2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_log_priv.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_log_recover.h"
41 #include "xfs_extfree_item.h"
42 #include "xfs_trans_priv.h"
43 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
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"
53 #include "xfs_dir2_priv.h"
54 #include "xfs_attr_leaf.h"
55 #include "xfs_attr_remote.h"
62 xlog_clear_stale_blocks(
67 xlog_recover_check_summary(
70 #define xlog_recover_check_summary(log)
74 * This structure is used during recovery to record the buf log items which
75 * have been canceled and should not be replayed.
77 struct xfs_buf_cancel
{
81 struct list_head bc_list
;
85 * Sector aligned buffer routines for buffer create/read/write/access
89 * Verify the given count of basic blocks is valid number of blocks
90 * to specify for an operation involving the given XFS log buffer.
91 * Returns nonzero if the count is valid, 0 otherwise.
95 xlog_buf_bbcount_valid(
99 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
103 * Allocate a buffer to hold log data. The buffer needs to be able
104 * to map to a range of nbblks basic blocks at any valid (basic
105 * block) offset within the log.
114 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
115 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
117 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
122 * We do log I/O in units of log sectors (a power-of-2
123 * multiple of the basic block size), so we round up the
124 * requested size to accommodate the basic blocks required
125 * for complete log sectors.
127 * In addition, the buffer may be used for a non-sector-
128 * aligned block offset, in which case an I/O of the
129 * requested size could extend beyond the end of the
130 * buffer. If the requested size is only 1 basic block it
131 * will never straddle a sector boundary, so this won't be
132 * an issue. Nor will this be a problem if the log I/O is
133 * done in basic blocks (sector size 1). But otherwise we
134 * extend the buffer by one extra log sector to ensure
135 * there's space to accommodate this possibility.
137 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
138 nbblks
+= log
->l_sectBBsize
;
139 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
141 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
155 * Return the address of the start of the given block number's data
156 * in a log buffer. The buffer covers a log sector-aligned region.
165 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
167 ASSERT(offset
+ nbblks
<= bp
->b_length
);
168 return bp
->b_addr
+ BBTOB(offset
);
173 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
184 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
185 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
187 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
191 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
192 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
195 ASSERT(nbblks
<= bp
->b_length
);
197 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
199 bp
->b_io_length
= nbblks
;
202 xfsbdstrat(log
->l_mp
, bp
);
203 error
= xfs_buf_iowait(bp
);
205 xfs_buf_ioerror_alert(bp
, __func__
);
219 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
223 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
228 * Read at an offset into the buffer. Returns with the buffer in it's original
229 * state regardless of the result of the read.
234 xfs_daddr_t blk_no
, /* block to read from */
235 int nbblks
, /* blocks to read */
239 xfs_caddr_t orig_offset
= bp
->b_addr
;
240 int orig_len
= BBTOB(bp
->b_length
);
243 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
247 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
249 /* must reset buffer pointer even on error */
250 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
257 * Write out the buffer at the given block for the given number of blocks.
258 * The buffer is kept locked across the write and is returned locked.
259 * This can only be used for synchronous log writes.
270 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
271 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
273 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
277 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
278 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
281 ASSERT(nbblks
<= bp
->b_length
);
283 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
284 XFS_BUF_ZEROFLAGS(bp
);
287 bp
->b_io_length
= nbblks
;
290 error
= xfs_bwrite(bp
);
292 xfs_buf_ioerror_alert(bp
, __func__
);
299 * dump debug superblock and log record information
302 xlog_header_check_dump(
304 xlog_rec_header_t
*head
)
306 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d\n",
307 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
308 xfs_debug(mp
, " log : uuid = %pU, fmt = %d\n",
309 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
312 #define xlog_header_check_dump(mp, head)
316 * check log record header for recovery
319 xlog_header_check_recover(
321 xlog_rec_header_t
*head
)
323 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
326 * IRIX doesn't write the h_fmt field and leaves it zeroed
327 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
328 * a dirty log created in IRIX.
330 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
332 "dirty log written in incompatible format - can't recover");
333 xlog_header_check_dump(mp
, head
);
334 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
335 XFS_ERRLEVEL_HIGH
, mp
);
336 return XFS_ERROR(EFSCORRUPTED
);
337 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
339 "dirty log entry has mismatched uuid - can't recover");
340 xlog_header_check_dump(mp
, head
);
341 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
342 XFS_ERRLEVEL_HIGH
, mp
);
343 return XFS_ERROR(EFSCORRUPTED
);
349 * read the head block of the log and check the header
352 xlog_header_check_mount(
354 xlog_rec_header_t
*head
)
356 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
358 if (uuid_is_nil(&head
->h_fs_uuid
)) {
360 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
361 * h_fs_uuid is nil, we assume this log was last mounted
362 * by IRIX and continue.
364 xfs_warn(mp
, "nil uuid in log - IRIX style log");
365 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
366 xfs_warn(mp
, "log has mismatched uuid - can't recover");
367 xlog_header_check_dump(mp
, head
);
368 XFS_ERROR_REPORT("xlog_header_check_mount",
369 XFS_ERRLEVEL_HIGH
, mp
);
370 return XFS_ERROR(EFSCORRUPTED
);
381 * We're not going to bother about retrying
382 * this during recovery. One strike!
384 xfs_buf_ioerror_alert(bp
, __func__
);
385 xfs_force_shutdown(bp
->b_target
->bt_mount
,
386 SHUTDOWN_META_IO_ERROR
);
389 xfs_buf_ioend(bp
, 0);
393 * This routine finds (to an approximation) the first block in the physical
394 * log which contains the given cycle. It uses a binary search algorithm.
395 * Note that the algorithm can not be perfect because the disk will not
396 * necessarily be perfect.
399 xlog_find_cycle_start(
402 xfs_daddr_t first_blk
,
403 xfs_daddr_t
*last_blk
,
413 mid_blk
= BLK_AVG(first_blk
, end_blk
);
414 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
415 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
418 mid_cycle
= xlog_get_cycle(offset
);
419 if (mid_cycle
== cycle
)
420 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
422 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
423 mid_blk
= BLK_AVG(first_blk
, end_blk
);
425 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
426 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
434 * Check that a range of blocks does not contain stop_on_cycle_no.
435 * Fill in *new_blk with the block offset where such a block is
436 * found, or with -1 (an invalid block number) if there is no such
437 * block in the range. The scan needs to occur from front to back
438 * and the pointer into the region must be updated since a later
439 * routine will need to perform another test.
442 xlog_find_verify_cycle(
444 xfs_daddr_t start_blk
,
446 uint stop_on_cycle_no
,
447 xfs_daddr_t
*new_blk
)
453 xfs_caddr_t buf
= NULL
;
457 * Greedily allocate a buffer big enough to handle the full
458 * range of basic blocks we'll be examining. If that fails,
459 * try a smaller size. We need to be able to read at least
460 * a log sector, or we're out of luck.
462 bufblks
= 1 << ffs(nbblks
);
463 while (bufblks
> log
->l_logBBsize
)
465 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
467 if (bufblks
< log
->l_sectBBsize
)
471 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
474 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
476 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
480 for (j
= 0; j
< bcount
; j
++) {
481 cycle
= xlog_get_cycle(buf
);
482 if (cycle
== stop_on_cycle_no
) {
499 * Potentially backup over partial log record write.
501 * In the typical case, last_blk is the number of the block directly after
502 * a good log record. Therefore, we subtract one to get the block number
503 * of the last block in the given buffer. extra_bblks contains the number
504 * of blocks we would have read on a previous read. This happens when the
505 * last log record is split over the end of the physical log.
507 * extra_bblks is the number of blocks potentially verified on a previous
508 * call to this routine.
511 xlog_find_verify_log_record(
513 xfs_daddr_t start_blk
,
514 xfs_daddr_t
*last_blk
,
519 xfs_caddr_t offset
= NULL
;
520 xlog_rec_header_t
*head
= NULL
;
523 int num_blks
= *last_blk
- start_blk
;
526 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
528 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
529 if (!(bp
= xlog_get_bp(log
, 1)))
533 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
536 offset
+= ((num_blks
- 1) << BBSHIFT
);
539 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
541 /* valid log record not found */
543 "Log inconsistent (didn't find previous header)");
545 error
= XFS_ERROR(EIO
);
550 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
555 head
= (xlog_rec_header_t
*)offset
;
557 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
565 * We hit the beginning of the physical log & still no header. Return
566 * to caller. If caller can handle a return of -1, then this routine
567 * will be called again for the end of the physical log.
575 * We have the final block of the good log (the first block
576 * of the log record _before_ the head. So we check the uuid.
578 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
582 * We may have found a log record header before we expected one.
583 * last_blk will be the 1st block # with a given cycle #. We may end
584 * up reading an entire log record. In this case, we don't want to
585 * reset last_blk. Only when last_blk points in the middle of a log
586 * record do we update last_blk.
588 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
589 uint h_size
= be32_to_cpu(head
->h_size
);
591 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
592 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
598 if (*last_blk
- i
+ extra_bblks
!=
599 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
608 * Head is defined to be the point of the log where the next log write
609 * write could go. This means that incomplete LR writes at the end are
610 * eliminated when calculating the head. We aren't guaranteed that previous
611 * LR have complete transactions. We only know that a cycle number of
612 * current cycle number -1 won't be present in the log if we start writing
613 * from our current block number.
615 * last_blk contains the block number of the first block with a given
618 * Return: zero if normal, non-zero if error.
623 xfs_daddr_t
*return_head_blk
)
627 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
629 uint first_half_cycle
, last_half_cycle
;
631 int error
, log_bbnum
= log
->l_logBBsize
;
633 /* Is the end of the log device zeroed? */
634 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
635 *return_head_blk
= first_blk
;
637 /* Is the whole lot zeroed? */
639 /* Linux XFS shouldn't generate totally zeroed logs -
640 * mkfs etc write a dummy unmount record to a fresh
641 * log so we can store the uuid in there
643 xfs_warn(log
->l_mp
, "totally zeroed log");
648 xfs_warn(log
->l_mp
, "empty log check failed");
652 first_blk
= 0; /* get cycle # of 1st block */
653 bp
= xlog_get_bp(log
, 1);
657 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
661 first_half_cycle
= xlog_get_cycle(offset
);
663 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
664 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
668 last_half_cycle
= xlog_get_cycle(offset
);
669 ASSERT(last_half_cycle
!= 0);
672 * If the 1st half cycle number is equal to the last half cycle number,
673 * then the entire log is stamped with the same cycle number. In this
674 * case, head_blk can't be set to zero (which makes sense). The below
675 * math doesn't work out properly with head_blk equal to zero. Instead,
676 * we set it to log_bbnum which is an invalid block number, but this
677 * value makes the math correct. If head_blk doesn't changed through
678 * all the tests below, *head_blk is set to zero at the very end rather
679 * than log_bbnum. In a sense, log_bbnum and zero are the same block
680 * in a circular file.
682 if (first_half_cycle
== last_half_cycle
) {
684 * In this case we believe that the entire log should have
685 * cycle number last_half_cycle. We need to scan backwards
686 * from the end verifying that there are no holes still
687 * containing last_half_cycle - 1. If we find such a hole,
688 * then the start of that hole will be the new head. The
689 * simple case looks like
690 * x | x ... | x - 1 | x
691 * Another case that fits this picture would be
692 * x | x + 1 | x ... | x
693 * In this case the head really is somewhere at the end of the
694 * log, as one of the latest writes at the beginning was
697 * x | x + 1 | x ... | x - 1 | x
698 * This is really the combination of the above two cases, and
699 * the head has to end up at the start of the x-1 hole at the
702 * In the 256k log case, we will read from the beginning to the
703 * end of the log and search for cycle numbers equal to x-1.
704 * We don't worry about the x+1 blocks that we encounter,
705 * because we know that they cannot be the head since the log
708 head_blk
= log_bbnum
;
709 stop_on_cycle
= last_half_cycle
- 1;
712 * In this case we want to find the first block with cycle
713 * number matching last_half_cycle. We expect the log to be
715 * x + 1 ... | x ... | x
716 * The first block with cycle number x (last_half_cycle) will
717 * be where the new head belongs. First we do a binary search
718 * for the first occurrence of last_half_cycle. The binary
719 * search may not be totally accurate, so then we scan back
720 * from there looking for occurrences of last_half_cycle before
721 * us. If that backwards scan wraps around the beginning of
722 * the log, then we look for occurrences of last_half_cycle - 1
723 * at the end of the log. The cases we're looking for look
725 * v binary search stopped here
726 * x + 1 ... | x | x + 1 | x ... | x
727 * ^ but we want to locate this spot
729 * <---------> less than scan distance
730 * x + 1 ... | x ... | x - 1 | x
731 * ^ we want to locate this spot
733 stop_on_cycle
= last_half_cycle
;
734 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
735 &head_blk
, last_half_cycle
)))
740 * Now validate the answer. Scan back some number of maximum possible
741 * blocks and make sure each one has the expected cycle number. The
742 * maximum is determined by the total possible amount of buffering
743 * in the in-core log. The following number can be made tighter if
744 * we actually look at the block size of the filesystem.
746 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
747 if (head_blk
>= num_scan_bblks
) {
749 * We are guaranteed that the entire check can be performed
752 start_blk
= head_blk
- num_scan_bblks
;
753 if ((error
= xlog_find_verify_cycle(log
,
754 start_blk
, num_scan_bblks
,
755 stop_on_cycle
, &new_blk
)))
759 } else { /* need to read 2 parts of log */
761 * We are going to scan backwards in the log in two parts.
762 * First we scan the physical end of the log. In this part
763 * of the log, we are looking for blocks with cycle number
764 * last_half_cycle - 1.
765 * If we find one, then we know that the log starts there, as
766 * we've found a hole that didn't get written in going around
767 * the end of the physical log. The simple case for this is
768 * x + 1 ... | x ... | x - 1 | x
769 * <---------> less than scan distance
770 * If all of the blocks at the end of the log have cycle number
771 * last_half_cycle, then we check the blocks at the start of
772 * the log looking for occurrences of last_half_cycle. If we
773 * find one, then our current estimate for the location of the
774 * first occurrence of last_half_cycle is wrong and we move
775 * back to the hole we've found. This case looks like
776 * x + 1 ... | x | x + 1 | x ...
777 * ^ binary search stopped here
778 * Another case we need to handle that only occurs in 256k
780 * x + 1 ... | x ... | x+1 | x ...
781 * ^ binary search stops here
782 * In a 256k log, the scan at the end of the log will see the
783 * x + 1 blocks. We need to skip past those since that is
784 * certainly not the head of the log. By searching for
785 * last_half_cycle-1 we accomplish that.
787 ASSERT(head_blk
<= INT_MAX
&&
788 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
789 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
790 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
791 num_scan_bblks
- (int)head_blk
,
792 (stop_on_cycle
- 1), &new_blk
)))
800 * Scan beginning of log now. The last part of the physical
801 * log is good. This scan needs to verify that it doesn't find
802 * the last_half_cycle.
805 ASSERT(head_blk
<= INT_MAX
);
806 if ((error
= xlog_find_verify_cycle(log
,
807 start_blk
, (int)head_blk
,
808 stop_on_cycle
, &new_blk
)))
816 * Now we need to make sure head_blk is not pointing to a block in
817 * the middle of a log record.
819 num_scan_bblks
= XLOG_REC_SHIFT(log
);
820 if (head_blk
>= num_scan_bblks
) {
821 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
823 /* start ptr at last block ptr before head_blk */
824 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
825 &head_blk
, 0)) == -1) {
826 error
= XFS_ERROR(EIO
);
832 ASSERT(head_blk
<= INT_MAX
);
833 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
834 &head_blk
, 0)) == -1) {
835 /* We hit the beginning of the log during our search */
836 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
838 ASSERT(start_blk
<= INT_MAX
&&
839 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
840 ASSERT(head_blk
<= INT_MAX
);
841 if ((error
= xlog_find_verify_log_record(log
,
843 (int)head_blk
)) == -1) {
844 error
= XFS_ERROR(EIO
);
848 if (new_blk
!= log_bbnum
)
855 if (head_blk
== log_bbnum
)
856 *return_head_blk
= 0;
858 *return_head_blk
= head_blk
;
860 * When returning here, we have a good block number. Bad block
861 * means that during a previous crash, we didn't have a clean break
862 * from cycle number N to cycle number N-1. In this case, we need
863 * to find the first block with cycle number N-1.
871 xfs_warn(log
->l_mp
, "failed to find log head");
876 * Find the sync block number or the tail of the log.
878 * This will be the block number of the last record to have its
879 * associated buffers synced to disk. Every log record header has
880 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
881 * to get a sync block number. The only concern is to figure out which
882 * log record header to believe.
884 * The following algorithm uses the log record header with the largest
885 * lsn. The entire log record does not need to be valid. We only care
886 * that the header is valid.
888 * We could speed up search by using current head_blk buffer, but it is not
894 xfs_daddr_t
*head_blk
,
895 xfs_daddr_t
*tail_blk
)
897 xlog_rec_header_t
*rhead
;
898 xlog_op_header_t
*op_head
;
899 xfs_caddr_t offset
= NULL
;
902 xfs_daddr_t umount_data_blk
;
903 xfs_daddr_t after_umount_blk
;
910 * Find previous log record
912 if ((error
= xlog_find_head(log
, head_blk
)))
915 bp
= xlog_get_bp(log
, 1);
918 if (*head_blk
== 0) { /* special case */
919 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
923 if (xlog_get_cycle(offset
) == 0) {
925 /* leave all other log inited values alone */
931 * Search backwards looking for log record header block
933 ASSERT(*head_blk
< INT_MAX
);
934 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
935 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
939 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
945 * If we haven't found the log record header block, start looking
946 * again from the end of the physical log. XXXmiken: There should be
947 * a check here to make sure we didn't search more than N blocks in
951 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
952 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
956 if (*(__be32
*)offset
==
957 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
964 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
966 return XFS_ERROR(EIO
);
969 /* find blk_no of tail of log */
970 rhead
= (xlog_rec_header_t
*)offset
;
971 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
974 * Reset log values according to the state of the log when we
975 * crashed. In the case where head_blk == 0, we bump curr_cycle
976 * one because the next write starts a new cycle rather than
977 * continuing the cycle of the last good log record. At this
978 * point we have guaranteed that all partial log records have been
979 * accounted for. Therefore, we know that the last good log record
980 * written was complete and ended exactly on the end boundary
981 * of the physical log.
983 log
->l_prev_block
= i
;
984 log
->l_curr_block
= (int)*head_blk
;
985 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
988 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
989 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
990 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
991 BBTOB(log
->l_curr_block
));
992 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
993 BBTOB(log
->l_curr_block
));
996 * Look for unmount record. If we find it, then we know there
997 * was a clean unmount. Since 'i' could be the last block in
998 * the physical log, we convert to a log block before comparing
1001 * Save the current tail lsn to use to pass to
1002 * xlog_clear_stale_blocks() below. We won't want to clear the
1003 * unmount record if there is one, so we pass the lsn of the
1004 * unmount record rather than the block after it.
1006 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1007 int h_size
= be32_to_cpu(rhead
->h_size
);
1008 int h_version
= be32_to_cpu(rhead
->h_version
);
1010 if ((h_version
& XLOG_VERSION_2
) &&
1011 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1012 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1013 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1021 after_umount_blk
= (i
+ hblks
+ (int)
1022 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1023 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1024 if (*head_blk
== after_umount_blk
&&
1025 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1026 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1027 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1031 op_head
= (xlog_op_header_t
*)offset
;
1032 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1034 * Set tail and last sync so that newly written
1035 * log records will point recovery to after the
1036 * current unmount record.
1038 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1039 log
->l_curr_cycle
, after_umount_blk
);
1040 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1041 log
->l_curr_cycle
, after_umount_blk
);
1042 *tail_blk
= after_umount_blk
;
1045 * Note that the unmount was clean. If the unmount
1046 * was not clean, we need to know this to rebuild the
1047 * superblock counters from the perag headers if we
1048 * have a filesystem using non-persistent counters.
1050 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1055 * Make sure that there are no blocks in front of the head
1056 * with the same cycle number as the head. This can happen
1057 * because we allow multiple outstanding log writes concurrently,
1058 * and the later writes might make it out before earlier ones.
1060 * We use the lsn from before modifying it so that we'll never
1061 * overwrite the unmount record after a clean unmount.
1063 * Do this only if we are going to recover the filesystem
1065 * NOTE: This used to say "if (!readonly)"
1066 * However on Linux, we can & do recover a read-only filesystem.
1067 * We only skip recovery if NORECOVERY is specified on mount,
1068 * in which case we would not be here.
1070 * But... if the -device- itself is readonly, just skip this.
1071 * We can't recover this device anyway, so it won't matter.
1073 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1074 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1080 xfs_warn(log
->l_mp
, "failed to locate log tail");
1085 * Is the log zeroed at all?
1087 * The last binary search should be changed to perform an X block read
1088 * once X becomes small enough. You can then search linearly through
1089 * the X blocks. This will cut down on the number of reads we need to do.
1091 * If the log is partially zeroed, this routine will pass back the blkno
1092 * of the first block with cycle number 0. It won't have a complete LR
1096 * 0 => the log is completely written to
1097 * -1 => use *blk_no as the first block of the log
1098 * >0 => error has occurred
1103 xfs_daddr_t
*blk_no
)
1107 uint first_cycle
, last_cycle
;
1108 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1109 xfs_daddr_t num_scan_bblks
;
1110 int error
, log_bbnum
= log
->l_logBBsize
;
1114 /* check totally zeroed log */
1115 bp
= xlog_get_bp(log
, 1);
1118 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1122 first_cycle
= xlog_get_cycle(offset
);
1123 if (first_cycle
== 0) { /* completely zeroed log */
1129 /* check partially zeroed log */
1130 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1134 last_cycle
= xlog_get_cycle(offset
);
1135 if (last_cycle
!= 0) { /* log completely written to */
1138 } else if (first_cycle
!= 1) {
1140 * If the cycle of the last block is zero, the cycle of
1141 * the first block must be 1. If it's not, maybe we're
1142 * not looking at a log... Bail out.
1145 "Log inconsistent or not a log (last==0, first!=1)");
1146 return XFS_ERROR(EINVAL
);
1149 /* we have a partially zeroed log */
1150 last_blk
= log_bbnum
-1;
1151 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1155 * Validate the answer. Because there is no way to guarantee that
1156 * the entire log is made up of log records which are the same size,
1157 * we scan over the defined maximum blocks. At this point, the maximum
1158 * is not chosen to mean anything special. XXXmiken
1160 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1161 ASSERT(num_scan_bblks
<= INT_MAX
);
1163 if (last_blk
< num_scan_bblks
)
1164 num_scan_bblks
= last_blk
;
1165 start_blk
= last_blk
- num_scan_bblks
;
1168 * We search for any instances of cycle number 0 that occur before
1169 * our current estimate of the head. What we're trying to detect is
1170 * 1 ... | 0 | 1 | 0...
1171 * ^ binary search ends here
1173 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1174 (int)num_scan_bblks
, 0, &new_blk
)))
1180 * Potentially backup over partial log record write. We don't need
1181 * to search the end of the log because we know it is zero.
1183 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1184 &last_blk
, 0)) == -1) {
1185 error
= XFS_ERROR(EIO
);
1199 * These are simple subroutines used by xlog_clear_stale_blocks() below
1200 * to initialize a buffer full of empty log record headers and write
1201 * them into the log.
1212 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1214 memset(buf
, 0, BBSIZE
);
1215 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1216 recp
->h_cycle
= cpu_to_be32(cycle
);
1217 recp
->h_version
= cpu_to_be32(
1218 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1219 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1220 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1221 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1222 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1226 xlog_write_log_records(
1237 int sectbb
= log
->l_sectBBsize
;
1238 int end_block
= start_block
+ blocks
;
1244 * Greedily allocate a buffer big enough to handle the full
1245 * range of basic blocks to be written. If that fails, try
1246 * a smaller size. We need to be able to write at least a
1247 * log sector, or we're out of luck.
1249 bufblks
= 1 << ffs(blocks
);
1250 while (bufblks
> log
->l_logBBsize
)
1252 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1254 if (bufblks
< sectbb
)
1258 /* We may need to do a read at the start to fill in part of
1259 * the buffer in the starting sector not covered by the first
1262 balign
= round_down(start_block
, sectbb
);
1263 if (balign
!= start_block
) {
1264 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1268 j
= start_block
- balign
;
1271 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1272 int bcount
, endcount
;
1274 bcount
= min(bufblks
, end_block
- start_block
);
1275 endcount
= bcount
- j
;
1277 /* We may need to do a read at the end to fill in part of
1278 * the buffer in the final sector not covered by the write.
1279 * If this is the same sector as the above read, skip it.
1281 ealign
= round_down(end_block
, sectbb
);
1282 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1283 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1284 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1291 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1292 for (; j
< endcount
; j
++) {
1293 xlog_add_record(log
, offset
, cycle
, i
+j
,
1294 tail_cycle
, tail_block
);
1297 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1300 start_block
+= endcount
;
1310 * This routine is called to blow away any incomplete log writes out
1311 * in front of the log head. We do this so that we won't become confused
1312 * if we come up, write only a little bit more, and then crash again.
1313 * If we leave the partial log records out there, this situation could
1314 * cause us to think those partial writes are valid blocks since they
1315 * have the current cycle number. We get rid of them by overwriting them
1316 * with empty log records with the old cycle number rather than the
1319 * The tail lsn is passed in rather than taken from
1320 * the log so that we will not write over the unmount record after a
1321 * clean unmount in a 512 block log. Doing so would leave the log without
1322 * any valid log records in it until a new one was written. If we crashed
1323 * during that time we would not be able to recover.
1326 xlog_clear_stale_blocks(
1330 int tail_cycle
, head_cycle
;
1331 int tail_block
, head_block
;
1332 int tail_distance
, max_distance
;
1336 tail_cycle
= CYCLE_LSN(tail_lsn
);
1337 tail_block
= BLOCK_LSN(tail_lsn
);
1338 head_cycle
= log
->l_curr_cycle
;
1339 head_block
= log
->l_curr_block
;
1342 * Figure out the distance between the new head of the log
1343 * and the tail. We want to write over any blocks beyond the
1344 * head that we may have written just before the crash, but
1345 * we don't want to overwrite the tail of the log.
1347 if (head_cycle
== tail_cycle
) {
1349 * The tail is behind the head in the physical log,
1350 * so the distance from the head to the tail is the
1351 * distance from the head to the end of the log plus
1352 * the distance from the beginning of the log to the
1355 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1356 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1357 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1358 return XFS_ERROR(EFSCORRUPTED
);
1360 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1363 * The head is behind the tail in the physical log,
1364 * so the distance from the head to the tail is just
1365 * the tail block minus the head block.
1367 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1368 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1369 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1370 return XFS_ERROR(EFSCORRUPTED
);
1372 tail_distance
= tail_block
- head_block
;
1376 * If the head is right up against the tail, we can't clear
1379 if (tail_distance
<= 0) {
1380 ASSERT(tail_distance
== 0);
1384 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1386 * Take the smaller of the maximum amount of outstanding I/O
1387 * we could have and the distance to the tail to clear out.
1388 * We take the smaller so that we don't overwrite the tail and
1389 * we don't waste all day writing from the head to the tail
1392 max_distance
= MIN(max_distance
, tail_distance
);
1394 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1396 * We can stomp all the blocks we need to without
1397 * wrapping around the end of the log. Just do it
1398 * in a single write. Use the cycle number of the
1399 * current cycle minus one so that the log will look like:
1402 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1403 head_block
, max_distance
, tail_cycle
,
1409 * We need to wrap around the end of the physical log in
1410 * order to clear all the blocks. Do it in two separate
1411 * I/Os. The first write should be from the head to the
1412 * end of the physical log, and it should use the current
1413 * cycle number minus one just like above.
1415 distance
= log
->l_logBBsize
- head_block
;
1416 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1417 head_block
, distance
, tail_cycle
,
1424 * Now write the blocks at the start of the physical log.
1425 * This writes the remainder of the blocks we want to clear.
1426 * It uses the current cycle number since we're now on the
1427 * same cycle as the head so that we get:
1428 * n ... n ... | n - 1 ...
1429 * ^^^^^ blocks we're writing
1431 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1432 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1433 tail_cycle
, tail_block
);
1441 /******************************************************************************
1443 * Log recover routines
1445 ******************************************************************************
1448 STATIC xlog_recover_t
*
1449 xlog_recover_find_tid(
1450 struct hlist_head
*head
,
1453 xlog_recover_t
*trans
;
1455 hlist_for_each_entry(trans
, head
, r_list
) {
1456 if (trans
->r_log_tid
== tid
)
1463 xlog_recover_new_tid(
1464 struct hlist_head
*head
,
1468 xlog_recover_t
*trans
;
1470 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1471 trans
->r_log_tid
= tid
;
1473 INIT_LIST_HEAD(&trans
->r_itemq
);
1475 INIT_HLIST_NODE(&trans
->r_list
);
1476 hlist_add_head(&trans
->r_list
, head
);
1480 xlog_recover_add_item(
1481 struct list_head
*head
)
1483 xlog_recover_item_t
*item
;
1485 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1486 INIT_LIST_HEAD(&item
->ri_list
);
1487 list_add_tail(&item
->ri_list
, head
);
1491 xlog_recover_add_to_cont_trans(
1493 struct xlog_recover
*trans
,
1497 xlog_recover_item_t
*item
;
1498 xfs_caddr_t ptr
, old_ptr
;
1501 if (list_empty(&trans
->r_itemq
)) {
1502 /* finish copying rest of trans header */
1503 xlog_recover_add_item(&trans
->r_itemq
);
1504 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1505 sizeof(xfs_trans_header_t
) - len
;
1506 memcpy(ptr
, dp
, len
); /* d, s, l */
1509 /* take the tail entry */
1510 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1512 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1513 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1515 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
1516 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1517 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1518 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1519 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1524 * The next region to add is the start of a new region. It could be
1525 * a whole region or it could be the first part of a new region. Because
1526 * of this, the assumption here is that the type and size fields of all
1527 * format structures fit into the first 32 bits of the structure.
1529 * This works because all regions must be 32 bit aligned. Therefore, we
1530 * either have both fields or we have neither field. In the case we have
1531 * neither field, the data part of the region is zero length. We only have
1532 * a log_op_header and can throw away the header since a new one will appear
1533 * later. If we have at least 4 bytes, then we can determine how many regions
1534 * will appear in the current log item.
1537 xlog_recover_add_to_trans(
1539 struct xlog_recover
*trans
,
1543 xfs_inode_log_format_t
*in_f
; /* any will do */
1544 xlog_recover_item_t
*item
;
1549 if (list_empty(&trans
->r_itemq
)) {
1550 /* we need to catch log corruptions here */
1551 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1552 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1555 return XFS_ERROR(EIO
);
1557 if (len
== sizeof(xfs_trans_header_t
))
1558 xlog_recover_add_item(&trans
->r_itemq
);
1559 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1563 ptr
= kmem_alloc(len
, KM_SLEEP
);
1564 memcpy(ptr
, dp
, len
);
1565 in_f
= (xfs_inode_log_format_t
*)ptr
;
1567 /* take the tail entry */
1568 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1569 if (item
->ri_total
!= 0 &&
1570 item
->ri_total
== item
->ri_cnt
) {
1571 /* tail item is in use, get a new one */
1572 xlog_recover_add_item(&trans
->r_itemq
);
1573 item
= list_entry(trans
->r_itemq
.prev
,
1574 xlog_recover_item_t
, ri_list
);
1577 if (item
->ri_total
== 0) { /* first region to be added */
1578 if (in_f
->ilf_size
== 0 ||
1579 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1581 "bad number of regions (%d) in inode log format",
1584 return XFS_ERROR(EIO
);
1587 item
->ri_total
= in_f
->ilf_size
;
1589 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1592 ASSERT(item
->ri_total
> item
->ri_cnt
);
1593 /* Description region is ri_buf[0] */
1594 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1595 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1597 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1602 * Sort the log items in the transaction.
1604 * The ordering constraints are defined by the inode allocation and unlink
1605 * behaviour. The rules are:
1607 * 1. Every item is only logged once in a given transaction. Hence it
1608 * represents the last logged state of the item. Hence ordering is
1609 * dependent on the order in which operations need to be performed so
1610 * required initial conditions are always met.
1612 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1613 * there's nothing to replay from them so we can simply cull them
1614 * from the transaction. However, we can't do that until after we've
1615 * replayed all the other items because they may be dependent on the
1616 * cancelled buffer and replaying the cancelled buffer can remove it
1617 * form the cancelled buffer table. Hence they have tobe done last.
1619 * 3. Inode allocation buffers must be replayed before inode items that
1620 * read the buffer and replay changes into it.
1622 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1623 * This ensures that inodes are completely flushed to the inode buffer
1624 * in a "free" state before we remove the unlinked inode list pointer.
1626 * Hence the ordering needs to be inode allocation buffers first, inode items
1627 * second, inode unlink buffers third and cancelled buffers last.
1629 * But there's a problem with that - we can't tell an inode allocation buffer
1630 * apart from a regular buffer, so we can't separate them. We can, however,
1631 * tell an inode unlink buffer from the others, and so we can separate them out
1632 * from all the other buffers and move them to last.
1634 * Hence, 4 lists, in order from head to tail:
1635 * - buffer_list for all buffers except cancelled/inode unlink buffers
1636 * - item_list for all non-buffer items
1637 * - inode_buffer_list for inode unlink buffers
1638 * - cancel_list for the cancelled buffers
1641 xlog_recover_reorder_trans(
1643 struct xlog_recover
*trans
,
1646 xlog_recover_item_t
*item
, *n
;
1647 LIST_HEAD(sort_list
);
1648 LIST_HEAD(cancel_list
);
1649 LIST_HEAD(buffer_list
);
1650 LIST_HEAD(inode_buffer_list
);
1651 LIST_HEAD(inode_list
);
1653 list_splice_init(&trans
->r_itemq
, &sort_list
);
1654 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1655 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1657 switch (ITEM_TYPE(item
)) {
1659 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1660 trace_xfs_log_recover_item_reorder_head(log
,
1662 list_move(&item
->ri_list
, &cancel_list
);
1665 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1666 list_move(&item
->ri_list
, &inode_buffer_list
);
1669 list_move_tail(&item
->ri_list
, &buffer_list
);
1673 case XFS_LI_QUOTAOFF
:
1676 trace_xfs_log_recover_item_reorder_tail(log
,
1678 list_move_tail(&item
->ri_list
, &inode_list
);
1682 "%s: unrecognized type of log operation",
1685 return XFS_ERROR(EIO
);
1688 ASSERT(list_empty(&sort_list
));
1689 if (!list_empty(&buffer_list
))
1690 list_splice(&buffer_list
, &trans
->r_itemq
);
1691 if (!list_empty(&inode_list
))
1692 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1693 if (!list_empty(&inode_buffer_list
))
1694 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1695 if (!list_empty(&cancel_list
))
1696 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1701 * Build up the table of buf cancel records so that we don't replay
1702 * cancelled data in the second pass. For buffer records that are
1703 * not cancel records, there is nothing to do here so we just return.
1705 * If we get a cancel record which is already in the table, this indicates
1706 * that the buffer was cancelled multiple times. In order to ensure
1707 * that during pass 2 we keep the record in the table until we reach its
1708 * last occurrence in the log, we keep a reference count in the cancel
1709 * record in the table to tell us how many times we expect to see this
1710 * record during the second pass.
1713 xlog_recover_buffer_pass1(
1715 struct xlog_recover_item
*item
)
1717 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1718 struct list_head
*bucket
;
1719 struct xfs_buf_cancel
*bcp
;
1722 * If this isn't a cancel buffer item, then just return.
1724 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1725 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1730 * Insert an xfs_buf_cancel record into the hash table of them.
1731 * If there is already an identical record, bump its reference count.
1733 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1734 list_for_each_entry(bcp
, bucket
, bc_list
) {
1735 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1736 bcp
->bc_len
== buf_f
->blf_len
) {
1738 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1743 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1744 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1745 bcp
->bc_len
= buf_f
->blf_len
;
1746 bcp
->bc_refcount
= 1;
1747 list_add_tail(&bcp
->bc_list
, bucket
);
1749 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1754 * Check to see whether the buffer being recovered has a corresponding
1755 * entry in the buffer cancel record table. If it does then return 1
1756 * so that it will be cancelled, otherwise return 0. If the buffer is
1757 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1758 * the refcount on the entry in the table and remove it from the table
1759 * if this is the last reference.
1761 * We remove the cancel record from the table when we encounter its
1762 * last occurrence in the log so that if the same buffer is re-used
1763 * again after its last cancellation we actually replay the changes
1764 * made at that point.
1767 xlog_check_buffer_cancelled(
1773 struct list_head
*bucket
;
1774 struct xfs_buf_cancel
*bcp
;
1776 if (log
->l_buf_cancel_table
== NULL
) {
1778 * There is nothing in the table built in pass one,
1779 * so this buffer must not be cancelled.
1781 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1786 * Search for an entry in the cancel table that matches our buffer.
1788 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1789 list_for_each_entry(bcp
, bucket
, bc_list
) {
1790 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1795 * We didn't find a corresponding entry in the table, so return 0 so
1796 * that the buffer is NOT cancelled.
1798 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1803 * We've go a match, so return 1 so that the recovery of this buffer
1804 * is cancelled. If this buffer is actually a buffer cancel log
1805 * item, then decrement the refcount on the one in the table and
1806 * remove it if this is the last reference.
1808 if (flags
& XFS_BLF_CANCEL
) {
1809 if (--bcp
->bc_refcount
== 0) {
1810 list_del(&bcp
->bc_list
);
1818 * Perform recovery for a buffer full of inodes. In these buffers, the only
1819 * data which should be recovered is that which corresponds to the
1820 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1821 * data for the inodes is always logged through the inodes themselves rather
1822 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1824 * The only time when buffers full of inodes are fully recovered is when the
1825 * buffer is full of newly allocated inodes. In this case the buffer will
1826 * not be marked as an inode buffer and so will be sent to
1827 * xlog_recover_do_reg_buffer() below during recovery.
1830 xlog_recover_do_inode_buffer(
1831 struct xfs_mount
*mp
,
1832 xlog_recover_item_t
*item
,
1834 xfs_buf_log_format_t
*buf_f
)
1840 int reg_buf_offset
= 0;
1841 int reg_buf_bytes
= 0;
1842 int next_unlinked_offset
;
1844 xfs_agino_t
*logged_nextp
;
1845 xfs_agino_t
*buffer_nextp
;
1847 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1848 bp
->b_ops
= &xfs_inode_buf_ops
;
1850 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1851 for (i
= 0; i
< inodes_per_buf
; i
++) {
1852 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1853 offsetof(xfs_dinode_t
, di_next_unlinked
);
1855 while (next_unlinked_offset
>=
1856 (reg_buf_offset
+ reg_buf_bytes
)) {
1858 * The next di_next_unlinked field is beyond
1859 * the current logged region. Find the next
1860 * logged region that contains or is beyond
1861 * the current di_next_unlinked field.
1864 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1865 buf_f
->blf_map_size
, bit
);
1868 * If there are no more logged regions in the
1869 * buffer, then we're done.
1874 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1875 buf_f
->blf_map_size
, bit
);
1877 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1878 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1883 * If the current logged region starts after the current
1884 * di_next_unlinked field, then move on to the next
1885 * di_next_unlinked field.
1887 if (next_unlinked_offset
< reg_buf_offset
)
1890 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1891 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1892 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1893 BBTOB(bp
->b_io_length
));
1896 * The current logged region contains a copy of the
1897 * current di_next_unlinked field. Extract its value
1898 * and copy it to the buffer copy.
1900 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1901 next_unlinked_offset
- reg_buf_offset
;
1902 if (unlikely(*logged_nextp
== 0)) {
1904 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1905 "Trying to replay bad (0) inode di_next_unlinked field.",
1907 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1908 XFS_ERRLEVEL_LOW
, mp
);
1909 return XFS_ERROR(EFSCORRUPTED
);
1912 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1913 next_unlinked_offset
);
1914 *buffer_nextp
= *logged_nextp
;
1917 * If necessary, recalculate the CRC in the on-disk inode. We
1918 * have to leave the inode in a consistent state for whoever
1921 xfs_dinode_calc_crc(mp
, (struct xfs_dinode
*)
1922 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
1930 * Validate the recovered buffer is of the correct type and attach the
1931 * appropriate buffer operations to them for writeback. Magic numbers are in a
1933 * the first 16 bits of the buffer (inode buffer, dquot buffer),
1934 * the first 32 bits of the buffer (most blocks),
1935 * inside a struct xfs_da_blkinfo at the start of the buffer.
1938 xlog_recovery_validate_buf_type(
1939 struct xfs_mount
*mp
,
1941 xfs_buf_log_format_t
*buf_f
)
1943 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
1948 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
1949 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
1950 magicda
= be16_to_cpu(info
->magic
);
1951 switch (xfs_blft_from_flags(buf_f
)) {
1952 case XFS_BLFT_BTREE_BUF
:
1954 case XFS_ABTB_CRC_MAGIC
:
1955 case XFS_ABTC_CRC_MAGIC
:
1956 case XFS_ABTB_MAGIC
:
1957 case XFS_ABTC_MAGIC
:
1958 bp
->b_ops
= &xfs_allocbt_buf_ops
;
1960 case XFS_IBT_CRC_MAGIC
:
1962 bp
->b_ops
= &xfs_inobt_buf_ops
;
1964 case XFS_BMAP_CRC_MAGIC
:
1965 case XFS_BMAP_MAGIC
:
1966 bp
->b_ops
= &xfs_bmbt_buf_ops
;
1969 xfs_warn(mp
, "Bad btree block magic!");
1974 case XFS_BLFT_AGF_BUF
:
1975 if (magic32
!= XFS_AGF_MAGIC
) {
1976 xfs_warn(mp
, "Bad AGF block magic!");
1980 bp
->b_ops
= &xfs_agf_buf_ops
;
1982 case XFS_BLFT_AGFL_BUF
:
1983 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1985 if (magic32
!= XFS_AGFL_MAGIC
) {
1986 xfs_warn(mp
, "Bad AGFL block magic!");
1990 bp
->b_ops
= &xfs_agfl_buf_ops
;
1992 case XFS_BLFT_AGI_BUF
:
1993 if (magic32
!= XFS_AGI_MAGIC
) {
1994 xfs_warn(mp
, "Bad AGI block magic!");
1998 bp
->b_ops
= &xfs_agi_buf_ops
;
2000 case XFS_BLFT_UDQUOT_BUF
:
2001 case XFS_BLFT_PDQUOT_BUF
:
2002 case XFS_BLFT_GDQUOT_BUF
:
2003 #ifdef CONFIG_XFS_QUOTA
2004 if (magic16
!= XFS_DQUOT_MAGIC
) {
2005 xfs_warn(mp
, "Bad DQUOT block magic!");
2009 bp
->b_ops
= &xfs_dquot_buf_ops
;
2012 "Trying to recover dquots without QUOTA support built in!");
2016 case XFS_BLFT_DINO_BUF
:
2018 * we get here with inode allocation buffers, not buffers that
2019 * track unlinked list changes.
2021 if (magic16
!= XFS_DINODE_MAGIC
) {
2022 xfs_warn(mp
, "Bad INODE block magic!");
2026 bp
->b_ops
= &xfs_inode_buf_ops
;
2028 case XFS_BLFT_SYMLINK_BUF
:
2029 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2030 xfs_warn(mp
, "Bad symlink block magic!");
2034 bp
->b_ops
= &xfs_symlink_buf_ops
;
2036 case XFS_BLFT_DIR_BLOCK_BUF
:
2037 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2038 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2039 xfs_warn(mp
, "Bad dir block magic!");
2043 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2045 case XFS_BLFT_DIR_DATA_BUF
:
2046 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2047 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2048 xfs_warn(mp
, "Bad dir data magic!");
2052 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2054 case XFS_BLFT_DIR_FREE_BUF
:
2055 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2056 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2057 xfs_warn(mp
, "Bad dir3 free magic!");
2061 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2063 case XFS_BLFT_DIR_LEAF1_BUF
:
2064 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2065 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2066 xfs_warn(mp
, "Bad dir leaf1 magic!");
2070 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2072 case XFS_BLFT_DIR_LEAFN_BUF
:
2073 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2074 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2075 xfs_warn(mp
, "Bad dir leafn magic!");
2079 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2081 case XFS_BLFT_DA_NODE_BUF
:
2082 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2083 magicda
!= XFS_DA3_NODE_MAGIC
) {
2084 xfs_warn(mp
, "Bad da node magic!");
2088 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2090 case XFS_BLFT_ATTR_LEAF_BUF
:
2091 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2092 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2093 xfs_warn(mp
, "Bad attr leaf magic!");
2097 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2099 case XFS_BLFT_ATTR_RMT_BUF
:
2100 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2102 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2103 xfs_warn(mp
, "Bad attr remote magic!");
2107 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2109 case XFS_BLFT_SB_BUF
:
2110 if (magic32
!= XFS_SB_MAGIC
) {
2111 xfs_warn(mp
, "Bad SB block magic!");
2115 bp
->b_ops
= &xfs_sb_buf_ops
;
2118 xfs_warn(mp
, "Unknown buffer type %d!",
2119 xfs_blft_from_flags(buf_f
));
2125 * Perform a 'normal' buffer recovery. Each logged region of the
2126 * buffer should be copied over the corresponding region in the
2127 * given buffer. The bitmap in the buf log format structure indicates
2128 * where to place the logged data.
2131 xlog_recover_do_reg_buffer(
2132 struct xfs_mount
*mp
,
2133 xlog_recover_item_t
*item
,
2135 xfs_buf_log_format_t
*buf_f
)
2142 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2145 i
= 1; /* 0 is the buf format structure */
2147 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2148 buf_f
->blf_map_size
, bit
);
2151 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2152 buf_f
->blf_map_size
, bit
);
2154 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2155 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2156 ASSERT(BBTOB(bp
->b_io_length
) >=
2157 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2160 * The dirty regions logged in the buffer, even though
2161 * contiguous, may span multiple chunks. This is because the
2162 * dirty region may span a physical page boundary in a buffer
2163 * and hence be split into two separate vectors for writing into
2164 * the log. Hence we need to trim nbits back to the length of
2165 * the current region being copied out of the log.
2167 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2168 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2171 * Do a sanity check if this is a dquot buffer. Just checking
2172 * the first dquot in the buffer should do. XXXThis is
2173 * probably a good thing to do for other buf types also.
2176 if (buf_f
->blf_flags
&
2177 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2178 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2180 "XFS: NULL dquot in %s.", __func__
);
2183 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2185 "XFS: dquot too small (%d) in %s.",
2186 item
->ri_buf
[i
].i_len
, __func__
);
2189 error
= xfs_qm_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2190 -1, 0, XFS_QMOPT_DOWARN
,
2191 "dquot_buf_recover");
2196 memcpy(xfs_buf_offset(bp
,
2197 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2198 item
->ri_buf
[i
].i_addr
, /* source */
2199 nbits
<<XFS_BLF_SHIFT
); /* length */
2205 /* Shouldn't be any more regions */
2206 ASSERT(i
== item
->ri_total
);
2208 xlog_recovery_validate_buf_type(mp
, bp
, buf_f
);
2212 * Do some primitive error checking on ondisk dquot data structures.
2216 struct xfs_mount
*mp
,
2217 xfs_disk_dquot_t
*ddq
,
2219 uint type
, /* used only when IO_dorepair is true */
2223 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
2227 * We can encounter an uninitialized dquot buffer for 2 reasons:
2228 * 1. If we crash while deleting the quotainode(s), and those blks got
2229 * used for user data. This is because we take the path of regular
2230 * file deletion; however, the size field of quotainodes is never
2231 * updated, so all the tricks that we play in itruncate_finish
2232 * don't quite matter.
2234 * 2. We don't play the quota buffers when there's a quotaoff logitem.
2235 * But the allocation will be replayed so we'll end up with an
2236 * uninitialized quota block.
2238 * This is all fine; things are still consistent, and we haven't lost
2239 * any quota information. Just don't complain about bad dquot blks.
2241 if (ddq
->d_magic
!= cpu_to_be16(XFS_DQUOT_MAGIC
)) {
2242 if (flags
& XFS_QMOPT_DOWARN
)
2244 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
2245 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
2248 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
2249 if (flags
& XFS_QMOPT_DOWARN
)
2251 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
2252 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
2256 if (ddq
->d_flags
!= XFS_DQ_USER
&&
2257 ddq
->d_flags
!= XFS_DQ_PROJ
&&
2258 ddq
->d_flags
!= XFS_DQ_GROUP
) {
2259 if (flags
& XFS_QMOPT_DOWARN
)
2261 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
2262 str
, id
, ddq
->d_flags
);
2266 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
2267 if (flags
& XFS_QMOPT_DOWARN
)
2269 "%s : ondisk-dquot 0x%p, ID mismatch: "
2270 "0x%x expected, found id 0x%x",
2271 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
2275 if (!errs
&& ddq
->d_id
) {
2276 if (ddq
->d_blk_softlimit
&&
2277 be64_to_cpu(ddq
->d_bcount
) >
2278 be64_to_cpu(ddq
->d_blk_softlimit
)) {
2279 if (!ddq
->d_btimer
) {
2280 if (flags
& XFS_QMOPT_DOWARN
)
2282 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
2283 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2287 if (ddq
->d_ino_softlimit
&&
2288 be64_to_cpu(ddq
->d_icount
) >
2289 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2290 if (!ddq
->d_itimer
) {
2291 if (flags
& XFS_QMOPT_DOWARN
)
2293 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2294 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2298 if (ddq
->d_rtb_softlimit
&&
2299 be64_to_cpu(ddq
->d_rtbcount
) >
2300 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2301 if (!ddq
->d_rtbtimer
) {
2302 if (flags
& XFS_QMOPT_DOWARN
)
2304 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2305 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2311 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2314 if (flags
& XFS_QMOPT_DOWARN
)
2315 xfs_notice(mp
, "Re-initializing dquot ID 0x%x", id
);
2318 * Typically, a repair is only requested by quotacheck.
2321 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2322 memset(d
, 0, sizeof(xfs_dqblk_t
));
2324 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2325 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2326 d
->dd_diskdq
.d_flags
= type
;
2327 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2329 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2330 uuid_copy(&d
->dd_uuid
, &mp
->m_sb
.sb_uuid
);
2331 xfs_update_cksum((char *)d
, sizeof(struct xfs_dqblk
),
2339 * Perform a dquot buffer recovery.
2340 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2341 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2342 * Else, treat it as a regular buffer and do recovery.
2345 xlog_recover_do_dquot_buffer(
2346 struct xfs_mount
*mp
,
2348 struct xlog_recover_item
*item
,
2350 struct xfs_buf_log_format
*buf_f
)
2354 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2357 * Filesystems are required to send in quota flags at mount time.
2359 if (mp
->m_qflags
== 0) {
2364 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2365 type
|= XFS_DQ_USER
;
2366 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2367 type
|= XFS_DQ_PROJ
;
2368 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2369 type
|= XFS_DQ_GROUP
;
2371 * This type of quotas was turned off, so ignore this buffer
2373 if (log
->l_quotaoffs_flag
& type
)
2376 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2380 * This routine replays a modification made to a buffer at runtime.
2381 * There are actually two types of buffer, regular and inode, which
2382 * are handled differently. Inode buffers are handled differently
2383 * in that we only recover a specific set of data from them, namely
2384 * the inode di_next_unlinked fields. This is because all other inode
2385 * data is actually logged via inode records and any data we replay
2386 * here which overlaps that may be stale.
2388 * When meta-data buffers are freed at run time we log a buffer item
2389 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2390 * of the buffer in the log should not be replayed at recovery time.
2391 * This is so that if the blocks covered by the buffer are reused for
2392 * file data before we crash we don't end up replaying old, freed
2393 * meta-data into a user's file.
2395 * To handle the cancellation of buffer log items, we make two passes
2396 * over the log during recovery. During the first we build a table of
2397 * those buffers which have been cancelled, and during the second we
2398 * only replay those buffers which do not have corresponding cancel
2399 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2400 * for more details on the implementation of the table of cancel records.
2403 xlog_recover_buffer_pass2(
2405 struct list_head
*buffer_list
,
2406 struct xlog_recover_item
*item
)
2408 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2409 xfs_mount_t
*mp
= log
->l_mp
;
2415 * In this pass we only want to recover all the buffers which have
2416 * not been cancelled and are not cancellation buffers themselves.
2418 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2419 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2420 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2424 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2427 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2428 buf_flags
|= XBF_UNMAPPED
;
2430 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2433 return XFS_ERROR(ENOMEM
);
2434 error
= bp
->b_error
;
2436 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2441 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2442 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2443 } else if (buf_f
->blf_flags
&
2444 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2445 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2447 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2450 return XFS_ERROR(error
);
2453 * Perform delayed write on the buffer. Asynchronous writes will be
2454 * slower when taking into account all the buffers to be flushed.
2456 * Also make sure that only inode buffers with good sizes stay in
2457 * the buffer cache. The kernel moves inodes in buffers of 1 block
2458 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2459 * buffers in the log can be a different size if the log was generated
2460 * by an older kernel using unclustered inode buffers or a newer kernel
2461 * running with a different inode cluster size. Regardless, if the
2462 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2463 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2464 * the buffer out of the buffer cache so that the buffer won't
2465 * overlap with future reads of those inodes.
2467 if (XFS_DINODE_MAGIC
==
2468 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2469 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2470 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2472 error
= xfs_bwrite(bp
);
2474 ASSERT(bp
->b_target
->bt_mount
== mp
);
2475 bp
->b_iodone
= xlog_recover_iodone
;
2476 xfs_buf_delwri_queue(bp
, buffer_list
);
2484 xlog_recover_inode_pass2(
2486 struct list_head
*buffer_list
,
2487 struct xlog_recover_item
*item
)
2489 xfs_inode_log_format_t
*in_f
;
2490 xfs_mount_t
*mp
= log
->l_mp
;
2499 xfs_icdinode_t
*dicp
;
2503 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2504 in_f
= item
->ri_buf
[0].i_addr
;
2506 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2508 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2514 * Inode buffers can be freed, look out for it,
2515 * and do not replay the inode.
2517 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2518 in_f
->ilf_len
, 0)) {
2520 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2523 trace_xfs_log_recover_inode_recover(log
, in_f
);
2525 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2526 &xfs_inode_buf_ops
);
2531 error
= bp
->b_error
;
2533 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2537 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2538 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2541 * Make sure the place we're flushing out to really looks
2544 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2547 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2548 __func__
, dip
, bp
, in_f
->ilf_ino
);
2549 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2550 XFS_ERRLEVEL_LOW
, mp
);
2551 error
= EFSCORRUPTED
;
2554 dicp
= item
->ri_buf
[1].i_addr
;
2555 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2558 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2559 __func__
, item
, in_f
->ilf_ino
);
2560 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2561 XFS_ERRLEVEL_LOW
, mp
);
2562 error
= EFSCORRUPTED
;
2566 /* Skip replay when the on disk inode is newer than the log one */
2567 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2569 * Deal with the wrap case, DI_MAX_FLUSH is less
2570 * than smaller numbers
2572 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2573 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2577 trace_xfs_log_recover_inode_skip(log
, in_f
);
2582 /* Take the opportunity to reset the flush iteration count */
2583 dicp
->di_flushiter
= 0;
2585 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2586 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2587 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2588 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2589 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2592 "%s: Bad regular inode log record, rec ptr 0x%p, "
2593 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2594 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2595 error
= EFSCORRUPTED
;
2598 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2599 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2600 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2601 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2602 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2603 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2606 "%s: Bad dir inode log record, rec ptr 0x%p, "
2607 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2608 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2609 error
= EFSCORRUPTED
;
2613 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2614 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2615 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2618 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2619 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2620 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2621 dicp
->di_nextents
+ dicp
->di_anextents
,
2623 error
= EFSCORRUPTED
;
2626 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2627 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2628 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2631 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2632 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2633 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2634 error
= EFSCORRUPTED
;
2637 isize
= xfs_icdinode_size(dicp
->di_version
);
2638 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2639 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2640 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2643 "%s: Bad inode log record length %d, rec ptr 0x%p",
2644 __func__
, item
->ri_buf
[1].i_len
, item
);
2645 error
= EFSCORRUPTED
;
2649 /* The core is in in-core format */
2650 xfs_dinode_to_disk(dip
, dicp
);
2652 /* the rest is in on-disk format */
2653 if (item
->ri_buf
[1].i_len
> isize
) {
2654 memcpy((char *)dip
+ isize
,
2655 item
->ri_buf
[1].i_addr
+ isize
,
2656 item
->ri_buf
[1].i_len
- isize
);
2659 fields
= in_f
->ilf_fields
;
2660 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2662 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2665 memcpy(XFS_DFORK_DPTR(dip
),
2666 &in_f
->ilf_u
.ilfu_uuid
,
2671 if (in_f
->ilf_size
== 2)
2672 goto write_inode_buffer
;
2673 len
= item
->ri_buf
[2].i_len
;
2674 src
= item
->ri_buf
[2].i_addr
;
2675 ASSERT(in_f
->ilf_size
<= 4);
2676 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2677 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2678 (len
== in_f
->ilf_dsize
));
2680 switch (fields
& XFS_ILOG_DFORK
) {
2681 case XFS_ILOG_DDATA
:
2683 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2686 case XFS_ILOG_DBROOT
:
2687 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2688 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2689 XFS_DFORK_DSIZE(dip
, mp
));
2694 * There are no data fork flags set.
2696 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2701 * If we logged any attribute data, recover it. There may or
2702 * may not have been any other non-core data logged in this
2705 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2706 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2711 len
= item
->ri_buf
[attr_index
].i_len
;
2712 src
= item
->ri_buf
[attr_index
].i_addr
;
2713 ASSERT(len
== in_f
->ilf_asize
);
2715 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2716 case XFS_ILOG_ADATA
:
2718 dest
= XFS_DFORK_APTR(dip
);
2719 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2720 memcpy(dest
, src
, len
);
2723 case XFS_ILOG_ABROOT
:
2724 dest
= XFS_DFORK_APTR(dip
);
2725 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2726 len
, (xfs_bmdr_block_t
*)dest
,
2727 XFS_DFORK_ASIZE(dip
, mp
));
2731 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2740 /* re-generate the checksum. */
2741 xfs_dinode_calc_crc(log
->l_mp
, dip
);
2743 ASSERT(bp
->b_target
->bt_mount
== mp
);
2744 bp
->b_iodone
= xlog_recover_iodone
;
2745 xfs_buf_delwri_queue(bp
, buffer_list
);
2750 return XFS_ERROR(error
);
2754 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2755 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2759 xlog_recover_quotaoff_pass1(
2761 struct xlog_recover_item
*item
)
2763 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2767 * The logitem format's flag tells us if this was user quotaoff,
2768 * group/project quotaoff or both.
2770 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2771 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2772 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2773 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2774 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2775 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2781 * Recover a dquot record
2784 xlog_recover_dquot_pass2(
2786 struct list_head
*buffer_list
,
2787 struct xlog_recover_item
*item
)
2789 xfs_mount_t
*mp
= log
->l_mp
;
2791 struct xfs_disk_dquot
*ddq
, *recddq
;
2793 xfs_dq_logformat_t
*dq_f
;
2798 * Filesystems are required to send in quota flags at mount time.
2800 if (mp
->m_qflags
== 0)
2803 recddq
= item
->ri_buf
[1].i_addr
;
2804 if (recddq
== NULL
) {
2805 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2806 return XFS_ERROR(EIO
);
2808 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2809 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2810 item
->ri_buf
[1].i_len
, __func__
);
2811 return XFS_ERROR(EIO
);
2815 * This type of quotas was turned off, so ignore this record.
2817 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2819 if (log
->l_quotaoffs_flag
& type
)
2823 * At this point we know that quota was _not_ turned off.
2824 * Since the mount flags are not indicating to us otherwise, this
2825 * must mean that quota is on, and the dquot needs to be replayed.
2826 * Remember that we may not have fully recovered the superblock yet,
2827 * so we can't do the usual trick of looking at the SB quota bits.
2829 * The other possibility, of course, is that the quota subsystem was
2830 * removed since the last mount - ENOSYS.
2832 dq_f
= item
->ri_buf
[0].i_addr
;
2834 error
= xfs_qm_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2835 "xlog_recover_dquot_pass2 (log copy)");
2837 return XFS_ERROR(EIO
);
2838 ASSERT(dq_f
->qlf_len
== 1);
2840 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
2841 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
2847 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2850 * At least the magic num portion should be on disk because this
2851 * was among a chunk of dquots created earlier, and we did some
2852 * minimal initialization then.
2854 error
= xfs_qm_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2855 "xlog_recover_dquot_pass2");
2858 return XFS_ERROR(EIO
);
2861 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2862 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
2863 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
2867 ASSERT(dq_f
->qlf_size
== 2);
2868 ASSERT(bp
->b_target
->bt_mount
== mp
);
2869 bp
->b_iodone
= xlog_recover_iodone
;
2870 xfs_buf_delwri_queue(bp
, buffer_list
);
2877 * This routine is called to create an in-core extent free intent
2878 * item from the efi format structure which was logged on disk.
2879 * It allocates an in-core efi, copies the extents from the format
2880 * structure into it, and adds the efi to the AIL with the given
2884 xlog_recover_efi_pass2(
2886 struct xlog_recover_item
*item
,
2890 xfs_mount_t
*mp
= log
->l_mp
;
2891 xfs_efi_log_item_t
*efip
;
2892 xfs_efi_log_format_t
*efi_formatp
;
2894 efi_formatp
= item
->ri_buf
[0].i_addr
;
2896 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2897 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2898 &(efip
->efi_format
)))) {
2899 xfs_efi_item_free(efip
);
2902 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
2904 spin_lock(&log
->l_ailp
->xa_lock
);
2906 * xfs_trans_ail_update() drops the AIL lock.
2908 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
2914 * This routine is called when an efd format structure is found in
2915 * a committed transaction in the log. It's purpose is to cancel
2916 * the corresponding efi if it was still in the log. To do this
2917 * it searches the AIL for the efi with an id equal to that in the
2918 * efd format structure. If we find it, we remove the efi from the
2922 xlog_recover_efd_pass2(
2924 struct xlog_recover_item
*item
)
2926 xfs_efd_log_format_t
*efd_formatp
;
2927 xfs_efi_log_item_t
*efip
= NULL
;
2928 xfs_log_item_t
*lip
;
2930 struct xfs_ail_cursor cur
;
2931 struct xfs_ail
*ailp
= log
->l_ailp
;
2933 efd_formatp
= item
->ri_buf
[0].i_addr
;
2934 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2935 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2936 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2937 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2938 efi_id
= efd_formatp
->efd_efi_id
;
2941 * Search for the efi with the id in the efd format structure
2944 spin_lock(&ailp
->xa_lock
);
2945 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2946 while (lip
!= NULL
) {
2947 if (lip
->li_type
== XFS_LI_EFI
) {
2948 efip
= (xfs_efi_log_item_t
*)lip
;
2949 if (efip
->efi_format
.efi_id
== efi_id
) {
2951 * xfs_trans_ail_delete() drops the
2954 xfs_trans_ail_delete(ailp
, lip
,
2955 SHUTDOWN_CORRUPT_INCORE
);
2956 xfs_efi_item_free(efip
);
2957 spin_lock(&ailp
->xa_lock
);
2961 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2963 xfs_trans_ail_cursor_done(ailp
, &cur
);
2964 spin_unlock(&ailp
->xa_lock
);
2970 * Free up any resources allocated by the transaction
2972 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2975 xlog_recover_free_trans(
2976 struct xlog_recover
*trans
)
2978 xlog_recover_item_t
*item
, *n
;
2981 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2982 /* Free the regions in the item. */
2983 list_del(&item
->ri_list
);
2984 for (i
= 0; i
< item
->ri_cnt
; i
++)
2985 kmem_free(item
->ri_buf
[i
].i_addr
);
2986 /* Free the item itself */
2987 kmem_free(item
->ri_buf
);
2990 /* Free the transaction recover structure */
2995 xlog_recover_commit_pass1(
2997 struct xlog_recover
*trans
,
2998 struct xlog_recover_item
*item
)
3000 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3002 switch (ITEM_TYPE(item
)) {
3004 return xlog_recover_buffer_pass1(log
, item
);
3005 case XFS_LI_QUOTAOFF
:
3006 return xlog_recover_quotaoff_pass1(log
, item
);
3011 /* nothing to do in pass 1 */
3014 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3015 __func__
, ITEM_TYPE(item
));
3017 return XFS_ERROR(EIO
);
3022 xlog_recover_commit_pass2(
3024 struct xlog_recover
*trans
,
3025 struct list_head
*buffer_list
,
3026 struct xlog_recover_item
*item
)
3028 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3030 switch (ITEM_TYPE(item
)) {
3032 return xlog_recover_buffer_pass2(log
, buffer_list
, item
);
3034 return xlog_recover_inode_pass2(log
, buffer_list
, item
);
3036 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3038 return xlog_recover_efd_pass2(log
, item
);
3040 return xlog_recover_dquot_pass2(log
, buffer_list
, item
);
3041 case XFS_LI_QUOTAOFF
:
3042 /* nothing to do in pass2 */
3045 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3046 __func__
, ITEM_TYPE(item
));
3048 return XFS_ERROR(EIO
);
3053 * Perform the transaction.
3055 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3056 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3059 xlog_recover_commit_trans(
3061 struct xlog_recover
*trans
,
3064 int error
= 0, error2
;
3065 xlog_recover_item_t
*item
;
3066 LIST_HEAD (buffer_list
);
3068 hlist_del(&trans
->r_list
);
3070 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3074 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
3076 case XLOG_RECOVER_PASS1
:
3077 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3079 case XLOG_RECOVER_PASS2
:
3080 error
= xlog_recover_commit_pass2(log
, trans
,
3081 &buffer_list
, item
);
3091 xlog_recover_free_trans(trans
);
3094 error2
= xfs_buf_delwri_submit(&buffer_list
);
3095 return error
? error
: error2
;
3099 xlog_recover_unmount_trans(
3101 struct xlog_recover
*trans
)
3103 /* Do nothing now */
3104 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3109 * There are two valid states of the r_state field. 0 indicates that the
3110 * transaction structure is in a normal state. We have either seen the
3111 * start of the transaction or the last operation we added was not a partial
3112 * operation. If the last operation we added to the transaction was a
3113 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3115 * NOTE: skip LRs with 0 data length.
3118 xlog_recover_process_data(
3120 struct hlist_head rhash
[],
3121 struct xlog_rec_header
*rhead
,
3127 xlog_op_header_t
*ohead
;
3128 xlog_recover_t
*trans
;
3134 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
3135 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
3137 /* check the log format matches our own - else we can't recover */
3138 if (xlog_header_check_recover(log
->l_mp
, rhead
))
3139 return (XFS_ERROR(EIO
));
3141 while ((dp
< lp
) && num_logops
) {
3142 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
3143 ohead
= (xlog_op_header_t
*)dp
;
3144 dp
+= sizeof(xlog_op_header_t
);
3145 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
3146 ohead
->oh_clientid
!= XFS_LOG
) {
3147 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
3148 __func__
, ohead
->oh_clientid
);
3150 return (XFS_ERROR(EIO
));
3152 tid
= be32_to_cpu(ohead
->oh_tid
);
3153 hash
= XLOG_RHASH(tid
);
3154 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
3155 if (trans
== NULL
) { /* not found; add new tid */
3156 if (ohead
->oh_flags
& XLOG_START_TRANS
)
3157 xlog_recover_new_tid(&rhash
[hash
], tid
,
3158 be64_to_cpu(rhead
->h_lsn
));
3160 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
3161 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
3162 __func__
, be32_to_cpu(ohead
->oh_len
));
3164 return (XFS_ERROR(EIO
));
3166 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
3167 if (flags
& XLOG_WAS_CONT_TRANS
)
3168 flags
&= ~XLOG_CONTINUE_TRANS
;
3170 case XLOG_COMMIT_TRANS
:
3171 error
= xlog_recover_commit_trans(log
,
3174 case XLOG_UNMOUNT_TRANS
:
3175 error
= xlog_recover_unmount_trans(log
, trans
);
3177 case XLOG_WAS_CONT_TRANS
:
3178 error
= xlog_recover_add_to_cont_trans(log
,
3180 be32_to_cpu(ohead
->oh_len
));
3182 case XLOG_START_TRANS
:
3183 xfs_warn(log
->l_mp
, "%s: bad transaction",
3186 error
= XFS_ERROR(EIO
);
3189 case XLOG_CONTINUE_TRANS
:
3190 error
= xlog_recover_add_to_trans(log
, trans
,
3191 dp
, be32_to_cpu(ohead
->oh_len
));
3194 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
3197 error
= XFS_ERROR(EIO
);
3203 dp
+= be32_to_cpu(ohead
->oh_len
);
3210 * Process an extent free intent item that was recovered from
3211 * the log. We need to free the extents that it describes.
3214 xlog_recover_process_efi(
3216 xfs_efi_log_item_t
*efip
)
3218 xfs_efd_log_item_t
*efdp
;
3223 xfs_fsblock_t startblock_fsb
;
3225 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
3228 * First check the validity of the extents described by the
3229 * EFI. If any are bad, then assume that all are bad and
3230 * just toss the EFI.
3232 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3233 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3234 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3235 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3236 if ((startblock_fsb
== 0) ||
3237 (extp
->ext_len
== 0) ||
3238 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3239 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3241 * This will pull the EFI from the AIL and
3242 * free the memory associated with it.
3244 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3245 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3246 return XFS_ERROR(EIO
);
3250 tp
= xfs_trans_alloc(mp
, 0);
3251 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
3254 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3256 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3257 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3258 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3261 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3265 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3266 error
= xfs_trans_commit(tp
, 0);
3270 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3275 * When this is called, all of the EFIs which did not have
3276 * corresponding EFDs should be in the AIL. What we do now
3277 * is free the extents associated with each one.
3279 * Since we process the EFIs in normal transactions, they
3280 * will be removed at some point after the commit. This prevents
3281 * us from just walking down the list processing each one.
3282 * We'll use a flag in the EFI to skip those that we've already
3283 * processed and use the AIL iteration mechanism's generation
3284 * count to try to speed this up at least a bit.
3286 * When we start, we know that the EFIs are the only things in
3287 * the AIL. As we process them, however, other items are added
3288 * to the AIL. Since everything added to the AIL must come after
3289 * everything already in the AIL, we stop processing as soon as
3290 * we see something other than an EFI in the AIL.
3293 xlog_recover_process_efis(
3296 xfs_log_item_t
*lip
;
3297 xfs_efi_log_item_t
*efip
;
3299 struct xfs_ail_cursor cur
;
3300 struct xfs_ail
*ailp
;
3303 spin_lock(&ailp
->xa_lock
);
3304 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3305 while (lip
!= NULL
) {
3307 * We're done when we see something other than an EFI.
3308 * There should be no EFIs left in the AIL now.
3310 if (lip
->li_type
!= XFS_LI_EFI
) {
3312 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3313 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3319 * Skip EFIs that we've already processed.
3321 efip
= (xfs_efi_log_item_t
*)lip
;
3322 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3323 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3327 spin_unlock(&ailp
->xa_lock
);
3328 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3329 spin_lock(&ailp
->xa_lock
);
3332 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3335 xfs_trans_ail_cursor_done(ailp
, &cur
);
3336 spin_unlock(&ailp
->xa_lock
);
3341 * This routine performs a transaction to null out a bad inode pointer
3342 * in an agi unlinked inode hash bucket.
3345 xlog_recover_clear_agi_bucket(
3347 xfs_agnumber_t agno
,
3356 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3357 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3362 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3366 agi
= XFS_BUF_TO_AGI(agibp
);
3367 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3368 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3369 (sizeof(xfs_agino_t
) * bucket
);
3370 xfs_trans_log_buf(tp
, agibp
, offset
,
3371 (offset
+ sizeof(xfs_agino_t
) - 1));
3373 error
= xfs_trans_commit(tp
, 0);
3379 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3381 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3386 xlog_recover_process_one_iunlink(
3387 struct xfs_mount
*mp
,
3388 xfs_agnumber_t agno
,
3392 struct xfs_buf
*ibp
;
3393 struct xfs_dinode
*dip
;
3394 struct xfs_inode
*ip
;
3398 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3399 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3404 * Get the on disk inode to find the next inode in the bucket.
3406 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3410 ASSERT(ip
->i_d
.di_nlink
== 0);
3411 ASSERT(ip
->i_d
.di_mode
!= 0);
3413 /* setup for the next pass */
3414 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3418 * Prevent any DMAPI event from being sent when the reference on
3419 * the inode is dropped.
3421 ip
->i_d
.di_dmevmask
= 0;
3430 * We can't read in the inode this bucket points to, or this inode
3431 * is messed up. Just ditch this bucket of inodes. We will lose
3432 * some inodes and space, but at least we won't hang.
3434 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3435 * clear the inode pointer in the bucket.
3437 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3442 * xlog_iunlink_recover
3444 * This is called during recovery to process any inodes which
3445 * we unlinked but not freed when the system crashed. These
3446 * inodes will be on the lists in the AGI blocks. What we do
3447 * here is scan all the AGIs and fully truncate and free any
3448 * inodes found on the lists. Each inode is removed from the
3449 * lists when it has been fully truncated and is freed. The
3450 * freeing of the inode and its removal from the list must be
3454 xlog_recover_process_iunlinks(
3458 xfs_agnumber_t agno
;
3469 * Prevent any DMAPI event from being sent while in this function.
3471 mp_dmevmask
= mp
->m_dmevmask
;
3474 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3476 * Find the agi for this ag.
3478 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3481 * AGI is b0rked. Don't process it.
3483 * We should probably mark the filesystem as corrupt
3484 * after we've recovered all the ag's we can....
3489 * Unlock the buffer so that it can be acquired in the normal
3490 * course of the transaction to truncate and free each inode.
3491 * Because we are not racing with anyone else here for the AGI
3492 * buffer, we don't even need to hold it locked to read the
3493 * initial unlinked bucket entries out of the buffer. We keep
3494 * buffer reference though, so that it stays pinned in memory
3495 * while we need the buffer.
3497 agi
= XFS_BUF_TO_AGI(agibp
);
3498 xfs_buf_unlock(agibp
);
3500 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3501 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3502 while (agino
!= NULLAGINO
) {
3503 agino
= xlog_recover_process_one_iunlink(mp
,
3504 agno
, agino
, bucket
);
3507 xfs_buf_rele(agibp
);
3510 mp
->m_dmevmask
= mp_dmevmask
;
3514 * Upack the log buffer data and crc check it. If the check fails, issue a
3515 * warning if and only if the CRC in the header is non-zero. This makes the
3516 * check an advisory warning, and the zero CRC check will prevent failure
3517 * warnings from being emitted when upgrading the kernel from one that does not
3518 * add CRCs by default.
3520 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3521 * corruption failure
3524 xlog_unpack_data_crc(
3525 struct xlog_rec_header
*rhead
,
3531 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
3532 if (crc
!= rhead
->h_crc
) {
3533 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
3534 xfs_alert(log
->l_mp
,
3535 "log record CRC mismatch: found 0x%x, expected 0x%x.\n",
3536 le32_to_cpu(rhead
->h_crc
),
3538 xfs_hex_dump(dp
, 32);
3542 * If we've detected a log record corruption, then we can't
3543 * recover past this point. Abort recovery if we are enforcing
3544 * CRC protection by punting an error back up the stack.
3546 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
3547 return EFSCORRUPTED
;
3555 struct xlog_rec_header
*rhead
,
3562 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
3566 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3567 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3568 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3572 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3573 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3574 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3575 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3576 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3577 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3586 xlog_valid_rec_header(
3588 struct xlog_rec_header
*rhead
,
3593 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
3594 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3595 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3596 return XFS_ERROR(EFSCORRUPTED
);
3599 (!rhead
->h_version
||
3600 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3601 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
3602 __func__
, be32_to_cpu(rhead
->h_version
));
3603 return XFS_ERROR(EIO
);
3606 /* LR body must have data or it wouldn't have been written */
3607 hlen
= be32_to_cpu(rhead
->h_len
);
3608 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3609 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3610 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3611 return XFS_ERROR(EFSCORRUPTED
);
3613 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3614 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3615 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3616 return XFS_ERROR(EFSCORRUPTED
);
3622 * Read the log from tail to head and process the log records found.
3623 * Handle the two cases where the tail and head are in the same cycle
3624 * and where the active portion of the log wraps around the end of
3625 * the physical log separately. The pass parameter is passed through
3626 * to the routines called to process the data and is not looked at
3630 xlog_do_recovery_pass(
3632 xfs_daddr_t head_blk
,
3633 xfs_daddr_t tail_blk
,
3636 xlog_rec_header_t
*rhead
;
3639 xfs_buf_t
*hbp
, *dbp
;
3640 int error
= 0, h_size
;
3641 int bblks
, split_bblks
;
3642 int hblks
, split_hblks
, wrapped_hblks
;
3643 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3645 ASSERT(head_blk
!= tail_blk
);
3648 * Read the header of the tail block and get the iclog buffer size from
3649 * h_size. Use this to tell how many sectors make up the log header.
3651 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3653 * When using variable length iclogs, read first sector of
3654 * iclog header and extract the header size from it. Get a
3655 * new hbp that is the correct size.
3657 hbp
= xlog_get_bp(log
, 1);
3661 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3665 rhead
= (xlog_rec_header_t
*)offset
;
3666 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3669 h_size
= be32_to_cpu(rhead
->h_size
);
3670 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3671 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3672 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3673 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3676 hbp
= xlog_get_bp(log
, hblks
);
3681 ASSERT(log
->l_sectBBsize
== 1);
3683 hbp
= xlog_get_bp(log
, 1);
3684 h_size
= XLOG_BIG_RECORD_BSIZE
;
3689 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3695 memset(rhash
, 0, sizeof(rhash
));
3696 if (tail_blk
<= head_blk
) {
3697 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3698 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3702 rhead
= (xlog_rec_header_t
*)offset
;
3703 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3707 /* blocks in data section */
3708 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3709 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3714 error
= xlog_unpack_data(rhead
, offset
, log
);
3718 error
= xlog_recover_process_data(log
,
3719 rhash
, rhead
, offset
, pass
);
3722 blk_no
+= bblks
+ hblks
;
3726 * Perform recovery around the end of the physical log.
3727 * When the head is not on the same cycle number as the tail,
3728 * we can't do a sequential recovery as above.
3731 while (blk_no
< log
->l_logBBsize
) {
3733 * Check for header wrapping around physical end-of-log
3735 offset
= hbp
->b_addr
;
3738 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3739 /* Read header in one read */
3740 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3745 /* This LR is split across physical log end */
3746 if (blk_no
!= log
->l_logBBsize
) {
3747 /* some data before physical log end */
3748 ASSERT(blk_no
<= INT_MAX
);
3749 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3750 ASSERT(split_hblks
> 0);
3751 error
= xlog_bread(log
, blk_no
,
3759 * Note: this black magic still works with
3760 * large sector sizes (non-512) only because:
3761 * - we increased the buffer size originally
3762 * by 1 sector giving us enough extra space
3763 * for the second read;
3764 * - the log start is guaranteed to be sector
3766 * - we read the log end (LR header start)
3767 * _first_, then the log start (LR header end)
3768 * - order is important.
3770 wrapped_hblks
= hblks
- split_hblks
;
3771 error
= xlog_bread_offset(log
, 0,
3773 offset
+ BBTOB(split_hblks
));
3777 rhead
= (xlog_rec_header_t
*)offset
;
3778 error
= xlog_valid_rec_header(log
, rhead
,
3779 split_hblks
? blk_no
: 0);
3783 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3786 /* Read in data for log record */
3787 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3788 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3793 /* This log record is split across the
3794 * physical end of log */
3795 offset
= dbp
->b_addr
;
3797 if (blk_no
!= log
->l_logBBsize
) {
3798 /* some data is before the physical
3800 ASSERT(!wrapped_hblks
);
3801 ASSERT(blk_no
<= INT_MAX
);
3803 log
->l_logBBsize
- (int)blk_no
;
3804 ASSERT(split_bblks
> 0);
3805 error
= xlog_bread(log
, blk_no
,
3813 * Note: this black magic still works with
3814 * large sector sizes (non-512) only because:
3815 * - we increased the buffer size originally
3816 * by 1 sector giving us enough extra space
3817 * for the second read;
3818 * - the log start is guaranteed to be sector
3820 * - we read the log end (LR header start)
3821 * _first_, then the log start (LR header end)
3822 * - order is important.
3824 error
= xlog_bread_offset(log
, 0,
3825 bblks
- split_bblks
, dbp
,
3826 offset
+ BBTOB(split_bblks
));
3831 error
= xlog_unpack_data(rhead
, offset
, log
);
3835 error
= xlog_recover_process_data(log
, rhash
,
3836 rhead
, offset
, pass
);
3842 ASSERT(blk_no
>= log
->l_logBBsize
);
3843 blk_no
-= log
->l_logBBsize
;
3845 /* read first part of physical log */
3846 while (blk_no
< head_blk
) {
3847 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3851 rhead
= (xlog_rec_header_t
*)offset
;
3852 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3856 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3857 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3862 error
= xlog_unpack_data(rhead
, offset
, log
);
3866 error
= xlog_recover_process_data(log
, rhash
,
3867 rhead
, offset
, pass
);
3870 blk_no
+= bblks
+ hblks
;
3882 * Do the recovery of the log. We actually do this in two phases.
3883 * The two passes are necessary in order to implement the function
3884 * of cancelling a record written into the log. The first pass
3885 * determines those things which have been cancelled, and the
3886 * second pass replays log items normally except for those which
3887 * have been cancelled. The handling of the replay and cancellations
3888 * takes place in the log item type specific routines.
3890 * The table of items which have cancel records in the log is allocated
3891 * and freed at this level, since only here do we know when all of
3892 * the log recovery has been completed.
3895 xlog_do_log_recovery(
3897 xfs_daddr_t head_blk
,
3898 xfs_daddr_t tail_blk
)
3902 ASSERT(head_blk
!= tail_blk
);
3905 * First do a pass to find all of the cancelled buf log items.
3906 * Store them in the buf_cancel_table for use in the second pass.
3908 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3909 sizeof(struct list_head
),
3911 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3912 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
3914 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3915 XLOG_RECOVER_PASS1
);
3917 kmem_free(log
->l_buf_cancel_table
);
3918 log
->l_buf_cancel_table
= NULL
;
3922 * Then do a second pass to actually recover the items in the log.
3923 * When it is complete free the table of buf cancel items.
3925 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3926 XLOG_RECOVER_PASS2
);
3931 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3932 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
3936 kmem_free(log
->l_buf_cancel_table
);
3937 log
->l_buf_cancel_table
= NULL
;
3943 * Do the actual recovery
3948 xfs_daddr_t head_blk
,
3949 xfs_daddr_t tail_blk
)
3956 * First replay the images in the log.
3958 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3963 * If IO errors happened during recovery, bail out.
3965 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3970 * We now update the tail_lsn since much of the recovery has completed
3971 * and there may be space available to use. If there were no extent
3972 * or iunlinks, we can free up the entire log and set the tail_lsn to
3973 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3974 * lsn of the last known good LR on disk. If there are extent frees
3975 * or iunlinks they will have some entries in the AIL; so we look at
3976 * the AIL to determine how to set the tail_lsn.
3978 xlog_assign_tail_lsn(log
->l_mp
);
3981 * Now that we've finished replaying all buffer and inode
3982 * updates, re-read in the superblock and reverify it.
3984 bp
= xfs_getsb(log
->l_mp
, 0);
3986 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3988 XFS_BUF_UNASYNC(bp
);
3989 bp
->b_ops
= &xfs_sb_buf_ops
;
3990 xfsbdstrat(log
->l_mp
, bp
);
3991 error
= xfs_buf_iowait(bp
);
3993 xfs_buf_ioerror_alert(bp
, __func__
);
3999 /* Convert superblock from on-disk format */
4000 sbp
= &log
->l_mp
->m_sb
;
4001 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4002 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4003 ASSERT(xfs_sb_good_version(sbp
));
4006 /* We've re-read the superblock so re-initialize per-cpu counters */
4007 xfs_icsb_reinit_counters(log
->l_mp
);
4009 xlog_recover_check_summary(log
);
4011 /* Normal transactions can now occur */
4012 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4017 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4019 * Return error or zero.
4025 xfs_daddr_t head_blk
, tail_blk
;
4028 /* find the tail of the log */
4029 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
4032 if (tail_blk
!= head_blk
) {
4033 /* There used to be a comment here:
4035 * disallow recovery on read-only mounts. note -- mount
4036 * checks for ENOSPC and turns it into an intelligent
4038 * ...but this is no longer true. Now, unless you specify
4039 * NORECOVERY (in which case this function would never be
4040 * called), we just go ahead and recover. We do this all
4041 * under the vfs layer, so we can get away with it unless
4042 * the device itself is read-only, in which case we fail.
4044 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
4049 * Version 5 superblock log feature mask validation. We know the
4050 * log is dirty so check if there are any unknown log features
4051 * in what we need to recover. If there are unknown features
4052 * (e.g. unsupported transactions, then simply reject the
4053 * attempt at recovery before touching anything.
4055 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
4056 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
4057 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
4059 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4060 "The log can not be fully and/or safely recovered by this kernel.\n"
4061 "Please recover the log on a kernel that supports the unknown features.",
4062 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
4063 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
4067 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
4068 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4071 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
4072 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
4078 * In the first part of recovery we replay inodes and buffers and build
4079 * up the list of extent free items which need to be processed. Here
4080 * we process the extent free items and clean up the on disk unlinked
4081 * inode lists. This is separated from the first part of recovery so
4082 * that the root and real-time bitmap inodes can be read in from disk in
4083 * between the two stages. This is necessary so that we can free space
4084 * in the real-time portion of the file system.
4087 xlog_recover_finish(
4091 * Now we're ready to do the transactions needed for the
4092 * rest of recovery. Start with completing all the extent
4093 * free intent records and then process the unlinked inode
4094 * lists. At this point, we essentially run in normal mode
4095 * except that we're still performing recovery actions
4096 * rather than accepting new requests.
4098 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
4100 error
= xlog_recover_process_efis(log
);
4102 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
4106 * Sync the log to get all the EFIs out of the AIL.
4107 * This isn't absolutely necessary, but it helps in
4108 * case the unlink transactions would have problems
4109 * pushing the EFIs out of the way.
4111 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
4113 xlog_recover_process_iunlinks(log
);
4115 xlog_recover_check_summary(log
);
4117 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
4118 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4120 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4122 xfs_info(log
->l_mp
, "Ending clean mount");
4130 * Read all of the agf and agi counters and check that they
4131 * are consistent with the superblock counters.
4134 xlog_recover_check_summary(
4141 xfs_agnumber_t agno
;
4142 __uint64_t freeblks
;
4152 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4153 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4155 xfs_alert(mp
, "%s agf read failed agno %d error %d",
4156 __func__
, agno
, error
);
4158 agfp
= XFS_BUF_TO_AGF(agfbp
);
4159 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4160 be32_to_cpu(agfp
->agf_flcount
);
4161 xfs_buf_relse(agfbp
);
4164 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4166 xfs_alert(mp
, "%s agi read failed agno %d error %d",
4167 __func__
, agno
, error
);
4169 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4171 itotal
+= be32_to_cpu(agi
->agi_count
);
4172 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4173 xfs_buf_relse(agibp
);