1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_trans.h"
21 #include "xfs_log_priv.h"
22 #include "xfs_log_recover.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_alloc.h"
27 #include "xfs_ialloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_cksum.h"
30 #include "xfs_trace.h"
31 #include "xfs_icache.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_error.h"
35 #include "xfs_rmap_item.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_refcount_item.h"
38 #include "xfs_bmap_item.h"
40 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
47 xlog_clear_stale_blocks(
52 xlog_recover_check_summary(
55 #define xlog_recover_check_summary(log)
58 xlog_do_recovery_pass(
59 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
62 * This structure is used during recovery to record the buf log items which
63 * have been canceled and should not be replayed.
65 struct xfs_buf_cancel {
69 struct list_head bc_list;
73 * Sector aligned buffer routines for buffer create/read/write/access
77 * Verify the log-relative block number and length in basic blocks are valid for
78 * an operation involving the given XFS log buffer. Returns true if the fields
79 * are valid, false otherwise.
87 if (blk_no < 0 || blk_no >= log->l_logBBsize)
89 if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
95 * Allocate a buffer to hold log data. The buffer needs to be able
96 * to map to a range of nbblks basic blocks at any valid (basic
97 * block) offset within the log.
107 * Pass log block 0 since we don't have an addr yet, buffer will be
110 if (!xlog_verify_bp(log, 0, nbblks)) {
111 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
113 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
118 * We do log I/O in units of log sectors (a power-of-2
119 * multiple of the basic block size), so we round up the
120 * requested size to accommodate the basic blocks required
121 * for complete log sectors.
123 * In addition, the buffer may be used for a non-sector-
124 * aligned block offset, in which case an I/O of the
125 * requested size could extend beyond the end of the
126 * buffer. If the requested size is only 1 basic block it
127 * will never straddle a sector boundary, so this won't be
128 * an issue. Nor will this be a problem if the log I/O is
129 * done in basic blocks (sector size 1). But otherwise we
130 * extend the buffer by one extra log sector to ensure
131 * there's space to accommodate this possibility.
133 if (nbblks > 1 && log->l_sectBBsize > 1)
134 nbblks += log->l_sectBBsize;
135 nbblks = round_up(nbblks, log->l_sectBBsize);
137 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
151 * Return the address of the start of the given block number's data
152 * in a log buffer. The buffer covers a log sector-aligned region.
161 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
163 ASSERT(offset + nbblks <= bp->b_length);
164 return bp->b_addr + BBTOB(offset);
169 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
180 if (!xlog_verify_bp(log, blk_no, nbblks)) {
182 "Invalid log block/length (0x%llx, 0x%x) for buffer",
184 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
185 return -EFSCORRUPTED;
188 blk_no = round_down(blk_no, log->l_sectBBsize);
189 nbblks = round_up(nbblks, log->l_sectBBsize);
192 ASSERT(nbblks <= bp->b_length);
194 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
195 bp->b_flags |= XBF_READ;
196 bp->b_io_length = nbblks;
199 error = xfs_buf_submit(bp);
200 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
201 xfs_buf_ioerror_alert(bp, __func__);
215 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
219 *offset = xlog_align(log, blk_no, nbblks, bp);
224 * Read at an offset into the buffer. Returns with the buffer in it's original
225 * state regardless of the result of the read.
230 xfs_daddr_t blk_no, /* block to read from */
231 int nbblks, /* blocks to read */
235 char *orig_offset = bp->b_addr;
236 int orig_len = BBTOB(bp->b_length);
239 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
243 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
245 /* must reset buffer pointer even on error */
246 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
253 * Write out the buffer at the given block for the given number of blocks.
254 * The buffer is kept locked across the write and is returned locked.
255 * This can only be used for synchronous log writes.
266 if (!xlog_verify_bp(log, blk_no, nbblks)) {
268 "Invalid log block/length (0x%llx, 0x%x) for buffer",
270 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
271 return -EFSCORRUPTED;
274 blk_no = round_down(blk_no, log->l_sectBBsize);
275 nbblks = round_up(nbblks, log->l_sectBBsize);
278 ASSERT(nbblks <= bp->b_length);
280 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
283 bp->b_io_length = nbblks;
286 error = xfs_bwrite(bp);
288 xfs_buf_ioerror_alert(bp, __func__);
295 * dump debug superblock and log record information
298 xlog_header_check_dump(
300 xlog_rec_header_t *head)
302 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
303 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
304 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
305 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
308 #define xlog_header_check_dump(mp, head)
312 * check log record header for recovery
315 xlog_header_check_recover(
317 xlog_rec_header_t *head)
319 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
322 * IRIX doesn't write the h_fmt field and leaves it zeroed
323 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
324 * a dirty log created in IRIX.
326 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
328 "dirty log written in incompatible format - can't recover");
329 xlog_header_check_dump(mp, head);
330 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
331 XFS_ERRLEVEL_HIGH, mp);
332 return -EFSCORRUPTED;
333 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
335 "dirty log entry has mismatched uuid - can't recover");
336 xlog_header_check_dump(mp, head);
337 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
338 XFS_ERRLEVEL_HIGH, mp);
339 return -EFSCORRUPTED;
345 * read the head block of the log and check the header
348 xlog_header_check_mount(
350 xlog_rec_header_t *head)
352 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
354 if (uuid_is_null(&head->h_fs_uuid)) {
356 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
357 * h_fs_uuid is null, we assume this log was last mounted
358 * by IRIX and continue.
360 xfs_warn(mp, "null uuid in log - IRIX style log");
361 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
362 xfs_warn(mp, "log has mismatched uuid - can't recover");
363 xlog_header_check_dump(mp, head);
364 XFS_ERROR_REPORT("xlog_header_check_mount",
365 XFS_ERRLEVEL_HIGH, mp);
366 return -EFSCORRUPTED;
377 * We're not going to bother about retrying
378 * this during recovery. One strike!
380 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
381 xfs_buf_ioerror_alert(bp, __func__);
382 xfs_force_shutdown(bp->b_target->bt_mount,
383 SHUTDOWN_META_IO_ERROR);
388 * On v5 supers, a bli could be attached to update the metadata LSN.
392 xfs_buf_item_relse(bp);
393 ASSERT(bp->b_log_item == NULL);
400 * This routine finds (to an approximation) the first block in the physical
401 * log which contains the given cycle. It uses a binary search algorithm.
402 * Note that the algorithm can not be perfect because the disk will not
403 * necessarily be perfect.
406 xlog_find_cycle_start(
409 xfs_daddr_t first_blk,
410 xfs_daddr_t *last_blk,
420 mid_blk = BLK_AVG(first_blk, end_blk);
421 while (mid_blk != first_blk && mid_blk != end_blk) {
422 error = xlog_bread(log, mid_blk, 1, bp, &offset);
425 mid_cycle = xlog_get_cycle(offset);
426 if (mid_cycle == cycle)
427 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
429 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
430 mid_blk = BLK_AVG(first_blk, end_blk);
432 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
433 (mid_blk == end_blk && mid_blk-1 == first_blk));
441 * Check that a range of blocks does not contain stop_on_cycle_no.
442 * Fill in *new_blk with the block offset where such a block is
443 * found, or with -1 (an invalid block number) if there is no such
444 * block in the range. The scan needs to occur from front to back
445 * and the pointer into the region must be updated since a later
446 * routine will need to perform another test.
449 xlog_find_verify_cycle(
451 xfs_daddr_t start_blk,
453 uint stop_on_cycle_no,
454 xfs_daddr_t *new_blk)
464 * Greedily allocate a buffer big enough to handle the full
465 * range of basic blocks we'll be examining. If that fails,
466 * try a smaller size. We need to be able to read at least
467 * a log sector, or we're out of luck.
469 bufblks = 1 << ffs(nbblks);
470 while (bufblks > log->l_logBBsize)
472 while (!(bp = xlog_get_bp(log, bufblks))) {
474 if (bufblks < log->l_sectBBsize)
478 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
481 bcount = min(bufblks, (start_blk + nbblks - i));
483 error = xlog_bread(log, i, bcount, bp, &buf);
487 for (j = 0; j < bcount; j++) {
488 cycle = xlog_get_cycle(buf);
489 if (cycle == stop_on_cycle_no) {
506 * Potentially backup over partial log record write.
508 * In the typical case, last_blk is the number of the block directly after
509 * a good log record. Therefore, we subtract one to get the block number
510 * of the last block in the given buffer. extra_bblks contains the number
511 * of blocks we would have read on a previous read. This happens when the
512 * last log record is split over the end of the physical log.
514 * extra_bblks is the number of blocks potentially verified on a previous
515 * call to this routine.
518 xlog_find_verify_log_record(
520 xfs_daddr_t start_blk,
521 xfs_daddr_t *last_blk,
527 xlog_rec_header_t *head = NULL;
530 int num_blks = *last_blk - start_blk;
533 ASSERT(start_blk != 0 || *last_blk != start_blk);
535 if (!(bp = xlog_get_bp(log, num_blks))) {
536 if (!(bp = xlog_get_bp(log, 1)))
540 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
543 offset += ((num_blks - 1) << BBSHIFT);
546 for (i = (*last_blk) - 1; i >= 0; i--) {
548 /* valid log record not found */
550 "Log inconsistent (didn't find previous header)");
557 error = xlog_bread(log, i, 1, bp, &offset);
562 head = (xlog_rec_header_t *)offset;
564 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
572 * We hit the beginning of the physical log & still no header. Return
573 * to caller. If caller can handle a return of -1, then this routine
574 * will be called again for the end of the physical log.
582 * We have the final block of the good log (the first block
583 * of the log record _before_ the head. So we check the uuid.
585 if ((error = xlog_header_check_mount(log->l_mp, head)))
589 * We may have found a log record header before we expected one.
590 * last_blk will be the 1st block # with a given cycle #. We may end
591 * up reading an entire log record. In this case, we don't want to
592 * reset last_blk. Only when last_blk points in the middle of a log
593 * record do we update last_blk.
595 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
596 uint h_size = be32_to_cpu(head->h_size);
598 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
599 if (h_size % XLOG_HEADER_CYCLE_SIZE)
605 if (*last_blk - i + extra_bblks !=
606 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
615 * Head is defined to be the point of the log where the next log write
616 * could go. This means that incomplete LR writes at the end are
617 * eliminated when calculating the head. We aren't guaranteed that previous
618 * LR have complete transactions. We only know that a cycle number of
619 * current cycle number -1 won't be present in the log if we start writing
620 * from our current block number.
622 * last_blk contains the block number of the first block with a given
625 * Return: zero if normal, non-zero if error.
630 xfs_daddr_t *return_head_blk)
634 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
636 uint first_half_cycle, last_half_cycle;
638 int error, log_bbnum = log->l_logBBsize;
640 /* Is the end of the log device zeroed? */
641 error = xlog_find_zeroed(log, &first_blk);
643 xfs_warn(log->l_mp, "empty log check failed");
647 *return_head_blk = first_blk;
649 /* Is the whole lot zeroed? */
651 /* Linux XFS shouldn't generate totally zeroed logs -
652 * mkfs etc write a dummy unmount record to a fresh
653 * log so we can store the uuid in there
655 xfs_warn(log->l_mp, "totally zeroed log");
661 first_blk = 0; /* get cycle # of 1st block */
662 bp = xlog_get_bp(log, 1);
666 error = xlog_bread(log, 0, 1, bp, &offset);
670 first_half_cycle = xlog_get_cycle(offset);
672 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
673 error = xlog_bread(log, last_blk, 1, bp, &offset);
677 last_half_cycle = xlog_get_cycle(offset);
678 ASSERT(last_half_cycle != 0);
681 * If the 1st half cycle number is equal to the last half cycle number,
682 * then the entire log is stamped with the same cycle number. In this
683 * case, head_blk can't be set to zero (which makes sense). The below
684 * math doesn't work out properly with head_blk equal to zero. Instead,
685 * we set it to log_bbnum which is an invalid block number, but this
686 * value makes the math correct. If head_blk doesn't changed through
687 * all the tests below, *head_blk is set to zero at the very end rather
688 * than log_bbnum. In a sense, log_bbnum and zero are the same block
689 * in a circular file.
691 if (first_half_cycle == last_half_cycle) {
693 * In this case we believe that the entire log should have
694 * cycle number last_half_cycle. We need to scan backwards
695 * from the end verifying that there are no holes still
696 * containing last_half_cycle - 1. If we find such a hole,
697 * then the start of that hole will be the new head. The
698 * simple case looks like
699 * x | x ... | x - 1 | x
700 * Another case that fits this picture would be
701 * x | x + 1 | x ... | x
702 * In this case the head really is somewhere at the end of the
703 * log, as one of the latest writes at the beginning was
706 * x | x + 1 | x ... | x - 1 | x
707 * This is really the combination of the above two cases, and
708 * the head has to end up at the start of the x-1 hole at the
711 * In the 256k log case, we will read from the beginning to the
712 * end of the log and search for cycle numbers equal to x-1.
713 * We don't worry about the x+1 blocks that we encounter,
714 * because we know that they cannot be the head since the log
717 head_blk = log_bbnum;
718 stop_on_cycle = last_half_cycle - 1;
721 * In this case we want to find the first block with cycle
722 * number matching last_half_cycle. We expect the log to be
724 * x + 1 ... | x ... | x
725 * The first block with cycle number x (last_half_cycle) will
726 * be where the new head belongs. First we do a binary search
727 * for the first occurrence of last_half_cycle. The binary
728 * search may not be totally accurate, so then we scan back
729 * from there looking for occurrences of last_half_cycle before
730 * us. If that backwards scan wraps around the beginning of
731 * the log, then we look for occurrences of last_half_cycle - 1
732 * at the end of the log. The cases we're looking for look
734 * v binary search stopped here
735 * x + 1 ... | x | x + 1 | x ... | x
736 * ^ but we want to locate this spot
738 * <---------> less than scan distance
739 * x + 1 ... | x ... | x - 1 | x
740 * ^ we want to locate this spot
742 stop_on_cycle = last_half_cycle;
743 if ((error = xlog_find_cycle_start(log, bp, first_blk,
744 &head_blk, last_half_cycle)))
749 * Now validate the answer. Scan back some number of maximum possible
750 * blocks and make sure each one has the expected cycle number. The
751 * maximum is determined by the total possible amount of buffering
752 * in the in-core log. The following number can be made tighter if
753 * we actually look at the block size of the filesystem.
755 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
756 if (head_blk >= num_scan_bblks) {
758 * We are guaranteed that the entire check can be performed
761 start_blk = head_blk - num_scan_bblks;
762 if ((error = xlog_find_verify_cycle(log,
763 start_blk, num_scan_bblks,
764 stop_on_cycle, &new_blk)))
768 } else { /* need to read 2 parts of log */
770 * We are going to scan backwards in the log in two parts.
771 * First we scan the physical end of the log. In this part
772 * of the log, we are looking for blocks with cycle number
773 * last_half_cycle - 1.
774 * If we find one, then we know that the log starts there, as
775 * we've found a hole that didn't get written in going around
776 * the end of the physical log. The simple case for this is
777 * x + 1 ... | x ... | x - 1 | x
778 * <---------> less than scan distance
779 * If all of the blocks at the end of the log have cycle number
780 * last_half_cycle, then we check the blocks at the start of
781 * the log looking for occurrences of last_half_cycle. If we
782 * find one, then our current estimate for the location of the
783 * first occurrence of last_half_cycle is wrong and we move
784 * back to the hole we've found. This case looks like
785 * x + 1 ... | x | x + 1 | x ...
786 * ^ binary search stopped here
787 * Another case we need to handle that only occurs in 256k
789 * x + 1 ... | x ... | x+1 | x ...
790 * ^ binary search stops here
791 * In a 256k log, the scan at the end of the log will see the
792 * x + 1 blocks. We need to skip past those since that is
793 * certainly not the head of the log. By searching for
794 * last_half_cycle-1 we accomplish that.
796 ASSERT(head_blk <= INT_MAX &&
797 (xfs_daddr_t) num_scan_bblks >= head_blk);
798 start_blk = log_bbnum - (num_scan_bblks - head_blk);
799 if ((error = xlog_find_verify_cycle(log, start_blk,
800 num_scan_bblks - (int)head_blk,
801 (stop_on_cycle - 1), &new_blk)))
809 * Scan beginning of log now. The last part of the physical
810 * log is good. This scan needs to verify that it doesn't find
811 * the last_half_cycle.
814 ASSERT(head_blk <= INT_MAX);
815 if ((error = xlog_find_verify_cycle(log,
816 start_blk, (int)head_blk,
817 stop_on_cycle, &new_blk)))
825 * Now we need to make sure head_blk is not pointing to a block in
826 * the middle of a log record.
828 num_scan_bblks = XLOG_REC_SHIFT(log);
829 if (head_blk >= num_scan_bblks) {
830 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
832 /* start ptr at last block ptr before head_blk */
833 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
840 ASSERT(head_blk <= INT_MAX);
841 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
845 /* We hit the beginning of the log during our search */
846 start_blk = log_bbnum - (num_scan_bblks - head_blk);
848 ASSERT(start_blk <= INT_MAX &&
849 (xfs_daddr_t) log_bbnum-start_blk >= 0);
850 ASSERT(head_blk <= INT_MAX);
851 error = xlog_find_verify_log_record(log, start_blk,
852 &new_blk, (int)head_blk);
857 if (new_blk != log_bbnum)
864 if (head_blk == log_bbnum)
865 *return_head_blk = 0;
867 *return_head_blk = head_blk;
869 * When returning here, we have a good block number. Bad block
870 * means that during a previous crash, we didn't have a clean break
871 * from cycle number N to cycle number N-1. In this case, we need
872 * to find the first block with cycle number N-1.
880 xfs_warn(log->l_mp, "failed to find log head");
885 * Seek backwards in the log for log record headers.
887 * Given a starting log block, walk backwards until we find the provided number
888 * of records or hit the provided tail block. The return value is the number of
889 * records encountered or a negative error code. The log block and buffer
890 * pointer of the last record seen are returned in rblk and rhead respectively.
893 xlog_rseek_logrec_hdr(
895 xfs_daddr_t head_blk,
896 xfs_daddr_t tail_blk,
900 struct xlog_rec_header **rhead,
912 * Walk backwards from the head block until we hit the tail or the first
915 end_blk = head_blk > tail_blk ? tail_blk : 0;
916 for (i = (int) head_blk - 1; i >= end_blk; i--) {
917 error = xlog_bread(log, i, 1, bp, &offset);
921 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
923 *rhead = (struct xlog_rec_header *) offset;
924 if (++found == count)
930 * If we haven't hit the tail block or the log record header count,
931 * start looking again from the end of the physical log. Note that
932 * callers can pass head == tail if the tail is not yet known.
934 if (tail_blk >= head_blk && found != count) {
935 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
936 error = xlog_bread(log, i, 1, bp, &offset);
940 if (*(__be32 *)offset ==
941 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
944 *rhead = (struct xlog_rec_header *) offset;
945 if (++found == count)
958 * Seek forward in the log for log record headers.
960 * Given head and tail blocks, walk forward from the tail block until we find
961 * the provided number of records or hit the head block. The return value is the
962 * number of records encountered or a negative error code. The log block and
963 * buffer pointer of the last record seen are returned in rblk and rhead
967 xlog_seek_logrec_hdr(
969 xfs_daddr_t head_blk,
970 xfs_daddr_t tail_blk,
974 struct xlog_rec_header **rhead,
986 * Walk forward from the tail block until we hit the head or the last
989 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
990 for (i = (int) tail_blk; i <= end_blk; i++) {
991 error = xlog_bread(log, i, 1, bp, &offset);
995 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
997 *rhead = (struct xlog_rec_header *) offset;
998 if (++found == count)
1004 * If we haven't hit the head block or the log record header count,
1005 * start looking again from the start of the physical log.
1007 if (tail_blk > head_blk && found != count) {
1008 for (i = 0; i < (int) head_blk; i++) {
1009 error = xlog_bread(log, i, 1, bp, &offset);
1013 if (*(__be32 *)offset ==
1014 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
1017 *rhead = (struct xlog_rec_header *) offset;
1018 if (++found == count)
1031 * Calculate distance from head to tail (i.e., unused space in the log).
1036 xfs_daddr_t head_blk,
1037 xfs_daddr_t tail_blk)
1039 if (head_blk < tail_blk)
1040 return tail_blk - head_blk;
1042 return tail_blk + (log->l_logBBsize - head_blk);
1046 * Verify the log tail. This is particularly important when torn or incomplete
1047 * writes have been detected near the front of the log and the head has been
1048 * walked back accordingly.
1050 * We also have to handle the case where the tail was pinned and the head
1051 * blocked behind the tail right before a crash. If the tail had been pushed
1052 * immediately prior to the crash and the subsequent checkpoint was only
1053 * partially written, it's possible it overwrote the last referenced tail in the
1054 * log with garbage. This is not a coherency problem because the tail must have
1055 * been pushed before it can be overwritten, but appears as log corruption to
1056 * recovery because we have no way to know the tail was updated if the
1057 * subsequent checkpoint didn't write successfully.
1059 * Therefore, CRC check the log from tail to head. If a failure occurs and the
1060 * offending record is within max iclog bufs from the head, walk the tail
1061 * forward and retry until a valid tail is found or corruption is detected out
1062 * of the range of a possible overwrite.
1067 xfs_daddr_t head_blk,
1068 xfs_daddr_t *tail_blk,
1071 struct xlog_rec_header *thead;
1073 xfs_daddr_t first_bad;
1076 xfs_daddr_t tmp_tail;
1077 xfs_daddr_t orig_tail = *tail_blk;
1079 bp = xlog_get_bp(log, 1);
1084 * Make sure the tail points to a record (returns positive count on
1087 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, bp,
1088 &tmp_tail, &thead, &wrapped);
1091 if (*tail_blk != tmp_tail)
1092 *tail_blk = tmp_tail;
1095 * Run a CRC check from the tail to the head. We can't just check
1096 * MAX_ICLOGS records past the tail because the tail may point to stale
1097 * blocks cleared during the search for the head/tail. These blocks are
1098 * overwritten with zero-length records and thus record count is not a
1099 * reliable indicator of the iclog state before a crash.
1102 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1103 XLOG_RECOVER_CRCPASS, &first_bad);
1104 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1108 * Is corruption within range of the head? If so, retry from
1109 * the next record. Otherwise return an error.
1111 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1112 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1115 /* skip to the next record; returns positive count on success */
1116 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2, bp,
1117 &tmp_tail, &thead, &wrapped);
1121 *tail_blk = tmp_tail;
1123 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1124 XLOG_RECOVER_CRCPASS, &first_bad);
1127 if (!error && *tail_blk != orig_tail)
1129 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1130 orig_tail, *tail_blk);
1137 * Detect and trim torn writes from the head of the log.
1139 * Storage without sector atomicity guarantees can result in torn writes in the
1140 * log in the event of a crash. Our only means to detect this scenario is via
1141 * CRC verification. While we can't always be certain that CRC verification
1142 * failure is due to a torn write vs. an unrelated corruption, we do know that
1143 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1144 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1145 * the log and treat failures in this range as torn writes as a matter of
1146 * policy. In the event of CRC failure, the head is walked back to the last good
1147 * record in the log and the tail is updated from that record and verified.
1152 xfs_daddr_t *head_blk, /* in/out: unverified head */
1153 xfs_daddr_t *tail_blk, /* out: tail block */
1155 xfs_daddr_t *rhead_blk, /* start blk of last record */
1156 struct xlog_rec_header **rhead, /* ptr to last record */
1157 bool *wrapped) /* last rec. wraps phys. log */
1159 struct xlog_rec_header *tmp_rhead;
1160 struct xfs_buf *tmp_bp;
1161 xfs_daddr_t first_bad;
1162 xfs_daddr_t tmp_rhead_blk;
1168 * Check the head of the log for torn writes. Search backwards from the
1169 * head until we hit the tail or the maximum number of log record I/Os
1170 * that could have been in flight at one time. Use a temporary buffer so
1171 * we don't trash the rhead/bp pointers from the caller.
1173 tmp_bp = xlog_get_bp(log, 1);
1176 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1177 XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk,
1178 &tmp_rhead, &tmp_wrapped);
1179 xlog_put_bp(tmp_bp);
1184 * Now run a CRC verification pass over the records starting at the
1185 * block found above to the current head. If a CRC failure occurs, the
1186 * log block of the first bad record is saved in first_bad.
1188 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1189 XLOG_RECOVER_CRCPASS, &first_bad);
1190 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1192 * We've hit a potential torn write. Reset the error and warn
1197 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1198 first_bad, *head_blk);
1201 * Get the header block and buffer pointer for the last good
1202 * record before the bad record.
1204 * Note that xlog_find_tail() clears the blocks at the new head
1205 * (i.e., the records with invalid CRC) if the cycle number
1206 * matches the the current cycle.
1208 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp,
1209 rhead_blk, rhead, wrapped);
1212 if (found == 0) /* XXX: right thing to do here? */
1216 * Reset the head block to the starting block of the first bad
1217 * log record and set the tail block based on the last good
1220 * Bail out if the updated head/tail match as this indicates
1221 * possible corruption outside of the acceptable
1222 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1224 *head_blk = first_bad;
1225 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1226 if (*head_blk == *tail_blk) {
1234 return xlog_verify_tail(log, *head_blk, tail_blk,
1235 be32_to_cpu((*rhead)->h_size));
1239 * We need to make sure we handle log wrapping properly, so we can't use the
1240 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1243 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1244 * operation here and cast it back to a 64 bit daddr on return.
1246 static inline xfs_daddr_t
1253 div_s64_rem(bno, log->l_logBBsize, &mod);
1258 * Check whether the head of the log points to an unmount record. In other
1259 * words, determine whether the log is clean. If so, update the in-core state
1263 xlog_check_unmount_rec(
1265 xfs_daddr_t *head_blk,
1266 xfs_daddr_t *tail_blk,
1267 struct xlog_rec_header *rhead,
1268 xfs_daddr_t rhead_blk,
1272 struct xlog_op_header *op_head;
1273 xfs_daddr_t umount_data_blk;
1274 xfs_daddr_t after_umount_blk;
1282 * Look for unmount record. If we find it, then we know there was a
1283 * clean unmount. Since 'i' could be the last block in the physical
1284 * log, we convert to a log block before comparing to the head_blk.
1286 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1287 * below. We won't want to clear the unmount record if there is one, so
1288 * we pass the lsn of the unmount record rather than the block after it.
1290 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1291 int h_size = be32_to_cpu(rhead->h_size);
1292 int h_version = be32_to_cpu(rhead->h_version);
1294 if ((h_version & XLOG_VERSION_2) &&
1295 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1296 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1297 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1306 after_umount_blk = xlog_wrap_logbno(log,
1307 rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
1309 if (*head_blk == after_umount_blk &&
1310 be32_to_cpu(rhead->h_num_logops) == 1) {
1311 umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
1312 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1316 op_head = (struct xlog_op_header *)offset;
1317 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1319 * Set tail and last sync so that newly written log
1320 * records will point recovery to after the current
1323 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1324 log->l_curr_cycle, after_umount_blk);
1325 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1326 log->l_curr_cycle, after_umount_blk);
1327 *tail_blk = after_umount_blk;
1339 xfs_daddr_t head_blk,
1340 struct xlog_rec_header *rhead,
1341 xfs_daddr_t rhead_blk,
1345 * Reset log values according to the state of the log when we
1346 * crashed. In the case where head_blk == 0, we bump curr_cycle
1347 * one because the next write starts a new cycle rather than
1348 * continuing the cycle of the last good log record. At this
1349 * point we have guaranteed that all partial log records have been
1350 * accounted for. Therefore, we know that the last good log record
1351 * written was complete and ended exactly on the end boundary
1352 * of the physical log.
1354 log->l_prev_block = rhead_blk;
1355 log->l_curr_block = (int)head_blk;
1356 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1358 log->l_curr_cycle++;
1359 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1360 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1361 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1362 BBTOB(log->l_curr_block));
1363 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1364 BBTOB(log->l_curr_block));
1368 * Find the sync block number or the tail of the log.
1370 * This will be the block number of the last record to have its
1371 * associated buffers synced to disk. Every log record header has
1372 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1373 * to get a sync block number. The only concern is to figure out which
1374 * log record header to believe.
1376 * The following algorithm uses the log record header with the largest
1377 * lsn. The entire log record does not need to be valid. We only care
1378 * that the header is valid.
1380 * We could speed up search by using current head_blk buffer, but it is not
1386 xfs_daddr_t *head_blk,
1387 xfs_daddr_t *tail_blk)
1389 xlog_rec_header_t *rhead;
1390 char *offset = NULL;
1393 xfs_daddr_t rhead_blk;
1395 bool wrapped = false;
1399 * Find previous log record
1401 if ((error = xlog_find_head(log, head_blk)))
1403 ASSERT(*head_blk < INT_MAX);
1405 bp = xlog_get_bp(log, 1);
1408 if (*head_blk == 0) { /* special case */
1409 error = xlog_bread(log, 0, 1, bp, &offset);
1413 if (xlog_get_cycle(offset) == 0) {
1415 /* leave all other log inited values alone */
1421 * Search backwards through the log looking for the log record header
1422 * block. This wraps all the way back around to the head so something is
1423 * seriously wrong if we can't find it.
1425 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp,
1426 &rhead_blk, &rhead, &wrapped);
1430 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1433 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1436 * Set the log state based on the current head record.
1438 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1439 tail_lsn = atomic64_read(&log->l_tail_lsn);
1442 * Look for an unmount record at the head of the log. This sets the log
1443 * state to determine whether recovery is necessary.
1445 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1446 rhead_blk, bp, &clean);
1451 * Verify the log head if the log is not clean (e.g., we have anything
1452 * but an unmount record at the head). This uses CRC verification to
1453 * detect and trim torn writes. If discovered, CRC failures are
1454 * considered torn writes and the log head is trimmed accordingly.
1456 * Note that we can only run CRC verification when the log is dirty
1457 * because there's no guarantee that the log data behind an unmount
1458 * record is compatible with the current architecture.
1461 xfs_daddr_t orig_head = *head_blk;
1463 error = xlog_verify_head(log, head_blk, tail_blk, bp,
1464 &rhead_blk, &rhead, &wrapped);
1468 /* update in-core state again if the head changed */
1469 if (*head_blk != orig_head) {
1470 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1472 tail_lsn = atomic64_read(&log->l_tail_lsn);
1473 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1474 rhead, rhead_blk, bp,
1482 * Note that the unmount was clean. If the unmount was not clean, we
1483 * need to know this to rebuild the superblock counters from the perag
1484 * headers if we have a filesystem using non-persistent counters.
1487 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1490 * Make sure that there are no blocks in front of the head
1491 * with the same cycle number as the head. This can happen
1492 * because we allow multiple outstanding log writes concurrently,
1493 * and the later writes might make it out before earlier ones.
1495 * We use the lsn from before modifying it so that we'll never
1496 * overwrite the unmount record after a clean unmount.
1498 * Do this only if we are going to recover the filesystem
1500 * NOTE: This used to say "if (!readonly)"
1501 * However on Linux, we can & do recover a read-only filesystem.
1502 * We only skip recovery if NORECOVERY is specified on mount,
1503 * in which case we would not be here.
1505 * But... if the -device- itself is readonly, just skip this.
1506 * We can't recover this device anyway, so it won't matter.
1508 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1509 error = xlog_clear_stale_blocks(log, tail_lsn);
1515 xfs_warn(log->l_mp, "failed to locate log tail");
1520 * Is the log zeroed at all?
1522 * The last binary search should be changed to perform an X block read
1523 * once X becomes small enough. You can then search linearly through
1524 * the X blocks. This will cut down on the number of reads we need to do.
1526 * If the log is partially zeroed, this routine will pass back the blkno
1527 * of the first block with cycle number 0. It won't have a complete LR
1531 * 0 => the log is completely written to
1532 * 1 => use *blk_no as the first block of the log
1533 * <0 => error has occurred
1538 xfs_daddr_t *blk_no)
1542 uint first_cycle, last_cycle;
1543 xfs_daddr_t new_blk, last_blk, start_blk;
1544 xfs_daddr_t num_scan_bblks;
1545 int error, log_bbnum = log->l_logBBsize;
1549 /* check totally zeroed log */
1550 bp = xlog_get_bp(log, 1);
1553 error = xlog_bread(log, 0, 1, bp, &offset);
1557 first_cycle = xlog_get_cycle(offset);
1558 if (first_cycle == 0) { /* completely zeroed log */
1564 /* check partially zeroed log */
1565 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1569 last_cycle = xlog_get_cycle(offset);
1570 if (last_cycle != 0) { /* log completely written to */
1575 /* we have a partially zeroed log */
1576 last_blk = log_bbnum-1;
1577 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1581 * Validate the answer. Because there is no way to guarantee that
1582 * the entire log is made up of log records which are the same size,
1583 * we scan over the defined maximum blocks. At this point, the maximum
1584 * is not chosen to mean anything special. XXXmiken
1586 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1587 ASSERT(num_scan_bblks <= INT_MAX);
1589 if (last_blk < num_scan_bblks)
1590 num_scan_bblks = last_blk;
1591 start_blk = last_blk - num_scan_bblks;
1594 * We search for any instances of cycle number 0 that occur before
1595 * our current estimate of the head. What we're trying to detect is
1596 * 1 ... | 0 | 1 | 0...
1597 * ^ binary search ends here
1599 if ((error = xlog_find_verify_cycle(log, start_blk,
1600 (int)num_scan_bblks, 0, &new_blk)))
1606 * Potentially backup over partial log record write. We don't need
1607 * to search the end of the log because we know it is zero.
1609 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1624 * These are simple subroutines used by xlog_clear_stale_blocks() below
1625 * to initialize a buffer full of empty log record headers and write
1626 * them into the log.
1637 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1639 memset(buf, 0, BBSIZE);
1640 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1641 recp->h_cycle = cpu_to_be32(cycle);
1642 recp->h_version = cpu_to_be32(
1643 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1644 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1645 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1646 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1647 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1651 xlog_write_log_records(
1662 int sectbb = log->l_sectBBsize;
1663 int end_block = start_block + blocks;
1669 * Greedily allocate a buffer big enough to handle the full
1670 * range of basic blocks to be written. If that fails, try
1671 * a smaller size. We need to be able to write at least a
1672 * log sector, or we're out of luck.
1674 bufblks = 1 << ffs(blocks);
1675 while (bufblks > log->l_logBBsize)
1677 while (!(bp = xlog_get_bp(log, bufblks))) {
1679 if (bufblks < sectbb)
1683 /* We may need to do a read at the start to fill in part of
1684 * the buffer in the starting sector not covered by the first
1687 balign = round_down(start_block, sectbb);
1688 if (balign != start_block) {
1689 error = xlog_bread_noalign(log, start_block, 1, bp);
1693 j = start_block - balign;
1696 for (i = start_block; i < end_block; i += bufblks) {
1697 int bcount, endcount;
1699 bcount = min(bufblks, end_block - start_block);
1700 endcount = bcount - j;
1702 /* We may need to do a read at the end to fill in part of
1703 * the buffer in the final sector not covered by the write.
1704 * If this is the same sector as the above read, skip it.
1706 ealign = round_down(end_block, sectbb);
1707 if (j == 0 && (start_block + endcount > ealign)) {
1708 offset = bp->b_addr + BBTOB(ealign - start_block);
1709 error = xlog_bread_offset(log, ealign, sectbb,
1716 offset = xlog_align(log, start_block, endcount, bp);
1717 for (; j < endcount; j++) {
1718 xlog_add_record(log, offset, cycle, i+j,
1719 tail_cycle, tail_block);
1722 error = xlog_bwrite(log, start_block, endcount, bp);
1725 start_block += endcount;
1735 * This routine is called to blow away any incomplete log writes out
1736 * in front of the log head. We do this so that we won't become confused
1737 * if we come up, write only a little bit more, and then crash again.
1738 * If we leave the partial log records out there, this situation could
1739 * cause us to think those partial writes are valid blocks since they
1740 * have the current cycle number. We get rid of them by overwriting them
1741 * with empty log records with the old cycle number rather than the
1744 * The tail lsn is passed in rather than taken from
1745 * the log so that we will not write over the unmount record after a
1746 * clean unmount in a 512 block log. Doing so would leave the log without
1747 * any valid log records in it until a new one was written. If we crashed
1748 * during that time we would not be able to recover.
1751 xlog_clear_stale_blocks(
1755 int tail_cycle, head_cycle;
1756 int tail_block, head_block;
1757 int tail_distance, max_distance;
1761 tail_cycle = CYCLE_LSN(tail_lsn);
1762 tail_block = BLOCK_LSN(tail_lsn);
1763 head_cycle = log->l_curr_cycle;
1764 head_block = log->l_curr_block;
1767 * Figure out the distance between the new head of the log
1768 * and the tail. We want to write over any blocks beyond the
1769 * head that we may have written just before the crash, but
1770 * we don't want to overwrite the tail of the log.
1772 if (head_cycle == tail_cycle) {
1774 * The tail is behind the head in the physical log,
1775 * so the distance from the head to the tail is the
1776 * distance from the head to the end of the log plus
1777 * the distance from the beginning of the log to the
1780 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1781 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1782 XFS_ERRLEVEL_LOW, log->l_mp);
1783 return -EFSCORRUPTED;
1785 tail_distance = tail_block + (log->l_logBBsize - head_block);
1788 * The head is behind the tail in the physical log,
1789 * so the distance from the head to the tail is just
1790 * the tail block minus the head block.
1792 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1793 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1794 XFS_ERRLEVEL_LOW, log->l_mp);
1795 return -EFSCORRUPTED;
1797 tail_distance = tail_block - head_block;
1801 * If the head is right up against the tail, we can't clear
1804 if (tail_distance <= 0) {
1805 ASSERT(tail_distance == 0);
1809 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1811 * Take the smaller of the maximum amount of outstanding I/O
1812 * we could have and the distance to the tail to clear out.
1813 * We take the smaller so that we don't overwrite the tail and
1814 * we don't waste all day writing from the head to the tail
1817 max_distance = min(max_distance, tail_distance);
1819 if ((head_block + max_distance) <= log->l_logBBsize) {
1821 * We can stomp all the blocks we need to without
1822 * wrapping around the end of the log. Just do it
1823 * in a single write. Use the cycle number of the
1824 * current cycle minus one so that the log will look like:
1827 error = xlog_write_log_records(log, (head_cycle - 1),
1828 head_block, max_distance, tail_cycle,
1834 * We need to wrap around the end of the physical log in
1835 * order to clear all the blocks. Do it in two separate
1836 * I/Os. The first write should be from the head to the
1837 * end of the physical log, and it should use the current
1838 * cycle number minus one just like above.
1840 distance = log->l_logBBsize - head_block;
1841 error = xlog_write_log_records(log, (head_cycle - 1),
1842 head_block, distance, tail_cycle,
1849 * Now write the blocks at the start of the physical log.
1850 * This writes the remainder of the blocks we want to clear.
1851 * It uses the current cycle number since we're now on the
1852 * same cycle as the head so that we get:
1853 * n ... n ... | n - 1 ...
1854 * ^^^^^ blocks we're writing
1856 distance = max_distance - (log->l_logBBsize - head_block);
1857 error = xlog_write_log_records(log, head_cycle, 0, distance,
1858 tail_cycle, tail_block);
1866 /******************************************************************************
1868 * Log recover routines
1870 ******************************************************************************
1874 * Sort the log items in the transaction.
1876 * The ordering constraints are defined by the inode allocation and unlink
1877 * behaviour. The rules are:
1879 * 1. Every item is only logged once in a given transaction. Hence it
1880 * represents the last logged state of the item. Hence ordering is
1881 * dependent on the order in which operations need to be performed so
1882 * required initial conditions are always met.
1884 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1885 * there's nothing to replay from them so we can simply cull them
1886 * from the transaction. However, we can't do that until after we've
1887 * replayed all the other items because they may be dependent on the
1888 * cancelled buffer and replaying the cancelled buffer can remove it
1889 * form the cancelled buffer table. Hence they have tobe done last.
1891 * 3. Inode allocation buffers must be replayed before inode items that
1892 * read the buffer and replay changes into it. For filesystems using the
1893 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1894 * treated the same as inode allocation buffers as they create and
1895 * initialise the buffers directly.
1897 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1898 * This ensures that inodes are completely flushed to the inode buffer
1899 * in a "free" state before we remove the unlinked inode list pointer.
1901 * Hence the ordering needs to be inode allocation buffers first, inode items
1902 * second, inode unlink buffers third and cancelled buffers last.
1904 * But there's a problem with that - we can't tell an inode allocation buffer
1905 * apart from a regular buffer, so we can't separate them. We can, however,
1906 * tell an inode unlink buffer from the others, and so we can separate them out
1907 * from all the other buffers and move them to last.
1909 * Hence, 4 lists, in order from head to tail:
1910 * - buffer_list for all buffers except cancelled/inode unlink buffers
1911 * - item_list for all non-buffer items
1912 * - inode_buffer_list for inode unlink buffers
1913 * - cancel_list for the cancelled buffers
1915 * Note that we add objects to the tail of the lists so that first-to-last
1916 * ordering is preserved within the lists. Adding objects to the head of the
1917 * list means when we traverse from the head we walk them in last-to-first
1918 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1919 * but for all other items there may be specific ordering that we need to
1923 xlog_recover_reorder_trans(
1925 struct xlog_recover *trans,
1928 xlog_recover_item_t *item, *n;
1930 LIST_HEAD(sort_list);
1931 LIST_HEAD(cancel_list);
1932 LIST_HEAD(buffer_list);
1933 LIST_HEAD(inode_buffer_list);
1934 LIST_HEAD(inode_list);
1936 list_splice_init(&trans->r_itemq, &sort_list);
1937 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1938 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1940 switch (ITEM_TYPE(item)) {
1941 case XFS_LI_ICREATE:
1942 list_move_tail(&item->ri_list, &buffer_list);
1945 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1946 trace_xfs_log_recover_item_reorder_head(log,
1948 list_move(&item->ri_list, &cancel_list);
1951 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1952 list_move(&item->ri_list, &inode_buffer_list);
1955 list_move_tail(&item->ri_list, &buffer_list);
1959 case XFS_LI_QUOTAOFF:
1968 trace_xfs_log_recover_item_reorder_tail(log,
1970 list_move_tail(&item->ri_list, &inode_list);
1974 "%s: unrecognized type of log operation",
1978 * return the remaining items back to the transaction
1979 * item list so they can be freed in caller.
1981 if (!list_empty(&sort_list))
1982 list_splice_init(&sort_list, &trans->r_itemq);
1988 ASSERT(list_empty(&sort_list));
1989 if (!list_empty(&buffer_list))
1990 list_splice(&buffer_list, &trans->r_itemq);
1991 if (!list_empty(&inode_list))
1992 list_splice_tail(&inode_list, &trans->r_itemq);
1993 if (!list_empty(&inode_buffer_list))
1994 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1995 if (!list_empty(&cancel_list))
1996 list_splice_tail(&cancel_list, &trans->r_itemq);
2001 * Build up the table of buf cancel records so that we don't replay
2002 * cancelled data in the second pass. For buffer records that are
2003 * not cancel records, there is nothing to do here so we just return.
2005 * If we get a cancel record which is already in the table, this indicates
2006 * that the buffer was cancelled multiple times. In order to ensure
2007 * that during pass 2 we keep the record in the table until we reach its
2008 * last occurrence in the log, we keep a reference count in the cancel
2009 * record in the table to tell us how many times we expect to see this
2010 * record during the second pass.
2013 xlog_recover_buffer_pass1(
2015 struct xlog_recover_item *item)
2017 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2018 struct list_head *bucket;
2019 struct xfs_buf_cancel *bcp;
2022 * If this isn't a cancel buffer item, then just return.
2024 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
2025 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
2030 * Insert an xfs_buf_cancel record into the hash table of them.
2031 * If there is already an identical record, bump its reference count.
2033 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
2034 list_for_each_entry(bcp, bucket, bc_list) {
2035 if (bcp->bc_blkno == buf_f->blf_blkno &&
2036 bcp->bc_len == buf_f->blf_len) {
2038 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
2043 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
2044 bcp->bc_blkno = buf_f->blf_blkno;
2045 bcp->bc_len = buf_f->blf_len;
2046 bcp->bc_refcount = 1;
2047 list_add_tail(&bcp->bc_list, bucket);
2049 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
2054 * Check to see whether the buffer being recovered has a corresponding
2055 * entry in the buffer cancel record table. If it is, return the cancel
2056 * buffer structure to the caller.
2058 STATIC struct xfs_buf_cancel *
2059 xlog_peek_buffer_cancelled(
2063 unsigned short flags)
2065 struct list_head *bucket;
2066 struct xfs_buf_cancel *bcp;
2068 if (!log->l_buf_cancel_table) {
2069 /* empty table means no cancelled buffers in the log */
2070 ASSERT(!(flags & XFS_BLF_CANCEL));
2074 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
2075 list_for_each_entry(bcp, bucket, bc_list) {
2076 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
2081 * We didn't find a corresponding entry in the table, so return 0 so
2082 * that the buffer is NOT cancelled.
2084 ASSERT(!(flags & XFS_BLF_CANCEL));
2089 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2090 * otherwise return 0. If the buffer is actually a buffer cancel item
2091 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2092 * table and remove it from the table if this is the last reference.
2094 * We remove the cancel record from the table when we encounter its last
2095 * occurrence in the log so that if the same buffer is re-used again after its
2096 * last cancellation we actually replay the changes made at that point.
2099 xlog_check_buffer_cancelled(
2103 unsigned short flags)
2105 struct xfs_buf_cancel *bcp;
2107 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2112 * We've go a match, so return 1 so that the recovery of this buffer
2113 * is cancelled. If this buffer is actually a buffer cancel log
2114 * item, then decrement the refcount on the one in the table and
2115 * remove it if this is the last reference.
2117 if (flags & XFS_BLF_CANCEL) {
2118 if (--bcp->bc_refcount == 0) {
2119 list_del(&bcp->bc_list);
2127 * Perform recovery for a buffer full of inodes. In these buffers, the only
2128 * data which should be recovered is that which corresponds to the
2129 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2130 * data for the inodes is always logged through the inodes themselves rather
2131 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2133 * The only time when buffers full of inodes are fully recovered is when the
2134 * buffer is full of newly allocated inodes. In this case the buffer will
2135 * not be marked as an inode buffer and so will be sent to
2136 * xlog_recover_do_reg_buffer() below during recovery.
2139 xlog_recover_do_inode_buffer(
2140 struct xfs_mount *mp,
2141 xlog_recover_item_t *item,
2143 xfs_buf_log_format_t *buf_f)
2149 int reg_buf_offset = 0;
2150 int reg_buf_bytes = 0;
2151 int next_unlinked_offset;
2153 xfs_agino_t *logged_nextp;
2154 xfs_agino_t *buffer_nextp;
2156 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2159 * Post recovery validation only works properly on CRC enabled
2162 if (xfs_sb_version_hascrc(&mp->m_sb))
2163 bp->b_ops = &xfs_inode_buf_ops;
2165 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
2166 for (i = 0; i < inodes_per_buf; i++) {
2167 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2168 offsetof(xfs_dinode_t, di_next_unlinked);
2170 while (next_unlinked_offset >=
2171 (reg_buf_offset + reg_buf_bytes)) {
2173 * The next di_next_unlinked field is beyond
2174 * the current logged region. Find the next
2175 * logged region that contains or is beyond
2176 * the current di_next_unlinked field.
2179 bit = xfs_next_bit(buf_f->blf_data_map,
2180 buf_f->blf_map_size, bit);
2183 * If there are no more logged regions in the
2184 * buffer, then we're done.
2189 nbits = xfs_contig_bits(buf_f->blf_data_map,
2190 buf_f->blf_map_size, bit);
2192 reg_buf_offset = bit << XFS_BLF_SHIFT;
2193 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2198 * If the current logged region starts after the current
2199 * di_next_unlinked field, then move on to the next
2200 * di_next_unlinked field.
2202 if (next_unlinked_offset < reg_buf_offset)
2205 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2206 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2207 ASSERT((reg_buf_offset + reg_buf_bytes) <=
2208 BBTOB(bp->b_io_length));
2211 * The current logged region contains a copy of the
2212 * current di_next_unlinked field. Extract its value
2213 * and copy it to the buffer copy.
2215 logged_nextp = item->ri_buf[item_index].i_addr +
2216 next_unlinked_offset - reg_buf_offset;
2217 if (unlikely(*logged_nextp == 0)) {
2219 "Bad inode buffer log record (ptr = "PTR_FMT", bp = "PTR_FMT"). "
2220 "Trying to replay bad (0) inode di_next_unlinked field.",
2222 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2223 XFS_ERRLEVEL_LOW, mp);
2224 return -EFSCORRUPTED;
2227 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2228 *buffer_nextp = *logged_nextp;
2231 * If necessary, recalculate the CRC in the on-disk inode. We
2232 * have to leave the inode in a consistent state for whoever
2235 xfs_dinode_calc_crc(mp,
2236 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2244 * V5 filesystems know the age of the buffer on disk being recovered. We can
2245 * have newer objects on disk than we are replaying, and so for these cases we
2246 * don't want to replay the current change as that will make the buffer contents
2247 * temporarily invalid on disk.
2249 * The magic number might not match the buffer type we are going to recover
2250 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2251 * extract the LSN of the existing object in the buffer based on it's current
2252 * magic number. If we don't recognise the magic number in the buffer, then
2253 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2254 * so can recover the buffer.
2256 * Note: we cannot rely solely on magic number matches to determine that the
2257 * buffer has a valid LSN - we also need to verify that it belongs to this
2258 * filesystem, so we need to extract the object's LSN and compare it to that
2259 * which we read from the superblock. If the UUIDs don't match, then we've got a
2260 * stale metadata block from an old filesystem instance that we need to recover
2264 xlog_recover_get_buf_lsn(
2265 struct xfs_mount *mp,
2271 void *blk = bp->b_addr;
2275 /* v4 filesystems always recover immediately */
2276 if (!xfs_sb_version_hascrc(&mp->m_sb))
2277 goto recover_immediately;
2279 magic32 = be32_to_cpu(*(__be32 *)blk);
2281 case XFS_ABTB_CRC_MAGIC:
2282 case XFS_ABTC_CRC_MAGIC:
2283 case XFS_ABTB_MAGIC:
2284 case XFS_ABTC_MAGIC:
2285 case XFS_RMAP_CRC_MAGIC:
2286 case XFS_REFC_CRC_MAGIC:
2287 case XFS_IBT_CRC_MAGIC:
2288 case XFS_IBT_MAGIC: {
2289 struct xfs_btree_block *btb = blk;
2291 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2292 uuid = &btb->bb_u.s.bb_uuid;
2295 case XFS_BMAP_CRC_MAGIC:
2296 case XFS_BMAP_MAGIC: {
2297 struct xfs_btree_block *btb = blk;
2299 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2300 uuid = &btb->bb_u.l.bb_uuid;
2304 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2305 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2307 case XFS_AGFL_MAGIC:
2308 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2309 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2312 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2313 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2315 case XFS_SYMLINK_MAGIC:
2316 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2317 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2319 case XFS_DIR3_BLOCK_MAGIC:
2320 case XFS_DIR3_DATA_MAGIC:
2321 case XFS_DIR3_FREE_MAGIC:
2322 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2323 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2325 case XFS_ATTR3_RMT_MAGIC:
2327 * Remote attr blocks are written synchronously, rather than
2328 * being logged. That means they do not contain a valid LSN
2329 * (i.e. transactionally ordered) in them, and hence any time we
2330 * see a buffer to replay over the top of a remote attribute
2331 * block we should simply do so.
2333 goto recover_immediately;
2336 * superblock uuids are magic. We may or may not have a
2337 * sb_meta_uuid on disk, but it will be set in the in-core
2338 * superblock. We set the uuid pointer for verification
2339 * according to the superblock feature mask to ensure we check
2340 * the relevant UUID in the superblock.
2342 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2343 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2344 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2346 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2352 if (lsn != (xfs_lsn_t)-1) {
2353 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2354 goto recover_immediately;
2358 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2360 case XFS_DIR3_LEAF1_MAGIC:
2361 case XFS_DIR3_LEAFN_MAGIC:
2362 case XFS_DA3_NODE_MAGIC:
2363 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2364 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2370 if (lsn != (xfs_lsn_t)-1) {
2371 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2372 goto recover_immediately;
2377 * We do individual object checks on dquot and inode buffers as they
2378 * have their own individual LSN records. Also, we could have a stale
2379 * buffer here, so we have to at least recognise these buffer types.
2381 * A notd complexity here is inode unlinked list processing - it logs
2382 * the inode directly in the buffer, but we don't know which inodes have
2383 * been modified, and there is no global buffer LSN. Hence we need to
2384 * recover all inode buffer types immediately. This problem will be
2385 * fixed by logical logging of the unlinked list modifications.
2387 magic16 = be16_to_cpu(*(__be16 *)blk);
2389 case XFS_DQUOT_MAGIC:
2390 case XFS_DINODE_MAGIC:
2391 goto recover_immediately;
2396 /* unknown buffer contents, recover immediately */
2398 recover_immediately:
2399 return (xfs_lsn_t)-1;
2404 * Validate the recovered buffer is of the correct type and attach the
2405 * appropriate buffer operations to them for writeback. Magic numbers are in a
2407 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2408 * the first 32 bits of the buffer (most blocks),
2409 * inside a struct xfs_da_blkinfo at the start of the buffer.
2412 xlog_recover_validate_buf_type(
2413 struct xfs_mount *mp,
2415 xfs_buf_log_format_t *buf_f,
2416 xfs_lsn_t current_lsn)
2418 struct xfs_da_blkinfo *info = bp->b_addr;
2422 char *warnmsg = NULL;
2425 * We can only do post recovery validation on items on CRC enabled
2426 * fielsystems as we need to know when the buffer was written to be able
2427 * to determine if we should have replayed the item. If we replay old
2428 * metadata over a newer buffer, then it will enter a temporarily
2429 * inconsistent state resulting in verification failures. Hence for now
2430 * just avoid the verification stage for non-crc filesystems
2432 if (!xfs_sb_version_hascrc(&mp->m_sb))
2435 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2436 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2437 magicda = be16_to_cpu(info->magic);
2438 switch (xfs_blft_from_flags(buf_f)) {
2439 case XFS_BLFT_BTREE_BUF:
2441 case XFS_ABTB_CRC_MAGIC:
2442 case XFS_ABTC_CRC_MAGIC:
2443 case XFS_ABTB_MAGIC:
2444 case XFS_ABTC_MAGIC:
2445 bp->b_ops = &xfs_allocbt_buf_ops;
2447 case XFS_IBT_CRC_MAGIC:
2448 case XFS_FIBT_CRC_MAGIC:
2450 case XFS_FIBT_MAGIC:
2451 bp->b_ops = &xfs_inobt_buf_ops;
2453 case XFS_BMAP_CRC_MAGIC:
2454 case XFS_BMAP_MAGIC:
2455 bp->b_ops = &xfs_bmbt_buf_ops;
2457 case XFS_RMAP_CRC_MAGIC:
2458 bp->b_ops = &xfs_rmapbt_buf_ops;
2460 case XFS_REFC_CRC_MAGIC:
2461 bp->b_ops = &xfs_refcountbt_buf_ops;
2464 warnmsg = "Bad btree block magic!";
2468 case XFS_BLFT_AGF_BUF:
2469 if (magic32 != XFS_AGF_MAGIC) {
2470 warnmsg = "Bad AGF block magic!";
2473 bp->b_ops = &xfs_agf_buf_ops;
2475 case XFS_BLFT_AGFL_BUF:
2476 if (magic32 != XFS_AGFL_MAGIC) {
2477 warnmsg = "Bad AGFL block magic!";
2480 bp->b_ops = &xfs_agfl_buf_ops;
2482 case XFS_BLFT_AGI_BUF:
2483 if (magic32 != XFS_AGI_MAGIC) {
2484 warnmsg = "Bad AGI block magic!";
2487 bp->b_ops = &xfs_agi_buf_ops;
2489 case XFS_BLFT_UDQUOT_BUF:
2490 case XFS_BLFT_PDQUOT_BUF:
2491 case XFS_BLFT_GDQUOT_BUF:
2492 #ifdef CONFIG_XFS_QUOTA
2493 if (magic16 != XFS_DQUOT_MAGIC) {
2494 warnmsg = "Bad DQUOT block magic!";
2497 bp->b_ops = &xfs_dquot_buf_ops;
2500 "Trying to recover dquots without QUOTA support built in!");
2504 case XFS_BLFT_DINO_BUF:
2505 if (magic16 != XFS_DINODE_MAGIC) {
2506 warnmsg = "Bad INODE block magic!";
2509 bp->b_ops = &xfs_inode_buf_ops;
2511 case XFS_BLFT_SYMLINK_BUF:
2512 if (magic32 != XFS_SYMLINK_MAGIC) {
2513 warnmsg = "Bad symlink block magic!";
2516 bp->b_ops = &xfs_symlink_buf_ops;
2518 case XFS_BLFT_DIR_BLOCK_BUF:
2519 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2520 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2521 warnmsg = "Bad dir block magic!";
2524 bp->b_ops = &xfs_dir3_block_buf_ops;
2526 case XFS_BLFT_DIR_DATA_BUF:
2527 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2528 magic32 != XFS_DIR3_DATA_MAGIC) {
2529 warnmsg = "Bad dir data magic!";
2532 bp->b_ops = &xfs_dir3_data_buf_ops;
2534 case XFS_BLFT_DIR_FREE_BUF:
2535 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2536 magic32 != XFS_DIR3_FREE_MAGIC) {
2537 warnmsg = "Bad dir3 free magic!";
2540 bp->b_ops = &xfs_dir3_free_buf_ops;
2542 case XFS_BLFT_DIR_LEAF1_BUF:
2543 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2544 magicda != XFS_DIR3_LEAF1_MAGIC) {
2545 warnmsg = "Bad dir leaf1 magic!";
2548 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2550 case XFS_BLFT_DIR_LEAFN_BUF:
2551 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2552 magicda != XFS_DIR3_LEAFN_MAGIC) {
2553 warnmsg = "Bad dir leafn magic!";
2556 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2558 case XFS_BLFT_DA_NODE_BUF:
2559 if (magicda != XFS_DA_NODE_MAGIC &&
2560 magicda != XFS_DA3_NODE_MAGIC) {
2561 warnmsg = "Bad da node magic!";
2564 bp->b_ops = &xfs_da3_node_buf_ops;
2566 case XFS_BLFT_ATTR_LEAF_BUF:
2567 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2568 magicda != XFS_ATTR3_LEAF_MAGIC) {
2569 warnmsg = "Bad attr leaf magic!";
2572 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2574 case XFS_BLFT_ATTR_RMT_BUF:
2575 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2576 warnmsg = "Bad attr remote magic!";
2579 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2581 case XFS_BLFT_SB_BUF:
2582 if (magic32 != XFS_SB_MAGIC) {
2583 warnmsg = "Bad SB block magic!";
2586 bp->b_ops = &xfs_sb_buf_ops;
2588 #ifdef CONFIG_XFS_RT
2589 case XFS_BLFT_RTBITMAP_BUF:
2590 case XFS_BLFT_RTSUMMARY_BUF:
2591 /* no magic numbers for verification of RT buffers */
2592 bp->b_ops = &xfs_rtbuf_ops;
2594 #endif /* CONFIG_XFS_RT */
2596 xfs_warn(mp, "Unknown buffer type %d!",
2597 xfs_blft_from_flags(buf_f));
2602 * Nothing else to do in the case of a NULL current LSN as this means
2603 * the buffer is more recent than the change in the log and will be
2606 if (current_lsn == NULLCOMMITLSN)
2610 xfs_warn(mp, warnmsg);
2615 * We must update the metadata LSN of the buffer as it is written out to
2616 * ensure that older transactions never replay over this one and corrupt
2617 * the buffer. This can occur if log recovery is interrupted at some
2618 * point after the current transaction completes, at which point a
2619 * subsequent mount starts recovery from the beginning.
2621 * Write verifiers update the metadata LSN from log items attached to
2622 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2623 * the verifier. We'll clean it up in our ->iodone() callback.
2626 struct xfs_buf_log_item *bip;
2628 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2629 bp->b_iodone = xlog_recover_iodone;
2630 xfs_buf_item_init(bp, mp);
2631 bip = bp->b_log_item;
2632 bip->bli_item.li_lsn = current_lsn;
2637 * Perform a 'normal' buffer recovery. Each logged region of the
2638 * buffer should be copied over the corresponding region in the
2639 * given buffer. The bitmap in the buf log format structure indicates
2640 * where to place the logged data.
2643 xlog_recover_do_reg_buffer(
2644 struct xfs_mount *mp,
2645 xlog_recover_item_t *item,
2647 xfs_buf_log_format_t *buf_f,
2648 xfs_lsn_t current_lsn)
2655 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2658 i = 1; /* 0 is the buf format structure */
2660 bit = xfs_next_bit(buf_f->blf_data_map,
2661 buf_f->blf_map_size, bit);
2664 nbits = xfs_contig_bits(buf_f->blf_data_map,
2665 buf_f->blf_map_size, bit);
2667 ASSERT(item->ri_buf[i].i_addr != NULL);
2668 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2669 ASSERT(BBTOB(bp->b_io_length) >=
2670 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2673 * The dirty regions logged in the buffer, even though
2674 * contiguous, may span multiple chunks. This is because the
2675 * dirty region may span a physical page boundary in a buffer
2676 * and hence be split into two separate vectors for writing into
2677 * the log. Hence we need to trim nbits back to the length of
2678 * the current region being copied out of the log.
2680 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2681 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2684 * Do a sanity check if this is a dquot buffer. Just checking
2685 * the first dquot in the buffer should do. XXXThis is
2686 * probably a good thing to do for other buf types also.
2689 if (buf_f->blf_flags &
2690 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2691 if (item->ri_buf[i].i_addr == NULL) {
2693 "XFS: NULL dquot in %s.", __func__);
2696 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2698 "XFS: dquot too small (%d) in %s.",
2699 item->ri_buf[i].i_len, __func__);
2702 fa = xfs_dquot_verify(mp, item->ri_buf[i].i_addr,
2706 "dquot corrupt at %pS trying to replay into block 0x%llx",
2712 memcpy(xfs_buf_offset(bp,
2713 (uint)bit << XFS_BLF_SHIFT), /* dest */
2714 item->ri_buf[i].i_addr, /* source */
2715 nbits<<XFS_BLF_SHIFT); /* length */
2721 /* Shouldn't be any more regions */
2722 ASSERT(i == item->ri_total);
2724 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2728 * Perform a dquot buffer recovery.
2729 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2730 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2731 * Else, treat it as a regular buffer and do recovery.
2733 * Return false if the buffer was tossed and true if we recovered the buffer to
2734 * indicate to the caller if the buffer needs writing.
2737 xlog_recover_do_dquot_buffer(
2738 struct xfs_mount *mp,
2740 struct xlog_recover_item *item,
2742 struct xfs_buf_log_format *buf_f)
2746 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2749 * Filesystems are required to send in quota flags at mount time.
2755 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2756 type |= XFS_DQ_USER;
2757 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2758 type |= XFS_DQ_PROJ;
2759 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2760 type |= XFS_DQ_GROUP;
2762 * This type of quotas was turned off, so ignore this buffer
2764 if (log->l_quotaoffs_flag & type)
2767 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2772 * This routine replays a modification made to a buffer at runtime.
2773 * There are actually two types of buffer, regular and inode, which
2774 * are handled differently. Inode buffers are handled differently
2775 * in that we only recover a specific set of data from them, namely
2776 * the inode di_next_unlinked fields. This is because all other inode
2777 * data is actually logged via inode records and any data we replay
2778 * here which overlaps that may be stale.
2780 * When meta-data buffers are freed at run time we log a buffer item
2781 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2782 * of the buffer in the log should not be replayed at recovery time.
2783 * This is so that if the blocks covered by the buffer are reused for
2784 * file data before we crash we don't end up replaying old, freed
2785 * meta-data into a user's file.
2787 * To handle the cancellation of buffer log items, we make two passes
2788 * over the log during recovery. During the first we build a table of
2789 * those buffers which have been cancelled, and during the second we
2790 * only replay those buffers which do not have corresponding cancel
2791 * records in the table. See xlog_recover_buffer_pass[1,2] above
2792 * for more details on the implementation of the table of cancel records.
2795 xlog_recover_buffer_pass2(
2797 struct list_head *buffer_list,
2798 struct xlog_recover_item *item,
2799 xfs_lsn_t current_lsn)
2801 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2802 xfs_mount_t *mp = log->l_mp;
2809 * In this pass we only want to recover all the buffers which have
2810 * not been cancelled and are not cancellation buffers themselves.
2812 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2813 buf_f->blf_len, buf_f->blf_flags)) {
2814 trace_xfs_log_recover_buf_cancel(log, buf_f);
2818 trace_xfs_log_recover_buf_recover(log, buf_f);
2821 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2822 buf_flags |= XBF_UNMAPPED;
2824 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2828 error = bp->b_error;
2830 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2835 * Recover the buffer only if we get an LSN from it and it's less than
2836 * the lsn of the transaction we are replaying.
2838 * Note that we have to be extremely careful of readahead here.
2839 * Readahead does not attach verfiers to the buffers so if we don't
2840 * actually do any replay after readahead because of the LSN we found
2841 * in the buffer if more recent than that current transaction then we
2842 * need to attach the verifier directly. Failure to do so can lead to
2843 * future recovery actions (e.g. EFI and unlinked list recovery) can
2844 * operate on the buffers and they won't get the verifier attached. This
2845 * can lead to blocks on disk having the correct content but a stale
2848 * It is safe to assume these clean buffers are currently up to date.
2849 * If the buffer is dirtied by a later transaction being replayed, then
2850 * the verifier will be reset to match whatever recover turns that
2853 lsn = xlog_recover_get_buf_lsn(mp, bp);
2854 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2855 trace_xfs_log_recover_buf_skip(log, buf_f);
2856 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2860 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2861 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2864 } else if (buf_f->blf_flags &
2865 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2868 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2872 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2876 * Perform delayed write on the buffer. Asynchronous writes will be
2877 * slower when taking into account all the buffers to be flushed.
2879 * Also make sure that only inode buffers with good sizes stay in
2880 * the buffer cache. The kernel moves inodes in buffers of 1 block
2881 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2882 * buffers in the log can be a different size if the log was generated
2883 * by an older kernel using unclustered inode buffers or a newer kernel
2884 * running with a different inode cluster size. Regardless, if the
2885 * the inode buffer size isn't max(blocksize, mp->m_inode_cluster_size)
2886 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2887 * the buffer out of the buffer cache so that the buffer won't
2888 * overlap with future reads of those inodes.
2890 if (XFS_DINODE_MAGIC ==
2891 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2892 (BBTOB(bp->b_io_length) != max(log->l_mp->m_sb.sb_blocksize,
2893 (uint32_t)log->l_mp->m_inode_cluster_size))) {
2895 error = xfs_bwrite(bp);
2897 ASSERT(bp->b_target->bt_mount == mp);
2898 bp->b_iodone = xlog_recover_iodone;
2899 xfs_buf_delwri_queue(bp, buffer_list);
2908 * Inode fork owner changes
2910 * If we have been told that we have to reparent the inode fork, it's because an
2911 * extent swap operation on a CRC enabled filesystem has been done and we are
2912 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2915 * The complexity here is that we don't have an inode context to work with, so
2916 * after we've replayed the inode we need to instantiate one. This is where the
2919 * We are in the middle of log recovery, so we can't run transactions. That
2920 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2921 * that will result in the corresponding iput() running the inode through
2922 * xfs_inactive(). If we've just replayed an inode core that changes the link
2923 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2924 * transactions (bad!).
2926 * So, to avoid this, we instantiate an inode directly from the inode core we've
2927 * just recovered. We have the buffer still locked, and all we really need to
2928 * instantiate is the inode core and the forks being modified. We can do this
2929 * manually, then run the inode btree owner change, and then tear down the
2930 * xfs_inode without having to run any transactions at all.
2932 * Also, because we don't have a transaction context available here but need to
2933 * gather all the buffers we modify for writeback so we pass the buffer_list
2934 * instead for the operation to use.
2938 xfs_recover_inode_owner_change(
2939 struct xfs_mount *mp,
2940 struct xfs_dinode *dip,
2941 struct xfs_inode_log_format *in_f,
2942 struct list_head *buffer_list)
2944 struct xfs_inode *ip;
2947 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2949 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2953 /* instantiate the inode */
2954 xfs_inode_from_disk(ip, dip);
2955 ASSERT(ip->i_d.di_version >= 3);
2957 error = xfs_iformat_fork(ip, dip);
2961 if (!xfs_inode_verify_forks(ip)) {
2962 error = -EFSCORRUPTED;
2966 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2967 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2968 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2969 ip->i_ino, buffer_list);
2974 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2975 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2976 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2977 ip->i_ino, buffer_list);
2988 xlog_recover_inode_pass2(
2990 struct list_head *buffer_list,
2991 struct xlog_recover_item *item,
2992 xfs_lsn_t current_lsn)
2994 struct xfs_inode_log_format *in_f;
2995 xfs_mount_t *mp = log->l_mp;
3004 struct xfs_log_dinode *ldip;
3008 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3009 in_f = item->ri_buf[0].i_addr;
3011 in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), KM_SLEEP);
3013 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
3019 * Inode buffers can be freed, look out for it,
3020 * and do not replay the inode.
3022 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
3023 in_f->ilf_len, 0)) {
3025 trace_xfs_log_recover_inode_cancel(log, in_f);
3028 trace_xfs_log_recover_inode_recover(log, in_f);
3030 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
3031 &xfs_inode_buf_ops);
3036 error = bp->b_error;
3038 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
3041 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
3042 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
3045 * Make sure the place we're flushing out to really looks
3048 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
3050 "%s: Bad inode magic number, dip = "PTR_FMT", dino bp = "PTR_FMT", ino = %Ld",
3051 __func__, dip, bp, in_f->ilf_ino);
3052 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
3053 XFS_ERRLEVEL_LOW, mp);
3054 error = -EFSCORRUPTED;
3057 ldip = item->ri_buf[1].i_addr;
3058 if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) {
3060 "%s: Bad inode log record, rec ptr "PTR_FMT", ino %Ld",
3061 __func__, item, in_f->ilf_ino);
3062 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
3063 XFS_ERRLEVEL_LOW, mp);
3064 error = -EFSCORRUPTED;
3069 * If the inode has an LSN in it, recover the inode only if it's less
3070 * than the lsn of the transaction we are replaying. Note: we still
3071 * need to replay an owner change even though the inode is more recent
3072 * than the transaction as there is no guarantee that all the btree
3073 * blocks are more recent than this transaction, too.
3075 if (dip->di_version >= 3) {
3076 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
3078 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3079 trace_xfs_log_recover_inode_skip(log, in_f);
3081 goto out_owner_change;
3086 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3087 * are transactional and if ordering is necessary we can determine that
3088 * more accurately by the LSN field in the V3 inode core. Don't trust
3089 * the inode versions we might be changing them here - use the
3090 * superblock flag to determine whether we need to look at di_flushiter
3091 * to skip replay when the on disk inode is newer than the log one
3093 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3094 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3096 * Deal with the wrap case, DI_MAX_FLUSH is less
3097 * than smaller numbers
3099 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3100 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3103 trace_xfs_log_recover_inode_skip(log, in_f);
3109 /* Take the opportunity to reset the flush iteration count */
3110 ldip->di_flushiter = 0;
3112 if (unlikely(S_ISREG(ldip->di_mode))) {
3113 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3114 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3115 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3116 XFS_ERRLEVEL_LOW, mp, ldip,
3119 "%s: Bad regular inode log record, rec ptr "PTR_FMT", "
3120 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3121 __func__, item, dip, bp, in_f->ilf_ino);
3122 error = -EFSCORRUPTED;
3125 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3126 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3127 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3128 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3129 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3130 XFS_ERRLEVEL_LOW, mp, ldip,
3133 "%s: Bad dir inode log record, rec ptr "PTR_FMT", "
3134 "ino ptr = "PTR_FMT", ino bp = "PTR_FMT", ino %Ld",
3135 __func__, item, dip, bp, in_f->ilf_ino);
3136 error = -EFSCORRUPTED;
3140 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3141 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3142 XFS_ERRLEVEL_LOW, mp, ldip,
3145 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3146 "dino bp "PTR_FMT", ino %Ld, total extents = %d, nblocks = %Ld",
3147 __func__, item, dip, bp, in_f->ilf_ino,
3148 ldip->di_nextents + ldip->di_anextents,
3150 error = -EFSCORRUPTED;
3153 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3154 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3155 XFS_ERRLEVEL_LOW, mp, ldip,
3158 "%s: Bad inode log record, rec ptr "PTR_FMT", dino ptr "PTR_FMT", "
3159 "dino bp "PTR_FMT", ino %Ld, forkoff 0x%x", __func__,
3160 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3161 error = -EFSCORRUPTED;
3164 isize = xfs_log_dinode_size(ldip->di_version);
3165 if (unlikely(item->ri_buf[1].i_len > isize)) {
3166 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3167 XFS_ERRLEVEL_LOW, mp, ldip,
3170 "%s: Bad inode log record length %d, rec ptr "PTR_FMT,
3171 __func__, item->ri_buf[1].i_len, item);
3172 error = -EFSCORRUPTED;
3176 /* recover the log dinode inode into the on disk inode */
3177 xfs_log_dinode_to_disk(ldip, dip);
3179 fields = in_f->ilf_fields;
3180 if (fields & XFS_ILOG_DEV)
3181 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3183 if (in_f->ilf_size == 2)
3184 goto out_owner_change;
3185 len = item->ri_buf[2].i_len;
3186 src = item->ri_buf[2].i_addr;
3187 ASSERT(in_f->ilf_size <= 4);
3188 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3189 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3190 (len == in_f->ilf_dsize));
3192 switch (fields & XFS_ILOG_DFORK) {
3193 case XFS_ILOG_DDATA:
3195 memcpy(XFS_DFORK_DPTR(dip), src, len);
3198 case XFS_ILOG_DBROOT:
3199 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3200 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3201 XFS_DFORK_DSIZE(dip, mp));
3206 * There are no data fork flags set.
3208 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3213 * If we logged any attribute data, recover it. There may or
3214 * may not have been any other non-core data logged in this
3217 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3218 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3223 len = item->ri_buf[attr_index].i_len;
3224 src = item->ri_buf[attr_index].i_addr;
3225 ASSERT(len == in_f->ilf_asize);
3227 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3228 case XFS_ILOG_ADATA:
3230 dest = XFS_DFORK_APTR(dip);
3231 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3232 memcpy(dest, src, len);
3235 case XFS_ILOG_ABROOT:
3236 dest = XFS_DFORK_APTR(dip);
3237 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3238 len, (xfs_bmdr_block_t*)dest,
3239 XFS_DFORK_ASIZE(dip, mp));
3243 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3251 /* Recover the swapext owner change unless inode has been deleted */
3252 if ((in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER)) &&
3253 (dip->di_mode != 0))
3254 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3256 /* re-generate the checksum. */
3257 xfs_dinode_calc_crc(log->l_mp, dip);
3259 ASSERT(bp->b_target->bt_mount == mp);
3260 bp->b_iodone = xlog_recover_iodone;
3261 xfs_buf_delwri_queue(bp, buffer_list);
3272 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3273 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3277 xlog_recover_quotaoff_pass1(
3279 struct xlog_recover_item *item)
3281 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3285 * The logitem format's flag tells us if this was user quotaoff,
3286 * group/project quotaoff or both.
3288 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3289 log->l_quotaoffs_flag |= XFS_DQ_USER;
3290 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3291 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3292 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3293 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3299 * Recover a dquot record
3302 xlog_recover_dquot_pass2(
3304 struct list_head *buffer_list,
3305 struct xlog_recover_item *item,
3306 xfs_lsn_t current_lsn)
3308 xfs_mount_t *mp = log->l_mp;
3310 struct xfs_disk_dquot *ddq, *recddq;
3313 xfs_dq_logformat_t *dq_f;
3318 * Filesystems are required to send in quota flags at mount time.
3320 if (mp->m_qflags == 0)
3323 recddq = item->ri_buf[1].i_addr;
3324 if (recddq == NULL) {
3325 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3328 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3329 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3330 item->ri_buf[1].i_len, __func__);
3335 * This type of quotas was turned off, so ignore this record.
3337 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3339 if (log->l_quotaoffs_flag & type)
3343 * At this point we know that quota was _not_ turned off.
3344 * Since the mount flags are not indicating to us otherwise, this
3345 * must mean that quota is on, and the dquot needs to be replayed.
3346 * Remember that we may not have fully recovered the superblock yet,
3347 * so we can't do the usual trick of looking at the SB quota bits.
3349 * The other possibility, of course, is that the quota subsystem was
3350 * removed since the last mount - ENOSYS.
3352 dq_f = item->ri_buf[0].i_addr;
3354 fa = xfs_dquot_verify(mp, recddq, dq_f->qlf_id, 0);
3356 xfs_alert(mp, "corrupt dquot ID 0x%x in log at %pS",
3360 ASSERT(dq_f->qlf_len == 1);
3363 * At this point we are assuming that the dquots have been allocated
3364 * and hence the buffer has valid dquots stamped in it. It should,
3365 * therefore, pass verifier validation. If the dquot is bad, then the
3366 * we'll return an error here, so we don't need to specifically check
3367 * the dquot in the buffer after the verifier has run.
3369 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3370 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3371 &xfs_dquot_buf_ops);
3376 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3379 * If the dquot has an LSN in it, recover the dquot only if it's less
3380 * than the lsn of the transaction we are replaying.
3382 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3383 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3384 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3386 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3391 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3392 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3393 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3397 ASSERT(dq_f->qlf_size == 2);
3398 ASSERT(bp->b_target->bt_mount == mp);
3399 bp->b_iodone = xlog_recover_iodone;
3400 xfs_buf_delwri_queue(bp, buffer_list);
3408 * This routine is called to create an in-core extent free intent
3409 * item from the efi format structure which was logged on disk.
3410 * It allocates an in-core efi, copies the extents from the format
3411 * structure into it, and adds the efi to the AIL with the given
3415 xlog_recover_efi_pass2(
3417 struct xlog_recover_item *item,
3421 struct xfs_mount *mp = log->l_mp;
3422 struct xfs_efi_log_item *efip;
3423 struct xfs_efi_log_format *efi_formatp;
3425 efi_formatp = item->ri_buf[0].i_addr;
3427 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3428 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3430 xfs_efi_item_free(efip);
3433 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3435 spin_lock(&log->l_ailp->ail_lock);
3437 * The EFI has two references. One for the EFD and one for EFI to ensure
3438 * it makes it into the AIL. Insert the EFI into the AIL directly and
3439 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3442 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3443 xfs_efi_release(efip);
3449 * This routine is called when an EFD format structure is found in a committed
3450 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3451 * was still in the log. To do this it searches the AIL for the EFI with an id
3452 * equal to that in the EFD format structure. If we find it we drop the EFD
3453 * reference, which removes the EFI from the AIL and frees it.
3456 xlog_recover_efd_pass2(
3458 struct xlog_recover_item *item)
3460 xfs_efd_log_format_t *efd_formatp;
3461 xfs_efi_log_item_t *efip = NULL;
3462 xfs_log_item_t *lip;
3464 struct xfs_ail_cursor cur;
3465 struct xfs_ail *ailp = log->l_ailp;
3467 efd_formatp = item->ri_buf[0].i_addr;
3468 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3469 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3470 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3471 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3472 efi_id = efd_formatp->efd_efi_id;
3475 * Search for the EFI with the id in the EFD format structure in the
3478 spin_lock(&ailp->ail_lock);
3479 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3480 while (lip != NULL) {
3481 if (lip->li_type == XFS_LI_EFI) {
3482 efip = (xfs_efi_log_item_t *)lip;
3483 if (efip->efi_format.efi_id == efi_id) {
3485 * Drop the EFD reference to the EFI. This
3486 * removes the EFI from the AIL and frees it.
3488 spin_unlock(&ailp->ail_lock);
3489 xfs_efi_release(efip);
3490 spin_lock(&ailp->ail_lock);
3494 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3497 xfs_trans_ail_cursor_done(&cur);
3498 spin_unlock(&ailp->ail_lock);
3504 * This routine is called to create an in-core extent rmap update
3505 * item from the rui format structure which was logged on disk.
3506 * It allocates an in-core rui, copies the extents from the format
3507 * structure into it, and adds the rui to the AIL with the given
3511 xlog_recover_rui_pass2(
3513 struct xlog_recover_item *item,
3517 struct xfs_mount *mp = log->l_mp;
3518 struct xfs_rui_log_item *ruip;
3519 struct xfs_rui_log_format *rui_formatp;
3521 rui_formatp = item->ri_buf[0].i_addr;
3523 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3524 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3526 xfs_rui_item_free(ruip);
3529 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3531 spin_lock(&log->l_ailp->ail_lock);
3533 * The RUI has two references. One for the RUD and one for RUI to ensure
3534 * it makes it into the AIL. Insert the RUI into the AIL directly and
3535 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3538 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3539 xfs_rui_release(ruip);
3545 * This routine is called when an RUD format structure is found in a committed
3546 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3547 * was still in the log. To do this it searches the AIL for the RUI with an id
3548 * equal to that in the RUD format structure. If we find it we drop the RUD
3549 * reference, which removes the RUI from the AIL and frees it.
3552 xlog_recover_rud_pass2(
3554 struct xlog_recover_item *item)
3556 struct xfs_rud_log_format *rud_formatp;
3557 struct xfs_rui_log_item *ruip = NULL;
3558 struct xfs_log_item *lip;
3560 struct xfs_ail_cursor cur;
3561 struct xfs_ail *ailp = log->l_ailp;
3563 rud_formatp = item->ri_buf[0].i_addr;
3564 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3565 rui_id = rud_formatp->rud_rui_id;
3568 * Search for the RUI with the id in the RUD format structure in the
3571 spin_lock(&ailp->ail_lock);
3572 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3573 while (lip != NULL) {
3574 if (lip->li_type == XFS_LI_RUI) {
3575 ruip = (struct xfs_rui_log_item *)lip;
3576 if (ruip->rui_format.rui_id == rui_id) {
3578 * Drop the RUD reference to the RUI. This
3579 * removes the RUI from the AIL and frees it.
3581 spin_unlock(&ailp->ail_lock);
3582 xfs_rui_release(ruip);
3583 spin_lock(&ailp->ail_lock);
3587 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3590 xfs_trans_ail_cursor_done(&cur);
3591 spin_unlock(&ailp->ail_lock);
3597 * Copy an CUI format buffer from the given buf, and into the destination
3598 * CUI format structure. The CUI/CUD items were designed not to need any
3599 * special alignment handling.
3602 xfs_cui_copy_format(
3603 struct xfs_log_iovec *buf,
3604 struct xfs_cui_log_format *dst_cui_fmt)
3606 struct xfs_cui_log_format *src_cui_fmt;
3609 src_cui_fmt = buf->i_addr;
3610 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3612 if (buf->i_len == len) {
3613 memcpy(dst_cui_fmt, src_cui_fmt, len);
3616 return -EFSCORRUPTED;
3620 * This routine is called to create an in-core extent refcount update
3621 * item from the cui format structure which was logged on disk.
3622 * It allocates an in-core cui, copies the extents from the format
3623 * structure into it, and adds the cui to the AIL with the given
3627 xlog_recover_cui_pass2(
3629 struct xlog_recover_item *item,
3633 struct xfs_mount *mp = log->l_mp;
3634 struct xfs_cui_log_item *cuip;
3635 struct xfs_cui_log_format *cui_formatp;
3637 cui_formatp = item->ri_buf[0].i_addr;
3639 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3640 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3642 xfs_cui_item_free(cuip);
3645 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3647 spin_lock(&log->l_ailp->ail_lock);
3649 * The CUI has two references. One for the CUD and one for CUI to ensure
3650 * it makes it into the AIL. Insert the CUI into the AIL directly and
3651 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3654 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3655 xfs_cui_release(cuip);
3661 * This routine is called when an CUD format structure is found in a committed
3662 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3663 * was still in the log. To do this it searches the AIL for the CUI with an id
3664 * equal to that in the CUD format structure. If we find it we drop the CUD
3665 * reference, which removes the CUI from the AIL and frees it.
3668 xlog_recover_cud_pass2(
3670 struct xlog_recover_item *item)
3672 struct xfs_cud_log_format *cud_formatp;
3673 struct xfs_cui_log_item *cuip = NULL;
3674 struct xfs_log_item *lip;
3676 struct xfs_ail_cursor cur;
3677 struct xfs_ail *ailp = log->l_ailp;
3679 cud_formatp = item->ri_buf[0].i_addr;
3680 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format))
3681 return -EFSCORRUPTED;
3682 cui_id = cud_formatp->cud_cui_id;
3685 * Search for the CUI with the id in the CUD format structure in the
3688 spin_lock(&ailp->ail_lock);
3689 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3690 while (lip != NULL) {
3691 if (lip->li_type == XFS_LI_CUI) {
3692 cuip = (struct xfs_cui_log_item *)lip;
3693 if (cuip->cui_format.cui_id == cui_id) {
3695 * Drop the CUD reference to the CUI. This
3696 * removes the CUI from the AIL and frees it.
3698 spin_unlock(&ailp->ail_lock);
3699 xfs_cui_release(cuip);
3700 spin_lock(&ailp->ail_lock);
3704 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3707 xfs_trans_ail_cursor_done(&cur);
3708 spin_unlock(&ailp->ail_lock);
3714 * Copy an BUI format buffer from the given buf, and into the destination
3715 * BUI format structure. The BUI/BUD items were designed not to need any
3716 * special alignment handling.
3719 xfs_bui_copy_format(
3720 struct xfs_log_iovec *buf,
3721 struct xfs_bui_log_format *dst_bui_fmt)
3723 struct xfs_bui_log_format *src_bui_fmt;
3726 src_bui_fmt = buf->i_addr;
3727 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3729 if (buf->i_len == len) {
3730 memcpy(dst_bui_fmt, src_bui_fmt, len);
3733 return -EFSCORRUPTED;
3737 * This routine is called to create an in-core extent bmap update
3738 * item from the bui format structure which was logged on disk.
3739 * It allocates an in-core bui, copies the extents from the format
3740 * structure into it, and adds the bui to the AIL with the given
3744 xlog_recover_bui_pass2(
3746 struct xlog_recover_item *item,
3750 struct xfs_mount *mp = log->l_mp;
3751 struct xfs_bui_log_item *buip;
3752 struct xfs_bui_log_format *bui_formatp;
3754 bui_formatp = item->ri_buf[0].i_addr;
3756 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
3757 return -EFSCORRUPTED;
3758 buip = xfs_bui_init(mp);
3759 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3761 xfs_bui_item_free(buip);
3764 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3766 spin_lock(&log->l_ailp->ail_lock);
3768 * The RUI has two references. One for the RUD and one for RUI to ensure
3769 * it makes it into the AIL. Insert the RUI into the AIL directly and
3770 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3773 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3774 xfs_bui_release(buip);
3780 * This routine is called when an BUD format structure is found in a committed
3781 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3782 * was still in the log. To do this it searches the AIL for the BUI with an id
3783 * equal to that in the BUD format structure. If we find it we drop the BUD
3784 * reference, which removes the BUI from the AIL and frees it.
3787 xlog_recover_bud_pass2(
3789 struct xlog_recover_item *item)
3791 struct xfs_bud_log_format *bud_formatp;
3792 struct xfs_bui_log_item *buip = NULL;
3793 struct xfs_log_item *lip;
3795 struct xfs_ail_cursor cur;
3796 struct xfs_ail *ailp = log->l_ailp;
3798 bud_formatp = item->ri_buf[0].i_addr;
3799 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format))
3800 return -EFSCORRUPTED;
3801 bui_id = bud_formatp->bud_bui_id;
3804 * Search for the BUI with the id in the BUD format structure in the
3807 spin_lock(&ailp->ail_lock);
3808 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3809 while (lip != NULL) {
3810 if (lip->li_type == XFS_LI_BUI) {
3811 buip = (struct xfs_bui_log_item *)lip;
3812 if (buip->bui_format.bui_id == bui_id) {
3814 * Drop the BUD reference to the BUI. This
3815 * removes the BUI from the AIL and frees it.
3817 spin_unlock(&ailp->ail_lock);
3818 xfs_bui_release(buip);
3819 spin_lock(&ailp->ail_lock);
3823 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3826 xfs_trans_ail_cursor_done(&cur);
3827 spin_unlock(&ailp->ail_lock);
3833 * This routine is called when an inode create format structure is found in a
3834 * committed transaction in the log. It's purpose is to initialise the inodes
3835 * being allocated on disk. This requires us to get inode cluster buffers that
3836 * match the range to be initialised, stamped with inode templates and written
3837 * by delayed write so that subsequent modifications will hit the cached buffer
3838 * and only need writing out at the end of recovery.
3841 xlog_recover_do_icreate_pass2(
3843 struct list_head *buffer_list,
3844 xlog_recover_item_t *item)
3846 struct xfs_mount *mp = log->l_mp;
3847 struct xfs_icreate_log *icl;
3848 xfs_agnumber_t agno;
3849 xfs_agblock_t agbno;
3852 xfs_agblock_t length;
3858 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3859 if (icl->icl_type != XFS_LI_ICREATE) {
3860 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3864 if (icl->icl_size != 1) {
3865 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3869 agno = be32_to_cpu(icl->icl_ag);
3870 if (agno >= mp->m_sb.sb_agcount) {
3871 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3874 agbno = be32_to_cpu(icl->icl_agbno);
3875 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3876 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3879 isize = be32_to_cpu(icl->icl_isize);
3880 if (isize != mp->m_sb.sb_inodesize) {
3881 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3884 count = be32_to_cpu(icl->icl_count);
3886 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3889 length = be32_to_cpu(icl->icl_length);
3890 if (!length || length >= mp->m_sb.sb_agblocks) {
3891 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3896 * The inode chunk is either full or sparse and we only support
3897 * m_ialloc_min_blks sized sparse allocations at this time.
3899 if (length != mp->m_ialloc_blks &&
3900 length != mp->m_ialloc_min_blks) {
3902 "%s: unsupported chunk length", __FUNCTION__);
3906 /* verify inode count is consistent with extent length */
3907 if ((count >> mp->m_sb.sb_inopblog) != length) {
3909 "%s: inconsistent inode count and chunk length",
3915 * The icreate transaction can cover multiple cluster buffers and these
3916 * buffers could have been freed and reused. Check the individual
3917 * buffers for cancellation so we don't overwrite anything written after
3920 bb_per_cluster = XFS_FSB_TO_BB(mp, mp->m_blocks_per_cluster);
3921 nbufs = length / mp->m_blocks_per_cluster;
3922 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3925 daddr = XFS_AGB_TO_DADDR(mp, agno,
3926 agbno + i * mp->m_blocks_per_cluster);
3927 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3932 * We currently only use icreate for a single allocation at a time. This
3933 * means we should expect either all or none of the buffers to be
3934 * cancelled. Be conservative and skip replay if at least one buffer is
3935 * cancelled, but warn the user that something is awry if the buffers
3936 * are not consistent.
3938 * XXX: This must be refined to only skip cancelled clusters once we use
3939 * icreate for multiple chunk allocations.
3941 ASSERT(!cancel_count || cancel_count == nbufs);
3943 if (cancel_count != nbufs)
3945 "WARNING: partial inode chunk cancellation, skipped icreate.");
3946 trace_xfs_log_recover_icreate_cancel(log, icl);
3950 trace_xfs_log_recover_icreate_recover(log, icl);
3951 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3952 length, be32_to_cpu(icl->icl_gen));
3956 xlog_recover_buffer_ra_pass2(
3958 struct xlog_recover_item *item)
3960 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3961 struct xfs_mount *mp = log->l_mp;
3963 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3964 buf_f->blf_len, buf_f->blf_flags)) {
3968 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3969 buf_f->blf_len, NULL);
3973 xlog_recover_inode_ra_pass2(
3975 struct xlog_recover_item *item)
3977 struct xfs_inode_log_format ilf_buf;
3978 struct xfs_inode_log_format *ilfp;
3979 struct xfs_mount *mp = log->l_mp;
3982 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3983 ilfp = item->ri_buf[0].i_addr;
3986 memset(ilfp, 0, sizeof(*ilfp));
3987 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3992 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3995 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3996 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
4000 xlog_recover_dquot_ra_pass2(
4002 struct xlog_recover_item *item)
4004 struct xfs_mount *mp = log->l_mp;
4005 struct xfs_disk_dquot *recddq;
4006 struct xfs_dq_logformat *dq_f;
4011 if (mp->m_qflags == 0)
4014 recddq = item->ri_buf[1].i_addr;
4017 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
4020 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
4022 if (log->l_quotaoffs_flag & type)
4025 dq_f = item->ri_buf[0].i_addr;
4027 ASSERT(dq_f->qlf_len == 1);
4029 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
4030 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
4033 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
4034 &xfs_dquot_buf_ra_ops);
4038 xlog_recover_ra_pass2(
4040 struct xlog_recover_item *item)
4042 switch (ITEM_TYPE(item)) {
4044 xlog_recover_buffer_ra_pass2(log, item);
4047 xlog_recover_inode_ra_pass2(log, item);
4050 xlog_recover_dquot_ra_pass2(log, item);
4054 case XFS_LI_QUOTAOFF:
4067 xlog_recover_commit_pass1(
4069 struct xlog_recover *trans,
4070 struct xlog_recover_item *item)
4072 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
4074 switch (ITEM_TYPE(item)) {
4076 return xlog_recover_buffer_pass1(log, item);
4077 case XFS_LI_QUOTAOFF:
4078 return xlog_recover_quotaoff_pass1(log, item);
4083 case XFS_LI_ICREATE:
4090 /* nothing to do in pass 1 */
4093 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4094 __func__, ITEM_TYPE(item));
4101 xlog_recover_commit_pass2(
4103 struct xlog_recover *trans,
4104 struct list_head *buffer_list,
4105 struct xlog_recover_item *item)
4107 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4109 switch (ITEM_TYPE(item)) {
4111 return xlog_recover_buffer_pass2(log, buffer_list, item,
4114 return xlog_recover_inode_pass2(log, buffer_list, item,
4117 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4119 return xlog_recover_efd_pass2(log, item);
4121 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4123 return xlog_recover_rud_pass2(log, item);
4125 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4127 return xlog_recover_cud_pass2(log, item);
4129 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4131 return xlog_recover_bud_pass2(log, item);
4133 return xlog_recover_dquot_pass2(log, buffer_list, item,
4135 case XFS_LI_ICREATE:
4136 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4137 case XFS_LI_QUOTAOFF:
4138 /* nothing to do in pass2 */
4141 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4142 __func__, ITEM_TYPE(item));
4149 xlog_recover_items_pass2(
4151 struct xlog_recover *trans,
4152 struct list_head *buffer_list,
4153 struct list_head *item_list)
4155 struct xlog_recover_item *item;
4158 list_for_each_entry(item, item_list, ri_list) {
4159 error = xlog_recover_commit_pass2(log, trans,
4169 * Perform the transaction.
4171 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4172 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4175 xlog_recover_commit_trans(
4177 struct xlog_recover *trans,
4179 struct list_head *buffer_list)
4182 int items_queued = 0;
4183 struct xlog_recover_item *item;
4184 struct xlog_recover_item *next;
4185 LIST_HEAD (ra_list);
4186 LIST_HEAD (done_list);
4188 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4190 hlist_del_init(&trans->r_list);
4192 error = xlog_recover_reorder_trans(log, trans, pass);
4196 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4198 case XLOG_RECOVER_PASS1:
4199 error = xlog_recover_commit_pass1(log, trans, item);
4201 case XLOG_RECOVER_PASS2:
4202 xlog_recover_ra_pass2(log, item);
4203 list_move_tail(&item->ri_list, &ra_list);
4205 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4206 error = xlog_recover_items_pass2(log, trans,
4207 buffer_list, &ra_list);
4208 list_splice_tail_init(&ra_list, &done_list);
4222 if (!list_empty(&ra_list)) {
4224 error = xlog_recover_items_pass2(log, trans,
4225 buffer_list, &ra_list);
4226 list_splice_tail_init(&ra_list, &done_list);
4229 if (!list_empty(&done_list))
4230 list_splice_init(&done_list, &trans->r_itemq);
4236 xlog_recover_add_item(
4237 struct list_head *head)
4239 xlog_recover_item_t *item;
4241 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
4242 INIT_LIST_HEAD(&item->ri_list);
4243 list_add_tail(&item->ri_list, head);
4247 xlog_recover_add_to_cont_trans(
4249 struct xlog_recover *trans,
4253 xlog_recover_item_t *item;
4254 char *ptr, *old_ptr;
4258 * If the transaction is empty, the header was split across this and the
4259 * previous record. Copy the rest of the header.
4261 if (list_empty(&trans->r_itemq)) {
4262 ASSERT(len <= sizeof(struct xfs_trans_header));
4263 if (len > sizeof(struct xfs_trans_header)) {
4264 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4268 xlog_recover_add_item(&trans->r_itemq);
4269 ptr = (char *)&trans->r_theader +
4270 sizeof(struct xfs_trans_header) - len;
4271 memcpy(ptr, dp, len);
4275 /* take the tail entry */
4276 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4278 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4279 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4281 ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP);
4282 memcpy(&ptr[old_len], dp, len);
4283 item->ri_buf[item->ri_cnt-1].i_len += len;
4284 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4285 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4290 * The next region to add is the start of a new region. It could be
4291 * a whole region or it could be the first part of a new region. Because
4292 * of this, the assumption here is that the type and size fields of all
4293 * format structures fit into the first 32 bits of the structure.
4295 * This works because all regions must be 32 bit aligned. Therefore, we
4296 * either have both fields or we have neither field. In the case we have
4297 * neither field, the data part of the region is zero length. We only have
4298 * a log_op_header and can throw away the header since a new one will appear
4299 * later. If we have at least 4 bytes, then we can determine how many regions
4300 * will appear in the current log item.
4303 xlog_recover_add_to_trans(
4305 struct xlog_recover *trans,
4309 struct xfs_inode_log_format *in_f; /* any will do */
4310 xlog_recover_item_t *item;
4315 if (list_empty(&trans->r_itemq)) {
4316 /* we need to catch log corruptions here */
4317 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4318 xfs_warn(log->l_mp, "%s: bad header magic number",
4324 if (len > sizeof(struct xfs_trans_header)) {
4325 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4331 * The transaction header can be arbitrarily split across op
4332 * records. If we don't have the whole thing here, copy what we
4333 * do have and handle the rest in the next record.
4335 if (len == sizeof(struct xfs_trans_header))
4336 xlog_recover_add_item(&trans->r_itemq);
4337 memcpy(&trans->r_theader, dp, len);
4341 ptr = kmem_alloc(len, KM_SLEEP);
4342 memcpy(ptr, dp, len);
4343 in_f = (struct xfs_inode_log_format *)ptr;
4345 /* take the tail entry */
4346 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4347 if (item->ri_total != 0 &&
4348 item->ri_total == item->ri_cnt) {
4349 /* tail item is in use, get a new one */
4350 xlog_recover_add_item(&trans->r_itemq);
4351 item = list_entry(trans->r_itemq.prev,
4352 xlog_recover_item_t, ri_list);
4355 if (item->ri_total == 0) { /* first region to be added */
4356 if (in_f->ilf_size == 0 ||
4357 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4359 "bad number of regions (%d) in inode log format",
4366 item->ri_total = in_f->ilf_size;
4368 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4371 ASSERT(item->ri_total > item->ri_cnt);
4372 /* Description region is ri_buf[0] */
4373 item->ri_buf[item->ri_cnt].i_addr = ptr;
4374 item->ri_buf[item->ri_cnt].i_len = len;
4376 trace_xfs_log_recover_item_add(log, trans, item, 0);
4381 * Free up any resources allocated by the transaction
4383 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4386 xlog_recover_free_trans(
4387 struct xlog_recover *trans)
4389 xlog_recover_item_t *item, *n;
4392 hlist_del_init(&trans->r_list);
4394 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4395 /* Free the regions in the item. */
4396 list_del(&item->ri_list);
4397 for (i = 0; i < item->ri_cnt; i++)
4398 kmem_free(item->ri_buf[i].i_addr);
4399 /* Free the item itself */
4400 kmem_free(item->ri_buf);
4403 /* Free the transaction recover structure */
4408 * On error or completion, trans is freed.
4411 xlog_recovery_process_trans(
4413 struct xlog_recover *trans,
4418 struct list_head *buffer_list)
4421 bool freeit = false;
4423 /* mask off ophdr transaction container flags */
4424 flags &= ~XLOG_END_TRANS;
4425 if (flags & XLOG_WAS_CONT_TRANS)
4426 flags &= ~XLOG_CONTINUE_TRANS;
4429 * Callees must not free the trans structure. We'll decide if we need to
4430 * free it or not based on the operation being done and it's result.
4433 /* expected flag values */
4435 case XLOG_CONTINUE_TRANS:
4436 error = xlog_recover_add_to_trans(log, trans, dp, len);
4438 case XLOG_WAS_CONT_TRANS:
4439 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4441 case XLOG_COMMIT_TRANS:
4442 error = xlog_recover_commit_trans(log, trans, pass,
4444 /* success or fail, we are now done with this transaction. */
4448 /* unexpected flag values */
4449 case XLOG_UNMOUNT_TRANS:
4450 /* just skip trans */
4451 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4454 case XLOG_START_TRANS:
4456 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4461 if (error || freeit)
4462 xlog_recover_free_trans(trans);
4467 * Lookup the transaction recovery structure associated with the ID in the
4468 * current ophdr. If the transaction doesn't exist and the start flag is set in
4469 * the ophdr, then allocate a new transaction for future ID matches to find.
4470 * Either way, return what we found during the lookup - an existing transaction
4473 STATIC struct xlog_recover *
4474 xlog_recover_ophdr_to_trans(
4475 struct hlist_head rhash[],
4476 struct xlog_rec_header *rhead,
4477 struct xlog_op_header *ohead)
4479 struct xlog_recover *trans;
4481 struct hlist_head *rhp;
4483 tid = be32_to_cpu(ohead->oh_tid);
4484 rhp = &rhash[XLOG_RHASH(tid)];
4485 hlist_for_each_entry(trans, rhp, r_list) {
4486 if (trans->r_log_tid == tid)
4491 * skip over non-start transaction headers - we could be
4492 * processing slack space before the next transaction starts
4494 if (!(ohead->oh_flags & XLOG_START_TRANS))
4497 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4500 * This is a new transaction so allocate a new recovery container to
4501 * hold the recovery ops that will follow.
4503 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
4504 trans->r_log_tid = tid;
4505 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4506 INIT_LIST_HEAD(&trans->r_itemq);
4507 INIT_HLIST_NODE(&trans->r_list);
4508 hlist_add_head(&trans->r_list, rhp);
4511 * Nothing more to do for this ophdr. Items to be added to this new
4512 * transaction will be in subsequent ophdr containers.
4518 xlog_recover_process_ophdr(
4520 struct hlist_head rhash[],
4521 struct xlog_rec_header *rhead,
4522 struct xlog_op_header *ohead,
4526 struct list_head *buffer_list)
4528 struct xlog_recover *trans;
4532 /* Do we understand who wrote this op? */
4533 if (ohead->oh_clientid != XFS_TRANSACTION &&
4534 ohead->oh_clientid != XFS_LOG) {
4535 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4536 __func__, ohead->oh_clientid);
4542 * Check the ophdr contains all the data it is supposed to contain.
4544 len = be32_to_cpu(ohead->oh_len);
4545 if (dp + len > end) {
4546 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4551 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4553 /* nothing to do, so skip over this ophdr */
4558 * The recovered buffer queue is drained only once we know that all
4559 * recovery items for the current LSN have been processed. This is
4562 * - Buffer write submission updates the metadata LSN of the buffer.
4563 * - Log recovery skips items with a metadata LSN >= the current LSN of
4564 * the recovery item.
4565 * - Separate recovery items against the same metadata buffer can share
4566 * a current LSN. I.e., consider that the LSN of a recovery item is
4567 * defined as the starting LSN of the first record in which its
4568 * transaction appears, that a record can hold multiple transactions,
4569 * and/or that a transaction can span multiple records.
4571 * In other words, we are allowed to submit a buffer from log recovery
4572 * once per current LSN. Otherwise, we may incorrectly skip recovery
4573 * items and cause corruption.
4575 * We don't know up front whether buffers are updated multiple times per
4576 * LSN. Therefore, track the current LSN of each commit log record as it
4577 * is processed and drain the queue when it changes. Use commit records
4578 * because they are ordered correctly by the logging code.
4580 if (log->l_recovery_lsn != trans->r_lsn &&
4581 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4582 error = xfs_buf_delwri_submit(buffer_list);
4585 log->l_recovery_lsn = trans->r_lsn;
4588 return xlog_recovery_process_trans(log, trans, dp, len,
4589 ohead->oh_flags, pass, buffer_list);
4593 * There are two valid states of the r_state field. 0 indicates that the
4594 * transaction structure is in a normal state. We have either seen the
4595 * start of the transaction or the last operation we added was not a partial
4596 * operation. If the last operation we added to the transaction was a
4597 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4599 * NOTE: skip LRs with 0 data length.
4602 xlog_recover_process_data(
4604 struct hlist_head rhash[],
4605 struct xlog_rec_header *rhead,
4608 struct list_head *buffer_list)
4610 struct xlog_op_header *ohead;
4615 end = dp + be32_to_cpu(rhead->h_len);
4616 num_logops = be32_to_cpu(rhead->h_num_logops);
4618 /* check the log format matches our own - else we can't recover */
4619 if (xlog_header_check_recover(log->l_mp, rhead))
4622 trace_xfs_log_recover_record(log, rhead, pass);
4623 while ((dp < end) && num_logops) {
4625 ohead = (struct xlog_op_header *)dp;
4626 dp += sizeof(*ohead);
4629 /* errors will abort recovery */
4630 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4631 dp, end, pass, buffer_list);
4635 dp += be32_to_cpu(ohead->oh_len);
4641 /* Recover the EFI if necessary. */
4643 xlog_recover_process_efi(
4644 struct xfs_mount *mp,
4645 struct xfs_ail *ailp,
4646 struct xfs_log_item *lip)
4648 struct xfs_efi_log_item *efip;
4652 * Skip EFIs that we've already processed.
4654 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4655 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4658 spin_unlock(&ailp->ail_lock);
4659 error = xfs_efi_recover(mp, efip);
4660 spin_lock(&ailp->ail_lock);
4665 /* Release the EFI since we're cancelling everything. */
4667 xlog_recover_cancel_efi(
4668 struct xfs_mount *mp,
4669 struct xfs_ail *ailp,
4670 struct xfs_log_item *lip)
4672 struct xfs_efi_log_item *efip;
4674 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4676 spin_unlock(&ailp->ail_lock);
4677 xfs_efi_release(efip);
4678 spin_lock(&ailp->ail_lock);
4681 /* Recover the RUI if necessary. */
4683 xlog_recover_process_rui(
4684 struct xfs_mount *mp,
4685 struct xfs_ail *ailp,
4686 struct xfs_log_item *lip)
4688 struct xfs_rui_log_item *ruip;
4692 * Skip RUIs that we've already processed.
4694 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4695 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4698 spin_unlock(&ailp->ail_lock);
4699 error = xfs_rui_recover(mp, ruip);
4700 spin_lock(&ailp->ail_lock);
4705 /* Release the RUI since we're cancelling everything. */
4707 xlog_recover_cancel_rui(
4708 struct xfs_mount *mp,
4709 struct xfs_ail *ailp,
4710 struct xfs_log_item *lip)
4712 struct xfs_rui_log_item *ruip;
4714 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4716 spin_unlock(&ailp->ail_lock);
4717 xfs_rui_release(ruip);
4718 spin_lock(&ailp->ail_lock);
4721 /* Recover the CUI if necessary. */
4723 xlog_recover_process_cui(
4724 struct xfs_trans *parent_tp,
4725 struct xfs_ail *ailp,
4726 struct xfs_log_item *lip)
4728 struct xfs_cui_log_item *cuip;
4732 * Skip CUIs that we've already processed.
4734 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4735 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4738 spin_unlock(&ailp->ail_lock);
4739 error = xfs_cui_recover(parent_tp, cuip);
4740 spin_lock(&ailp->ail_lock);
4745 /* Release the CUI since we're cancelling everything. */
4747 xlog_recover_cancel_cui(
4748 struct xfs_mount *mp,
4749 struct xfs_ail *ailp,
4750 struct xfs_log_item *lip)
4752 struct xfs_cui_log_item *cuip;
4754 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4756 spin_unlock(&ailp->ail_lock);
4757 xfs_cui_release(cuip);
4758 spin_lock(&ailp->ail_lock);
4761 /* Recover the BUI if necessary. */
4763 xlog_recover_process_bui(
4764 struct xfs_trans *parent_tp,
4765 struct xfs_ail *ailp,
4766 struct xfs_log_item *lip)
4768 struct xfs_bui_log_item *buip;
4772 * Skip BUIs that we've already processed.
4774 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4775 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4778 spin_unlock(&ailp->ail_lock);
4779 error = xfs_bui_recover(parent_tp, buip);
4780 spin_lock(&ailp->ail_lock);
4785 /* Release the BUI since we're cancelling everything. */
4787 xlog_recover_cancel_bui(
4788 struct xfs_mount *mp,
4789 struct xfs_ail *ailp,
4790 struct xfs_log_item *lip)
4792 struct xfs_bui_log_item *buip;
4794 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4796 spin_unlock(&ailp->ail_lock);
4797 xfs_bui_release(buip);
4798 spin_lock(&ailp->ail_lock);
4801 /* Is this log item a deferred action intent? */
4802 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4804 switch (lip->li_type) {
4815 /* Take all the collected deferred ops and finish them in order. */
4817 xlog_finish_defer_ops(
4818 struct xfs_trans *parent_tp)
4820 struct xfs_mount *mp = parent_tp->t_mountp;
4821 struct xfs_trans *tp;
4827 * We're finishing the defer_ops that accumulated as a result of
4828 * recovering unfinished intent items during log recovery. We
4829 * reserve an itruncate transaction because it is the largest
4830 * permanent transaction type. Since we're the only user of the fs
4831 * right now, take 93% (15/16) of the available free blocks. Use
4832 * weird math to avoid a 64-bit division.
4834 freeblks = percpu_counter_sum(&mp->m_fdblocks);
4837 resblks = min_t(int64_t, UINT_MAX, freeblks);
4838 resblks = (resblks * 15) >> 4;
4839 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
4840 0, XFS_TRANS_RESERVE, &tp);
4843 /* transfer all collected dfops to this transaction */
4844 xfs_defer_move(tp, parent_tp);
4846 return xfs_trans_commit(tp);
4850 * When this is called, all of the log intent items which did not have
4851 * corresponding log done items should be in the AIL. What we do now
4852 * is update the data structures associated with each one.
4854 * Since we process the log intent items in normal transactions, they
4855 * will be removed at some point after the commit. This prevents us
4856 * from just walking down the list processing each one. We'll use a
4857 * flag in the intent item to skip those that we've already processed
4858 * and use the AIL iteration mechanism's generation count to try to
4859 * speed this up at least a bit.
4861 * When we start, we know that the intents are the only things in the
4862 * AIL. As we process them, however, other items are added to the
4866 xlog_recover_process_intents(
4869 struct xfs_trans *parent_tp;
4870 struct xfs_ail_cursor cur;
4871 struct xfs_log_item *lip;
4872 struct xfs_ail *ailp;
4874 #if defined(DEBUG) || defined(XFS_WARN)
4879 * The intent recovery handlers commit transactions to complete recovery
4880 * for individual intents, but any new deferred operations that are
4881 * queued during that process are held off until the very end. The
4882 * purpose of this transaction is to serve as a container for deferred
4883 * operations. Each intent recovery handler must transfer dfops here
4884 * before its local transaction commits, and we'll finish the entire
4887 error = xfs_trans_alloc_empty(log->l_mp, &parent_tp);
4892 spin_lock(&ailp->ail_lock);
4893 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4894 #if defined(DEBUG) || defined(XFS_WARN)
4895 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4897 while (lip != NULL) {
4899 * We're done when we see something other than an intent.
4900 * There should be no intents left in the AIL now.
4902 if (!xlog_item_is_intent(lip)) {
4904 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4905 ASSERT(!xlog_item_is_intent(lip));
4911 * We should never see a redo item with a LSN higher than
4912 * the last transaction we found in the log at the start
4915 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4918 * NOTE: If your intent processing routine can create more
4919 * deferred ops, you /must/ attach them to the dfops in this
4920 * routine or else those subsequent intents will get
4921 * replayed in the wrong order!
4923 switch (lip->li_type) {
4925 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4928 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4931 error = xlog_recover_process_cui(parent_tp, ailp, lip);
4934 error = xlog_recover_process_bui(parent_tp, ailp, lip);
4939 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4942 xfs_trans_ail_cursor_done(&cur);
4943 spin_unlock(&ailp->ail_lock);
4945 error = xlog_finish_defer_ops(parent_tp);
4946 xfs_trans_cancel(parent_tp);
4952 * A cancel occurs when the mount has failed and we're bailing out.
4953 * Release all pending log intent items so they don't pin the AIL.
4956 xlog_recover_cancel_intents(
4959 struct xfs_log_item *lip;
4961 struct xfs_ail_cursor cur;
4962 struct xfs_ail *ailp;
4965 spin_lock(&ailp->ail_lock);
4966 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4967 while (lip != NULL) {
4969 * We're done when we see something other than an intent.
4970 * There should be no intents left in the AIL now.
4972 if (!xlog_item_is_intent(lip)) {
4974 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4975 ASSERT(!xlog_item_is_intent(lip));
4980 switch (lip->li_type) {
4982 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4985 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4988 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4991 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4995 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4998 xfs_trans_ail_cursor_done(&cur);
4999 spin_unlock(&ailp->ail_lock);
5004 * This routine performs a transaction to null out a bad inode pointer
5005 * in an agi unlinked inode hash bucket.
5008 xlog_recover_clear_agi_bucket(
5010 xfs_agnumber_t agno,
5019 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
5023 error = xfs_read_agi(mp, tp, agno, &agibp);
5027 agi = XFS_BUF_TO_AGI(agibp);
5028 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
5029 offset = offsetof(xfs_agi_t, agi_unlinked) +
5030 (sizeof(xfs_agino_t) * bucket);
5031 xfs_trans_log_buf(tp, agibp, offset,
5032 (offset + sizeof(xfs_agino_t) - 1));
5034 error = xfs_trans_commit(tp);
5040 xfs_trans_cancel(tp);
5042 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
5047 xlog_recover_process_one_iunlink(
5048 struct xfs_mount *mp,
5049 xfs_agnumber_t agno,
5053 struct xfs_buf *ibp;
5054 struct xfs_dinode *dip;
5055 struct xfs_inode *ip;
5059 ino = XFS_AGINO_TO_INO(mp, agno, agino);
5060 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
5065 * Get the on disk inode to find the next inode in the bucket.
5067 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
5071 xfs_iflags_clear(ip, XFS_IRECOVERY);
5072 ASSERT(VFS_I(ip)->i_nlink == 0);
5073 ASSERT(VFS_I(ip)->i_mode != 0);
5075 /* setup for the next pass */
5076 agino = be32_to_cpu(dip->di_next_unlinked);
5080 * Prevent any DMAPI event from being sent when the reference on
5081 * the inode is dropped.
5083 ip->i_d.di_dmevmask = 0;
5092 * We can't read in the inode this bucket points to, or this inode
5093 * is messed up. Just ditch this bucket of inodes. We will lose
5094 * some inodes and space, but at least we won't hang.
5096 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5097 * clear the inode pointer in the bucket.
5099 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5104 * xlog_iunlink_recover
5106 * This is called during recovery to process any inodes which
5107 * we unlinked but not freed when the system crashed. These
5108 * inodes will be on the lists in the AGI blocks. What we do
5109 * here is scan all the AGIs and fully truncate and free any
5110 * inodes found on the lists. Each inode is removed from the
5111 * lists when it has been fully truncated and is freed. The
5112 * freeing of the inode and its removal from the list must be
5116 xlog_recover_process_iunlinks(
5120 xfs_agnumber_t agno;
5129 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5131 * Find the agi for this ag.
5133 error = xfs_read_agi(mp, NULL, agno, &agibp);
5136 * AGI is b0rked. Don't process it.
5138 * We should probably mark the filesystem as corrupt
5139 * after we've recovered all the ag's we can....
5144 * Unlock the buffer so that it can be acquired in the normal
5145 * course of the transaction to truncate and free each inode.
5146 * Because we are not racing with anyone else here for the AGI
5147 * buffer, we don't even need to hold it locked to read the
5148 * initial unlinked bucket entries out of the buffer. We keep
5149 * buffer reference though, so that it stays pinned in memory
5150 * while we need the buffer.
5152 agi = XFS_BUF_TO_AGI(agibp);
5153 xfs_buf_unlock(agibp);
5155 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5156 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5157 while (agino != NULLAGINO) {
5158 agino = xlog_recover_process_one_iunlink(mp,
5159 agno, agino, bucket);
5162 xfs_buf_rele(agibp);
5168 struct xlog_rec_header *rhead,
5174 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5175 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5176 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5180 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5181 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5182 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5183 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5184 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5185 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5194 * CRC check, unpack and process a log record.
5197 xlog_recover_process(
5199 struct hlist_head rhash[],
5200 struct xlog_rec_header *rhead,
5203 struct list_head *buffer_list)
5206 __le32 old_crc = rhead->h_crc;
5210 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5213 * Nothing else to do if this is a CRC verification pass. Just return
5214 * if this a record with a non-zero crc. Unfortunately, mkfs always
5215 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5216 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5217 * know precisely what failed.
5219 if (pass == XLOG_RECOVER_CRCPASS) {
5220 if (old_crc && crc != old_crc)
5226 * We're in the normal recovery path. Issue a warning if and only if the
5227 * CRC in the header is non-zero. This is an advisory warning and the
5228 * zero CRC check prevents warnings from being emitted when upgrading
5229 * the kernel from one that does not add CRCs by default.
5231 if (crc != old_crc) {
5232 if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5233 xfs_alert(log->l_mp,
5234 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5235 le32_to_cpu(old_crc),
5237 xfs_hex_dump(dp, 32);
5241 * If the filesystem is CRC enabled, this mismatch becomes a
5242 * fatal log corruption failure.
5244 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
5245 return -EFSCORRUPTED;
5248 error = xlog_unpack_data(rhead, dp, log);
5252 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5257 xlog_valid_rec_header(
5259 struct xlog_rec_header *rhead,
5264 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
5265 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5266 XFS_ERRLEVEL_LOW, log->l_mp);
5267 return -EFSCORRUPTED;
5270 (!rhead->h_version ||
5271 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
5272 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5273 __func__, be32_to_cpu(rhead->h_version));
5277 /* LR body must have data or it wouldn't have been written */
5278 hlen = be32_to_cpu(rhead->h_len);
5279 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
5280 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5281 XFS_ERRLEVEL_LOW, log->l_mp);
5282 return -EFSCORRUPTED;
5284 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
5285 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5286 XFS_ERRLEVEL_LOW, log->l_mp);
5287 return -EFSCORRUPTED;
5293 * Read the log from tail to head and process the log records found.
5294 * Handle the two cases where the tail and head are in the same cycle
5295 * and where the active portion of the log wraps around the end of
5296 * the physical log separately. The pass parameter is passed through
5297 * to the routines called to process the data and is not looked at
5301 xlog_do_recovery_pass(
5303 xfs_daddr_t head_blk,
5304 xfs_daddr_t tail_blk,
5306 xfs_daddr_t *first_bad) /* out: first bad log rec */
5308 xlog_rec_header_t *rhead;
5309 xfs_daddr_t blk_no, rblk_no;
5310 xfs_daddr_t rhead_blk;
5312 xfs_buf_t *hbp, *dbp;
5313 int error = 0, h_size, h_len;
5315 int bblks, split_bblks;
5316 int hblks, split_hblks, wrapped_hblks;
5318 struct hlist_head rhash[XLOG_RHASH_SIZE];
5319 LIST_HEAD (buffer_list);
5321 ASSERT(head_blk != tail_blk);
5322 blk_no = rhead_blk = tail_blk;
5324 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5325 INIT_HLIST_HEAD(&rhash[i]);
5328 * Read the header of the tail block and get the iclog buffer size from
5329 * h_size. Use this to tell how many sectors make up the log header.
5331 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5333 * When using variable length iclogs, read first sector of
5334 * iclog header and extract the header size from it. Get a
5335 * new hbp that is the correct size.
5337 hbp = xlog_get_bp(log, 1);
5341 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5345 rhead = (xlog_rec_header_t *)offset;
5346 error = xlog_valid_rec_header(log, rhead, tail_blk);
5351 * xfsprogs has a bug where record length is based on lsunit but
5352 * h_size (iclog size) is hardcoded to 32k. Now that we
5353 * unconditionally CRC verify the unmount record, this means the
5354 * log buffer can be too small for the record and cause an
5357 * Detect this condition here. Use lsunit for the buffer size as
5358 * long as this looks like the mkfs case. Otherwise, return an
5359 * error to avoid a buffer overrun.
5361 h_size = be32_to_cpu(rhead->h_size);
5362 h_len = be32_to_cpu(rhead->h_len);
5363 if (h_len > h_size) {
5364 if (h_len <= log->l_mp->m_logbsize &&
5365 be32_to_cpu(rhead->h_num_logops) == 1) {
5367 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5368 h_size, log->l_mp->m_logbsize);
5369 h_size = log->l_mp->m_logbsize;
5371 return -EFSCORRUPTED;
5374 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5375 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5376 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5377 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5380 hbp = xlog_get_bp(log, hblks);
5385 ASSERT(log->l_sectBBsize == 1);
5387 hbp = xlog_get_bp(log, 1);
5388 h_size = XLOG_BIG_RECORD_BSIZE;
5393 dbp = xlog_get_bp(log, BTOBB(h_size));
5399 memset(rhash, 0, sizeof(rhash));
5400 if (tail_blk > head_blk) {
5402 * Perform recovery around the end of the physical log.
5403 * When the head is not on the same cycle number as the tail,
5404 * we can't do a sequential recovery.
5406 while (blk_no < log->l_logBBsize) {
5408 * Check for header wrapping around physical end-of-log
5410 offset = hbp->b_addr;
5413 if (blk_no + hblks <= log->l_logBBsize) {
5414 /* Read header in one read */
5415 error = xlog_bread(log, blk_no, hblks, hbp,
5420 /* This LR is split across physical log end */
5421 if (blk_no != log->l_logBBsize) {
5422 /* some data before physical log end */
5423 ASSERT(blk_no <= INT_MAX);
5424 split_hblks = log->l_logBBsize - (int)blk_no;
5425 ASSERT(split_hblks > 0);
5426 error = xlog_bread(log, blk_no,
5434 * Note: this black magic still works with
5435 * large sector sizes (non-512) only because:
5436 * - we increased the buffer size originally
5437 * by 1 sector giving us enough extra space
5438 * for the second read;
5439 * - the log start is guaranteed to be sector
5441 * - we read the log end (LR header start)
5442 * _first_, then the log start (LR header end)
5443 * - order is important.
5445 wrapped_hblks = hblks - split_hblks;
5446 error = xlog_bread_offset(log, 0,
5448 offset + BBTOB(split_hblks));
5452 rhead = (xlog_rec_header_t *)offset;
5453 error = xlog_valid_rec_header(log, rhead,
5454 split_hblks ? blk_no : 0);
5458 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5462 * Read the log record data in multiple reads if it
5463 * wraps around the end of the log. Note that if the
5464 * header already wrapped, blk_no could point past the
5465 * end of the log. The record data is contiguous in
5468 if (blk_no + bblks <= log->l_logBBsize ||
5469 blk_no >= log->l_logBBsize) {
5470 rblk_no = xlog_wrap_logbno(log, blk_no);
5471 error = xlog_bread(log, rblk_no, bblks, dbp,
5476 /* This log record is split across the
5477 * physical end of log */
5478 offset = dbp->b_addr;
5480 if (blk_no != log->l_logBBsize) {
5481 /* some data is before the physical
5483 ASSERT(!wrapped_hblks);
5484 ASSERT(blk_no <= INT_MAX);
5486 log->l_logBBsize - (int)blk_no;
5487 ASSERT(split_bblks > 0);
5488 error = xlog_bread(log, blk_no,
5496 * Note: this black magic still works with
5497 * large sector sizes (non-512) only because:
5498 * - we increased the buffer size originally
5499 * by 1 sector giving us enough extra space
5500 * for the second read;
5501 * - the log start is guaranteed to be sector
5503 * - we read the log end (LR header start)
5504 * _first_, then the log start (LR header end)
5505 * - order is important.
5507 error = xlog_bread_offset(log, 0,
5508 bblks - split_bblks, dbp,
5509 offset + BBTOB(split_bblks));
5514 error = xlog_recover_process(log, rhash, rhead, offset,
5515 pass, &buffer_list);
5523 ASSERT(blk_no >= log->l_logBBsize);
5524 blk_no -= log->l_logBBsize;
5528 /* read first part of physical log */
5529 while (blk_no < head_blk) {
5530 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5534 rhead = (xlog_rec_header_t *)offset;
5535 error = xlog_valid_rec_header(log, rhead, blk_no);
5539 /* blocks in data section */
5540 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5541 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5546 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5551 blk_no += bblks + hblks;
5561 * Submit buffers that have been added from the last record processed,
5562 * regardless of error status.
5564 if (!list_empty(&buffer_list))
5565 error2 = xfs_buf_delwri_submit(&buffer_list);
5567 if (error && first_bad)
5568 *first_bad = rhead_blk;
5571 * Transactions are freed at commit time but transactions without commit
5572 * records on disk are never committed. Free any that may be left in the
5575 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5576 struct hlist_node *tmp;
5577 struct xlog_recover *trans;
5579 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5580 xlog_recover_free_trans(trans);
5583 return error ? error : error2;
5587 * Do the recovery of the log. We actually do this in two phases.
5588 * The two passes are necessary in order to implement the function
5589 * of cancelling a record written into the log. The first pass
5590 * determines those things which have been cancelled, and the
5591 * second pass replays log items normally except for those which
5592 * have been cancelled. The handling of the replay and cancellations
5593 * takes place in the log item type specific routines.
5595 * The table of items which have cancel records in the log is allocated
5596 * and freed at this level, since only here do we know when all of
5597 * the log recovery has been completed.
5600 xlog_do_log_recovery(
5602 xfs_daddr_t head_blk,
5603 xfs_daddr_t tail_blk)
5607 ASSERT(head_blk != tail_blk);
5610 * First do a pass to find all of the cancelled buf log items.
5611 * Store them in the buf_cancel_table for use in the second pass.
5613 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5614 sizeof(struct list_head),
5616 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5617 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5619 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5620 XLOG_RECOVER_PASS1, NULL);
5622 kmem_free(log->l_buf_cancel_table);
5623 log->l_buf_cancel_table = NULL;
5627 * Then do a second pass to actually recover the items in the log.
5628 * When it is complete free the table of buf cancel items.
5630 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5631 XLOG_RECOVER_PASS2, NULL);
5636 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5637 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5641 kmem_free(log->l_buf_cancel_table);
5642 log->l_buf_cancel_table = NULL;
5648 * Do the actual recovery
5653 xfs_daddr_t head_blk,
5654 xfs_daddr_t tail_blk)
5656 struct xfs_mount *mp = log->l_mp;
5661 trace_xfs_log_recover(log, head_blk, tail_blk);
5664 * First replay the images in the log.
5666 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5671 * If IO errors happened during recovery, bail out.
5673 if (XFS_FORCED_SHUTDOWN(mp)) {
5678 * We now update the tail_lsn since much of the recovery has completed
5679 * and there may be space available to use. If there were no extent
5680 * or iunlinks, we can free up the entire log and set the tail_lsn to
5681 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5682 * lsn of the last known good LR on disk. If there are extent frees
5683 * or iunlinks they will have some entries in the AIL; so we look at
5684 * the AIL to determine how to set the tail_lsn.
5686 xlog_assign_tail_lsn(mp);
5689 * Now that we've finished replaying all buffer and inode
5690 * updates, re-read in the superblock and reverify it.
5692 bp = xfs_getsb(mp, 0);
5693 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5694 ASSERT(!(bp->b_flags & XBF_WRITE));
5695 bp->b_flags |= XBF_READ;
5696 bp->b_ops = &xfs_sb_buf_ops;
5698 error = xfs_buf_submit(bp);
5700 if (!XFS_FORCED_SHUTDOWN(mp)) {
5701 xfs_buf_ioerror_alert(bp, __func__);
5708 /* Convert superblock from on-disk format */
5710 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5713 /* re-initialise in-core superblock and geometry structures */
5714 xfs_reinit_percpu_counters(mp);
5715 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5717 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5720 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5722 xlog_recover_check_summary(log);
5724 /* Normal transactions can now occur */
5725 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5730 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5732 * Return error or zero.
5738 xfs_daddr_t head_blk, tail_blk;
5741 /* find the tail of the log */
5742 error = xlog_find_tail(log, &head_blk, &tail_blk);
5747 * The superblock was read before the log was available and thus the LSN
5748 * could not be verified. Check the superblock LSN against the current
5749 * LSN now that it's known.
5751 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5752 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5755 if (tail_blk != head_blk) {
5756 /* There used to be a comment here:
5758 * disallow recovery on read-only mounts. note -- mount
5759 * checks for ENOSPC and turns it into an intelligent
5761 * ...but this is no longer true. Now, unless you specify
5762 * NORECOVERY (in which case this function would never be
5763 * called), we just go ahead and recover. We do this all
5764 * under the vfs layer, so we can get away with it unless
5765 * the device itself is read-only, in which case we fail.
5767 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5772 * Version 5 superblock log feature mask validation. We know the
5773 * log is dirty so check if there are any unknown log features
5774 * in what we need to recover. If there are unknown features
5775 * (e.g. unsupported transactions, then simply reject the
5776 * attempt at recovery before touching anything.
5778 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5779 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5780 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5782 "Superblock has unknown incompatible log features (0x%x) enabled.",
5783 (log->l_mp->m_sb.sb_features_log_incompat &
5784 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5786 "The log can not be fully and/or safely recovered by this kernel.");
5788 "Please recover the log on a kernel that supports the unknown features.");
5793 * Delay log recovery if the debug hook is set. This is debug
5794 * instrumention to coordinate simulation of I/O failures with
5797 if (xfs_globals.log_recovery_delay) {
5798 xfs_notice(log->l_mp,
5799 "Delaying log recovery for %d seconds.",
5800 xfs_globals.log_recovery_delay);
5801 msleep(xfs_globals.log_recovery_delay * 1000);
5804 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5805 log->l_mp->m_logname ? log->l_mp->m_logname
5808 error = xlog_do_recover(log, head_blk, tail_blk);
5809 log->l_flags |= XLOG_RECOVERY_NEEDED;
5815 * In the first part of recovery we replay inodes and buffers and build
5816 * up the list of extent free items which need to be processed. Here
5817 * we process the extent free items and clean up the on disk unlinked
5818 * inode lists. This is separated from the first part of recovery so
5819 * that the root and real-time bitmap inodes can be read in from disk in
5820 * between the two stages. This is necessary so that we can free space
5821 * in the real-time portion of the file system.
5824 xlog_recover_finish(
5828 * Now we're ready to do the transactions needed for the
5829 * rest of recovery. Start with completing all the extent
5830 * free intent records and then process the unlinked inode
5831 * lists. At this point, we essentially run in normal mode
5832 * except that we're still performing recovery actions
5833 * rather than accepting new requests.
5835 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5837 error = xlog_recover_process_intents(log);
5839 xfs_alert(log->l_mp, "Failed to recover intents");
5844 * Sync the log to get all the intents out of the AIL.
5845 * This isn't absolutely necessary, but it helps in
5846 * case the unlink transactions would have problems
5847 * pushing the intents out of the way.
5849 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5851 xlog_recover_process_iunlinks(log);
5853 xlog_recover_check_summary(log);
5855 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5856 log->l_mp->m_logname ? log->l_mp->m_logname
5858 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5860 xfs_info(log->l_mp, "Ending clean mount");
5866 xlog_recover_cancel(
5871 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5872 error = xlog_recover_cancel_intents(log);
5879 * Read all of the agf and agi counters and check that they
5880 * are consistent with the superblock counters.
5883 xlog_recover_check_summary(
5890 xfs_agnumber_t agno;
5901 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5902 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5904 xfs_alert(mp, "%s agf read failed agno %d error %d",
5905 __func__, agno, error);
5907 agfp = XFS_BUF_TO_AGF(agfbp);
5908 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5909 be32_to_cpu(agfp->agf_flcount);
5910 xfs_buf_relse(agfbp);
5913 error = xfs_read_agi(mp, NULL, agno, &agibp);
5915 xfs_alert(mp, "%s agi read failed agno %d error %d",
5916 __func__, agno, error);
5918 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5920 itotal += be32_to_cpu(agi->agi_count);
5921 ifree += be32_to_cpu(agi->agi_freecount);
5922 xfs_buf_relse(agibp);