2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_log_recover.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_trans_priv.h"
37 #include "xfs_alloc.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_quota.h"
40 #include "xfs_cksum.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
43 #include "xfs_bmap_btree.h"
44 #include "xfs_error.h"
46 #include "xfs_rmap_item.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_refcount_item.h"
49 #include "xfs_bmap_item.h"
51 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
58 xlog_clear_stale_blocks(
63 xlog_recover_check_summary(
66 #define xlog_recover_check_summary(log)
69 xlog_do_recovery_pass(
70 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
73 * This structure is used during recovery to record the buf log items which
74 * have been canceled and should not be replayed.
76 struct xfs_buf_cancel {
80 struct list_head bc_list;
84 * Sector aligned buffer routines for buffer create/read/write/access
88 * Verify the given count of basic blocks is valid number of blocks
89 * to specify for an operation involving the given XFS log buffer.
90 * Returns nonzero if the count is valid, 0 otherwise.
94 xlog_buf_bbcount_valid(
98 return bbcount > 0 && bbcount <= log->l_logBBsize;
102 * Allocate a buffer to hold log data. The buffer needs to be able
103 * to map to a range of nbblks basic blocks at any valid (basic
104 * block) offset within the log.
113 if (!xlog_buf_bbcount_valid(log, nbblks)) {
114 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
116 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
121 * We do log I/O in units of log sectors (a power-of-2
122 * multiple of the basic block size), so we round up the
123 * requested size to accommodate the basic blocks required
124 * for complete log sectors.
126 * In addition, the buffer may be used for a non-sector-
127 * aligned block offset, in which case an I/O of the
128 * requested size could extend beyond the end of the
129 * buffer. If the requested size is only 1 basic block it
130 * will never straddle a sector boundary, so this won't be
131 * an issue. Nor will this be a problem if the log I/O is
132 * done in basic blocks (sector size 1). But otherwise we
133 * extend the buffer by one extra log sector to ensure
134 * there's space to accommodate this possibility.
136 if (nbblks > 1 && log->l_sectBBsize > 1)
137 nbblks += log->l_sectBBsize;
138 nbblks = round_up(nbblks, log->l_sectBBsize);
140 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
154 * Return the address of the start of the given block number's data
155 * in a log buffer. The buffer covers a log sector-aligned region.
164 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
166 ASSERT(offset + nbblks <= bp->b_length);
167 return bp->b_addr + BBTOB(offset);
172 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
183 if (!xlog_buf_bbcount_valid(log, nbblks)) {
184 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
186 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
187 return -EFSCORRUPTED;
190 blk_no = round_down(blk_no, log->l_sectBBsize);
191 nbblks = round_up(nbblks, log->l_sectBBsize);
194 ASSERT(nbblks <= bp->b_length);
196 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
197 bp->b_flags |= XBF_READ;
198 bp->b_io_length = nbblks;
201 error = xfs_buf_submit_wait(bp);
202 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
203 xfs_buf_ioerror_alert(bp, __func__);
217 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
221 *offset = xlog_align(log, blk_no, nbblks, bp);
226 * Read at an offset into the buffer. Returns with the buffer in it's original
227 * state regardless of the result of the read.
232 xfs_daddr_t blk_no, /* block to read from */
233 int nbblks, /* blocks to read */
237 char *orig_offset = bp->b_addr;
238 int orig_len = BBTOB(bp->b_length);
241 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
245 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
247 /* must reset buffer pointer even on error */
248 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
255 * Write out the buffer at the given block for the given number of blocks.
256 * The buffer is kept locked across the write and is returned locked.
257 * This can only be used for synchronous log writes.
268 if (!xlog_buf_bbcount_valid(log, nbblks)) {
269 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
271 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
272 return -EFSCORRUPTED;
275 blk_no = round_down(blk_no, log->l_sectBBsize);
276 nbblks = round_up(nbblks, log->l_sectBBsize);
279 ASSERT(nbblks <= bp->b_length);
281 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
284 bp->b_io_length = nbblks;
287 error = xfs_bwrite(bp);
289 xfs_buf_ioerror_alert(bp, __func__);
296 * dump debug superblock and log record information
299 xlog_header_check_dump(
301 xlog_rec_header_t *head)
303 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
304 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
305 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
306 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
309 #define xlog_header_check_dump(mp, head)
313 * check log record header for recovery
316 xlog_header_check_recover(
318 xlog_rec_header_t *head)
320 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
323 * IRIX doesn't write the h_fmt field and leaves it zeroed
324 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
325 * a dirty log created in IRIX.
327 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
329 "dirty log written in incompatible format - can't recover");
330 xlog_header_check_dump(mp, head);
331 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
332 XFS_ERRLEVEL_HIGH, mp);
333 return -EFSCORRUPTED;
334 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
336 "dirty log entry has mismatched uuid - can't recover");
337 xlog_header_check_dump(mp, head);
338 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
339 XFS_ERRLEVEL_HIGH, mp);
340 return -EFSCORRUPTED;
346 * read the head block of the log and check the header
349 xlog_header_check_mount(
351 xlog_rec_header_t *head)
353 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
355 if (uuid_is_nil(&head->h_fs_uuid)) {
357 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
358 * h_fs_uuid is nil, we assume this log was last mounted
359 * by IRIX and continue.
361 xfs_warn(mp, "nil uuid in log - IRIX style log");
362 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
363 xfs_warn(mp, "log has mismatched uuid - can't recover");
364 xlog_header_check_dump(mp, head);
365 XFS_ERROR_REPORT("xlog_header_check_mount",
366 XFS_ERRLEVEL_HIGH, mp);
367 return -EFSCORRUPTED;
378 * We're not going to bother about retrying
379 * this during recovery. One strike!
381 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
382 xfs_buf_ioerror_alert(bp, __func__);
383 xfs_force_shutdown(bp->b_target->bt_mount,
384 SHUTDOWN_META_IO_ERROR);
389 * On v5 supers, a bli could be attached to update the metadata LSN.
393 xfs_buf_item_relse(bp);
394 ASSERT(bp->b_fspriv == NULL);
401 * This routine finds (to an approximation) the first block in the physical
402 * log which contains the given cycle. It uses a binary search algorithm.
403 * Note that the algorithm can not be perfect because the disk will not
404 * necessarily be perfect.
407 xlog_find_cycle_start(
410 xfs_daddr_t first_blk,
411 xfs_daddr_t *last_blk,
421 mid_blk = BLK_AVG(first_blk, end_blk);
422 while (mid_blk != first_blk && mid_blk != end_blk) {
423 error = xlog_bread(log, mid_blk, 1, bp, &offset);
426 mid_cycle = xlog_get_cycle(offset);
427 if (mid_cycle == cycle)
428 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
430 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
431 mid_blk = BLK_AVG(first_blk, end_blk);
433 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
434 (mid_blk == end_blk && mid_blk-1 == first_blk));
442 * Check that a range of blocks does not contain stop_on_cycle_no.
443 * Fill in *new_blk with the block offset where such a block is
444 * found, or with -1 (an invalid block number) if there is no such
445 * block in the range. The scan needs to occur from front to back
446 * and the pointer into the region must be updated since a later
447 * routine will need to perform another test.
450 xlog_find_verify_cycle(
452 xfs_daddr_t start_blk,
454 uint stop_on_cycle_no,
455 xfs_daddr_t *new_blk)
465 * Greedily allocate a buffer big enough to handle the full
466 * range of basic blocks we'll be examining. If that fails,
467 * try a smaller size. We need to be able to read at least
468 * a log sector, or we're out of luck.
470 bufblks = 1 << ffs(nbblks);
471 while (bufblks > log->l_logBBsize)
473 while (!(bp = xlog_get_bp(log, bufblks))) {
475 if (bufblks < log->l_sectBBsize)
479 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
482 bcount = min(bufblks, (start_blk + nbblks - i));
484 error = xlog_bread(log, i, bcount, bp, &buf);
488 for (j = 0; j < bcount; j++) {
489 cycle = xlog_get_cycle(buf);
490 if (cycle == stop_on_cycle_no) {
507 * Potentially backup over partial log record write.
509 * In the typical case, last_blk is the number of the block directly after
510 * a good log record. Therefore, we subtract one to get the block number
511 * of the last block in the given buffer. extra_bblks contains the number
512 * of blocks we would have read on a previous read. This happens when the
513 * last log record is split over the end of the physical log.
515 * extra_bblks is the number of blocks potentially verified on a previous
516 * call to this routine.
519 xlog_find_verify_log_record(
521 xfs_daddr_t start_blk,
522 xfs_daddr_t *last_blk,
528 xlog_rec_header_t *head = NULL;
531 int num_blks = *last_blk - start_blk;
534 ASSERT(start_blk != 0 || *last_blk != start_blk);
536 if (!(bp = xlog_get_bp(log, num_blks))) {
537 if (!(bp = xlog_get_bp(log, 1)))
541 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
544 offset += ((num_blks - 1) << BBSHIFT);
547 for (i = (*last_blk) - 1; i >= 0; i--) {
549 /* valid log record not found */
551 "Log inconsistent (didn't find previous header)");
558 error = xlog_bread(log, i, 1, bp, &offset);
563 head = (xlog_rec_header_t *)offset;
565 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
573 * We hit the beginning of the physical log & still no header. Return
574 * to caller. If caller can handle a return of -1, then this routine
575 * will be called again for the end of the physical log.
583 * We have the final block of the good log (the first block
584 * of the log record _before_ the head. So we check the uuid.
586 if ((error = xlog_header_check_mount(log->l_mp, head)))
590 * We may have found a log record header before we expected one.
591 * last_blk will be the 1st block # with a given cycle #. We may end
592 * up reading an entire log record. In this case, we don't want to
593 * reset last_blk. Only when last_blk points in the middle of a log
594 * record do we update last_blk.
596 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
597 uint h_size = be32_to_cpu(head->h_size);
599 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
600 if (h_size % XLOG_HEADER_CYCLE_SIZE)
606 if (*last_blk - i + extra_bblks !=
607 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
616 * Head is defined to be the point of the log where the next log write
617 * could go. This means that incomplete LR writes at the end are
618 * eliminated when calculating the head. We aren't guaranteed that previous
619 * LR have complete transactions. We only know that a cycle number of
620 * current cycle number -1 won't be present in the log if we start writing
621 * from our current block number.
623 * last_blk contains the block number of the first block with a given
626 * Return: zero if normal, non-zero if error.
631 xfs_daddr_t *return_head_blk)
635 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
637 uint first_half_cycle, last_half_cycle;
639 int error, log_bbnum = log->l_logBBsize;
641 /* Is the end of the log device zeroed? */
642 error = xlog_find_zeroed(log, &first_blk);
644 xfs_warn(log->l_mp, "empty log check failed");
648 *return_head_blk = first_blk;
650 /* Is the whole lot zeroed? */
652 /* Linux XFS shouldn't generate totally zeroed logs -
653 * mkfs etc write a dummy unmount record to a fresh
654 * log so we can store the uuid in there
656 xfs_warn(log->l_mp, "totally zeroed log");
662 first_blk = 0; /* get cycle # of 1st block */
663 bp = xlog_get_bp(log, 1);
667 error = xlog_bread(log, 0, 1, bp, &offset);
671 first_half_cycle = xlog_get_cycle(offset);
673 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
674 error = xlog_bread(log, last_blk, 1, bp, &offset);
678 last_half_cycle = xlog_get_cycle(offset);
679 ASSERT(last_half_cycle != 0);
682 * If the 1st half cycle number is equal to the last half cycle number,
683 * then the entire log is stamped with the same cycle number. In this
684 * case, head_blk can't be set to zero (which makes sense). The below
685 * math doesn't work out properly with head_blk equal to zero. Instead,
686 * we set it to log_bbnum which is an invalid block number, but this
687 * value makes the math correct. If head_blk doesn't changed through
688 * all the tests below, *head_blk is set to zero at the very end rather
689 * than log_bbnum. In a sense, log_bbnum and zero are the same block
690 * in a circular file.
692 if (first_half_cycle == last_half_cycle) {
694 * In this case we believe that the entire log should have
695 * cycle number last_half_cycle. We need to scan backwards
696 * from the end verifying that there are no holes still
697 * containing last_half_cycle - 1. If we find such a hole,
698 * then the start of that hole will be the new head. The
699 * simple case looks like
700 * x | x ... | x - 1 | x
701 * Another case that fits this picture would be
702 * x | x + 1 | x ... | x
703 * In this case the head really is somewhere at the end of the
704 * log, as one of the latest writes at the beginning was
707 * x | x + 1 | x ... | x - 1 | x
708 * This is really the combination of the above two cases, and
709 * the head has to end up at the start of the x-1 hole at the
712 * In the 256k log case, we will read from the beginning to the
713 * end of the log and search for cycle numbers equal to x-1.
714 * We don't worry about the x+1 blocks that we encounter,
715 * because we know that they cannot be the head since the log
718 head_blk = log_bbnum;
719 stop_on_cycle = last_half_cycle - 1;
722 * In this case we want to find the first block with cycle
723 * number matching last_half_cycle. We expect the log to be
725 * x + 1 ... | x ... | x
726 * The first block with cycle number x (last_half_cycle) will
727 * be where the new head belongs. First we do a binary search
728 * for the first occurrence of last_half_cycle. The binary
729 * search may not be totally accurate, so then we scan back
730 * from there looking for occurrences of last_half_cycle before
731 * us. If that backwards scan wraps around the beginning of
732 * the log, then we look for occurrences of last_half_cycle - 1
733 * at the end of the log. The cases we're looking for look
735 * v binary search stopped here
736 * x + 1 ... | x | x + 1 | x ... | x
737 * ^ but we want to locate this spot
739 * <---------> less than scan distance
740 * x + 1 ... | x ... | x - 1 | x
741 * ^ we want to locate this spot
743 stop_on_cycle = last_half_cycle;
744 if ((error = xlog_find_cycle_start(log, bp, first_blk,
745 &head_blk, last_half_cycle)))
750 * Now validate the answer. Scan back some number of maximum possible
751 * blocks and make sure each one has the expected cycle number. The
752 * maximum is determined by the total possible amount of buffering
753 * in the in-core log. The following number can be made tighter if
754 * we actually look at the block size of the filesystem.
756 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
757 if (head_blk >= num_scan_bblks) {
759 * We are guaranteed that the entire check can be performed
762 start_blk = head_blk - num_scan_bblks;
763 if ((error = xlog_find_verify_cycle(log,
764 start_blk, num_scan_bblks,
765 stop_on_cycle, &new_blk)))
769 } else { /* need to read 2 parts of log */
771 * We are going to scan backwards in the log in two parts.
772 * First we scan the physical end of the log. In this part
773 * of the log, we are looking for blocks with cycle number
774 * last_half_cycle - 1.
775 * If we find one, then we know that the log starts there, as
776 * we've found a hole that didn't get written in going around
777 * the end of the physical log. The simple case for this is
778 * x + 1 ... | x ... | x - 1 | x
779 * <---------> less than scan distance
780 * If all of the blocks at the end of the log have cycle number
781 * last_half_cycle, then we check the blocks at the start of
782 * the log looking for occurrences of last_half_cycle. If we
783 * find one, then our current estimate for the location of the
784 * first occurrence of last_half_cycle is wrong and we move
785 * back to the hole we've found. This case looks like
786 * x + 1 ... | x | x + 1 | x ...
787 * ^ binary search stopped here
788 * Another case we need to handle that only occurs in 256k
790 * x + 1 ... | x ... | x+1 | x ...
791 * ^ binary search stops here
792 * In a 256k log, the scan at the end of the log will see the
793 * x + 1 blocks. We need to skip past those since that is
794 * certainly not the head of the log. By searching for
795 * last_half_cycle-1 we accomplish that.
797 ASSERT(head_blk <= INT_MAX &&
798 (xfs_daddr_t) num_scan_bblks >= head_blk);
799 start_blk = log_bbnum - (num_scan_bblks - head_blk);
800 if ((error = xlog_find_verify_cycle(log, start_blk,
801 num_scan_bblks - (int)head_blk,
802 (stop_on_cycle - 1), &new_blk)))
810 * Scan beginning of log now. The last part of the physical
811 * log is good. This scan needs to verify that it doesn't find
812 * the last_half_cycle.
815 ASSERT(head_blk <= INT_MAX);
816 if ((error = xlog_find_verify_cycle(log,
817 start_blk, (int)head_blk,
818 stop_on_cycle, &new_blk)))
826 * Now we need to make sure head_blk is not pointing to a block in
827 * the middle of a log record.
829 num_scan_bblks = XLOG_REC_SHIFT(log);
830 if (head_blk >= num_scan_bblks) {
831 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
833 /* start ptr at last block ptr before head_blk */
834 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
841 ASSERT(head_blk <= INT_MAX);
842 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
846 /* We hit the beginning of the log during our search */
847 start_blk = log_bbnum - (num_scan_bblks - head_blk);
849 ASSERT(start_blk <= INT_MAX &&
850 (xfs_daddr_t) log_bbnum-start_blk >= 0);
851 ASSERT(head_blk <= INT_MAX);
852 error = xlog_find_verify_log_record(log, start_blk,
853 &new_blk, (int)head_blk);
858 if (new_blk != log_bbnum)
865 if (head_blk == log_bbnum)
866 *return_head_blk = 0;
868 *return_head_blk = head_blk;
870 * When returning here, we have a good block number. Bad block
871 * means that during a previous crash, we didn't have a clean break
872 * from cycle number N to cycle number N-1. In this case, we need
873 * to find the first block with cycle number N-1.
881 xfs_warn(log->l_mp, "failed to find log head");
886 * Seek backwards in the log for log record headers.
888 * Given a starting log block, walk backwards until we find the provided number
889 * of records or hit the provided tail block. The return value is the number of
890 * records encountered or a negative error code. The log block and buffer
891 * pointer of the last record seen are returned in rblk and rhead respectively.
894 xlog_rseek_logrec_hdr(
896 xfs_daddr_t head_blk,
897 xfs_daddr_t tail_blk,
901 struct xlog_rec_header **rhead,
913 * Walk backwards from the head block until we hit the tail or the first
916 end_blk = head_blk > tail_blk ? tail_blk : 0;
917 for (i = (int) head_blk - 1; i >= end_blk; i--) {
918 error = xlog_bread(log, i, 1, bp, &offset);
922 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
924 *rhead = (struct xlog_rec_header *) offset;
925 if (++found == count)
931 * If we haven't hit the tail block or the log record header count,
932 * start looking again from the end of the physical log. Note that
933 * callers can pass head == tail if the tail is not yet known.
935 if (tail_blk >= head_blk && found != count) {
936 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
937 error = xlog_bread(log, i, 1, bp, &offset);
941 if (*(__be32 *)offset ==
942 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
945 *rhead = (struct xlog_rec_header *) offset;
946 if (++found == count)
959 * Seek forward in the log for log record headers.
961 * Given head and tail blocks, walk forward from the tail block until we find
962 * the provided number of records or hit the head block. The return value is the
963 * number of records encountered or a negative error code. The log block and
964 * buffer pointer of the last record seen are returned in rblk and rhead
968 xlog_seek_logrec_hdr(
970 xfs_daddr_t head_blk,
971 xfs_daddr_t tail_blk,
975 struct xlog_rec_header **rhead,
987 * Walk forward from the tail block until we hit the head or the last
990 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
991 for (i = (int) tail_blk; i <= end_blk; i++) {
992 error = xlog_bread(log, i, 1, bp, &offset);
996 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
998 *rhead = (struct xlog_rec_header *) offset;
999 if (++found == count)
1005 * If we haven't hit the head block or the log record header count,
1006 * start looking again from the start of the physical log.
1008 if (tail_blk > head_blk && found != count) {
1009 for (i = 0; i < (int) head_blk; i++) {
1010 error = xlog_bread(log, i, 1, bp, &offset);
1014 if (*(__be32 *)offset ==
1015 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
1018 *rhead = (struct xlog_rec_header *) offset;
1019 if (++found == count)
1032 * Calculate distance from head to tail (i.e., unused space in the log).
1037 xfs_daddr_t head_blk,
1038 xfs_daddr_t tail_blk)
1040 if (head_blk < tail_blk)
1041 return tail_blk - head_blk;
1043 return tail_blk + (log->l_logBBsize - head_blk);
1047 * Verify the log tail. This is particularly important when torn or incomplete
1048 * writes have been detected near the front of the log and the head has been
1049 * walked back accordingly.
1051 * We also have to handle the case where the tail was pinned and the head
1052 * blocked behind the tail right before a crash. If the tail had been pushed
1053 * immediately prior to the crash and the subsequent checkpoint was only
1054 * partially written, it's possible it overwrote the last referenced tail in the
1055 * log with garbage. This is not a coherency problem because the tail must have
1056 * been pushed before it can be overwritten, but appears as log corruption to
1057 * recovery because we have no way to know the tail was updated if the
1058 * subsequent checkpoint didn't write successfully.
1060 * Therefore, CRC check the log from tail to head. If a failure occurs and the
1061 * offending record is within max iclog bufs from the head, walk the tail
1062 * forward and retry until a valid tail is found or corruption is detected out
1063 * of the range of a possible overwrite.
1068 xfs_daddr_t head_blk,
1069 xfs_daddr_t *tail_blk,
1072 struct xlog_rec_header *thead;
1074 xfs_daddr_t first_bad;
1077 xfs_daddr_t tmp_tail;
1078 xfs_daddr_t orig_tail = *tail_blk;
1080 bp = xlog_get_bp(log, 1);
1085 * Make sure the tail points to a record (returns positive count on
1088 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, bp,
1089 &tmp_tail, &thead, &wrapped);
1092 if (*tail_blk != tmp_tail)
1093 *tail_blk = tmp_tail;
1096 * Run a CRC check from the tail to the head. We can't just check
1097 * MAX_ICLOGS records past the tail because the tail may point to stale
1098 * blocks cleared during the search for the head/tail. These blocks are
1099 * overwritten with zero-length records and thus record count is not a
1100 * reliable indicator of the iclog state before a crash.
1103 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1104 XLOG_RECOVER_CRCPASS, &first_bad);
1105 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1109 * Is corruption within range of the head? If so, retry from
1110 * the next record. Otherwise return an error.
1112 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1113 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1116 /* skip to the next record; returns positive count on success */
1117 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2, bp,
1118 &tmp_tail, &thead, &wrapped);
1122 *tail_blk = tmp_tail;
1124 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1125 XLOG_RECOVER_CRCPASS, &first_bad);
1128 if (!error && *tail_blk != orig_tail)
1130 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1131 orig_tail, *tail_blk);
1138 * Detect and trim torn writes from the head of the log.
1140 * Storage without sector atomicity guarantees can result in torn writes in the
1141 * log in the event of a crash. Our only means to detect this scenario is via
1142 * CRC verification. While we can't always be certain that CRC verification
1143 * failure is due to a torn write vs. an unrelated corruption, we do know that
1144 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1145 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1146 * the log and treat failures in this range as torn writes as a matter of
1147 * policy. In the event of CRC failure, the head is walked back to the last good
1148 * record in the log and the tail is updated from that record and verified.
1153 xfs_daddr_t *head_blk, /* in/out: unverified head */
1154 xfs_daddr_t *tail_blk, /* out: tail block */
1156 xfs_daddr_t *rhead_blk, /* start blk of last record */
1157 struct xlog_rec_header **rhead, /* ptr to last record */
1158 bool *wrapped) /* last rec. wraps phys. log */
1160 struct xlog_rec_header *tmp_rhead;
1161 struct xfs_buf *tmp_bp;
1162 xfs_daddr_t first_bad;
1163 xfs_daddr_t tmp_rhead_blk;
1169 * Check the head of the log for torn writes. Search backwards from the
1170 * head until we hit the tail or the maximum number of log record I/Os
1171 * that could have been in flight at one time. Use a temporary buffer so
1172 * we don't trash the rhead/bp pointers from the caller.
1174 tmp_bp = xlog_get_bp(log, 1);
1177 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1178 XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk,
1179 &tmp_rhead, &tmp_wrapped);
1180 xlog_put_bp(tmp_bp);
1185 * Now run a CRC verification pass over the records starting at the
1186 * block found above to the current head. If a CRC failure occurs, the
1187 * log block of the first bad record is saved in first_bad.
1189 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1190 XLOG_RECOVER_CRCPASS, &first_bad);
1191 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1193 * We've hit a potential torn write. Reset the error and warn
1198 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1199 first_bad, *head_blk);
1202 * Get the header block and buffer pointer for the last good
1203 * record before the bad record.
1205 * Note that xlog_find_tail() clears the blocks at the new head
1206 * (i.e., the records with invalid CRC) if the cycle number
1207 * matches the the current cycle.
1209 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp,
1210 rhead_blk, rhead, wrapped);
1213 if (found == 0) /* XXX: right thing to do here? */
1217 * Reset the head block to the starting block of the first bad
1218 * log record and set the tail block based on the last good
1221 * Bail out if the updated head/tail match as this indicates
1222 * possible corruption outside of the acceptable
1223 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1225 *head_blk = first_bad;
1226 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1227 if (*head_blk == *tail_blk) {
1235 return xlog_verify_tail(log, *head_blk, tail_blk,
1236 be32_to_cpu((*rhead)->h_size));
1240 * Check whether the head of the log points to an unmount record. In other
1241 * words, determine whether the log is clean. If so, update the in-core state
1245 xlog_check_unmount_rec(
1247 xfs_daddr_t *head_blk,
1248 xfs_daddr_t *tail_blk,
1249 struct xlog_rec_header *rhead,
1250 xfs_daddr_t rhead_blk,
1254 struct xlog_op_header *op_head;
1255 xfs_daddr_t umount_data_blk;
1256 xfs_daddr_t after_umount_blk;
1264 * Look for unmount record. If we find it, then we know there was a
1265 * clean unmount. Since 'i' could be the last block in the physical
1266 * log, we convert to a log block before comparing to the head_blk.
1268 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1269 * below. We won't want to clear the unmount record if there is one, so
1270 * we pass the lsn of the unmount record rather than the block after it.
1272 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1273 int h_size = be32_to_cpu(rhead->h_size);
1274 int h_version = be32_to_cpu(rhead->h_version);
1276 if ((h_version & XLOG_VERSION_2) &&
1277 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1278 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1279 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1287 after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len));
1288 after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize);
1289 if (*head_blk == after_umount_blk &&
1290 be32_to_cpu(rhead->h_num_logops) == 1) {
1291 umount_data_blk = rhead_blk + hblks;
1292 umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize);
1293 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1297 op_head = (struct xlog_op_header *)offset;
1298 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1300 * Set tail and last sync so that newly written log
1301 * records will point recovery to after the current
1304 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1305 log->l_curr_cycle, after_umount_blk);
1306 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1307 log->l_curr_cycle, after_umount_blk);
1308 *tail_blk = after_umount_blk;
1320 xfs_daddr_t head_blk,
1321 struct xlog_rec_header *rhead,
1322 xfs_daddr_t rhead_blk,
1326 * Reset log values according to the state of the log when we
1327 * crashed. In the case where head_blk == 0, we bump curr_cycle
1328 * one because the next write starts a new cycle rather than
1329 * continuing the cycle of the last good log record. At this
1330 * point we have guaranteed that all partial log records have been
1331 * accounted for. Therefore, we know that the last good log record
1332 * written was complete and ended exactly on the end boundary
1333 * of the physical log.
1335 log->l_prev_block = rhead_blk;
1336 log->l_curr_block = (int)head_blk;
1337 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1339 log->l_curr_cycle++;
1340 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1341 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1342 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1343 BBTOB(log->l_curr_block));
1344 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1345 BBTOB(log->l_curr_block));
1349 * Find the sync block number or the tail of the log.
1351 * This will be the block number of the last record to have its
1352 * associated buffers synced to disk. Every log record header has
1353 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1354 * to get a sync block number. The only concern is to figure out which
1355 * log record header to believe.
1357 * The following algorithm uses the log record header with the largest
1358 * lsn. The entire log record does not need to be valid. We only care
1359 * that the header is valid.
1361 * We could speed up search by using current head_blk buffer, but it is not
1367 xfs_daddr_t *head_blk,
1368 xfs_daddr_t *tail_blk)
1370 xlog_rec_header_t *rhead;
1371 char *offset = NULL;
1374 xfs_daddr_t rhead_blk;
1376 bool wrapped = false;
1380 * Find previous log record
1382 if ((error = xlog_find_head(log, head_blk)))
1384 ASSERT(*head_blk < INT_MAX);
1386 bp = xlog_get_bp(log, 1);
1389 if (*head_blk == 0) { /* special case */
1390 error = xlog_bread(log, 0, 1, bp, &offset);
1394 if (xlog_get_cycle(offset) == 0) {
1396 /* leave all other log inited values alone */
1402 * Search backwards through the log looking for the log record header
1403 * block. This wraps all the way back around to the head so something is
1404 * seriously wrong if we can't find it.
1406 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp,
1407 &rhead_blk, &rhead, &wrapped);
1411 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1414 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1417 * Set the log state based on the current head record.
1419 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1420 tail_lsn = atomic64_read(&log->l_tail_lsn);
1423 * Look for an unmount record at the head of the log. This sets the log
1424 * state to determine whether recovery is necessary.
1426 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1427 rhead_blk, bp, &clean);
1432 * Verify the log head if the log is not clean (e.g., we have anything
1433 * but an unmount record at the head). This uses CRC verification to
1434 * detect and trim torn writes. If discovered, CRC failures are
1435 * considered torn writes and the log head is trimmed accordingly.
1437 * Note that we can only run CRC verification when the log is dirty
1438 * because there's no guarantee that the log data behind an unmount
1439 * record is compatible with the current architecture.
1442 xfs_daddr_t orig_head = *head_blk;
1444 error = xlog_verify_head(log, head_blk, tail_blk, bp,
1445 &rhead_blk, &rhead, &wrapped);
1449 /* update in-core state again if the head changed */
1450 if (*head_blk != orig_head) {
1451 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1453 tail_lsn = atomic64_read(&log->l_tail_lsn);
1454 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1455 rhead, rhead_blk, bp,
1463 * Note that the unmount was clean. If the unmount was not clean, we
1464 * need to know this to rebuild the superblock counters from the perag
1465 * headers if we have a filesystem using non-persistent counters.
1468 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1471 * Make sure that there are no blocks in front of the head
1472 * with the same cycle number as the head. This can happen
1473 * because we allow multiple outstanding log writes concurrently,
1474 * and the later writes might make it out before earlier ones.
1476 * We use the lsn from before modifying it so that we'll never
1477 * overwrite the unmount record after a clean unmount.
1479 * Do this only if we are going to recover the filesystem
1481 * NOTE: This used to say "if (!readonly)"
1482 * However on Linux, we can & do recover a read-only filesystem.
1483 * We only skip recovery if NORECOVERY is specified on mount,
1484 * in which case we would not be here.
1486 * But... if the -device- itself is readonly, just skip this.
1487 * We can't recover this device anyway, so it won't matter.
1489 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1490 error = xlog_clear_stale_blocks(log, tail_lsn);
1496 xfs_warn(log->l_mp, "failed to locate log tail");
1501 * Is the log zeroed at all?
1503 * The last binary search should be changed to perform an X block read
1504 * once X becomes small enough. You can then search linearly through
1505 * the X blocks. This will cut down on the number of reads we need to do.
1507 * If the log is partially zeroed, this routine will pass back the blkno
1508 * of the first block with cycle number 0. It won't have a complete LR
1512 * 0 => the log is completely written to
1513 * 1 => use *blk_no as the first block of the log
1514 * <0 => error has occurred
1519 xfs_daddr_t *blk_no)
1523 uint first_cycle, last_cycle;
1524 xfs_daddr_t new_blk, last_blk, start_blk;
1525 xfs_daddr_t num_scan_bblks;
1526 int error, log_bbnum = log->l_logBBsize;
1530 /* check totally zeroed log */
1531 bp = xlog_get_bp(log, 1);
1534 error = xlog_bread(log, 0, 1, bp, &offset);
1538 first_cycle = xlog_get_cycle(offset);
1539 if (first_cycle == 0) { /* completely zeroed log */
1545 /* check partially zeroed log */
1546 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1550 last_cycle = xlog_get_cycle(offset);
1551 if (last_cycle != 0) { /* log completely written to */
1554 } else if (first_cycle != 1) {
1556 * If the cycle of the last block is zero, the cycle of
1557 * the first block must be 1. If it's not, maybe we're
1558 * not looking at a log... Bail out.
1561 "Log inconsistent or not a log (last==0, first!=1)");
1566 /* we have a partially zeroed log */
1567 last_blk = log_bbnum-1;
1568 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1572 * Validate the answer. Because there is no way to guarantee that
1573 * the entire log is made up of log records which are the same size,
1574 * we scan over the defined maximum blocks. At this point, the maximum
1575 * is not chosen to mean anything special. XXXmiken
1577 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1578 ASSERT(num_scan_bblks <= INT_MAX);
1580 if (last_blk < num_scan_bblks)
1581 num_scan_bblks = last_blk;
1582 start_blk = last_blk - num_scan_bblks;
1585 * We search for any instances of cycle number 0 that occur before
1586 * our current estimate of the head. What we're trying to detect is
1587 * 1 ... | 0 | 1 | 0...
1588 * ^ binary search ends here
1590 if ((error = xlog_find_verify_cycle(log, start_blk,
1591 (int)num_scan_bblks, 0, &new_blk)))
1597 * Potentially backup over partial log record write. We don't need
1598 * to search the end of the log because we know it is zero.
1600 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1615 * These are simple subroutines used by xlog_clear_stale_blocks() below
1616 * to initialize a buffer full of empty log record headers and write
1617 * them into the log.
1628 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1630 memset(buf, 0, BBSIZE);
1631 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1632 recp->h_cycle = cpu_to_be32(cycle);
1633 recp->h_version = cpu_to_be32(
1634 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1635 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1636 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1637 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1638 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1642 xlog_write_log_records(
1653 int sectbb = log->l_sectBBsize;
1654 int end_block = start_block + blocks;
1660 * Greedily allocate a buffer big enough to handle the full
1661 * range of basic blocks to be written. If that fails, try
1662 * a smaller size. We need to be able to write at least a
1663 * log sector, or we're out of luck.
1665 bufblks = 1 << ffs(blocks);
1666 while (bufblks > log->l_logBBsize)
1668 while (!(bp = xlog_get_bp(log, bufblks))) {
1670 if (bufblks < sectbb)
1674 /* We may need to do a read at the start to fill in part of
1675 * the buffer in the starting sector not covered by the first
1678 balign = round_down(start_block, sectbb);
1679 if (balign != start_block) {
1680 error = xlog_bread_noalign(log, start_block, 1, bp);
1684 j = start_block - balign;
1687 for (i = start_block; i < end_block; i += bufblks) {
1688 int bcount, endcount;
1690 bcount = min(bufblks, end_block - start_block);
1691 endcount = bcount - j;
1693 /* We may need to do a read at the end to fill in part of
1694 * the buffer in the final sector not covered by the write.
1695 * If this is the same sector as the above read, skip it.
1697 ealign = round_down(end_block, sectbb);
1698 if (j == 0 && (start_block + endcount > ealign)) {
1699 offset = bp->b_addr + BBTOB(ealign - start_block);
1700 error = xlog_bread_offset(log, ealign, sectbb,
1707 offset = xlog_align(log, start_block, endcount, bp);
1708 for (; j < endcount; j++) {
1709 xlog_add_record(log, offset, cycle, i+j,
1710 tail_cycle, tail_block);
1713 error = xlog_bwrite(log, start_block, endcount, bp);
1716 start_block += endcount;
1726 * This routine is called to blow away any incomplete log writes out
1727 * in front of the log head. We do this so that we won't become confused
1728 * if we come up, write only a little bit more, and then crash again.
1729 * If we leave the partial log records out there, this situation could
1730 * cause us to think those partial writes are valid blocks since they
1731 * have the current cycle number. We get rid of them by overwriting them
1732 * with empty log records with the old cycle number rather than the
1735 * The tail lsn is passed in rather than taken from
1736 * the log so that we will not write over the unmount record after a
1737 * clean unmount in a 512 block log. Doing so would leave the log without
1738 * any valid log records in it until a new one was written. If we crashed
1739 * during that time we would not be able to recover.
1742 xlog_clear_stale_blocks(
1746 int tail_cycle, head_cycle;
1747 int tail_block, head_block;
1748 int tail_distance, max_distance;
1752 tail_cycle = CYCLE_LSN(tail_lsn);
1753 tail_block = BLOCK_LSN(tail_lsn);
1754 head_cycle = log->l_curr_cycle;
1755 head_block = log->l_curr_block;
1758 * Figure out the distance between the new head of the log
1759 * and the tail. We want to write over any blocks beyond the
1760 * head that we may have written just before the crash, but
1761 * we don't want to overwrite the tail of the log.
1763 if (head_cycle == tail_cycle) {
1765 * The tail is behind the head in the physical log,
1766 * so the distance from the head to the tail is the
1767 * distance from the head to the end of the log plus
1768 * the distance from the beginning of the log to the
1771 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1772 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1773 XFS_ERRLEVEL_LOW, log->l_mp);
1774 return -EFSCORRUPTED;
1776 tail_distance = tail_block + (log->l_logBBsize - head_block);
1779 * The head is behind the tail in the physical log,
1780 * so the distance from the head to the tail is just
1781 * the tail block minus the head block.
1783 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1784 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1785 XFS_ERRLEVEL_LOW, log->l_mp);
1786 return -EFSCORRUPTED;
1788 tail_distance = tail_block - head_block;
1792 * If the head is right up against the tail, we can't clear
1795 if (tail_distance <= 0) {
1796 ASSERT(tail_distance == 0);
1800 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1802 * Take the smaller of the maximum amount of outstanding I/O
1803 * we could have and the distance to the tail to clear out.
1804 * We take the smaller so that we don't overwrite the tail and
1805 * we don't waste all day writing from the head to the tail
1808 max_distance = MIN(max_distance, tail_distance);
1810 if ((head_block + max_distance) <= log->l_logBBsize) {
1812 * We can stomp all the blocks we need to without
1813 * wrapping around the end of the log. Just do it
1814 * in a single write. Use the cycle number of the
1815 * current cycle minus one so that the log will look like:
1818 error = xlog_write_log_records(log, (head_cycle - 1),
1819 head_block, max_distance, tail_cycle,
1825 * We need to wrap around the end of the physical log in
1826 * order to clear all the blocks. Do it in two separate
1827 * I/Os. The first write should be from the head to the
1828 * end of the physical log, and it should use the current
1829 * cycle number minus one just like above.
1831 distance = log->l_logBBsize - head_block;
1832 error = xlog_write_log_records(log, (head_cycle - 1),
1833 head_block, distance, tail_cycle,
1840 * Now write the blocks at the start of the physical log.
1841 * This writes the remainder of the blocks we want to clear.
1842 * It uses the current cycle number since we're now on the
1843 * same cycle as the head so that we get:
1844 * n ... n ... | n - 1 ...
1845 * ^^^^^ blocks we're writing
1847 distance = max_distance - (log->l_logBBsize - head_block);
1848 error = xlog_write_log_records(log, head_cycle, 0, distance,
1849 tail_cycle, tail_block);
1857 /******************************************************************************
1859 * Log recover routines
1861 ******************************************************************************
1865 * Sort the log items in the transaction.
1867 * The ordering constraints are defined by the inode allocation and unlink
1868 * behaviour. The rules are:
1870 * 1. Every item is only logged once in a given transaction. Hence it
1871 * represents the last logged state of the item. Hence ordering is
1872 * dependent on the order in which operations need to be performed so
1873 * required initial conditions are always met.
1875 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1876 * there's nothing to replay from them so we can simply cull them
1877 * from the transaction. However, we can't do that until after we've
1878 * replayed all the other items because they may be dependent on the
1879 * cancelled buffer and replaying the cancelled buffer can remove it
1880 * form the cancelled buffer table. Hence they have tobe done last.
1882 * 3. Inode allocation buffers must be replayed before inode items that
1883 * read the buffer and replay changes into it. For filesystems using the
1884 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1885 * treated the same as inode allocation buffers as they create and
1886 * initialise the buffers directly.
1888 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1889 * This ensures that inodes are completely flushed to the inode buffer
1890 * in a "free" state before we remove the unlinked inode list pointer.
1892 * Hence the ordering needs to be inode allocation buffers first, inode items
1893 * second, inode unlink buffers third and cancelled buffers last.
1895 * But there's a problem with that - we can't tell an inode allocation buffer
1896 * apart from a regular buffer, so we can't separate them. We can, however,
1897 * tell an inode unlink buffer from the others, and so we can separate them out
1898 * from all the other buffers and move them to last.
1900 * Hence, 4 lists, in order from head to tail:
1901 * - buffer_list for all buffers except cancelled/inode unlink buffers
1902 * - item_list for all non-buffer items
1903 * - inode_buffer_list for inode unlink buffers
1904 * - cancel_list for the cancelled buffers
1906 * Note that we add objects to the tail of the lists so that first-to-last
1907 * ordering is preserved within the lists. Adding objects to the head of the
1908 * list means when we traverse from the head we walk them in last-to-first
1909 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1910 * but for all other items there may be specific ordering that we need to
1914 xlog_recover_reorder_trans(
1916 struct xlog_recover *trans,
1919 xlog_recover_item_t *item, *n;
1921 LIST_HEAD(sort_list);
1922 LIST_HEAD(cancel_list);
1923 LIST_HEAD(buffer_list);
1924 LIST_HEAD(inode_buffer_list);
1925 LIST_HEAD(inode_list);
1927 list_splice_init(&trans->r_itemq, &sort_list);
1928 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1929 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1931 switch (ITEM_TYPE(item)) {
1932 case XFS_LI_ICREATE:
1933 list_move_tail(&item->ri_list, &buffer_list);
1936 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1937 trace_xfs_log_recover_item_reorder_head(log,
1939 list_move(&item->ri_list, &cancel_list);
1942 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1943 list_move(&item->ri_list, &inode_buffer_list);
1946 list_move_tail(&item->ri_list, &buffer_list);
1950 case XFS_LI_QUOTAOFF:
1959 trace_xfs_log_recover_item_reorder_tail(log,
1961 list_move_tail(&item->ri_list, &inode_list);
1965 "%s: unrecognized type of log operation",
1969 * return the remaining items back to the transaction
1970 * item list so they can be freed in caller.
1972 if (!list_empty(&sort_list))
1973 list_splice_init(&sort_list, &trans->r_itemq);
1979 ASSERT(list_empty(&sort_list));
1980 if (!list_empty(&buffer_list))
1981 list_splice(&buffer_list, &trans->r_itemq);
1982 if (!list_empty(&inode_list))
1983 list_splice_tail(&inode_list, &trans->r_itemq);
1984 if (!list_empty(&inode_buffer_list))
1985 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1986 if (!list_empty(&cancel_list))
1987 list_splice_tail(&cancel_list, &trans->r_itemq);
1992 * Build up the table of buf cancel records so that we don't replay
1993 * cancelled data in the second pass. For buffer records that are
1994 * not cancel records, there is nothing to do here so we just return.
1996 * If we get a cancel record which is already in the table, this indicates
1997 * that the buffer was cancelled multiple times. In order to ensure
1998 * that during pass 2 we keep the record in the table until we reach its
1999 * last occurrence in the log, we keep a reference count in the cancel
2000 * record in the table to tell us how many times we expect to see this
2001 * record during the second pass.
2004 xlog_recover_buffer_pass1(
2006 struct xlog_recover_item *item)
2008 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2009 struct list_head *bucket;
2010 struct xfs_buf_cancel *bcp;
2013 * If this isn't a cancel buffer item, then just return.
2015 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
2016 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
2021 * Insert an xfs_buf_cancel record into the hash table of them.
2022 * If there is already an identical record, bump its reference count.
2024 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
2025 list_for_each_entry(bcp, bucket, bc_list) {
2026 if (bcp->bc_blkno == buf_f->blf_blkno &&
2027 bcp->bc_len == buf_f->blf_len) {
2029 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
2034 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
2035 bcp->bc_blkno = buf_f->blf_blkno;
2036 bcp->bc_len = buf_f->blf_len;
2037 bcp->bc_refcount = 1;
2038 list_add_tail(&bcp->bc_list, bucket);
2040 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
2045 * Check to see whether the buffer being recovered has a corresponding
2046 * entry in the buffer cancel record table. If it is, return the cancel
2047 * buffer structure to the caller.
2049 STATIC struct xfs_buf_cancel *
2050 xlog_peek_buffer_cancelled(
2056 struct list_head *bucket;
2057 struct xfs_buf_cancel *bcp;
2059 if (!log->l_buf_cancel_table) {
2060 /* empty table means no cancelled buffers in the log */
2061 ASSERT(!(flags & XFS_BLF_CANCEL));
2065 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
2066 list_for_each_entry(bcp, bucket, bc_list) {
2067 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
2072 * We didn't find a corresponding entry in the table, so return 0 so
2073 * that the buffer is NOT cancelled.
2075 ASSERT(!(flags & XFS_BLF_CANCEL));
2080 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2081 * otherwise return 0. If the buffer is actually a buffer cancel item
2082 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2083 * table and remove it from the table if this is the last reference.
2085 * We remove the cancel record from the table when we encounter its last
2086 * occurrence in the log so that if the same buffer is re-used again after its
2087 * last cancellation we actually replay the changes made at that point.
2090 xlog_check_buffer_cancelled(
2096 struct xfs_buf_cancel *bcp;
2098 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2103 * We've go a match, so return 1 so that the recovery of this buffer
2104 * is cancelled. If this buffer is actually a buffer cancel log
2105 * item, then decrement the refcount on the one in the table and
2106 * remove it if this is the last reference.
2108 if (flags & XFS_BLF_CANCEL) {
2109 if (--bcp->bc_refcount == 0) {
2110 list_del(&bcp->bc_list);
2118 * Perform recovery for a buffer full of inodes. In these buffers, the only
2119 * data which should be recovered is that which corresponds to the
2120 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2121 * data for the inodes is always logged through the inodes themselves rather
2122 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2124 * The only time when buffers full of inodes are fully recovered is when the
2125 * buffer is full of newly allocated inodes. In this case the buffer will
2126 * not be marked as an inode buffer and so will be sent to
2127 * xlog_recover_do_reg_buffer() below during recovery.
2130 xlog_recover_do_inode_buffer(
2131 struct xfs_mount *mp,
2132 xlog_recover_item_t *item,
2134 xfs_buf_log_format_t *buf_f)
2140 int reg_buf_offset = 0;
2141 int reg_buf_bytes = 0;
2142 int next_unlinked_offset;
2144 xfs_agino_t *logged_nextp;
2145 xfs_agino_t *buffer_nextp;
2147 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2150 * Post recovery validation only works properly on CRC enabled
2153 if (xfs_sb_version_hascrc(&mp->m_sb))
2154 bp->b_ops = &xfs_inode_buf_ops;
2156 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
2157 for (i = 0; i < inodes_per_buf; i++) {
2158 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2159 offsetof(xfs_dinode_t, di_next_unlinked);
2161 while (next_unlinked_offset >=
2162 (reg_buf_offset + reg_buf_bytes)) {
2164 * The next di_next_unlinked field is beyond
2165 * the current logged region. Find the next
2166 * logged region that contains or is beyond
2167 * the current di_next_unlinked field.
2170 bit = xfs_next_bit(buf_f->blf_data_map,
2171 buf_f->blf_map_size, bit);
2174 * If there are no more logged regions in the
2175 * buffer, then we're done.
2180 nbits = xfs_contig_bits(buf_f->blf_data_map,
2181 buf_f->blf_map_size, bit);
2183 reg_buf_offset = bit << XFS_BLF_SHIFT;
2184 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2189 * If the current logged region starts after the current
2190 * di_next_unlinked field, then move on to the next
2191 * di_next_unlinked field.
2193 if (next_unlinked_offset < reg_buf_offset)
2196 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2197 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2198 ASSERT((reg_buf_offset + reg_buf_bytes) <=
2199 BBTOB(bp->b_io_length));
2202 * The current logged region contains a copy of the
2203 * current di_next_unlinked field. Extract its value
2204 * and copy it to the buffer copy.
2206 logged_nextp = item->ri_buf[item_index].i_addr +
2207 next_unlinked_offset - reg_buf_offset;
2208 if (unlikely(*logged_nextp == 0)) {
2210 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2211 "Trying to replay bad (0) inode di_next_unlinked field.",
2213 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2214 XFS_ERRLEVEL_LOW, mp);
2215 return -EFSCORRUPTED;
2218 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2219 *buffer_nextp = *logged_nextp;
2222 * If necessary, recalculate the CRC in the on-disk inode. We
2223 * have to leave the inode in a consistent state for whoever
2226 xfs_dinode_calc_crc(mp,
2227 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2235 * V5 filesystems know the age of the buffer on disk being recovered. We can
2236 * have newer objects on disk than we are replaying, and so for these cases we
2237 * don't want to replay the current change as that will make the buffer contents
2238 * temporarily invalid on disk.
2240 * The magic number might not match the buffer type we are going to recover
2241 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2242 * extract the LSN of the existing object in the buffer based on it's current
2243 * magic number. If we don't recognise the magic number in the buffer, then
2244 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2245 * so can recover the buffer.
2247 * Note: we cannot rely solely on magic number matches to determine that the
2248 * buffer has a valid LSN - we also need to verify that it belongs to this
2249 * filesystem, so we need to extract the object's LSN and compare it to that
2250 * which we read from the superblock. If the UUIDs don't match, then we've got a
2251 * stale metadata block from an old filesystem instance that we need to recover
2255 xlog_recover_get_buf_lsn(
2256 struct xfs_mount *mp,
2262 void *blk = bp->b_addr;
2266 /* v4 filesystems always recover immediately */
2267 if (!xfs_sb_version_hascrc(&mp->m_sb))
2268 goto recover_immediately;
2270 magic32 = be32_to_cpu(*(__be32 *)blk);
2272 case XFS_ABTB_CRC_MAGIC:
2273 case XFS_ABTC_CRC_MAGIC:
2274 case XFS_ABTB_MAGIC:
2275 case XFS_ABTC_MAGIC:
2276 case XFS_RMAP_CRC_MAGIC:
2277 case XFS_REFC_CRC_MAGIC:
2278 case XFS_IBT_CRC_MAGIC:
2279 case XFS_IBT_MAGIC: {
2280 struct xfs_btree_block *btb = blk;
2282 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2283 uuid = &btb->bb_u.s.bb_uuid;
2286 case XFS_BMAP_CRC_MAGIC:
2287 case XFS_BMAP_MAGIC: {
2288 struct xfs_btree_block *btb = blk;
2290 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2291 uuid = &btb->bb_u.l.bb_uuid;
2295 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2296 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2298 case XFS_AGFL_MAGIC:
2299 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2300 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2303 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2304 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2306 case XFS_SYMLINK_MAGIC:
2307 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2308 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2310 case XFS_DIR3_BLOCK_MAGIC:
2311 case XFS_DIR3_DATA_MAGIC:
2312 case XFS_DIR3_FREE_MAGIC:
2313 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2314 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2316 case XFS_ATTR3_RMT_MAGIC:
2318 * Remote attr blocks are written synchronously, rather than
2319 * being logged. That means they do not contain a valid LSN
2320 * (i.e. transactionally ordered) in them, and hence any time we
2321 * see a buffer to replay over the top of a remote attribute
2322 * block we should simply do so.
2324 goto recover_immediately;
2327 * superblock uuids are magic. We may or may not have a
2328 * sb_meta_uuid on disk, but it will be set in the in-core
2329 * superblock. We set the uuid pointer for verification
2330 * according to the superblock feature mask to ensure we check
2331 * the relevant UUID in the superblock.
2333 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2334 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2335 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2337 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2343 if (lsn != (xfs_lsn_t)-1) {
2344 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2345 goto recover_immediately;
2349 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2351 case XFS_DIR3_LEAF1_MAGIC:
2352 case XFS_DIR3_LEAFN_MAGIC:
2353 case XFS_DA3_NODE_MAGIC:
2354 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2355 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2361 if (lsn != (xfs_lsn_t)-1) {
2362 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2363 goto recover_immediately;
2368 * We do individual object checks on dquot and inode buffers as they
2369 * have their own individual LSN records. Also, we could have a stale
2370 * buffer here, so we have to at least recognise these buffer types.
2372 * A notd complexity here is inode unlinked list processing - it logs
2373 * the inode directly in the buffer, but we don't know which inodes have
2374 * been modified, and there is no global buffer LSN. Hence we need to
2375 * recover all inode buffer types immediately. This problem will be
2376 * fixed by logical logging of the unlinked list modifications.
2378 magic16 = be16_to_cpu(*(__be16 *)blk);
2380 case XFS_DQUOT_MAGIC:
2381 case XFS_DINODE_MAGIC:
2382 goto recover_immediately;
2387 /* unknown buffer contents, recover immediately */
2389 recover_immediately:
2390 return (xfs_lsn_t)-1;
2395 * Validate the recovered buffer is of the correct type and attach the
2396 * appropriate buffer operations to them for writeback. Magic numbers are in a
2398 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2399 * the first 32 bits of the buffer (most blocks),
2400 * inside a struct xfs_da_blkinfo at the start of the buffer.
2403 xlog_recover_validate_buf_type(
2404 struct xfs_mount *mp,
2406 xfs_buf_log_format_t *buf_f,
2407 xfs_lsn_t current_lsn)
2409 struct xfs_da_blkinfo *info = bp->b_addr;
2413 char *warnmsg = NULL;
2416 * We can only do post recovery validation on items on CRC enabled
2417 * fielsystems as we need to know when the buffer was written to be able
2418 * to determine if we should have replayed the item. If we replay old
2419 * metadata over a newer buffer, then it will enter a temporarily
2420 * inconsistent state resulting in verification failures. Hence for now
2421 * just avoid the verification stage for non-crc filesystems
2423 if (!xfs_sb_version_hascrc(&mp->m_sb))
2426 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2427 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2428 magicda = be16_to_cpu(info->magic);
2429 switch (xfs_blft_from_flags(buf_f)) {
2430 case XFS_BLFT_BTREE_BUF:
2432 case XFS_ABTB_CRC_MAGIC:
2433 case XFS_ABTC_CRC_MAGIC:
2434 case XFS_ABTB_MAGIC:
2435 case XFS_ABTC_MAGIC:
2436 bp->b_ops = &xfs_allocbt_buf_ops;
2438 case XFS_IBT_CRC_MAGIC:
2439 case XFS_FIBT_CRC_MAGIC:
2441 case XFS_FIBT_MAGIC:
2442 bp->b_ops = &xfs_inobt_buf_ops;
2444 case XFS_BMAP_CRC_MAGIC:
2445 case XFS_BMAP_MAGIC:
2446 bp->b_ops = &xfs_bmbt_buf_ops;
2448 case XFS_RMAP_CRC_MAGIC:
2449 bp->b_ops = &xfs_rmapbt_buf_ops;
2451 case XFS_REFC_CRC_MAGIC:
2452 bp->b_ops = &xfs_refcountbt_buf_ops;
2455 warnmsg = "Bad btree block magic!";
2459 case XFS_BLFT_AGF_BUF:
2460 if (magic32 != XFS_AGF_MAGIC) {
2461 warnmsg = "Bad AGF block magic!";
2464 bp->b_ops = &xfs_agf_buf_ops;
2466 case XFS_BLFT_AGFL_BUF:
2467 if (magic32 != XFS_AGFL_MAGIC) {
2468 warnmsg = "Bad AGFL block magic!";
2471 bp->b_ops = &xfs_agfl_buf_ops;
2473 case XFS_BLFT_AGI_BUF:
2474 if (magic32 != XFS_AGI_MAGIC) {
2475 warnmsg = "Bad AGI block magic!";
2478 bp->b_ops = &xfs_agi_buf_ops;
2480 case XFS_BLFT_UDQUOT_BUF:
2481 case XFS_BLFT_PDQUOT_BUF:
2482 case XFS_BLFT_GDQUOT_BUF:
2483 #ifdef CONFIG_XFS_QUOTA
2484 if (magic16 != XFS_DQUOT_MAGIC) {
2485 warnmsg = "Bad DQUOT block magic!";
2488 bp->b_ops = &xfs_dquot_buf_ops;
2491 "Trying to recover dquots without QUOTA support built in!");
2495 case XFS_BLFT_DINO_BUF:
2496 if (magic16 != XFS_DINODE_MAGIC) {
2497 warnmsg = "Bad INODE block magic!";
2500 bp->b_ops = &xfs_inode_buf_ops;
2502 case XFS_BLFT_SYMLINK_BUF:
2503 if (magic32 != XFS_SYMLINK_MAGIC) {
2504 warnmsg = "Bad symlink block magic!";
2507 bp->b_ops = &xfs_symlink_buf_ops;
2509 case XFS_BLFT_DIR_BLOCK_BUF:
2510 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2511 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2512 warnmsg = "Bad dir block magic!";
2515 bp->b_ops = &xfs_dir3_block_buf_ops;
2517 case XFS_BLFT_DIR_DATA_BUF:
2518 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2519 magic32 != XFS_DIR3_DATA_MAGIC) {
2520 warnmsg = "Bad dir data magic!";
2523 bp->b_ops = &xfs_dir3_data_buf_ops;
2525 case XFS_BLFT_DIR_FREE_BUF:
2526 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2527 magic32 != XFS_DIR3_FREE_MAGIC) {
2528 warnmsg = "Bad dir3 free magic!";
2531 bp->b_ops = &xfs_dir3_free_buf_ops;
2533 case XFS_BLFT_DIR_LEAF1_BUF:
2534 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2535 magicda != XFS_DIR3_LEAF1_MAGIC) {
2536 warnmsg = "Bad dir leaf1 magic!";
2539 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2541 case XFS_BLFT_DIR_LEAFN_BUF:
2542 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2543 magicda != XFS_DIR3_LEAFN_MAGIC) {
2544 warnmsg = "Bad dir leafn magic!";
2547 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2549 case XFS_BLFT_DA_NODE_BUF:
2550 if (magicda != XFS_DA_NODE_MAGIC &&
2551 magicda != XFS_DA3_NODE_MAGIC) {
2552 warnmsg = "Bad da node magic!";
2555 bp->b_ops = &xfs_da3_node_buf_ops;
2557 case XFS_BLFT_ATTR_LEAF_BUF:
2558 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2559 magicda != XFS_ATTR3_LEAF_MAGIC) {
2560 warnmsg = "Bad attr leaf magic!";
2563 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2565 case XFS_BLFT_ATTR_RMT_BUF:
2566 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2567 warnmsg = "Bad attr remote magic!";
2570 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2572 case XFS_BLFT_SB_BUF:
2573 if (magic32 != XFS_SB_MAGIC) {
2574 warnmsg = "Bad SB block magic!";
2577 bp->b_ops = &xfs_sb_buf_ops;
2579 #ifdef CONFIG_XFS_RT
2580 case XFS_BLFT_RTBITMAP_BUF:
2581 case XFS_BLFT_RTSUMMARY_BUF:
2582 /* no magic numbers for verification of RT buffers */
2583 bp->b_ops = &xfs_rtbuf_ops;
2585 #endif /* CONFIG_XFS_RT */
2587 xfs_warn(mp, "Unknown buffer type %d!",
2588 xfs_blft_from_flags(buf_f));
2593 * Nothing else to do in the case of a NULL current LSN as this means
2594 * the buffer is more recent than the change in the log and will be
2597 if (current_lsn == NULLCOMMITLSN)
2601 xfs_warn(mp, warnmsg);
2606 * We must update the metadata LSN of the buffer as it is written out to
2607 * ensure that older transactions never replay over this one and corrupt
2608 * the buffer. This can occur if log recovery is interrupted at some
2609 * point after the current transaction completes, at which point a
2610 * subsequent mount starts recovery from the beginning.
2612 * Write verifiers update the metadata LSN from log items attached to
2613 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2614 * the verifier. We'll clean it up in our ->iodone() callback.
2617 struct xfs_buf_log_item *bip;
2619 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2620 bp->b_iodone = xlog_recover_iodone;
2621 xfs_buf_item_init(bp, mp);
2623 bip->bli_item.li_lsn = current_lsn;
2628 * Perform a 'normal' buffer recovery. Each logged region of the
2629 * buffer should be copied over the corresponding region in the
2630 * given buffer. The bitmap in the buf log format structure indicates
2631 * where to place the logged data.
2634 xlog_recover_do_reg_buffer(
2635 struct xfs_mount *mp,
2636 xlog_recover_item_t *item,
2638 xfs_buf_log_format_t *buf_f,
2639 xfs_lsn_t current_lsn)
2646 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2649 i = 1; /* 0 is the buf format structure */
2651 bit = xfs_next_bit(buf_f->blf_data_map,
2652 buf_f->blf_map_size, bit);
2655 nbits = xfs_contig_bits(buf_f->blf_data_map,
2656 buf_f->blf_map_size, bit);
2658 ASSERT(item->ri_buf[i].i_addr != NULL);
2659 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2660 ASSERT(BBTOB(bp->b_io_length) >=
2661 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2664 * The dirty regions logged in the buffer, even though
2665 * contiguous, may span multiple chunks. This is because the
2666 * dirty region may span a physical page boundary in a buffer
2667 * and hence be split into two separate vectors for writing into
2668 * the log. Hence we need to trim nbits back to the length of
2669 * the current region being copied out of the log.
2671 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2672 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2675 * Do a sanity check if this is a dquot buffer. Just checking
2676 * the first dquot in the buffer should do. XXXThis is
2677 * probably a good thing to do for other buf types also.
2680 if (buf_f->blf_flags &
2681 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2682 if (item->ri_buf[i].i_addr == NULL) {
2684 "XFS: NULL dquot in %s.", __func__);
2687 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2689 "XFS: dquot too small (%d) in %s.",
2690 item->ri_buf[i].i_len, __func__);
2693 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2694 -1, 0, XFS_QMOPT_DOWARN,
2695 "dquot_buf_recover");
2700 memcpy(xfs_buf_offset(bp,
2701 (uint)bit << XFS_BLF_SHIFT), /* dest */
2702 item->ri_buf[i].i_addr, /* source */
2703 nbits<<XFS_BLF_SHIFT); /* length */
2709 /* Shouldn't be any more regions */
2710 ASSERT(i == item->ri_total);
2712 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2716 * Perform a dquot buffer recovery.
2717 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2718 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2719 * Else, treat it as a regular buffer and do recovery.
2721 * Return false if the buffer was tossed and true if we recovered the buffer to
2722 * indicate to the caller if the buffer needs writing.
2725 xlog_recover_do_dquot_buffer(
2726 struct xfs_mount *mp,
2728 struct xlog_recover_item *item,
2730 struct xfs_buf_log_format *buf_f)
2734 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2737 * Filesystems are required to send in quota flags at mount time.
2743 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2744 type |= XFS_DQ_USER;
2745 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2746 type |= XFS_DQ_PROJ;
2747 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2748 type |= XFS_DQ_GROUP;
2750 * This type of quotas was turned off, so ignore this buffer
2752 if (log->l_quotaoffs_flag & type)
2755 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2760 * This routine replays a modification made to a buffer at runtime.
2761 * There are actually two types of buffer, regular and inode, which
2762 * are handled differently. Inode buffers are handled differently
2763 * in that we only recover a specific set of data from them, namely
2764 * the inode di_next_unlinked fields. This is because all other inode
2765 * data is actually logged via inode records and any data we replay
2766 * here which overlaps that may be stale.
2768 * When meta-data buffers are freed at run time we log a buffer item
2769 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2770 * of the buffer in the log should not be replayed at recovery time.
2771 * This is so that if the blocks covered by the buffer are reused for
2772 * file data before we crash we don't end up replaying old, freed
2773 * meta-data into a user's file.
2775 * To handle the cancellation of buffer log items, we make two passes
2776 * over the log during recovery. During the first we build a table of
2777 * those buffers which have been cancelled, and during the second we
2778 * only replay those buffers which do not have corresponding cancel
2779 * records in the table. See xlog_recover_buffer_pass[1,2] above
2780 * for more details on the implementation of the table of cancel records.
2783 xlog_recover_buffer_pass2(
2785 struct list_head *buffer_list,
2786 struct xlog_recover_item *item,
2787 xfs_lsn_t current_lsn)
2789 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2790 xfs_mount_t *mp = log->l_mp;
2797 * In this pass we only want to recover all the buffers which have
2798 * not been cancelled and are not cancellation buffers themselves.
2800 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2801 buf_f->blf_len, buf_f->blf_flags)) {
2802 trace_xfs_log_recover_buf_cancel(log, buf_f);
2806 trace_xfs_log_recover_buf_recover(log, buf_f);
2809 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2810 buf_flags |= XBF_UNMAPPED;
2812 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2816 error = bp->b_error;
2818 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2823 * Recover the buffer only if we get an LSN from it and it's less than
2824 * the lsn of the transaction we are replaying.
2826 * Note that we have to be extremely careful of readahead here.
2827 * Readahead does not attach verfiers to the buffers so if we don't
2828 * actually do any replay after readahead because of the LSN we found
2829 * in the buffer if more recent than that current transaction then we
2830 * need to attach the verifier directly. Failure to do so can lead to
2831 * future recovery actions (e.g. EFI and unlinked list recovery) can
2832 * operate on the buffers and they won't get the verifier attached. This
2833 * can lead to blocks on disk having the correct content but a stale
2836 * It is safe to assume these clean buffers are currently up to date.
2837 * If the buffer is dirtied by a later transaction being replayed, then
2838 * the verifier will be reset to match whatever recover turns that
2841 lsn = xlog_recover_get_buf_lsn(mp, bp);
2842 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2843 trace_xfs_log_recover_buf_skip(log, buf_f);
2844 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2848 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2849 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2852 } else if (buf_f->blf_flags &
2853 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2856 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2860 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2864 * Perform delayed write on the buffer. Asynchronous writes will be
2865 * slower when taking into account all the buffers to be flushed.
2867 * Also make sure that only inode buffers with good sizes stay in
2868 * the buffer cache. The kernel moves inodes in buffers of 1 block
2869 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2870 * buffers in the log can be a different size if the log was generated
2871 * by an older kernel using unclustered inode buffers or a newer kernel
2872 * running with a different inode cluster size. Regardless, if the
2873 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2874 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2875 * the buffer out of the buffer cache so that the buffer won't
2876 * overlap with future reads of those inodes.
2878 if (XFS_DINODE_MAGIC ==
2879 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2880 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2881 (__uint32_t)log->l_mp->m_inode_cluster_size))) {
2883 error = xfs_bwrite(bp);
2885 ASSERT(bp->b_target->bt_mount == mp);
2886 bp->b_iodone = xlog_recover_iodone;
2887 xfs_buf_delwri_queue(bp, buffer_list);
2896 * Inode fork owner changes
2898 * If we have been told that we have to reparent the inode fork, it's because an
2899 * extent swap operation on a CRC enabled filesystem has been done and we are
2900 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2903 * The complexity here is that we don't have an inode context to work with, so
2904 * after we've replayed the inode we need to instantiate one. This is where the
2907 * We are in the middle of log recovery, so we can't run transactions. That
2908 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2909 * that will result in the corresponding iput() running the inode through
2910 * xfs_inactive(). If we've just replayed an inode core that changes the link
2911 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2912 * transactions (bad!).
2914 * So, to avoid this, we instantiate an inode directly from the inode core we've
2915 * just recovered. We have the buffer still locked, and all we really need to
2916 * instantiate is the inode core and the forks being modified. We can do this
2917 * manually, then run the inode btree owner change, and then tear down the
2918 * xfs_inode without having to run any transactions at all.
2920 * Also, because we don't have a transaction context available here but need to
2921 * gather all the buffers we modify for writeback so we pass the buffer_list
2922 * instead for the operation to use.
2926 xfs_recover_inode_owner_change(
2927 struct xfs_mount *mp,
2928 struct xfs_dinode *dip,
2929 struct xfs_inode_log_format *in_f,
2930 struct list_head *buffer_list)
2932 struct xfs_inode *ip;
2935 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2937 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2941 /* instantiate the inode */
2942 xfs_inode_from_disk(ip, dip);
2943 ASSERT(ip->i_d.di_version >= 3);
2945 error = xfs_iformat_fork(ip, dip);
2950 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2951 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2952 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2953 ip->i_ino, buffer_list);
2958 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2959 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2960 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2961 ip->i_ino, buffer_list);
2972 xlog_recover_inode_pass2(
2974 struct list_head *buffer_list,
2975 struct xlog_recover_item *item,
2976 xfs_lsn_t current_lsn)
2978 xfs_inode_log_format_t *in_f;
2979 xfs_mount_t *mp = log->l_mp;
2988 struct xfs_log_dinode *ldip;
2992 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2993 in_f = item->ri_buf[0].i_addr;
2995 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2997 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
3003 * Inode buffers can be freed, look out for it,
3004 * and do not replay the inode.
3006 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
3007 in_f->ilf_len, 0)) {
3009 trace_xfs_log_recover_inode_cancel(log, in_f);
3012 trace_xfs_log_recover_inode_recover(log, in_f);
3014 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
3015 &xfs_inode_buf_ops);
3020 error = bp->b_error;
3022 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
3025 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
3026 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
3029 * Make sure the place we're flushing out to really looks
3032 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
3034 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
3035 __func__, dip, bp, in_f->ilf_ino);
3036 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
3037 XFS_ERRLEVEL_LOW, mp);
3038 error = -EFSCORRUPTED;
3041 ldip = item->ri_buf[1].i_addr;
3042 if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) {
3044 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
3045 __func__, item, in_f->ilf_ino);
3046 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
3047 XFS_ERRLEVEL_LOW, mp);
3048 error = -EFSCORRUPTED;
3053 * If the inode has an LSN in it, recover the inode only if it's less
3054 * than the lsn of the transaction we are replaying. Note: we still
3055 * need to replay an owner change even though the inode is more recent
3056 * than the transaction as there is no guarantee that all the btree
3057 * blocks are more recent than this transaction, too.
3059 if (dip->di_version >= 3) {
3060 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
3062 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3063 trace_xfs_log_recover_inode_skip(log, in_f);
3065 goto out_owner_change;
3070 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3071 * are transactional and if ordering is necessary we can determine that
3072 * more accurately by the LSN field in the V3 inode core. Don't trust
3073 * the inode versions we might be changing them here - use the
3074 * superblock flag to determine whether we need to look at di_flushiter
3075 * to skip replay when the on disk inode is newer than the log one
3077 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3078 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3080 * Deal with the wrap case, DI_MAX_FLUSH is less
3081 * than smaller numbers
3083 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3084 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3087 trace_xfs_log_recover_inode_skip(log, in_f);
3093 /* Take the opportunity to reset the flush iteration count */
3094 ldip->di_flushiter = 0;
3096 if (unlikely(S_ISREG(ldip->di_mode))) {
3097 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3098 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3099 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3100 XFS_ERRLEVEL_LOW, mp, ldip);
3102 "%s: Bad regular inode log record, rec ptr 0x%p, "
3103 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3104 __func__, item, dip, bp, in_f->ilf_ino);
3105 error = -EFSCORRUPTED;
3108 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3109 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3110 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3111 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3112 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3113 XFS_ERRLEVEL_LOW, mp, ldip);
3115 "%s: Bad dir inode log record, rec ptr 0x%p, "
3116 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3117 __func__, item, dip, bp, in_f->ilf_ino);
3118 error = -EFSCORRUPTED;
3122 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3123 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3124 XFS_ERRLEVEL_LOW, mp, ldip);
3126 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3127 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
3128 __func__, item, dip, bp, in_f->ilf_ino,
3129 ldip->di_nextents + ldip->di_anextents,
3131 error = -EFSCORRUPTED;
3134 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3135 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3136 XFS_ERRLEVEL_LOW, mp, ldip);
3138 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3139 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
3140 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3141 error = -EFSCORRUPTED;
3144 isize = xfs_log_dinode_size(ldip->di_version);
3145 if (unlikely(item->ri_buf[1].i_len > isize)) {
3146 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3147 XFS_ERRLEVEL_LOW, mp, ldip);
3149 "%s: Bad inode log record length %d, rec ptr 0x%p",
3150 __func__, item->ri_buf[1].i_len, item);
3151 error = -EFSCORRUPTED;
3155 /* recover the log dinode inode into the on disk inode */
3156 xfs_log_dinode_to_disk(ldip, dip);
3158 /* the rest is in on-disk format */
3159 if (item->ri_buf[1].i_len > isize) {
3160 memcpy((char *)dip + isize,
3161 item->ri_buf[1].i_addr + isize,
3162 item->ri_buf[1].i_len - isize);
3165 fields = in_f->ilf_fields;
3166 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
3168 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3171 memcpy(XFS_DFORK_DPTR(dip),
3172 &in_f->ilf_u.ilfu_uuid,
3177 if (in_f->ilf_size == 2)
3178 goto out_owner_change;
3179 len = item->ri_buf[2].i_len;
3180 src = item->ri_buf[2].i_addr;
3181 ASSERT(in_f->ilf_size <= 4);
3182 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3183 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3184 (len == in_f->ilf_dsize));
3186 switch (fields & XFS_ILOG_DFORK) {
3187 case XFS_ILOG_DDATA:
3189 memcpy(XFS_DFORK_DPTR(dip), src, len);
3192 case XFS_ILOG_DBROOT:
3193 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3194 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3195 XFS_DFORK_DSIZE(dip, mp));
3200 * There are no data fork flags set.
3202 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3207 * If we logged any attribute data, recover it. There may or
3208 * may not have been any other non-core data logged in this
3211 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3212 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3217 len = item->ri_buf[attr_index].i_len;
3218 src = item->ri_buf[attr_index].i_addr;
3219 ASSERT(len == in_f->ilf_asize);
3221 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3222 case XFS_ILOG_ADATA:
3224 dest = XFS_DFORK_APTR(dip);
3225 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3226 memcpy(dest, src, len);
3229 case XFS_ILOG_ABROOT:
3230 dest = XFS_DFORK_APTR(dip);
3231 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3232 len, (xfs_bmdr_block_t*)dest,
3233 XFS_DFORK_ASIZE(dip, mp));
3237 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3245 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
3246 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3248 /* re-generate the checksum. */
3249 xfs_dinode_calc_crc(log->l_mp, dip);
3251 ASSERT(bp->b_target->bt_mount == mp);
3252 bp->b_iodone = xlog_recover_iodone;
3253 xfs_buf_delwri_queue(bp, buffer_list);
3264 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3265 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3269 xlog_recover_quotaoff_pass1(
3271 struct xlog_recover_item *item)
3273 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3277 * The logitem format's flag tells us if this was user quotaoff,
3278 * group/project quotaoff or both.
3280 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3281 log->l_quotaoffs_flag |= XFS_DQ_USER;
3282 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3283 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3284 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3285 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3291 * Recover a dquot record
3294 xlog_recover_dquot_pass2(
3296 struct list_head *buffer_list,
3297 struct xlog_recover_item *item,
3298 xfs_lsn_t current_lsn)
3300 xfs_mount_t *mp = log->l_mp;
3302 struct xfs_disk_dquot *ddq, *recddq;
3304 xfs_dq_logformat_t *dq_f;
3309 * Filesystems are required to send in quota flags at mount time.
3311 if (mp->m_qflags == 0)
3314 recddq = item->ri_buf[1].i_addr;
3315 if (recddq == NULL) {
3316 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3319 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3320 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3321 item->ri_buf[1].i_len, __func__);
3326 * This type of quotas was turned off, so ignore this record.
3328 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3330 if (log->l_quotaoffs_flag & type)
3334 * At this point we know that quota was _not_ turned off.
3335 * Since the mount flags are not indicating to us otherwise, this
3336 * must mean that quota is on, and the dquot needs to be replayed.
3337 * Remember that we may not have fully recovered the superblock yet,
3338 * so we can't do the usual trick of looking at the SB quota bits.
3340 * The other possibility, of course, is that the quota subsystem was
3341 * removed since the last mount - ENOSYS.
3343 dq_f = item->ri_buf[0].i_addr;
3345 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3346 "xlog_recover_dquot_pass2 (log copy)");
3349 ASSERT(dq_f->qlf_len == 1);
3352 * At this point we are assuming that the dquots have been allocated
3353 * and hence the buffer has valid dquots stamped in it. It should,
3354 * therefore, pass verifier validation. If the dquot is bad, then the
3355 * we'll return an error here, so we don't need to specifically check
3356 * the dquot in the buffer after the verifier has run.
3358 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3359 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3360 &xfs_dquot_buf_ops);
3365 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3368 * If the dquot has an LSN in it, recover the dquot only if it's less
3369 * than the lsn of the transaction we are replaying.
3371 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3372 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3373 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3375 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3380 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3381 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3382 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3386 ASSERT(dq_f->qlf_size == 2);
3387 ASSERT(bp->b_target->bt_mount == mp);
3388 bp->b_iodone = xlog_recover_iodone;
3389 xfs_buf_delwri_queue(bp, buffer_list);
3397 * This routine is called to create an in-core extent free intent
3398 * item from the efi format structure which was logged on disk.
3399 * It allocates an in-core efi, copies the extents from the format
3400 * structure into it, and adds the efi to the AIL with the given
3404 xlog_recover_efi_pass2(
3406 struct xlog_recover_item *item,
3410 struct xfs_mount *mp = log->l_mp;
3411 struct xfs_efi_log_item *efip;
3412 struct xfs_efi_log_format *efi_formatp;
3414 efi_formatp = item->ri_buf[0].i_addr;
3416 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3417 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3419 xfs_efi_item_free(efip);
3422 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3424 spin_lock(&log->l_ailp->xa_lock);
3426 * The EFI has two references. One for the EFD and one for EFI to ensure
3427 * it makes it into the AIL. Insert the EFI into the AIL directly and
3428 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3431 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3432 xfs_efi_release(efip);
3438 * This routine is called when an EFD format structure is found in a committed
3439 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3440 * was still in the log. To do this it searches the AIL for the EFI with an id
3441 * equal to that in the EFD format structure. If we find it we drop the EFD
3442 * reference, which removes the EFI from the AIL and frees it.
3445 xlog_recover_efd_pass2(
3447 struct xlog_recover_item *item)
3449 xfs_efd_log_format_t *efd_formatp;
3450 xfs_efi_log_item_t *efip = NULL;
3451 xfs_log_item_t *lip;
3453 struct xfs_ail_cursor cur;
3454 struct xfs_ail *ailp = log->l_ailp;
3456 efd_formatp = item->ri_buf[0].i_addr;
3457 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3458 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3459 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3460 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3461 efi_id = efd_formatp->efd_efi_id;
3464 * Search for the EFI with the id in the EFD format structure in the
3467 spin_lock(&ailp->xa_lock);
3468 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3469 while (lip != NULL) {
3470 if (lip->li_type == XFS_LI_EFI) {
3471 efip = (xfs_efi_log_item_t *)lip;
3472 if (efip->efi_format.efi_id == efi_id) {
3474 * Drop the EFD reference to the EFI. This
3475 * removes the EFI from the AIL and frees it.
3477 spin_unlock(&ailp->xa_lock);
3478 xfs_efi_release(efip);
3479 spin_lock(&ailp->xa_lock);
3483 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3486 xfs_trans_ail_cursor_done(&cur);
3487 spin_unlock(&ailp->xa_lock);
3493 * This routine is called to create an in-core extent rmap update
3494 * item from the rui format structure which was logged on disk.
3495 * It allocates an in-core rui, copies the extents from the format
3496 * structure into it, and adds the rui to the AIL with the given
3500 xlog_recover_rui_pass2(
3502 struct xlog_recover_item *item,
3506 struct xfs_mount *mp = log->l_mp;
3507 struct xfs_rui_log_item *ruip;
3508 struct xfs_rui_log_format *rui_formatp;
3510 rui_formatp = item->ri_buf[0].i_addr;
3512 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3513 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3515 xfs_rui_item_free(ruip);
3518 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3520 spin_lock(&log->l_ailp->xa_lock);
3522 * The RUI has two references. One for the RUD and one for RUI to ensure
3523 * it makes it into the AIL. Insert the RUI into the AIL directly and
3524 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3527 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3528 xfs_rui_release(ruip);
3534 * This routine is called when an RUD format structure is found in a committed
3535 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3536 * was still in the log. To do this it searches the AIL for the RUI with an id
3537 * equal to that in the RUD format structure. If we find it we drop the RUD
3538 * reference, which removes the RUI from the AIL and frees it.
3541 xlog_recover_rud_pass2(
3543 struct xlog_recover_item *item)
3545 struct xfs_rud_log_format *rud_formatp;
3546 struct xfs_rui_log_item *ruip = NULL;
3547 struct xfs_log_item *lip;
3549 struct xfs_ail_cursor cur;
3550 struct xfs_ail *ailp = log->l_ailp;
3552 rud_formatp = item->ri_buf[0].i_addr;
3553 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3554 rui_id = rud_formatp->rud_rui_id;
3557 * Search for the RUI with the id in the RUD format structure in the
3560 spin_lock(&ailp->xa_lock);
3561 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3562 while (lip != NULL) {
3563 if (lip->li_type == XFS_LI_RUI) {
3564 ruip = (struct xfs_rui_log_item *)lip;
3565 if (ruip->rui_format.rui_id == rui_id) {
3567 * Drop the RUD reference to the RUI. This
3568 * removes the RUI from the AIL and frees it.
3570 spin_unlock(&ailp->xa_lock);
3571 xfs_rui_release(ruip);
3572 spin_lock(&ailp->xa_lock);
3576 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3579 xfs_trans_ail_cursor_done(&cur);
3580 spin_unlock(&ailp->xa_lock);
3586 * Copy an CUI format buffer from the given buf, and into the destination
3587 * CUI format structure. The CUI/CUD items were designed not to need any
3588 * special alignment handling.
3591 xfs_cui_copy_format(
3592 struct xfs_log_iovec *buf,
3593 struct xfs_cui_log_format *dst_cui_fmt)
3595 struct xfs_cui_log_format *src_cui_fmt;
3598 src_cui_fmt = buf->i_addr;
3599 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3601 if (buf->i_len == len) {
3602 memcpy(dst_cui_fmt, src_cui_fmt, len);
3605 return -EFSCORRUPTED;
3609 * This routine is called to create an in-core extent refcount update
3610 * item from the cui format structure which was logged on disk.
3611 * It allocates an in-core cui, copies the extents from the format
3612 * structure into it, and adds the cui to the AIL with the given
3616 xlog_recover_cui_pass2(
3618 struct xlog_recover_item *item,
3622 struct xfs_mount *mp = log->l_mp;
3623 struct xfs_cui_log_item *cuip;
3624 struct xfs_cui_log_format *cui_formatp;
3626 cui_formatp = item->ri_buf[0].i_addr;
3628 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3629 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3631 xfs_cui_item_free(cuip);
3634 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3636 spin_lock(&log->l_ailp->xa_lock);
3638 * The CUI has two references. One for the CUD and one for CUI to ensure
3639 * it makes it into the AIL. Insert the CUI into the AIL directly and
3640 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3643 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3644 xfs_cui_release(cuip);
3650 * This routine is called when an CUD format structure is found in a committed
3651 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3652 * was still in the log. To do this it searches the AIL for the CUI with an id
3653 * equal to that in the CUD format structure. If we find it we drop the CUD
3654 * reference, which removes the CUI from the AIL and frees it.
3657 xlog_recover_cud_pass2(
3659 struct xlog_recover_item *item)
3661 struct xfs_cud_log_format *cud_formatp;
3662 struct xfs_cui_log_item *cuip = NULL;
3663 struct xfs_log_item *lip;
3665 struct xfs_ail_cursor cur;
3666 struct xfs_ail *ailp = log->l_ailp;
3668 cud_formatp = item->ri_buf[0].i_addr;
3669 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format))
3670 return -EFSCORRUPTED;
3671 cui_id = cud_formatp->cud_cui_id;
3674 * Search for the CUI with the id in the CUD format structure in the
3677 spin_lock(&ailp->xa_lock);
3678 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3679 while (lip != NULL) {
3680 if (lip->li_type == XFS_LI_CUI) {
3681 cuip = (struct xfs_cui_log_item *)lip;
3682 if (cuip->cui_format.cui_id == cui_id) {
3684 * Drop the CUD reference to the CUI. This
3685 * removes the CUI from the AIL and frees it.
3687 spin_unlock(&ailp->xa_lock);
3688 xfs_cui_release(cuip);
3689 spin_lock(&ailp->xa_lock);
3693 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3696 xfs_trans_ail_cursor_done(&cur);
3697 spin_unlock(&ailp->xa_lock);
3703 * Copy an BUI format buffer from the given buf, and into the destination
3704 * BUI format structure. The BUI/BUD items were designed not to need any
3705 * special alignment handling.
3708 xfs_bui_copy_format(
3709 struct xfs_log_iovec *buf,
3710 struct xfs_bui_log_format *dst_bui_fmt)
3712 struct xfs_bui_log_format *src_bui_fmt;
3715 src_bui_fmt = buf->i_addr;
3716 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3718 if (buf->i_len == len) {
3719 memcpy(dst_bui_fmt, src_bui_fmt, len);
3722 return -EFSCORRUPTED;
3726 * This routine is called to create an in-core extent bmap update
3727 * item from the bui format structure which was logged on disk.
3728 * It allocates an in-core bui, copies the extents from the format
3729 * structure into it, and adds the bui to the AIL with the given
3733 xlog_recover_bui_pass2(
3735 struct xlog_recover_item *item,
3739 struct xfs_mount *mp = log->l_mp;
3740 struct xfs_bui_log_item *buip;
3741 struct xfs_bui_log_format *bui_formatp;
3743 bui_formatp = item->ri_buf[0].i_addr;
3745 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
3746 return -EFSCORRUPTED;
3747 buip = xfs_bui_init(mp);
3748 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3750 xfs_bui_item_free(buip);
3753 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3755 spin_lock(&log->l_ailp->xa_lock);
3757 * The RUI has two references. One for the RUD and one for RUI to ensure
3758 * it makes it into the AIL. Insert the RUI into the AIL directly and
3759 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3762 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3763 xfs_bui_release(buip);
3769 * This routine is called when an BUD format structure is found in a committed
3770 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3771 * was still in the log. To do this it searches the AIL for the BUI with an id
3772 * equal to that in the BUD format structure. If we find it we drop the BUD
3773 * reference, which removes the BUI from the AIL and frees it.
3776 xlog_recover_bud_pass2(
3778 struct xlog_recover_item *item)
3780 struct xfs_bud_log_format *bud_formatp;
3781 struct xfs_bui_log_item *buip = NULL;
3782 struct xfs_log_item *lip;
3784 struct xfs_ail_cursor cur;
3785 struct xfs_ail *ailp = log->l_ailp;
3787 bud_formatp = item->ri_buf[0].i_addr;
3788 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format))
3789 return -EFSCORRUPTED;
3790 bui_id = bud_formatp->bud_bui_id;
3793 * Search for the BUI with the id in the BUD format structure in the
3796 spin_lock(&ailp->xa_lock);
3797 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3798 while (lip != NULL) {
3799 if (lip->li_type == XFS_LI_BUI) {
3800 buip = (struct xfs_bui_log_item *)lip;
3801 if (buip->bui_format.bui_id == bui_id) {
3803 * Drop the BUD reference to the BUI. This
3804 * removes the BUI from the AIL and frees it.
3806 spin_unlock(&ailp->xa_lock);
3807 xfs_bui_release(buip);
3808 spin_lock(&ailp->xa_lock);
3812 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3815 xfs_trans_ail_cursor_done(&cur);
3816 spin_unlock(&ailp->xa_lock);
3822 * This routine is called when an inode create format structure is found in a
3823 * committed transaction in the log. It's purpose is to initialise the inodes
3824 * being allocated on disk. This requires us to get inode cluster buffers that
3825 * match the range to be intialised, stamped with inode templates and written
3826 * by delayed write so that subsequent modifications will hit the cached buffer
3827 * and only need writing out at the end of recovery.
3830 xlog_recover_do_icreate_pass2(
3832 struct list_head *buffer_list,
3833 xlog_recover_item_t *item)
3835 struct xfs_mount *mp = log->l_mp;
3836 struct xfs_icreate_log *icl;
3837 xfs_agnumber_t agno;
3838 xfs_agblock_t agbno;
3841 xfs_agblock_t length;
3842 int blks_per_cluster;
3848 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3849 if (icl->icl_type != XFS_LI_ICREATE) {
3850 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3854 if (icl->icl_size != 1) {
3855 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3859 agno = be32_to_cpu(icl->icl_ag);
3860 if (agno >= mp->m_sb.sb_agcount) {
3861 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3864 agbno = be32_to_cpu(icl->icl_agbno);
3865 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3866 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3869 isize = be32_to_cpu(icl->icl_isize);
3870 if (isize != mp->m_sb.sb_inodesize) {
3871 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3874 count = be32_to_cpu(icl->icl_count);
3876 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3879 length = be32_to_cpu(icl->icl_length);
3880 if (!length || length >= mp->m_sb.sb_agblocks) {
3881 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3886 * The inode chunk is either full or sparse and we only support
3887 * m_ialloc_min_blks sized sparse allocations at this time.
3889 if (length != mp->m_ialloc_blks &&
3890 length != mp->m_ialloc_min_blks) {
3892 "%s: unsupported chunk length", __FUNCTION__);
3896 /* verify inode count is consistent with extent length */
3897 if ((count >> mp->m_sb.sb_inopblog) != length) {
3899 "%s: inconsistent inode count and chunk length",
3905 * The icreate transaction can cover multiple cluster buffers and these
3906 * buffers could have been freed and reused. Check the individual
3907 * buffers for cancellation so we don't overwrite anything written after
3910 blks_per_cluster = xfs_icluster_size_fsb(mp);
3911 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3912 nbufs = length / blks_per_cluster;
3913 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3916 daddr = XFS_AGB_TO_DADDR(mp, agno,
3917 agbno + i * blks_per_cluster);
3918 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3923 * We currently only use icreate for a single allocation at a time. This
3924 * means we should expect either all or none of the buffers to be
3925 * cancelled. Be conservative and skip replay if at least one buffer is
3926 * cancelled, but warn the user that something is awry if the buffers
3927 * are not consistent.
3929 * XXX: This must be refined to only skip cancelled clusters once we use
3930 * icreate for multiple chunk allocations.
3932 ASSERT(!cancel_count || cancel_count == nbufs);
3934 if (cancel_count != nbufs)
3936 "WARNING: partial inode chunk cancellation, skipped icreate.");
3937 trace_xfs_log_recover_icreate_cancel(log, icl);
3941 trace_xfs_log_recover_icreate_recover(log, icl);
3942 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3943 length, be32_to_cpu(icl->icl_gen));
3947 xlog_recover_buffer_ra_pass2(
3949 struct xlog_recover_item *item)
3951 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3952 struct xfs_mount *mp = log->l_mp;
3954 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3955 buf_f->blf_len, buf_f->blf_flags)) {
3959 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3960 buf_f->blf_len, NULL);
3964 xlog_recover_inode_ra_pass2(
3966 struct xlog_recover_item *item)
3968 struct xfs_inode_log_format ilf_buf;
3969 struct xfs_inode_log_format *ilfp;
3970 struct xfs_mount *mp = log->l_mp;
3973 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3974 ilfp = item->ri_buf[0].i_addr;
3977 memset(ilfp, 0, sizeof(*ilfp));
3978 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3983 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3986 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3987 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3991 xlog_recover_dquot_ra_pass2(
3993 struct xlog_recover_item *item)
3995 struct xfs_mount *mp = log->l_mp;
3996 struct xfs_disk_dquot *recddq;
3997 struct xfs_dq_logformat *dq_f;
4002 if (mp->m_qflags == 0)
4005 recddq = item->ri_buf[1].i_addr;
4008 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
4011 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
4013 if (log->l_quotaoffs_flag & type)
4016 dq_f = item->ri_buf[0].i_addr;
4018 ASSERT(dq_f->qlf_len == 1);
4020 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
4021 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
4024 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
4025 &xfs_dquot_buf_ra_ops);
4029 xlog_recover_ra_pass2(
4031 struct xlog_recover_item *item)
4033 switch (ITEM_TYPE(item)) {
4035 xlog_recover_buffer_ra_pass2(log, item);
4038 xlog_recover_inode_ra_pass2(log, item);
4041 xlog_recover_dquot_ra_pass2(log, item);
4045 case XFS_LI_QUOTAOFF:
4058 xlog_recover_commit_pass1(
4060 struct xlog_recover *trans,
4061 struct xlog_recover_item *item)
4063 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
4065 switch (ITEM_TYPE(item)) {
4067 return xlog_recover_buffer_pass1(log, item);
4068 case XFS_LI_QUOTAOFF:
4069 return xlog_recover_quotaoff_pass1(log, item);
4074 case XFS_LI_ICREATE:
4081 /* nothing to do in pass 1 */
4084 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4085 __func__, ITEM_TYPE(item));
4092 xlog_recover_commit_pass2(
4094 struct xlog_recover *trans,
4095 struct list_head *buffer_list,
4096 struct xlog_recover_item *item)
4098 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4100 switch (ITEM_TYPE(item)) {
4102 return xlog_recover_buffer_pass2(log, buffer_list, item,
4105 return xlog_recover_inode_pass2(log, buffer_list, item,
4108 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4110 return xlog_recover_efd_pass2(log, item);
4112 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4114 return xlog_recover_rud_pass2(log, item);
4116 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4118 return xlog_recover_cud_pass2(log, item);
4120 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4122 return xlog_recover_bud_pass2(log, item);
4124 return xlog_recover_dquot_pass2(log, buffer_list, item,
4126 case XFS_LI_ICREATE:
4127 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4128 case XFS_LI_QUOTAOFF:
4129 /* nothing to do in pass2 */
4132 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4133 __func__, ITEM_TYPE(item));
4140 xlog_recover_items_pass2(
4142 struct xlog_recover *trans,
4143 struct list_head *buffer_list,
4144 struct list_head *item_list)
4146 struct xlog_recover_item *item;
4149 list_for_each_entry(item, item_list, ri_list) {
4150 error = xlog_recover_commit_pass2(log, trans,
4160 * Perform the transaction.
4162 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4163 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4166 xlog_recover_commit_trans(
4168 struct xlog_recover *trans,
4170 struct list_head *buffer_list)
4173 int items_queued = 0;
4174 struct xlog_recover_item *item;
4175 struct xlog_recover_item *next;
4176 LIST_HEAD (ra_list);
4177 LIST_HEAD (done_list);
4179 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4181 hlist_del_init(&trans->r_list);
4183 error = xlog_recover_reorder_trans(log, trans, pass);
4187 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4189 case XLOG_RECOVER_PASS1:
4190 error = xlog_recover_commit_pass1(log, trans, item);
4192 case XLOG_RECOVER_PASS2:
4193 xlog_recover_ra_pass2(log, item);
4194 list_move_tail(&item->ri_list, &ra_list);
4196 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4197 error = xlog_recover_items_pass2(log, trans,
4198 buffer_list, &ra_list);
4199 list_splice_tail_init(&ra_list, &done_list);
4213 if (!list_empty(&ra_list)) {
4215 error = xlog_recover_items_pass2(log, trans,
4216 buffer_list, &ra_list);
4217 list_splice_tail_init(&ra_list, &done_list);
4220 if (!list_empty(&done_list))
4221 list_splice_init(&done_list, &trans->r_itemq);
4227 xlog_recover_add_item(
4228 struct list_head *head)
4230 xlog_recover_item_t *item;
4232 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
4233 INIT_LIST_HEAD(&item->ri_list);
4234 list_add_tail(&item->ri_list, head);
4238 xlog_recover_add_to_cont_trans(
4240 struct xlog_recover *trans,
4244 xlog_recover_item_t *item;
4245 char *ptr, *old_ptr;
4249 * If the transaction is empty, the header was split across this and the
4250 * previous record. Copy the rest of the header.
4252 if (list_empty(&trans->r_itemq)) {
4253 ASSERT(len <= sizeof(struct xfs_trans_header));
4254 if (len > sizeof(struct xfs_trans_header)) {
4255 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4259 xlog_recover_add_item(&trans->r_itemq);
4260 ptr = (char *)&trans->r_theader +
4261 sizeof(struct xfs_trans_header) - len;
4262 memcpy(ptr, dp, len);
4266 /* take the tail entry */
4267 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4269 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4270 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4272 ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP);
4273 memcpy(&ptr[old_len], dp, len);
4274 item->ri_buf[item->ri_cnt-1].i_len += len;
4275 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4276 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4281 * The next region to add is the start of a new region. It could be
4282 * a whole region or it could be the first part of a new region. Because
4283 * of this, the assumption here is that the type and size fields of all
4284 * format structures fit into the first 32 bits of the structure.
4286 * This works because all regions must be 32 bit aligned. Therefore, we
4287 * either have both fields or we have neither field. In the case we have
4288 * neither field, the data part of the region is zero length. We only have
4289 * a log_op_header and can throw away the header since a new one will appear
4290 * later. If we have at least 4 bytes, then we can determine how many regions
4291 * will appear in the current log item.
4294 xlog_recover_add_to_trans(
4296 struct xlog_recover *trans,
4300 xfs_inode_log_format_t *in_f; /* any will do */
4301 xlog_recover_item_t *item;
4306 if (list_empty(&trans->r_itemq)) {
4307 /* we need to catch log corruptions here */
4308 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4309 xfs_warn(log->l_mp, "%s: bad header magic number",
4315 if (len > sizeof(struct xfs_trans_header)) {
4316 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4322 * The transaction header can be arbitrarily split across op
4323 * records. If we don't have the whole thing here, copy what we
4324 * do have and handle the rest in the next record.
4326 if (len == sizeof(struct xfs_trans_header))
4327 xlog_recover_add_item(&trans->r_itemq);
4328 memcpy(&trans->r_theader, dp, len);
4332 ptr = kmem_alloc(len, KM_SLEEP);
4333 memcpy(ptr, dp, len);
4334 in_f = (xfs_inode_log_format_t *)ptr;
4336 /* take the tail entry */
4337 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4338 if (item->ri_total != 0 &&
4339 item->ri_total == item->ri_cnt) {
4340 /* tail item is in use, get a new one */
4341 xlog_recover_add_item(&trans->r_itemq);
4342 item = list_entry(trans->r_itemq.prev,
4343 xlog_recover_item_t, ri_list);
4346 if (item->ri_total == 0) { /* first region to be added */
4347 if (in_f->ilf_size == 0 ||
4348 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4350 "bad number of regions (%d) in inode log format",
4357 item->ri_total = in_f->ilf_size;
4359 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4362 ASSERT(item->ri_total > item->ri_cnt);
4363 /* Description region is ri_buf[0] */
4364 item->ri_buf[item->ri_cnt].i_addr = ptr;
4365 item->ri_buf[item->ri_cnt].i_len = len;
4367 trace_xfs_log_recover_item_add(log, trans, item, 0);
4372 * Free up any resources allocated by the transaction
4374 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4377 xlog_recover_free_trans(
4378 struct xlog_recover *trans)
4380 xlog_recover_item_t *item, *n;
4383 hlist_del_init(&trans->r_list);
4385 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4386 /* Free the regions in the item. */
4387 list_del(&item->ri_list);
4388 for (i = 0; i < item->ri_cnt; i++)
4389 kmem_free(item->ri_buf[i].i_addr);
4390 /* Free the item itself */
4391 kmem_free(item->ri_buf);
4394 /* Free the transaction recover structure */
4399 * On error or completion, trans is freed.
4402 xlog_recovery_process_trans(
4404 struct xlog_recover *trans,
4409 struct list_head *buffer_list)
4412 bool freeit = false;
4414 /* mask off ophdr transaction container flags */
4415 flags &= ~XLOG_END_TRANS;
4416 if (flags & XLOG_WAS_CONT_TRANS)
4417 flags &= ~XLOG_CONTINUE_TRANS;
4420 * Callees must not free the trans structure. We'll decide if we need to
4421 * free it or not based on the operation being done and it's result.
4424 /* expected flag values */
4426 case XLOG_CONTINUE_TRANS:
4427 error = xlog_recover_add_to_trans(log, trans, dp, len);
4429 case XLOG_WAS_CONT_TRANS:
4430 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4432 case XLOG_COMMIT_TRANS:
4433 error = xlog_recover_commit_trans(log, trans, pass,
4435 /* success or fail, we are now done with this transaction. */
4439 /* unexpected flag values */
4440 case XLOG_UNMOUNT_TRANS:
4441 /* just skip trans */
4442 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4445 case XLOG_START_TRANS:
4447 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4452 if (error || freeit)
4453 xlog_recover_free_trans(trans);
4458 * Lookup the transaction recovery structure associated with the ID in the
4459 * current ophdr. If the transaction doesn't exist and the start flag is set in
4460 * the ophdr, then allocate a new transaction for future ID matches to find.
4461 * Either way, return what we found during the lookup - an existing transaction
4464 STATIC struct xlog_recover *
4465 xlog_recover_ophdr_to_trans(
4466 struct hlist_head rhash[],
4467 struct xlog_rec_header *rhead,
4468 struct xlog_op_header *ohead)
4470 struct xlog_recover *trans;
4472 struct hlist_head *rhp;
4474 tid = be32_to_cpu(ohead->oh_tid);
4475 rhp = &rhash[XLOG_RHASH(tid)];
4476 hlist_for_each_entry(trans, rhp, r_list) {
4477 if (trans->r_log_tid == tid)
4482 * skip over non-start transaction headers - we could be
4483 * processing slack space before the next transaction starts
4485 if (!(ohead->oh_flags & XLOG_START_TRANS))
4488 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4491 * This is a new transaction so allocate a new recovery container to
4492 * hold the recovery ops that will follow.
4494 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
4495 trans->r_log_tid = tid;
4496 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4497 INIT_LIST_HEAD(&trans->r_itemq);
4498 INIT_HLIST_NODE(&trans->r_list);
4499 hlist_add_head(&trans->r_list, rhp);
4502 * Nothing more to do for this ophdr. Items to be added to this new
4503 * transaction will be in subsequent ophdr containers.
4509 xlog_recover_process_ophdr(
4511 struct hlist_head rhash[],
4512 struct xlog_rec_header *rhead,
4513 struct xlog_op_header *ohead,
4517 struct list_head *buffer_list)
4519 struct xlog_recover *trans;
4523 /* Do we understand who wrote this op? */
4524 if (ohead->oh_clientid != XFS_TRANSACTION &&
4525 ohead->oh_clientid != XFS_LOG) {
4526 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4527 __func__, ohead->oh_clientid);
4533 * Check the ophdr contains all the data it is supposed to contain.
4535 len = be32_to_cpu(ohead->oh_len);
4536 if (dp + len > end) {
4537 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4542 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4544 /* nothing to do, so skip over this ophdr */
4549 * The recovered buffer queue is drained only once we know that all
4550 * recovery items for the current LSN have been processed. This is
4553 * - Buffer write submission updates the metadata LSN of the buffer.
4554 * - Log recovery skips items with a metadata LSN >= the current LSN of
4555 * the recovery item.
4556 * - Separate recovery items against the same metadata buffer can share
4557 * a current LSN. I.e., consider that the LSN of a recovery item is
4558 * defined as the starting LSN of the first record in which its
4559 * transaction appears, that a record can hold multiple transactions,
4560 * and/or that a transaction can span multiple records.
4562 * In other words, we are allowed to submit a buffer from log recovery
4563 * once per current LSN. Otherwise, we may incorrectly skip recovery
4564 * items and cause corruption.
4566 * We don't know up front whether buffers are updated multiple times per
4567 * LSN. Therefore, track the current LSN of each commit log record as it
4568 * is processed and drain the queue when it changes. Use commit records
4569 * because they are ordered correctly by the logging code.
4571 if (log->l_recovery_lsn != trans->r_lsn &&
4572 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4573 error = xfs_buf_delwri_submit(buffer_list);
4576 log->l_recovery_lsn = trans->r_lsn;
4579 return xlog_recovery_process_trans(log, trans, dp, len,
4580 ohead->oh_flags, pass, buffer_list);
4584 * There are two valid states of the r_state field. 0 indicates that the
4585 * transaction structure is in a normal state. We have either seen the
4586 * start of the transaction or the last operation we added was not a partial
4587 * operation. If the last operation we added to the transaction was a
4588 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4590 * NOTE: skip LRs with 0 data length.
4593 xlog_recover_process_data(
4595 struct hlist_head rhash[],
4596 struct xlog_rec_header *rhead,
4599 struct list_head *buffer_list)
4601 struct xlog_op_header *ohead;
4606 end = dp + be32_to_cpu(rhead->h_len);
4607 num_logops = be32_to_cpu(rhead->h_num_logops);
4609 /* check the log format matches our own - else we can't recover */
4610 if (xlog_header_check_recover(log->l_mp, rhead))
4613 trace_xfs_log_recover_record(log, rhead, pass);
4614 while ((dp < end) && num_logops) {
4616 ohead = (struct xlog_op_header *)dp;
4617 dp += sizeof(*ohead);
4620 /* errors will abort recovery */
4621 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4622 dp, end, pass, buffer_list);
4626 dp += be32_to_cpu(ohead->oh_len);
4632 /* Recover the EFI if necessary. */
4634 xlog_recover_process_efi(
4635 struct xfs_mount *mp,
4636 struct xfs_ail *ailp,
4637 struct xfs_log_item *lip)
4639 struct xfs_efi_log_item *efip;
4643 * Skip EFIs that we've already processed.
4645 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4646 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4649 spin_unlock(&ailp->xa_lock);
4650 error = xfs_efi_recover(mp, efip);
4651 spin_lock(&ailp->xa_lock);
4656 /* Release the EFI since we're cancelling everything. */
4658 xlog_recover_cancel_efi(
4659 struct xfs_mount *mp,
4660 struct xfs_ail *ailp,
4661 struct xfs_log_item *lip)
4663 struct xfs_efi_log_item *efip;
4665 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4667 spin_unlock(&ailp->xa_lock);
4668 xfs_efi_release(efip);
4669 spin_lock(&ailp->xa_lock);
4672 /* Recover the RUI if necessary. */
4674 xlog_recover_process_rui(
4675 struct xfs_mount *mp,
4676 struct xfs_ail *ailp,
4677 struct xfs_log_item *lip)
4679 struct xfs_rui_log_item *ruip;
4683 * Skip RUIs that we've already processed.
4685 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4686 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4689 spin_unlock(&ailp->xa_lock);
4690 error = xfs_rui_recover(mp, ruip);
4691 spin_lock(&ailp->xa_lock);
4696 /* Release the RUI since we're cancelling everything. */
4698 xlog_recover_cancel_rui(
4699 struct xfs_mount *mp,
4700 struct xfs_ail *ailp,
4701 struct xfs_log_item *lip)
4703 struct xfs_rui_log_item *ruip;
4705 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4707 spin_unlock(&ailp->xa_lock);
4708 xfs_rui_release(ruip);
4709 spin_lock(&ailp->xa_lock);
4712 /* Recover the CUI if necessary. */
4714 xlog_recover_process_cui(
4715 struct xfs_mount *mp,
4716 struct xfs_ail *ailp,
4717 struct xfs_log_item *lip)
4719 struct xfs_cui_log_item *cuip;
4723 * Skip CUIs that we've already processed.
4725 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4726 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4729 spin_unlock(&ailp->xa_lock);
4730 error = xfs_cui_recover(mp, cuip);
4731 spin_lock(&ailp->xa_lock);
4736 /* Release the CUI since we're cancelling everything. */
4738 xlog_recover_cancel_cui(
4739 struct xfs_mount *mp,
4740 struct xfs_ail *ailp,
4741 struct xfs_log_item *lip)
4743 struct xfs_cui_log_item *cuip;
4745 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4747 spin_unlock(&ailp->xa_lock);
4748 xfs_cui_release(cuip);
4749 spin_lock(&ailp->xa_lock);
4752 /* Recover the BUI if necessary. */
4754 xlog_recover_process_bui(
4755 struct xfs_mount *mp,
4756 struct xfs_ail *ailp,
4757 struct xfs_log_item *lip)
4759 struct xfs_bui_log_item *buip;
4763 * Skip BUIs that we've already processed.
4765 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4766 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4769 spin_unlock(&ailp->xa_lock);
4770 error = xfs_bui_recover(mp, buip);
4771 spin_lock(&ailp->xa_lock);
4776 /* Release the BUI since we're cancelling everything. */
4778 xlog_recover_cancel_bui(
4779 struct xfs_mount *mp,
4780 struct xfs_ail *ailp,
4781 struct xfs_log_item *lip)
4783 struct xfs_bui_log_item *buip;
4785 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4787 spin_unlock(&ailp->xa_lock);
4788 xfs_bui_release(buip);
4789 spin_lock(&ailp->xa_lock);
4792 /* Is this log item a deferred action intent? */
4793 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4795 switch (lip->li_type) {
4807 * When this is called, all of the log intent items which did not have
4808 * corresponding log done items should be in the AIL. What we do now
4809 * is update the data structures associated with each one.
4811 * Since we process the log intent items in normal transactions, they
4812 * will be removed at some point after the commit. This prevents us
4813 * from just walking down the list processing each one. We'll use a
4814 * flag in the intent item to skip those that we've already processed
4815 * and use the AIL iteration mechanism's generation count to try to
4816 * speed this up at least a bit.
4818 * When we start, we know that the intents are the only things in the
4819 * AIL. As we process them, however, other items are added to the
4823 xlog_recover_process_intents(
4826 struct xfs_log_item *lip;
4828 struct xfs_ail_cursor cur;
4829 struct xfs_ail *ailp;
4830 #if defined(DEBUG) || defined(XFS_WARN)
4835 spin_lock(&ailp->xa_lock);
4836 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4837 #if defined(DEBUG) || defined(XFS_WARN)
4838 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4840 while (lip != NULL) {
4842 * We're done when we see something other than an intent.
4843 * There should be no intents left in the AIL now.
4845 if (!xlog_item_is_intent(lip)) {
4847 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4848 ASSERT(!xlog_item_is_intent(lip));
4854 * We should never see a redo item with a LSN higher than
4855 * the last transaction we found in the log at the start
4858 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4860 switch (lip->li_type) {
4862 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4865 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4868 error = xlog_recover_process_cui(log->l_mp, ailp, lip);
4871 error = xlog_recover_process_bui(log->l_mp, ailp, lip);
4876 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4879 xfs_trans_ail_cursor_done(&cur);
4880 spin_unlock(&ailp->xa_lock);
4885 * A cancel occurs when the mount has failed and we're bailing out.
4886 * Release all pending log intent items so they don't pin the AIL.
4889 xlog_recover_cancel_intents(
4892 struct xfs_log_item *lip;
4894 struct xfs_ail_cursor cur;
4895 struct xfs_ail *ailp;
4898 spin_lock(&ailp->xa_lock);
4899 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4900 while (lip != NULL) {
4902 * We're done when we see something other than an intent.
4903 * There should be no intents left in the AIL now.
4905 if (!xlog_item_is_intent(lip)) {
4907 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4908 ASSERT(!xlog_item_is_intent(lip));
4913 switch (lip->li_type) {
4915 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4918 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4921 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4924 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4928 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4931 xfs_trans_ail_cursor_done(&cur);
4932 spin_unlock(&ailp->xa_lock);
4937 * This routine performs a transaction to null out a bad inode pointer
4938 * in an agi unlinked inode hash bucket.
4941 xlog_recover_clear_agi_bucket(
4943 xfs_agnumber_t agno,
4952 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
4956 error = xfs_read_agi(mp, tp, agno, &agibp);
4960 agi = XFS_BUF_TO_AGI(agibp);
4961 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
4962 offset = offsetof(xfs_agi_t, agi_unlinked) +
4963 (sizeof(xfs_agino_t) * bucket);
4964 xfs_trans_log_buf(tp, agibp, offset,
4965 (offset + sizeof(xfs_agino_t) - 1));
4967 error = xfs_trans_commit(tp);
4973 xfs_trans_cancel(tp);
4975 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
4980 xlog_recover_process_one_iunlink(
4981 struct xfs_mount *mp,
4982 xfs_agnumber_t agno,
4986 struct xfs_buf *ibp;
4987 struct xfs_dinode *dip;
4988 struct xfs_inode *ip;
4992 ino = XFS_AGINO_TO_INO(mp, agno, agino);
4993 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
4998 * Get the on disk inode to find the next inode in the bucket.
5000 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
5004 xfs_iflags_clear(ip, XFS_IRECOVERY);
5005 ASSERT(VFS_I(ip)->i_nlink == 0);
5006 ASSERT(VFS_I(ip)->i_mode != 0);
5008 /* setup for the next pass */
5009 agino = be32_to_cpu(dip->di_next_unlinked);
5013 * Prevent any DMAPI event from being sent when the reference on
5014 * the inode is dropped.
5016 ip->i_d.di_dmevmask = 0;
5025 * We can't read in the inode this bucket points to, or this inode
5026 * is messed up. Just ditch this bucket of inodes. We will lose
5027 * some inodes and space, but at least we won't hang.
5029 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5030 * clear the inode pointer in the bucket.
5032 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5037 * xlog_iunlink_recover
5039 * This is called during recovery to process any inodes which
5040 * we unlinked but not freed when the system crashed. These
5041 * inodes will be on the lists in the AGI blocks. What we do
5042 * here is scan all the AGIs and fully truncate and free any
5043 * inodes found on the lists. Each inode is removed from the
5044 * lists when it has been fully truncated and is freed. The
5045 * freeing of the inode and its removal from the list must be
5049 xlog_recover_process_iunlinks(
5053 xfs_agnumber_t agno;
5064 * Prevent any DMAPI event from being sent while in this function.
5066 mp_dmevmask = mp->m_dmevmask;
5069 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5071 * Find the agi for this ag.
5073 error = xfs_read_agi(mp, NULL, agno, &agibp);
5076 * AGI is b0rked. Don't process it.
5078 * We should probably mark the filesystem as corrupt
5079 * after we've recovered all the ag's we can....
5084 * Unlock the buffer so that it can be acquired in the normal
5085 * course of the transaction to truncate and free each inode.
5086 * Because we are not racing with anyone else here for the AGI
5087 * buffer, we don't even need to hold it locked to read the
5088 * initial unlinked bucket entries out of the buffer. We keep
5089 * buffer reference though, so that it stays pinned in memory
5090 * while we need the buffer.
5092 agi = XFS_BUF_TO_AGI(agibp);
5093 xfs_buf_unlock(agibp);
5095 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5096 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5097 while (agino != NULLAGINO) {
5098 agino = xlog_recover_process_one_iunlink(mp,
5099 agno, agino, bucket);
5102 xfs_buf_rele(agibp);
5105 mp->m_dmevmask = mp_dmevmask;
5110 struct xlog_rec_header *rhead,
5116 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5117 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5118 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5122 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5123 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5124 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5125 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5126 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5127 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5136 * CRC check, unpack and process a log record.
5139 xlog_recover_process(
5141 struct hlist_head rhash[],
5142 struct xlog_rec_header *rhead,
5145 struct list_head *buffer_list)
5150 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5153 * Nothing else to do if this is a CRC verification pass. Just return
5154 * if this a record with a non-zero crc. Unfortunately, mkfs always
5155 * sets h_crc to 0 so we must consider this valid even on v5 supers.
5156 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5157 * know precisely what failed.
5159 if (pass == XLOG_RECOVER_CRCPASS) {
5160 if (rhead->h_crc && crc != rhead->h_crc)
5166 * We're in the normal recovery path. Issue a warning if and only if the
5167 * CRC in the header is non-zero. This is an advisory warning and the
5168 * zero CRC check prevents warnings from being emitted when upgrading
5169 * the kernel from one that does not add CRCs by default.
5171 if (crc != rhead->h_crc) {
5172 if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5173 xfs_alert(log->l_mp,
5174 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5175 le32_to_cpu(rhead->h_crc),
5177 xfs_hex_dump(dp, 32);
5181 * If the filesystem is CRC enabled, this mismatch becomes a
5182 * fatal log corruption failure.
5184 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
5185 return -EFSCORRUPTED;
5188 error = xlog_unpack_data(rhead, dp, log);
5192 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5197 xlog_valid_rec_header(
5199 struct xlog_rec_header *rhead,
5204 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
5205 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5206 XFS_ERRLEVEL_LOW, log->l_mp);
5207 return -EFSCORRUPTED;
5210 (!rhead->h_version ||
5211 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
5212 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5213 __func__, be32_to_cpu(rhead->h_version));
5217 /* LR body must have data or it wouldn't have been written */
5218 hlen = be32_to_cpu(rhead->h_len);
5219 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
5220 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5221 XFS_ERRLEVEL_LOW, log->l_mp);
5222 return -EFSCORRUPTED;
5224 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
5225 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5226 XFS_ERRLEVEL_LOW, log->l_mp);
5227 return -EFSCORRUPTED;
5233 * Read the log from tail to head and process the log records found.
5234 * Handle the two cases where the tail and head are in the same cycle
5235 * and where the active portion of the log wraps around the end of
5236 * the physical log separately. The pass parameter is passed through
5237 * to the routines called to process the data and is not looked at
5241 xlog_do_recovery_pass(
5243 xfs_daddr_t head_blk,
5244 xfs_daddr_t tail_blk,
5246 xfs_daddr_t *first_bad) /* out: first bad log rec */
5248 xlog_rec_header_t *rhead;
5249 xfs_daddr_t blk_no, rblk_no;
5250 xfs_daddr_t rhead_blk;
5252 xfs_buf_t *hbp, *dbp;
5253 int error = 0, h_size, h_len;
5255 int bblks, split_bblks;
5256 int hblks, split_hblks, wrapped_hblks;
5258 struct hlist_head rhash[XLOG_RHASH_SIZE];
5259 LIST_HEAD (buffer_list);
5261 ASSERT(head_blk != tail_blk);
5262 blk_no = rhead_blk = tail_blk;
5264 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5265 INIT_HLIST_HEAD(&rhash[i]);
5268 * Read the header of the tail block and get the iclog buffer size from
5269 * h_size. Use this to tell how many sectors make up the log header.
5271 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5273 * When using variable length iclogs, read first sector of
5274 * iclog header and extract the header size from it. Get a
5275 * new hbp that is the correct size.
5277 hbp = xlog_get_bp(log, 1);
5281 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5285 rhead = (xlog_rec_header_t *)offset;
5286 error = xlog_valid_rec_header(log, rhead, tail_blk);
5291 * xfsprogs has a bug where record length is based on lsunit but
5292 * h_size (iclog size) is hardcoded to 32k. Now that we
5293 * unconditionally CRC verify the unmount record, this means the
5294 * log buffer can be too small for the record and cause an
5297 * Detect this condition here. Use lsunit for the buffer size as
5298 * long as this looks like the mkfs case. Otherwise, return an
5299 * error to avoid a buffer overrun.
5301 h_size = be32_to_cpu(rhead->h_size);
5302 h_len = be32_to_cpu(rhead->h_len);
5303 if (h_len > h_size) {
5304 if (h_len <= log->l_mp->m_logbsize &&
5305 be32_to_cpu(rhead->h_num_logops) == 1) {
5307 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5308 h_size, log->l_mp->m_logbsize);
5309 h_size = log->l_mp->m_logbsize;
5311 return -EFSCORRUPTED;
5314 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5315 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5316 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5317 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5320 hbp = xlog_get_bp(log, hblks);
5325 ASSERT(log->l_sectBBsize == 1);
5327 hbp = xlog_get_bp(log, 1);
5328 h_size = XLOG_BIG_RECORD_BSIZE;
5333 dbp = xlog_get_bp(log, BTOBB(h_size));
5339 memset(rhash, 0, sizeof(rhash));
5340 if (tail_blk > head_blk) {
5342 * Perform recovery around the end of the physical log.
5343 * When the head is not on the same cycle number as the tail,
5344 * we can't do a sequential recovery.
5346 while (blk_no < log->l_logBBsize) {
5348 * Check for header wrapping around physical end-of-log
5350 offset = hbp->b_addr;
5353 if (blk_no + hblks <= log->l_logBBsize) {
5354 /* Read header in one read */
5355 error = xlog_bread(log, blk_no, hblks, hbp,
5360 /* This LR is split across physical log end */
5361 if (blk_no != log->l_logBBsize) {
5362 /* some data before physical log end */
5363 ASSERT(blk_no <= INT_MAX);
5364 split_hblks = log->l_logBBsize - (int)blk_no;
5365 ASSERT(split_hblks > 0);
5366 error = xlog_bread(log, blk_no,
5374 * Note: this black magic still works with
5375 * large sector sizes (non-512) only because:
5376 * - we increased the buffer size originally
5377 * by 1 sector giving us enough extra space
5378 * for the second read;
5379 * - the log start is guaranteed to be sector
5381 * - we read the log end (LR header start)
5382 * _first_, then the log start (LR header end)
5383 * - order is important.
5385 wrapped_hblks = hblks - split_hblks;
5386 error = xlog_bread_offset(log, 0,
5388 offset + BBTOB(split_hblks));
5392 rhead = (xlog_rec_header_t *)offset;
5393 error = xlog_valid_rec_header(log, rhead,
5394 split_hblks ? blk_no : 0);
5398 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5402 * Read the log record data in multiple reads if it
5403 * wraps around the end of the log. Note that if the
5404 * header already wrapped, blk_no could point past the
5405 * end of the log. The record data is contiguous in
5408 if (blk_no + bblks <= log->l_logBBsize ||
5409 blk_no >= log->l_logBBsize) {
5410 /* mod blk_no in case the header wrapped and
5411 * pushed it beyond the end of the log */
5412 rblk_no = do_mod(blk_no, log->l_logBBsize);
5413 error = xlog_bread(log, rblk_no, bblks, dbp,
5418 /* This log record is split across the
5419 * physical end of log */
5420 offset = dbp->b_addr;
5422 if (blk_no != log->l_logBBsize) {
5423 /* some data is before the physical
5425 ASSERT(!wrapped_hblks);
5426 ASSERT(blk_no <= INT_MAX);
5428 log->l_logBBsize - (int)blk_no;
5429 ASSERT(split_bblks > 0);
5430 error = xlog_bread(log, blk_no,
5438 * Note: this black magic still works with
5439 * large sector sizes (non-512) only because:
5440 * - we increased the buffer size originally
5441 * by 1 sector giving us enough extra space
5442 * for the second read;
5443 * - the log start is guaranteed to be sector
5445 * - we read the log end (LR header start)
5446 * _first_, then the log start (LR header end)
5447 * - order is important.
5449 error = xlog_bread_offset(log, 0,
5450 bblks - split_bblks, dbp,
5451 offset + BBTOB(split_bblks));
5456 error = xlog_recover_process(log, rhash, rhead, offset,
5457 pass, &buffer_list);
5465 ASSERT(blk_no >= log->l_logBBsize);
5466 blk_no -= log->l_logBBsize;
5470 /* read first part of physical log */
5471 while (blk_no < head_blk) {
5472 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5476 rhead = (xlog_rec_header_t *)offset;
5477 error = xlog_valid_rec_header(log, rhead, blk_no);
5481 /* blocks in data section */
5482 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5483 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5488 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5493 blk_no += bblks + hblks;
5503 * Submit buffers that have been added from the last record processed,
5504 * regardless of error status.
5506 if (!list_empty(&buffer_list))
5507 error2 = xfs_buf_delwri_submit(&buffer_list);
5509 if (error && first_bad)
5510 *first_bad = rhead_blk;
5513 * Transactions are freed at commit time but transactions without commit
5514 * records on disk are never committed. Free any that may be left in the
5517 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5518 struct hlist_node *tmp;
5519 struct xlog_recover *trans;
5521 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5522 xlog_recover_free_trans(trans);
5525 return error ? error : error2;
5529 * Do the recovery of the log. We actually do this in two phases.
5530 * The two passes are necessary in order to implement the function
5531 * of cancelling a record written into the log. The first pass
5532 * determines those things which have been cancelled, and the
5533 * second pass replays log items normally except for those which
5534 * have been cancelled. The handling of the replay and cancellations
5535 * takes place in the log item type specific routines.
5537 * The table of items which have cancel records in the log is allocated
5538 * and freed at this level, since only here do we know when all of
5539 * the log recovery has been completed.
5542 xlog_do_log_recovery(
5544 xfs_daddr_t head_blk,
5545 xfs_daddr_t tail_blk)
5549 ASSERT(head_blk != tail_blk);
5552 * First do a pass to find all of the cancelled buf log items.
5553 * Store them in the buf_cancel_table for use in the second pass.
5555 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5556 sizeof(struct list_head),
5558 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5559 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5561 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5562 XLOG_RECOVER_PASS1, NULL);
5564 kmem_free(log->l_buf_cancel_table);
5565 log->l_buf_cancel_table = NULL;
5569 * Then do a second pass to actually recover the items in the log.
5570 * When it is complete free the table of buf cancel items.
5572 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5573 XLOG_RECOVER_PASS2, NULL);
5578 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5579 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5583 kmem_free(log->l_buf_cancel_table);
5584 log->l_buf_cancel_table = NULL;
5590 * Do the actual recovery
5595 xfs_daddr_t head_blk,
5596 xfs_daddr_t tail_blk)
5598 struct xfs_mount *mp = log->l_mp;
5603 trace_xfs_log_recover(log, head_blk, tail_blk);
5606 * First replay the images in the log.
5608 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5613 * If IO errors happened during recovery, bail out.
5615 if (XFS_FORCED_SHUTDOWN(mp)) {
5620 * We now update the tail_lsn since much of the recovery has completed
5621 * and there may be space available to use. If there were no extent
5622 * or iunlinks, we can free up the entire log and set the tail_lsn to
5623 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5624 * lsn of the last known good LR on disk. If there are extent frees
5625 * or iunlinks they will have some entries in the AIL; so we look at
5626 * the AIL to determine how to set the tail_lsn.
5628 xlog_assign_tail_lsn(mp);
5631 * Now that we've finished replaying all buffer and inode
5632 * updates, re-read in the superblock and reverify it.
5634 bp = xfs_getsb(mp, 0);
5635 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5636 ASSERT(!(bp->b_flags & XBF_WRITE));
5637 bp->b_flags |= XBF_READ;
5638 bp->b_ops = &xfs_sb_buf_ops;
5640 error = xfs_buf_submit_wait(bp);
5642 if (!XFS_FORCED_SHUTDOWN(mp)) {
5643 xfs_buf_ioerror_alert(bp, __func__);
5650 /* Convert superblock from on-disk format */
5652 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5655 /* re-initialise in-core superblock and geometry structures */
5656 xfs_reinit_percpu_counters(mp);
5657 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5659 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5662 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5664 xlog_recover_check_summary(log);
5666 /* Normal transactions can now occur */
5667 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5672 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5674 * Return error or zero.
5680 xfs_daddr_t head_blk, tail_blk;
5683 /* find the tail of the log */
5684 error = xlog_find_tail(log, &head_blk, &tail_blk);
5689 * The superblock was read before the log was available and thus the LSN
5690 * could not be verified. Check the superblock LSN against the current
5691 * LSN now that it's known.
5693 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5694 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5697 if (tail_blk != head_blk) {
5698 /* There used to be a comment here:
5700 * disallow recovery on read-only mounts. note -- mount
5701 * checks for ENOSPC and turns it into an intelligent
5703 * ...but this is no longer true. Now, unless you specify
5704 * NORECOVERY (in which case this function would never be
5705 * called), we just go ahead and recover. We do this all
5706 * under the vfs layer, so we can get away with it unless
5707 * the device itself is read-only, in which case we fail.
5709 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5714 * Version 5 superblock log feature mask validation. We know the
5715 * log is dirty so check if there are any unknown log features
5716 * in what we need to recover. If there are unknown features
5717 * (e.g. unsupported transactions, then simply reject the
5718 * attempt at recovery before touching anything.
5720 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5721 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5722 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5724 "Superblock has unknown incompatible log features (0x%x) enabled.",
5725 (log->l_mp->m_sb.sb_features_log_incompat &
5726 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5728 "The log can not be fully and/or safely recovered by this kernel.");
5730 "Please recover the log on a kernel that supports the unknown features.");
5735 * Delay log recovery if the debug hook is set. This is debug
5736 * instrumention to coordinate simulation of I/O failures with
5739 if (xfs_globals.log_recovery_delay) {
5740 xfs_notice(log->l_mp,
5741 "Delaying log recovery for %d seconds.",
5742 xfs_globals.log_recovery_delay);
5743 msleep(xfs_globals.log_recovery_delay * 1000);
5746 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5747 log->l_mp->m_logname ? log->l_mp->m_logname
5750 error = xlog_do_recover(log, head_blk, tail_blk);
5751 log->l_flags |= XLOG_RECOVERY_NEEDED;
5757 * In the first part of recovery we replay inodes and buffers and build
5758 * up the list of extent free items which need to be processed. Here
5759 * we process the extent free items and clean up the on disk unlinked
5760 * inode lists. This is separated from the first part of recovery so
5761 * that the root and real-time bitmap inodes can be read in from disk in
5762 * between the two stages. This is necessary so that we can free space
5763 * in the real-time portion of the file system.
5766 xlog_recover_finish(
5770 * Now we're ready to do the transactions needed for the
5771 * rest of recovery. Start with completing all the extent
5772 * free intent records and then process the unlinked inode
5773 * lists. At this point, we essentially run in normal mode
5774 * except that we're still performing recovery actions
5775 * rather than accepting new requests.
5777 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5779 error = xlog_recover_process_intents(log);
5781 xfs_alert(log->l_mp, "Failed to recover intents");
5786 * Sync the log to get all the intents out of the AIL.
5787 * This isn't absolutely necessary, but it helps in
5788 * case the unlink transactions would have problems
5789 * pushing the intents out of the way.
5791 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5793 xlog_recover_process_iunlinks(log);
5795 xlog_recover_check_summary(log);
5797 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5798 log->l_mp->m_logname ? log->l_mp->m_logname
5800 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5802 xfs_info(log->l_mp, "Ending clean mount");
5808 xlog_recover_cancel(
5813 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5814 error = xlog_recover_cancel_intents(log);
5821 * Read all of the agf and agi counters and check that they
5822 * are consistent with the superblock counters.
5825 xlog_recover_check_summary(
5832 xfs_agnumber_t agno;
5833 __uint64_t freeblks;
5843 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5844 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5846 xfs_alert(mp, "%s agf read failed agno %d error %d",
5847 __func__, agno, error);
5849 agfp = XFS_BUF_TO_AGF(agfbp);
5850 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5851 be32_to_cpu(agfp->agf_flcount);
5852 xfs_buf_relse(agfbp);
5855 error = xfs_read_agi(mp, NULL, agno, &agibp);
5857 xfs_alert(mp, "%s agi read failed agno %d error %d",
5858 __func__, agno, error);
5860 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5862 itotal += be32_to_cpu(agi->agi_count);
5863 ifree += be32_to_cpu(agi->agi_freecount);
5864 xfs_buf_relse(agibp);