, length
,
3091 be32_to_cpu(icl
->icl_gen
));
3096 * Free up any resources allocated by the transaction
3098 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3101 xlog_recover_free_trans(
3102 struct xlog_recover
*trans
)
3104 xlog_recover_item_t
*item
, *n
;
3107 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3108 /* Free the regions in the item. */
3109 list_del(&item
->ri_list
);
3110 for (i
= 0; i
< item
->ri_cnt
; i
++)
3111 kmem_free(item
->ri_buf
[i
].i_addr
);
3112 /* Free the item itself */
3113 kmem_free(item
->ri_buf
);
3116 /* Free the transaction recover structure */
3121 xlog_recover_commit_pass1(
3123 struct xlog_recover
*trans
,
3124 struct xlog_recover_item
*item
)
3126 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3128 switch (ITEM_TYPE(item
)) {
3130 return xlog_recover_buffer_pass1(log
, item
);
3131 case XFS_LI_QUOTAOFF
:
3132 return xlog_recover_quotaoff_pass1(log
, item
);
3137 case XFS_LI_ICREATE
:
3138 /* nothing to do in pass 1 */
3141 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3142 __func__
, ITEM_TYPE(item
));
3144 return XFS_ERROR(EIO
);
3149 xlog_recover_commit_pass2(
3151 struct xlog_recover
*trans
,
3152 struct list_head
*buffer_list
,
3153 struct xlog_recover_item
*item
)
3155 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3157 switch (ITEM_TYPE(item
)) {
3159 return xlog_recover_buffer_pass2(log
, buffer_list
, item
);
3161 return xlog_recover_inode_pass2(log
, buffer_list
, item
);
3163 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3165 return xlog_recover_efd_pass2(log
, item
);
3167 return xlog_recover_dquot_pass2(log
, buffer_list
, item
);
3168 case XFS_LI_ICREATE
:
3169 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3170 case XFS_LI_QUOTAOFF
:
3171 /* nothing to do in pass2 */
3174 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3175 __func__
, ITEM_TYPE(item
));
3177 return XFS_ERROR(EIO
);
3182 * Perform the transaction.
3184 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3185 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3188 xlog_recover_commit_trans(
3190 struct xlog_recover
*trans
,
3193 int error
= 0, error2
;
3194 xlog_recover_item_t
*item
;
3195 LIST_HEAD (buffer_list
);
3197 hlist_del(&trans
->r_list
);
3199 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3203 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
3205 case XLOG_RECOVER_PASS1
:
3206 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3208 case XLOG_RECOVER_PASS2
:
3209 error
= xlog_recover_commit_pass2(log
, trans
,
3210 &buffer_list
, item
);
3220 xlog_recover_free_trans(trans
);
3223 error2
= xfs_buf_delwri_submit(&buffer_list
);
3224 return error
? error
: error2
;
3228 xlog_recover_unmount_trans(
3230 struct xlog_recover
*trans
)
3232 /* Do nothing now */
3233 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3238 * There are two valid states of the r_state field. 0 indicates that the
3239 * transaction structure is in a normal state. We have either seen the
3240 * start of the transaction or the last operation we added was not a partial
3241 * operation. If the last operation we added to the transaction was a
3242 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3244 * NOTE: skip LRs with 0 data length.
3247 xlog_recover_process_data(
3249 struct hlist_head rhash
[],
3250 struct xlog_rec_header
*rhead
,
3256 xlog_op_header_t
*ohead
;
3257 xlog_recover_t
*trans
;
3263 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
3264 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
3266 /* check the log format matches our own - else we can't recover */
3267 if (xlog_header_check_recover(log
->l_mp
, rhead
))
3268 return (XFS_ERROR(EIO
));
3270 while ((dp
< lp
) && num_logops
) {
3271 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
3272 ohead
= (xlog_op_header_t
*)dp
;
3273 dp
+= sizeof(xlog_op_header_t
);
3274 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
3275 ohead
->oh_clientid
!= XFS_LOG
) {
3276 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
3277 __func__
, ohead
->oh_clientid
);
3279 return (XFS_ERROR(EIO
));
3281 tid
= be32_to_cpu(ohead
->oh_tid
);
3282 hash
= XLOG_RHASH(tid
);
3283 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
3284 if (trans
== NULL
) { /* not found; add new tid */
3285 if (ohead
->oh_flags
& XLOG_START_TRANS
)
3286 xlog_recover_new_tid(&rhash
[hash
], tid
,
3287 be64_to_cpu(rhead
->h_lsn
));
3289 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
3290 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
3291 __func__
, be32_to_cpu(ohead
->oh_len
));
3293 return (XFS_ERROR(EIO
));
3295 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
3296 if (flags
& XLOG_WAS_CONT_TRANS
)
3297 flags
&= ~XLOG_CONTINUE_TRANS
;
3299 case XLOG_COMMIT_TRANS
:
3300 error
= xlog_recover_commit_trans(log
,
3303 case XLOG_UNMOUNT_TRANS
:
3304 error
= xlog_recover_unmount_trans(log
, trans
);
3306 case XLOG_WAS_CONT_TRANS
:
3307 error
= xlog_recover_add_to_cont_trans(log
,
3309 be32_to_cpu(ohead
->oh_len
));
3311 case XLOG_START_TRANS
:
3312 xfs_warn(log
->l_mp
, "%s: bad transaction",
3315 error
= XFS_ERROR(EIO
);
3318 case XLOG_CONTINUE_TRANS
:
3319 error
= xlog_recover_add_to_trans(log
, trans
,
3320 dp
, be32_to_cpu(ohead
->oh_len
));
3323 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
3326 error
= XFS_ERROR(EIO
);
3332 dp
+= be32_to_cpu(ohead
->oh_len
);
3339 * Process an extent free intent item that was recovered from
3340 * the log. We need to free the extents that it describes.
3343 xlog_recover_process_efi(
3345 xfs_efi_log_item_t
*efip
)
3347 xfs_efd_log_item_t
*efdp
;
3352 xfs_fsblock_t startblock_fsb
;
3354 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
3357 * First check the validity of the extents described by the
3358 * EFI. If any are bad, then assume that all are bad and
3359 * just toss the EFI.
3361 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3362 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3363 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3364 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3365 if ((startblock_fsb
== 0) ||
3366 (extp
->ext_len
== 0) ||
3367 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3368 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3370 * This will pull the EFI from the AIL and
3371 * free the memory associated with it.
3373 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3374 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3375 return XFS_ERROR(EIO
);
3379 tp
= xfs_trans_alloc(mp
, 0);
3380 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
3383 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3385 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3386 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3387 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3390 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3394 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3395 error
= xfs_trans_commit(tp
, 0);
3399 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3404 * When this is called, all of the EFIs which did not have
3405 * corresponding EFDs should be in the AIL. What we do now
3406 * is free the extents associated with each one.
3408 * Since we process the EFIs in normal transactions, they
3409 * will be removed at some point after the commit. This prevents
3410 * us from just walking down the list processing each one.
3411 * We'll use a flag in the EFI to skip those that we've already
3412 * processed and use the AIL iteration mechanism's generation
3413 * count to try to speed this up at least a bit.
3415 * When we start, we know that the EFIs are the only things in
3416 * the AIL. As we process them, however, other items are added
3417 * to the AIL. Since everything added to the AIL must come after
3418 * everything already in the AIL, we stop processing as soon as
3419 * we see something other than an EFI in the AIL.
3422 xlog_recover_process_efis(
3425 xfs_log_item_t
*lip
;
3426 xfs_efi_log_item_t
*efip
;
3428 struct xfs_ail_cursor cur
;
3429 struct xfs_ail
*ailp
;
3432 spin_lock(&ailp
->xa_lock
);
3433 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3434 while (lip
!= NULL
) {
3436 * We're done when we see something other than an EFI.
3437 * There should be no EFIs left in the AIL now.
3439 if (lip
->li_type
!= XFS_LI_EFI
) {
3441 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3442 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3448 * Skip EFIs that we've already processed.
3450 efip
= (xfs_efi_log_item_t
*)lip
;
3451 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3452 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3456 spin_unlock(&ailp
->xa_lock
);
3457 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3458 spin_lock(&ailp
->xa_lock
);
3461 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3464 xfs_trans_ail_cursor_done(ailp
, &cur
);
3465 spin_unlock(&ailp
->xa_lock
);
3470 * This routine performs a transaction to null out a bad inode pointer
3471 * in an agi unlinked inode hash bucket.
3474 xlog_recover_clear_agi_bucket(
3476 xfs_agnumber_t agno
,
3485 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3486 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
3490 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3494 agi
= XFS_BUF_TO_AGI(agibp
);
3495 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3496 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3497 (sizeof(xfs_agino_t
) * bucket
);
3498 xfs_trans_log_buf(tp
, agibp
, offset
,
3499 (offset
+ sizeof(xfs_agino_t
) - 1));
3501 error
= xfs_trans_commit(tp
, 0);
3507 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3509 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3514 xlog_recover_process_one_iunlink(
3515 struct xfs_mount
*mp
,
3516 xfs_agnumber_t agno
,
3520 struct xfs_buf
*ibp
;
3521 struct xfs_dinode
*dip
;
3522 struct xfs_inode
*ip
;
3526 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3527 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3532 * Get the on disk inode to find the next inode in the bucket.
3534 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3538 ASSERT(ip
->i_d
.di_nlink
== 0);
3539 ASSERT(ip
->i_d
.di_mode
!= 0);
3541 /* setup for the next pass */
3542 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3546 * Prevent any DMAPI event from being sent when the reference on
3547 * the inode is dropped.
3549 ip
->i_d
.di_dmevmask
= 0;
3558 * We can't read in the inode this bucket points to, or this inode
3559 * is messed up. Just ditch this bucket of inodes. We will lose
3560 * some inodes and space, but at least we won't hang.
3562 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3563 * clear the inode pointer in the bucket.
3565 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3570 * xlog_iunlink_recover
3572 * This is called during recovery to process any inodes which
3573 * we unlinked but not freed when the system crashed. These
3574 * inodes will be on the lists in the AGI blocks. What we do
3575 * here is scan all the AGIs and fully truncate and free any
3576 * inodes found on the lists. Each inode is removed from the
3577 * lists when it has been fully truncated and is freed. The
3578 * freeing of the inode and its removal from the list must be
3582 xlog_recover_process_iunlinks(
3586 xfs_agnumber_t agno
;
3597 * Prevent any DMAPI event from being sent while in this function.
3599 mp_dmevmask
= mp
->m_dmevmask
;
3602 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3604 * Find the agi for this ag.
3606 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3609 * AGI is b0rked. Don't process it.
3611 * We should probably mark the filesystem as corrupt
3612 * after we've recovered all the ag's we can....
3617 * Unlock the buffer so that it can be acquired in the normal
3618 * course of the transaction to truncate and free each inode.
3619 * Because we are not racing with anyone else here for the AGI
3620 * buffer, we don't even need to hold it locked to read the
3621 * initial unlinked bucket entries out of the buffer. We keep
3622 * buffer reference though, so that it stays pinned in memory
3623 * while we need the buffer.
3625 agi
= XFS_BUF_TO_AGI(agibp
);
3626 xfs_buf_unlock(agibp
);
3628 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3629 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3630 while (agino
!= NULLAGINO
) {
3631 agino
= xlog_recover_process_one_iunlink(mp
,
3632 agno
, agino
, bucket
);
3635 xfs_buf_rele(agibp
);
3638 mp
->m_dmevmask
= mp_dmevmask
;
3642 * Upack the log buffer data and crc check it. If the check fails, issue a
3643 * warning if and only if the CRC in the header is non-zero. This makes the
3644 * check an advisory warning, and the zero CRC check will prevent failure
3645 * warnings from being emitted when upgrading the kernel from one that does not
3646 * add CRCs by default.
3648 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3649 * corruption failure
3652 xlog_unpack_data_crc(
3653 struct xlog_rec_header
*rhead
,
3659 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
3660 if (crc
!= rhead
->h_crc
) {
3661 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
3662 xfs_alert(log
->l_mp
,
3663 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
3664 le32_to_cpu(rhead
->h_crc
),
3666 xfs_hex_dump(dp
, 32);
3670 * If we've detected a log record corruption, then we can't
3671 * recover past this point. Abort recovery if we are enforcing
3672 * CRC protection by punting an error back up the stack.
3674 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
3675 return EFSCORRUPTED
;
3683 struct xlog_rec_header
*rhead
,
3690 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
3694 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3695 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3696 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3700 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3701 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3702 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3703 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3704 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3705 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3714 xlog_valid_rec_header(
3716 struct xlog_rec_header
*rhead
,
3721 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
3722 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3723 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3724 return XFS_ERROR(EFSCORRUPTED
);
3727 (!rhead
->h_version
||
3728 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3729 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
3730 __func__
, be32_to_cpu(rhead
->h_version
));
3731 return XFS_ERROR(EIO
);
3734 /* LR body must have data or it wouldn't have been written */
3735 hlen
= be32_to_cpu(rhead
->h_len
);
3736 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3737 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3738 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3739 return XFS_ERROR(EFSCORRUPTED
);
3741 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3742 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3743 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3744 return XFS_ERROR(EFSCORRUPTED
);
3750 * Read the log from tail to head and process the log records found.
3751 * Handle the two cases where the tail and head are in the same cycle
3752 * and where the active portion of the log wraps around the end of
3753 * the physical log separately. The pass parameter is passed through
3754 * to the routines called to process the data and is not looked at
3758 xlog_do_recovery_pass(
3760 xfs_daddr_t head_blk
,
3761 xfs_daddr_t tail_blk
,
3764 xlog_rec_header_t
*rhead
;
3767 xfs_buf_t
*hbp
, *dbp
;
3768 int error
= 0, h_size
;
3769 int bblks
, split_bblks
;
3770 int hblks
, split_hblks
, wrapped_hblks
;
3771 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3773 ASSERT(head_blk
!= tail_blk
);
3776 * Read the header of the tail block and get the iclog buffer size from
3777 * h_size. Use this to tell how many sectors make up the log header.
3779 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3781 * When using variable length iclogs, read first sector of
3782 * iclog header and extract the header size from it. Get a
3783 * new hbp that is the correct size.
3785 hbp
= xlog_get_bp(log
, 1);
3789 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3793 rhead
= (xlog_rec_header_t
*)offset
;
3794 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3797 h_size
= be32_to_cpu(rhead
->h_size
);
3798 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3799 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3800 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3801 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3804 hbp
= xlog_get_bp(log
, hblks
);
3809 ASSERT(log
->l_sectBBsize
== 1);
3811 hbp
= xlog_get_bp(log
, 1);
3812 h_size
= XLOG_BIG_RECORD_BSIZE
;
3817 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3823 memset(rhash
, 0, sizeof(rhash
));
3824 if (tail_blk
<= head_blk
) {
3825 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3826 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3830 rhead
= (xlog_rec_header_t
*)offset
;
3831 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3835 /* blocks in data section */
3836 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3837 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3842 error
= xlog_unpack_data(rhead
, offset
, log
);
3846 error
= xlog_recover_process_data(log
,
3847 rhash
, rhead
, offset
, pass
);
3850 blk_no
+= bblks
+ hblks
;
3854 * Perform recovery around the end of the physical log.
3855 * When the head is not on the same cycle number as the tail,
3856 * we can't do a sequential recovery as above.
3859 while (blk_no
< log
->l_logBBsize
) {
3861 * Check for header wrapping around physical end-of-log
3863 offset
= hbp
->b_addr
;
3866 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3867 /* Read header in one read */
3868 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3873 /* This LR is split across physical log end */
3874 if (blk_no
!= log
->l_logBBsize
) {
3875 /* some data before physical log end */
3876 ASSERT(blk_no
<= INT_MAX
);
3877 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3878 ASSERT(split_hblks
> 0);
3879 error
= xlog_bread(log
, blk_no
,
3887 * Note: this black magic still works with
3888 * large sector sizes (non-512) only because:
3889 * - we increased the buffer size originally
3890 * by 1 sector giving us enough extra space
3891 * for the second read;
3892 * - the log start is guaranteed to be sector
3894 * - we read the log end (LR header start)
3895 * _first_, then the log start (LR header end)
3896 * - order is important.
3898 wrapped_hblks
= hblks
- split_hblks
;
3899 error
= xlog_bread_offset(log
, 0,
3901 offset
+ BBTOB(split_hblks
));
3905 rhead
= (xlog_rec_header_t
*)offset
;
3906 error
= xlog_valid_rec_header(log
, rhead
,
3907 split_hblks
? blk_no
: 0);
3911 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3914 /* Read in data for log record */
3915 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3916 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3921 /* This log record is split across the
3922 * physical end of log */
3923 offset
= dbp
->b_addr
;
3925 if (blk_no
!= log
->l_logBBsize
) {
3926 /* some data is before the physical
3928 ASSERT(!wrapped_hblks
);
3929 ASSERT(blk_no
<= INT_MAX
);
3931 log
->l_logBBsize
- (int)blk_no
;
3932 ASSERT(split_bblks
> 0);
3933 error
= xlog_bread(log
, blk_no
,
3941 * Note: this black magic still works with
3942 * large sector sizes (non-512) only because:
3943 * - we increased the buffer size originally
3944 * by 1 sector giving us enough extra space
3945 * for the second read;
3946 * - the log start is guaranteed to be sector
3948 * - we read the log end (LR header start)
3949 * _first_, then the log start (LR header end)
3950 * - order is important.
3952 error
= xlog_bread_offset(log
, 0,
3953 bblks
- split_bblks
, dbp
,
3954 offset
+ BBTOB(split_bblks
));
3959 error
= xlog_unpack_data(rhead
, offset
, log
);
3963 error
= xlog_recover_process_data(log
, rhash
,
3964 rhead
, offset
, pass
);
3970 ASSERT(blk_no
>= log
->l_logBBsize
);
3971 blk_no
-= log
->l_logBBsize
;
3973 /* read first part of physical log */
3974 while (blk_no
< head_blk
) {
3975 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3979 rhead
= (xlog_rec_header_t
*)offset
;
3980 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3984 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3985 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3990 error
= xlog_unpack_data(rhead
, offset
, log
);
3994 error
= xlog_recover_process_data(log
, rhash
,
3995 rhead
, offset
, pass
);
3998 blk_no
+= bblks
+ hblks
;
4010 * Do the recovery of the log. We actually do this in two phases.
4011 * The two passes are necessary in order to implement the function
4012 * of cancelling a record written into the log. The first pass
4013 * determines those things which have been cancelled, and the
4014 * second pass replays log items normally except for those which
4015 * have been cancelled. The handling of the replay and cancellations
4016 * takes place in the log item type specific routines.
4018 * The table of items which have cancel records in the log is allocated
4019 * and freed at this level, since only here do we know when all of
4020 * the log recovery has been completed.
4023 xlog_do_log_recovery(
4025 xfs_daddr_t head_blk
,
4026 xfs_daddr_t tail_blk
)
4030 ASSERT(head_blk
!= tail_blk
);
4033 * First do a pass to find all of the cancelled buf log items.
4034 * Store them in the buf_cancel_table for use in the second pass.
4036 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4037 sizeof(struct list_head
),
4039 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4040 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4042 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4043 XLOG_RECOVER_PASS1
);
4045 kmem_free(log
->l_buf_cancel_table
);
4046 log
->l_buf_cancel_table
= NULL
;
4050 * Then do a second pass to actually recover the items in the log.
4051 * When it is complete free the table of buf cancel items.
4053 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4054 XLOG_RECOVER_PASS2
);
4059 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4060 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4064 kmem_free(log
->l_buf_cancel_table
);
4065 log
->l_buf_cancel_table
= NULL
;
4071 * Do the actual recovery
4076 xfs_daddr_t head_blk
,
4077 xfs_daddr_t tail_blk
)
4084 * First replay the images in the log.
4086 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4091 * If IO errors happened during recovery, bail out.
4093 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4098 * We now update the tail_lsn since much of the recovery has completed
4099 * and there may be space available to use. If there were no extent
4100 * or iunlinks, we can free up the entire log and set the tail_lsn to
4101 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4102 * lsn of the last known good LR on disk. If there are extent frees
4103 * or iunlinks they will have some entries in the AIL; so we look at
4104 * the AIL to determine how to set the tail_lsn.
4106 xlog_assign_tail_lsn(log
->l_mp
);
4109 * Now that we've finished replaying all buffer and inode
4110 * updates, re-read in the superblock and reverify it.
4112 bp
= xfs_getsb(log
->l_mp
, 0);
4114 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
4116 XFS_BUF_UNASYNC(bp
);
4117 bp
->b_ops
= &xfs_sb_buf_ops
;
4118 xfsbdstrat(log
->l_mp
, bp
);
4119 error
= xfs_buf_iowait(bp
);
4121 xfs_buf_ioerror_alert(bp
, __func__
);
4127 /* Convert superblock from on-disk format */
4128 sbp
= &log
->l_mp
->m_sb
;
4129 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4130 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4131 ASSERT(xfs_sb_good_version(sbp
));
4134 /* We've re-read the superblock so re-initialize per-cpu counters */
4135 xfs_icsb_reinit_counters(log
->l_mp
);
4137 xlog_recover_check_summary(log
);
4139 /* Normal transactions can now occur */
4140 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4145 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4147 * Return error or zero.
4153 xfs_daddr_t head_blk
, tail_blk
;
4156 /* find the tail of the log */
4157 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
4160 if (tail_blk
!= head_blk
) {
4161 /* There used to be a comment here:
4163 * disallow recovery on read-only mounts. note -- mount
4164 * checks for ENOSPC and turns it into an intelligent
4166 * ...but this is no longer true. Now, unless you specify
4167 * NORECOVERY (in which case this function would never be
4168 * called), we just go ahead and recover. We do this all
4169 * under the vfs layer, so we can get away with it unless
4170 * the device itself is read-only, in which case we fail.
4172 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
4177 * Version 5 superblock log feature mask validation. We know the
4178 * log is dirty so check if there are any unknown log features
4179 * in what we need to recover. If there are unknown features
4180 * (e.g. unsupported transactions, then simply reject the
4181 * attempt at recovery before touching anything.
4183 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
4184 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
4185 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
4187 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4188 "The log can not be fully and/or safely recovered by this kernel.\n"
4189 "Please recover the log on a kernel that supports the unknown features.",
4190 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
4191 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
4195 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
4196 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4199 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
4200 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
4206 * In the first part of recovery we replay inodes and buffers and build
4207 * up the list of extent free items which need to be processed. Here
4208 * we process the extent free items and clean up the on disk unlinked
4209 * inode lists. This is separated from the first part of recovery so
4210 * that the root and real-time bitmap inodes can be read in from disk in
4211 * between the two stages. This is necessary so that we can free space
4212 * in the real-time portion of the file system.
4215 xlog_recover_finish(
4219 * Now we're ready to do the transactions needed for the
4220 * rest of recovery. Start with completing all the extent
4221 * free intent records and then process the unlinked inode
4222 * lists. At this point, we essentially run in normal mode
4223 * except that we're still performing recovery actions
4224 * rather than accepting new requests.
4226 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
4228 error
= xlog_recover_process_efis(log
);
4230 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
4234 * Sync the log to get all the EFIs out of the AIL.
4235 * This isn't absolutely necessary, but it helps in
4236 * case the unlink transactions would have problems
4237 * pushing the EFIs out of the way.
4239 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
4241 xlog_recover_process_iunlinks(log
);
4243 xlog_recover_check_summary(log
);
4245 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
4246 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4248 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4250 xfs_info(log
->l_mp
, "Ending clean mount");
4258 * Read all of the agf and agi counters and check that they
4259 * are consistent with the superblock counters.
4262 xlog_recover_check_summary(
4269 xfs_agnumber_t agno
;
4270 __uint64_t freeblks
;
4280 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4281 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4283 xfs_alert(mp
, "%s agf read failed agno %d error %d",
4284 __func__
, agno
, error
);
4286 agfp
= XFS_BUF_TO_AGF(agfbp
);
4287 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4288 be32_to_cpu(agfp
->agf_flcount
);
4289 xfs_buf_relse(agfbp
);
4292 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4294 xfs_alert(mp
, "%s agi read failed agno %d error %d",
4295 __func__
, agno
, error
);
4297 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4299 itotal
+= be32_to_cpu(agi
->agi_count
);
4300 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4301 xfs_buf_relse(agibp
);