2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_alloc.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_log_priv.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_log_recover.h"
40 #include "xfs_extfree_item.h"
41 #include "xfs_trans_priv.h"
42 #include "xfs_quota.h"
44 #include "xfs_utils.h"
45 #include "xfs_trace.h"
47 STATIC
int xlog_find_zeroed(xlog_t
*, xfs_daddr_t
*);
48 STATIC
int xlog_clear_stale_blocks(xlog_t
*, xfs_lsn_t
);
50 STATIC
void xlog_recover_check_summary(xlog_t
*);
52 #define xlog_recover_check_summary(log)
56 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
59 struct xfs_buf_cancel
{
63 struct list_head bc_list
;
67 * Sector aligned buffer routines for buffer create/read/write/access
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
77 xlog_buf_bbcount_valid(
81 return bbcount
> 0 && bbcount
<= log
->l_logBBsize
;
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
96 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
97 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
99 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
104 * We do log I/O in units of log sectors (a power-of-2
105 * multiple of the basic block size), so we round up the
106 * requested size to accommodate the basic blocks required
107 * for complete log sectors.
109 * In addition, the buffer may be used for a non-sector-
110 * aligned block offset, in which case an I/O of the
111 * requested size could extend beyond the end of the
112 * buffer. If the requested size is only 1 basic block it
113 * will never straddle a sector boundary, so this won't be
114 * an issue. Nor will this be a problem if the log I/O is
115 * done in basic blocks (sector size 1). But otherwise we
116 * extend the buffer by one extra log sector to ensure
117 * there's space to accommodate this possibility.
119 if (nbblks
> 1 && log
->l_sectBBsize
> 1)
120 nbblks
+= log
->l_sectBBsize
;
121 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
123 bp
= xfs_buf_get_uncached(log
->l_mp
->m_logdev_targp
, BBTOB(nbblks
), 0);
137 * Return the address of the start of the given block number's data
138 * in a log buffer. The buffer covers a log sector-aligned region.
147 xfs_daddr_t offset
= blk_no
& ((xfs_daddr_t
)log
->l_sectBBsize
- 1);
149 ASSERT(BBTOB(offset
+ nbblks
) <= XFS_BUF_SIZE(bp
));
150 return XFS_BUF_PTR(bp
) + BBTOB(offset
);
155 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
166 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
167 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
169 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
173 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
174 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
177 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
179 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
182 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
183 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
185 xfsbdstrat(log
->l_mp
, bp
);
186 error
= xfs_buf_iowait(bp
);
188 xfs_ioerror_alert("xlog_bread", log
->l_mp
,
189 bp
, XFS_BUF_ADDR(bp
));
203 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
207 *offset
= xlog_align(log
, blk_no
, nbblks
, bp
);
212 * Read at an offset into the buffer. Returns with the buffer in it's original
213 * state regardless of the result of the read.
218 xfs_daddr_t blk_no
, /* block to read from */
219 int nbblks
, /* blocks to read */
223 xfs_caddr_t orig_offset
= XFS_BUF_PTR(bp
);
224 int orig_len
= bp
->b_buffer_length
;
227 error
= XFS_BUF_SET_PTR(bp
, offset
, BBTOB(nbblks
));
231 error
= xlog_bread_noalign(log
, blk_no
, nbblks
, bp
);
233 /* must reset buffer pointer even on error */
234 error2
= XFS_BUF_SET_PTR(bp
, orig_offset
, orig_len
);
241 * Write out the buffer at the given block for the given number of blocks.
242 * The buffer is kept locked across the write and is returned locked.
243 * This can only be used for synchronous log writes.
254 if (!xlog_buf_bbcount_valid(log
, nbblks
)) {
255 xfs_warn(log
->l_mp
, "Invalid block length (0x%x) for buffer",
257 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_HIGH
, log
->l_mp
);
261 blk_no
= round_down(blk_no
, log
->l_sectBBsize
);
262 nbblks
= round_up(nbblks
, log
->l_sectBBsize
);
265 ASSERT(BBTOB(nbblks
) <= XFS_BUF_SIZE(bp
));
267 XFS_BUF_SET_ADDR(bp
, log
->l_logBBstart
+ blk_no
);
268 XFS_BUF_ZEROFLAGS(bp
);
272 XFS_BUF_SET_COUNT(bp
, BBTOB(nbblks
));
273 XFS_BUF_SET_TARGET(bp
, log
->l_mp
->m_logdev_targp
);
275 if ((error
= xfs_bwrite(log
->l_mp
, bp
)))
276 xfs_ioerror_alert("xlog_bwrite", log
->l_mp
,
277 bp
, XFS_BUF_ADDR(bp
));
283 * dump debug superblock and log record information
286 xlog_header_check_dump(
288 xlog_rec_header_t
*head
)
290 xfs_debug(mp
, "%s: SB : uuid = %pU, fmt = %d\n",
291 __func__
, &mp
->m_sb
.sb_uuid
, XLOG_FMT
);
292 xfs_debug(mp
, " log : uuid = %pU, fmt = %d\n",
293 &head
->h_fs_uuid
, be32_to_cpu(head
->h_fmt
));
296 #define xlog_header_check_dump(mp, head)
300 * check log record header for recovery
303 xlog_header_check_recover(
305 xlog_rec_header_t
*head
)
307 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
310 * IRIX doesn't write the h_fmt field and leaves it zeroed
311 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
312 * a dirty log created in IRIX.
314 if (unlikely(head
->h_fmt
!= cpu_to_be32(XLOG_FMT
))) {
316 "dirty log written in incompatible format - can't recover");
317 xlog_header_check_dump(mp
, head
);
318 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
319 XFS_ERRLEVEL_HIGH
, mp
);
320 return XFS_ERROR(EFSCORRUPTED
);
321 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
323 "dirty log entry has mismatched uuid - can't recover");
324 xlog_header_check_dump(mp
, head
);
325 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
326 XFS_ERRLEVEL_HIGH
, mp
);
327 return XFS_ERROR(EFSCORRUPTED
);
333 * read the head block of the log and check the header
336 xlog_header_check_mount(
338 xlog_rec_header_t
*head
)
340 ASSERT(head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
));
342 if (uuid_is_nil(&head
->h_fs_uuid
)) {
344 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
345 * h_fs_uuid is nil, we assume this log was last mounted
346 * by IRIX and continue.
348 xfs_warn(mp
, "nil uuid in log - IRIX style log");
349 } else if (unlikely(!uuid_equal(&mp
->m_sb
.sb_uuid
, &head
->h_fs_uuid
))) {
350 xfs_warn(mp
, "log has mismatched uuid - can't recover");
351 xlog_header_check_dump(mp
, head
);
352 XFS_ERROR_REPORT("xlog_header_check_mount",
353 XFS_ERRLEVEL_HIGH
, mp
);
354 return XFS_ERROR(EFSCORRUPTED
);
365 * We're not going to bother about retrying
366 * this during recovery. One strike!
368 xfs_ioerror_alert("xlog_recover_iodone",
369 bp
->b_target
->bt_mount
, bp
,
371 xfs_force_shutdown(bp
->b_target
->bt_mount
,
372 SHUTDOWN_META_IO_ERROR
);
375 xfs_buf_ioend(bp
, 0);
379 * This routine finds (to an approximation) the first block in the physical
380 * log which contains the given cycle. It uses a binary search algorithm.
381 * Note that the algorithm can not be perfect because the disk will not
382 * necessarily be perfect.
385 xlog_find_cycle_start(
388 xfs_daddr_t first_blk
,
389 xfs_daddr_t
*last_blk
,
399 mid_blk
= BLK_AVG(first_blk
, end_blk
);
400 while (mid_blk
!= first_blk
&& mid_blk
!= end_blk
) {
401 error
= xlog_bread(log
, mid_blk
, 1, bp
, &offset
);
404 mid_cycle
= xlog_get_cycle(offset
);
405 if (mid_cycle
== cycle
)
406 end_blk
= mid_blk
; /* last_half_cycle == mid_cycle */
408 first_blk
= mid_blk
; /* first_half_cycle == mid_cycle */
409 mid_blk
= BLK_AVG(first_blk
, end_blk
);
411 ASSERT((mid_blk
== first_blk
&& mid_blk
+1 == end_blk
) ||
412 (mid_blk
== end_blk
&& mid_blk
-1 == first_blk
));
420 * Check that a range of blocks does not contain stop_on_cycle_no.
421 * Fill in *new_blk with the block offset where such a block is
422 * found, or with -1 (an invalid block number) if there is no such
423 * block in the range. The scan needs to occur from front to back
424 * and the pointer into the region must be updated since a later
425 * routine will need to perform another test.
428 xlog_find_verify_cycle(
430 xfs_daddr_t start_blk
,
432 uint stop_on_cycle_no
,
433 xfs_daddr_t
*new_blk
)
439 xfs_caddr_t buf
= NULL
;
443 * Greedily allocate a buffer big enough to handle the full
444 * range of basic blocks we'll be examining. If that fails,
445 * try a smaller size. We need to be able to read at least
446 * a log sector, or we're out of luck.
448 bufblks
= 1 << ffs(nbblks
);
449 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
451 if (bufblks
< log
->l_sectBBsize
)
455 for (i
= start_blk
; i
< start_blk
+ nbblks
; i
+= bufblks
) {
458 bcount
= min(bufblks
, (start_blk
+ nbblks
- i
));
460 error
= xlog_bread(log
, i
, bcount
, bp
, &buf
);
464 for (j
= 0; j
< bcount
; j
++) {
465 cycle
= xlog_get_cycle(buf
);
466 if (cycle
== stop_on_cycle_no
) {
483 * Potentially backup over partial log record write.
485 * In the typical case, last_blk is the number of the block directly after
486 * a good log record. Therefore, we subtract one to get the block number
487 * of the last block in the given buffer. extra_bblks contains the number
488 * of blocks we would have read on a previous read. This happens when the
489 * last log record is split over the end of the physical log.
491 * extra_bblks is the number of blocks potentially verified on a previous
492 * call to this routine.
495 xlog_find_verify_log_record(
497 xfs_daddr_t start_blk
,
498 xfs_daddr_t
*last_blk
,
503 xfs_caddr_t offset
= NULL
;
504 xlog_rec_header_t
*head
= NULL
;
507 int num_blks
= *last_blk
- start_blk
;
510 ASSERT(start_blk
!= 0 || *last_blk
!= start_blk
);
512 if (!(bp
= xlog_get_bp(log
, num_blks
))) {
513 if (!(bp
= xlog_get_bp(log
, 1)))
517 error
= xlog_bread(log
, start_blk
, num_blks
, bp
, &offset
);
520 offset
+= ((num_blks
- 1) << BBSHIFT
);
523 for (i
= (*last_blk
) - 1; i
>= 0; i
--) {
525 /* valid log record not found */
527 "Log inconsistent (didn't find previous header)");
529 error
= XFS_ERROR(EIO
);
534 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
539 head
= (xlog_rec_header_t
*)offset
;
541 if (head
->h_magicno
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
549 * We hit the beginning of the physical log & still no header. Return
550 * to caller. If caller can handle a return of -1, then this routine
551 * will be called again for the end of the physical log.
559 * We have the final block of the good log (the first block
560 * of the log record _before_ the head. So we check the uuid.
562 if ((error
= xlog_header_check_mount(log
->l_mp
, head
)))
566 * We may have found a log record header before we expected one.
567 * last_blk will be the 1st block # with a given cycle #. We may end
568 * up reading an entire log record. In this case, we don't want to
569 * reset last_blk. Only when last_blk points in the middle of a log
570 * record do we update last_blk.
572 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
573 uint h_size
= be32_to_cpu(head
->h_size
);
575 xhdrs
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
576 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
582 if (*last_blk
- i
+ extra_bblks
!=
583 BTOBB(be32_to_cpu(head
->h_len
)) + xhdrs
)
592 * Head is defined to be the point of the log where the next log write
593 * write could go. This means that incomplete LR writes at the end are
594 * eliminated when calculating the head. We aren't guaranteed that previous
595 * LR have complete transactions. We only know that a cycle number of
596 * current cycle number -1 won't be present in the log if we start writing
597 * from our current block number.
599 * last_blk contains the block number of the first block with a given
602 * Return: zero if normal, non-zero if error.
607 xfs_daddr_t
*return_head_blk
)
611 xfs_daddr_t new_blk
, first_blk
, start_blk
, last_blk
, head_blk
;
613 uint first_half_cycle
, last_half_cycle
;
615 int error
, log_bbnum
= log
->l_logBBsize
;
617 /* Is the end of the log device zeroed? */
618 if ((error
= xlog_find_zeroed(log
, &first_blk
)) == -1) {
619 *return_head_blk
= first_blk
;
621 /* Is the whole lot zeroed? */
623 /* Linux XFS shouldn't generate totally zeroed logs -
624 * mkfs etc write a dummy unmount record to a fresh
625 * log so we can store the uuid in there
627 xfs_warn(log
->l_mp
, "totally zeroed log");
632 xfs_warn(log
->l_mp
, "empty log check failed");
636 first_blk
= 0; /* get cycle # of 1st block */
637 bp
= xlog_get_bp(log
, 1);
641 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
645 first_half_cycle
= xlog_get_cycle(offset
);
647 last_blk
= head_blk
= log_bbnum
- 1; /* get cycle # of last block */
648 error
= xlog_bread(log
, last_blk
, 1, bp
, &offset
);
652 last_half_cycle
= xlog_get_cycle(offset
);
653 ASSERT(last_half_cycle
!= 0);
656 * If the 1st half cycle number is equal to the last half cycle number,
657 * then the entire log is stamped with the same cycle number. In this
658 * case, head_blk can't be set to zero (which makes sense). The below
659 * math doesn't work out properly with head_blk equal to zero. Instead,
660 * we set it to log_bbnum which is an invalid block number, but this
661 * value makes the math correct. If head_blk doesn't changed through
662 * all the tests below, *head_blk is set to zero at the very end rather
663 * than log_bbnum. In a sense, log_bbnum and zero are the same block
664 * in a circular file.
666 if (first_half_cycle
== last_half_cycle
) {
668 * In this case we believe that the entire log should have
669 * cycle number last_half_cycle. We need to scan backwards
670 * from the end verifying that there are no holes still
671 * containing last_half_cycle - 1. If we find such a hole,
672 * then the start of that hole will be the new head. The
673 * simple case looks like
674 * x | x ... | x - 1 | x
675 * Another case that fits this picture would be
676 * x | x + 1 | x ... | x
677 * In this case the head really is somewhere at the end of the
678 * log, as one of the latest writes at the beginning was
681 * x | x + 1 | x ... | x - 1 | x
682 * This is really the combination of the above two cases, and
683 * the head has to end up at the start of the x-1 hole at the
686 * In the 256k log case, we will read from the beginning to the
687 * end of the log and search for cycle numbers equal to x-1.
688 * We don't worry about the x+1 blocks that we encounter,
689 * because we know that they cannot be the head since the log
692 head_blk
= log_bbnum
;
693 stop_on_cycle
= last_half_cycle
- 1;
696 * In this case we want to find the first block with cycle
697 * number matching last_half_cycle. We expect the log to be
699 * x + 1 ... | x ... | x
700 * The first block with cycle number x (last_half_cycle) will
701 * be where the new head belongs. First we do a binary search
702 * for the first occurrence of last_half_cycle. The binary
703 * search may not be totally accurate, so then we scan back
704 * from there looking for occurrences of last_half_cycle before
705 * us. If that backwards scan wraps around the beginning of
706 * the log, then we look for occurrences of last_half_cycle - 1
707 * at the end of the log. The cases we're looking for look
709 * v binary search stopped here
710 * x + 1 ... | x | x + 1 | x ... | x
711 * ^ but we want to locate this spot
713 * <---------> less than scan distance
714 * x + 1 ... | x ... | x - 1 | x
715 * ^ we want to locate this spot
717 stop_on_cycle
= last_half_cycle
;
718 if ((error
= xlog_find_cycle_start(log
, bp
, first_blk
,
719 &head_blk
, last_half_cycle
)))
724 * Now validate the answer. Scan back some number of maximum possible
725 * blocks and make sure each one has the expected cycle number. The
726 * maximum is determined by the total possible amount of buffering
727 * in the in-core log. The following number can be made tighter if
728 * we actually look at the block size of the filesystem.
730 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
731 if (head_blk
>= num_scan_bblks
) {
733 * We are guaranteed that the entire check can be performed
736 start_blk
= head_blk
- num_scan_bblks
;
737 if ((error
= xlog_find_verify_cycle(log
,
738 start_blk
, num_scan_bblks
,
739 stop_on_cycle
, &new_blk
)))
743 } else { /* need to read 2 parts of log */
745 * We are going to scan backwards in the log in two parts.
746 * First we scan the physical end of the log. In this part
747 * of the log, we are looking for blocks with cycle number
748 * last_half_cycle - 1.
749 * If we find one, then we know that the log starts there, as
750 * we've found a hole that didn't get written in going around
751 * the end of the physical log. The simple case for this is
752 * x + 1 ... | x ... | x - 1 | x
753 * <---------> less than scan distance
754 * If all of the blocks at the end of the log have cycle number
755 * last_half_cycle, then we check the blocks at the start of
756 * the log looking for occurrences of last_half_cycle. If we
757 * find one, then our current estimate for the location of the
758 * first occurrence of last_half_cycle is wrong and we move
759 * back to the hole we've found. This case looks like
760 * x + 1 ... | x | x + 1 | x ...
761 * ^ binary search stopped here
762 * Another case we need to handle that only occurs in 256k
764 * x + 1 ... | x ... | x+1 | x ...
765 * ^ binary search stops here
766 * In a 256k log, the scan at the end of the log will see the
767 * x + 1 blocks. We need to skip past those since that is
768 * certainly not the head of the log. By searching for
769 * last_half_cycle-1 we accomplish that.
771 ASSERT(head_blk
<= INT_MAX
&&
772 (xfs_daddr_t
) num_scan_bblks
>= head_blk
);
773 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
774 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
775 num_scan_bblks
- (int)head_blk
,
776 (stop_on_cycle
- 1), &new_blk
)))
784 * Scan beginning of log now. The last part of the physical
785 * log is good. This scan needs to verify that it doesn't find
786 * the last_half_cycle.
789 ASSERT(head_blk
<= INT_MAX
);
790 if ((error
= xlog_find_verify_cycle(log
,
791 start_blk
, (int)head_blk
,
792 stop_on_cycle
, &new_blk
)))
800 * Now we need to make sure head_blk is not pointing to a block in
801 * the middle of a log record.
803 num_scan_bblks
= XLOG_REC_SHIFT(log
);
804 if (head_blk
>= num_scan_bblks
) {
805 start_blk
= head_blk
- num_scan_bblks
; /* don't read head_blk */
807 /* start ptr at last block ptr before head_blk */
808 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
809 &head_blk
, 0)) == -1) {
810 error
= XFS_ERROR(EIO
);
816 ASSERT(head_blk
<= INT_MAX
);
817 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
818 &head_blk
, 0)) == -1) {
819 /* We hit the beginning of the log during our search */
820 start_blk
= log_bbnum
- (num_scan_bblks
- head_blk
);
822 ASSERT(start_blk
<= INT_MAX
&&
823 (xfs_daddr_t
) log_bbnum
-start_blk
>= 0);
824 ASSERT(head_blk
<= INT_MAX
);
825 if ((error
= xlog_find_verify_log_record(log
,
827 (int)head_blk
)) == -1) {
828 error
= XFS_ERROR(EIO
);
832 if (new_blk
!= log_bbnum
)
839 if (head_blk
== log_bbnum
)
840 *return_head_blk
= 0;
842 *return_head_blk
= head_blk
;
844 * When returning here, we have a good block number. Bad block
845 * means that during a previous crash, we didn't have a clean break
846 * from cycle number N to cycle number N-1. In this case, we need
847 * to find the first block with cycle number N-1.
855 xfs_warn(log
->l_mp
, "failed to find log head");
860 * Find the sync block number or the tail of the log.
862 * This will be the block number of the last record to have its
863 * associated buffers synced to disk. Every log record header has
864 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
865 * to get a sync block number. The only concern is to figure out which
866 * log record header to believe.
868 * The following algorithm uses the log record header with the largest
869 * lsn. The entire log record does not need to be valid. We only care
870 * that the header is valid.
872 * We could speed up search by using current head_blk buffer, but it is not
878 xfs_daddr_t
*head_blk
,
879 xfs_daddr_t
*tail_blk
)
881 xlog_rec_header_t
*rhead
;
882 xlog_op_header_t
*op_head
;
883 xfs_caddr_t offset
= NULL
;
886 xfs_daddr_t umount_data_blk
;
887 xfs_daddr_t after_umount_blk
;
894 * Find previous log record
896 if ((error
= xlog_find_head(log
, head_blk
)))
899 bp
= xlog_get_bp(log
, 1);
902 if (*head_blk
== 0) { /* special case */
903 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
907 if (xlog_get_cycle(offset
) == 0) {
909 /* leave all other log inited values alone */
915 * Search backwards looking for log record header block
917 ASSERT(*head_blk
< INT_MAX
);
918 for (i
= (int)(*head_blk
) - 1; i
>= 0; i
--) {
919 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
923 if (*(__be32
*)offset
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
929 * If we haven't found the log record header block, start looking
930 * again from the end of the physical log. XXXmiken: There should be
931 * a check here to make sure we didn't search more than N blocks in
935 for (i
= log
->l_logBBsize
- 1; i
>= (int)(*head_blk
); i
--) {
936 error
= xlog_bread(log
, i
, 1, bp
, &offset
);
940 if (*(__be32
*)offset
==
941 cpu_to_be32(XLOG_HEADER_MAGIC_NUM
)) {
948 xfs_warn(log
->l_mp
, "%s: couldn't find sync record", __func__
);
950 return XFS_ERROR(EIO
);
953 /* find blk_no of tail of log */
954 rhead
= (xlog_rec_header_t
*)offset
;
955 *tail_blk
= BLOCK_LSN(be64_to_cpu(rhead
->h_tail_lsn
));
958 * Reset log values according to the state of the log when we
959 * crashed. In the case where head_blk == 0, we bump curr_cycle
960 * one because the next write starts a new cycle rather than
961 * continuing the cycle of the last good log record. At this
962 * point we have guaranteed that all partial log records have been
963 * accounted for. Therefore, we know that the last good log record
964 * written was complete and ended exactly on the end boundary
965 * of the physical log.
967 log
->l_prev_block
= i
;
968 log
->l_curr_block
= (int)*head_blk
;
969 log
->l_curr_cycle
= be32_to_cpu(rhead
->h_cycle
);
972 atomic64_set(&log
->l_tail_lsn
, be64_to_cpu(rhead
->h_tail_lsn
));
973 atomic64_set(&log
->l_last_sync_lsn
, be64_to_cpu(rhead
->h_lsn
));
974 xlog_assign_grant_head(&log
->l_grant_reserve_head
, log
->l_curr_cycle
,
975 BBTOB(log
->l_curr_block
));
976 xlog_assign_grant_head(&log
->l_grant_write_head
, log
->l_curr_cycle
,
977 BBTOB(log
->l_curr_block
));
980 * Look for unmount record. If we find it, then we know there
981 * was a clean unmount. Since 'i' could be the last block in
982 * the physical log, we convert to a log block before comparing
985 * Save the current tail lsn to use to pass to
986 * xlog_clear_stale_blocks() below. We won't want to clear the
987 * unmount record if there is one, so we pass the lsn of the
988 * unmount record rather than the block after it.
990 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
991 int h_size
= be32_to_cpu(rhead
->h_size
);
992 int h_version
= be32_to_cpu(rhead
->h_version
);
994 if ((h_version
& XLOG_VERSION_2
) &&
995 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
996 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
997 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
1005 after_umount_blk
= (i
+ hblks
+ (int)
1006 BTOBB(be32_to_cpu(rhead
->h_len
))) % log
->l_logBBsize
;
1007 tail_lsn
= atomic64_read(&log
->l_tail_lsn
);
1008 if (*head_blk
== after_umount_blk
&&
1009 be32_to_cpu(rhead
->h_num_logops
) == 1) {
1010 umount_data_blk
= (i
+ hblks
) % log
->l_logBBsize
;
1011 error
= xlog_bread(log
, umount_data_blk
, 1, bp
, &offset
);
1015 op_head
= (xlog_op_header_t
*)offset
;
1016 if (op_head
->oh_flags
& XLOG_UNMOUNT_TRANS
) {
1018 * Set tail and last sync so that newly written
1019 * log records will point recovery to after the
1020 * current unmount record.
1022 xlog_assign_atomic_lsn(&log
->l_tail_lsn
,
1023 log
->l_curr_cycle
, after_umount_blk
);
1024 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
,
1025 log
->l_curr_cycle
, after_umount_blk
);
1026 *tail_blk
= after_umount_blk
;
1029 * Note that the unmount was clean. If the unmount
1030 * was not clean, we need to know this to rebuild the
1031 * superblock counters from the perag headers if we
1032 * have a filesystem using non-persistent counters.
1034 log
->l_mp
->m_flags
|= XFS_MOUNT_WAS_CLEAN
;
1039 * Make sure that there are no blocks in front of the head
1040 * with the same cycle number as the head. This can happen
1041 * because we allow multiple outstanding log writes concurrently,
1042 * and the later writes might make it out before earlier ones.
1044 * We use the lsn from before modifying it so that we'll never
1045 * overwrite the unmount record after a clean unmount.
1047 * Do this only if we are going to recover the filesystem
1049 * NOTE: This used to say "if (!readonly)"
1050 * However on Linux, we can & do recover a read-only filesystem.
1051 * We only skip recovery if NORECOVERY is specified on mount,
1052 * in which case we would not be here.
1054 * But... if the -device- itself is readonly, just skip this.
1055 * We can't recover this device anyway, so it won't matter.
1057 if (!xfs_readonly_buftarg(log
->l_mp
->m_logdev_targp
))
1058 error
= xlog_clear_stale_blocks(log
, tail_lsn
);
1064 xfs_warn(log
->l_mp
, "failed to locate log tail");
1069 * Is the log zeroed at all?
1071 * The last binary search should be changed to perform an X block read
1072 * once X becomes small enough. You can then search linearly through
1073 * the X blocks. This will cut down on the number of reads we need to do.
1075 * If the log is partially zeroed, this routine will pass back the blkno
1076 * of the first block with cycle number 0. It won't have a complete LR
1080 * 0 => the log is completely written to
1081 * -1 => use *blk_no as the first block of the log
1082 * >0 => error has occurred
1087 xfs_daddr_t
*blk_no
)
1091 uint first_cycle
, last_cycle
;
1092 xfs_daddr_t new_blk
, last_blk
, start_blk
;
1093 xfs_daddr_t num_scan_bblks
;
1094 int error
, log_bbnum
= log
->l_logBBsize
;
1098 /* check totally zeroed log */
1099 bp
= xlog_get_bp(log
, 1);
1102 error
= xlog_bread(log
, 0, 1, bp
, &offset
);
1106 first_cycle
= xlog_get_cycle(offset
);
1107 if (first_cycle
== 0) { /* completely zeroed log */
1113 /* check partially zeroed log */
1114 error
= xlog_bread(log
, log_bbnum
-1, 1, bp
, &offset
);
1118 last_cycle
= xlog_get_cycle(offset
);
1119 if (last_cycle
!= 0) { /* log completely written to */
1122 } else if (first_cycle
!= 1) {
1124 * If the cycle of the last block is zero, the cycle of
1125 * the first block must be 1. If it's not, maybe we're
1126 * not looking at a log... Bail out.
1129 "Log inconsistent or not a log (last==0, first!=1)");
1130 return XFS_ERROR(EINVAL
);
1133 /* we have a partially zeroed log */
1134 last_blk
= log_bbnum
-1;
1135 if ((error
= xlog_find_cycle_start(log
, bp
, 0, &last_blk
, 0)))
1139 * Validate the answer. Because there is no way to guarantee that
1140 * the entire log is made up of log records which are the same size,
1141 * we scan over the defined maximum blocks. At this point, the maximum
1142 * is not chosen to mean anything special. XXXmiken
1144 num_scan_bblks
= XLOG_TOTAL_REC_SHIFT(log
);
1145 ASSERT(num_scan_bblks
<= INT_MAX
);
1147 if (last_blk
< num_scan_bblks
)
1148 num_scan_bblks
= last_blk
;
1149 start_blk
= last_blk
- num_scan_bblks
;
1152 * We search for any instances of cycle number 0 that occur before
1153 * our current estimate of the head. What we're trying to detect is
1154 * 1 ... | 0 | 1 | 0...
1155 * ^ binary search ends here
1157 if ((error
= xlog_find_verify_cycle(log
, start_blk
,
1158 (int)num_scan_bblks
, 0, &new_blk
)))
1164 * Potentially backup over partial log record write. We don't need
1165 * to search the end of the log because we know it is zero.
1167 if ((error
= xlog_find_verify_log_record(log
, start_blk
,
1168 &last_blk
, 0)) == -1) {
1169 error
= XFS_ERROR(EIO
);
1183 * These are simple subroutines used by xlog_clear_stale_blocks() below
1184 * to initialize a buffer full of empty log record headers and write
1185 * them into the log.
1196 xlog_rec_header_t
*recp
= (xlog_rec_header_t
*)buf
;
1198 memset(buf
, 0, BBSIZE
);
1199 recp
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1200 recp
->h_cycle
= cpu_to_be32(cycle
);
1201 recp
->h_version
= cpu_to_be32(
1202 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1203 recp
->h_lsn
= cpu_to_be64(xlog_assign_lsn(cycle
, block
));
1204 recp
->h_tail_lsn
= cpu_to_be64(xlog_assign_lsn(tail_cycle
, tail_block
));
1205 recp
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1206 memcpy(&recp
->h_fs_uuid
, &log
->l_mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1210 xlog_write_log_records(
1221 int sectbb
= log
->l_sectBBsize
;
1222 int end_block
= start_block
+ blocks
;
1228 * Greedily allocate a buffer big enough to handle the full
1229 * range of basic blocks to be written. If that fails, try
1230 * a smaller size. We need to be able to write at least a
1231 * log sector, or we're out of luck.
1233 bufblks
= 1 << ffs(blocks
);
1234 while (!(bp
= xlog_get_bp(log
, bufblks
))) {
1236 if (bufblks
< sectbb
)
1240 /* We may need to do a read at the start to fill in part of
1241 * the buffer in the starting sector not covered by the first
1244 balign
= round_down(start_block
, sectbb
);
1245 if (balign
!= start_block
) {
1246 error
= xlog_bread_noalign(log
, start_block
, 1, bp
);
1250 j
= start_block
- balign
;
1253 for (i
= start_block
; i
< end_block
; i
+= bufblks
) {
1254 int bcount
, endcount
;
1256 bcount
= min(bufblks
, end_block
- start_block
);
1257 endcount
= bcount
- j
;
1259 /* We may need to do a read at the end to fill in part of
1260 * the buffer in the final sector not covered by the write.
1261 * If this is the same sector as the above read, skip it.
1263 ealign
= round_down(end_block
, sectbb
);
1264 if (j
== 0 && (start_block
+ endcount
> ealign
)) {
1265 offset
= XFS_BUF_PTR(bp
) + BBTOB(ealign
- start_block
);
1266 error
= xlog_bread_offset(log
, ealign
, sectbb
,
1273 offset
= xlog_align(log
, start_block
, endcount
, bp
);
1274 for (; j
< endcount
; j
++) {
1275 xlog_add_record(log
, offset
, cycle
, i
+j
,
1276 tail_cycle
, tail_block
);
1279 error
= xlog_bwrite(log
, start_block
, endcount
, bp
);
1282 start_block
+= endcount
;
1292 * This routine is called to blow away any incomplete log writes out
1293 * in front of the log head. We do this so that we won't become confused
1294 * if we come up, write only a little bit more, and then crash again.
1295 * If we leave the partial log records out there, this situation could
1296 * cause us to think those partial writes are valid blocks since they
1297 * have the current cycle number. We get rid of them by overwriting them
1298 * with empty log records with the old cycle number rather than the
1301 * The tail lsn is passed in rather than taken from
1302 * the log so that we will not write over the unmount record after a
1303 * clean unmount in a 512 block log. Doing so would leave the log without
1304 * any valid log records in it until a new one was written. If we crashed
1305 * during that time we would not be able to recover.
1308 xlog_clear_stale_blocks(
1312 int tail_cycle
, head_cycle
;
1313 int tail_block
, head_block
;
1314 int tail_distance
, max_distance
;
1318 tail_cycle
= CYCLE_LSN(tail_lsn
);
1319 tail_block
= BLOCK_LSN(tail_lsn
);
1320 head_cycle
= log
->l_curr_cycle
;
1321 head_block
= log
->l_curr_block
;
1324 * Figure out the distance between the new head of the log
1325 * and the tail. We want to write over any blocks beyond the
1326 * head that we may have written just before the crash, but
1327 * we don't want to overwrite the tail of the log.
1329 if (head_cycle
== tail_cycle
) {
1331 * The tail is behind the head in the physical log,
1332 * so the distance from the head to the tail is the
1333 * distance from the head to the end of the log plus
1334 * the distance from the beginning of the log to the
1337 if (unlikely(head_block
< tail_block
|| head_block
>= log
->l_logBBsize
)) {
1338 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1339 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1340 return XFS_ERROR(EFSCORRUPTED
);
1342 tail_distance
= tail_block
+ (log
->l_logBBsize
- head_block
);
1345 * The head is behind the tail in the physical log,
1346 * so the distance from the head to the tail is just
1347 * the tail block minus the head block.
1349 if (unlikely(head_block
>= tail_block
|| head_cycle
!= (tail_cycle
+ 1))){
1350 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1351 XFS_ERRLEVEL_LOW
, log
->l_mp
);
1352 return XFS_ERROR(EFSCORRUPTED
);
1354 tail_distance
= tail_block
- head_block
;
1358 * If the head is right up against the tail, we can't clear
1361 if (tail_distance
<= 0) {
1362 ASSERT(tail_distance
== 0);
1366 max_distance
= XLOG_TOTAL_REC_SHIFT(log
);
1368 * Take the smaller of the maximum amount of outstanding I/O
1369 * we could have and the distance to the tail to clear out.
1370 * We take the smaller so that we don't overwrite the tail and
1371 * we don't waste all day writing from the head to the tail
1374 max_distance
= MIN(max_distance
, tail_distance
);
1376 if ((head_block
+ max_distance
) <= log
->l_logBBsize
) {
1378 * We can stomp all the blocks we need to without
1379 * wrapping around the end of the log. Just do it
1380 * in a single write. Use the cycle number of the
1381 * current cycle minus one so that the log will look like:
1384 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1385 head_block
, max_distance
, tail_cycle
,
1391 * We need to wrap around the end of the physical log in
1392 * order to clear all the blocks. Do it in two separate
1393 * I/Os. The first write should be from the head to the
1394 * end of the physical log, and it should use the current
1395 * cycle number minus one just like above.
1397 distance
= log
->l_logBBsize
- head_block
;
1398 error
= xlog_write_log_records(log
, (head_cycle
- 1),
1399 head_block
, distance
, tail_cycle
,
1406 * Now write the blocks at the start of the physical log.
1407 * This writes the remainder of the blocks we want to clear.
1408 * It uses the current cycle number since we're now on the
1409 * same cycle as the head so that we get:
1410 * n ... n ... | n - 1 ...
1411 * ^^^^^ blocks we're writing
1413 distance
= max_distance
- (log
->l_logBBsize
- head_block
);
1414 error
= xlog_write_log_records(log
, head_cycle
, 0, distance
,
1415 tail_cycle
, tail_block
);
1423 /******************************************************************************
1425 * Log recover routines
1427 ******************************************************************************
1430 STATIC xlog_recover_t
*
1431 xlog_recover_find_tid(
1432 struct hlist_head
*head
,
1435 xlog_recover_t
*trans
;
1436 struct hlist_node
*n
;
1438 hlist_for_each_entry(trans
, n
, head
, r_list
) {
1439 if (trans
->r_log_tid
== tid
)
1446 xlog_recover_new_tid(
1447 struct hlist_head
*head
,
1451 xlog_recover_t
*trans
;
1453 trans
= kmem_zalloc(sizeof(xlog_recover_t
), KM_SLEEP
);
1454 trans
->r_log_tid
= tid
;
1456 INIT_LIST_HEAD(&trans
->r_itemq
);
1458 INIT_HLIST_NODE(&trans
->r_list
);
1459 hlist_add_head(&trans
->r_list
, head
);
1463 xlog_recover_add_item(
1464 struct list_head
*head
)
1466 xlog_recover_item_t
*item
;
1468 item
= kmem_zalloc(sizeof(xlog_recover_item_t
), KM_SLEEP
);
1469 INIT_LIST_HEAD(&item
->ri_list
);
1470 list_add_tail(&item
->ri_list
, head
);
1474 xlog_recover_add_to_cont_trans(
1476 xlog_recover_t
*trans
,
1480 xlog_recover_item_t
*item
;
1481 xfs_caddr_t ptr
, old_ptr
;
1484 if (list_empty(&trans
->r_itemq
)) {
1485 /* finish copying rest of trans header */
1486 xlog_recover_add_item(&trans
->r_itemq
);
1487 ptr
= (xfs_caddr_t
) &trans
->r_theader
+
1488 sizeof(xfs_trans_header_t
) - len
;
1489 memcpy(ptr
, dp
, len
); /* d, s, l */
1492 /* take the tail entry */
1493 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1495 old_ptr
= item
->ri_buf
[item
->ri_cnt
-1].i_addr
;
1496 old_len
= item
->ri_buf
[item
->ri_cnt
-1].i_len
;
1498 ptr
= kmem_realloc(old_ptr
, len
+old_len
, old_len
, 0u);
1499 memcpy(&ptr
[old_len
], dp
, len
); /* d, s, l */
1500 item
->ri_buf
[item
->ri_cnt
-1].i_len
+= len
;
1501 item
->ri_buf
[item
->ri_cnt
-1].i_addr
= ptr
;
1502 trace_xfs_log_recover_item_add_cont(log
, trans
, item
, 0);
1507 * The next region to add is the start of a new region. It could be
1508 * a whole region or it could be the first part of a new region. Because
1509 * of this, the assumption here is that the type and size fields of all
1510 * format structures fit into the first 32 bits of the structure.
1512 * This works because all regions must be 32 bit aligned. Therefore, we
1513 * either have both fields or we have neither field. In the case we have
1514 * neither field, the data part of the region is zero length. We only have
1515 * a log_op_header and can throw away the header since a new one will appear
1516 * later. If we have at least 4 bytes, then we can determine how many regions
1517 * will appear in the current log item.
1520 xlog_recover_add_to_trans(
1522 xlog_recover_t
*trans
,
1526 xfs_inode_log_format_t
*in_f
; /* any will do */
1527 xlog_recover_item_t
*item
;
1532 if (list_empty(&trans
->r_itemq
)) {
1533 /* we need to catch log corruptions here */
1534 if (*(uint
*)dp
!= XFS_TRANS_HEADER_MAGIC
) {
1535 xfs_warn(log
->l_mp
, "%s: bad header magic number",
1538 return XFS_ERROR(EIO
);
1540 if (len
== sizeof(xfs_trans_header_t
))
1541 xlog_recover_add_item(&trans
->r_itemq
);
1542 memcpy(&trans
->r_theader
, dp
, len
); /* d, s, l */
1546 ptr
= kmem_alloc(len
, KM_SLEEP
);
1547 memcpy(ptr
, dp
, len
);
1548 in_f
= (xfs_inode_log_format_t
*)ptr
;
1550 /* take the tail entry */
1551 item
= list_entry(trans
->r_itemq
.prev
, xlog_recover_item_t
, ri_list
);
1552 if (item
->ri_total
!= 0 &&
1553 item
->ri_total
== item
->ri_cnt
) {
1554 /* tail item is in use, get a new one */
1555 xlog_recover_add_item(&trans
->r_itemq
);
1556 item
= list_entry(trans
->r_itemq
.prev
,
1557 xlog_recover_item_t
, ri_list
);
1560 if (item
->ri_total
== 0) { /* first region to be added */
1561 if (in_f
->ilf_size
== 0 ||
1562 in_f
->ilf_size
> XLOG_MAX_REGIONS_IN_ITEM
) {
1564 "bad number of regions (%d) in inode log format",
1567 return XFS_ERROR(EIO
);
1570 item
->ri_total
= in_f
->ilf_size
;
1572 kmem_zalloc(item
->ri_total
* sizeof(xfs_log_iovec_t
),
1575 ASSERT(item
->ri_total
> item
->ri_cnt
);
1576 /* Description region is ri_buf[0] */
1577 item
->ri_buf
[item
->ri_cnt
].i_addr
= ptr
;
1578 item
->ri_buf
[item
->ri_cnt
].i_len
= len
;
1580 trace_xfs_log_recover_item_add(log
, trans
, item
, 0);
1585 * Sort the log items in the transaction. Cancelled buffers need
1586 * to be put first so they are processed before any items that might
1587 * modify the buffers. If they are cancelled, then the modifications
1588 * don't need to be replayed.
1591 xlog_recover_reorder_trans(
1593 xlog_recover_t
*trans
,
1596 xlog_recover_item_t
*item
, *n
;
1597 LIST_HEAD(sort_list
);
1599 list_splice_init(&trans
->r_itemq
, &sort_list
);
1600 list_for_each_entry_safe(item
, n
, &sort_list
, ri_list
) {
1601 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1603 switch (ITEM_TYPE(item
)) {
1605 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1606 trace_xfs_log_recover_item_reorder_head(log
,
1608 list_move(&item
->ri_list
, &trans
->r_itemq
);
1613 case XFS_LI_QUOTAOFF
:
1616 trace_xfs_log_recover_item_reorder_tail(log
,
1618 list_move_tail(&item
->ri_list
, &trans
->r_itemq
);
1622 "%s: unrecognized type of log operation",
1625 return XFS_ERROR(EIO
);
1628 ASSERT(list_empty(&sort_list
));
1633 * Build up the table of buf cancel records so that we don't replay
1634 * cancelled data in the second pass. For buffer records that are
1635 * not cancel records, there is nothing to do here so we just return.
1637 * If we get a cancel record which is already in the table, this indicates
1638 * that the buffer was cancelled multiple times. In order to ensure
1639 * that during pass 2 we keep the record in the table until we reach its
1640 * last occurrence in the log, we keep a reference count in the cancel
1641 * record in the table to tell us how many times we expect to see this
1642 * record during the second pass.
1645 xlog_recover_buffer_pass1(
1647 xlog_recover_item_t
*item
)
1649 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
1650 struct list_head
*bucket
;
1651 struct xfs_buf_cancel
*bcp
;
1654 * If this isn't a cancel buffer item, then just return.
1656 if (!(buf_f
->blf_flags
& XFS_BLF_CANCEL
)) {
1657 trace_xfs_log_recover_buf_not_cancel(log
, buf_f
);
1662 * Insert an xfs_buf_cancel record into the hash table of them.
1663 * If there is already an identical record, bump its reference count.
1665 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, buf_f
->blf_blkno
);
1666 list_for_each_entry(bcp
, bucket
, bc_list
) {
1667 if (bcp
->bc_blkno
== buf_f
->blf_blkno
&&
1668 bcp
->bc_len
== buf_f
->blf_len
) {
1670 trace_xfs_log_recover_buf_cancel_ref_inc(log
, buf_f
);
1675 bcp
= kmem_alloc(sizeof(struct xfs_buf_cancel
), KM_SLEEP
);
1676 bcp
->bc_blkno
= buf_f
->blf_blkno
;
1677 bcp
->bc_len
= buf_f
->blf_len
;
1678 bcp
->bc_refcount
= 1;
1679 list_add_tail(&bcp
->bc_list
, bucket
);
1681 trace_xfs_log_recover_buf_cancel_add(log
, buf_f
);
1686 * Check to see whether the buffer being recovered has a corresponding
1687 * entry in the buffer cancel record table. If it does then return 1
1688 * so that it will be cancelled, otherwise return 0. If the buffer is
1689 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1690 * the refcount on the entry in the table and remove it from the table
1691 * if this is the last reference.
1693 * We remove the cancel record from the table when we encounter its
1694 * last occurrence in the log so that if the same buffer is re-used
1695 * again after its last cancellation we actually replay the changes
1696 * made at that point.
1699 xlog_check_buffer_cancelled(
1705 struct list_head
*bucket
;
1706 struct xfs_buf_cancel
*bcp
;
1708 if (log
->l_buf_cancel_table
== NULL
) {
1710 * There is nothing in the table built in pass one,
1711 * so this buffer must not be cancelled.
1713 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1718 * Search for an entry in the cancel table that matches our buffer.
1720 bucket
= XLOG_BUF_CANCEL_BUCKET(log
, blkno
);
1721 list_for_each_entry(bcp
, bucket
, bc_list
) {
1722 if (bcp
->bc_blkno
== blkno
&& bcp
->bc_len
== len
)
1727 * We didn't find a corresponding entry in the table, so return 0 so
1728 * that the buffer is NOT cancelled.
1730 ASSERT(!(flags
& XFS_BLF_CANCEL
));
1735 * We've go a match, so return 1 so that the recovery of this buffer
1736 * is cancelled. If this buffer is actually a buffer cancel log
1737 * item, then decrement the refcount on the one in the table and
1738 * remove it if this is the last reference.
1740 if (flags
& XFS_BLF_CANCEL
) {
1741 if (--bcp
->bc_refcount
== 0) {
1742 list_del(&bcp
->bc_list
);
1750 * Perform recovery for a buffer full of inodes. In these buffers, the only
1751 * data which should be recovered is that which corresponds to the
1752 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1753 * data for the inodes is always logged through the inodes themselves rather
1754 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1756 * The only time when buffers full of inodes are fully recovered is when the
1757 * buffer is full of newly allocated inodes. In this case the buffer will
1758 * not be marked as an inode buffer and so will be sent to
1759 * xlog_recover_do_reg_buffer() below during recovery.
1762 xlog_recover_do_inode_buffer(
1763 struct xfs_mount
*mp
,
1764 xlog_recover_item_t
*item
,
1766 xfs_buf_log_format_t
*buf_f
)
1772 int reg_buf_offset
= 0;
1773 int reg_buf_bytes
= 0;
1774 int next_unlinked_offset
;
1776 xfs_agino_t
*logged_nextp
;
1777 xfs_agino_t
*buffer_nextp
;
1779 trace_xfs_log_recover_buf_inode_buf(mp
->m_log
, buf_f
);
1781 inodes_per_buf
= XFS_BUF_COUNT(bp
) >> mp
->m_sb
.sb_inodelog
;
1782 for (i
= 0; i
< inodes_per_buf
; i
++) {
1783 next_unlinked_offset
= (i
* mp
->m_sb
.sb_inodesize
) +
1784 offsetof(xfs_dinode_t
, di_next_unlinked
);
1786 while (next_unlinked_offset
>=
1787 (reg_buf_offset
+ reg_buf_bytes
)) {
1789 * The next di_next_unlinked field is beyond
1790 * the current logged region. Find the next
1791 * logged region that contains or is beyond
1792 * the current di_next_unlinked field.
1795 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1796 buf_f
->blf_map_size
, bit
);
1799 * If there are no more logged regions in the
1800 * buffer, then we're done.
1805 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1806 buf_f
->blf_map_size
, bit
);
1808 reg_buf_offset
= bit
<< XFS_BLF_SHIFT
;
1809 reg_buf_bytes
= nbits
<< XFS_BLF_SHIFT
;
1814 * If the current logged region starts after the current
1815 * di_next_unlinked field, then move on to the next
1816 * di_next_unlinked field.
1818 if (next_unlinked_offset
< reg_buf_offset
)
1821 ASSERT(item
->ri_buf
[item_index
].i_addr
!= NULL
);
1822 ASSERT((item
->ri_buf
[item_index
].i_len
% XFS_BLF_CHUNK
) == 0);
1823 ASSERT((reg_buf_offset
+ reg_buf_bytes
) <= XFS_BUF_COUNT(bp
));
1826 * The current logged region contains a copy of the
1827 * current di_next_unlinked field. Extract its value
1828 * and copy it to the buffer copy.
1830 logged_nextp
= item
->ri_buf
[item_index
].i_addr
+
1831 next_unlinked_offset
- reg_buf_offset
;
1832 if (unlikely(*logged_nextp
== 0)) {
1834 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1835 "Trying to replay bad (0) inode di_next_unlinked field.",
1837 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1838 XFS_ERRLEVEL_LOW
, mp
);
1839 return XFS_ERROR(EFSCORRUPTED
);
1842 buffer_nextp
= (xfs_agino_t
*)xfs_buf_offset(bp
,
1843 next_unlinked_offset
);
1844 *buffer_nextp
= *logged_nextp
;
1851 * Perform a 'normal' buffer recovery. Each logged region of the
1852 * buffer should be copied over the corresponding region in the
1853 * given buffer. The bitmap in the buf log format structure indicates
1854 * where to place the logged data.
1857 xlog_recover_do_reg_buffer(
1858 struct xfs_mount
*mp
,
1859 xlog_recover_item_t
*item
,
1861 xfs_buf_log_format_t
*buf_f
)
1868 trace_xfs_log_recover_buf_reg_buf(mp
->m_log
, buf_f
);
1871 i
= 1; /* 0 is the buf format structure */
1873 bit
= xfs_next_bit(buf_f
->blf_data_map
,
1874 buf_f
->blf_map_size
, bit
);
1877 nbits
= xfs_contig_bits(buf_f
->blf_data_map
,
1878 buf_f
->blf_map_size
, bit
);
1880 ASSERT(item
->ri_buf
[i
].i_addr
!= NULL
);
1881 ASSERT(item
->ri_buf
[i
].i_len
% XFS_BLF_CHUNK
== 0);
1882 ASSERT(XFS_BUF_COUNT(bp
) >=
1883 ((uint
)bit
<< XFS_BLF_SHIFT
)+(nbits
<<XFS_BLF_SHIFT
));
1886 * Do a sanity check if this is a dquot buffer. Just checking
1887 * the first dquot in the buffer should do. XXXThis is
1888 * probably a good thing to do for other buf types also.
1891 if (buf_f
->blf_flags
&
1892 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
1893 if (item
->ri_buf
[i
].i_addr
== NULL
) {
1895 "XFS: NULL dquot in %s.", __func__
);
1898 if (item
->ri_buf
[i
].i_len
< sizeof(xfs_disk_dquot_t
)) {
1900 "XFS: dquot too small (%d) in %s.",
1901 item
->ri_buf
[i
].i_len
, __func__
);
1904 error
= xfs_qm_dqcheck(mp
, item
->ri_buf
[i
].i_addr
,
1905 -1, 0, XFS_QMOPT_DOWARN
,
1906 "dquot_buf_recover");
1911 memcpy(xfs_buf_offset(bp
,
1912 (uint
)bit
<< XFS_BLF_SHIFT
), /* dest */
1913 item
->ri_buf
[i
].i_addr
, /* source */
1914 nbits
<<XFS_BLF_SHIFT
); /* length */
1920 /* Shouldn't be any more regions */
1921 ASSERT(i
== item
->ri_total
);
1925 * Do some primitive error checking on ondisk dquot data structures.
1929 struct xfs_mount
*mp
,
1930 xfs_disk_dquot_t
*ddq
,
1932 uint type
, /* used only when IO_dorepair is true */
1936 xfs_dqblk_t
*d
= (xfs_dqblk_t
*)ddq
;
1940 * We can encounter an uninitialized dquot buffer for 2 reasons:
1941 * 1. If we crash while deleting the quotainode(s), and those blks got
1942 * used for user data. This is because we take the path of regular
1943 * file deletion; however, the size field of quotainodes is never
1944 * updated, so all the tricks that we play in itruncate_finish
1945 * don't quite matter.
1947 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1948 * But the allocation will be replayed so we'll end up with an
1949 * uninitialized quota block.
1951 * This is all fine; things are still consistent, and we haven't lost
1952 * any quota information. Just don't complain about bad dquot blks.
1954 if (ddq
->d_magic
!= cpu_to_be16(XFS_DQUOT_MAGIC
)) {
1955 if (flags
& XFS_QMOPT_DOWARN
)
1957 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
1958 str
, id
, be16_to_cpu(ddq
->d_magic
), XFS_DQUOT_MAGIC
);
1961 if (ddq
->d_version
!= XFS_DQUOT_VERSION
) {
1962 if (flags
& XFS_QMOPT_DOWARN
)
1964 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
1965 str
, id
, ddq
->d_version
, XFS_DQUOT_VERSION
);
1969 if (ddq
->d_flags
!= XFS_DQ_USER
&&
1970 ddq
->d_flags
!= XFS_DQ_PROJ
&&
1971 ddq
->d_flags
!= XFS_DQ_GROUP
) {
1972 if (flags
& XFS_QMOPT_DOWARN
)
1974 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
1975 str
, id
, ddq
->d_flags
);
1979 if (id
!= -1 && id
!= be32_to_cpu(ddq
->d_id
)) {
1980 if (flags
& XFS_QMOPT_DOWARN
)
1982 "%s : ondisk-dquot 0x%p, ID mismatch: "
1983 "0x%x expected, found id 0x%x",
1984 str
, ddq
, id
, be32_to_cpu(ddq
->d_id
));
1988 if (!errs
&& ddq
->d_id
) {
1989 if (ddq
->d_blk_softlimit
&&
1990 be64_to_cpu(ddq
->d_bcount
) >=
1991 be64_to_cpu(ddq
->d_blk_softlimit
)) {
1992 if (!ddq
->d_btimer
) {
1993 if (flags
& XFS_QMOPT_DOWARN
)
1995 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
1996 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2000 if (ddq
->d_ino_softlimit
&&
2001 be64_to_cpu(ddq
->d_icount
) >=
2002 be64_to_cpu(ddq
->d_ino_softlimit
)) {
2003 if (!ddq
->d_itimer
) {
2004 if (flags
& XFS_QMOPT_DOWARN
)
2006 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
2007 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2011 if (ddq
->d_rtb_softlimit
&&
2012 be64_to_cpu(ddq
->d_rtbcount
) >=
2013 be64_to_cpu(ddq
->d_rtb_softlimit
)) {
2014 if (!ddq
->d_rtbtimer
) {
2015 if (flags
& XFS_QMOPT_DOWARN
)
2017 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
2018 str
, (int)be32_to_cpu(ddq
->d_id
), ddq
);
2024 if (!errs
|| !(flags
& XFS_QMOPT_DQREPAIR
))
2027 if (flags
& XFS_QMOPT_DOWARN
)
2028 xfs_notice(mp
, "Re-initializing dquot ID 0x%x", id
);
2031 * Typically, a repair is only requested by quotacheck.
2034 ASSERT(flags
& XFS_QMOPT_DQREPAIR
);
2035 memset(d
, 0, sizeof(xfs_dqblk_t
));
2037 d
->dd_diskdq
.d_magic
= cpu_to_be16(XFS_DQUOT_MAGIC
);
2038 d
->dd_diskdq
.d_version
= XFS_DQUOT_VERSION
;
2039 d
->dd_diskdq
.d_flags
= type
;
2040 d
->dd_diskdq
.d_id
= cpu_to_be32(id
);
2046 * Perform a dquot buffer recovery.
2047 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2048 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2049 * Else, treat it as a regular buffer and do recovery.
2052 xlog_recover_do_dquot_buffer(
2055 xlog_recover_item_t
*item
,
2057 xfs_buf_log_format_t
*buf_f
)
2061 trace_xfs_log_recover_buf_dquot_buf(log
, buf_f
);
2064 * Filesystems are required to send in quota flags at mount time.
2066 if (mp
->m_qflags
== 0) {
2071 if (buf_f
->blf_flags
& XFS_BLF_UDQUOT_BUF
)
2072 type
|= XFS_DQ_USER
;
2073 if (buf_f
->blf_flags
& XFS_BLF_PDQUOT_BUF
)
2074 type
|= XFS_DQ_PROJ
;
2075 if (buf_f
->blf_flags
& XFS_BLF_GDQUOT_BUF
)
2076 type
|= XFS_DQ_GROUP
;
2078 * This type of quotas was turned off, so ignore this buffer
2080 if (log
->l_quotaoffs_flag
& type
)
2083 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2087 * This routine replays a modification made to a buffer at runtime.
2088 * There are actually two types of buffer, regular and inode, which
2089 * are handled differently. Inode buffers are handled differently
2090 * in that we only recover a specific set of data from them, namely
2091 * the inode di_next_unlinked fields. This is because all other inode
2092 * data is actually logged via inode records and any data we replay
2093 * here which overlaps that may be stale.
2095 * When meta-data buffers are freed at run time we log a buffer item
2096 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2097 * of the buffer in the log should not be replayed at recovery time.
2098 * This is so that if the blocks covered by the buffer are reused for
2099 * file data before we crash we don't end up replaying old, freed
2100 * meta-data into a user's file.
2102 * To handle the cancellation of buffer log items, we make two passes
2103 * over the log during recovery. During the first we build a table of
2104 * those buffers which have been cancelled, and during the second we
2105 * only replay those buffers which do not have corresponding cancel
2106 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2107 * for more details on the implementation of the table of cancel records.
2110 xlog_recover_buffer_pass2(
2112 xlog_recover_item_t
*item
)
2114 xfs_buf_log_format_t
*buf_f
= item
->ri_buf
[0].i_addr
;
2115 xfs_mount_t
*mp
= log
->l_mp
;
2121 * In this pass we only want to recover all the buffers which have
2122 * not been cancelled and are not cancellation buffers themselves.
2124 if (xlog_check_buffer_cancelled(log
, buf_f
->blf_blkno
,
2125 buf_f
->blf_len
, buf_f
->blf_flags
)) {
2126 trace_xfs_log_recover_buf_cancel(log
, buf_f
);
2130 trace_xfs_log_recover_buf_recover(log
, buf_f
);
2132 buf_flags
= XBF_LOCK
;
2133 if (!(buf_f
->blf_flags
& XFS_BLF_INODE_BUF
))
2134 buf_flags
|= XBF_MAPPED
;
2136 bp
= xfs_buf_read(mp
->m_ddev_targp
, buf_f
->blf_blkno
, buf_f
->blf_len
,
2138 error
= xfs_buf_geterror(bp
);
2140 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp
,
2141 bp
, buf_f
->blf_blkno
);
2146 if (buf_f
->blf_flags
& XFS_BLF_INODE_BUF
) {
2147 error
= xlog_recover_do_inode_buffer(mp
, item
, bp
, buf_f
);
2148 } else if (buf_f
->blf_flags
&
2149 (XFS_BLF_UDQUOT_BUF
|XFS_BLF_PDQUOT_BUF
|XFS_BLF_GDQUOT_BUF
)) {
2150 xlog_recover_do_dquot_buffer(mp
, log
, item
, bp
, buf_f
);
2152 xlog_recover_do_reg_buffer(mp
, item
, bp
, buf_f
);
2155 return XFS_ERROR(error
);
2158 * Perform delayed write on the buffer. Asynchronous writes will be
2159 * slower when taking into account all the buffers to be flushed.
2161 * Also make sure that only inode buffers with good sizes stay in
2162 * the buffer cache. The kernel moves inodes in buffers of 1 block
2163 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2164 * buffers in the log can be a different size if the log was generated
2165 * by an older kernel using unclustered inode buffers or a newer kernel
2166 * running with a different inode cluster size. Regardless, if the
2167 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2168 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2169 * the buffer out of the buffer cache so that the buffer won't
2170 * overlap with future reads of those inodes.
2172 if (XFS_DINODE_MAGIC
==
2173 be16_to_cpu(*((__be16
*)xfs_buf_offset(bp
, 0))) &&
2174 (XFS_BUF_COUNT(bp
) != MAX(log
->l_mp
->m_sb
.sb_blocksize
,
2175 (__uint32_t
)XFS_INODE_CLUSTER_SIZE(log
->l_mp
)))) {
2177 error
= xfs_bwrite(mp
, bp
);
2179 ASSERT(bp
->b_target
->bt_mount
== mp
);
2180 bp
->b_iodone
= xlog_recover_iodone
;
2181 xfs_bdwrite(mp
, bp
);
2188 xlog_recover_inode_pass2(
2190 xlog_recover_item_t
*item
)
2192 xfs_inode_log_format_t
*in_f
;
2193 xfs_mount_t
*mp
= log
->l_mp
;
2202 xfs_icdinode_t
*dicp
;
2205 if (item
->ri_buf
[0].i_len
== sizeof(xfs_inode_log_format_t
)) {
2206 in_f
= item
->ri_buf
[0].i_addr
;
2208 in_f
= kmem_alloc(sizeof(xfs_inode_log_format_t
), KM_SLEEP
);
2210 error
= xfs_inode_item_format_convert(&item
->ri_buf
[0], in_f
);
2216 * Inode buffers can be freed, look out for it,
2217 * and do not replay the inode.
2219 if (xlog_check_buffer_cancelled(log
, in_f
->ilf_blkno
,
2220 in_f
->ilf_len
, 0)) {
2222 trace_xfs_log_recover_inode_cancel(log
, in_f
);
2225 trace_xfs_log_recover_inode_recover(log
, in_f
);
2227 bp
= xfs_buf_read(mp
->m_ddev_targp
, in_f
->ilf_blkno
, in_f
->ilf_len
,
2229 error
= xfs_buf_geterror(bp
);
2231 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp
,
2232 bp
, in_f
->ilf_blkno
);
2236 ASSERT(in_f
->ilf_fields
& XFS_ILOG_CORE
);
2237 dip
= (xfs_dinode_t
*)xfs_buf_offset(bp
, in_f
->ilf_boffset
);
2240 * Make sure the place we're flushing out to really looks
2243 if (unlikely(dip
->di_magic
!= cpu_to_be16(XFS_DINODE_MAGIC
))) {
2246 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2247 __func__
, dip
, bp
, in_f
->ilf_ino
);
2248 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2249 XFS_ERRLEVEL_LOW
, mp
);
2250 error
= EFSCORRUPTED
;
2253 dicp
= item
->ri_buf
[1].i_addr
;
2254 if (unlikely(dicp
->di_magic
!= XFS_DINODE_MAGIC
)) {
2257 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2258 __func__
, item
, in_f
->ilf_ino
);
2259 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2260 XFS_ERRLEVEL_LOW
, mp
);
2261 error
= EFSCORRUPTED
;
2265 /* Skip replay when the on disk inode is newer than the log one */
2266 if (dicp
->di_flushiter
< be16_to_cpu(dip
->di_flushiter
)) {
2268 * Deal with the wrap case, DI_MAX_FLUSH is less
2269 * than smaller numbers
2271 if (be16_to_cpu(dip
->di_flushiter
) == DI_MAX_FLUSH
&&
2272 dicp
->di_flushiter
< (DI_MAX_FLUSH
>> 1)) {
2276 trace_xfs_log_recover_inode_skip(log
, in_f
);
2281 /* Take the opportunity to reset the flush iteration count */
2282 dicp
->di_flushiter
= 0;
2284 if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFREG
)) {
2285 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2286 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
)) {
2287 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2288 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2291 "%s: Bad regular inode log record, rec ptr 0x%p, "
2292 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2293 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2294 error
= EFSCORRUPTED
;
2297 } else if (unlikely((dicp
->di_mode
& S_IFMT
) == S_IFDIR
)) {
2298 if ((dicp
->di_format
!= XFS_DINODE_FMT_EXTENTS
) &&
2299 (dicp
->di_format
!= XFS_DINODE_FMT_BTREE
) &&
2300 (dicp
->di_format
!= XFS_DINODE_FMT_LOCAL
)) {
2301 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2302 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2305 "%s: Bad dir inode log record, rec ptr 0x%p, "
2306 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2307 __func__
, item
, dip
, bp
, in_f
->ilf_ino
);
2308 error
= EFSCORRUPTED
;
2312 if (unlikely(dicp
->di_nextents
+ dicp
->di_anextents
> dicp
->di_nblocks
)){
2313 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2314 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2317 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2318 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2319 __func__
, item
, dip
, bp
, in_f
->ilf_ino
,
2320 dicp
->di_nextents
+ dicp
->di_anextents
,
2322 error
= EFSCORRUPTED
;
2325 if (unlikely(dicp
->di_forkoff
> mp
->m_sb
.sb_inodesize
)) {
2326 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2327 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2330 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2331 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__
,
2332 item
, dip
, bp
, in_f
->ilf_ino
, dicp
->di_forkoff
);
2333 error
= EFSCORRUPTED
;
2336 if (unlikely(item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
))) {
2337 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2338 XFS_ERRLEVEL_LOW
, mp
, dicp
);
2341 "%s: Bad inode log record length %d, rec ptr 0x%p",
2342 __func__
, item
->ri_buf
[1].i_len
, item
);
2343 error
= EFSCORRUPTED
;
2347 /* The core is in in-core format */
2348 xfs_dinode_to_disk(dip
, item
->ri_buf
[1].i_addr
);
2350 /* the rest is in on-disk format */
2351 if (item
->ri_buf
[1].i_len
> sizeof(struct xfs_icdinode
)) {
2352 memcpy((xfs_caddr_t
) dip
+ sizeof(struct xfs_icdinode
),
2353 item
->ri_buf
[1].i_addr
+ sizeof(struct xfs_icdinode
),
2354 item
->ri_buf
[1].i_len
- sizeof(struct xfs_icdinode
));
2357 fields
= in_f
->ilf_fields
;
2358 switch (fields
& (XFS_ILOG_DEV
| XFS_ILOG_UUID
)) {
2360 xfs_dinode_put_rdev(dip
, in_f
->ilf_u
.ilfu_rdev
);
2363 memcpy(XFS_DFORK_DPTR(dip
),
2364 &in_f
->ilf_u
.ilfu_uuid
,
2369 if (in_f
->ilf_size
== 2)
2370 goto write_inode_buffer
;
2371 len
= item
->ri_buf
[2].i_len
;
2372 src
= item
->ri_buf
[2].i_addr
;
2373 ASSERT(in_f
->ilf_size
<= 4);
2374 ASSERT((in_f
->ilf_size
== 3) || (fields
& XFS_ILOG_AFORK
));
2375 ASSERT(!(fields
& XFS_ILOG_DFORK
) ||
2376 (len
== in_f
->ilf_dsize
));
2378 switch (fields
& XFS_ILOG_DFORK
) {
2379 case XFS_ILOG_DDATA
:
2381 memcpy(XFS_DFORK_DPTR(dip
), src
, len
);
2384 case XFS_ILOG_DBROOT
:
2385 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
, len
,
2386 (xfs_bmdr_block_t
*)XFS_DFORK_DPTR(dip
),
2387 XFS_DFORK_DSIZE(dip
, mp
));
2392 * There are no data fork flags set.
2394 ASSERT((fields
& XFS_ILOG_DFORK
) == 0);
2399 * If we logged any attribute data, recover it. There may or
2400 * may not have been any other non-core data logged in this
2403 if (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2404 if (in_f
->ilf_fields
& XFS_ILOG_DFORK
) {
2409 len
= item
->ri_buf
[attr_index
].i_len
;
2410 src
= item
->ri_buf
[attr_index
].i_addr
;
2411 ASSERT(len
== in_f
->ilf_asize
);
2413 switch (in_f
->ilf_fields
& XFS_ILOG_AFORK
) {
2414 case XFS_ILOG_ADATA
:
2416 dest
= XFS_DFORK_APTR(dip
);
2417 ASSERT(len
<= XFS_DFORK_ASIZE(dip
, mp
));
2418 memcpy(dest
, src
, len
);
2421 case XFS_ILOG_ABROOT
:
2422 dest
= XFS_DFORK_APTR(dip
);
2423 xfs_bmbt_to_bmdr(mp
, (struct xfs_btree_block
*)src
,
2424 len
, (xfs_bmdr_block_t
*)dest
,
2425 XFS_DFORK_ASIZE(dip
, mp
));
2429 xfs_warn(log
->l_mp
, "%s: Invalid flag", __func__
);
2438 ASSERT(bp
->b_target
->bt_mount
== mp
);
2439 bp
->b_iodone
= xlog_recover_iodone
;
2440 xfs_bdwrite(mp
, bp
);
2444 return XFS_ERROR(error
);
2448 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2449 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2453 xlog_recover_quotaoff_pass1(
2455 xlog_recover_item_t
*item
)
2457 xfs_qoff_logformat_t
*qoff_f
= item
->ri_buf
[0].i_addr
;
2461 * The logitem format's flag tells us if this was user quotaoff,
2462 * group/project quotaoff or both.
2464 if (qoff_f
->qf_flags
& XFS_UQUOTA_ACCT
)
2465 log
->l_quotaoffs_flag
|= XFS_DQ_USER
;
2466 if (qoff_f
->qf_flags
& XFS_PQUOTA_ACCT
)
2467 log
->l_quotaoffs_flag
|= XFS_DQ_PROJ
;
2468 if (qoff_f
->qf_flags
& XFS_GQUOTA_ACCT
)
2469 log
->l_quotaoffs_flag
|= XFS_DQ_GROUP
;
2475 * Recover a dquot record
2478 xlog_recover_dquot_pass2(
2480 xlog_recover_item_t
*item
)
2482 xfs_mount_t
*mp
= log
->l_mp
;
2484 struct xfs_disk_dquot
*ddq
, *recddq
;
2486 xfs_dq_logformat_t
*dq_f
;
2491 * Filesystems are required to send in quota flags at mount time.
2493 if (mp
->m_qflags
== 0)
2496 recddq
= item
->ri_buf
[1].i_addr
;
2497 if (recddq
== NULL
) {
2498 xfs_alert(log
->l_mp
, "NULL dquot in %s.", __func__
);
2499 return XFS_ERROR(EIO
);
2501 if (item
->ri_buf
[1].i_len
< sizeof(xfs_disk_dquot_t
)) {
2502 xfs_alert(log
->l_mp
, "dquot too small (%d) in %s.",
2503 item
->ri_buf
[1].i_len
, __func__
);
2504 return XFS_ERROR(EIO
);
2508 * This type of quotas was turned off, so ignore this record.
2510 type
= recddq
->d_flags
& (XFS_DQ_USER
| XFS_DQ_PROJ
| XFS_DQ_GROUP
);
2512 if (log
->l_quotaoffs_flag
& type
)
2516 * At this point we know that quota was _not_ turned off.
2517 * Since the mount flags are not indicating to us otherwise, this
2518 * must mean that quota is on, and the dquot needs to be replayed.
2519 * Remember that we may not have fully recovered the superblock yet,
2520 * so we can't do the usual trick of looking at the SB quota bits.
2522 * The other possibility, of course, is that the quota subsystem was
2523 * removed since the last mount - ENOSYS.
2525 dq_f
= item
->ri_buf
[0].i_addr
;
2527 error
= xfs_qm_dqcheck(mp
, recddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2528 "xlog_recover_dquot_pass2 (log copy)");
2530 return XFS_ERROR(EIO
);
2531 ASSERT(dq_f
->qlf_len
== 1);
2533 error
= xfs_read_buf(mp
, mp
->m_ddev_targp
,
2535 XFS_FSB_TO_BB(mp
, dq_f
->qlf_len
),
2538 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp
,
2539 bp
, dq_f
->qlf_blkno
);
2543 ddq
= (xfs_disk_dquot_t
*)xfs_buf_offset(bp
, dq_f
->qlf_boffset
);
2546 * At least the magic num portion should be on disk because this
2547 * was among a chunk of dquots created earlier, and we did some
2548 * minimal initialization then.
2550 error
= xfs_qm_dqcheck(mp
, ddq
, dq_f
->qlf_id
, 0, XFS_QMOPT_DOWARN
,
2551 "xlog_recover_dquot_pass2");
2554 return XFS_ERROR(EIO
);
2557 memcpy(ddq
, recddq
, item
->ri_buf
[1].i_len
);
2559 ASSERT(dq_f
->qlf_size
== 2);
2560 ASSERT(bp
->b_target
->bt_mount
== mp
);
2561 bp
->b_iodone
= xlog_recover_iodone
;
2562 xfs_bdwrite(mp
, bp
);
2568 * This routine is called to create an in-core extent free intent
2569 * item from the efi format structure which was logged on disk.
2570 * It allocates an in-core efi, copies the extents from the format
2571 * structure into it, and adds the efi to the AIL with the given
2575 xlog_recover_efi_pass2(
2577 xlog_recover_item_t
*item
,
2581 xfs_mount_t
*mp
= log
->l_mp
;
2582 xfs_efi_log_item_t
*efip
;
2583 xfs_efi_log_format_t
*efi_formatp
;
2585 efi_formatp
= item
->ri_buf
[0].i_addr
;
2587 efip
= xfs_efi_init(mp
, efi_formatp
->efi_nextents
);
2588 if ((error
= xfs_efi_copy_format(&(item
->ri_buf
[0]),
2589 &(efip
->efi_format
)))) {
2590 xfs_efi_item_free(efip
);
2593 atomic_set(&efip
->efi_next_extent
, efi_formatp
->efi_nextents
);
2595 spin_lock(&log
->l_ailp
->xa_lock
);
2597 * xfs_trans_ail_update() drops the AIL lock.
2599 xfs_trans_ail_update(log
->l_ailp
, &efip
->efi_item
, lsn
);
2605 * This routine is called when an efd format structure is found in
2606 * a committed transaction in the log. It's purpose is to cancel
2607 * the corresponding efi if it was still in the log. To do this
2608 * it searches the AIL for the efi with an id equal to that in the
2609 * efd format structure. If we find it, we remove the efi from the
2613 xlog_recover_efd_pass2(
2615 xlog_recover_item_t
*item
)
2617 xfs_efd_log_format_t
*efd_formatp
;
2618 xfs_efi_log_item_t
*efip
= NULL
;
2619 xfs_log_item_t
*lip
;
2621 struct xfs_ail_cursor cur
;
2622 struct xfs_ail
*ailp
= log
->l_ailp
;
2624 efd_formatp
= item
->ri_buf
[0].i_addr
;
2625 ASSERT((item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_32_t
) +
2626 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_32_t
)))) ||
2627 (item
->ri_buf
[0].i_len
== (sizeof(xfs_efd_log_format_64_t
) +
2628 ((efd_formatp
->efd_nextents
- 1) * sizeof(xfs_extent_64_t
)))));
2629 efi_id
= efd_formatp
->efd_efi_id
;
2632 * Search for the efi with the id in the efd format structure
2635 spin_lock(&ailp
->xa_lock
);
2636 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2637 while (lip
!= NULL
) {
2638 if (lip
->li_type
== XFS_LI_EFI
) {
2639 efip
= (xfs_efi_log_item_t
*)lip
;
2640 if (efip
->efi_format
.efi_id
== efi_id
) {
2642 * xfs_trans_ail_delete() drops the
2645 xfs_trans_ail_delete(ailp
, lip
);
2646 xfs_efi_item_free(efip
);
2647 spin_lock(&ailp
->xa_lock
);
2651 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
2653 xfs_trans_ail_cursor_done(ailp
, &cur
);
2654 spin_unlock(&ailp
->xa_lock
);
2660 * Free up any resources allocated by the transaction
2662 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2665 xlog_recover_free_trans(
2666 struct xlog_recover
*trans
)
2668 xlog_recover_item_t
*item
, *n
;
2671 list_for_each_entry_safe(item
, n
, &trans
->r_itemq
, ri_list
) {
2672 /* Free the regions in the item. */
2673 list_del(&item
->ri_list
);
2674 for (i
= 0; i
< item
->ri_cnt
; i
++)
2675 kmem_free(item
->ri_buf
[i
].i_addr
);
2676 /* Free the item itself */
2677 kmem_free(item
->ri_buf
);
2680 /* Free the transaction recover structure */
2685 xlog_recover_commit_pass1(
2687 struct xlog_recover
*trans
,
2688 xlog_recover_item_t
*item
)
2690 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS1
);
2692 switch (ITEM_TYPE(item
)) {
2694 return xlog_recover_buffer_pass1(log
, item
);
2695 case XFS_LI_QUOTAOFF
:
2696 return xlog_recover_quotaoff_pass1(log
, item
);
2701 /* nothing to do in pass 1 */
2704 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
2705 __func__
, ITEM_TYPE(item
));
2707 return XFS_ERROR(EIO
);
2712 xlog_recover_commit_pass2(
2714 struct xlog_recover
*trans
,
2715 xlog_recover_item_t
*item
)
2717 trace_xfs_log_recover_item_recover(log
, trans
, item
, XLOG_RECOVER_PASS2
);
2719 switch (ITEM_TYPE(item
)) {
2721 return xlog_recover_buffer_pass2(log
, item
);
2723 return xlog_recover_inode_pass2(log
, item
);
2725 return xlog_recover_efi_pass2(log
, item
, trans
->r_lsn
);
2727 return xlog_recover_efd_pass2(log
, item
);
2729 return xlog_recover_dquot_pass2(log
, item
);
2730 case XFS_LI_QUOTAOFF
:
2731 /* nothing to do in pass2 */
2734 xfs_warn(log
->l_mp
, "%s: invalid item type (%d)",
2735 __func__
, ITEM_TYPE(item
));
2737 return XFS_ERROR(EIO
);
2742 * Perform the transaction.
2744 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2745 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2748 xlog_recover_commit_trans(
2750 struct xlog_recover
*trans
,
2754 xlog_recover_item_t
*item
;
2756 hlist_del(&trans
->r_list
);
2758 error
= xlog_recover_reorder_trans(log
, trans
, pass
);
2762 list_for_each_entry(item
, &trans
->r_itemq
, ri_list
) {
2763 if (pass
== XLOG_RECOVER_PASS1
)
2764 error
= xlog_recover_commit_pass1(log
, trans
, item
);
2766 error
= xlog_recover_commit_pass2(log
, trans
, item
);
2771 xlog_recover_free_trans(trans
);
2776 xlog_recover_unmount_trans(
2778 xlog_recover_t
*trans
)
2780 /* Do nothing now */
2781 xfs_warn(log
->l_mp
, "%s: Unmount LR", __func__
);
2786 * There are two valid states of the r_state field. 0 indicates that the
2787 * transaction structure is in a normal state. We have either seen the
2788 * start of the transaction or the last operation we added was not a partial
2789 * operation. If the last operation we added to the transaction was a
2790 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2792 * NOTE: skip LRs with 0 data length.
2795 xlog_recover_process_data(
2797 struct hlist_head rhash
[],
2798 xlog_rec_header_t
*rhead
,
2804 xlog_op_header_t
*ohead
;
2805 xlog_recover_t
*trans
;
2811 lp
= dp
+ be32_to_cpu(rhead
->h_len
);
2812 num_logops
= be32_to_cpu(rhead
->h_num_logops
);
2814 /* check the log format matches our own - else we can't recover */
2815 if (xlog_header_check_recover(log
->l_mp
, rhead
))
2816 return (XFS_ERROR(EIO
));
2818 while ((dp
< lp
) && num_logops
) {
2819 ASSERT(dp
+ sizeof(xlog_op_header_t
) <= lp
);
2820 ohead
= (xlog_op_header_t
*)dp
;
2821 dp
+= sizeof(xlog_op_header_t
);
2822 if (ohead
->oh_clientid
!= XFS_TRANSACTION
&&
2823 ohead
->oh_clientid
!= XFS_LOG
) {
2824 xfs_warn(log
->l_mp
, "%s: bad clientid 0x%x",
2825 __func__
, ohead
->oh_clientid
);
2827 return (XFS_ERROR(EIO
));
2829 tid
= be32_to_cpu(ohead
->oh_tid
);
2830 hash
= XLOG_RHASH(tid
);
2831 trans
= xlog_recover_find_tid(&rhash
[hash
], tid
);
2832 if (trans
== NULL
) { /* not found; add new tid */
2833 if (ohead
->oh_flags
& XLOG_START_TRANS
)
2834 xlog_recover_new_tid(&rhash
[hash
], tid
,
2835 be64_to_cpu(rhead
->h_lsn
));
2837 if (dp
+ be32_to_cpu(ohead
->oh_len
) > lp
) {
2838 xfs_warn(log
->l_mp
, "%s: bad length 0x%x",
2839 __func__
, be32_to_cpu(ohead
->oh_len
));
2841 return (XFS_ERROR(EIO
));
2843 flags
= ohead
->oh_flags
& ~XLOG_END_TRANS
;
2844 if (flags
& XLOG_WAS_CONT_TRANS
)
2845 flags
&= ~XLOG_CONTINUE_TRANS
;
2847 case XLOG_COMMIT_TRANS
:
2848 error
= xlog_recover_commit_trans(log
,
2851 case XLOG_UNMOUNT_TRANS
:
2852 error
= xlog_recover_unmount_trans(log
, trans
);
2854 case XLOG_WAS_CONT_TRANS
:
2855 error
= xlog_recover_add_to_cont_trans(log
,
2857 be32_to_cpu(ohead
->oh_len
));
2859 case XLOG_START_TRANS
:
2860 xfs_warn(log
->l_mp
, "%s: bad transaction",
2863 error
= XFS_ERROR(EIO
);
2866 case XLOG_CONTINUE_TRANS
:
2867 error
= xlog_recover_add_to_trans(log
, trans
,
2868 dp
, be32_to_cpu(ohead
->oh_len
));
2871 xfs_warn(log
->l_mp
, "%s: bad flag 0x%x",
2874 error
= XFS_ERROR(EIO
);
2880 dp
+= be32_to_cpu(ohead
->oh_len
);
2887 * Process an extent free intent item that was recovered from
2888 * the log. We need to free the extents that it describes.
2891 xlog_recover_process_efi(
2893 xfs_efi_log_item_t
*efip
)
2895 xfs_efd_log_item_t
*efdp
;
2900 xfs_fsblock_t startblock_fsb
;
2902 ASSERT(!test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
));
2905 * First check the validity of the extents described by the
2906 * EFI. If any are bad, then assume that all are bad and
2907 * just toss the EFI.
2909 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2910 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2911 startblock_fsb
= XFS_BB_TO_FSB(mp
,
2912 XFS_FSB_TO_DADDR(mp
, extp
->ext_start
));
2913 if ((startblock_fsb
== 0) ||
2914 (extp
->ext_len
== 0) ||
2915 (startblock_fsb
>= mp
->m_sb
.sb_dblocks
) ||
2916 (extp
->ext_len
>= mp
->m_sb
.sb_agblocks
)) {
2918 * This will pull the EFI from the AIL and
2919 * free the memory associated with it.
2921 xfs_efi_release(efip
, efip
->efi_format
.efi_nextents
);
2922 return XFS_ERROR(EIO
);
2926 tp
= xfs_trans_alloc(mp
, 0);
2927 error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0, 0, 0);
2930 efdp
= xfs_trans_get_efd(tp
, efip
, efip
->efi_format
.efi_nextents
);
2932 for (i
= 0; i
< efip
->efi_format
.efi_nextents
; i
++) {
2933 extp
= &(efip
->efi_format
.efi_extents
[i
]);
2934 error
= xfs_free_extent(tp
, extp
->ext_start
, extp
->ext_len
);
2937 xfs_trans_log_efd_extent(tp
, efdp
, extp
->ext_start
,
2941 set_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
);
2942 error
= xfs_trans_commit(tp
, 0);
2946 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
2951 * When this is called, all of the EFIs which did not have
2952 * corresponding EFDs should be in the AIL. What we do now
2953 * is free the extents associated with each one.
2955 * Since we process the EFIs in normal transactions, they
2956 * will be removed at some point after the commit. This prevents
2957 * us from just walking down the list processing each one.
2958 * We'll use a flag in the EFI to skip those that we've already
2959 * processed and use the AIL iteration mechanism's generation
2960 * count to try to speed this up at least a bit.
2962 * When we start, we know that the EFIs are the only things in
2963 * the AIL. As we process them, however, other items are added
2964 * to the AIL. Since everything added to the AIL must come after
2965 * everything already in the AIL, we stop processing as soon as
2966 * we see something other than an EFI in the AIL.
2969 xlog_recover_process_efis(
2972 xfs_log_item_t
*lip
;
2973 xfs_efi_log_item_t
*efip
;
2975 struct xfs_ail_cursor cur
;
2976 struct xfs_ail
*ailp
;
2979 spin_lock(&ailp
->xa_lock
);
2980 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, 0);
2981 while (lip
!= NULL
) {
2983 * We're done when we see something other than an EFI.
2984 * There should be no EFIs left in the AIL now.
2986 if (lip
->li_type
!= XFS_LI_EFI
) {
2988 for (; lip
; lip
= xfs_trans_ail_cursor_next(ailp
, &cur
))
2989 ASSERT(lip
->li_type
!= XFS_LI_EFI
);
2995 * Skip EFIs that we've already processed.
2997 efip
= (xfs_efi_log_item_t
*)lip
;
2998 if (test_bit(XFS_EFI_RECOVERED
, &efip
->efi_flags
)) {
2999 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3003 spin_unlock(&ailp
->xa_lock
);
3004 error
= xlog_recover_process_efi(log
->l_mp
, efip
);
3005 spin_lock(&ailp
->xa_lock
);
3008 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
3011 xfs_trans_ail_cursor_done(ailp
, &cur
);
3012 spin_unlock(&ailp
->xa_lock
);
3017 * This routine performs a transaction to null out a bad inode pointer
3018 * in an agi unlinked inode hash bucket.
3021 xlog_recover_clear_agi_bucket(
3023 xfs_agnumber_t agno
,
3032 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CLEAR_AGI_BUCKET
);
3033 error
= xfs_trans_reserve(tp
, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp
),
3038 error
= xfs_read_agi(mp
, tp
, agno
, &agibp
);
3042 agi
= XFS_BUF_TO_AGI(agibp
);
3043 agi
->agi_unlinked
[bucket
] = cpu_to_be32(NULLAGINO
);
3044 offset
= offsetof(xfs_agi_t
, agi_unlinked
) +
3045 (sizeof(xfs_agino_t
) * bucket
);
3046 xfs_trans_log_buf(tp
, agibp
, offset
,
3047 (offset
+ sizeof(xfs_agino_t
) - 1));
3049 error
= xfs_trans_commit(tp
, 0);
3055 xfs_trans_cancel(tp
, XFS_TRANS_ABORT
);
3057 xfs_warn(mp
, "%s: failed to clear agi %d. Continuing.", __func__
, agno
);
3062 xlog_recover_process_one_iunlink(
3063 struct xfs_mount
*mp
,
3064 xfs_agnumber_t agno
,
3068 struct xfs_buf
*ibp
;
3069 struct xfs_dinode
*dip
;
3070 struct xfs_inode
*ip
;
3074 ino
= XFS_AGINO_TO_INO(mp
, agno
, agino
);
3075 error
= xfs_iget(mp
, NULL
, ino
, 0, 0, &ip
);
3080 * Get the on disk inode to find the next inode in the bucket.
3082 error
= xfs_itobp(mp
, NULL
, ip
, &dip
, &ibp
, XBF_LOCK
);
3086 ASSERT(ip
->i_d
.di_nlink
== 0);
3087 ASSERT(ip
->i_d
.di_mode
!= 0);
3089 /* setup for the next pass */
3090 agino
= be32_to_cpu(dip
->di_next_unlinked
);
3094 * Prevent any DMAPI event from being sent when the reference on
3095 * the inode is dropped.
3097 ip
->i_d
.di_dmevmask
= 0;
3106 * We can't read in the inode this bucket points to, or this inode
3107 * is messed up. Just ditch this bucket of inodes. We will lose
3108 * some inodes and space, but at least we won't hang.
3110 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3111 * clear the inode pointer in the bucket.
3113 xlog_recover_clear_agi_bucket(mp
, agno
, bucket
);
3118 * xlog_iunlink_recover
3120 * This is called during recovery to process any inodes which
3121 * we unlinked but not freed when the system crashed. These
3122 * inodes will be on the lists in the AGI blocks. What we do
3123 * here is scan all the AGIs and fully truncate and free any
3124 * inodes found on the lists. Each inode is removed from the
3125 * lists when it has been fully truncated and is freed. The
3126 * freeing of the inode and its removal from the list must be
3130 xlog_recover_process_iunlinks(
3134 xfs_agnumber_t agno
;
3145 * Prevent any DMAPI event from being sent while in this function.
3147 mp_dmevmask
= mp
->m_dmevmask
;
3150 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3152 * Find the agi for this ag.
3154 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3157 * AGI is b0rked. Don't process it.
3159 * We should probably mark the filesystem as corrupt
3160 * after we've recovered all the ag's we can....
3164 agi
= XFS_BUF_TO_AGI(agibp
);
3166 for (bucket
= 0; bucket
< XFS_AGI_UNLINKED_BUCKETS
; bucket
++) {
3167 agino
= be32_to_cpu(agi
->agi_unlinked
[bucket
]);
3168 while (agino
!= NULLAGINO
) {
3170 * Release the agi buffer so that it can
3171 * be acquired in the normal course of the
3172 * transaction to truncate and free the inode.
3174 xfs_buf_relse(agibp
);
3176 agino
= xlog_recover_process_one_iunlink(mp
,
3177 agno
, agino
, bucket
);
3180 * Reacquire the agibuffer and continue around
3181 * the loop. This should never fail as we know
3182 * the buffer was good earlier on.
3184 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3186 agi
= XFS_BUF_TO_AGI(agibp
);
3191 * Release the buffer for the current agi so we can
3192 * go on to the next one.
3194 xfs_buf_relse(agibp
);
3197 mp
->m_dmevmask
= mp_dmevmask
;
3203 xlog_pack_data_checksum(
3205 xlog_in_core_t
*iclog
,
3212 up
= (__be32
*)iclog
->ic_datap
;
3213 /* divide length by 4 to get # words */
3214 for (i
= 0; i
< (size
>> 2); i
++) {
3215 chksum
^= be32_to_cpu(*up
);
3218 iclog
->ic_header
.h_chksum
= cpu_to_be32(chksum
);
3221 #define xlog_pack_data_checksum(log, iclog, size)
3225 * Stamp cycle number in every block
3230 xlog_in_core_t
*iclog
,
3234 int size
= iclog
->ic_offset
+ roundoff
;
3238 xlog_pack_data_checksum(log
, iclog
, size
);
3240 cycle_lsn
= CYCLE_LSN_DISK(iclog
->ic_header
.h_lsn
);
3242 dp
= iclog
->ic_datap
;
3243 for (i
= 0; i
< BTOBB(size
) &&
3244 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3245 iclog
->ic_header
.h_cycle_data
[i
] = *(__be32
*)dp
;
3246 *(__be32
*)dp
= cycle_lsn
;
3250 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3251 xlog_in_core_2_t
*xhdr
= iclog
->ic_data
;
3253 for ( ; i
< BTOBB(size
); i
++) {
3254 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3255 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3256 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
] = *(__be32
*)dp
;
3257 *(__be32
*)dp
= cycle_lsn
;
3261 for (i
= 1; i
< log
->l_iclog_heads
; i
++) {
3262 xhdr
[i
].hic_xheader
.xh_cycle
= cycle_lsn
;
3269 xlog_rec_header_t
*rhead
,
3275 for (i
= 0; i
< BTOBB(be32_to_cpu(rhead
->h_len
)) &&
3276 i
< (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
); i
++) {
3277 *(__be32
*)dp
= *(__be32
*)&rhead
->h_cycle_data
[i
];
3281 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3282 xlog_in_core_2_t
*xhdr
= (xlog_in_core_2_t
*)rhead
;
3283 for ( ; i
< BTOBB(be32_to_cpu(rhead
->h_len
)); i
++) {
3284 j
= i
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3285 k
= i
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3286 *(__be32
*)dp
= xhdr
[j
].hic_xheader
.xh_cycle_data
[k
];
3293 xlog_valid_rec_header(
3295 xlog_rec_header_t
*rhead
,
3300 if (unlikely(rhead
->h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))) {
3301 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3302 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3303 return XFS_ERROR(EFSCORRUPTED
);
3306 (!rhead
->h_version
||
3307 (be32_to_cpu(rhead
->h_version
) & (~XLOG_VERSION_OKBITS
))))) {
3308 xfs_warn(log
->l_mp
, "%s: unrecognised log version (%d).",
3309 __func__
, be32_to_cpu(rhead
->h_version
));
3310 return XFS_ERROR(EIO
);
3313 /* LR body must have data or it wouldn't have been written */
3314 hlen
= be32_to_cpu(rhead
->h_len
);
3315 if (unlikely( hlen
<= 0 || hlen
> INT_MAX
)) {
3316 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3317 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3318 return XFS_ERROR(EFSCORRUPTED
);
3320 if (unlikely( blkno
> log
->l_logBBsize
|| blkno
> INT_MAX
)) {
3321 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3322 XFS_ERRLEVEL_LOW
, log
->l_mp
);
3323 return XFS_ERROR(EFSCORRUPTED
);
3329 * Read the log from tail to head and process the log records found.
3330 * Handle the two cases where the tail and head are in the same cycle
3331 * and where the active portion of the log wraps around the end of
3332 * the physical log separately. The pass parameter is passed through
3333 * to the routines called to process the data and is not looked at
3337 xlog_do_recovery_pass(
3339 xfs_daddr_t head_blk
,
3340 xfs_daddr_t tail_blk
,
3343 xlog_rec_header_t
*rhead
;
3346 xfs_buf_t
*hbp
, *dbp
;
3347 int error
= 0, h_size
;
3348 int bblks
, split_bblks
;
3349 int hblks
, split_hblks
, wrapped_hblks
;
3350 struct hlist_head rhash
[XLOG_RHASH_SIZE
];
3352 ASSERT(head_blk
!= tail_blk
);
3355 * Read the header of the tail block and get the iclog buffer size from
3356 * h_size. Use this to tell how many sectors make up the log header.
3358 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
)) {
3360 * When using variable length iclogs, read first sector of
3361 * iclog header and extract the header size from it. Get a
3362 * new hbp that is the correct size.
3364 hbp
= xlog_get_bp(log
, 1);
3368 error
= xlog_bread(log
, tail_blk
, 1, hbp
, &offset
);
3372 rhead
= (xlog_rec_header_t
*)offset
;
3373 error
= xlog_valid_rec_header(log
, rhead
, tail_blk
);
3376 h_size
= be32_to_cpu(rhead
->h_size
);
3377 if ((be32_to_cpu(rhead
->h_version
) & XLOG_VERSION_2
) &&
3378 (h_size
> XLOG_HEADER_CYCLE_SIZE
)) {
3379 hblks
= h_size
/ XLOG_HEADER_CYCLE_SIZE
;
3380 if (h_size
% XLOG_HEADER_CYCLE_SIZE
)
3383 hbp
= xlog_get_bp(log
, hblks
);
3388 ASSERT(log
->l_sectBBsize
== 1);
3390 hbp
= xlog_get_bp(log
, 1);
3391 h_size
= XLOG_BIG_RECORD_BSIZE
;
3396 dbp
= xlog_get_bp(log
, BTOBB(h_size
));
3402 memset(rhash
, 0, sizeof(rhash
));
3403 if (tail_blk
<= head_blk
) {
3404 for (blk_no
= tail_blk
; blk_no
< head_blk
; ) {
3405 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3409 rhead
= (xlog_rec_header_t
*)offset
;
3410 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3414 /* blocks in data section */
3415 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3416 error
= xlog_bread(log
, blk_no
+ hblks
, bblks
, dbp
,
3421 xlog_unpack_data(rhead
, offset
, log
);
3422 if ((error
= xlog_recover_process_data(log
,
3423 rhash
, rhead
, offset
, pass
)))
3425 blk_no
+= bblks
+ hblks
;
3429 * Perform recovery around the end of the physical log.
3430 * When the head is not on the same cycle number as the tail,
3431 * we can't do a sequential recovery as above.
3434 while (blk_no
< log
->l_logBBsize
) {
3436 * Check for header wrapping around physical end-of-log
3438 offset
= XFS_BUF_PTR(hbp
);
3441 if (blk_no
+ hblks
<= log
->l_logBBsize
) {
3442 /* Read header in one read */
3443 error
= xlog_bread(log
, blk_no
, hblks
, hbp
,
3448 /* This LR is split across physical log end */
3449 if (blk_no
!= log
->l_logBBsize
) {
3450 /* some data before physical log end */
3451 ASSERT(blk_no
<= INT_MAX
);
3452 split_hblks
= log
->l_logBBsize
- (int)blk_no
;
3453 ASSERT(split_hblks
> 0);
3454 error
= xlog_bread(log
, blk_no
,
3462 * Note: this black magic still works with
3463 * large sector sizes (non-512) only because:
3464 * - we increased the buffer size originally
3465 * by 1 sector giving us enough extra space
3466 * for the second read;
3467 * - the log start is guaranteed to be sector
3469 * - we read the log end (LR header start)
3470 * _first_, then the log start (LR header end)
3471 * - order is important.
3473 wrapped_hblks
= hblks
- split_hblks
;
3474 error
= xlog_bread_offset(log
, 0,
3476 offset
+ BBTOB(split_hblks
));
3480 rhead
= (xlog_rec_header_t
*)offset
;
3481 error
= xlog_valid_rec_header(log
, rhead
,
3482 split_hblks
? blk_no
: 0);
3486 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3489 /* Read in data for log record */
3490 if (blk_no
+ bblks
<= log
->l_logBBsize
) {
3491 error
= xlog_bread(log
, blk_no
, bblks
, dbp
,
3496 /* This log record is split across the
3497 * physical end of log */
3498 offset
= XFS_BUF_PTR(dbp
);
3500 if (blk_no
!= log
->l_logBBsize
) {
3501 /* some data is before the physical
3503 ASSERT(!wrapped_hblks
);
3504 ASSERT(blk_no
<= INT_MAX
);
3506 log
->l_logBBsize
- (int)blk_no
;
3507 ASSERT(split_bblks
> 0);
3508 error
= xlog_bread(log
, blk_no
,
3516 * Note: this black magic still works with
3517 * large sector sizes (non-512) only because:
3518 * - we increased the buffer size originally
3519 * by 1 sector giving us enough extra space
3520 * for the second read;
3521 * - the log start is guaranteed to be sector
3523 * - we read the log end (LR header start)
3524 * _first_, then the log start (LR header end)
3525 * - order is important.
3527 error
= xlog_bread_offset(log
, 0,
3528 bblks
- split_bblks
, hbp
,
3529 offset
+ BBTOB(split_bblks
));
3533 xlog_unpack_data(rhead
, offset
, log
);
3534 if ((error
= xlog_recover_process_data(log
, rhash
,
3535 rhead
, offset
, pass
)))
3540 ASSERT(blk_no
>= log
->l_logBBsize
);
3541 blk_no
-= log
->l_logBBsize
;
3543 /* read first part of physical log */
3544 while (blk_no
< head_blk
) {
3545 error
= xlog_bread(log
, blk_no
, hblks
, hbp
, &offset
);
3549 rhead
= (xlog_rec_header_t
*)offset
;
3550 error
= xlog_valid_rec_header(log
, rhead
, blk_no
);
3554 bblks
= (int)BTOBB(be32_to_cpu(rhead
->h_len
));
3555 error
= xlog_bread(log
, blk_no
+hblks
, bblks
, dbp
,
3560 xlog_unpack_data(rhead
, offset
, log
);
3561 if ((error
= xlog_recover_process_data(log
, rhash
,
3562 rhead
, offset
, pass
)))
3564 blk_no
+= bblks
+ hblks
;
3576 * Do the recovery of the log. We actually do this in two phases.
3577 * The two passes are necessary in order to implement the function
3578 * of cancelling a record written into the log. The first pass
3579 * determines those things which have been cancelled, and the
3580 * second pass replays log items normally except for those which
3581 * have been cancelled. The handling of the replay and cancellations
3582 * takes place in the log item type specific routines.
3584 * The table of items which have cancel records in the log is allocated
3585 * and freed at this level, since only here do we know when all of
3586 * the log recovery has been completed.
3589 xlog_do_log_recovery(
3591 xfs_daddr_t head_blk
,
3592 xfs_daddr_t tail_blk
)
3596 ASSERT(head_blk
!= tail_blk
);
3599 * First do a pass to find all of the cancelled buf log items.
3600 * Store them in the buf_cancel_table for use in the second pass.
3602 log
->l_buf_cancel_table
= kmem_zalloc(XLOG_BC_TABLE_SIZE
*
3603 sizeof(struct list_head
),
3605 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3606 INIT_LIST_HEAD(&log
->l_buf_cancel_table
[i
]);
3608 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3609 XLOG_RECOVER_PASS1
);
3611 kmem_free(log
->l_buf_cancel_table
);
3612 log
->l_buf_cancel_table
= NULL
;
3616 * Then do a second pass to actually recover the items in the log.
3617 * When it is complete free the table of buf cancel items.
3619 error
= xlog_do_recovery_pass(log
, head_blk
, tail_blk
,
3620 XLOG_RECOVER_PASS2
);
3625 for (i
= 0; i
< XLOG_BC_TABLE_SIZE
; i
++)
3626 ASSERT(list_empty(&log
->l_buf_cancel_table
[i
]));
3630 kmem_free(log
->l_buf_cancel_table
);
3631 log
->l_buf_cancel_table
= NULL
;
3637 * Do the actual recovery
3642 xfs_daddr_t head_blk
,
3643 xfs_daddr_t tail_blk
)
3650 * First replay the images in the log.
3652 error
= xlog_do_log_recovery(log
, head_blk
, tail_blk
);
3657 XFS_bflush(log
->l_mp
->m_ddev_targp
);
3660 * If IO errors happened during recovery, bail out.
3662 if (XFS_FORCED_SHUTDOWN(log
->l_mp
)) {
3667 * We now update the tail_lsn since much of the recovery has completed
3668 * and there may be space available to use. If there were no extent
3669 * or iunlinks, we can free up the entire log and set the tail_lsn to
3670 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3671 * lsn of the last known good LR on disk. If there are extent frees
3672 * or iunlinks they will have some entries in the AIL; so we look at
3673 * the AIL to determine how to set the tail_lsn.
3675 xlog_assign_tail_lsn(log
->l_mp
);
3678 * Now that we've finished replaying all buffer and inode
3679 * updates, re-read in the superblock.
3681 bp
= xfs_getsb(log
->l_mp
, 0);
3683 ASSERT(!(XFS_BUF_ISWRITE(bp
)));
3684 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
3686 XFS_BUF_UNASYNC(bp
);
3687 xfsbdstrat(log
->l_mp
, bp
);
3688 error
= xfs_buf_iowait(bp
);
3690 xfs_ioerror_alert("xlog_do_recover",
3691 log
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
3697 /* Convert superblock from on-disk format */
3698 sbp
= &log
->l_mp
->m_sb
;
3699 xfs_sb_from_disk(sbp
, XFS_BUF_TO_SBP(bp
));
3700 ASSERT(sbp
->sb_magicnum
== XFS_SB_MAGIC
);
3701 ASSERT(xfs_sb_good_version(sbp
));
3704 /* We've re-read the superblock so re-initialize per-cpu counters */
3705 xfs_icsb_reinit_counters(log
->l_mp
);
3707 xlog_recover_check_summary(log
);
3709 /* Normal transactions can now occur */
3710 log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
3715 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3717 * Return error or zero.
3723 xfs_daddr_t head_blk
, tail_blk
;
3726 /* find the tail of the log */
3727 if ((error
= xlog_find_tail(log
, &head_blk
, &tail_blk
)))
3730 if (tail_blk
!= head_blk
) {
3731 /* There used to be a comment here:
3733 * disallow recovery on read-only mounts. note -- mount
3734 * checks for ENOSPC and turns it into an intelligent
3736 * ...but this is no longer true. Now, unless you specify
3737 * NORECOVERY (in which case this function would never be
3738 * called), we just go ahead and recover. We do this all
3739 * under the vfs layer, so we can get away with it unless
3740 * the device itself is read-only, in which case we fail.
3742 if ((error
= xfs_dev_is_read_only(log
->l_mp
, "recovery"))) {
3746 xfs_notice(log
->l_mp
, "Starting recovery (logdev: %s)",
3747 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
3750 error
= xlog_do_recover(log
, head_blk
, tail_blk
);
3751 log
->l_flags
|= XLOG_RECOVERY_NEEDED
;
3757 * In the first part of recovery we replay inodes and buffers and build
3758 * up the list of extent free items which need to be processed. Here
3759 * we process the extent free items and clean up the on disk unlinked
3760 * inode lists. This is separated from the first part of recovery so
3761 * that the root and real-time bitmap inodes can be read in from disk in
3762 * between the two stages. This is necessary so that we can free space
3763 * in the real-time portion of the file system.
3766 xlog_recover_finish(
3770 * Now we're ready to do the transactions needed for the
3771 * rest of recovery. Start with completing all the extent
3772 * free intent records and then process the unlinked inode
3773 * lists. At this point, we essentially run in normal mode
3774 * except that we're still performing recovery actions
3775 * rather than accepting new requests.
3777 if (log
->l_flags
& XLOG_RECOVERY_NEEDED
) {
3779 error
= xlog_recover_process_efis(log
);
3781 xfs_alert(log
->l_mp
, "Failed to recover EFIs");
3785 * Sync the log to get all the EFIs out of the AIL.
3786 * This isn't absolutely necessary, but it helps in
3787 * case the unlink transactions would have problems
3788 * pushing the EFIs out of the way.
3790 xfs_log_force(log
->l_mp
, XFS_LOG_SYNC
);
3792 xlog_recover_process_iunlinks(log
);
3794 xlog_recover_check_summary(log
);
3796 xfs_notice(log
->l_mp
, "Ending recovery (logdev: %s)",
3797 log
->l_mp
->m_logname
? log
->l_mp
->m_logname
3799 log
->l_flags
&= ~XLOG_RECOVERY_NEEDED
;
3801 xfs_info(log
->l_mp
, "Ending clean mount");
3809 * Read all of the agf and agi counters and check that they
3810 * are consistent with the superblock counters.
3813 xlog_recover_check_summary(
3820 xfs_agnumber_t agno
;
3821 __uint64_t freeblks
;
3831 for (agno
= 0; agno
< mp
->m_sb
.sb_agcount
; agno
++) {
3832 error
= xfs_read_agf(mp
, NULL
, agno
, 0, &agfbp
);
3834 xfs_alert(mp
, "%s agf read failed agno %d error %d",
3835 __func__
, agno
, error
);
3837 agfp
= XFS_BUF_TO_AGF(agfbp
);
3838 freeblks
+= be32_to_cpu(agfp
->agf_freeblks
) +
3839 be32_to_cpu(agfp
->agf_flcount
);
3840 xfs_buf_relse(agfbp
);
3843 error
= xfs_read_agi(mp
, NULL
, agno
, &agibp
);
3845 xfs_alert(mp
, "%s agi read failed agno %d error %d",
3846 __func__
, agno
, error
);
3848 struct xfs_agi
*agi
= XFS_BUF_TO_AGI(agibp
);
3850 itotal
+= be32_to_cpu(agi
->agi_count
);
3851 ifree
+= be32_to_cpu(agi
->agi_freecount
);
3852 xfs_buf_relse(agibp
);