1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_trans.h"
21 #include "xfs_log_priv.h"
22 #include "xfs_log_recover.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_alloc.h"
27 #include "xfs_ialloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_cksum.h"
30 #include "xfs_trace.h"
31 #include "xfs_icache.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_error.h"
35 #include "xfs_rmap_item.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_refcount_item.h"
38 #include "xfs_bmap_item.h"
40 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
47 xlog_clear_stale_blocks(
52 xlog_recover_check_summary(
55 #define xlog_recover_check_summary(log)
58 xlog_do_recovery_pass(
59 struct xlog
*, xfs_daddr_t
, xfs_daddr_t
, int, xfs_daddr_t
*);
62 * This structure is used during recovery to record the buf log items which
63 * have been canceled and should not be replayed.
65 struct xfs_buf_cancel
{
69 struct list_head bc_list
;
73 * Sector aligned buffer routines for buffer create/read/write/access
77 * Verify the log-relative block number and length in basic blocks are valid for
78 * an operation involving the given XFS log buffer. Returns true if the fields
79 * are valid, false otherwise.
87 if (blk_no
< 0 || blk_no
>= log
->l_logBBsize
)
89 if (bbcount
<= 0 || (blk_no
+ bbcount
) > log
->l_logBBsize
)
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
107 * Pass log block 0 since we don't have an addr yet, buffer will be
110 if (!xlog_verify_bp(log
, 0, nbblks
)) {
111 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
113 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
118 * We do log I/O in units of log sectors (a power-of-2
119 * multiple of the basic block size), so we round up the
120 * requested size to accommodate the basic blocks required
121 * for complete log sectors.
123 * In addition, the buffer may be used for a non-sector-
124 * aligned block offset, in which case an I/O of the
125 * requested size could extend beyond the end of the
126 * buffer. If the requested size is only 1 basic block it
127 * will never straddle a sector boundary, so this won't be
128 * an issue. Nor will this be a problem if the log I/O is
129 * done in basic blocks (sector size 1). But otherwise we
130 * extend the buffer by one extra log sector to ensure
131 * there's space to accommodate this possibility.
133 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
134 nbblks
+= log
->l_sectBBsize
;
135 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
137 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, nbblks
, 0);
151 * Return the address of the start of the given block number's data
152 * in a log buffer. The buffer covers a log sector-aligned region.
161 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
163 ASSERT(offset
+ nbblks
<= bp
->b_length
);
164 return bp
->b_addr
+ BBTOB(offset
);
169 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
180 if (!xlog_verify_bp(log
, blk_no
, nbblks
)) {
182 "Invalid log block/length (0x%llx, 0x%x) for buffer",
184 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
185 return -EFSCORRUPTED
;
188 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
189 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
192 ASSERT(nbblks
<= bp
->b_length
);
194 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
195 bp
->b_flags
|= XBF_READ
;
196 bp
->b_io_length
= nbblks
;
199 error
= xfs_buf_submit_wait(bp
);
200 if (error
&& !XFS_FORCED_SHUTDOWN(log
->l_mp
))
201 xfs_buf_ioerror_alert(bp
, __func__
);
215 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
219 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
224 * Read at an offset into the buffer. Returns with the buffer in it's original
225 * state regardless of the result of the read.
230 xfs_daddr_t blk_no
, /* block to read from */
231 int nbblks
, /* blocks to read */
235 char *orig_offset
= bp
->b_addr
;
236 int orig_len
= BBTOB(bp
->b_length
);
239 error
= xfs_buf_associate_memory(bp
, offset
, BBTOB(nbblks
));
243 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
245 /* must reset buffer pointer even on error */
246 error2
= xfs_buf_associate_memory(bp
, orig_offset
, orig_len
);
253 * Write out the buffer at the given block for the given number of blocks.
254 * The buffer is kept locked across the write and is returned locked.
255 * This can only be used for synchronous log writes.
266 if (!xlog_verify_bp(log
, blk_no
, nbblks
)) {
268 "Invalid log block/length (0x%llx, 0x%x) for buffer",
270 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
271 return -EFSCORRUPTED
;
274 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
275 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
278 ASSERT(nbblks
<= bp
->b_length
);
280 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
283 bp
->b_io_length
= nbblks
;
286 error
= xfs_bwrite(bp
);
288 xfs_buf_ioerror_alert(bp
, __func__
);
295 * dump debug superblock and log record information
298 xlog_header_check_dump(
300 xlog_rec_header_t
*head
)
302 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d",
303 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
304 xfs_debug(mp
, " log : uuid = %pU, fmt = %d",
305 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
308 #define xlog_header_check_dump(mp, head)
312 * check log record header for recovery
315 xlog_header_check_recover(
317 xlog_rec_header_t
*head
)
319 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
322 * IRIX doesn't write the h_fmt field and leaves it zeroed
323 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
324 * a dirty log created in IRIX.
326 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
328 "dirty log written in incompatible format - can't recover");
329 xlog_header_check_dump(mp
, head
);
330 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
331 XFS_ERRLEVEL_HIGH
, mp
);
332 return -EFSCORRUPTED
;
333 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
335 "dirty log entry has mismatched uuid - can't recover");
336 xlog_header_check_dump(mp
, head
);
337 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
338 XFS_ERRLEVEL_HIGH
, mp
);
339 return -EFSCORRUPTED
;
345 * read the head block of the log and check the header
348 xlog_header_check_mount(
350 xlog_rec_header_t
*head
)
352 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
354 if (uuid_is_null(&head
->h_fs_uuid
)) {
356 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
357 * h_fs_uuid is null, we assume this log was last mounted
358 * by IRIX and continue.
360 xfs_warn(mp
, "null uuid in log - IRIX style log");
361 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
362 xfs_warn(mp
, "log has mismatched uuid - can't recover");
363 xlog_header_check_dump(mp
, head
);
364 XFS_ERROR_REPORT("xlog_header_check_mount",
365 XFS_ERRLEVEL_HIGH
, mp
);
366 return -EFSCORRUPTED
;
377 * We're not going to bother about retrying
378 * this during recovery. One strike!
380 if (!XFS_FORCED_SHUTDOWN(bp
->b_target
->bt_mount
)) {
381 xfs_buf_ioerror_alert(bp
, __func__
);
382 xfs_force_shutdown(bp
->b_target
->bt_mount
,
383 SHUTDOWN_META_IO_ERROR
);
388 * On v5 supers, a bli could be attached to update the metadata LSN.
392 xfs_buf_item_relse(bp
);
393 ASSERT(bp
->b_log_item
== NULL
);
400 * This routine finds (to an approximation) the first block in the physical
401 * log which contains the given cycle. It uses a binary search algorithm.
402 * Note that the algorithm can not be perfect because the disk will not
403 * necessarily be perfect.
406 xlog_find_cycle_start(
409 xfs_daddr_t first_blk
,
410 xfs_daddr_t
*last_blk
,
420 mid_blk
= BLK_AVG(first_blk
, end_blk
);
421 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
422 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
425 mid_cycle
= xlog_get_cycle(offset
);
426 if (mid_cycle
== cycle
)
427 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
429 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
430 mid_blk
= BLK_AVG(first_blk
, end_blk
);
432 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
433 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
441 * Check that a range of blocks does not contain stop_on_cycle_no.
442 * Fill in *new_blk with the block offset where such a block is
443 * found, or with -1 (an invalid block number) if there is no such
444 * block in the range. The scan needs to occur from front to back
445 * and the pointer into the region must be updated since a later
446 * routine will need to perform another test.
449 xlog_find_verify_cycle(
451 xfs_daddr_t start_blk
,
453 uint stop_on_cycle_no
,
454 xfs_daddr_t
*new_blk
)
464 * Greedily allocate a buffer big enough to handle the full
465 * range of basic blocks we'll be examining. If that fails,
466 * try a smaller size. We need to be able to read at least
467 * a log sector, or we're out of luck.
469 bufblks
= 1 << ffs(nbblks
);
470 while (bufblks
> log
->l_logBBsize
)
472 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
474 if (bufblks
< log
->l_sectBBsize
)
478 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
481 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
483 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
487 for (j
= 0; j
< bcount
; j
++) {
488 cycle
= xlog_get_cycle(buf
);
489 if (cycle
== stop_on_cycle_no
) {
506 * Potentially backup over partial log record write.
508 * In the typical case, last_blk is the number of the block directly after
509 * a good log record. Therefore, we subtract one to get the block number
510 * of the last block in the given buffer. extra_bblks contains the number
511 * of blocks we would have read on a previous read. This happens when the
512 * last log record is split over the end of the physical log.
514 * extra_bblks is the number of blocks potentially verified on a previous
515 * call to this routine.
518 xlog_find_verify_log_record(
520 xfs_daddr_t start_blk
,
521 xfs_daddr_t
*last_blk
,
527 xlog_rec_header_t
*head
= NULL
;
530 int num_blks
= *last_blk
- start_blk
;
533 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
535 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
536 if (!(bp
= xlog_get_bp(log
, 1)))
540 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
543 offset
+= ((num_blks
- 1) << BBSHIFT
);
546 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
548 /* valid log record not found */
550 "Log inconsistent (didn't find previous header)");
557 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
562 head
= (xlog_rec_header_t
*)offset
;
564 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
572 * We hit the beginning of the physical log & still no header. Return
573 * to caller. If caller can handle a return of -1, then this routine
574 * will be called again for the end of the physical log.
582 * We have the final block of the good log (the first block
583 * of the log record _before_ the head. So we check the uuid.
585 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
589 * We may have found a log record header before we expected one.
590 * last_blk will be the 1st block # with a given cycle #. We may end
591 * up reading an entire log record. In this case, we don't want to
592 * reset last_blk. Only when last_blk points in the middle of a log
593 * record do we update last_blk.
595 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
596 uint h_size
= be32_to_cpu(head
->h_size
);
598 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
599 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
605 if (*last_blk
- i
+ extra_bblks
!=
606 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
615 * Head is defined to be the point of the log where the next log write
616 * could go. This means that incomplete LR writes at the end are
617 * eliminated when calculating the head. We aren't guaranteed that previous
618 * LR have complete transactions. We only know that a cycle number of
619 * current cycle number -1 won't be present in the log if we start writing
620 * from our current block number.
622 * last_blk contains the block number of the first block with a given
625 * Return: zero if normal, non-zero if error.
630 xfs_daddr_t
*return_head_blk
)
634 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
636 uint first_half_cycle
, last_half_cycle
;
638 int error
, log_bbnum
= log
->l_logBBsize
;
640 /* Is the end of the log device zeroed? */
641 error
= xlog_find_zeroed(log
, &first_blk
);
643 xfs_warn(log
->l_mp
, "empty log check failed");
647 *return_head_blk
= first_blk
;
649 /* Is the whole lot zeroed? */
651 /* Linux XFS shouldn't generate totally zeroed logs -
652 * mkfs etc write a dummy unmount record to a fresh
653 * log so we can store the uuid in there
655 xfs_warn(log
->l_mp
, "totally zeroed log");
661 first_blk
= 0; /* get cycle # of 1st block */
662 bp
= xlog_get_bp(log
, 1);
666 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
670 first_half_cycle
= xlog_get_cycle(offset
);
672 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
673 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
677 last_half_cycle
= xlog_get_cycle(offset
);
678 ASSERT(last_half_cycle
!= 0);
681 * If the 1st half cycle number is equal to the last half cycle number,
682 * then the entire log is stamped with the same cycle number. In this
683 * case, head_blk can't be set to zero (which makes sense). The below
684 * math doesn't work out properly with head_blk equal to zero. Instead,
685 * we set it to log_bbnum which is an invalid block number, but this
686 * value makes the math correct. If head_blk doesn't changed through
687 * all the tests below, *head_blk is set to zero at the very end rather
688 * than log_bbnum. In a sense, log_bbnum and zero are the same block
689 * in a circular file.
691 if (first_half_cycle
== last_half_cycle
) {
693 * In this case we believe that the entire log should have
694 * cycle number last_half_cycle. We need to scan backwards
695 * from the end verifying that there are no holes still
696 * containing last_half_cycle - 1. If we find such a hole,
697 * then the start of that hole will be the new head. The
698 * simple case looks like
699 * x | x ... | x - 1 | x
700 * Another case that fits this picture would be
701 * x | x + 1 | x ... | x
702 * In this case the head really is somewhere at the end of the
703 * log, as one of the latest writes at the beginning was
706 * x | x + 1 | x ... | x - 1 | x
707 * This is really the combination of the above two cases, and
708 * the head has to end up at the start of the x-1 hole at the
711 * In the 256k log case, we will read from the beginning to the
712 * end of the log and search for cycle numbers equal to x-1.
713 * We don't worry about the x+1 blocks that we encounter,
714 * because we know that they cannot be the head since the log
717 head_blk
= log_bbnum
;
718 stop_on_cycle
= last_half_cycle
- 1;
721 * In this case we want to find the first block with cycle
722 * number matching last_half_cycle. We expect the log to be
724 * x + 1 ... | x ... | x
725 * The first block with cycle number x (last_half_cycle) will
726 * be where the new head belongs. First we do a binary search
727 * for the first occurrence of last_half_cycle. The binary
728 * search may not be totally accurate, so then we scan back
729 * from there looking for occurrences of last_half_cycle before
730 * us. If that backwards scan wraps around the beginning of
731 * the log, then we look for occurrences of last_half_cycle - 1
732 * at the end of the log. The cases we're looking for look
734 * v binary search stopped here
735 * x + 1 ... | x | x + 1 | x ... | x
736 * ^ but we want to locate this spot
738 * <---------> less than scan distance
739 * x + 1 ... | x ... | x - 1 | x
740 * ^ we want to locate this spot
742 stop_on_cycle
= last_half_cycle
;
743 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
744 &head_blk
, last_half_cycle
)))
749 * Now validate the answer. Scan back some number of maximum possible
750 * blocks and make sure each one has the expected cycle number. The
751 * maximum is determined by the total possible amount of buffering
752 * in the in-core log. The following number can be made tighter if
753 * we actually look at the block size of the filesystem.
755 num_scan_bblks
= min_t(int, log_bbnum
, XLOG_TOTAL_REC_SHIFT(log
));
756 if (head_blk
>= num_scan_bblks
) {
758 * We are guaranteed that the entire check can be performed
761 start_blk
= head_blk
- num_scan_bblks
;
762 if ((error
= xlog_find_verify_cycle(log
,
763 start_blk
, num_scan_bblks
,
764 stop_on_cycle
, &new_blk
)))
768 } else { /* need to read 2 parts of log */
770 * We are going to scan backwards in the log in two parts.
771 * First we scan the physical end of the log. In this part
772 * of the log, we are looking for blocks with cycle number
773 * last_half_cycle - 1.
774 * If we find one, then we know that the log starts there, as
775 * we've found a hole that didn't get written in going around
776 * the end of the physical log. The simple case for this is
777 * x + 1 ... | x ... | x - 1 | x
778 * <---------> less than scan distance
779 * If all of the blocks at the end of the log have cycle number
780 * last_half_cycle, then we check the blocks at the start of
781 * the log looking for occurrences of last_half_cycle. If we
782 * find one, then our current estimate for the location of the
783 * first occurrence of last_half_cycle is wrong and we move
784 * back to the hole we've found. This case looks like
785 * x + 1 ... | x | x + 1 | x ...
786 * ^ binary search stopped here
787 * Another case we need to handle that only occurs in 256k
789 * x + 1 ... | x ... | x+1 | x ...
790 * ^ binary search stops here
791 * In a 256k log, the scan at the end of the log will see the
792 * x + 1 blocks. We need to skip past those since that is
793 * certainly not the head of the log. By searching for
794 * last_half_cycle-1 we accomplish that.
796 ASSERT(head_blk
<= INT_MAX
&&
797 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
798 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
799 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
800 num_scan_bblks
- (int)head_blk
,
801 (stop_on_cycle
- 1), &new_blk
)))
809 * Scan beginning of log now. The last part of the physical
810 * log is good. This scan needs to verify that it doesn't find
811 * the last_half_cycle.
814 ASSERT(head_blk
<= INT_MAX
);
815 if ((error
= xlog_find_verify_cycle(log
,
816 start_blk
, (int)head_blk
,
817 stop_on_cycle
, &new_blk
)))
825 * Now we need to make sure head_blk is not pointing to a block in
826 * the middle of a log record.
828 num_scan_bblks
= XLOG_REC_SHIFT(log
);
829 if (head_blk
>= num_scan_bblks
) {
830 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
832 /* start ptr at last block ptr before head_blk */
833 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
840 ASSERT(head_blk
<= INT_MAX
);
841 error
= xlog_find_verify_log_record(log
, start_blk
, &head_blk
, 0);
845 /* We hit the beginning of the log during our search */
846 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
848 ASSERT(start_blk
<= INT_MAX
&&
849 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
850 ASSERT(head_blk
<= INT_MAX
);
851 error
= xlog_find_verify_log_record(log
, start_blk
,
852 &new_blk
, (int)head_blk
);
857 if (new_blk
!= log_bbnum
)
864 if (head_blk
== log_bbnum
)
865 *return_head_blk
= 0;
867 *return_head_blk
= head_blk
;
869 * When returning here, we have a good block number. Bad block
870 * means that during a previous crash, we didn't have a clean break
871 * from cycle number N to cycle number N-1. In this case, we need
872 * to find the first block with cycle number N-1.
880 xfs_warn(log
->l_mp
, "failed to find log head");
885 * Seek backwards in the log for log record headers.
887 * Given a starting log block, walk backwards until we find the provided number
888 * of records or hit the provided tail block. The return value is the number of
889 * records encountered or a negative error code. The log block and buffer
890 * pointer of the last record seen are returned in rblk and rhead respectively.
893 xlog_rseek_logrec_hdr(
895 xfs_daddr_t head_blk
,
896 xfs_daddr_t tail_blk
,
900 struct xlog_rec_header
**rhead
,
912 * Walk backwards from the head block until we hit the tail or the first
915 end_blk
= head_blk
> tail_blk
? tail_blk
: 0;
916 for (i
= (int) head_blk
- 1; i
>= end_blk
; i
--) {
917 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
921 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
923 *rhead
= (struct xlog_rec_header
*) offset
;
924 if (++found
== count
)
930 * If we haven't hit the tail block or the log record header count,
931 * start looking again from the end of the physical log. Note that
932 * callers can pass head == tail if the tail is not yet known.
934 if (tail_blk
>= head_blk
&& found
!= count
) {
935 for (i
= log
->l_logBBsize
- 1; i
>= (int) tail_blk
; i
--) {
936 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
940 if (*(__be32
*)offset
==
941 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
944 *rhead
= (struct xlog_rec_header
*) offset
;
945 if (++found
== count
)
958 * Seek forward in the log for log record headers.
960 * Given head and tail blocks, walk forward from the tail block until we find
961 * the provided number of records or hit the head block. The return value is the
962 * number of records encountered or a negative error code. The log block and
963 * buffer pointer of the last record seen are returned in rblk and rhead
967 xlog_seek_logrec_hdr(
969 xfs_daddr_t head_blk
,
970 xfs_daddr_t tail_blk
,
974 struct xlog_rec_header
**rhead
,
986 * Walk forward from the tail block until we hit the head or the last
989 end_blk
= head_blk
> tail_blk
? head_blk
: log
->l_logBBsize
- 1;
990 for (i
= (int) tail_blk
; i
<= end_blk
; i
++) {
991 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
995 if (*(__be32
*) offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
997 *rhead
= (struct xlog_rec_header
*) offset
;
998 if (++found
== count
)
1004 * If we haven't hit the head block or the log record header count,
1005 * start looking again from the start of the physical log.
1007 if (tail_blk
> head_blk
&& found
!= count
) {
1008 for (i
= 0; i
< (int) head_blk
; i
++) {
1009 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
1013 if (*(__be32
*)offset
==
1014 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
1017 *rhead
= (struct xlog_rec_header
*) offset
;
1018 if (++found
== count
)
1031 * Calculate distance from head to tail (i.e., unused space in the log).
1036 xfs_daddr_t head_blk
,
1037 xfs_daddr_t tail_blk
)
1039 if (head_blk
< tail_blk
)
1040 return tail_blk
- head_blk
;
1042 return tail_blk
+ (log
->l_logBBsize
- head_blk
);
1046 * Verify the log tail. This is particularly important when torn or incomplete
1047 * writes have been detected near the front of the log and the head has been
1048 * walked back accordingly.
1050 * We also have to handle the case where the tail was pinned and the head
1051 * blocked behind the tail right before a crash. If the tail had been pushed
1052 * immediately prior to the crash and the subsequent checkpoint was only
1053 * partially written, it's possible it overwrote the last referenced tail in the
1054 * log with garbage. This is not a coherency problem because the tail must have
1055 * been pushed before it can be overwritten, but appears as log corruption to
1056 * recovery because we have no way to know the tail was updated if the
1057 * subsequent checkpoint didn't write successfully.
1059 * Therefore, CRC check the log from tail to head. If a failure occurs and the
1060 * offending record is within max iclog bufs from the head, walk the tail
1061 * forward and retry until a valid tail is found or corruption is detected out
1062 * of the range of a possible overwrite.
1067 xfs_daddr_t head_blk
,
1068 xfs_daddr_t
*tail_blk
,
1071 struct xlog_rec_header
*thead
;
1073 xfs_daddr_t first_bad
;
1076 xfs_daddr_t tmp_tail
;
1077 xfs_daddr_t orig_tail
= *tail_blk
;
1079 bp
= xlog_get_bp(log
, 1);
1084 * Make sure the tail points to a record (returns positive count on
1087 error
= xlog_seek_logrec_hdr(log
, head_blk
, *tail_blk
, 1, bp
,
1088 &tmp_tail
, &thead
, &wrapped
);
1091 if (*tail_blk
!= tmp_tail
)
1092 *tail_blk
= tmp_tail
;
1095 * Run a CRC check from the tail to the head. We can't just check
1096 * MAX_ICLOGS records past the tail because the tail may point to stale
1097 * blocks cleared during the search for the head/tail. These blocks are
1098 * overwritten with zero-length records and thus record count is not a
1099 * reliable indicator of the iclog state before a crash.
1102 error
= xlog_do_recovery_pass(log
, head_blk
, *tail_blk
,
1103 XLOG_RECOVER_CRCPASS
, &first_bad
);
1104 while ((error
== -EFSBADCRC
|| error
== -EFSCORRUPTED
) && first_bad
) {
1108 * Is corruption within range of the head? If so, retry from
1109 * the next record. Otherwise return an error.
1111 tail_distance
= xlog_tail_distance(log
, head_blk
, first_bad
);
1112 if (tail_distance
> BTOBB(XLOG_MAX_ICLOGS
* hsize
))
1115 /* skip to the next record; returns positive count on success */
1116 error
= xlog_seek_logrec_hdr(log
, head_blk
, first_bad
, 2, bp
,
1117 &tmp_tail
, &thead
, &wrapped
);
1121 *tail_blk
= tmp_tail
;
1123 error
= xlog_do_recovery_pass(log
, head_blk
, *tail_blk
,
1124 XLOG_RECOVER_CRCPASS
, &first_bad
);
1127 if (!error
&& *tail_blk
!= orig_tail
)
1129 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1130 orig_tail
, *tail_blk
);
1137 * Detect and trim torn writes from the head of the log.
1139 * Storage without sector atomicity guarantees can result in torn writes in the
1140 * log in the event of a crash. Our only means to detect this scenario is via
1141 * CRC verification. While we can't always be certain that CRC verification
1142 * failure is due to a torn write vs. an unrelated corruption, we do know that
1143 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1144 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1145 * the log and treat failures in this range as torn writes as a matter of
1146 * policy. In the event of CRC failure, the head is walked back to the last good
1147 * record in the log and the tail is updated from that record and verified.
1152 xfs_daddr_t
*head_blk
, /* in/out: unverified head */
1153 xfs_daddr_t
*tail_blk
, /* out: tail block */
1155 xfs_daddr_t
*rhead_blk
, /* start blk of last record */
1156 struct xlog_rec_header
**rhead
, /* ptr to last record */
1157 bool *wrapped
) /* last rec. wraps phys. log */
1159 struct xlog_rec_header
*tmp_rhead
;
1160 struct xfs_buf
*tmp_bp
;
1161 xfs_daddr_t first_bad
;
1162 xfs_daddr_t tmp_rhead_blk
;
1168 * Check the head of the log for torn writes. Search backwards from the
1169 * head until we hit the tail or the maximum number of log record I/Os
1170 * that could have been in flight at one time. Use a temporary buffer so
1171 * we don't trash the rhead/bp pointers from the caller.
1173 tmp_bp
= xlog_get_bp(log
, 1);
1176 error
= xlog_rseek_logrec_hdr(log
, *head_blk
, *tail_blk
,
1177 XLOG_MAX_ICLOGS
, tmp_bp
, &tmp_rhead_blk
,
1178 &tmp_rhead
, &tmp_wrapped
);
1179 xlog_put_bp(tmp_bp
);
1184 * Now run a CRC verification pass over the records starting at the
1185 * block found above to the current head. If a CRC failure occurs, the
1186 * log block of the first bad record is saved in first_bad.
1188 error
= xlog_do_recovery_pass(log
, *head_blk
, tmp_rhead_blk
,
1189 XLOG_RECOVER_CRCPASS
, &first_bad
);
1190 if ((error
== -EFSBADCRC
|| error
== -EFSCORRUPTED
) && first_bad
) {
1192 * We've hit a potential torn write. Reset the error and warn
1197 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1198 first_bad
, *head_blk
);
1201 * Get the header block and buffer pointer for the last good
1202 * record before the bad record.
1204 * Note that xlog_find_tail() clears the blocks at the new head
1205 * (i.e., the records with invalid CRC) if the cycle number
1206 * matches the the current cycle.
1208 found
= xlog_rseek_logrec_hdr(log
, first_bad
, *tail_blk
, 1, bp
,
1209 rhead_blk
, rhead
, wrapped
);
1212 if (found
== 0) /* XXX: right thing to do here? */
1216 * Reset the head block to the starting block of the first bad
1217 * log record and set the tail block based on the last good
1220 * Bail out if the updated head/tail match as this indicates
1221 * possible corruption outside of the acceptable
1222 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1224 *head_blk
= first_bad
;
1225 *tail_blk
= BLOCK_LSN(be64_to_cpu((*rhead
)->h_tail_lsn
));
1226 if (*head_blk
== *tail_blk
) {
1234 return xlog_verify_tail(log
, *head_blk
, tail_blk
,
1235 be32_to_cpu((*rhead
)->h_size
));
1239 * Check whether the head of the log points to an unmount record. In other
1240 * words, determine whether the log is clean. If so, update the in-core state
1244 xlog_check_unmount_rec(
1246 xfs_daddr_t
*head_blk
,
1247 xfs_daddr_t
*tail_blk
,
1248 struct xlog_rec_header
*rhead
,
1249 xfs_daddr_t rhead_blk
,
1253 struct xlog_op_header
*op_head
;
1254 xfs_daddr_t umount_data_blk
;
1255 xfs_daddr_t after_umount_blk
;
1263 * Look for unmount record. If we find it, then we know there was a
1264 * clean unmount. Since 'i' could be the last block in the physical
1265 * log, we convert to a log block before comparing to the head_blk.
1267 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1268 * below. We won't want to clear the unmount record if there is one, so
1269 * we pass the lsn of the unmount record rather than the block after it.
1271 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
1272 int h_size
= be32_to_cpu(rhead
->h_size
);
1273 int h_version
= be32_to_cpu(rhead
->h_version
);
1275 if ((h_version
& XLOG_VERSION_2
) &&
1276 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
1277 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
1278 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1286 after_umount_blk
= rhead_blk
+ hblks
+ BTOBB(be32_to_cpu(rhead
->h_len
));
1287 after_umount_blk
= do_mod(after_umount_blk
, log
->l_logBBsize
);
1288 if (*head_blk
== after_umount_blk
&&
1289 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1290 umount_data_blk
= rhead_blk
+ hblks
;
1291 umount_data_blk
= do_mod(umount_data_blk
, log
->l_logBBsize
);
1292 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1296 op_head
= (struct xlog_op_header
*)offset
;
1297 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1299 * Set tail and last sync so that newly written log
1300 * records will point recovery to after the current
1303 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1304 log
->l_curr_cycle
, after_umount_blk
);
1305 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1306 log
->l_curr_cycle
, after_umount_blk
);
1307 *tail_blk
= after_umount_blk
;
1319 xfs_daddr_t head_blk
,
1320 struct xlog_rec_header
*rhead
,
1321 xfs_daddr_t rhead_blk
,
1325 * Reset log values according to the state of the log when we
1326 * crashed. In the case where head_blk == 0, we bump curr_cycle
1327 * one because the next write starts a new cycle rather than
1328 * continuing the cycle of the last good log record. At this
1329 * point we have guaranteed that all partial log records have been
1330 * accounted for. Therefore, we know that the last good log record
1331 * written was complete and ended exactly on the end boundary
1332 * of the physical log.
1334 log
->l_prev_block
= rhead_blk
;
1335 log
->l_curr_block
= (int)head_blk
;
1336 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
1338 log
->l_curr_cycle
++;
1339 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
1340 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
1341 xlog_assign_grant_head(&log
->l_reserve_head
.grant
, log
->l_curr_cycle
,
1342 BBTOB(log
->l_curr_block
));
1343 xlog_assign_grant_head(&log
->l_write_head
.grant
, log
->l_curr_cycle
,
1344 BBTOB(log
->l_curr_block
));
1348 * Find the sync block number or the tail of the log.
1350 * This will be the block number of the last record to have its
1351 * associated buffers synced to disk. Every log record header has
1352 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1353 * to get a sync block number. The only concern is to figure out which
1354 * log record header to believe.
1356 * The following algorithm uses the log record header with the largest
1357 * lsn. The entire log record does not need to be valid. We only care
1358 * that the header is valid.
1360 * We could speed up search by using current head_blk buffer, but it is not
1366 xfs_daddr_t
*head_blk
,
1367 xfs_daddr_t
*tail_blk
)
1369 xlog_rec_header_t
*rhead
;
1370 char *offset
= NULL
;
1373 xfs_daddr_t rhead_blk
;
1375 bool wrapped
= false;
1379 * Find previous log record
1381 if ((error
= xlog_find_head(log
, head_blk
)))
1383 ASSERT(*head_blk
< INT_MAX
);
1385 bp
= xlog_get_bp(log
, 1);
1388 if (*head_blk
== 0) { /* special case */
1389 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1393 if (xlog_get_cycle(offset
) == 0) {
1395 /* leave all other log inited values alone */
1401 * Search backwards through the log looking for the log record header
1402 * block. This wraps all the way back around to the head so something is
1403 * seriously wrong if we can't find it.
1405 error
= xlog_rseek_logrec_hdr(log
, *head_blk
, *head_blk
, 1, bp
,
1406 &rhead_blk
, &rhead
, &wrapped
);
1410 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
1413 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
1416 * Set the log state based on the current head record.
1418 xlog_set_state(log
, *head_blk
, rhead
, rhead_blk
, wrapped
);
1419 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1422 * Look for an unmount record at the head of the log. This sets the log
1423 * state to determine whether recovery is necessary.
1425 error
= xlog_check_unmount_rec(log
, head_blk
, tail_blk
, rhead
,
1426 rhead_blk
, bp
, &clean
);
1431 * Verify the log head if the log is not clean (e.g., we have anything
1432 * but an unmount record at the head). This uses CRC verification to
1433 * detect and trim torn writes. If discovered, CRC failures are
1434 * considered torn writes and the log head is trimmed accordingly.
1436 * Note that we can only run CRC verification when the log is dirty
1437 * because there's no guarantee that the log data behind an unmount
1438 * record is compatible with the current architecture.
1441 xfs_daddr_t orig_head
= *head_blk
;
1443 error
= xlog_verify_head(log
, head_blk
, tail_blk
, bp
,
1444 &rhead_blk
, &rhead
, &wrapped
);
1448 /* update in-core state again if the head changed */
1449 if (*head_blk
!= orig_head
) {
1450 xlog_set_state(log
, *head_blk
, rhead
, rhead_blk
,
1452 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1453 error
= xlog_check_unmount_rec(log
, head_blk
, tail_blk
,
1454 rhead
, rhead_blk
, bp
,
1462 * Note that the unmount was clean. If the unmount was not clean, we
1463 * need to know this to rebuild the superblock counters from the perag
1464 * headers if we have a filesystem using non-persistent counters.
1467 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1470 * Make sure that there are no blocks in front of the head
1471 * with the same cycle number as the head. This can happen
1472 * because we allow multiple outstanding log writes concurrently,
1473 * and the later writes might make it out before earlier ones.
1475 * We use the lsn from before modifying it so that we'll never
1476 * overwrite the unmount record after a clean unmount.
1478 * Do this only if we are going to recover the filesystem
1480 * NOTE: This used to say "if (!readonly)"
1481 * However on Linux, we can & do recover a read-only filesystem.
1482 * We only skip recovery if NORECOVERY is specified on mount,
1483 * in which case we would not be here.
1485 * But... if the -device- itself is readonly, just skip this.
1486 * We can't recover this device anyway, so it won't matter.
1488 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1489 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1495 xfs_warn(log
->l_mp
, "failed to locate log tail");
1500 * Is the log zeroed at all?
1502 * The last binary search should be changed to perform an X block read
1503 * once X becomes small enough. You can then search linearly through
1504 * the X blocks. This will cut down on the number of reads we need to do.
1506 * If the log is partially zeroed, this routine will pass back the blkno
1507 * of the first block with cycle number 0. It won't have a complete LR
1511 * 0 => the log is completely written to
1512 * 1 => use *blk_no as the first block of the log
1513 * <0 => error has occurred
1518 xfs_daddr_t
*blk_no
)
1522 uint first_cycle
, last_cycle
;
1523 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1524 xfs_daddr_t num_scan_bblks
;
1525 int error
, log_bbnum
= log
->l_logBBsize
;
1529 /* check totally zeroed log */
1530 bp
= xlog_get_bp(log
, 1);
1533 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1537 first_cycle
= xlog_get_cycle(offset
);
1538 if (first_cycle
== 0) { /* completely zeroed log */
1544 /* check partially zeroed log */
1545 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1549 last_cycle
= xlog_get_cycle(offset
);
1550 if (last_cycle
!= 0) { /* log completely written to */
1553 } else if (first_cycle
!= 1) {
1555 * If the cycle of the last block is zero, the cycle of
1556 * the first block must be 1. If it's not, maybe we're
1557 * not looking at a log... Bail out.
1560 "Log inconsistent or not a log (last==0, first!=1)");
1565 /* we have a partially zeroed log */
1566 last_blk
= log_bbnum
-1;
1567 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1571 * Validate the answer. Because there is no way to guarantee that
1572 * the entire log is made up of log records which are the same size,
1573 * we scan over the defined maximum blocks. At this point, the maximum
1574 * is not chosen to mean anything special. XXXmiken
1576 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1577 ASSERT(num_scan_bblks
<= INT_MAX
);
1579 if (last_blk
< num_scan_bblks
)
1580 num_scan_bblks
= last_blk
;
1581 start_blk
= last_blk
- num_scan_bblks
;
1584 * We search for any instances of cycle number 0 that occur before
1585 * our current estimate of the head. What we're trying to detect is
1586 * 1 ... | 0 | 1 | 0...
1587 * ^ binary search ends here
1589 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1590 (int)num_scan_bblks
, 0, &new_blk
)))
1596 * Potentially backup over partial log record write. We don't need
1597 * to search the end of the log because we know it is zero.
1599 error
= xlog_find_verify_log_record(log
, start_blk
, &last_blk
, 0);
1614 * These are simple subroutines used by xlog_clear_stale_blocks() below
1615 * to initialize a buffer full of empty log record headers and write
1616 * them into the log.
1627 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1629 memset(buf
, 0, BBSIZE
);
1630 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1631 recp
->h_cycle
= cpu_to_be32(cycle
);
1632 recp
->h_version
= cpu_to_be32(
1633 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1634 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1635 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1636 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1637 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1641 xlog_write_log_records(
1652 int sectbb
= log
->l_sectBBsize
;
1653 int end_block
= start_block
+ blocks
;
1659 * Greedily allocate a buffer big enough to handle the full
1660 * range of basic blocks to be written. If that fails, try
1661 * a smaller size. We need to be able to write at least a
1662 * log sector, or we're out of luck.
1664 bufblks
= 1 << ffs(blocks
);
1665 while (bufblks
> log
->l_logBBsize
)
1667 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1669 if (bufblks
< sectbb
)
1673 /* We may need to do a read at the start to fill in part of
1674 * the buffer in the starting sector not covered by the first
1677 balign
= round_down(start_block
, sectbb
);
1678 if (balign
!= start_block
) {
1679 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1683 j
= start_block
- balign
;
1686 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1687 int bcount
, endcount
;
1689 bcount
= min(bufblks
, end_block
- start_block
);
1690 endcount
= bcount
- j
;
1692 /* We may need to do a read at the end to fill in part of
1693 * the buffer in the final sector not covered by the write.
1694 * If this is the same sector as the above read, skip it.
1696 ealign
= round_down(end_block
, sectbb
);
1697 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1698 offset
= bp
->b_addr
+ BBTOB(ealign
- start_block
);
1699 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1706 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1707 for (; j
< endcount
; j
++) {
1708 xlog_add_record(log
, offset
, cycle
, i
+j
,
1709 tail_cycle
, tail_block
);
1712 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1715 start_block
+= endcount
;
1725 * This routine is called to blow away any incomplete log writes out
1726 * in front of the log head. We do this so that we won't become confused
1727 * if we come up, write only a little bit more, and then crash again.
1728 * If we leave the partial log records out there, this situation could
1729 * cause us to think those partial writes are valid blocks since they
1730 * have the current cycle number. We get rid of them by overwriting them
1731 * with empty log records with the old cycle number rather than the
1734 * The tail lsn is passed in rather than taken from
1735 * the log so that we will not write over the unmount record after a
1736 * clean unmount in a 512 block log. Doing so would leave the log without
1737 * any valid log records in it until a new one was written. If we crashed
1738 * during that time we would not be able to recover.
1741 xlog_clear_stale_blocks(
1745 int tail_cycle
, head_cycle
;
1746 int tail_block
, head_block
;
1747 int tail_distance
, max_distance
;
1751 tail_cycle
= CYCLE_LSN(tail_lsn
);
1752 tail_block
= BLOCK_LSN(tail_lsn
);
1753 head_cycle
= log
->l_curr_cycle
;
1754 head_block
= log
->l_curr_block
;
1757 * Figure out the distance between the new head of the log
1758 * and the tail. We want to write over any blocks beyond the
1759 * head that we may have written just before the crash, but
1760 * we don't want to overwrite the tail of the log.
1762 if (head_cycle
== tail_cycle
) {
1764 * The tail is behind the head in the physical log,
1765 * so the distance from the head to the tail is the
1766 * distance from the head to the end of the log plus
1767 * the distance from the beginning of the log to the
1770 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1771 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1772 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1773 return -EFSCORRUPTED
;
1775 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1778 * The head is behind the tail in the physical log,
1779 * so the distance from the head to the tail is just
1780 * the tail block minus the head block.
1782 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1783 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1784 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1785 return -EFSCORRUPTED
;
1787 tail_distance
= tail_block
- head_block
;
1791 * If the head is right up against the tail, we can't clear
1794 if (tail_distance
<= 0) {
1795 ASSERT(tail_distance
== 0);
1799 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1801 * Take the smaller of the maximum amount of outstanding I/O
1802 * we could have and the distance to the tail to clear out.
1803 * We take the smaller so that we don't overwrite the tail and
1804 * we don't waste all day writing from the head to the tail
1807 max_distance
= MIN(max_distance
, tail_distance
);
1809 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1811 * We can stomp all the blocks we need to without
1812 * wrapping around the end of the log. Just do it
1813 * in a single write. Use the cycle number of the
1814 * current cycle minus one so that the log will look like:
1817 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1818 head_block
, max_distance
, tail_cycle
,
1824 * We need to wrap around the end of the physical log in
1825 * order to clear all the blocks. Do it in two separate
1826 * I/Os. The first write should be from the head to the
1827 * end of the physical log, and it should use the current
1828 * cycle number minus one just like above.
1830 distance
= log
->l_logBBsize
- head_block
;
1831 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1832 head_block
, distance
, tail_cycle
,
1839 * Now write the blocks at the start of the physical log.
1840 * This writes the remainder of the blocks we want to clear.
1841 * It uses the current cycle number since we're now on the
1842 * same cycle as the head so that we get:
1843 * n ... n ... | n - 1 ...
1844 * ^^^^^ blocks we're writing
1846 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1847 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1848 tail_cycle
, tail_block
);
1856 /******************************************************************************
1858 * Log recover routines
1860 ******************************************************************************
1864 * Sort the log items in the transaction.
1866 * The ordering constraints are defined by the inode allocation and unlink
1867 * behaviour. The rules are:
1869 * 1. Every item is only logged once in a given transaction. Hence it
1870 * represents the last logged state of the item. Hence ordering is
1871 * dependent on the order in which operations need to be performed so
1872 * required initial conditions are always met.
1874 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1875 * there's nothing to replay from them so we can simply cull them
1876 * from the transaction. However, we can't do that until after we've
1877 * replayed all the other items because they may be dependent on the
1878 * cancelled buffer and replaying the cancelled buffer can remove it
1879 * form the cancelled buffer table. Hence they have tobe done last.
1881 * 3. Inode allocation buffers must be replayed before inode items that
1882 * read the buffer and replay changes into it. For filesystems using the
1883 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1884 * treated the same as inode allocation buffers as they create and
1885 * initialise the buffers directly.
1887 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1888 * This ensures that inodes are completely flushed to the inode buffer
1889 * in a "free" state before we remove the unlinked inode list pointer.
1891 * Hence the ordering needs to be inode allocation buffers first, inode items
1892 * second, inode unlink buffers third and cancelled buffers last.
1894 * But there's a problem with that - we can't tell an inode allocation buffer
1895 * apart from a regular buffer, so we can't separate them. We can, however,
1896 * tell an inode unlink buffer from the others, and so we can separate them out
1897 * from all the other buffers and move them to last.
1899 * Hence, 4 lists, in order from head to tail:
1900 * - buffer_list for all buffers except cancelled/inode unlink buffers
1901 * - item_list for all non-buffer items
1902 * - inode_buffer_list for inode unlink buffers
1903 * - cancel_list for the cancelled buffers
1905 * Note that we add objects to the tail of the lists so that first-to-last
1906 * ordering is preserved within the lists. Adding objects to the head of the
1907 * list means when we traverse from the head we walk them in last-to-first
1908 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1909 * but for all other items there may be specific ordering that we need to
1913 xlog_recover_reorder_trans(
1915 struct xlog_recover
*trans
,
1918 xlog_recover_item_t
*item
, *n
;
1920 LIST_HEAD(sort_list
);
1921 LIST_HEAD(cancel_list
);
1922 LIST_HEAD(buffer_list
);
1923 LIST_HEAD(inode_buffer_list
);
1924 LIST_HEAD(inode_list
);
1926 list_splice_init(&trans
->r_itemq
, &sort_list
);
1927 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1928 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1930 switch (ITEM_TYPE(item
)) {
1931 case XFS_LI_ICREATE
:
1932 list_move_tail(&item
->ri_list
, &buffer_list
);
1935 if (buf_f
->blf_flags
& XFS_BLF_CANCEL
) {
1936 trace_xfs_log_recover_item_reorder_head(log
,
1938 list_move(&item
->ri_list
, &cancel_list
);
1941 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
1942 list_move(&item
->ri_list
, &inode_buffer_list
);
1945 list_move_tail(&item
->ri_list
, &buffer_list
);
1949 case XFS_LI_QUOTAOFF
:
1958 trace_xfs_log_recover_item_reorder_tail(log
,
1960 list_move_tail(&item
->ri_list
, &inode_list
);
1964 "%s: unrecognized type of log operation",
1968 * return the remaining items back to the transaction
1969 * item list so they can be freed in caller.
1971 if (!list_empty(&sort_list
))
1972 list_splice_init(&sort_list
, &trans
->r_itemq
);
1978 ASSERT(list_empty(&sort_list
));
1979 if (!list_empty(&buffer_list
))
1980 list_splice(&buffer_list
, &trans
->r_itemq
);
1981 if (!list_empty(&inode_list
))
1982 list_splice_tail(&inode_list
, &trans
->r_itemq
);
1983 if (!list_empty(&inode_buffer_list
))
1984 list_splice_tail(&inode_buffer_list
, &trans
->r_itemq
);
1985 if (!list_empty(&cancel_list
))
1986 list_splice_tail(&cancel_list
, &trans
->r_itemq
);
1991 * Build up the table of buf cancel records so that we don't replay
1992 * cancelled data in the second pass. For buffer records that are
1993 * not cancel records, there is nothing to do here so we just return.
1995 * If we get a cancel record which is already in the table, this indicates
1996 * that the buffer was cancelled multiple times. In order to ensure
1997 * that during pass 2 we keep the record in the table until we reach its
1998 * last occurrence in the log, we keep a reference count in the cancel
1999 * record in the table to tell us how many times we expect to see this
2000 * record during the second pass.
2003 xlog_recover_buffer_pass1(
2005 struct xlog_recover_item
*item
)
2007 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2008 struct list_head
*bucket
;
2009 struct xfs_buf_cancel
*bcp
;
2012 * If this isn't a cancel buffer item, then just return.
2014 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
2015 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
2020 * Insert an xfs_buf_cancel record into the hash table of them.
2021 * If there is already an identical record, bump its reference count.
2023 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
2024 list_for_each_entry(bcp
, bucket
, bc_list
) {
2025 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
2026 bcp
->bc_len
== buf_f
->blf_len
) {
2028 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
2033 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
2034 bcp
->bc_blkno
= buf_f
->blf_blkno
;
2035 bcp
->bc_len
= buf_f
->blf_len
;
2036 bcp
->bc_refcount
= 1;
2037 list_add_tail(&bcp
->bc_list
, bucket
);
2039 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
2044 * Check to see whether the buffer being recovered has a corresponding
2045 * entry in the buffer cancel record table. If it is, return the cancel
2046 * buffer structure to the caller.
2048 STATIC
struct xfs_buf_cancel
*
2049 xlog_peek_buffer_cancelled(
2053 unsigned short flags
)
2055 struct list_head
*bucket
;
2056 struct xfs_buf_cancel
*bcp
;
2058 if (!log
->l_buf_cancel_table
) {
2059 /* empty table means no cancelled buffers in the log */
2060 ASSERT(!(flags
& XFS_BLF_CANCEL
));
2064 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
2065 list_for_each_entry(bcp
, bucket
, bc_list
) {
2066 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
2071 * We didn't find a corresponding entry in the table, so return 0 so
2072 * that the buffer is NOT cancelled.
2074 ASSERT(!(flags
& XFS_BLF_CANCEL
));
2079 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2080 * otherwise return 0. If the buffer is actually a buffer cancel item
2081 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2082 * table and remove it from the table if this is the last reference.
2084 * We remove the cancel record from the table when we encounter its last
2085 * occurrence in the log so that if the same buffer is re-used again after its
2086 * last cancellation we actually replay the changes made at that point.
2089 xlog_check_buffer_cancelled(
2093 unsigned short flags
)
2095 struct xfs_buf_cancel
*bcp
;
2097 bcp
= xlog_peek_buffer_cancelled(log
, blkno
, len
, flags
);
2102 * We've go a match, so return 1 so that the recovery of this buffer
2103 * is cancelled. If this buffer is actually a buffer cancel log
2104 * item, then decrement the refcount on the one in the table and
2105 * remove it if this is the last reference.
2107 if (flags
& XFS_BLF_CANCEL
) {
2108 if (--bcp
->bc_refcount
== 0) {
2109 list_del(&bcp
->bc_list
);
2117 * Perform recovery for a buffer full of inodes. In these buffers, the only
2118 * data which should be recovered is that which corresponds to the
2119 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2120 * data for the inodes is always logged through the inodes themselves rather
2121 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2123 * The only time when buffers full of inodes are fully recovered is when the
2124 * buffer is full of newly allocated inodes. In this case the buffer will
2125 * not be marked as an inode buffer and so will be sent to
2126 * xlog_recover_do_reg_buffer() below during recovery.
2129 xlog_recover_do_inode_buffer(
2130 struct xfs_mount
*mp
,
2131 xlog_recover_item_t
*item
,
2133 xfs_buf_log_format_t
*buf_f
)
2139 int reg_buf_offset
= 0;
2140 int reg_buf_bytes
= 0;
2141 int next_unlinked_offset
;
2143 xfs_agino_t
*logged_nextp
;
2144 xfs_agino_t
*buffer_nextp
;
2146 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
2149 * Post recovery validation only works properly on CRC enabled
2152 if (xfs_sb_version_hascrc(&mp
->m_sb
))
2153 bp
->b_ops
= &xfs_inode_buf_ops
;
2155 inodes_per_buf
= BBTOB(bp
->b_io_length
) >> mp
->m_sb
.sb_inodelog
;
2156 for (i
= 0; i
< inodes_per_buf
; i
++) {
2157 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
2158 offsetof(xfs_dinode_t
, di_next_unlinked
);
2160 while (next_unlinked_offset
>=
2161 (reg_buf_offset
+ reg_buf_bytes
)) {
2163 * The next di_next_unlinked field is beyond
2164 * the current logged region. Find the next
2165 * logged region that contains or is beyond
2166 * the current di_next_unlinked field.
2169 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2170 buf_f
->blf_map_size
, bit
);
2173 * If there are no more logged regions in the
2174 * buffer, then we're done.
2179 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2180 buf_f
->blf_map_size
, bit
);
2182 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
2183 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
2188 * If the current logged region starts after the current
2189 * di_next_unlinked field, then move on to the next
2190 * di_next_unlinked field.
2192 if (next_unlinked_offset
< reg_buf_offset
)
2195 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
2196 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
2197 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <=
2198 BBTOB(bp
->b_io_length
));
2201 * The current logged region contains a copy of the
2202 * current di_next_unlinked field. Extract its value
2203 * and copy it to the buffer copy.
2205 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
2206 next_unlinked_offset
- reg_buf_offset
;
2207 if (unlikely(*logged_nextp
== 0)) {
2209 "Bad inode buffer log record (ptr = "PTR_FMT
", bp = "PTR_FMT
"). "
2210 "Trying to replay bad (0) inode di_next_unlinked field.",
2212 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2213 XFS_ERRLEVEL_LOW
, mp
);
2214 return -EFSCORRUPTED
;
2217 buffer_nextp
= xfs_buf_offset(bp
, next_unlinked_offset
);
2218 *buffer_nextp
= *logged_nextp
;
2221 * If necessary, recalculate the CRC in the on-disk inode. We
2222 * have to leave the inode in a consistent state for whoever
2225 xfs_dinode_calc_crc(mp
,
2226 xfs_buf_offset(bp
, i
* mp
->m_sb
.sb_inodesize
));
2234 * V5 filesystems know the age of the buffer on disk being recovered. We can
2235 * have newer objects on disk than we are replaying, and so for these cases we
2236 * don't want to replay the current change as that will make the buffer contents
2237 * temporarily invalid on disk.
2239 * The magic number might not match the buffer type we are going to recover
2240 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2241 * extract the LSN of the existing object in the buffer based on it's current
2242 * magic number. If we don't recognise the magic number in the buffer, then
2243 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2244 * so can recover the buffer.
2246 * Note: we cannot rely solely on magic number matches to determine that the
2247 * buffer has a valid LSN - we also need to verify that it belongs to this
2248 * filesystem, so we need to extract the object's LSN and compare it to that
2249 * which we read from the superblock. If the UUIDs don't match, then we've got a
2250 * stale metadata block from an old filesystem instance that we need to recover
2254 xlog_recover_get_buf_lsn(
2255 struct xfs_mount
*mp
,
2261 void *blk
= bp
->b_addr
;
2265 /* v4 filesystems always recover immediately */
2266 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2267 goto recover_immediately
;
2269 magic32
= be32_to_cpu(*(__be32
*)blk
);
2271 case XFS_ABTB_CRC_MAGIC
:
2272 case XFS_ABTC_CRC_MAGIC
:
2273 case XFS_ABTB_MAGIC
:
2274 case XFS_ABTC_MAGIC
:
2275 case XFS_RMAP_CRC_MAGIC
:
2276 case XFS_REFC_CRC_MAGIC
:
2277 case XFS_IBT_CRC_MAGIC
:
2278 case XFS_IBT_MAGIC
: {
2279 struct xfs_btree_block
*btb
= blk
;
2281 lsn
= be64_to_cpu(btb
->bb_u
.s
.bb_lsn
);
2282 uuid
= &btb
->bb_u
.s
.bb_uuid
;
2285 case XFS_BMAP_CRC_MAGIC
:
2286 case XFS_BMAP_MAGIC
: {
2287 struct xfs_btree_block
*btb
= blk
;
2289 lsn
= be64_to_cpu(btb
->bb_u
.l
.bb_lsn
);
2290 uuid
= &btb
->bb_u
.l
.bb_uuid
;
2294 lsn
= be64_to_cpu(((struct xfs_agf
*)blk
)->agf_lsn
);
2295 uuid
= &((struct xfs_agf
*)blk
)->agf_uuid
;
2297 case XFS_AGFL_MAGIC
:
2298 lsn
= be64_to_cpu(((struct xfs_agfl
*)blk
)->agfl_lsn
);
2299 uuid
= &((struct xfs_agfl
*)blk
)->agfl_uuid
;
2302 lsn
= be64_to_cpu(((struct xfs_agi
*)blk
)->agi_lsn
);
2303 uuid
= &((struct xfs_agi
*)blk
)->agi_uuid
;
2305 case XFS_SYMLINK_MAGIC
:
2306 lsn
= be64_to_cpu(((struct xfs_dsymlink_hdr
*)blk
)->sl_lsn
);
2307 uuid
= &((struct xfs_dsymlink_hdr
*)blk
)->sl_uuid
;
2309 case XFS_DIR3_BLOCK_MAGIC
:
2310 case XFS_DIR3_DATA_MAGIC
:
2311 case XFS_DIR3_FREE_MAGIC
:
2312 lsn
= be64_to_cpu(((struct xfs_dir3_blk_hdr
*)blk
)->lsn
);
2313 uuid
= &((struct xfs_dir3_blk_hdr
*)blk
)->uuid
;
2315 case XFS_ATTR3_RMT_MAGIC
:
2317 * Remote attr blocks are written synchronously, rather than
2318 * being logged. That means they do not contain a valid LSN
2319 * (i.e. transactionally ordered) in them, and hence any time we
2320 * see a buffer to replay over the top of a remote attribute
2321 * block we should simply do so.
2323 goto recover_immediately
;
2326 * superblock uuids are magic. We may or may not have a
2327 * sb_meta_uuid on disk, but it will be set in the in-core
2328 * superblock. We set the uuid pointer for verification
2329 * according to the superblock feature mask to ensure we check
2330 * the relevant UUID in the superblock.
2332 lsn
= be64_to_cpu(((struct xfs_dsb
*)blk
)->sb_lsn
);
2333 if (xfs_sb_version_hasmetauuid(&mp
->m_sb
))
2334 uuid
= &((struct xfs_dsb
*)blk
)->sb_meta_uuid
;
2336 uuid
= &((struct xfs_dsb
*)blk
)->sb_uuid
;
2342 if (lsn
!= (xfs_lsn_t
)-1) {
2343 if (!uuid_equal(&mp
->m_sb
.sb_meta_uuid
, uuid
))
2344 goto recover_immediately
;
2348 magicda
= be16_to_cpu(((struct xfs_da_blkinfo
*)blk
)->magic
);
2350 case XFS_DIR3_LEAF1_MAGIC
:
2351 case XFS_DIR3_LEAFN_MAGIC
:
2352 case XFS_DA3_NODE_MAGIC
:
2353 lsn
= be64_to_cpu(((struct xfs_da3_blkinfo
*)blk
)->lsn
);
2354 uuid
= &((struct xfs_da3_blkinfo
*)blk
)->uuid
;
2360 if (lsn
!= (xfs_lsn_t
)-1) {
2361 if (!uuid_equal(&mp
->m_sb
.sb_uuid
, uuid
))
2362 goto recover_immediately
;
2367 * We do individual object checks on dquot and inode buffers as they
2368 * have their own individual LSN records. Also, we could have a stale
2369 * buffer here, so we have to at least recognise these buffer types.
2371 * A notd complexity here is inode unlinked list processing - it logs
2372 * the inode directly in the buffer, but we don't know which inodes have
2373 * been modified, and there is no global buffer LSN. Hence we need to
2374 * recover all inode buffer types immediately. This problem will be
2375 * fixed by logical logging of the unlinked list modifications.
2377 magic16
= be16_to_cpu(*(__be16
*)blk
);
2379 case XFS_DQUOT_MAGIC
:
2380 case XFS_DINODE_MAGIC
:
2381 goto recover_immediately
;
2386 /* unknown buffer contents, recover immediately */
2388 recover_immediately
:
2389 return (xfs_lsn_t
)-1;
2394 * Validate the recovered buffer is of the correct type and attach the
2395 * appropriate buffer operations to them for writeback. Magic numbers are in a
2397 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2398 * the first 32 bits of the buffer (most blocks),
2399 * inside a struct xfs_da_blkinfo at the start of the buffer.
2402 xlog_recover_validate_buf_type(
2403 struct xfs_mount
*mp
,
2405 xfs_buf_log_format_t
*buf_f
,
2406 xfs_lsn_t current_lsn
)
2408 struct xfs_da_blkinfo
*info
= bp
->b_addr
;
2412 char *warnmsg
= NULL
;
2415 * We can only do post recovery validation on items on CRC enabled
2416 * fielsystems as we need to know when the buffer was written to be able
2417 * to determine if we should have replayed the item. If we replay old
2418 * metadata over a newer buffer, then it will enter a temporarily
2419 * inconsistent state resulting in verification failures. Hence for now
2420 * just avoid the verification stage for non-crc filesystems
2422 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
2425 magic32
= be32_to_cpu(*(__be32
*)bp
->b_addr
);
2426 magic16
= be16_to_cpu(*(__be16
*)bp
->b_addr
);
2427 magicda
= be16_to_cpu(info
->magic
);
2428 switch (xfs_blft_from_flags(buf_f
)) {
2429 case XFS_BLFT_BTREE_BUF
:
2431 case XFS_ABTB_CRC_MAGIC
:
2432 case XFS_ABTC_CRC_MAGIC
:
2433 case XFS_ABTB_MAGIC
:
2434 case XFS_ABTC_MAGIC
:
2435 bp
->b_ops
= &xfs_allocbt_buf_ops
;
2437 case XFS_IBT_CRC_MAGIC
:
2438 case XFS_FIBT_CRC_MAGIC
:
2440 case XFS_FIBT_MAGIC
:
2441 bp
->b_ops
= &xfs_inobt_buf_ops
;
2443 case XFS_BMAP_CRC_MAGIC
:
2444 case XFS_BMAP_MAGIC
:
2445 bp
->b_ops
= &xfs_bmbt_buf_ops
;
2447 case XFS_RMAP_CRC_MAGIC
:
2448 bp
->b_ops
= &xfs_rmapbt_buf_ops
;
2450 case XFS_REFC_CRC_MAGIC
:
2451 bp
->b_ops
= &xfs_refcountbt_buf_ops
;
2454 warnmsg
= "Bad btree block magic!";
2458 case XFS_BLFT_AGF_BUF
:
2459 if (magic32
!= XFS_AGF_MAGIC
) {
2460 warnmsg
= "Bad AGF block magic!";
2463 bp
->b_ops
= &xfs_agf_buf_ops
;
2465 case XFS_BLFT_AGFL_BUF
:
2466 if (magic32
!= XFS_AGFL_MAGIC
) {
2467 warnmsg
= "Bad AGFL block magic!";
2470 bp
->b_ops
= &xfs_agfl_buf_ops
;
2472 case XFS_BLFT_AGI_BUF
:
2473 if (magic32
!= XFS_AGI_MAGIC
) {
2474 warnmsg
= "Bad AGI block magic!";
2477 bp
->b_ops
= &xfs_agi_buf_ops
;
2479 case XFS_BLFT_UDQUOT_BUF
:
2480 case XFS_BLFT_PDQUOT_BUF
:
2481 case XFS_BLFT_GDQUOT_BUF
:
2482 #ifdef CONFIG_XFS_QUOTA
2483 if (magic16
!= XFS_DQUOT_MAGIC
) {
2484 warnmsg
= "Bad DQUOT block magic!";
2487 bp
->b_ops
= &xfs_dquot_buf_ops
;
2490 "Trying to recover dquots without QUOTA support built in!");
2494 case XFS_BLFT_DINO_BUF
:
2495 if (magic16
!= XFS_DINODE_MAGIC
) {
2496 warnmsg
= "Bad INODE block magic!";
2499 bp
->b_ops
= &xfs_inode_buf_ops
;
2501 case XFS_BLFT_SYMLINK_BUF
:
2502 if (magic32
!= XFS_SYMLINK_MAGIC
) {
2503 warnmsg
= "Bad symlink block magic!";
2506 bp
->b_ops
= &xfs_symlink_buf_ops
;
2508 case XFS_BLFT_DIR_BLOCK_BUF
:
2509 if (magic32
!= XFS_DIR2_BLOCK_MAGIC
&&
2510 magic32
!= XFS_DIR3_BLOCK_MAGIC
) {
2511 warnmsg
= "Bad dir block magic!";
2514 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
2516 case XFS_BLFT_DIR_DATA_BUF
:
2517 if (magic32
!= XFS_DIR2_DATA_MAGIC
&&
2518 magic32
!= XFS_DIR3_DATA_MAGIC
) {
2519 warnmsg
= "Bad dir data magic!";
2522 bp
->b_ops
= &xfs_dir3_data_buf_ops
;
2524 case XFS_BLFT_DIR_FREE_BUF
:
2525 if (magic32
!= XFS_DIR2_FREE_MAGIC
&&
2526 magic32
!= XFS_DIR3_FREE_MAGIC
) {
2527 warnmsg
= "Bad dir3 free magic!";
2530 bp
->b_ops
= &xfs_dir3_free_buf_ops
;
2532 case XFS_BLFT_DIR_LEAF1_BUF
:
2533 if (magicda
!= XFS_DIR2_LEAF1_MAGIC
&&
2534 magicda
!= XFS_DIR3_LEAF1_MAGIC
) {
2535 warnmsg
= "Bad dir leaf1 magic!";
2538 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
2540 case XFS_BLFT_DIR_LEAFN_BUF
:
2541 if (magicda
!= XFS_DIR2_LEAFN_MAGIC
&&
2542 magicda
!= XFS_DIR3_LEAFN_MAGIC
) {
2543 warnmsg
= "Bad dir leafn magic!";
2546 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
2548 case XFS_BLFT_DA_NODE_BUF
:
2549 if (magicda
!= XFS_DA_NODE_MAGIC
&&
2550 magicda
!= XFS_DA3_NODE_MAGIC
) {
2551 warnmsg
= "Bad da node magic!";
2554 bp
->b_ops
= &xfs_da3_node_buf_ops
;
2556 case XFS_BLFT_ATTR_LEAF_BUF
:
2557 if (magicda
!= XFS_ATTR_LEAF_MAGIC
&&
2558 magicda
!= XFS_ATTR3_LEAF_MAGIC
) {
2559 warnmsg
= "Bad attr leaf magic!";
2562 bp
->b_ops
= &xfs_attr3_leaf_buf_ops
;
2564 case XFS_BLFT_ATTR_RMT_BUF
:
2565 if (magic32
!= XFS_ATTR3_RMT_MAGIC
) {
2566 warnmsg
= "Bad attr remote magic!";
2569 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
2571 case XFS_BLFT_SB_BUF
:
2572 if (magic32
!= XFS_SB_MAGIC
) {
2573 warnmsg
= "Bad SB block magic!";
2576 bp
->b_ops
= &xfs_sb_buf_ops
;
2578 #ifdef CONFIG_XFS_RT
2579 case XFS_BLFT_RTBITMAP_BUF
:
2580 case XFS_BLFT_RTSUMMARY_BUF
:
2581 /* no magic numbers for verification of RT buffers */
2582 bp
->b_ops
= &xfs_rtbuf_ops
;
2584 #endif /* CONFIG_XFS_RT */
2586 xfs_warn(mp
, "Unknown buffer type %d!",
2587 xfs_blft_from_flags(buf_f
));
2592 * Nothing else to do in the case of a NULL current LSN as this means
2593 * the buffer is more recent than the change in the log and will be
2596 if (current_lsn
== NULLCOMMITLSN
)
2600 xfs_warn(mp
, warnmsg
);
2605 * We must update the metadata LSN of the buffer as it is written out to
2606 * ensure that older transactions never replay over this one and corrupt
2607 * the buffer. This can occur if log recovery is interrupted at some
2608 * point after the current transaction completes, at which point a
2609 * subsequent mount starts recovery from the beginning.
2611 * Write verifiers update the metadata LSN from log items attached to
2612 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2613 * the verifier. We'll clean it up in our ->iodone() callback.
2616 struct xfs_buf_log_item
*bip
;
2618 ASSERT(!bp
->b_iodone
|| bp
->b_iodone
== xlog_recover_iodone
);
2619 bp
->b_iodone
= xlog_recover_iodone
;
2620 xfs_buf_item_init(bp
, mp
);
2621 bip
= bp
->b_log_item
;
2622 bip
->bli_item
.li_lsn
= current_lsn
;
2627 * Perform a 'normal' buffer recovery. Each logged region of the
2628 * buffer should be copied over the corresponding region in the
2629 * given buffer. The bitmap in the buf log format structure indicates
2630 * where to place the logged data.
2633 xlog_recover_do_reg_buffer(
2634 struct xfs_mount
*mp
,
2635 xlog_recover_item_t
*item
,
2637 xfs_buf_log_format_t
*buf_f
,
2638 xfs_lsn_t current_lsn
)
2645 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
2648 i
= 1; /* 0 is the buf format structure */
2650 bit
= xfs_next_bit(buf_f
->blf_data_map
,
2651 buf_f
->blf_map_size
, bit
);
2654 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
2655 buf_f
->blf_map_size
, bit
);
2657 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
2658 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
2659 ASSERT(BBTOB(bp
->b_io_length
) >=
2660 ((uint
)bit
<< XFS_BLF_SHIFT
) + (nbits
<< XFS_BLF_SHIFT
));
2663 * The dirty regions logged in the buffer, even though
2664 * contiguous, may span multiple chunks. This is because the
2665 * dirty region may span a physical page boundary in a buffer
2666 * and hence be split into two separate vectors for writing into
2667 * the log. Hence we need to trim nbits back to the length of
2668 * the current region being copied out of the log.
2670 if (item
->ri_buf
[i
].i_len
< (nbits
<< XFS_BLF_SHIFT
))
2671 nbits
= item
->ri_buf
[i
].i_len
>> XFS_BLF_SHIFT
;
2674 * Do a sanity check if this is a dquot buffer. Just checking
2675 * the first dquot in the buffer should do. XXXThis is
2676 * probably a good thing to do for other buf types also.
2679 if (buf_f
->blf_flags
&
2680 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2681 if (item
->ri_buf
[i
].i_addr
== NULL
) {
2683 "XFS: NULL dquot in %s.", __func__
);
2686 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
2688 "XFS: dquot too small (%d) in %s.",
2689 item
->ri_buf
[i
].i_len
, __func__
);
2692 fa
= xfs_dquot_verify(mp
, item
->ri_buf
[i
].i_addr
,
2696 "dquot corrupt at %pS trying to replay into block 0x%llx",
2702 memcpy(xfs_buf_offset(bp
,
2703 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
2704 item
->ri_buf
[i
].i_addr
, /* source */
2705 nbits
<<XFS_BLF_SHIFT
); /* length */
2711 /* Shouldn't be any more regions */
2712 ASSERT(i
== item
->ri_total
);
2714 xlog_recover_validate_buf_type(mp
, bp
, buf_f
, current_lsn
);
2718 * Perform a dquot buffer recovery.
2719 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2720 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2721 * Else, treat it as a regular buffer and do recovery.
2723 * Return false if the buffer was tossed and true if we recovered the buffer to
2724 * indicate to the caller if the buffer needs writing.
2727 xlog_recover_do_dquot_buffer(
2728 struct xfs_mount
*mp
,
2730 struct xlog_recover_item
*item
,
2732 struct xfs_buf_log_format
*buf_f
)
2736 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2739 * Filesystems are required to send in quota flags at mount time.
2745 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2746 type
|= XFS_DQ_USER
;
2747 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2748 type
|= XFS_DQ_PROJ
;
2749 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2750 type
|= XFS_DQ_GROUP
;
2752 * This type of quotas was turned off, so ignore this buffer
2754 if (log
->l_quotaoffs_flag
& type
)
2757 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
, NULLCOMMITLSN
);
2762 * This routine replays a modification made to a buffer at runtime.
2763 * There are actually two types of buffer, regular and inode, which
2764 * are handled differently. Inode buffers are handled differently
2765 * in that we only recover a specific set of data from them, namely
2766 * the inode di_next_unlinked fields. This is because all other inode
2767 * data is actually logged via inode records and any data we replay
2768 * here which overlaps that may be stale.
2770 * When meta-data buffers are freed at run time we log a buffer item
2771 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2772 * of the buffer in the log should not be replayed at recovery time.
2773 * This is so that if the blocks covered by the buffer are reused for
2774 * file data before we crash we don't end up replaying old, freed
2775 * meta-data into a user's file.
2777 * To handle the cancellation of buffer log items, we make two passes
2778 * over the log during recovery. During the first we build a table of
2779 * those buffers which have been cancelled, and during the second we
2780 * only replay those buffers which do not have corresponding cancel
2781 * records in the table. See xlog_recover_buffer_pass[1,2] above
2782 * for more details on the implementation of the table of cancel records.
2785 xlog_recover_buffer_pass2(
2787 struct list_head
*buffer_list
,
2788 struct xlog_recover_item
*item
,
2789 xfs_lsn_t current_lsn
)
2791 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2792 xfs_mount_t
*mp
= log
->l_mp
;
2799 * In this pass we only want to recover all the buffers which have
2800 * not been cancelled and are not cancellation buffers themselves.
2802 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2803 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2804 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2808 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2811 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
)
2812 buf_flags
|= XBF_UNMAPPED
;
2814 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2818 error
= bp
->b_error
;
2820 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#1)");
2825 * Recover the buffer only if we get an LSN from it and it's less than
2826 * the lsn of the transaction we are replaying.
2828 * Note that we have to be extremely careful of readahead here.
2829 * Readahead does not attach verfiers to the buffers so if we don't
2830 * actually do any replay after readahead because of the LSN we found
2831 * in the buffer if more recent than that current transaction then we
2832 * need to attach the verifier directly. Failure to do so can lead to
2833 * future recovery actions (e.g. EFI and unlinked list recovery) can
2834 * operate on the buffers and they won't get the verifier attached. This
2835 * can lead to blocks on disk having the correct content but a stale
2838 * It is safe to assume these clean buffers are currently up to date.
2839 * If the buffer is dirtied by a later transaction being replayed, then
2840 * the verifier will be reset to match whatever recover turns that
2843 lsn
= xlog_recover_get_buf_lsn(mp
, bp
);
2844 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
2845 trace_xfs_log_recover_buf_skip(log
, buf_f
);
2846 xlog_recover_validate_buf_type(mp
, bp
, buf_f
, NULLCOMMITLSN
);
2850 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2851 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2854 } else if (buf_f
->blf_flags
&
2855 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2858 dirty
= xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2862 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
, current_lsn
);
2866 * Perform delayed write on the buffer. Asynchronous writes will be
2867 * slower when taking into account all the buffers to be flushed.
2869 * Also make sure that only inode buffers with good sizes stay in
2870 * the buffer cache. The kernel moves inodes in buffers of 1 block
2871 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2872 * buffers in the log can be a different size if the log was generated
2873 * by an older kernel using unclustered inode buffers or a newer kernel
2874 * running with a different inode cluster size. Regardless, if the
2875 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2876 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2877 * the buffer out of the buffer cache so that the buffer won't
2878 * overlap with future reads of those inodes.
2880 if (XFS_DINODE_MAGIC
==
2881 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2882 (BBTOB(bp
->b_io_length
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2883 (uint32_t)log
->l_mp
->m_inode_cluster_size
))) {
2885 error
= xfs_bwrite(bp
);
2887 ASSERT(bp
->b_target
->bt_mount
== mp
);
2888 bp
->b_iodone
= xlog_recover_iodone
;
2889 xfs_buf_delwri_queue(bp
, buffer_list
);
2898 * Inode fork owner changes
2900 * If we have been told that we have to reparent the inode fork, it's because an
2901 * extent swap operation on a CRC enabled filesystem has been done and we are
2902 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2905 * The complexity here is that we don't have an inode context to work with, so
2906 * after we've replayed the inode we need to instantiate one. This is where the
2909 * We are in the middle of log recovery, so we can't run transactions. That
2910 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2911 * that will result in the corresponding iput() running the inode through
2912 * xfs_inactive(). If we've just replayed an inode core that changes the link
2913 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2914 * transactions (bad!).
2916 * So, to avoid this, we instantiate an inode directly from the inode core we've
2917 * just recovered. We have the buffer still locked, and all we really need to
2918 * instantiate is the inode core and the forks being modified. We can do this
2919 * manually, then run the inode btree owner change, and then tear down the
2920 * xfs_inode without having to run any transactions at all.
2922 * Also, because we don't have a transaction context available here but need to
2923 * gather all the buffers we modify for writeback so we pass the buffer_list
2924 * instead for the operation to use.
2928 xfs_recover_inode_owner_change(
2929 struct xfs_mount
*mp
,
2930 struct xfs_dinode
*dip
,
2931 struct xfs_inode_log_format
*in_f
,
2932 struct list_head
*buffer_list
)
2934 struct xfs_inode
*ip
;
2937 ASSERT(in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
));
2939 ip
= xfs_inode_alloc(mp
, in_f
->ilf_ino
);
2943 /* instantiate the inode */
2944 xfs_inode_from_disk(ip
, dip
);
2945 ASSERT(ip
->i_d
.di_version
>= 3);
2947 error
= xfs_iformat_fork(ip
, dip
);
2951 if (!xfs_inode_verify_forks(ip
)) {
2952 error
= -EFSCORRUPTED
;
2956 if (in_f
->ilf_fields
& XFS_ILOG_DOWNER
) {
2957 ASSERT(in_f
->ilf_fields
& XFS_ILOG_DBROOT
);
2958 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_DATA_FORK
,
2959 ip
->i_ino
, buffer_list
);
2964 if (in_f
->ilf_fields
& XFS_ILOG_AOWNER
) {
2965 ASSERT(in_f
->ilf_fields
& XFS_ILOG_ABROOT
);
2966 error
= xfs_bmbt_change_owner(NULL
, ip
, XFS_ATTR_FORK
,
2967 ip
->i_ino
, buffer_list
);
2978 xlog_recover_inode_pass2(
2980 struct list_head
*buffer_list
,
2981 struct xlog_recover_item
*item
,
2982 xfs_lsn_t current_lsn
)
2984 struct xfs_inode_log_format
*in_f
;
2985 xfs_mount_t
*mp
= log
->l_mp
;
2994 struct xfs_log_dinode
*ldip
;
2998 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
2999 in_f
= item
->ri_buf
[0].i_addr
;
3001 in_f
= kmem_alloc(sizeof(struct xfs_inode_log_format
), KM_SLEEP
);
3003 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
3009 * Inode buffers can be freed, look out for it,
3010 * and do not replay the inode.
3012 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
3013 in_f
->ilf_len
, 0)) {
3015 trace_xfs_log_recover_inode_cancel(log
, in_f
);
3018 trace_xfs_log_recover_inode_recover(log
, in_f
);
3020 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
, 0,
3021 &xfs_inode_buf_ops
);
3026 error
= bp
->b_error
;
3028 xfs_buf_ioerror_alert(bp
, "xlog_recover_do..(read#2)");
3031 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
3032 dip
= xfs_buf_offset(bp
, in_f
->ilf_boffset
);
3035 * Make sure the place we're flushing out to really looks
3038 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
3040 "%s: Bad inode magic number, dip = "PTR_FMT
", dino bp = "PTR_FMT
", ino = %Ld",
3041 __func__
, dip
, bp
, in_f
->ilf_ino
);
3042 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
3043 XFS_ERRLEVEL_LOW
, mp
);
3044 error
= -EFSCORRUPTED
;
3047 ldip
= item
->ri_buf
[1].i_addr
;
3048 if (unlikely(ldip
->di_magic
!= XFS_DINODE_MAGIC
)) {
3050 "%s: Bad inode log record, rec ptr "PTR_FMT
", ino %Ld",
3051 __func__
, item
, in_f
->ilf_ino
);
3052 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
3053 XFS_ERRLEVEL_LOW
, mp
);
3054 error
= -EFSCORRUPTED
;
3059 * If the inode has an LSN in it, recover the inode only if it's less
3060 * than the lsn of the transaction we are replaying. Note: we still
3061 * need to replay an owner change even though the inode is more recent
3062 * than the transaction as there is no guarantee that all the btree
3063 * blocks are more recent than this transaction, too.
3065 if (dip
->di_version
>= 3) {
3066 xfs_lsn_t lsn
= be64_to_cpu(dip
->di_lsn
);
3068 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3069 trace_xfs_log_recover_inode_skip(log
, in_f
);
3071 goto out_owner_change
;
3076 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3077 * are transactional and if ordering is necessary we can determine that
3078 * more accurately by the LSN field in the V3 inode core. Don't trust
3079 * the inode versions we might be changing them here - use the
3080 * superblock flag to determine whether we need to look at di_flushiter
3081 * to skip replay when the on disk inode is newer than the log one
3083 if (!xfs_sb_version_hascrc(&mp
->m_sb
) &&
3084 ldip
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
3086 * Deal with the wrap case, DI_MAX_FLUSH is less
3087 * than smaller numbers
3089 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
3090 ldip
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
3093 trace_xfs_log_recover_inode_skip(log
, in_f
);
3099 /* Take the opportunity to reset the flush iteration count */
3100 ldip
->di_flushiter
= 0;
3102 if (unlikely(S_ISREG(ldip
->di_mode
))) {
3103 if ((ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3104 (ldip
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
3105 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3106 XFS_ERRLEVEL_LOW
, mp
, ldip
,
3109 "%s: Bad regular inode log record, rec ptr "PTR_FMT
", "
3110 "ino ptr = "PTR_FMT
", ino bp = "PTR_FMT
", ino %Ld",
3111 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
3112 error
= -EFSCORRUPTED
;
3115 } else if (unlikely(S_ISDIR(ldip
->di_mode
))) {
3116 if ((ldip
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
3117 (ldip
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
3118 (ldip
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
3119 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3120 XFS_ERRLEVEL_LOW
, mp
, ldip
,
3123 "%s: Bad dir inode log record, rec ptr "PTR_FMT
", "
3124 "ino ptr = "PTR_FMT
", ino bp = "PTR_FMT
", ino %Ld",
3125 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
3126 error
= -EFSCORRUPTED
;
3130 if (unlikely(ldip
->di_nextents
+ ldip
->di_anextents
> ldip
->di_nblocks
)){
3131 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3132 XFS_ERRLEVEL_LOW
, mp
, ldip
,
3135 "%s: Bad inode log record, rec ptr "PTR_FMT
", dino ptr "PTR_FMT
", "
3136 "dino bp "PTR_FMT
", ino %Ld, total extents = %d, nblocks = %Ld",
3137 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
3138 ldip
->di_nextents
+ ldip
->di_anextents
,
3140 error
= -EFSCORRUPTED
;
3143 if (unlikely(ldip
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
3144 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3145 XFS_ERRLEVEL_LOW
, mp
, ldip
,
3148 "%s: Bad inode log record, rec ptr "PTR_FMT
", dino ptr "PTR_FMT
", "
3149 "dino bp "PTR_FMT
", ino %Ld, forkoff 0x%x", __func__
,
3150 item
, dip
, bp
, in_f
->ilf_ino
, ldip
->di_forkoff
);
3151 error
= -EFSCORRUPTED
;
3154 isize
= xfs_log_dinode_size(ldip
->di_version
);
3155 if (unlikely(item
->ri_buf
[1].i_len
> isize
)) {
3156 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3157 XFS_ERRLEVEL_LOW
, mp
, ldip
,
3160 "%s: Bad inode log record length %d, rec ptr "PTR_FMT
,
3161 __func__
, item
->ri_buf
[1].i_len
, item
);
3162 error
= -EFSCORRUPTED
;
3166 /* recover the log dinode inode into the on disk inode */
3167 xfs_log_dinode_to_disk(ldip
, dip
);
3169 fields
= in_f
->ilf_fields
;
3170 if (fields
& XFS_ILOG_DEV
)
3171 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
3173 if (in_f
->ilf_size
== 2)
3174 goto out_owner_change
;
3175 len
= item
->ri_buf
[2].i_len
;
3176 src
= item
->ri_buf
[2].i_addr
;
3177 ASSERT(in_f
->ilf_size
<= 4);
3178 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
3179 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
3180 (len
== in_f
->ilf_dsize
));
3182 switch (fields
& XFS_ILOG_DFORK
) {
3183 case XFS_ILOG_DDATA
:
3185 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
3188 case XFS_ILOG_DBROOT
:
3189 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
3190 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
3191 XFS_DFORK_DSIZE(dip
, mp
));
3196 * There are no data fork flags set.
3198 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
3203 * If we logged any attribute data, recover it. There may or
3204 * may not have been any other non-core data logged in this
3207 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3208 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
3213 len
= item
->ri_buf
[attr_index
].i_len
;
3214 src
= item
->ri_buf
[attr_index
].i_addr
;
3215 ASSERT(len
== in_f
->ilf_asize
);
3217 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
3218 case XFS_ILOG_ADATA
:
3220 dest
= XFS_DFORK_APTR(dip
);
3221 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
3222 memcpy(dest
, src
, len
);
3225 case XFS_ILOG_ABROOT
:
3226 dest
= XFS_DFORK_APTR(dip
);
3227 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
3228 len
, (xfs_bmdr_block_t
*)dest
,
3229 XFS_DFORK_ASIZE(dip
, mp
));
3233 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
3241 /* Recover the swapext owner change unless inode has been deleted */
3242 if ((in_f
->ilf_fields
& (XFS_ILOG_DOWNER
|XFS_ILOG_AOWNER
)) &&
3243 (dip
->di_mode
!= 0))
3244 error
= xfs_recover_inode_owner_change(mp
, dip
, in_f
,
3246 /* re-generate the checksum. */
3247 xfs_dinode_calc_crc(log
->l_mp
, dip
);
3249 ASSERT(bp
->b_target
->bt_mount
== mp
);
3250 bp
->b_iodone
= xlog_recover_iodone
;
3251 xfs_buf_delwri_queue(bp
, buffer_list
);
3262 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3263 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3267 xlog_recover_quotaoff_pass1(
3269 struct xlog_recover_item
*item
)
3271 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
3275 * The logitem format's flag tells us if this was user quotaoff,
3276 * group/project quotaoff or both.
3278 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
3279 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
3280 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
3281 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
3282 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
3283 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
3289 * Recover a dquot record
3292 xlog_recover_dquot_pass2(
3294 struct list_head
*buffer_list
,
3295 struct xlog_recover_item
*item
,
3296 xfs_lsn_t current_lsn
)
3298 xfs_mount_t
*mp
= log
->l_mp
;
3300 struct xfs_disk_dquot
*ddq
, *recddq
;
3303 xfs_dq_logformat_t
*dq_f
;
3308 * Filesystems are required to send in quota flags at mount time.
3310 if (mp
->m_qflags
== 0)
3313 recddq
= item
->ri_buf
[1].i_addr
;
3314 if (recddq
== NULL
) {
3315 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
3318 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
3319 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
3320 item
->ri_buf
[1].i_len
, __func__
);
3325 * This type of quotas was turned off, so ignore this record.
3327 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
3329 if (log
->l_quotaoffs_flag
& type
)
3333 * At this point we know that quota was _not_ turned off.
3334 * Since the mount flags are not indicating to us otherwise, this
3335 * must mean that quota is on, and the dquot needs to be replayed.
3336 * Remember that we may not have fully recovered the superblock yet,
3337 * so we can't do the usual trick of looking at the SB quota bits.
3339 * The other possibility, of course, is that the quota subsystem was
3340 * removed since the last mount - ENOSYS.
3342 dq_f
= item
->ri_buf
[0].i_addr
;
3344 fa
= xfs_dquot_verify(mp
, recddq
, dq_f
->qlf_id
, 0);
3346 xfs_alert(mp
, "corrupt dquot ID 0x%x in log at %pS",
3350 ASSERT(dq_f
->qlf_len
== 1);
3353 * At this point we are assuming that the dquots have been allocated
3354 * and hence the buffer has valid dquots stamped in it. It should,
3355 * therefore, pass verifier validation. If the dquot is bad, then the
3356 * we'll return an error here, so we don't need to specifically check
3357 * the dquot in the buffer after the verifier has run.
3359 error
= xfs_trans_read_buf(mp
, NULL
, mp
->m_ddev_targp
, dq_f
->qlf_blkno
,
3360 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
), 0, &bp
,
3361 &xfs_dquot_buf_ops
);
3366 ddq
= xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
3369 * If the dquot has an LSN in it, recover the dquot only if it's less
3370 * than the lsn of the transaction we are replaying.
3372 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3373 struct xfs_dqblk
*dqb
= (struct xfs_dqblk
*)ddq
;
3374 xfs_lsn_t lsn
= be64_to_cpu(dqb
->dd_lsn
);
3376 if (lsn
&& lsn
!= -1 && XFS_LSN_CMP(lsn
, current_lsn
) >= 0) {
3381 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
3382 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
3383 xfs_update_cksum((char *)ddq
, sizeof(struct xfs_dqblk
),
3387 ASSERT(dq_f
->qlf_size
== 2);
3388 ASSERT(bp
->b_target
->bt_mount
== mp
);
3389 bp
->b_iodone
= xlog_recover_iodone
;
3390 xfs_buf_delwri_queue(bp
, buffer_list
);
3398 * This routine is called to create an in-core extent free intent
3399 * item from the efi format structure which was logged on disk.
3400 * It allocates an in-core efi, copies the extents from the format
3401 * structure into it, and adds the efi to the AIL with the given
3405 xlog_recover_efi_pass2(
3407 struct xlog_recover_item
*item
,
3411 struct xfs_mount
*mp
= log
->l_mp
;
3412 struct xfs_efi_log_item
*efip
;
3413 struct xfs_efi_log_format
*efi_formatp
;
3415 efi_formatp
= item
->ri_buf
[0].i_addr
;
3417 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
3418 error
= xfs_efi_copy_format(&item
->ri_buf
[0], &efip
->efi_format
);
3420 xfs_efi_item_free(efip
);
3423 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
3425 spin_lock(&log
->l_ailp
->ail_lock
);
3427 * The EFI has two references. One for the EFD and one for EFI to ensure
3428 * it makes it into the AIL. Insert the EFI into the AIL directly and
3429 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3432 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
3433 xfs_efi_release(efip
);
3439 * This routine is called when an EFD format structure is found in a committed
3440 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3441 * was still in the log. To do this it searches the AIL for the EFI with an id
3442 * equal to that in the EFD format structure. If we find it we drop the EFD
3443 * reference, which removes the EFI from the AIL and frees it.
3446 xlog_recover_efd_pass2(
3448 struct xlog_recover_item
*item
)
3450 xfs_efd_log_format_t
*efd_formatp
;
3451 xfs_efi_log_item_t
*efip
= NULL
;
3452 xfs_log_item_t
*lip
;
3454 struct xfs_ail_cursor cur
;
3455 struct xfs_ail
*ailp
= log
->l_ailp
;
3457 efd_formatp
= item
->ri_buf
[0].i_addr
;
3458 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
3459 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
3460 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
3461 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
3462 efi_id
= efd_formatp
->efd_efi_id
;
3465 * Search for the EFI with the id in the EFD format structure in the
3468 spin_lock(&ailp
->ail_lock
);
3469 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3470 while (lip
!= NULL
) {
3471 if (lip
->li_type
== XFS_LI_EFI
) {
3472 efip
= (xfs_efi_log_item_t
*)lip
;
3473 if (efip
->efi_format
.efi_id
== efi_id
) {
3475 * Drop the EFD reference to the EFI. This
3476 * removes the EFI from the AIL and frees it.
3478 spin_unlock(&ailp
->ail_lock
);
3479 xfs_efi_release(efip
);
3480 spin_lock(&ailp
->ail_lock
);
3484 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3487 xfs_trans_ail_cursor_done(&cur
);
3488 spin_unlock(&ailp
->ail_lock
);
3494 * This routine is called to create an in-core extent rmap update
3495 * item from the rui format structure which was logged on disk.
3496 * It allocates an in-core rui, copies the extents from the format
3497 * structure into it, and adds the rui to the AIL with the given
3501 xlog_recover_rui_pass2(
3503 struct xlog_recover_item
*item
,
3507 struct xfs_mount
*mp
= log
->l_mp
;
3508 struct xfs_rui_log_item
*ruip
;
3509 struct xfs_rui_log_format
*rui_formatp
;
3511 rui_formatp
= item
->ri_buf
[0].i_addr
;
3513 ruip
= xfs_rui_init(mp
, rui_formatp
->rui_nextents
);
3514 error
= xfs_rui_copy_format(&item
->ri_buf
[0], &ruip
->rui_format
);
3516 xfs_rui_item_free(ruip
);
3519 atomic_set(&ruip
->rui_next_extent
, rui_formatp
->rui_nextents
);
3521 spin_lock(&log
->l_ailp
->ail_lock
);
3523 * The RUI has two references. One for the RUD and one for RUI to ensure
3524 * it makes it into the AIL. Insert the RUI into the AIL directly and
3525 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3528 xfs_trans_ail_update(log
->l_ailp
, &ruip
->rui_item
, lsn
);
3529 xfs_rui_release(ruip
);
3535 * This routine is called when an RUD format structure is found in a committed
3536 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3537 * was still in the log. To do this it searches the AIL for the RUI with an id
3538 * equal to that in the RUD format structure. If we find it we drop the RUD
3539 * reference, which removes the RUI from the AIL and frees it.
3542 xlog_recover_rud_pass2(
3544 struct xlog_recover_item
*item
)
3546 struct xfs_rud_log_format
*rud_formatp
;
3547 struct xfs_rui_log_item
*ruip
= NULL
;
3548 struct xfs_log_item
*lip
;
3550 struct xfs_ail_cursor cur
;
3551 struct xfs_ail
*ailp
= log
->l_ailp
;
3553 rud_formatp
= item
->ri_buf
[0].i_addr
;
3554 ASSERT(item
->ri_buf
[0].i_len
== sizeof(struct xfs_rud_log_format
));
3555 rui_id
= rud_formatp
->rud_rui_id
;
3558 * Search for the RUI with the id in the RUD format structure in the
3561 spin_lock(&ailp
->ail_lock
);
3562 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3563 while (lip
!= NULL
) {
3564 if (lip
->li_type
== XFS_LI_RUI
) {
3565 ruip
= (struct xfs_rui_log_item
*)lip
;
3566 if (ruip
->rui_format
.rui_id
== rui_id
) {
3568 * Drop the RUD reference to the RUI. This
3569 * removes the RUI from the AIL and frees it.
3571 spin_unlock(&ailp
->ail_lock
);
3572 xfs_rui_release(ruip
);
3573 spin_lock(&ailp
->ail_lock
);
3577 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3580 xfs_trans_ail_cursor_done(&cur
);
3581 spin_unlock(&ailp
->ail_lock
);
3587 * Copy an CUI format buffer from the given buf, and into the destination
3588 * CUI format structure. The CUI/CUD items were designed not to need any
3589 * special alignment handling.
3592 xfs_cui_copy_format(
3593 struct xfs_log_iovec
*buf
,
3594 struct xfs_cui_log_format
*dst_cui_fmt
)
3596 struct xfs_cui_log_format
*src_cui_fmt
;
3599 src_cui_fmt
= buf
->i_addr
;
3600 len
= xfs_cui_log_format_sizeof(src_cui_fmt
->cui_nextents
);
3602 if (buf
->i_len
== len
) {
3603 memcpy(dst_cui_fmt
, src_cui_fmt
, len
);
3606 return -EFSCORRUPTED
;
3610 * This routine is called to create an in-core extent refcount update
3611 * item from the cui format structure which was logged on disk.
3612 * It allocates an in-core cui, copies the extents from the format
3613 * structure into it, and adds the cui to the AIL with the given
3617 xlog_recover_cui_pass2(
3619 struct xlog_recover_item
*item
,
3623 struct xfs_mount
*mp
= log
->l_mp
;
3624 struct xfs_cui_log_item
*cuip
;
3625 struct xfs_cui_log_format
*cui_formatp
;
3627 cui_formatp
= item
->ri_buf
[0].i_addr
;
3629 cuip
= xfs_cui_init(mp
, cui_formatp
->cui_nextents
);
3630 error
= xfs_cui_copy_format(&item
->ri_buf
[0], &cuip
->cui_format
);
3632 xfs_cui_item_free(cuip
);
3635 atomic_set(&cuip
->cui_next_extent
, cui_formatp
->cui_nextents
);
3637 spin_lock(&log
->l_ailp
->ail_lock
);
3639 * The CUI has two references. One for the CUD and one for CUI to ensure
3640 * it makes it into the AIL. Insert the CUI into the AIL directly and
3641 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3644 xfs_trans_ail_update(log
->l_ailp
, &cuip
->cui_item
, lsn
);
3645 xfs_cui_release(cuip
);
3651 * This routine is called when an CUD format structure is found in a committed
3652 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3653 * was still in the log. To do this it searches the AIL for the CUI with an id
3654 * equal to that in the CUD format structure. If we find it we drop the CUD
3655 * reference, which removes the CUI from the AIL and frees it.
3658 xlog_recover_cud_pass2(
3660 struct xlog_recover_item
*item
)
3662 struct xfs_cud_log_format
*cud_formatp
;
3663 struct xfs_cui_log_item
*cuip
= NULL
;
3664 struct xfs_log_item
*lip
;
3666 struct xfs_ail_cursor cur
;
3667 struct xfs_ail
*ailp
= log
->l_ailp
;
3669 cud_formatp
= item
->ri_buf
[0].i_addr
;
3670 if (item
->ri_buf
[0].i_len
!= sizeof(struct xfs_cud_log_format
))
3671 return -EFSCORRUPTED
;
3672 cui_id
= cud_formatp
->cud_cui_id
;
3675 * Search for the CUI with the id in the CUD format structure in the
3678 spin_lock(&ailp
->ail_lock
);
3679 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3680 while (lip
!= NULL
) {
3681 if (lip
->li_type
== XFS_LI_CUI
) {
3682 cuip
= (struct xfs_cui_log_item
*)lip
;
3683 if (cuip
->cui_format
.cui_id
== cui_id
) {
3685 * Drop the CUD reference to the CUI. This
3686 * removes the CUI from the AIL and frees it.
3688 spin_unlock(&ailp
->ail_lock
);
3689 xfs_cui_release(cuip
);
3690 spin_lock(&ailp
->ail_lock
);
3694 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3697 xfs_trans_ail_cursor_done(&cur
);
3698 spin_unlock(&ailp
->ail_lock
);
3704 * Copy an BUI format buffer from the given buf, and into the destination
3705 * BUI format structure. The BUI/BUD items were designed not to need any
3706 * special alignment handling.
3709 xfs_bui_copy_format(
3710 struct xfs_log_iovec
*buf
,
3711 struct xfs_bui_log_format
*dst_bui_fmt
)
3713 struct xfs_bui_log_format
*src_bui_fmt
;
3716 src_bui_fmt
= buf
->i_addr
;
3717 len
= xfs_bui_log_format_sizeof(src_bui_fmt
->bui_nextents
);
3719 if (buf
->i_len
== len
) {
3720 memcpy(dst_bui_fmt
, src_bui_fmt
, len
);
3723 return -EFSCORRUPTED
;
3727 * This routine is called to create an in-core extent bmap update
3728 * item from the bui format structure which was logged on disk.
3729 * It allocates an in-core bui, copies the extents from the format
3730 * structure into it, and adds the bui to the AIL with the given
3734 xlog_recover_bui_pass2(
3736 struct xlog_recover_item
*item
,
3740 struct xfs_mount
*mp
= log
->l_mp
;
3741 struct xfs_bui_log_item
*buip
;
3742 struct xfs_bui_log_format
*bui_formatp
;
3744 bui_formatp
= item
->ri_buf
[0].i_addr
;
3746 if (bui_formatp
->bui_nextents
!= XFS_BUI_MAX_FAST_EXTENTS
)
3747 return -EFSCORRUPTED
;
3748 buip
= xfs_bui_init(mp
);
3749 error
= xfs_bui_copy_format(&item
->ri_buf
[0], &buip
->bui_format
);
3751 xfs_bui_item_free(buip
);
3754 atomic_set(&buip
->bui_next_extent
, bui_formatp
->bui_nextents
);
3756 spin_lock(&log
->l_ailp
->ail_lock
);
3758 * The RUI has two references. One for the RUD and one for RUI to ensure
3759 * it makes it into the AIL. Insert the RUI into the AIL directly and
3760 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3763 xfs_trans_ail_update(log
->l_ailp
, &buip
->bui_item
, lsn
);
3764 xfs_bui_release(buip
);
3770 * This routine is called when an BUD format structure is found in a committed
3771 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3772 * was still in the log. To do this it searches the AIL for the BUI with an id
3773 * equal to that in the BUD format structure. If we find it we drop the BUD
3774 * reference, which removes the BUI from the AIL and frees it.
3777 xlog_recover_bud_pass2(
3779 struct xlog_recover_item
*item
)
3781 struct xfs_bud_log_format
*bud_formatp
;
3782 struct xfs_bui_log_item
*buip
= NULL
;
3783 struct xfs_log_item
*lip
;
3785 struct xfs_ail_cursor cur
;
3786 struct xfs_ail
*ailp
= log
->l_ailp
;
3788 bud_formatp
= item
->ri_buf
[0].i_addr
;
3789 if (item
->ri_buf
[0].i_len
!= sizeof(struct xfs_bud_log_format
))
3790 return -EFSCORRUPTED
;
3791 bui_id
= bud_formatp
->bud_bui_id
;
3794 * Search for the BUI with the id in the BUD format structure in the
3797 spin_lock(&ailp
->ail_lock
);
3798 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
3799 while (lip
!= NULL
) {
3800 if (lip
->li_type
== XFS_LI_BUI
) {
3801 buip
= (struct xfs_bui_log_item
*)lip
;
3802 if (buip
->bui_format
.bui_id
== bui_id
) {
3804 * Drop the BUD reference to the BUI. This
3805 * removes the BUI from the AIL and frees it.
3807 spin_unlock(&ailp
->ail_lock
);
3808 xfs_bui_release(buip
);
3809 spin_lock(&ailp
->ail_lock
);
3813 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3816 xfs_trans_ail_cursor_done(&cur
);
3817 spin_unlock(&ailp
->ail_lock
);
3823 * This routine is called when an inode create format structure is found in a
3824 * committed transaction in the log. It's purpose is to initialise the inodes
3825 * being allocated on disk. This requires us to get inode cluster buffers that
3826 * match the range to be initialised, stamped with inode templates and written
3827 * by delayed write so that subsequent modifications will hit the cached buffer
3828 * and only need writing out at the end of recovery.
3831 xlog_recover_do_icreate_pass2(
3833 struct list_head
*buffer_list
,
3834 xlog_recover_item_t
*item
)
3836 struct xfs_mount
*mp
= log
->l_mp
;
3837 struct xfs_icreate_log
*icl
;
3838 xfs_agnumber_t agno
;
3839 xfs_agblock_t agbno
;
3842 xfs_agblock_t length
;
3843 int blks_per_cluster
;
3849 icl
= (struct xfs_icreate_log
*)item
->ri_buf
[0].i_addr
;
3850 if (icl
->icl_type
!= XFS_LI_ICREATE
) {
3851 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad type");
3855 if (icl
->icl_size
!= 1) {
3856 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad icl size");
3860 agno
= be32_to_cpu(icl
->icl_ag
);
3861 if (agno
>= mp
->m_sb
.sb_agcount
) {
3862 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agno");
3865 agbno
= be32_to_cpu(icl
->icl_agbno
);
3866 if (!agbno
|| agbno
== NULLAGBLOCK
|| agbno
>= mp
->m_sb
.sb_agblocks
) {
3867 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad agbno");
3870 isize
= be32_to_cpu(icl
->icl_isize
);
3871 if (isize
!= mp
->m_sb
.sb_inodesize
) {
3872 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad isize");
3875 count
= be32_to_cpu(icl
->icl_count
);
3877 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad count");
3880 length
= be32_to_cpu(icl
->icl_length
);
3881 if (!length
|| length
>= mp
->m_sb
.sb_agblocks
) {
3882 xfs_warn(log
->l_mp
, "xlog_recover_do_icreate_trans: bad length");
3887 * The inode chunk is either full or sparse and we only support
3888 * m_ialloc_min_blks sized sparse allocations at this time.
3890 if (length
!= mp
->m_ialloc_blks
&&
3891 length
!= mp
->m_ialloc_min_blks
) {
3893 "%s: unsupported chunk length", __FUNCTION__
);
3897 /* verify inode count is consistent with extent length */
3898 if ((count
>> mp
->m_sb
.sb_inopblog
) != length
) {
3900 "%s: inconsistent inode count and chunk length",
3906 * The icreate transaction can cover multiple cluster buffers and these
3907 * buffers could have been freed and reused. Check the individual
3908 * buffers for cancellation so we don't overwrite anything written after
3911 blks_per_cluster
= xfs_icluster_size_fsb(mp
);
3912 bb_per_cluster
= XFS_FSB_TO_BB(mp
, blks_per_cluster
);
3913 nbufs
= length
/ blks_per_cluster
;
3914 for (i
= 0, cancel_count
= 0; i
< nbufs
; i
++) {
3917 daddr
= XFS_AGB_TO_DADDR(mp
, agno
,
3918 agbno
+ i
* blks_per_cluster
);
3919 if (xlog_check_buffer_cancelled(log
, daddr
, bb_per_cluster
, 0))
3924 * We currently only use icreate for a single allocation at a time. This
3925 * means we should expect either all or none of the buffers to be
3926 * cancelled. Be conservative and skip replay if at least one buffer is
3927 * cancelled, but warn the user that something is awry if the buffers
3928 * are not consistent.
3930 * XXX: This must be refined to only skip cancelled clusters once we use
3931 * icreate for multiple chunk allocations.
3933 ASSERT(!cancel_count
|| cancel_count
== nbufs
);
3935 if (cancel_count
!= nbufs
)
3937 "WARNING: partial inode chunk cancellation, skipped icreate.");
3938 trace_xfs_log_recover_icreate_cancel(log
, icl
);
3942 trace_xfs_log_recover_icreate_recover(log
, icl
);
3943 return xfs_ialloc_inode_init(mp
, NULL
, buffer_list
, count
, agno
, agbno
,
3944 length
, be32_to_cpu(icl
->icl_gen
));
3948 xlog_recover_buffer_ra_pass2(
3950 struct xlog_recover_item
*item
)
3952 struct xfs_buf_log_format
*buf_f
= item
->ri_buf
[0].i_addr
;
3953 struct xfs_mount
*mp
= log
->l_mp
;
3955 if (xlog_peek_buffer_cancelled(log
, buf_f
->blf_blkno
,
3956 buf_f
->blf_len
, buf_f
->blf_flags
)) {
3960 xfs_buf_readahead(mp
->m_ddev_targp
, buf_f
->blf_blkno
,
3961 buf_f
->blf_len
, NULL
);
3965 xlog_recover_inode_ra_pass2(
3967 struct xlog_recover_item
*item
)
3969 struct xfs_inode_log_format ilf_buf
;
3970 struct xfs_inode_log_format
*ilfp
;
3971 struct xfs_mount
*mp
= log
->l_mp
;
3974 if (item
->ri_buf
[0].i_len
== sizeof(struct xfs_inode_log_format
)) {
3975 ilfp
= item
->ri_buf
[0].i_addr
;
3978 memset(ilfp
, 0, sizeof(*ilfp
));
3979 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], ilfp
);
3984 if (xlog_peek_buffer_cancelled(log
, ilfp
->ilf_blkno
, ilfp
->ilf_len
, 0))
3987 xfs_buf_readahead(mp
->m_ddev_targp
, ilfp
->ilf_blkno
,
3988 ilfp
->ilf_len
, &xfs_inode_buf_ra_ops
);
3992 xlog_recover_dquot_ra_pass2(
3994 struct xlog_recover_item
*item
)
3996 struct xfs_mount
*mp
= log
->l_mp
;
3997 struct xfs_disk_dquot
*recddq
;
3998 struct xfs_dq_logformat
*dq_f
;
4003 if (mp
->m_qflags
== 0)
4006 recddq
= item
->ri_buf
[1].i_addr
;
4009 if (item
->ri_buf
[1].i_len
< sizeof(struct xfs_disk_dquot
))
4012 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
4014 if (log
->l_quotaoffs_flag
& type
)
4017 dq_f
= item
->ri_buf
[0].i_addr
;
4019 ASSERT(dq_f
->qlf_len
== 1);
4021 len
= XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
);
4022 if (xlog_peek_buffer_cancelled(log
, dq_f
->qlf_blkno
, len
, 0))
4025 xfs_buf_readahead(mp
->m_ddev_targp
, dq_f
->qlf_blkno
, len
,
4026 &xfs_dquot_buf_ra_ops
);
4030 xlog_recover_ra_pass2(
4032 struct xlog_recover_item
*item
)
4034 switch (ITEM_TYPE(item
)) {
4036 xlog_recover_buffer_ra_pass2(log
, item
);
4039 xlog_recover_inode_ra_pass2(log
, item
);
4042 xlog_recover_dquot_ra_pass2(log
, item
);
4046 case XFS_LI_QUOTAOFF
:
4059 xlog_recover_commit_pass1(
4061 struct xlog_recover
*trans
,
4062 struct xlog_recover_item
*item
)
4064 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
4066 switch (ITEM_TYPE(item
)) {
4068 return xlog_recover_buffer_pass1(log
, item
);
4069 case XFS_LI_QUOTAOFF
:
4070 return xlog_recover_quotaoff_pass1(log
, item
);
4075 case XFS_LI_ICREATE
:
4082 /* nothing to do in pass 1 */
4085 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
4086 __func__
, ITEM_TYPE(item
));
4093 xlog_recover_commit_pass2(
4095 struct xlog_recover
*trans
,
4096 struct list_head
*buffer_list
,
4097 struct xlog_recover_item
*item
)
4099 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
4101 switch (ITEM_TYPE(item
)) {
4103 return xlog_recover_buffer_pass2(log
, buffer_list
, item
,
4106 return xlog_recover_inode_pass2(log
, buffer_list
, item
,
4109 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
4111 return xlog_recover_efd_pass2(log
, item
);
4113 return xlog_recover_rui_pass2(log
, item
, trans
->r_lsn
);
4115 return xlog_recover_rud_pass2(log
, item
);
4117 return xlog_recover_cui_pass2(log
, item
, trans
->r_lsn
);
4119 return xlog_recover_cud_pass2(log
, item
);
4121 return xlog_recover_bui_pass2(log
, item
, trans
->r_lsn
);
4123 return xlog_recover_bud_pass2(log
, item
);
4125 return xlog_recover_dquot_pass2(log
, buffer_list
, item
,
4127 case XFS_LI_ICREATE
:
4128 return xlog_recover_do_icreate_pass2(log
, buffer_list
, item
);
4129 case XFS_LI_QUOTAOFF
:
4130 /* nothing to do in pass2 */
4133 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
4134 __func__
, ITEM_TYPE(item
));
4141 xlog_recover_items_pass2(
4143 struct xlog_recover
*trans
,
4144 struct list_head
*buffer_list
,
4145 struct list_head
*item_list
)
4147 struct xlog_recover_item
*item
;
4150 list_for_each_entry(item
, item_list
, ri_list
) {
4151 error
= xlog_recover_commit_pass2(log
, trans
,
4161 * Perform the transaction.
4163 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4164 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4167 xlog_recover_commit_trans(
4169 struct xlog_recover
*trans
,
4171 struct list_head
*buffer_list
)
4174 int items_queued
= 0;
4175 struct xlog_recover_item
*item
;
4176 struct xlog_recover_item
*next
;
4177 LIST_HEAD (ra_list
);
4178 LIST_HEAD (done_list
);
4180 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4182 hlist_del_init(&trans
->r_list
);
4184 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
4188 list_for_each_entry_safe(item
, next
, &trans
->r_itemq
, ri_list
) {
4190 case XLOG_RECOVER_PASS1
:
4191 error
= xlog_recover_commit_pass1(log
, trans
, item
);
4193 case XLOG_RECOVER_PASS2
:
4194 xlog_recover_ra_pass2(log
, item
);
4195 list_move_tail(&item
->ri_list
, &ra_list
);
4197 if (items_queued
>= XLOG_RECOVER_COMMIT_QUEUE_MAX
) {
4198 error
= xlog_recover_items_pass2(log
, trans
,
4199 buffer_list
, &ra_list
);
4200 list_splice_tail_init(&ra_list
, &done_list
);
4214 if (!list_empty(&ra_list
)) {
4216 error
= xlog_recover_items_pass2(log
, trans
,
4217 buffer_list
, &ra_list
);
4218 list_splice_tail_init(&ra_list
, &done_list
);
4221 if (!list_empty(&done_list
))
4222 list_splice_init(&done_list
, &trans
->r_itemq
);
4228 xlog_recover_add_item(
4229 struct list_head
*head
)
4231 xlog_recover_item_t
*item
;
4233 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
4234 INIT_LIST_HEAD(&item
->ri_list
);
4235 list_add_tail(&item
->ri_list
, head
);
4239 xlog_recover_add_to_cont_trans(
4241 struct xlog_recover
*trans
,
4245 xlog_recover_item_t
*item
;
4246 char *ptr
, *old_ptr
;
4250 * If the transaction is empty, the header was split across this and the
4251 * previous record. Copy the rest of the header.
4253 if (list_empty(&trans
->r_itemq
)) {
4254 ASSERT(len
<= sizeof(struct xfs_trans_header
));
4255 if (len
> sizeof(struct xfs_trans_header
)) {
4256 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
4260 xlog_recover_add_item(&trans
->r_itemq
);
4261 ptr
= (char *)&trans
->r_theader
+
4262 sizeof(struct xfs_trans_header
) - len
;
4263 memcpy(ptr
, dp
, len
);
4267 /* take the tail entry */
4268 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
4270 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
4271 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
4273 ptr
= kmem_realloc(old_ptr
, len
+ old_len
, KM_SLEEP
);
4274 memcpy(&ptr
[old_len
], dp
, len
);
4275 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
4276 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
4277 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
4282 * The next region to add is the start of a new region. It could be
4283 * a whole region or it could be the first part of a new region. Because
4284 * of this, the assumption here is that the type and size fields of all
4285 * format structures fit into the first 32 bits of the structure.
4287 * This works because all regions must be 32 bit aligned. Therefore, we
4288 * either have both fields or we have neither field. In the case we have
4289 * neither field, the data part of the region is zero length. We only have
4290 * a log_op_header and can throw away the header since a new one will appear
4291 * later. If we have at least 4 bytes, then we can determine how many regions
4292 * will appear in the current log item.
4295 xlog_recover_add_to_trans(
4297 struct xlog_recover
*trans
,
4301 struct xfs_inode_log_format
*in_f
; /* any will do */
4302 xlog_recover_item_t
*item
;
4307 if (list_empty(&trans
->r_itemq
)) {
4308 /* we need to catch log corruptions here */
4309 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
4310 xfs_warn(log
->l_mp
, "%s: bad header magic number",
4316 if (len
> sizeof(struct xfs_trans_header
)) {
4317 xfs_warn(log
->l_mp
, "%s: bad header length", __func__
);
4323 * The transaction header can be arbitrarily split across op
4324 * records. If we don't have the whole thing here, copy what we
4325 * do have and handle the rest in the next record.
4327 if (len
== sizeof(struct xfs_trans_header
))
4328 xlog_recover_add_item(&trans
->r_itemq
);
4329 memcpy(&trans
->r_theader
, dp
, len
);
4333 ptr
= kmem_alloc(len
, KM_SLEEP
);
4334 memcpy(ptr
, dp
, len
);
4335 in_f
= (struct xfs_inode_log_format
*)ptr
;
4337 /* take the tail entry */
4338 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
4339 if (item
->ri_total
!= 0 &&
4340 item
->ri_total
== item
->ri_cnt
) {
4341 /* tail item is in use, get a new one */
4342 xlog_recover_add_item(&trans
->r_itemq
);
4343 item
= list_entry(trans
->r_itemq
.prev
,
4344 xlog_recover_item_t
, ri_list
);
4347 if (item
->ri_total
== 0) { /* first region to be added */
4348 if (in_f
->ilf_size
== 0 ||
4349 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
4351 "bad number of regions (%d) in inode log format",
4358 item
->ri_total
= in_f
->ilf_size
;
4360 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
4363 ASSERT(item
->ri_total
> item
->ri_cnt
);
4364 /* Description region is ri_buf[0] */
4365 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
4366 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
4368 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
4373 * Free up any resources allocated by the transaction
4375 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4378 xlog_recover_free_trans(
4379 struct xlog_recover
*trans
)
4381 xlog_recover_item_t
*item
, *n
;
4384 hlist_del_init(&trans
->r_list
);
4386 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
4387 /* Free the regions in the item. */
4388 list_del(&item
->ri_list
);
4389 for (i
= 0; i
< item
->ri_cnt
; i
++)
4390 kmem_free(item
->ri_buf
[i
].i_addr
);
4391 /* Free the item itself */
4392 kmem_free(item
->ri_buf
);
4395 /* Free the transaction recover structure */
4400 * On error or completion, trans is freed.
4403 xlog_recovery_process_trans(
4405 struct xlog_recover
*trans
,
4410 struct list_head
*buffer_list
)
4413 bool freeit
= false;
4415 /* mask off ophdr transaction container flags */
4416 flags
&= ~XLOG_END_TRANS
;
4417 if (flags
& XLOG_WAS_CONT_TRANS
)
4418 flags
&= ~XLOG_CONTINUE_TRANS
;
4421 * Callees must not free the trans structure. We'll decide if we need to
4422 * free it or not based on the operation being done and it's result.
4425 /* expected flag values */
4427 case XLOG_CONTINUE_TRANS
:
4428 error
= xlog_recover_add_to_trans(log
, trans
, dp
, len
);
4430 case XLOG_WAS_CONT_TRANS
:
4431 error
= xlog_recover_add_to_cont_trans(log
, trans
, dp
, len
);
4433 case XLOG_COMMIT_TRANS
:
4434 error
= xlog_recover_commit_trans(log
, trans
, pass
,
4436 /* success or fail, we are now done with this transaction. */
4440 /* unexpected flag values */
4441 case XLOG_UNMOUNT_TRANS
:
4442 /* just skip trans */
4443 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
4446 case XLOG_START_TRANS
:
4448 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x", __func__
, flags
);
4453 if (error
|| freeit
)
4454 xlog_recover_free_trans(trans
);
4459 * Lookup the transaction recovery structure associated with the ID in the
4460 * current ophdr. If the transaction doesn't exist and the start flag is set in
4461 * the ophdr, then allocate a new transaction for future ID matches to find.
4462 * Either way, return what we found during the lookup - an existing transaction
4465 STATIC
struct xlog_recover
*
4466 xlog_recover_ophdr_to_trans(
4467 struct hlist_head rhash
[],
4468 struct xlog_rec_header
*rhead
,
4469 struct xlog_op_header
*ohead
)
4471 struct xlog_recover
*trans
;
4473 struct hlist_head
*rhp
;
4475 tid
= be32_to_cpu(ohead
->oh_tid
);
4476 rhp
= &rhash
[XLOG_RHASH(tid
)];
4477 hlist_for_each_entry(trans
, rhp
, r_list
) {
4478 if (trans
->r_log_tid
== tid
)
4483 * skip over non-start transaction headers - we could be
4484 * processing slack space before the next transaction starts
4486 if (!(ohead
->oh_flags
& XLOG_START_TRANS
))
4489 ASSERT(be32_to_cpu(ohead
->oh_len
) == 0);
4492 * This is a new transaction so allocate a new recovery container to
4493 * hold the recovery ops that will follow.
4495 trans
= kmem_zalloc(sizeof(struct xlog_recover
), KM_SLEEP
);
4496 trans
->r_log_tid
= tid
;
4497 trans
->r_lsn
= be64_to_cpu(rhead
->h_lsn
);
4498 INIT_LIST_HEAD(&trans
->r_itemq
);
4499 INIT_HLIST_NODE(&trans
->r_list
);
4500 hlist_add_head(&trans
->r_list
, rhp
);
4503 * Nothing more to do for this ophdr. Items to be added to this new
4504 * transaction will be in subsequent ophdr containers.
4510 xlog_recover_process_ophdr(
4512 struct hlist_head rhash
[],
4513 struct xlog_rec_header
*rhead
,
4514 struct xlog_op_header
*ohead
,
4518 struct list_head
*buffer_list
)
4520 struct xlog_recover
*trans
;
4524 /* Do we understand who wrote this op? */
4525 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
4526 ohead
->oh_clientid
!= XFS_LOG
) {
4527 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
4528 __func__
, ohead
->oh_clientid
);
4534 * Check the ophdr contains all the data it is supposed to contain.
4536 len
= be32_to_cpu(ohead
->oh_len
);
4537 if (dp
+ len
> end
) {
4538 xfs_warn(log
->l_mp
, "%s: bad length 0x%x", __func__
, len
);
4543 trans
= xlog_recover_ophdr_to_trans(rhash
, rhead
, ohead
);
4545 /* nothing to do, so skip over this ophdr */
4550 * The recovered buffer queue is drained only once we know that all
4551 * recovery items for the current LSN have been processed. This is
4554 * - Buffer write submission updates the metadata LSN of the buffer.
4555 * - Log recovery skips items with a metadata LSN >= the current LSN of
4556 * the recovery item.
4557 * - Separate recovery items against the same metadata buffer can share
4558 * a current LSN. I.e., consider that the LSN of a recovery item is
4559 * defined as the starting LSN of the first record in which its
4560 * transaction appears, that a record can hold multiple transactions,
4561 * and/or that a transaction can span multiple records.
4563 * In other words, we are allowed to submit a buffer from log recovery
4564 * once per current LSN. Otherwise, we may incorrectly skip recovery
4565 * items and cause corruption.
4567 * We don't know up front whether buffers are updated multiple times per
4568 * LSN. Therefore, track the current LSN of each commit log record as it
4569 * is processed and drain the queue when it changes. Use commit records
4570 * because they are ordered correctly by the logging code.
4572 if (log
->l_recovery_lsn
!= trans
->r_lsn
&&
4573 ohead
->oh_flags
& XLOG_COMMIT_TRANS
) {
4574 error
= xfs_buf_delwri_submit(buffer_list
);
4577 log
->l_recovery_lsn
= trans
->r_lsn
;
4580 return xlog_recovery_process_trans(log
, trans
, dp
, len
,
4581 ohead
->oh_flags
, pass
, buffer_list
);
4585 * There are two valid states of the r_state field. 0 indicates that the
4586 * transaction structure is in a normal state. We have either seen the
4587 * start of the transaction or the last operation we added was not a partial
4588 * operation. If the last operation we added to the transaction was a
4589 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4591 * NOTE: skip LRs with 0 data length.
4594 xlog_recover_process_data(
4596 struct hlist_head rhash
[],
4597 struct xlog_rec_header
*rhead
,
4600 struct list_head
*buffer_list
)
4602 struct xlog_op_header
*ohead
;
4607 end
= dp
+ be32_to_cpu(rhead
->h_len
);
4608 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
4610 /* check the log format matches our own - else we can't recover */
4611 if (xlog_header_check_recover(log
->l_mp
, rhead
))
4614 trace_xfs_log_recover_record(log
, rhead
, pass
);
4615 while ((dp
< end
) && num_logops
) {
4617 ohead
= (struct xlog_op_header
*)dp
;
4618 dp
+= sizeof(*ohead
);
4621 /* errors will abort recovery */
4622 error
= xlog_recover_process_ophdr(log
, rhash
, rhead
, ohead
,
4623 dp
, end
, pass
, buffer_list
);
4627 dp
+= be32_to_cpu(ohead
->oh_len
);
4633 /* Recover the EFI if necessary. */
4635 xlog_recover_process_efi(
4636 struct xfs_mount
*mp
,
4637 struct xfs_ail
*ailp
,
4638 struct xfs_log_item
*lip
)
4640 struct xfs_efi_log_item
*efip
;
4644 * Skip EFIs that we've already processed.
4646 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4647 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
))
4650 spin_unlock(&ailp
->ail_lock
);
4651 error
= xfs_efi_recover(mp
, efip
);
4652 spin_lock(&ailp
->ail_lock
);
4657 /* Release the EFI since we're cancelling everything. */
4659 xlog_recover_cancel_efi(
4660 struct xfs_mount
*mp
,
4661 struct xfs_ail
*ailp
,
4662 struct xfs_log_item
*lip
)
4664 struct xfs_efi_log_item
*efip
;
4666 efip
= container_of(lip
, struct xfs_efi_log_item
, efi_item
);
4668 spin_unlock(&ailp
->ail_lock
);
4669 xfs_efi_release(efip
);
4670 spin_lock(&ailp
->ail_lock
);
4673 /* Recover the RUI if necessary. */
4675 xlog_recover_process_rui(
4676 struct xfs_mount
*mp
,
4677 struct xfs_ail
*ailp
,
4678 struct xfs_log_item
*lip
)
4680 struct xfs_rui_log_item
*ruip
;
4684 * Skip RUIs that we've already processed.
4686 ruip
= container_of(lip
, struct xfs_rui_log_item
, rui_item
);
4687 if (test_bit(XFS_RUI_RECOVERED
, &ruip
->rui_flags
))
4690 spin_unlock(&ailp
->ail_lock
);
4691 error
= xfs_rui_recover(mp
, ruip
);
4692 spin_lock(&ailp
->ail_lock
);
4697 /* Release the RUI since we're cancelling everything. */
4699 xlog_recover_cancel_rui(
4700 struct xfs_mount
*mp
,
4701 struct xfs_ail
*ailp
,
4702 struct xfs_log_item
*lip
)
4704 struct xfs_rui_log_item
*ruip
;
4706 ruip
= container_of(lip
, struct xfs_rui_log_item
, rui_item
);
4708 spin_unlock(&ailp
->ail_lock
);
4709 xfs_rui_release(ruip
);
4710 spin_lock(&ailp
->ail_lock
);
4713 /* Recover the CUI if necessary. */
4715 xlog_recover_process_cui(
4716 struct xfs_mount
*mp
,
4717 struct xfs_ail
*ailp
,
4718 struct xfs_log_item
*lip
,
4719 struct xfs_defer_ops
*dfops
)
4721 struct xfs_cui_log_item
*cuip
;
4725 * Skip CUIs that we've already processed.
4727 cuip
= container_of(lip
, struct xfs_cui_log_item
, cui_item
);
4728 if (test_bit(XFS_CUI_RECOVERED
, &cuip
->cui_flags
))
4731 spin_unlock(&ailp
->ail_lock
);
4732 error
= xfs_cui_recover(mp
, cuip
, dfops
);
4733 spin_lock(&ailp
->ail_lock
);
4738 /* Release the CUI since we're cancelling everything. */
4740 xlog_recover_cancel_cui(
4741 struct xfs_mount
*mp
,
4742 struct xfs_ail
*ailp
,
4743 struct xfs_log_item
*lip
)
4745 struct xfs_cui_log_item
*cuip
;
4747 cuip
= container_of(lip
, struct xfs_cui_log_item
, cui_item
);
4749 spin_unlock(&ailp
->ail_lock
);
4750 xfs_cui_release(cuip
);
4751 spin_lock(&ailp
->ail_lock
);
4754 /* Recover the BUI if necessary. */
4756 xlog_recover_process_bui(
4757 struct xfs_mount
*mp
,
4758 struct xfs_ail
*ailp
,
4759 struct xfs_log_item
*lip
,
4760 struct xfs_defer_ops
*dfops
)
4762 struct xfs_bui_log_item
*buip
;
4766 * Skip BUIs that we've already processed.
4768 buip
= container_of(lip
, struct xfs_bui_log_item
, bui_item
);
4769 if (test_bit(XFS_BUI_RECOVERED
, &buip
->bui_flags
))
4772 spin_unlock(&ailp
->ail_lock
);
4773 error
= xfs_bui_recover(mp
, buip
, dfops
);
4774 spin_lock(&ailp
->ail_lock
);
4779 /* Release the BUI since we're cancelling everything. */
4781 xlog_recover_cancel_bui(
4782 struct xfs_mount
*mp
,
4783 struct xfs_ail
*ailp
,
4784 struct xfs_log_item
*lip
)
4786 struct xfs_bui_log_item
*buip
;
4788 buip
= container_of(lip
, struct xfs_bui_log_item
, bui_item
);
4790 spin_unlock(&ailp
->ail_lock
);
4791 xfs_bui_release(buip
);
4792 spin_lock(&ailp
->ail_lock
);
4795 /* Is this log item a deferred action intent? */
4796 static inline bool xlog_item_is_intent(struct xfs_log_item
*lip
)
4798 switch (lip
->li_type
) {
4809 /* Take all the collected deferred ops and finish them in order. */
4811 xlog_finish_defer_ops(
4812 struct xfs_mount
*mp
,
4813 struct xfs_defer_ops
*dfops
)
4815 struct xfs_trans
*tp
;
4821 * We're finishing the defer_ops that accumulated as a result of
4822 * recovering unfinished intent items during log recovery. We
4823 * reserve an itruncate transaction because it is the largest
4824 * permanent transaction type. Since we're the only user of the fs
4825 * right now, take 93% (15/16) of the available free blocks. Use
4826 * weird math to avoid a 64-bit division.
4828 freeblks
= percpu_counter_sum(&mp
->m_fdblocks
);
4831 resblks
= min_t(int64_t, UINT_MAX
, freeblks
);
4832 resblks
= (resblks
* 15) >> 4;
4833 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_itruncate
, resblks
,
4834 0, XFS_TRANS_RESERVE
, &tp
);
4838 error
= xfs_defer_finish(&tp
, dfops
);
4842 return xfs_trans_commit(tp
);
4845 xfs_trans_cancel(tp
);
4850 * When this is called, all of the log intent items which did not have
4851 * corresponding log done items should be in the AIL. What we do now
4852 * is update the data structures associated with each one.
4854 * Since we process the log intent items in normal transactions, they
4855 * will be removed at some point after the commit. This prevents us
4856 * from just walking down the list processing each one. We'll use a
4857 * flag in the intent item to skip those that we've already processed
4858 * and use the AIL iteration mechanism's generation count to try to
4859 * speed this up at least a bit.
4861 * When we start, we know that the intents are the only things in the
4862 * AIL. As we process them, however, other items are added to the
4866 xlog_recover_process_intents(
4869 struct xfs_defer_ops dfops
;
4870 struct xfs_ail_cursor cur
;
4871 struct xfs_log_item
*lip
;
4872 struct xfs_ail
*ailp
;
4873 xfs_fsblock_t firstfsb
;
4875 #if defined(DEBUG) || defined(XFS_WARN)
4880 spin_lock(&ailp
->ail_lock
);
4881 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4882 #if defined(DEBUG) || defined(XFS_WARN)
4883 last_lsn
= xlog_assign_lsn(log
->l_curr_cycle
, log
->l_curr_block
);
4885 xfs_defer_init(&dfops
, &firstfsb
);
4886 while (lip
!= NULL
) {
4888 * We're done when we see something other than an intent.
4889 * There should be no intents left in the AIL now.
4891 if (!xlog_item_is_intent(lip
)) {
4893 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4894 ASSERT(!xlog_item_is_intent(lip
));
4900 * We should never see a redo item with a LSN higher than
4901 * the last transaction we found in the log at the start
4904 ASSERT(XFS_LSN_CMP(last_lsn
, lip
->li_lsn
) >= 0);
4907 * NOTE: If your intent processing routine can create more
4908 * deferred ops, you /must/ attach them to the dfops in this
4909 * routine or else those subsequent intents will get
4910 * replayed in the wrong order!
4912 switch (lip
->li_type
) {
4914 error
= xlog_recover_process_efi(log
->l_mp
, ailp
, lip
);
4917 error
= xlog_recover_process_rui(log
->l_mp
, ailp
, lip
);
4920 error
= xlog_recover_process_cui(log
->l_mp
, ailp
, lip
,
4924 error
= xlog_recover_process_bui(log
->l_mp
, ailp
, lip
,
4930 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4933 xfs_trans_ail_cursor_done(&cur
);
4934 spin_unlock(&ailp
->ail_lock
);
4936 xfs_defer_cancel(&dfops
);
4938 error
= xlog_finish_defer_ops(log
->l_mp
, &dfops
);
4944 * A cancel occurs when the mount has failed and we're bailing out.
4945 * Release all pending log intent items so they don't pin the AIL.
4948 xlog_recover_cancel_intents(
4951 struct xfs_log_item
*lip
;
4953 struct xfs_ail_cursor cur
;
4954 struct xfs_ail
*ailp
;
4957 spin_lock(&ailp
->ail_lock
);
4958 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
4959 while (lip
!= NULL
) {
4961 * We're done when we see something other than an intent.
4962 * There should be no intents left in the AIL now.
4964 if (!xlog_item_is_intent(lip
)) {
4966 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
4967 ASSERT(!xlog_item_is_intent(lip
));
4972 switch (lip
->li_type
) {
4974 xlog_recover_cancel_efi(log
->l_mp
, ailp
, lip
);
4977 xlog_recover_cancel_rui(log
->l_mp
, ailp
, lip
);
4980 xlog_recover_cancel_cui(log
->l_mp
, ailp
, lip
);
4983 xlog_recover_cancel_bui(log
->l_mp
, ailp
, lip
);
4987 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
4990 xfs_trans_ail_cursor_done(&cur
);
4991 spin_unlock(&ailp
->ail_lock
);
4996 * This routine performs a transaction to null out a bad inode pointer
4997 * in an agi unlinked inode hash bucket.
5000 xlog_recover_clear_agi_bucket(
5002 xfs_agnumber_t agno
,
5011 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_clearagi
, 0, 0, 0, &tp
);
5015 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
5019 agi
= XFS_BUF_TO_AGI(agibp
);
5020 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
5021 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
5022 (sizeof(xfs_agino_t
) * bucket
);
5023 xfs_trans_log_buf(tp
, agibp
, offset
,
5024 (offset
+ sizeof(xfs_agino_t
) - 1));
5026 error
= xfs_trans_commit(tp
);
5032 xfs_trans_cancel(tp
);
5034 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
5039 xlog_recover_process_one_iunlink(
5040 struct xfs_mount
*mp
,
5041 xfs_agnumber_t agno
,
5045 struct xfs_buf
*ibp
;
5046 struct xfs_dinode
*dip
;
5047 struct xfs_inode
*ip
;
5051 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
5052 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
5057 * Get the on disk inode to find the next inode in the bucket.
5059 error
= xfs_imap_to_bp(mp
, NULL
, &ip
->i_imap
, &dip
, &ibp
, 0, 0);
5063 xfs_iflags_clear(ip
, XFS_IRECOVERY
);
5064 ASSERT(VFS_I(ip
)->i_nlink
== 0);
5065 ASSERT(VFS_I(ip
)->i_mode
!= 0);
5067 /* setup for the next pass */
5068 agino
= be32_to_cpu(dip
->di_next_unlinked
);
5072 * Prevent any DMAPI event from being sent when the reference on
5073 * the inode is dropped.
5075 ip
->i_d
.di_dmevmask
= 0;
5084 * We can't read in the inode this bucket points to, or this inode
5085 * is messed up. Just ditch this bucket of inodes. We will lose
5086 * some inodes and space, but at least we won't hang.
5088 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5089 * clear the inode pointer in the bucket.
5091 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
5096 * xlog_iunlink_recover
5098 * This is called during recovery to process any inodes which
5099 * we unlinked but not freed when the system crashed. These
5100 * inodes will be on the lists in the AGI blocks. What we do
5101 * here is scan all the AGIs and fully truncate and free any
5102 * inodes found on the lists. Each inode is removed from the
5103 * lists when it has been fully truncated and is freed. The
5104 * freeing of the inode and its removal from the list must be
5108 xlog_recover_process_iunlinks(
5112 xfs_agnumber_t agno
;
5121 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
5123 * Find the agi for this ag.
5125 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
5128 * AGI is b0rked. Don't process it.
5130 * We should probably mark the filesystem as corrupt
5131 * after we've recovered all the ag's we can....
5136 * Unlock the buffer so that it can be acquired in the normal
5137 * course of the transaction to truncate and free each inode.
5138 * Because we are not racing with anyone else here for the AGI
5139 * buffer, we don't even need to hold it locked to read the
5140 * initial unlinked bucket entries out of the buffer. We keep
5141 * buffer reference though, so that it stays pinned in memory
5142 * while we need the buffer.
5144 agi
= XFS_BUF_TO_AGI(agibp
);
5145 xfs_buf_unlock(agibp
);
5147 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
5148 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
5149 while (agino
!= NULLAGINO
) {
5150 agino
= xlog_recover_process_one_iunlink(mp
,
5151 agno
, agino
, bucket
);
5154 xfs_buf_rele(agibp
);
5160 struct xlog_rec_header
*rhead
,
5166 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
5167 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
5168 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
5172 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
5173 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
5174 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
5175 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
5176 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
5177 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
5186 * CRC check, unpack and process a log record.
5189 xlog_recover_process(
5191 struct hlist_head rhash
[],
5192 struct xlog_rec_header
*rhead
,
5195 struct list_head
*buffer_list
)
5198 __le32 old_crc
= rhead
->h_crc
;
5202 crc
= xlog_cksum(log
, rhead
, dp
, be32_to_cpu(rhead
->h_len
));
5205 * Nothing else to do if this is a CRC verification pass. Just return
5206 * if this a record with a non-zero crc. Unfortunately, mkfs always
5207 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5208 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5209 * know precisely what failed.
5211 if (pass
== XLOG_RECOVER_CRCPASS
) {
5212 if (old_crc
&& crc
!= old_crc
)
5218 * We're in the normal recovery path. Issue a warning if and only if the
5219 * CRC in the header is non-zero. This is an advisory warning and the
5220 * zero CRC check prevents warnings from being emitted when upgrading
5221 * the kernel from one that does not add CRCs by default.
5223 if (crc
!= old_crc
) {
5224 if (old_crc
|| xfs_sb_version_hascrc(&log
->l_mp
->m_sb
)) {
5225 xfs_alert(log
->l_mp
,
5226 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5227 le32_to_cpu(old_crc
),
5229 xfs_hex_dump(dp
, 32);
5233 * If the filesystem is CRC enabled, this mismatch becomes a
5234 * fatal log corruption failure.
5236 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
))
5237 return -EFSCORRUPTED
;
5240 error
= xlog_unpack_data(rhead
, dp
, log
);
5244 return xlog_recover_process_data(log
, rhash
, rhead
, dp
, pass
,
5249 xlog_valid_rec_header(
5251 struct xlog_rec_header
*rhead
,
5256 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
5257 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5258 XFS_ERRLEVEL_LOW
, log
->l_mp
);
5259 return -EFSCORRUPTED
;
5262 (!rhead
->h_version
||
5263 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
5264 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
5265 __func__
, be32_to_cpu(rhead
->h_version
));
5269 /* LR body must have data or it wouldn't have been written */
5270 hlen
= be32_to_cpu(rhead
->h_len
);
5271 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
5272 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5273 XFS_ERRLEVEL_LOW
, log
->l_mp
);
5274 return -EFSCORRUPTED
;
5276 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
5277 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5278 XFS_ERRLEVEL_LOW
, log
->l_mp
);
5279 return -EFSCORRUPTED
;
5285 * Read the log from tail to head and process the log records found.
5286 * Handle the two cases where the tail and head are in the same cycle
5287 * and where the active portion of the log wraps around the end of
5288 * the physical log separately. The pass parameter is passed through
5289 * to the routines called to process the data and is not looked at
5293 xlog_do_recovery_pass(
5295 xfs_daddr_t head_blk
,
5296 xfs_daddr_t tail_blk
,
5298 xfs_daddr_t
*first_bad
) /* out: first bad log rec */
5300 xlog_rec_header_t
*rhead
;
5301 xfs_daddr_t blk_no
, rblk_no
;
5302 xfs_daddr_t rhead_blk
;
5304 xfs_buf_t
*hbp
, *dbp
;
5305 int error
= 0, h_size
, h_len
;
5307 int bblks
, split_bblks
;
5308 int hblks
, split_hblks
, wrapped_hblks
;
5310 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
5311 LIST_HEAD (buffer_list
);
5313 ASSERT(head_blk
!= tail_blk
);
5314 blk_no
= rhead_blk
= tail_blk
;
5316 for (i
= 0; i
< XLOG_RHASH_SIZE
; i
++)
5317 INIT_HLIST_HEAD(&rhash
[i
]);
5320 * Read the header of the tail block and get the iclog buffer size from
5321 * h_size. Use this to tell how many sectors make up the log header.
5323 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
5325 * When using variable length iclogs, read first sector of
5326 * iclog header and extract the header size from it. Get a
5327 * new hbp that is the correct size.
5329 hbp
= xlog_get_bp(log
, 1);
5333 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
5337 rhead
= (xlog_rec_header_t
*)offset
;
5338 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
5343 * xfsprogs has a bug where record length is based on lsunit but
5344 * h_size (iclog size) is hardcoded to 32k. Now that we
5345 * unconditionally CRC verify the unmount record, this means the
5346 * log buffer can be too small for the record and cause an
5349 * Detect this condition here. Use lsunit for the buffer size as
5350 * long as this looks like the mkfs case. Otherwise, return an
5351 * error to avoid a buffer overrun.
5353 h_size
= be32_to_cpu(rhead
->h_size
);
5354 h_len
= be32_to_cpu(rhead
->h_len
);
5355 if (h_len
> h_size
) {
5356 if (h_len
<= log
->l_mp
->m_logbsize
&&
5357 be32_to_cpu(rhead
->h_num_logops
) == 1) {
5359 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5360 h_size
, log
->l_mp
->m_logbsize
);
5361 h_size
= log
->l_mp
->m_logbsize
;
5363 return -EFSCORRUPTED
;
5366 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
5367 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
5368 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
5369 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
5372 hbp
= xlog_get_bp(log
, hblks
);
5377 ASSERT(log
->l_sectBBsize
== 1);
5379 hbp
= xlog_get_bp(log
, 1);
5380 h_size
= XLOG_BIG_RECORD_BSIZE
;
5385 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
5391 memset(rhash
, 0, sizeof(rhash
));
5392 if (tail_blk
> head_blk
) {
5394 * Perform recovery around the end of the physical log.
5395 * When the head is not on the same cycle number as the tail,
5396 * we can't do a sequential recovery.
5398 while (blk_no
< log
->l_logBBsize
) {
5400 * Check for header wrapping around physical end-of-log
5402 offset
= hbp
->b_addr
;
5405 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
5406 /* Read header in one read */
5407 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
5412 /* This LR is split across physical log end */
5413 if (blk_no
!= log
->l_logBBsize
) {
5414 /* some data before physical log end */
5415 ASSERT(blk_no
<= INT_MAX
);
5416 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
5417 ASSERT(split_hblks
> 0);
5418 error
= xlog_bread(log
, blk_no
,
5426 * Note: this black magic still works with
5427 * large sector sizes (non-512) only because:
5428 * - we increased the buffer size originally
5429 * by 1 sector giving us enough extra space
5430 * for the second read;
5431 * - the log start is guaranteed to be sector
5433 * - we read the log end (LR header start)
5434 * _first_, then the log start (LR header end)
5435 * - order is important.
5437 wrapped_hblks
= hblks
- split_hblks
;
5438 error
= xlog_bread_offset(log
, 0,
5440 offset
+ BBTOB(split_hblks
));
5444 rhead
= (xlog_rec_header_t
*)offset
;
5445 error
= xlog_valid_rec_header(log
, rhead
,
5446 split_hblks
? blk_no
: 0);
5450 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
5454 * Read the log record data in multiple reads if it
5455 * wraps around the end of the log. Note that if the
5456 * header already wrapped, blk_no could point past the
5457 * end of the log. The record data is contiguous in
5460 if (blk_no
+ bblks
<= log
->l_logBBsize
||
5461 blk_no
>= log
->l_logBBsize
) {
5462 /* mod blk_no in case the header wrapped and
5463 * pushed it beyond the end of the log */
5464 rblk_no
= do_mod(blk_no
, log
->l_logBBsize
);
5465 error
= xlog_bread(log
, rblk_no
, bblks
, dbp
,
5470 /* This log record is split across the
5471 * physical end of log */
5472 offset
= dbp
->b_addr
;
5474 if (blk_no
!= log
->l_logBBsize
) {
5475 /* some data is before the physical
5477 ASSERT(!wrapped_hblks
);
5478 ASSERT(blk_no
<= INT_MAX
);
5480 log
->l_logBBsize
- (int)blk_no
;
5481 ASSERT(split_bblks
> 0);
5482 error
= xlog_bread(log
, blk_no
,
5490 * Note: this black magic still works with
5491 * large sector sizes (non-512) only because:
5492 * - we increased the buffer size originally
5493 * by 1 sector giving us enough extra space
5494 * for the second read;
5495 * - the log start is guaranteed to be sector
5497 * - we read the log end (LR header start)
5498 * _first_, then the log start (LR header end)
5499 * - order is important.
5501 error
= xlog_bread_offset(log
, 0,
5502 bblks
- split_bblks
, dbp
,
5503 offset
+ BBTOB(split_bblks
));
5508 error
= xlog_recover_process(log
, rhash
, rhead
, offset
,
5509 pass
, &buffer_list
);
5517 ASSERT(blk_no
>= log
->l_logBBsize
);
5518 blk_no
-= log
->l_logBBsize
;
5522 /* read first part of physical log */
5523 while (blk_no
< head_blk
) {
5524 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
5528 rhead
= (xlog_rec_header_t
*)offset
;
5529 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
5533 /* blocks in data section */
5534 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
5535 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
5540 error
= xlog_recover_process(log
, rhash
, rhead
, offset
, pass
,
5545 blk_no
+= bblks
+ hblks
;
5555 * Submit buffers that have been added from the last record processed,
5556 * regardless of error status.
5558 if (!list_empty(&buffer_list
))
5559 error2
= xfs_buf_delwri_submit(&buffer_list
);
5561 if (error
&& first_bad
)
5562 *first_bad
= rhead_blk
;
5565 * Transactions are freed at commit time but transactions without commit
5566 * records on disk are never committed. Free any that may be left in the
5569 for (i
= 0; i
< XLOG_RHASH_SIZE
; i
++) {
5570 struct hlist_node
*tmp
;
5571 struct xlog_recover
*trans
;
5573 hlist_for_each_entry_safe(trans
, tmp
, &rhash
[i
], r_list
)
5574 xlog_recover_free_trans(trans
);
5577 return error
? error
: error2
;
5581 * Do the recovery of the log. We actually do this in two phases.
5582 * The two passes are necessary in order to implement the function
5583 * of cancelling a record written into the log. The first pass
5584 * determines those things which have been cancelled, and the
5585 * second pass replays log items normally except for those which
5586 * have been cancelled. The handling of the replay and cancellations
5587 * takes place in the log item type specific routines.
5589 * The table of items which have cancel records in the log is allocated
5590 * and freed at this level, since only here do we know when all of
5591 * the log recovery has been completed.
5594 xlog_do_log_recovery(
5596 xfs_daddr_t head_blk
,
5597 xfs_daddr_t tail_blk
)
5601 ASSERT(head_blk
!= tail_blk
);
5604 * First do a pass to find all of the cancelled buf log items.
5605 * Store them in the buf_cancel_table for use in the second pass.
5607 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
5608 sizeof(struct list_head
),
5610 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
5611 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
5613 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
5614 XLOG_RECOVER_PASS1
, NULL
);
5616 kmem_free(log
->l_buf_cancel_table
);
5617 log
->l_buf_cancel_table
= NULL
;
5621 * Then do a second pass to actually recover the items in the log.
5622 * When it is complete free the table of buf cancel items.
5624 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
5625 XLOG_RECOVER_PASS2
, NULL
);
5630 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
5631 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
5635 kmem_free(log
->l_buf_cancel_table
);
5636 log
->l_buf_cancel_table
= NULL
;
5642 * Do the actual recovery
5647 xfs_daddr_t head_blk
,
5648 xfs_daddr_t tail_blk
)
5650 struct xfs_mount
*mp
= log
->l_mp
;
5655 trace_xfs_log_recover(log
, head_blk
, tail_blk
);
5658 * First replay the images in the log.
5660 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
5665 * If IO errors happened during recovery, bail out.
5667 if (XFS_FORCED_SHUTDOWN(mp
)) {
5672 * We now update the tail_lsn since much of the recovery has completed
5673 * and there may be space available to use. If there were no extent
5674 * or iunlinks, we can free up the entire log and set the tail_lsn to
5675 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5676 * lsn of the last known good LR on disk. If there are extent frees
5677 * or iunlinks they will have some entries in the AIL; so we look at
5678 * the AIL to determine how to set the tail_lsn.
5680 xlog_assign_tail_lsn(mp
);
5683 * Now that we've finished replaying all buffer and inode
5684 * updates, re-read in the superblock and reverify it.
5686 bp
= xfs_getsb(mp
, 0);
5687 bp
->b_flags
&= ~(XBF_DONE
| XBF_ASYNC
);
5688 ASSERT(!(bp
->b_flags
& XBF_WRITE
));
5689 bp
->b_flags
|= XBF_READ
;
5690 bp
->b_ops
= &xfs_sb_buf_ops
;
5692 error
= xfs_buf_submit_wait(bp
);
5694 if (!XFS_FORCED_SHUTDOWN(mp
)) {
5695 xfs_buf_ioerror_alert(bp
, __func__
);
5702 /* Convert superblock from on-disk format */
5704 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
5707 /* re-initialise in-core superblock and geometry structures */
5708 xfs_reinit_percpu_counters(mp
);
5709 error
= xfs_initialize_perag(mp
, sbp
->sb_agcount
, &mp
->m_maxagi
);
5711 xfs_warn(mp
, "Failed post-recovery per-ag init: %d", error
);
5714 mp
->m_alloc_set_aside
= xfs_alloc_set_aside(mp
);
5716 xlog_recover_check_summary(log
);
5718 /* Normal transactions can now occur */
5719 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
5724 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5726 * Return error or zero.
5732 xfs_daddr_t head_blk
, tail_blk
;
5735 /* find the tail of the log */
5736 error
= xlog_find_tail(log
, &head_blk
, &tail_blk
);
5741 * The superblock was read before the log was available and thus the LSN
5742 * could not be verified. Check the superblock LSN against the current
5743 * LSN now that it's known.
5745 if (xfs_sb_version_hascrc(&log
->l_mp
->m_sb
) &&
5746 !xfs_log_check_lsn(log
->l_mp
, log
->l_mp
->m_sb
.sb_lsn
))
5749 if (tail_blk
!= head_blk
) {
5750 /* There used to be a comment here:
5752 * disallow recovery on read-only mounts. note -- mount
5753 * checks for ENOSPC and turns it into an intelligent
5755 * ...but this is no longer true. Now, unless you specify
5756 * NORECOVERY (in which case this function would never be
5757 * called), we just go ahead and recover. We do this all
5758 * under the vfs layer, so we can get away with it unless
5759 * the device itself is read-only, in which case we fail.
5761 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
5766 * Version 5 superblock log feature mask validation. We know the
5767 * log is dirty so check if there are any unknown log features
5768 * in what we need to recover. If there are unknown features
5769 * (e.g. unsupported transactions, then simply reject the
5770 * attempt at recovery before touching anything.
5772 if (XFS_SB_VERSION_NUM(&log
->l_mp
->m_sb
) == XFS_SB_VERSION_5
&&
5773 xfs_sb_has_incompat_log_feature(&log
->l_mp
->m_sb
,
5774 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
)) {
5776 "Superblock has unknown incompatible log features (0x%x) enabled.",
5777 (log
->l_mp
->m_sb
.sb_features_log_incompat
&
5778 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN
));
5780 "The log can not be fully and/or safely recovered by this kernel.");
5782 "Please recover the log on a kernel that supports the unknown features.");
5787 * Delay log recovery if the debug hook is set. This is debug
5788 * instrumention to coordinate simulation of I/O failures with
5791 if (xfs_globals
.log_recovery_delay
) {
5792 xfs_notice(log
->l_mp
,
5793 "Delaying log recovery for %d seconds.",
5794 xfs_globals
.log_recovery_delay
);
5795 msleep(xfs_globals
.log_recovery_delay
* 1000);
5798 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
5799 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5802 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
5803 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
5809 * In the first part of recovery we replay inodes and buffers and build
5810 * up the list of extent free items which need to be processed. Here
5811 * we process the extent free items and clean up the on disk unlinked
5812 * inode lists. This is separated from the first part of recovery so
5813 * that the root and real-time bitmap inodes can be read in from disk in
5814 * between the two stages. This is necessary so that we can free space
5815 * in the real-time portion of the file system.
5818 xlog_recover_finish(
5822 * Now we're ready to do the transactions needed for the
5823 * rest of recovery. Start with completing all the extent
5824 * free intent records and then process the unlinked inode
5825 * lists. At this point, we essentially run in normal mode
5826 * except that we're still performing recovery actions
5827 * rather than accepting new requests.
5829 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
5831 error
= xlog_recover_process_intents(log
);
5833 xfs_alert(log
->l_mp
, "Failed to recover intents");
5838 * Sync the log to get all the intents out of the AIL.
5839 * This isn't absolutely necessary, but it helps in
5840 * case the unlink transactions would have problems
5841 * pushing the intents out of the way.
5843 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
5845 xlog_recover_process_iunlinks(log
);
5847 xlog_recover_check_summary(log
);
5849 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
5850 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
5852 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
5854 xfs_info(log
->l_mp
, "Ending clean mount");
5860 xlog_recover_cancel(
5865 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
)
5866 error
= xlog_recover_cancel_intents(log
);
5873 * Read all of the agf and agi counters and check that they
5874 * are consistent with the superblock counters.
5877 xlog_recover_check_summary(
5884 xfs_agnumber_t agno
;
5895 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
5896 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
5898 xfs_alert(mp
, "%s agf read failed agno %d error %d",
5899 __func__
, agno
, error
);
5901 agfp
= XFS_BUF_TO_AGF(agfbp
);
5902 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
5903 be32_to_cpu(agfp
->agf_flcount
);
5904 xfs_buf_relse(agfbp
);
5907 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
5909 xfs_alert(mp
, "%s agi read failed agno %d error %d",
5910 __func__
, agno
, error
);
5912 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
5914 itotal
+= be32_to_cpu(agi
->agi_count
);
5915 ifree
+= be32_to_cpu(agi
->agi_freecount
);
5916 xfs_buf_relse(agibp
);