2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_error.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_btree.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_trans.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_log_priv.h"
43 #include "xfs_log_recover.h"
44 #include "xfs_extfree_item.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_quota.h"
47 #include "xfs_cksum.h"
48 #include "xfs_trace.h"
49 #include "xfs_icache.h"
51 /* Need all the magic numbers and buffer ops structures from these headers */
52 #include "xfs_da_btree.h"
55 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
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",
307 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
308 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
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 * 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__
);
967 return XFS_ERROR(EIO
);
970 /* find blk_no of tail of log */
971 rhead
= (xlog_rec_header_t
*)offset
;
972 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
975 * Reset log values according to the state of the log when we
976 * crashed. In the case where head_blk == 0, we bump curr_cycle
977 * one because the next write starts a new cycle rather than
978 * continuing the cycle of the last good log record. At this
979 * point we have guaranteed that all partial log records have been
980 * accounted for. Therefore, we know that the last good log record
981 * written was complete and ended exactly on the end boundary
982 * of the physical log.
984 log
->l_prev_block
= i
;
985 log
->l_curr_block
= (int)*head_blk
;
986 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
989 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
990 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
991 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
992 BBTOB(log
->l_curr_block
));
993 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
994 BBTOB(log
->l_curr_block
));
997 * Look for unmount record. If we find it, then we know there
998 * was a clean unmount. Since 'i' could be the last block in
999 * the physical log, we convert to a log block before comparing
1002 * Save the current tail lsn to use to pass to
1003 * xlog_clear_stale_blocks() below. We won't want to clear the
1004 * unmount record if there is one, so we pass the lsn of the
1005 * unmount record rather than the block after it.
1007 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1008 int h_size
= be32_to_cpu(rhead
->h_size
);
1009 int h_version
= be32_to_cpu(rhead
->h_version
);
1011 if ((h_version
& XLOG_VERSION_2
) &&
1012 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1013 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1014 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1022 after_umount_blk
= (i
+ hblks
+ (int)
1023 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1024 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1025 if (*head_blk
== after_umount_blk
&&
1026 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1027 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1028 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1032 op_head
= (xlog_op_header_t
*)offset
;
1033 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1035 * Set tail and last sync so that newly written
1036 * log records will point recovery to after the
1037 * current unmount record.
1039 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1040 log
->l_curr_cycle
, after_umount_blk
);
1041 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1042 log
->l_curr_cycle
, after_umount_blk
);
1043 *tail_blk
= after_umount_blk
;
1046 * Note that the unmount was clean. If the unmount
1047 * was not clean, we need to know this to rebuild the
1048 * superblock counters from the perag headers if we
1049 * have a filesystem using non-persistent counters.
1051 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1056 * Make sure that there are no blocks in front of the head
1057 * with the same cycle number as the head. This can happen
1058 * because we allow multiple outstanding log writes concurrently,
1059 * and the later writes might make it out before earlier ones.
1061 * We use the lsn from before modifying it so that we'll never
1062 * overwrite the unmount record after a clean unmount.
1064 * Do this only if we are going to recover the filesystem
1066 * NOTE: This used to say "if (!readonly)"
1067 * However on Linux, we can & do recover a read-only filesystem.
1068 * We only skip recovery if NORECOVERY is specified on mount,
1069 * in which case we would not be here.
1071 * But... if the -device- itself is readonly, just skip this.
1072 * We can't recover this device anyway, so it won't matter.
1074 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1075 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1081 xfs_warn(log
->l_mp
, "failed to locate log tail");
1086 * Is the log zeroed at all?
1088 * The last binary search should be changed to perform an X block read
1089 * once X becomes small enough. You can then search linearly through
1090 * the X blocks. This will cut down on the number of reads we need to do.
1092 * If the log is partially zeroed, this routine will pass back the blkno
1093 * of the first block with cycle number 0. It won't have a complete LR
1097 * 0 => the log is completely written to
1098 * -1 => use *blk_no as the first block of the log
1099 * >0 => error has occurred
1104 xfs_daddr_t
*blk_no
)
1108 uint first_cycle
, last_cycle
;
1109 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1110 xfs_daddr_t num_scan_bblks
;
1111 int error
, log_bbnum
= log
->l_logBBsize
;
1115 /* check totally zeroed log */
1116 bp
= xlog_get_bp(log
, 1);
1119 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1123 first_cycle
= xlog_get_cycle(offset
);
1124 if (first_cycle
== 0) { /* completely zeroed log */
1130 /* check partially zeroed log */
1131 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1135 last_cycle
= xlog_get_cycle(offset
);
1136 if (last_cycle
!= 0) { /* log completely written to */
1139 } else if (first_cycle
!= 1) {
1141 * If the cycle of the last block is zero, the cycle of
1142 * the first block must be 1. If it's not, maybe we're
1143 * not looking at a log... Bail out.
1146 "Log inconsistent or not a log (last==0, first!=1)");
1147 error
= XFS_ERROR(EINVAL
);
1151 /* we have a partially zeroed log */
1152 last_blk
= log_bbnum
-1;
1153 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1157 * Validate the answer. Because there is no way to guarantee that
1158 * the entire log is made up of log records which are the same size,
1159 * we scan over the defined maximum blocks. At this point, the maximum
1160 * is not chosen to mean anything special. XXXmiken
1162 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1163 ASSERT(num_scan_bblks
<= INT_MAX
);
1165 if (last_blk
< num_scan_bblks
)
1166 num_scan_bblks
= last_blk
;
1167 start_blk
= last_blk
- num_scan_bblks
;
1170 * We search for any instances of cycle number 0 that occur before
1171 * our current estimate of the head. What we're trying to detect is
1172 * 1 ... | 0 | 1 | 0...
1173 * ^ binary search ends here
1175 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1176 (int)num_scan_bblks
, 0, &new_blk
)))
1182 * Potentially backup over partial log record write. We don't need
1183 * to search the end of the log because we know it is zero.
1185 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1186 &last_blk
, 0)) == -1) {
1187 error
= XFS_ERROR(EIO
);
1201 * These are simple subroutines used by xlog_clear_stale_blocks() below
1202 * to initialize a buffer full of empty log record headers and write
1203 * them into the log.
1214 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1216 memset(buf
, 0, BBSIZE
);
1217 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1218 recp
->h_cycle
= cpu_to_be32(cycle
);
1219 recp
->h_version
= cpu_to_be32(
1220 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1221 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1222 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1223 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1224 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1228 xlog_write_log_records(
1239 int sectbb
= log
->l_sectBBsize
;
1240 int end_block
= start_block
+ blocks
;
1246 * Greedily allocate a buffer big enough to handle the full
1247 * range of basic blocks to be written. If that fails, try
1248 * a smaller size. We need to be able to write at least a
1249 * log sector, or we're out of luck.
1251 bufblks
= 1 << ffs(blocks
);
1252 while (bufblks
> log
->l_logBBsize
)
1254 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1256 if (bufblks
< sectbb
)
1260 /* We may need to do a read at the start to fill in part of
1261 * the buffer in the starting sector not covered by the first
1264 balign
= round_down(start_block
, sectbb
);
1265 if (balign
!= start_block
) {
1266 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1270 j
= start_block
- balign
;
1273 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1274 int bcount
, endcount
;
1276 bcount
= min(bufblks
, end_block
- start_block
);
1277 endcount
= bcount
- j
;
1279 /* We may need to do a read at the end to fill in part of
1280 * the buffer in the final sector not covered by the write.
1281 * If this is the same sector as the above read, skip it.
1283 ealign
= round_down(end_block
, sectbb
);
1284 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1285 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1286 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1293 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1294 for (; j
< endcount
; j
++) {
1295 xlog_add_record(log
, offset
, cycle
, i
+j
,
1296 tail_cycle
, tail_block
);
1299 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1302 start_block
+= endcount
;
1312 * This routine is called to blow away any incomplete log writes out
1313 * in front of the log head. We do this so that we won't become confused
1314 * if we come up, write only a little bit more, and then crash again.
1315 * If we leave the partial log records out there, this situation could
1316 * cause us to think those partial writes are valid blocks since they
1317 * have the current cycle number. We get rid of them by overwriting them
1318 * with empty log records with the old cycle number rather than the
1321 * The tail lsn is passed in rather than taken from
1322 * the log so that we will not write over the unmount record after a
1323 * clean unmount in a 512 block log. Doing so would leave the log without
1324 * any valid log records in it until a new one was written. If we crashed
1325 * during that time we would not be able to recover.
1328 xlog_clear_stale_blocks(
1332 int tail_cycle
, head_cycle
;
1333 int tail_block
, head_block
;
1334 int tail_distance
, max_distance
;
1338 tail_cycle
= CYCLE_LSN(tail_lsn
);
1339 tail_block
= BLOCK_LSN(tail_lsn
);
1340 head_cycle
= log
->l_curr_cycle
;
1341 head_block
= log
->l_curr_block
;
1344 * Figure out the distance between the new head of the log
1345 * and the tail. We want to write over any blocks beyond the
1346 * head that we may have written just before the crash, but
1347 * we don't want to overwrite the tail of the log.
1349 if (head_cycle
== tail_cycle
) {
1351 * The tail is behind the head in the physical log,
1352 * so the distance from the head to the tail is the
1353 * distance from the head to the end of the log plus
1354 * the distance from the beginning of the log to the
1357 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1358 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1359 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1360 return XFS_ERROR(EFSCORRUPTED
);
1362 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1365 * The head is behind the tail in the physical log,
1366 * so the distance from the head to the tail is just
1367 * the tail block minus the head block.
1369 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1370 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1371 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1372 return XFS_ERROR(EFSCORRUPTED
);
1374 tail_distance
= tail_block
- head_block
;
1378 * If the head is right up against the tail, we can't clear
1381 if (tail_distance
<= 0) {
1382 ASSERT(tail_distance
== 0);
1386 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1388 * Take the smaller of the maximum amount of outstanding I/O
1389 * we could have and the distance to the tail to clear out.
1390 * We take the smaller so that we don't overwrite the tail and
1391 * we don't waste all day writing from the head to the tail
1394 max_distance
= MIN(max_distance
, tail_distance
);
1396 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1398 * We can stomp all the blocks we need to without
1399 * wrapping around the end of the log. Just do it
1400 * in a single write. Use the cycle number of the
1401 * current cycle minus one so that the log will look like:
1404 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1405 head_block
, max_distance
, tail_cycle
,
1411 * We need to wrap around the end of the physical log in
1412 * order to clear all the blocks. Do it in two separate
1413 * I/Os. The first write should be from the head to the
1414 * end of the physical log, and it should use the current
1415 * cycle number minus one just like above.
1417 distance
= log
->l_logBBsize
- head_block
;
1418 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1419 head_block
, distance
, tail_cycle
,
1426 * Now write the blocks at the start of the physical log.
1427 * This writes the remainder of the blocks we want to clear.
1428 * It uses the current cycle number since we're now on the
1429 * same cycle as the head so that we get:
1430 * n ... n ... | n - 1 ...
1431 * ^^^^^ blocks we're writing
1433 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1434 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1435 tail_cycle
, tail_block
);
1443 /******************************************************************************
1445 * Log recover routines
1447 ******************************************************************************
1450 STATIC xlog_recover_t
*
1451 xlog_recover_find_tid(
1452 struct hlist_head
*head
,
1455 xlog_recover_t
*trans
;
1457 hlist_for_each_entry(trans
, head
, r_list
) {
1458 if (trans
->r_log_tid
== tid
)
1465 xlog_recover_new_tid(
1466 struct hlist_head
*head
,
1470 xlog_recover_t
*trans
;
1472 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1473 trans
->r_log_tid
= tid
;
1475 INIT_LIST_HEAD(&trans
->r_itemq
);
1477 INIT_HLIST_NODE(&trans
->r_list
);
1478 hlist_add_head(&trans
->r_list
, head
);
1482 xlog_recover_add_item(
1483 struct list_head
*head
)
1485 xlog_recover_item_t
*item
;
1487 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1488 INIT_LIST_HEAD(&item
->ri_list
);
1489 list_add_tail(&item
->ri_list
, head
);
1493 xlog_recover_add_to_cont_trans(
1495 struct xlog_recover
*trans
,
1499 xlog_recover_item_t
*item
;
1500 xfs_caddr_t ptr
, old_ptr
;
1503 if (list_empty(&trans
->r_itemq
)) {
1504 /* finish copying rest of trans header */
1505 xlog_recover_add_item(&trans
->r_itemq
);
1506 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1507 sizeof(xfs_trans_header_t
) - len
;
1508 memcpy(ptr
, dp
, len
); /* d, s, l */
1511 /* take the tail entry */
1512 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1514 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1515 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1517 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, KM_SLEEP
);
1518 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1519 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1520 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1521 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1526 * The next region to add is the start of a new region. It could be
1527 * a whole region or it could be the first part of a new region. Because
1528 * of this, the assumption here is that the type and size fields of all
1529 * format structures fit into the first 32 bits of the structure.
1531 * This works because all regions must be 32 bit aligned. Therefore, we
1532 * either have both fields or we have neither field. In the case we have
1533 * neither field, the data part of the region is zero length. We only have
1534 * a log_op_header and can throw away the header since a new one will appear
1535 * later. If we have at least 4 bytes, then we can determine how many regions
1536 * will appear in the current log item.
1539 xlog_recover_add_to_trans(
1541 struct xlog_recover
*trans
,
1545 xfs_inode_log_format_t
*in_f
; /* any will do */
1546 xlog_recover_item_t
*item
;
1551 if (list_empty(&trans
->r_itemq
)) {
1552 /* we need to catch log corruptions here */
1553 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1554 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1557 return XFS_ERROR(EIO
);
1559 if (len
== sizeof(xfs_trans_header_t
))
1560 xlog_recover_add_item(&trans
->r_itemq
);
1561 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1565 ptr
= kmem_alloc(len
, KM_SLEEP
);
1566 memcpy(ptr
, dp
, len
);
1567 in_f
= (xfs_inode_log_format_t
*)ptr
;
1569 /* take the tail entry */
1570 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1571 if (item
->ri_total
!= 0 &&
1572 item
->ri_total
== item
->ri_cnt
) {
1573 /* tail item is in use, get a new one */
1574 xlog_recover_add_item(&trans
->r_itemq
);
1575 item
= list_entry(trans
->r_itemq
.prev
,
1576 xlog_recover_item_t
, ri_list
);
1579 if (item
->ri_total
== 0) { /* first region to be added */
1580 if (in_f
->ilf_size
== 0 ||
1581 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1583 "bad number of regions (%d) in inode log format",
1587 return XFS_ERROR(EIO
);
1590 item
->ri_total
= in_f
->ilf_size
;
1592 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1595 ASSERT(item
->ri_total
> item
->ri_cnt
);
1596 /* Description region is ri_buf[0] */
1597 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1598 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1600 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1605 * Sort the log items in the transaction.
1607 * The ordering constraints are defined by the inode allocation and unlink
1608 * behaviour. The rules are:
1610 * 1. Every item is only logged once in a given transaction. Hence it
1611 * represents the last logged state of the item. Hence ordering is
1612 * dependent on the order in which operations need to be performed so
1613 * required initial conditions are always met.
1615 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1616 * there's nothing to replay from them so we can simply cull them
1617 * from the transaction. However, we can't do that until after we've
1618 * replayed all the other items because they may be dependent on the
1619 * cancelled buffer and replaying the cancelled buffer can remove it
1620 * form the cancelled buffer table. Hence they have tobe done last.
1622 * 3. Inode allocation buffers must be replayed before inode items that
1623 * read the buffer and replay changes into it. For filesystems using the
1624 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1625 * treated the same as inode allocation buffers as they create and
1626 * initialise the buffers directly.
1628 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1629 * This ensures that inodes are completely flushed to the inode buffer
1630 * in a "free" state before we remove the unlinked inode list pointer.
1632 * Hence the ordering needs to be inode allocation buffers first, inode items
1633 * second, inode unlink buffers third and cancelled buffers last.
1635 * But there's a problem with that - we can't tell an inode allocation buffer
1636 * apart from a regular buffer, so we can't separate them. We can, however,
1637 * tell an inode unlink buffer from the others, and so we can separate them out
1638 * from all the other buffers and move them to last.
1640 * Hence, 4 lists, in order from head to tail:
1641 * - buffer_list for all buffers except cancelled/inode unlink buffers
1642 * - item_list for all non-buffer items
1643 * - inode_buffer_list for inode unlink buffers
1644 * - cancel_list for the cancelled buffers
1646 * Note that we add objects to the tail of the lists so that first-to-last
1647 * ordering is preserved within the lists. Adding objects to the head of the
1648 * list means when we traverse from the head we walk them in last-to-first
1649 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1650 * but for all other items there may be specific ordering that we need to
1654 xlog_recover_reorder_trans(
1656 struct xlog_recover
*trans
,
1659 xlog_recover_item_t
*item
, *n
;
1660 LIST_HEAD(sort_list
);
1661 LIST_HEAD(cancel_list
);
1662 LIST_HEAD(buffer_list
);
1663 LIST_HEAD(inode_buffer_list
);
1664 LIST_HEAD(inode_list
);
1666 list_splice_init(&trans
->r_itemq
, &sort_list
);
1667 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1668 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1670 switch (ITEM_TYPE(item
)) {
1671 case XFS_LI_ICREATE
:
1672 list_move_tail(&item
->ri_list
, &buffer_list
);
1675 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1676 trace_xfs_log_recover_item_reorder_head(log
,
1678 list_move(&item
->ri_list
, &cancel_list
);
1681 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1682 list_move(&item
->ri_list
, &inode_buffer_list
);
1685 list_move_tail(&item
->ri_list
, &buffer_list
);
1689 case XFS_LI_QUOTAOFF
:
1692 trace_xfs_log_recover_item_reorder_tail(log
,
1694 list_move_tail(&item
->ri_list
, &inode_list
);
1698 "%s: unrecognized type of log operation",
1701 return XFS_ERROR(EIO
);
1704 ASSERT(list_empty(&sort_list
));
1705 if (!list_empty(&buffer_list
))
1706 list_splice(&buffer_list
, &trans
->r_itemq
);
1707 if (!list_empty(&inode_list
))
1708 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1709 if (!list_empty(&inode_buffer_list
))
1710 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1711 if (!list_empty(&cancel_list
))
1712 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1717 * Build up the table of buf cancel records so that we don't replay
1718 * cancelled data in the second pass. For buffer records that are
1719 * not cancel records, there is nothing to do here so we just return.
1721 * If we get a cancel record which is already in the table, this indicates
1722 * that the buffer was cancelled multiple times. In order to ensure
1723 * that during pass 2 we keep the record in the table until we reach its
1724 * last occurrence in the log, we keep a reference count in the cancel
1725 * record in the table to tell us how many times we expect to see this
1726 * record during the second pass.
1729 xlog_recover_buffer_pass1(
1731 struct xlog_recover_item
*item
)
1733 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1734 struct list_head
*bucket
;
1735 struct xfs_buf_cancel
*bcp
;
1738 * If this isn't a cancel buffer item, then just return.
1740 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1741 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1746 * Insert an xfs_buf_cancel record into the hash table of them.
1747 * If there is already an identical record, bump its reference count.
1749 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1750 list_for_each_entry(bcp
, bucket
, bc_list
) {
1751 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1752 bcp
->bc_len
== buf_f
->blf_len
) {
1754 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1759 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1760 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1761 bcp
->bc_len
= buf_f
->blf_len
;
1762 bcp
->bc_refcount
= 1;
1763 list_add_tail(&bcp
->bc_list
, bucket
);
1765 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1770 * Check to see whether the buffer being recovered has a corresponding
1771 * entry in the buffer cancel record table. If it is, return the cancel
1772 * buffer structure to the caller.
1774 STATIC
struct xfs_buf_cancel
*
1775 xlog_peek_buffer_cancelled(
1781 struct list_head
*bucket
;
1782 struct xfs_buf_cancel
*bcp
;
1784 if (!log
->l_buf_cancel_table
) {
1785 /* empty table means no cancelled buffers in the log */
1786 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1790 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1791 list_for_each_entry(bcp
, bucket
, bc_list
) {
1792 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1797 * We didn't find a corresponding entry in the table, so return 0 so
1798 * that the buffer is NOT cancelled.
1800 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1805 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1806 * otherwise return 0. If the buffer is actually a buffer cancel item
1807 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1808 * table and remove it from the table if this is the last reference.
1810 * We remove the cancel record from the table when we encounter its last
1811 * occurrence in the log so that if the same buffer is re-used again after its
1812 * last cancellation we actually replay the changes made at that point.
1815 xlog_check_buffer_cancelled(
1821 struct xfs_buf_cancel
*bcp
;
1823 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
1828 * We've go a match, so return 1 so that the recovery of this buffer
1829 * is cancelled. If this buffer is actually a buffer cancel log
1830 * item, then decrement the refcount on the one in the table and
1831 * remove it if this is the last reference.
1833 if (flags
& XFS_BLF_CANCEL
) {
1834 if (--bcp
->bc_refcount
== 0) {
1835 list_del(&bcp
->bc_list
);
1843 * Perform recovery for a buffer full of inodes. In these buffers, the only
1844 * data which should be recovered is that which corresponds to the
1845 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1846 * data for the inodes is always logged through the inodes themselves rather
1847 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1849 * The only time when buffers full of inodes are fully recovered is when the
1850 * buffer is full of newly allocated inodes. In this case the buffer will
1851 * not be marked as an inode buffer and so will be sent to
1852 * xlog_recover_do_reg_buffer() below during recovery.
1855 xlog_recover_do_inode_buffer(
1856 struct xfs_mount
*mp
,
1857 xlog_recover_item_t
*item
,
1859 xfs_buf_log_format_t
*buf_f
)
1865 int reg_buf_offset
= 0;
1866 int reg_buf_bytes
= 0;
1867 int next_unlinked_offset
;
1869 xfs_agino_t
*logged_nextp
;
1870 xfs_agino_t
*buffer_nextp
;
1872 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1875 * Post recovery validation only works properly on CRC enabled
1878 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1879 bp
->b_ops
= &xfs_inode_buf_ops
;
1881 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
1882 for (i
= 0; i
< inodes_per_buf
; i
++) {
1883 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1884 offsetof(xfs_dinode_t
, di_next_unlinked
);
1886 while (next_unlinked_offset
>=
1887 (reg_buf_offset
+ reg_buf_bytes
)) {
1889 * The next di_next_unlinked field is beyond
1890 * the current logged region. Find the next
1891 * logged region that contains or is beyond
1892 * the current di_next_unlinked field.
1895 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1896 buf_f
->blf_map_size
, bit
);
1899 * If there are no more logged regions in the
1900 * buffer, then we're done.
1905 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1906 buf_f
->blf_map_size
, bit
);
1908 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1909 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1914 * If the current logged region starts after the current
1915 * di_next_unlinked field, then move on to the next
1916 * di_next_unlinked field.
1918 if (next_unlinked_offset
< reg_buf_offset
)
1921 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1922 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1923 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
1924 BBTOB(bp
->b_io_length
));
1927 * The current logged region contains a copy of the
1928 * current di_next_unlinked field. Extract its value
1929 * and copy it to the buffer copy.
1931 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1932 next_unlinked_offset
- reg_buf_offset
;
1933 if (unlikely(*logged_nextp
== 0)) {
1935 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1936 "Trying to replay bad (0) inode di_next_unlinked field.",
1938 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1939 XFS_ERRLEVEL_LOW
, mp
);
1940 return XFS_ERROR(EFSCORRUPTED
);
1943 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1944 next_unlinked_offset
);
1945 *buffer_nextp
= *logged_nextp
;
1948 * If necessary, recalculate the CRC in the on-disk inode. We
1949 * have to leave the inode in a consistent state for whoever
1952 xfs_dinode_calc_crc(mp
, (struct xfs_dinode
*)
1953 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
1961 * V5 filesystems know the age of the buffer on disk being recovered. We can
1962 * have newer objects on disk than we are replaying, and so for these cases we
1963 * don't want to replay the current change as that will make the buffer contents
1964 * temporarily invalid on disk.
1966 * The magic number might not match the buffer type we are going to recover
1967 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
1968 * extract the LSN of the existing object in the buffer based on it's current
1969 * magic number. If we don't recognise the magic number in the buffer, then
1970 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1971 * so can recover the buffer.
1973 * Note: we cannot rely solely on magic number matches to determine that the
1974 * buffer has a valid LSN - we also need to verify that it belongs to this
1975 * filesystem, so we need to extract the object's LSN and compare it to that
1976 * which we read from the superblock. If the UUIDs don't match, then we've got a
1977 * stale metadata block from an old filesystem instance that we need to recover
1981 xlog_recover_get_buf_lsn(
1982 struct xfs_mount
*mp
,
1988 void *blk
= bp
->b_addr
;
1992 /* v4 filesystems always recover immediately */
1993 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
1994 goto recover_immediately
;
1996 magic32
= be32_to_cpu(*(__be32
*)blk
);
1998 case XFS_ABTB_CRC_MAGIC
:
1999 case XFS_ABTC_CRC_MAGIC
:
2000 case XFS_ABTB_MAGIC
:
2001 case XFS_ABTC_MAGIC
:
2002 case XFS_IBT_CRC_MAGIC
:
2003 case XFS_IBT_MAGIC
: {
2004 struct xfs_btree_block
*btb
= blk
;
2006 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
2007 uuid
= &btb
->bb_u
.s
.bb_uuid
;
2010 case XFS_BMAP_CRC_MAGIC
:
2011 case XFS_BMAP_MAGIC
: {
2012 struct xfs_btree_block
*btb
= blk
;
2014 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
2015 uuid
= &btb
->bb_u
.l
.bb_uuid
;
2019 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
2020 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
2022 case XFS_AGFL_MAGIC
:
2023 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
2024 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
2027 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
2028 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
2030 case XFS_SYMLINK_MAGIC
:
2031 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
2032 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
2034 case XFS_DIR3_BLOCK_MAGIC
:
2035 case XFS_DIR3_DATA_MAGIC
:
2036 case XFS_DIR3_FREE_MAGIC
:
2037 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
2038 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
2040 case XFS_ATTR3_RMT_MAGIC
:
2041 lsn
= be64_to_cpu(((struct xfs_attr3_rmt_hdr
*)blk
)->rm_lsn
);
2042 uuid
= &((struct xfs_attr3_rmt_hdr
*)blk
)->rm_uuid
;
2045 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
2046 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
2052 if (lsn
!= (xfs_lsn_t
)-1) {
2053 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2054 goto recover_immediately
;
2058 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
2060 case XFS_DIR3_LEAF1_MAGIC
:
2061 case XFS_DIR3_LEAFN_MAGIC
:
2062 case XFS_DA3_NODE_MAGIC
:
2063 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
2064 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
2070 if (lsn
!= (xfs_lsn_t
)-1) {
2071 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2072 goto recover_immediately
;
2077 * We do individual object checks on dquot and inode buffers as they
2078 * have their own individual LSN records. Also, we could have a stale
2079 * buffer here, so we have to at least recognise these buffer types.
2081 * A notd complexity here is inode unlinked list processing - it logs
2082 * the inode directly in the buffer, but we don't know which inodes have
2083 * been modified, and there is no global buffer LSN. Hence we need to
2084 * recover all inode buffer types immediately. This problem will be
2085 * fixed by logical logging of the unlinked list modifications.
2087 magic16
= be16_to_cpu(*(__be16
*)blk
);
2089 case XFS_DQUOT_MAGIC
:
2090 case XFS_DINODE_MAGIC
:
2091 goto recover_immediately
;
2096 /* unknown buffer contents, recover immediately */
2098 recover_immediately
:
2099 return (xfs_lsn_t
)-1;
2104 * Validate the recovered buffer is of the correct type and attach the
2105 * appropriate buffer operations to them for writeback. Magic numbers are in a
2107 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2108 * the first 32 bits of the buffer (most blocks),
2109 * inside a struct xfs_da_blkinfo at the start of the buffer.
2112 xlog_recover_validate_buf_type(
2113 struct xfs_mount
*mp
,
2115 xfs_buf_log_format_t
*buf_f
)
2117 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
2122 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
2123 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
2124 magicda
= be16_to_cpu(info
->magic
);
2125 switch (xfs_blft_from_flags(buf_f
)) {
2126 case XFS_BLFT_BTREE_BUF
:
2128 case XFS_ABTB_CRC_MAGIC
:
2129 case XFS_ABTC_CRC_MAGIC
:
2130 case XFS_ABTB_MAGIC
:
2131 case XFS_ABTC_MAGIC
:
2132 bp
->b_ops
= &xfs_allocbt_buf_ops
;
2134 case XFS_IBT_CRC_MAGIC
:
2136 bp
->b_ops
= &xfs_inobt_buf_ops
;
2138 case XFS_BMAP_CRC_MAGIC
:
2139 case XFS_BMAP_MAGIC
:
2140 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2143 xfs_warn(mp
, "Bad btree block magic!");
2148 case XFS_BLFT_AGF_BUF
:
2149 if (magic32
!= XFS_AGF_MAGIC
) {
2150 xfs_warn(mp
, "Bad AGF block magic!");
2154 bp
->b_ops
= &xfs_agf_buf_ops
;
2156 case XFS_BLFT_AGFL_BUF
:
2157 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2159 if (magic32
!= XFS_AGFL_MAGIC
) {
2160 xfs_warn(mp
, "Bad AGFL block magic!");
2164 bp
->b_ops
= &xfs_agfl_buf_ops
;
2166 case XFS_BLFT_AGI_BUF
:
2167 if (magic32
!= XFS_AGI_MAGIC
) {
2168 xfs_warn(mp
, "Bad AGI block magic!");
2172 bp
->b_ops
= &xfs_agi_buf_ops
;
2174 case XFS_BLFT_UDQUOT_BUF
:
2175 case XFS_BLFT_PDQUOT_BUF
:
2176 case XFS_BLFT_GDQUOT_BUF
:
2177 #ifdef CONFIG_XFS_QUOTA
2178 if (magic16
!= XFS_DQUOT_MAGIC
) {
2179 xfs_warn(mp
, "Bad DQUOT block magic!");
2183 bp
->b_ops
= &xfs_dquot_buf_ops
;
2186 "Trying to recover dquots without QUOTA support built in!");
2190 case XFS_BLFT_DINO_BUF
:
2192 * we get here with inode allocation buffers, not buffers that
2193 * track unlinked list changes.
2195 if (magic16
!= XFS_DINODE_MAGIC
) {
2196 xfs_warn(mp
, "Bad INODE block magic!");
2200 bp
->b_ops
= &xfs_inode_buf_ops
;
2202 case XFS_BLFT_SYMLINK_BUF
:
2203 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2204 xfs_warn(mp
, "Bad symlink block magic!");
2208 bp
->b_ops
= &xfs_symlink_buf_ops
;
2210 case XFS_BLFT_DIR_BLOCK_BUF
:
2211 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2212 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2213 xfs_warn(mp
, "Bad dir block magic!");
2217 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2219 case XFS_BLFT_DIR_DATA_BUF
:
2220 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2221 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2222 xfs_warn(mp
, "Bad dir data magic!");
2226 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2228 case XFS_BLFT_DIR_FREE_BUF
:
2229 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2230 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2231 xfs_warn(mp
, "Bad dir3 free magic!");
2235 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2237 case XFS_BLFT_DIR_LEAF1_BUF
:
2238 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2239 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2240 xfs_warn(mp
, "Bad dir leaf1 magic!");
2244 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2246 case XFS_BLFT_DIR_LEAFN_BUF
:
2247 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2248 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2249 xfs_warn(mp
, "Bad dir leafn magic!");
2253 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2255 case XFS_BLFT_DA_NODE_BUF
:
2256 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2257 magicda
!= XFS_DA3_NODE_MAGIC
) {
2258 xfs_warn(mp
, "Bad da node magic!");
2262 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2264 case XFS_BLFT_ATTR_LEAF_BUF
:
2265 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2266 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2267 xfs_warn(mp
, "Bad attr leaf magic!");
2271 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2273 case XFS_BLFT_ATTR_RMT_BUF
:
2274 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2276 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2277 xfs_warn(mp
, "Bad attr remote magic!");
2281 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2283 case XFS_BLFT_SB_BUF
:
2284 if (magic32
!= XFS_SB_MAGIC
) {
2285 xfs_warn(mp
, "Bad SB block magic!");
2289 bp
->b_ops
= &xfs_sb_buf_ops
;
2292 xfs_warn(mp
, "Unknown buffer type %d!",
2293 xfs_blft_from_flags(buf_f
));
2299 * Perform a 'normal' buffer recovery. Each logged region of the
2300 * buffer should be copied over the corresponding region in the
2301 * given buffer. The bitmap in the buf log format structure indicates
2302 * where to place the logged data.
2305 xlog_recover_do_reg_buffer(
2306 struct xfs_mount
*mp
,
2307 xlog_recover_item_t
*item
,
2309 xfs_buf_log_format_t
*buf_f
)
2316 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2319 i
= 1; /* 0 is the buf format structure */
2321 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2322 buf_f
->blf_map_size
, bit
);
2325 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2326 buf_f
->blf_map_size
, bit
);
2328 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2329 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2330 ASSERT(BBTOB(bp
->b_io_length
) >=
2331 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2334 * The dirty regions logged in the buffer, even though
2335 * contiguous, may span multiple chunks. This is because the
2336 * dirty region may span a physical page boundary in a buffer
2337 * and hence be split into two separate vectors for writing into
2338 * the log. Hence we need to trim nbits back to the length of
2339 * the current region being copied out of the log.
2341 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2342 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2345 * Do a sanity check if this is a dquot buffer. Just checking
2346 * the first dquot in the buffer should do. XXXThis is
2347 * probably a good thing to do for other buf types also.
2350 if (buf_f
->blf_flags
&
2351 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2352 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2354 "XFS: NULL dquot in %s.", __func__
);
2357 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2359 "XFS: dquot too small (%d) in %s.",
2360 item
->ri_buf
[i
].i_len
, __func__
);
2363 error
= xfs_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
2364 -1, 0, XFS_QMOPT_DOWARN
,
2365 "dquot_buf_recover");
2370 memcpy(xfs_buf_offset(bp
,
2371 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2372 item
->ri_buf
[i
].i_addr
, /* source */
2373 nbits
<<XFS_BLF_SHIFT
); /* length */
2379 /* Shouldn't be any more regions */
2380 ASSERT(i
== item
->ri_total
);
2383 * We can only do post recovery validation on items on CRC enabled
2384 * fielsystems as we need to know when the buffer was written to be able
2385 * to determine if we should have replayed the item. If we replay old
2386 * metadata over a newer buffer, then it will enter a temporarily
2387 * inconsistent state resulting in verification failures. Hence for now
2388 * just avoid the verification stage for non-crc filesystems
2390 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2391 xlog_recover_validate_buf_type(mp
, bp
, buf_f
);
2395 * Perform a dquot buffer recovery.
2396 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2397 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2398 * Else, treat it as a regular buffer and do recovery.
2401 xlog_recover_do_dquot_buffer(
2402 struct xfs_mount
*mp
,
2404 struct xlog_recover_item
*item
,
2406 struct xfs_buf_log_format
*buf_f
)
2410 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2413 * Filesystems are required to send in quota flags at mount time.
2415 if (mp
->m_qflags
== 0) {
2420 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2421 type
|= XFS_DQ_USER
;
2422 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2423 type
|= XFS_DQ_PROJ
;
2424 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2425 type
|= XFS_DQ_GROUP
;
2427 * This type of quotas was turned off, so ignore this buffer
2429 if (log
->l_quotaoffs_flag
& type
)
2432 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2436 * This routine replays a modification made to a buffer at runtime.
2437 * There are actually two types of buffer, regular and inode, which
2438 * are handled differently. Inode buffers are handled differently
2439 * in that we only recover a specific set of data from them, namely
2440 * the inode di_next_unlinked fields. This is because all other inode
2441 * data is actually logged via inode records and any data we replay
2442 * here which overlaps that may be stale.
2444 * When meta-data buffers are freed at run time we log a buffer item
2445 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2446 * of the buffer in the log should not be replayed at recovery time.
2447 * This is so that if the blocks covered by the buffer are reused for
2448 * file data before we crash we don't end up replaying old, freed
2449 * meta-data into a user's file.
2451 * To handle the cancellation of buffer log items, we make two passes
2452 * over the log during recovery. During the first we build a table of
2453 * those buffers which have been cancelled, and during the second we
2454 * only replay those buffers which do not have corresponding cancel
2455 * records in the table. See xlog_recover_buffer_pass[1,2] above
2456 * for more details on the implementation of the table of cancel records.
2459 xlog_recover_buffer_pass2(
2461 struct list_head
*buffer_list
,
2462 struct xlog_recover_item
*item
,
2463 xfs_lsn_t current_lsn
)
2465 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2466 xfs_mount_t
*mp
= log
->l_mp
;
2473 * In this pass we only want to recover all the buffers which have
2474 * not been cancelled and are not cancellation buffers themselves.
2476 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2477 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2478 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2482 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2485 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2486 buf_flags
|= XBF_UNMAPPED
;
2488 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2491 return XFS_ERROR(ENOMEM
);
2492 error
= bp
->b_error
;
2494 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2499 * recover the buffer only if we get an LSN from it and it's less than
2500 * the lsn of the transaction we are replaying.
2502 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2503 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0)
2506 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2507 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2508 } else if (buf_f
->blf_flags
&
2509 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2510 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2512 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2518 * Perform delayed write on the buffer. Asynchronous writes will be
2519 * slower when taking into account all the buffers to be flushed.
2521 * Also make sure that only inode buffers with good sizes stay in
2522 * the buffer cache. The kernel moves inodes in buffers of 1 block
2523 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2524 * buffers in the log can be a different size if the log was generated
2525 * by an older kernel using unclustered inode buffers or a newer kernel
2526 * running with a different inode cluster size. Regardless, if the
2527 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2528 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2529 * the buffer out of the buffer cache so that the buffer won't
2530 * overlap with future reads of those inodes.
2532 if (XFS_DINODE_MAGIC
==
2533 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2534 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2535 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2537 error
= xfs_bwrite(bp
);
2539 ASSERT(bp
->b_target
->bt_mount
== mp
);
2540 bp
->b_iodone
= xlog_recover_iodone
;
2541 xfs_buf_delwri_queue(bp
, buffer_list
);
2550 * Inode fork owner changes
2552 * If we have been told that we have to reparent the inode fork, it's because an
2553 * extent swap operation on a CRC enabled filesystem has been done and we are
2554 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2557 * The complexity here is that we don't have an inode context to work with, so
2558 * after we've replayed the inode we need to instantiate one. This is where the
2561 * We are in the middle of log recovery, so we can't run transactions. That
2562 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2563 * that will result in the corresponding iput() running the inode through
2564 * xfs_inactive(). If we've just replayed an inode core that changes the link
2565 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2566 * transactions (bad!).
2568 * So, to avoid this, we instantiate an inode directly from the inode core we've
2569 * just recovered. We have the buffer still locked, and all we really need to
2570 * instantiate is the inode core and the forks being modified. We can do this
2571 * manually, then run the inode btree owner change, and then tear down the
2572 * xfs_inode without having to run any transactions at all.
2574 * Also, because we don't have a transaction context available here but need to
2575 * gather all the buffers we modify for writeback so we pass the buffer_list
2576 * instead for the operation to use.
2580 xfs_recover_inode_owner_change(
2581 struct xfs_mount
*mp
,
2582 struct xfs_dinode
*dip
,
2583 struct xfs_inode_log_format
*in_f
,
2584 struct list_head
*buffer_list
)
2586 struct xfs_inode
*ip
;
2589 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2591 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2595 /* instantiate the inode */
2596 xfs_dinode_from_disk(&ip
->i_d
, dip
);
2597 ASSERT(ip
->i_d
.di_version
>= 3);
2599 error
= xfs_iformat_fork(ip
, dip
);
2604 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2605 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2606 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2607 ip
->i_ino
, buffer_list
);
2612 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2613 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2614 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2615 ip
->i_ino
, buffer_list
);
2626 xlog_recover_inode_pass2(
2628 struct list_head
*buffer_list
,
2629 struct xlog_recover_item
*item
,
2630 xfs_lsn_t current_lsn
)
2632 xfs_inode_log_format_t
*in_f
;
2633 xfs_mount_t
*mp
= log
->l_mp
;
2642 xfs_icdinode_t
*dicp
;
2646 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2647 in_f
= item
->ri_buf
[0].i_addr
;
2649 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2651 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2657 * Inode buffers can be freed, look out for it,
2658 * and do not replay the inode.
2660 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2661 in_f
->ilf_len
, 0)) {
2663 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2666 trace_xfs_log_recover_inode_recover(log
, in_f
);
2668 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
2669 &xfs_inode_buf_ops
);
2674 error
= bp
->b_error
;
2676 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
2679 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2680 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2683 * Make sure the place we're flushing out to really looks
2686 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2688 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2689 __func__
, dip
, bp
, in_f
->ilf_ino
);
2690 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2691 XFS_ERRLEVEL_LOW
, mp
);
2692 error
= EFSCORRUPTED
;
2695 dicp
= item
->ri_buf
[1].i_addr
;
2696 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2698 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2699 __func__
, item
, in_f
->ilf_ino
);
2700 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2701 XFS_ERRLEVEL_LOW
, mp
);
2702 error
= EFSCORRUPTED
;
2707 * If the inode has an LSN in it, recover the inode only if it's less
2708 * than the lsn of the transaction we are replaying. Note: we still
2709 * need to replay an owner change even though the inode is more recent
2710 * than the transaction as there is no guarantee that all the btree
2711 * blocks are more recent than this transaction, too.
2713 if (dip
->di_version
>= 3) {
2714 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
2716 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2717 trace_xfs_log_recover_inode_skip(log
, in_f
);
2719 goto out_owner_change
;
2724 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2725 * are transactional and if ordering is necessary we can determine that
2726 * more accurately by the LSN field in the V3 inode core. Don't trust
2727 * the inode versions we might be changing them here - use the
2728 * superblock flag to determine whether we need to look at di_flushiter
2729 * to skip replay when the on disk inode is newer than the log one
2731 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
2732 dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2734 * Deal with the wrap case, DI_MAX_FLUSH is less
2735 * than smaller numbers
2737 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2738 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2741 trace_xfs_log_recover_inode_skip(log
, in_f
);
2747 /* Take the opportunity to reset the flush iteration count */
2748 dicp
->di_flushiter
= 0;
2750 if (unlikely(S_ISREG(dicp
->di_mode
))) {
2751 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2752 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2753 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2754 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2756 "%s: Bad regular inode log record, rec ptr 0x%p, "
2757 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2758 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2759 error
= EFSCORRUPTED
;
2762 } else if (unlikely(S_ISDIR(dicp
->di_mode
))) {
2763 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2764 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2765 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2766 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2767 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2769 "%s: Bad dir inode log record, rec ptr 0x%p, "
2770 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2771 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2772 error
= EFSCORRUPTED
;
2776 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2777 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2778 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2780 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2781 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2782 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2783 dicp
->di_nextents
+ dicp
->di_anextents
,
2785 error
= EFSCORRUPTED
;
2788 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2789 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2790 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2792 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2793 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2794 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2795 error
= EFSCORRUPTED
;
2798 isize
= xfs_icdinode_size(dicp
->di_version
);
2799 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
2800 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2801 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2803 "%s: Bad inode log record length %d, rec ptr 0x%p",
2804 __func__
, item
->ri_buf
[1].i_len
, item
);
2805 error
= EFSCORRUPTED
;
2809 /* The core is in in-core format */
2810 xfs_dinode_to_disk(dip
, dicp
);
2812 /* the rest is in on-disk format */
2813 if (item
->ri_buf
[1].i_len
> isize
) {
2814 memcpy((char *)dip
+ isize
,
2815 item
->ri_buf
[1].i_addr
+ isize
,
2816 item
->ri_buf
[1].i_len
- isize
);
2819 fields
= in_f
->ilf_fields
;
2820 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2822 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2825 memcpy(XFS_DFORK_DPTR(dip
),
2826 &in_f
->ilf_u
.ilfu_uuid
,
2831 if (in_f
->ilf_size
== 2)
2832 goto out_owner_change
;
2833 len
= item
->ri_buf
[2].i_len
;
2834 src
= item
->ri_buf
[2].i_addr
;
2835 ASSERT(in_f
->ilf_size
<= 4);
2836 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2837 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2838 (len
== in_f
->ilf_dsize
));
2840 switch (fields
& XFS_ILOG_DFORK
) {
2841 case XFS_ILOG_DDATA
:
2843 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2846 case XFS_ILOG_DBROOT
:
2847 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2848 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2849 XFS_DFORK_DSIZE(dip
, mp
));
2854 * There are no data fork flags set.
2856 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2861 * If we logged any attribute data, recover it. There may or
2862 * may not have been any other non-core data logged in this
2865 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2866 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2871 len
= item
->ri_buf
[attr_index
].i_len
;
2872 src
= item
->ri_buf
[attr_index
].i_addr
;
2873 ASSERT(len
== in_f
->ilf_asize
);
2875 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2876 case XFS_ILOG_ADATA
:
2878 dest
= XFS_DFORK_APTR(dip
);
2879 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2880 memcpy(dest
, src
, len
);
2883 case XFS_ILOG_ABROOT
:
2884 dest
= XFS_DFORK_APTR(dip
);
2885 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2886 len
, (xfs_bmdr_block_t
*)dest
,
2887 XFS_DFORK_ASIZE(dip
, mp
));
2891 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2899 if (in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
))
2900 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
2902 /* re-generate the checksum. */
2903 xfs_dinode_calc_crc(log
->l_mp
, dip
);
2905 ASSERT(bp
->b_target
->bt_mount
== mp
);
2906 bp
->b_iodone
= xlog_recover_iodone
;
2907 xfs_buf_delwri_queue(bp
, buffer_list
);
2914 return XFS_ERROR(error
);
2918 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2919 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2923 xlog_recover_quotaoff_pass1(
2925 struct xlog_recover_item
*item
)
2927 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2931 * The logitem format's flag tells us if this was user quotaoff,
2932 * group/project quotaoff or both.
2934 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2935 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2936 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2937 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2938 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2939 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2945 * Recover a dquot record
2948 xlog_recover_dquot_pass2(
2950 struct list_head
*buffer_list
,
2951 struct xlog_recover_item
*item
,
2952 xfs_lsn_t current_lsn
)
2954 xfs_mount_t
*mp
= log
->l_mp
;
2956 struct xfs_disk_dquot
*ddq
, *recddq
;
2958 xfs_dq_logformat_t
*dq_f
;
2963 * Filesystems are required to send in quota flags at mount time.
2965 if (mp
->m_qflags
== 0)
2968 recddq
= item
->ri_buf
[1].i_addr
;
2969 if (recddq
== NULL
) {
2970 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2971 return XFS_ERROR(EIO
);
2973 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2974 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2975 item
->ri_buf
[1].i_len
, __func__
);
2976 return XFS_ERROR(EIO
);
2980 * This type of quotas was turned off, so ignore this record.
2982 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2984 if (log
->l_quotaoffs_flag
& type
)
2988 * At this point we know that quota was _not_ turned off.
2989 * Since the mount flags are not indicating to us otherwise, this
2990 * must mean that quota is on, and the dquot needs to be replayed.
2991 * Remember that we may not have fully recovered the superblock yet,
2992 * so we can't do the usual trick of looking at the SB quota bits.
2994 * The other possibility, of course, is that the quota subsystem was
2995 * removed since the last mount - ENOSYS.
2997 dq_f
= item
->ri_buf
[0].i_addr
;
2999 error
= xfs_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3000 "xlog_recover_dquot_pass2 (log copy)");
3002 return XFS_ERROR(EIO
);
3003 ASSERT(dq_f
->qlf_len
== 1);
3005 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3006 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
3012 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
3015 * At least the magic num portion should be on disk because this
3016 * was among a chunk of dquots created earlier, and we did some
3017 * minimal initialization then.
3019 error
= xfs_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
3020 "xlog_recover_dquot_pass2");
3023 return XFS_ERROR(EIO
);
3027 * If the dquot has an LSN in it, recover the dquot only if it's less
3028 * than the lsn of the transaction we are replaying.
3030 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3031 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
3032 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
3034 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3039 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
3040 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3041 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
3045 ASSERT(dq_f
->qlf_size
== 2);
3046 ASSERT(bp
->b_target
->bt_mount
== mp
);
3047 bp
->b_iodone
= xlog_recover_iodone
;
3048 xfs_buf_delwri_queue(bp
, buffer_list
);
3056 * This routine is called to create an in-core extent free intent
3057 * item from the efi format structure which was logged on disk.
3058 * It allocates an in-core efi, copies the extents from the format
3059 * structure into it, and adds the efi to the AIL with the given
3063 xlog_recover_efi_pass2(
3065 struct xlog_recover_item
*item
,
3069 xfs_mount_t
*mp
= log
->l_mp
;
3070 xfs_efi_log_item_t
*efip
;
3071 xfs_efi_log_format_t
*efi_formatp
;
3073 efi_formatp
= item
->ri_buf
[0].i_addr
;
3075 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
3076 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
3077 &(efip
->efi_format
)))) {
3078 xfs_efi_item_free(efip
);
3081 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
3083 spin_lock(&log
->l_ailp
->xa_lock
);
3085 * xfs_trans_ail_update() drops the AIL lock.
3087 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
3093 * This routine is called when an efd format structure is found in
3094 * a committed transaction in the log. It's purpose is to cancel
3095 * the corresponding efi if it was still in the log. To do this
3096 * it searches the AIL for the efi with an id equal to that in the
3097 * efd format structure. If we find it, we remove the efi from the
3101 xlog_recover_efd_pass2(
3103 struct xlog_recover_item
*item
)
3105 xfs_efd_log_format_t
*efd_formatp
;
3106 xfs_efi_log_item_t
*efip
= NULL
;
3107 xfs_log_item_t
*lip
;
3109 struct xfs_ail_cursor cur
;
3110 struct xfs_ail
*ailp
= log
->l_ailp
;
3112 efd_formatp
= item
->ri_buf
[0].i_addr
;
3113 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
3114 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
3115 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
3116 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
3117 efi_id
= efd_formatp
->efd_efi_id
;
3120 * Search for the efi with the id in the efd format structure
3123 spin_lock(&ailp
->xa_lock
);
3124 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3125 while (lip
!= NULL
) {
3126 if (lip
->li_type
== XFS_LI_EFI
) {
3127 efip
= (xfs_efi_log_item_t
*)lip
;
3128 if (efip
->efi_format
.efi_id
== efi_id
) {
3130 * xfs_trans_ail_delete() drops the
3133 xfs_trans_ail_delete(ailp
, lip
,
3134 SHUTDOWN_CORRUPT_INCORE
);
3135 xfs_efi_item_free(efip
);
3136 spin_lock(&ailp
->xa_lock
);
3140 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3142 xfs_trans_ail_cursor_done(ailp
, &cur
);
3143 spin_unlock(&ailp
->xa_lock
);
3149 * This routine is called when an inode create format structure is found in a
3150 * committed transaction in the log. It's purpose is to initialise the inodes
3151 * being allocated on disk. This requires us to get inode cluster buffers that
3152 * match the range to be intialised, stamped with inode templates and written
3153 * by delayed write so that subsequent modifications will hit the cached buffer
3154 * and only need writing out at the end of recovery.
3157 xlog_recover_do_icreate_pass2(
3159 struct list_head
*buffer_list
,
3160 xlog_recover_item_t
*item
)
3162 struct xfs_mount
*mp
= log
->l_mp
;
3163 struct xfs_icreate_log
*icl
;
3164 xfs_agnumber_t agno
;
3165 xfs_agblock_t agbno
;
3168 xfs_agblock_t length
;
3170 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3171 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3172 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3176 if (icl
->icl_size
!= 1) {
3177 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3181 agno
= be32_to_cpu(icl
->icl_ag
);
3182 if (agno
>= mp
->m_sb
.sb_agcount
) {
3183 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3186 agbno
= be32_to_cpu(icl
->icl_agbno
);
3187 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3188 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3191 isize
= be32_to_cpu(icl
->icl_isize
);
3192 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3193 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3196 count
= be32_to_cpu(icl
->icl_count
);
3198 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3201 length
= be32_to_cpu(icl
->icl_length
);
3202 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3203 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3207 /* existing allocation is fixed value */
3208 ASSERT(count
== XFS_IALLOC_INODES(mp
));
3209 ASSERT(length
== XFS_IALLOC_BLOCKS(mp
));
3210 if (count
!= XFS_IALLOC_INODES(mp
) ||
3211 length
!= XFS_IALLOC_BLOCKS(mp
)) {
3212 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count 2");
3217 * Inode buffers can be freed. Do not replay the inode initialisation as
3218 * we could be overwriting something written after this inode buffer was
3221 * XXX: we need to iterate all buffers and only init those that are not
3222 * cancelled. I think that a more fine grained factoring of
3223 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3226 if (xlog_check_buffer_cancelled(log
,
3227 XFS_AGB_TO_DADDR(mp
, agno
, agbno
), length
, 0))
3230 xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, agno
, agbno
, length
,
3231 be32_to_cpu(icl
->icl_gen
));
3236 * Free up any resources allocated by the transaction
3238 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3241 xlog_recover_free_trans(
3242 struct xlog_recover
*trans
)
3244 xlog_recover_item_t
*item
, *n
;
3247 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
3248 /* Free the regions in the item. */
3249 list_del(&item
->ri_list
);
3250 for (i
= 0; i
< item
->ri_cnt
; i
++)
3251 kmem_free(item
->ri_buf
[i
].i_addr
);
3252 /* Free the item itself */
3253 kmem_free(item
->ri_buf
);
3256 /* Free the transaction recover structure */
3261 xlog_recover_buffer_ra_pass2(
3263 struct xlog_recover_item
*item
)
3265 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3266 struct xfs_mount
*mp
= log
->l_mp
;
3268 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3269 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3273 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3274 buf_f
->blf_len
, NULL
);
3278 xlog_recover_inode_ra_pass2(
3280 struct xlog_recover_item
*item
)
3282 struct xfs_inode_log_format ilf_buf
;
3283 struct xfs_inode_log_format
*ilfp
;
3284 struct xfs_mount
*mp
= log
->l_mp
;
3287 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3288 ilfp
= item
->ri_buf
[0].i_addr
;
3291 memset(ilfp
, 0, sizeof(*ilfp
));
3292 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3297 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3300 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3301 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3305 xlog_recover_dquot_ra_pass2(
3307 struct xlog_recover_item
*item
)
3309 struct xfs_mount
*mp
= log
->l_mp
;
3310 struct xfs_disk_dquot
*recddq
;
3311 struct xfs_dq_logformat
*dq_f
;
3315 if (mp
->m_qflags
== 0)
3318 recddq
= item
->ri_buf
[1].i_addr
;
3321 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
3324 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3326 if (log
->l_quotaoffs_flag
& type
)
3329 dq_f
= item
->ri_buf
[0].i_addr
;
3331 ASSERT(dq_f
->qlf_len
== 1);
3333 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3334 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), NULL
);
3338 xlog_recover_ra_pass2(
3340 struct xlog_recover_item
*item
)
3342 switch (ITEM_TYPE(item
)) {
3344 xlog_recover_buffer_ra_pass2(log
, item
);
3347 xlog_recover_inode_ra_pass2(log
, item
);
3350 xlog_recover_dquot_ra_pass2(log
, item
);
3354 case XFS_LI_QUOTAOFF
:
3361 xlog_recover_commit_pass1(
3363 struct xlog_recover
*trans
,
3364 struct xlog_recover_item
*item
)
3366 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
3368 switch (ITEM_TYPE(item
)) {
3370 return xlog_recover_buffer_pass1(log
, item
);
3371 case XFS_LI_QUOTAOFF
:
3372 return xlog_recover_quotaoff_pass1(log
, item
);
3377 case XFS_LI_ICREATE
:
3378 /* nothing to do in pass 1 */
3381 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3382 __func__
, ITEM_TYPE(item
));
3384 return XFS_ERROR(EIO
);
3389 xlog_recover_commit_pass2(
3391 struct xlog_recover
*trans
,
3392 struct list_head
*buffer_list
,
3393 struct xlog_recover_item
*item
)
3395 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
3397 switch (ITEM_TYPE(item
)) {
3399 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
3402 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
3405 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
3407 return xlog_recover_efd_pass2(log
, item
);
3409 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
3411 case XFS_LI_ICREATE
:
3412 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
3413 case XFS_LI_QUOTAOFF
:
3414 /* nothing to do in pass2 */
3417 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
3418 __func__
, ITEM_TYPE(item
));
3420 return XFS_ERROR(EIO
);
3425 xlog_recover_items_pass2(
3427 struct xlog_recover
*trans
,
3428 struct list_head
*buffer_list
,
3429 struct list_head
*item_list
)
3431 struct xlog_recover_item
*item
;
3434 list_for_each_entry(item
, item_list
, ri_list
) {
3435 error
= xlog_recover_commit_pass2(log
, trans
,
3445 * Perform the transaction.
3447 * If the transaction modifies a buffer or inode, do it now. Otherwise,
3448 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3451 xlog_recover_commit_trans(
3453 struct xlog_recover
*trans
,
3458 int items_queued
= 0;
3459 struct xlog_recover_item
*item
;
3460 struct xlog_recover_item
*next
;
3461 LIST_HEAD (buffer_list
);
3462 LIST_HEAD (ra_list
);
3463 LIST_HEAD (done_list
);
3465 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3467 hlist_del(&trans
->r_list
);
3469 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
3473 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
3475 case XLOG_RECOVER_PASS1
:
3476 error
= xlog_recover_commit_pass1(log
, trans
, item
);
3478 case XLOG_RECOVER_PASS2
:
3479 xlog_recover_ra_pass2(log
, item
);
3480 list_move_tail(&item
->ri_list
, &ra_list
);
3482 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
3483 error
= xlog_recover_items_pass2(log
, trans
,
3484 &buffer_list
, &ra_list
);
3485 list_splice_tail_init(&ra_list
, &done_list
);
3499 if (!list_empty(&ra_list
)) {
3501 error
= xlog_recover_items_pass2(log
, trans
,
3502 &buffer_list
, &ra_list
);
3503 list_splice_tail_init(&ra_list
, &done_list
);
3506 if (!list_empty(&done_list
))
3507 list_splice_init(&done_list
, &trans
->r_itemq
);
3509 xlog_recover_free_trans(trans
);
3511 error2
= xfs_buf_delwri_submit(&buffer_list
);
3512 return error
? error
: error2
;
3516 xlog_recover_unmount_trans(
3518 struct xlog_recover
*trans
)
3520 /* Do nothing now */
3521 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
3526 * There are two valid states of the r_state field. 0 indicates that the
3527 * transaction structure is in a normal state. We have either seen the
3528 * start of the transaction or the last operation we added was not a partial
3529 * operation. If the last operation we added to the transaction was a
3530 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3532 * NOTE: skip LRs with 0 data length.
3535 xlog_recover_process_data(
3537 struct hlist_head rhash
[],
3538 struct xlog_rec_header
*rhead
,
3544 xlog_op_header_t
*ohead
;
3545 xlog_recover_t
*trans
;
3551 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
3552 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
3554 /* check the log format matches our own - else we can't recover */
3555 if (xlog_header_check_recover(log
->l_mp
, rhead
))
3556 return (XFS_ERROR(EIO
));
3558 while ((dp
< lp
) && num_logops
) {
3559 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
3560 ohead
= (xlog_op_header_t
*)dp
;
3561 dp
+= sizeof(xlog_op_header_t
);
3562 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
3563 ohead
->oh_clientid
!= XFS_LOG
) {
3564 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
3565 __func__
, ohead
->oh_clientid
);
3567 return (XFS_ERROR(EIO
));
3569 tid
= be32_to_cpu(ohead
->oh_tid
);
3570 hash
= XLOG_RHASH(tid
);
3571 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
3572 if (trans
== NULL
) { /* not found; add new tid */
3573 if (ohead
->oh_flags
& XLOG_START_TRANS
)
3574 xlog_recover_new_tid(&rhash
[hash
], tid
,
3575 be64_to_cpu(rhead
->h_lsn
));
3577 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
3578 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
3579 __func__
, be32_to_cpu(ohead
->oh_len
));
3581 return (XFS_ERROR(EIO
));
3583 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
3584 if (flags
& XLOG_WAS_CONT_TRANS
)
3585 flags
&= ~XLOG_CONTINUE_TRANS
;
3587 case XLOG_COMMIT_TRANS
:
3588 error
= xlog_recover_commit_trans(log
,
3591 case XLOG_UNMOUNT_TRANS
:
3592 error
= xlog_recover_unmount_trans(log
, trans
);
3594 case XLOG_WAS_CONT_TRANS
:
3595 error
= xlog_recover_add_to_cont_trans(log
,
3597 be32_to_cpu(ohead
->oh_len
));
3599 case XLOG_START_TRANS
:
3600 xfs_warn(log
->l_mp
, "%s: bad transaction",
3603 error
= XFS_ERROR(EIO
);
3606 case XLOG_CONTINUE_TRANS
:
3607 error
= xlog_recover_add_to_trans(log
, trans
,
3608 dp
, be32_to_cpu(ohead
->oh_len
));
3611 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
3614 error
= XFS_ERROR(EIO
);
3620 dp
+= be32_to_cpu(ohead
->oh_len
);
3627 * Process an extent free intent item that was recovered from
3628 * the log. We need to free the extents that it describes.
3631 xlog_recover_process_efi(
3633 xfs_efi_log_item_t
*efip
)
3635 xfs_efd_log_item_t
*efdp
;
3640 xfs_fsblock_t startblock_fsb
;
3642 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
3645 * First check the validity of the extents described by the
3646 * EFI. If any are bad, then assume that all are bad and
3647 * just toss the EFI.
3649 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3650 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3651 startblock_fsb
= XFS_BB_TO_FSB(mp
,
3652 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
3653 if ((startblock_fsb
== 0) ||
3654 (extp
->ext_len
== 0) ||
3655 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
3656 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
3658 * This will pull the EFI from the AIL and
3659 * free the memory associated with it.
3661 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3662 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
3663 return XFS_ERROR(EIO
);
3667 tp
= xfs_trans_alloc(mp
, 0);
3668 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_itruncate
, 0, 0);
3671 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
3673 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
3674 extp
= &(efip
->efi_format
.efi_extents
[i
]);
3675 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
3678 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
3682 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
3683 error
= xfs_trans_commit(tp
, 0);
3687 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3692 * When this is called, all of the EFIs which did not have
3693 * corresponding EFDs should be in the AIL. What we do now
3694 * is free the extents associated with each one.
3696 * Since we process the EFIs in normal transactions, they
3697 * will be removed at some point after the commit. This prevents
3698 * us from just walking down the list processing each one.
3699 * We'll use a flag in the EFI to skip those that we've already
3700 * processed and use the AIL iteration mechanism's generation
3701 * count to try to speed this up at least a bit.
3703 * When we start, we know that the EFIs are the only things in
3704 * the AIL. As we process them, however, other items are added
3705 * to the AIL. Since everything added to the AIL must come after
3706 * everything already in the AIL, we stop processing as soon as
3707 * we see something other than an EFI in the AIL.
3710 xlog_recover_process_efis(
3713 xfs_log_item_t
*lip
;
3714 xfs_efi_log_item_t
*efip
;
3716 struct xfs_ail_cursor cur
;
3717 struct xfs_ail
*ailp
;
3720 spin_lock(&ailp
->xa_lock
);
3721 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3722 while (lip
!= NULL
) {
3724 * We're done when we see something other than an EFI.
3725 * There should be no EFIs left in the AIL now.
3727 if (lip
->li_type
!= XFS_LI_EFI
) {
3729 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
3730 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
3736 * Skip EFIs that we've already processed.
3738 efip
= (xfs_efi_log_item_t
*)lip
;
3739 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
3740 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3744 spin_unlock(&ailp
->xa_lock
);
3745 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3746 spin_lock(&ailp
->xa_lock
);
3749 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3752 xfs_trans_ail_cursor_done(ailp
, &cur
);
3753 spin_unlock(&ailp
->xa_lock
);
3758 * This routine performs a transaction to null out a bad inode pointer
3759 * in an agi unlinked inode hash bucket.
3762 xlog_recover_clear_agi_bucket(
3764 xfs_agnumber_t agno
,
3773 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3774 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_clearagi
, 0, 0);
3778 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3782 agi
= XFS_BUF_TO_AGI(agibp
);
3783 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3784 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3785 (sizeof(xfs_agino_t
) * bucket
);
3786 xfs_trans_log_buf(tp
, agibp
, offset
,
3787 (offset
+ sizeof(xfs_agino_t
) - 1));
3789 error
= xfs_trans_commit(tp
, 0);
3795 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3797 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3802 xlog_recover_process_one_iunlink(
3803 struct xfs_mount
*mp
,
3804 xfs_agnumber_t agno
,
3808 struct xfs_buf
*ibp
;
3809 struct xfs_dinode
*dip
;
3810 struct xfs_inode
*ip
;
3814 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3815 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3820 * Get the on disk inode to find the next inode in the bucket.
3822 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
3826 ASSERT(ip
->i_d
.di_nlink
== 0);
3827 ASSERT(ip
->i_d
.di_mode
!= 0);
3829 /* setup for the next pass */
3830 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3834 * Prevent any DMAPI event from being sent when the reference on
3835 * the inode is dropped.
3837 ip
->i_d
.di_dmevmask
= 0;
3846 * We can't read in the inode this bucket points to, or this inode
3847 * is messed up. Just ditch this bucket of inodes. We will lose
3848 * some inodes and space, but at least we won't hang.
3850 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3851 * clear the inode pointer in the bucket.
3853 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3858 * xlog_iunlink_recover
3860 * This is called during recovery to process any inodes which
3861 * we unlinked but not freed when the system crashed. These
3862 * inodes will be on the lists in the AGI blocks. What we do
3863 * here is scan all the AGIs and fully truncate and free any
3864 * inodes found on the lists. Each inode is removed from the
3865 * lists when it has been fully truncated and is freed. The
3866 * freeing of the inode and its removal from the list must be
3870 xlog_recover_process_iunlinks(
3874 xfs_agnumber_t agno
;
3885 * Prevent any DMAPI event from being sent while in this function.
3887 mp_dmevmask
= mp
->m_dmevmask
;
3890 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3892 * Find the agi for this ag.
3894 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3897 * AGI is b0rked. Don't process it.
3899 * We should probably mark the filesystem as corrupt
3900 * after we've recovered all the ag's we can....
3905 * Unlock the buffer so that it can be acquired in the normal
3906 * course of the transaction to truncate and free each inode.
3907 * Because we are not racing with anyone else here for the AGI
3908 * buffer, we don't even need to hold it locked to read the
3909 * initial unlinked bucket entries out of the buffer. We keep
3910 * buffer reference though, so that it stays pinned in memory
3911 * while we need the buffer.
3913 agi
= XFS_BUF_TO_AGI(agibp
);
3914 xfs_buf_unlock(agibp
);
3916 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3917 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3918 while (agino
!= NULLAGINO
) {
3919 agino
= xlog_recover_process_one_iunlink(mp
,
3920 agno
, agino
, bucket
);
3923 xfs_buf_rele(agibp
);
3926 mp
->m_dmevmask
= mp_dmevmask
;
3930 * Upack the log buffer data and crc check it. If the check fails, issue a
3931 * warning if and only if the CRC in the header is non-zero. This makes the
3932 * check an advisory warning, and the zero CRC check will prevent failure
3933 * warnings from being emitted when upgrading the kernel from one that does not
3934 * add CRCs by default.
3936 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3937 * corruption failure
3940 xlog_unpack_data_crc(
3941 struct xlog_rec_header
*rhead
,
3947 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
3948 if (crc
!= rhead
->h_crc
) {
3949 if (rhead
->h_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
3950 xfs_alert(log
->l_mp
,
3951 "log record CRC mismatch: found 0x%x, expected 0x%x.",
3952 le32_to_cpu(rhead
->h_crc
),
3954 xfs_hex_dump(dp
, 32);
3958 * If we've detected a log record corruption, then we can't
3959 * recover past this point. Abort recovery if we are enforcing
3960 * CRC protection by punting an error back up the stack.
3962 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
3963 return EFSCORRUPTED
;
3971 struct xlog_rec_header
*rhead
,
3978 error
= xlog_unpack_data_crc(rhead
, dp
, log
);
3982 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3983 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3984 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3988 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3989 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3990 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3991 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3992 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3993 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
4002 xlog_valid_rec_header(
4004 struct xlog_rec_header
*rhead
,
4009 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
4010 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4011 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4012 return XFS_ERROR(EFSCORRUPTED
);
4015 (!rhead
->h_version
||
4016 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
4017 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
4018 __func__
, be32_to_cpu(rhead
->h_version
));
4019 return XFS_ERROR(EIO
);
4022 /* LR body must have data or it wouldn't have been written */
4023 hlen
= be32_to_cpu(rhead
->h_len
);
4024 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
4025 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4026 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4027 return XFS_ERROR(EFSCORRUPTED
);
4029 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
4030 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4031 XFS_ERRLEVEL_LOW
, log
->l_mp
);
4032 return XFS_ERROR(EFSCORRUPTED
);
4038 * Read the log from tail to head and process the log records found.
4039 * Handle the two cases where the tail and head are in the same cycle
4040 * and where the active portion of the log wraps around the end of
4041 * the physical log separately. The pass parameter is passed through
4042 * to the routines called to process the data and is not looked at
4046 xlog_do_recovery_pass(
4048 xfs_daddr_t head_blk
,
4049 xfs_daddr_t tail_blk
,
4052 xlog_rec_header_t
*rhead
;
4055 xfs_buf_t
*hbp
, *dbp
;
4056 int error
= 0, h_size
;
4057 int bblks
, split_bblks
;
4058 int hblks
, split_hblks
, wrapped_hblks
;
4059 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
4061 ASSERT(head_blk
!= tail_blk
);
4064 * Read the header of the tail block and get the iclog buffer size from
4065 * h_size. Use this to tell how many sectors make up the log header.
4067 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
4069 * When using variable length iclogs, read first sector of
4070 * iclog header and extract the header size from it. Get a
4071 * new hbp that is the correct size.
4073 hbp
= xlog_get_bp(log
, 1);
4077 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
4081 rhead
= (xlog_rec_header_t
*)offset
;
4082 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
4085 h_size
= be32_to_cpu(rhead
->h_size
);
4086 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
4087 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
4088 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
4089 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
4092 hbp
= xlog_get_bp(log
, hblks
);
4097 ASSERT(log
->l_sectBBsize
== 1);
4099 hbp
= xlog_get_bp(log
, 1);
4100 h_size
= XLOG_BIG_RECORD_BSIZE
;
4105 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
4111 memset(rhash
, 0, sizeof(rhash
));
4112 if (tail_blk
<= head_blk
) {
4113 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
4114 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4118 rhead
= (xlog_rec_header_t
*)offset
;
4119 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4123 /* blocks in data section */
4124 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4125 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
4130 error
= xlog_unpack_data(rhead
, offset
, log
);
4134 error
= xlog_recover_process_data(log
,
4135 rhash
, rhead
, offset
, pass
);
4138 blk_no
+= bblks
+ hblks
;
4142 * Perform recovery around the end of the physical log.
4143 * When the head is not on the same cycle number as the tail,
4144 * we can't do a sequential recovery as above.
4147 while (blk_no
< log
->l_logBBsize
) {
4149 * Check for header wrapping around physical end-of-log
4151 offset
= hbp
->b_addr
;
4154 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
4155 /* Read header in one read */
4156 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
4161 /* This LR is split across physical log end */
4162 if (blk_no
!= log
->l_logBBsize
) {
4163 /* some data before physical log end */
4164 ASSERT(blk_no
<= INT_MAX
);
4165 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
4166 ASSERT(split_hblks
> 0);
4167 error
= xlog_bread(log
, blk_no
,
4175 * Note: this black magic still works with
4176 * large sector sizes (non-512) only because:
4177 * - we increased the buffer size originally
4178 * by 1 sector giving us enough extra space
4179 * for the second read;
4180 * - the log start is guaranteed to be sector
4182 * - we read the log end (LR header start)
4183 * _first_, then the log start (LR header end)
4184 * - order is important.
4186 wrapped_hblks
= hblks
- split_hblks
;
4187 error
= xlog_bread_offset(log
, 0,
4189 offset
+ BBTOB(split_hblks
));
4193 rhead
= (xlog_rec_header_t
*)offset
;
4194 error
= xlog_valid_rec_header(log
, rhead
,
4195 split_hblks
? blk_no
: 0);
4199 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4202 /* Read in data for log record */
4203 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
4204 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
4209 /* This log record is split across the
4210 * physical end of log */
4211 offset
= dbp
->b_addr
;
4213 if (blk_no
!= log
->l_logBBsize
) {
4214 /* some data is before the physical
4216 ASSERT(!wrapped_hblks
);
4217 ASSERT(blk_no
<= INT_MAX
);
4219 log
->l_logBBsize
- (int)blk_no
;
4220 ASSERT(split_bblks
> 0);
4221 error
= xlog_bread(log
, blk_no
,
4229 * Note: this black magic still works with
4230 * large sector sizes (non-512) only because:
4231 * - we increased the buffer size originally
4232 * by 1 sector giving us enough extra space
4233 * for the second read;
4234 * - the log start is guaranteed to be sector
4236 * - we read the log end (LR header start)
4237 * _first_, then the log start (LR header end)
4238 * - order is important.
4240 error
= xlog_bread_offset(log
, 0,
4241 bblks
- split_bblks
, dbp
,
4242 offset
+ BBTOB(split_bblks
));
4247 error
= xlog_unpack_data(rhead
, offset
, log
);
4251 error
= xlog_recover_process_data(log
, rhash
,
4252 rhead
, offset
, pass
);
4258 ASSERT(blk_no
>= log
->l_logBBsize
);
4259 blk_no
-= log
->l_logBBsize
;
4261 /* read first part of physical log */
4262 while (blk_no
< head_blk
) {
4263 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
4267 rhead
= (xlog_rec_header_t
*)offset
;
4268 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
4272 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
4273 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
4278 error
= xlog_unpack_data(rhead
, offset
, log
);
4282 error
= xlog_recover_process_data(log
, rhash
,
4283 rhead
, offset
, pass
);
4286 blk_no
+= bblks
+ hblks
;
4298 * Do the recovery of the log. We actually do this in two phases.
4299 * The two passes are necessary in order to implement the function
4300 * of cancelling a record written into the log. The first pass
4301 * determines those things which have been cancelled, and the
4302 * second pass replays log items normally except for those which
4303 * have been cancelled. The handling of the replay and cancellations
4304 * takes place in the log item type specific routines.
4306 * The table of items which have cancel records in the log is allocated
4307 * and freed at this level, since only here do we know when all of
4308 * the log recovery has been completed.
4311 xlog_do_log_recovery(
4313 xfs_daddr_t head_blk
,
4314 xfs_daddr_t tail_blk
)
4318 ASSERT(head_blk
!= tail_blk
);
4321 * First do a pass to find all of the cancelled buf log items.
4322 * Store them in the buf_cancel_table for use in the second pass.
4324 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
4325 sizeof(struct list_head
),
4327 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4328 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
4330 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4331 XLOG_RECOVER_PASS1
);
4333 kmem_free(log
->l_buf_cancel_table
);
4334 log
->l_buf_cancel_table
= NULL
;
4338 * Then do a second pass to actually recover the items in the log.
4339 * When it is complete free the table of buf cancel items.
4341 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
4342 XLOG_RECOVER_PASS2
);
4347 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
4348 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
4352 kmem_free(log
->l_buf_cancel_table
);
4353 log
->l_buf_cancel_table
= NULL
;
4359 * Do the actual recovery
4364 xfs_daddr_t head_blk
,
4365 xfs_daddr_t tail_blk
)
4372 * First replay the images in the log.
4374 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
4379 * If IO errors happened during recovery, bail out.
4381 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
4386 * We now update the tail_lsn since much of the recovery has completed
4387 * and there may be space available to use. If there were no extent
4388 * or iunlinks, we can free up the entire log and set the tail_lsn to
4389 * be the last_sync_lsn. This was set in xlog_find_tail to be the
4390 * lsn of the last known good LR on disk. If there are extent frees
4391 * or iunlinks they will have some entries in the AIL; so we look at
4392 * the AIL to determine how to set the tail_lsn.
4394 xlog_assign_tail_lsn(log
->l_mp
);
4397 * Now that we've finished replaying all buffer and inode
4398 * updates, re-read in the superblock and reverify it.
4400 bp
= xfs_getsb(log
->l_mp
, 0);
4402 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
4404 XFS_BUF_UNASYNC(bp
);
4405 bp
->b_ops
= &xfs_sb_buf_ops
;
4406 xfsbdstrat(log
->l_mp
, bp
);
4407 error
= xfs_buf_iowait(bp
);
4409 xfs_buf_ioerror_alert(bp
, __func__
);
4415 /* Convert superblock from on-disk format */
4416 sbp
= &log
->l_mp
->m_sb
;
4417 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
4418 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
4419 ASSERT(xfs_sb_good_version(sbp
));
4422 /* We've re-read the superblock so re-initialize per-cpu counters */
4423 xfs_icsb_reinit_counters(log
->l_mp
);
4425 xlog_recover_check_summary(log
);
4427 /* Normal transactions can now occur */
4428 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
4433 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4435 * Return error or zero.
4441 xfs_daddr_t head_blk
, tail_blk
;
4444 /* find the tail of the log */
4445 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
4448 if (tail_blk
!= head_blk
) {
4449 /* There used to be a comment here:
4451 * disallow recovery on read-only mounts. note -- mount
4452 * checks for ENOSPC and turns it into an intelligent
4454 * ...but this is no longer true. Now, unless you specify
4455 * NORECOVERY (in which case this function would never be
4456 * called), we just go ahead and recover. We do this all
4457 * under the vfs layer, so we can get away with it unless
4458 * the device itself is read-only, in which case we fail.
4460 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
4465 * Version 5 superblock log feature mask validation. We know the
4466 * log is dirty so check if there are any unknown log features
4467 * in what we need to recover. If there are unknown features
4468 * (e.g. unsupported transactions, then simply reject the
4469 * attempt at recovery before touching anything.
4471 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
4472 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
4473 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
4475 "Superblock has unknown incompatible log features (0x%x) enabled.\n"
4476 "The log can not be fully and/or safely recovered by this kernel.\n"
4477 "Please recover the log on a kernel that supports the unknown features.",
4478 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
4479 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
4483 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
4484 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4487 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
4488 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
4494 * In the first part of recovery we replay inodes and buffers and build
4495 * up the list of extent free items which need to be processed. Here
4496 * we process the extent free items and clean up the on disk unlinked
4497 * inode lists. This is separated from the first part of recovery so
4498 * that the root and real-time bitmap inodes can be read in from disk in
4499 * between the two stages. This is necessary so that we can free space
4500 * in the real-time portion of the file system.
4503 xlog_recover_finish(
4507 * Now we're ready to do the transactions needed for the
4508 * rest of recovery. Start with completing all the extent
4509 * free intent records and then process the unlinked inode
4510 * lists. At this point, we essentially run in normal mode
4511 * except that we're still performing recovery actions
4512 * rather than accepting new requests.
4514 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
4516 error
= xlog_recover_process_efis(log
);
4518 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
4522 * Sync the log to get all the EFIs out of the AIL.
4523 * This isn't absolutely necessary, but it helps in
4524 * case the unlink transactions would have problems
4525 * pushing the EFIs out of the way.
4527 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
4529 xlog_recover_process_iunlinks(log
);
4531 xlog_recover_check_summary(log
);
4533 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
4534 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
4536 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
4538 xfs_info(log
->l_mp
, "Ending clean mount");
4546 * Read all of the agf and agi counters and check that they
4547 * are consistent with the superblock counters.
4550 xlog_recover_check_summary(
4557 xfs_agnumber_t agno
;
4558 __uint64_t freeblks
;
4568 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
4569 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
4571 xfs_alert(mp
, "%s agf read failed agno %d error %d",
4572 __func__
, agno
, error
);
4574 agfp
= XFS_BUF_TO_AGF(agfbp
);
4575 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
4576 be32_to_cpu(agfp
->agf_flcount
);
4577 xfs_buf_relse(agfbp
);
4580 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
4582 xfs_alert(mp
, "%s agi read failed agno %d error %d",
4583 __func__
, agno
, error
);
4585 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
4587 itotal
+= be32_to_cpu(agi
->agi_count
);
4588 ifree
+= be32_to_cpu(agi
->agi_freecount
);
4589 xfs_buf_relse(agibp
);