2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
33 #include "xfs_log_priv.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_alloc.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_quota.h"
41 #include "xfs_cksum.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_bmap_btree.h"
45 #include "xfs_error.h"
47 #include "xfs_rmap_item.h"
48 #include "xfs_buf_item.h"
49 #include "xfs_refcount_item.h"
50 #include "xfs_bmap_item.h"
52 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
59 xlog_clear_stale_blocks(
64 xlog_recover_check_summary(
67 #define xlog_recover_check_summary(log)
70 xlog_do_recovery_pass(
71 struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
74 * This structure is used during recovery to record the buf log items which
75 * have been canceled and should not be replayed.
77 struct xfs_buf_cancel {
81 struct list_head bc_list;
85 * Sector aligned buffer routines for buffer create/read/write/access
89 * Verify the given count of basic blocks is valid number of blocks
90 * to specify for an operation involving the given XFS log buffer.
91 * Returns nonzero if the count is valid, 0 otherwise.
95 xlog_buf_bbcount_valid(
99 return bbcount > 0 && bbcount <= log->l_logBBsize;
103 * Allocate a buffer to hold log data. The buffer needs to be able
104 * to map to a range of nbblks basic blocks at any valid (basic
105 * block) offset within the log.
114 if (!xlog_buf_bbcount_valid(log, nbblks)) {
115 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
117 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
122 * We do log I/O in units of log sectors (a power-of-2
123 * multiple of the basic block size), so we round up the
124 * requested size to accommodate the basic blocks required
125 * for complete log sectors.
127 * In addition, the buffer may be used for a non-sector-
128 * aligned block offset, in which case an I/O of the
129 * requested size could extend beyond the end of the
130 * buffer. If the requested size is only 1 basic block it
131 * will never straddle a sector boundary, so this won't be
132 * an issue. Nor will this be a problem if the log I/O is
133 * done in basic blocks (sector size 1). But otherwise we
134 * extend the buffer by one extra log sector to ensure
135 * there's space to accommodate this possibility.
137 if (nbblks > 1 && log->l_sectBBsize > 1)
138 nbblks += log->l_sectBBsize;
139 nbblks = round_up(nbblks, log->l_sectBBsize);
141 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
155 * Return the address of the start of the given block number's data
156 * in a log buffer. The buffer covers a log sector-aligned region.
165 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
167 ASSERT(offset + nbblks <= bp->b_length);
168 return bp->b_addr + BBTOB(offset);
173 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
184 if (!xlog_buf_bbcount_valid(log, nbblks)) {
185 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
187 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
188 return -EFSCORRUPTED;
191 blk_no = round_down(blk_no, log->l_sectBBsize);
192 nbblks = round_up(nbblks, log->l_sectBBsize);
195 ASSERT(nbblks <= bp->b_length);
197 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
198 bp->b_flags |= XBF_READ;
199 bp->b_io_length = nbblks;
202 error = xfs_buf_submit_wait(bp);
203 if (error && !XFS_FORCED_SHUTDOWN(log->l_mp))
204 xfs_buf_ioerror_alert(bp, __func__);
218 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
222 *offset = xlog_align(log, blk_no, nbblks, bp);
227 * Read at an offset into the buffer. Returns with the buffer in it's original
228 * state regardless of the result of the read.
233 xfs_daddr_t blk_no, /* block to read from */
234 int nbblks, /* blocks to read */
238 char *orig_offset = bp->b_addr;
239 int orig_len = BBTOB(bp->b_length);
242 error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
246 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
248 /* must reset buffer pointer even on error */
249 error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
256 * Write out the buffer at the given block for the given number of blocks.
257 * The buffer is kept locked across the write and is returned locked.
258 * This can only be used for synchronous log writes.
269 if (!xlog_buf_bbcount_valid(log, nbblks)) {
270 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
272 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
273 return -EFSCORRUPTED;
276 blk_no = round_down(blk_no, log->l_sectBBsize);
277 nbblks = round_up(nbblks, log->l_sectBBsize);
280 ASSERT(nbblks <= bp->b_length);
282 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
285 bp->b_io_length = nbblks;
288 error = xfs_bwrite(bp);
290 xfs_buf_ioerror_alert(bp, __func__);
297 * dump debug superblock and log record information
300 xlog_header_check_dump(
302 xlog_rec_header_t *head)
304 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d",
305 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
306 xfs_debug(mp, " log : uuid = %pU, fmt = %d",
307 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
310 #define xlog_header_check_dump(mp, head)
314 * check log record header for recovery
317 xlog_header_check_recover(
319 xlog_rec_header_t *head)
321 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
324 * IRIX doesn't write the h_fmt field and leaves it zeroed
325 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
326 * a dirty log created in IRIX.
328 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
330 "dirty log written in incompatible format - can't recover");
331 xlog_header_check_dump(mp, head);
332 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
333 XFS_ERRLEVEL_HIGH, mp);
334 return -EFSCORRUPTED;
335 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
337 "dirty log entry has mismatched uuid - can't recover");
338 xlog_header_check_dump(mp, head);
339 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
340 XFS_ERRLEVEL_HIGH, mp);
341 return -EFSCORRUPTED;
347 * read the head block of the log and check the header
350 xlog_header_check_mount(
352 xlog_rec_header_t *head)
354 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
356 if (uuid_is_null(&head->h_fs_uuid)) {
358 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
359 * h_fs_uuid is null, we assume this log was last mounted
360 * by IRIX and continue.
362 xfs_warn(mp, "null uuid in log - IRIX style log");
363 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
364 xfs_warn(mp, "log has mismatched uuid - can't recover");
365 xlog_header_check_dump(mp, head);
366 XFS_ERROR_REPORT("xlog_header_check_mount",
367 XFS_ERRLEVEL_HIGH, mp);
368 return -EFSCORRUPTED;
379 * We're not going to bother about retrying
380 * this during recovery. One strike!
382 if (!XFS_FORCED_SHUTDOWN(bp->b_target->bt_mount)) {
383 xfs_buf_ioerror_alert(bp, __func__);
384 xfs_force_shutdown(bp->b_target->bt_mount,
385 SHUTDOWN_META_IO_ERROR);
390 * On v5 supers, a bli could be attached to update the metadata LSN.
394 xfs_buf_item_relse(bp);
395 ASSERT(bp->b_fspriv == NULL);
402 * This routine finds (to an approximation) the first block in the physical
403 * log which contains the given cycle. It uses a binary search algorithm.
404 * Note that the algorithm can not be perfect because the disk will not
405 * necessarily be perfect.
408 xlog_find_cycle_start(
411 xfs_daddr_t first_blk,
412 xfs_daddr_t *last_blk,
422 mid_blk = BLK_AVG(first_blk, end_blk);
423 while (mid_blk != first_blk && mid_blk != end_blk) {
424 error = xlog_bread(log, mid_blk, 1, bp, &offset);
427 mid_cycle = xlog_get_cycle(offset);
428 if (mid_cycle == cycle)
429 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
431 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
432 mid_blk = BLK_AVG(first_blk, end_blk);
434 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
435 (mid_blk == end_blk && mid_blk-1 == first_blk));
443 * Check that a range of blocks does not contain stop_on_cycle_no.
444 * Fill in *new_blk with the block offset where such a block is
445 * found, or with -1 (an invalid block number) if there is no such
446 * block in the range. The scan needs to occur from front to back
447 * and the pointer into the region must be updated since a later
448 * routine will need to perform another test.
451 xlog_find_verify_cycle(
453 xfs_daddr_t start_blk,
455 uint stop_on_cycle_no,
456 xfs_daddr_t *new_blk)
466 * Greedily allocate a buffer big enough to handle the full
467 * range of basic blocks we'll be examining. If that fails,
468 * try a smaller size. We need to be able to read at least
469 * a log sector, or we're out of luck.
471 bufblks = 1 << ffs(nbblks);
472 while (bufblks > log->l_logBBsize)
474 while (!(bp = xlog_get_bp(log, bufblks))) {
476 if (bufblks < log->l_sectBBsize)
480 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
483 bcount = min(bufblks, (start_blk + nbblks - i));
485 error = xlog_bread(log, i, bcount, bp, &buf);
489 for (j = 0; j < bcount; j++) {
490 cycle = xlog_get_cycle(buf);
491 if (cycle == stop_on_cycle_no) {
508 * Potentially backup over partial log record write.
510 * In the typical case, last_blk is the number of the block directly after
511 * a good log record. Therefore, we subtract one to get the block number
512 * of the last block in the given buffer. extra_bblks contains the number
513 * of blocks we would have read on a previous read. This happens when the
514 * last log record is split over the end of the physical log.
516 * extra_bblks is the number of blocks potentially verified on a previous
517 * call to this routine.
520 xlog_find_verify_log_record(
522 xfs_daddr_t start_blk,
523 xfs_daddr_t *last_blk,
529 xlog_rec_header_t *head = NULL;
532 int num_blks = *last_blk - start_blk;
535 ASSERT(start_blk != 0 || *last_blk != start_blk);
537 if (!(bp = xlog_get_bp(log, num_blks))) {
538 if (!(bp = xlog_get_bp(log, 1)))
542 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
545 offset += ((num_blks - 1) << BBSHIFT);
548 for (i = (*last_blk) - 1; i >= 0; i--) {
550 /* valid log record not found */
552 "Log inconsistent (didn't find previous header)");
559 error = xlog_bread(log, i, 1, bp, &offset);
564 head = (xlog_rec_header_t *)offset;
566 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
574 * We hit the beginning of the physical log & still no header. Return
575 * to caller. If caller can handle a return of -1, then this routine
576 * will be called again for the end of the physical log.
584 * We have the final block of the good log (the first block
585 * of the log record _before_ the head. So we check the uuid.
587 if ((error = xlog_header_check_mount(log->l_mp, head)))
591 * We may have found a log record header before we expected one.
592 * last_blk will be the 1st block # with a given cycle #. We may end
593 * up reading an entire log record. In this case, we don't want to
594 * reset last_blk. Only when last_blk points in the middle of a log
595 * record do we update last_blk.
597 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
598 uint h_size = be32_to_cpu(head->h_size);
600 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
601 if (h_size % XLOG_HEADER_CYCLE_SIZE)
607 if (*last_blk - i + extra_bblks !=
608 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
617 * Head is defined to be the point of the log where the next log write
618 * could go. This means that incomplete LR writes at the end are
619 * eliminated when calculating the head. We aren't guaranteed that previous
620 * LR have complete transactions. We only know that a cycle number of
621 * current cycle number -1 won't be present in the log if we start writing
622 * from our current block number.
624 * last_blk contains the block number of the first block with a given
627 * Return: zero if normal, non-zero if error.
632 xfs_daddr_t *return_head_blk)
636 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
638 uint first_half_cycle, last_half_cycle;
640 int error, log_bbnum = log->l_logBBsize;
642 /* Is the end of the log device zeroed? */
643 error = xlog_find_zeroed(log, &first_blk);
645 xfs_warn(log->l_mp, "empty log check failed");
649 *return_head_blk = first_blk;
651 /* Is the whole lot zeroed? */
653 /* Linux XFS shouldn't generate totally zeroed logs -
654 * mkfs etc write a dummy unmount record to a fresh
655 * log so we can store the uuid in there
657 xfs_warn(log->l_mp, "totally zeroed log");
663 first_blk = 0; /* get cycle # of 1st block */
664 bp = xlog_get_bp(log, 1);
668 error = xlog_bread(log, 0, 1, bp, &offset);
672 first_half_cycle = xlog_get_cycle(offset);
674 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
675 error = xlog_bread(log, last_blk, 1, bp, &offset);
679 last_half_cycle = xlog_get_cycle(offset);
680 ASSERT(last_half_cycle != 0);
683 * If the 1st half cycle number is equal to the last half cycle number,
684 * then the entire log is stamped with the same cycle number. In this
685 * case, head_blk can't be set to zero (which makes sense). The below
686 * math doesn't work out properly with head_blk equal to zero. Instead,
687 * we set it to log_bbnum which is an invalid block number, but this
688 * value makes the math correct. If head_blk doesn't changed through
689 * all the tests below, *head_blk is set to zero at the very end rather
690 * than log_bbnum. In a sense, log_bbnum and zero are the same block
691 * in a circular file.
693 if (first_half_cycle == last_half_cycle) {
695 * In this case we believe that the entire log should have
696 * cycle number last_half_cycle. We need to scan backwards
697 * from the end verifying that there are no holes still
698 * containing last_half_cycle - 1. If we find such a hole,
699 * then the start of that hole will be the new head. The
700 * simple case looks like
701 * x | x ... | x - 1 | x
702 * Another case that fits this picture would be
703 * x | x + 1 | x ... | x
704 * In this case the head really is somewhere at the end of the
705 * log, as one of the latest writes at the beginning was
708 * x | x + 1 | x ... | x - 1 | x
709 * This is really the combination of the above two cases, and
710 * the head has to end up at the start of the x-1 hole at the
713 * In the 256k log case, we will read from the beginning to the
714 * end of the log and search for cycle numbers equal to x-1.
715 * We don't worry about the x+1 blocks that we encounter,
716 * because we know that they cannot be the head since the log
719 head_blk = log_bbnum;
720 stop_on_cycle = last_half_cycle - 1;
723 * In this case we want to find the first block with cycle
724 * number matching last_half_cycle. We expect the log to be
726 * x + 1 ... | x ... | x
727 * The first block with cycle number x (last_half_cycle) will
728 * be where the new head belongs. First we do a binary search
729 * for the first occurrence of last_half_cycle. The binary
730 * search may not be totally accurate, so then we scan back
731 * from there looking for occurrences of last_half_cycle before
732 * us. If that backwards scan wraps around the beginning of
733 * the log, then we look for occurrences of last_half_cycle - 1
734 * at the end of the log. The cases we're looking for look
736 * v binary search stopped here
737 * x + 1 ... | x | x + 1 | x ... | x
738 * ^ but we want to locate this spot
740 * <---------> less than scan distance
741 * x + 1 ... | x ... | x - 1 | x
742 * ^ we want to locate this spot
744 stop_on_cycle = last_half_cycle;
745 if ((error = xlog_find_cycle_start(log, bp, first_blk,
746 &head_blk, last_half_cycle)))
751 * Now validate the answer. Scan back some number of maximum possible
752 * blocks and make sure each one has the expected cycle number. The
753 * maximum is determined by the total possible amount of buffering
754 * in the in-core log. The following number can be made tighter if
755 * we actually look at the block size of the filesystem.
757 num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
758 if (head_blk >= num_scan_bblks) {
760 * We are guaranteed that the entire check can be performed
763 start_blk = head_blk - num_scan_bblks;
764 if ((error = xlog_find_verify_cycle(log,
765 start_blk, num_scan_bblks,
766 stop_on_cycle, &new_blk)))
770 } else { /* need to read 2 parts of log */
772 * We are going to scan backwards in the log in two parts.
773 * First we scan the physical end of the log. In this part
774 * of the log, we are looking for blocks with cycle number
775 * last_half_cycle - 1.
776 * If we find one, then we know that the log starts there, as
777 * we've found a hole that didn't get written in going around
778 * the end of the physical log. The simple case for this is
779 * x + 1 ... | x ... | x - 1 | x
780 * <---------> less than scan distance
781 * If all of the blocks at the end of the log have cycle number
782 * last_half_cycle, then we check the blocks at the start of
783 * the log looking for occurrences of last_half_cycle. If we
784 * find one, then our current estimate for the location of the
785 * first occurrence of last_half_cycle is wrong and we move
786 * back to the hole we've found. This case looks like
787 * x + 1 ... | x | x + 1 | x ...
788 * ^ binary search stopped here
789 * Another case we need to handle that only occurs in 256k
791 * x + 1 ... | x ... | x+1 | x ...
792 * ^ binary search stops here
793 * In a 256k log, the scan at the end of the log will see the
794 * x + 1 blocks. We need to skip past those since that is
795 * certainly not the head of the log. By searching for
796 * last_half_cycle-1 we accomplish that.
798 ASSERT(head_blk <= INT_MAX &&
799 (xfs_daddr_t) num_scan_bblks >= head_blk);
800 start_blk = log_bbnum - (num_scan_bblks - head_blk);
801 if ((error = xlog_find_verify_cycle(log, start_blk,
802 num_scan_bblks - (int)head_blk,
803 (stop_on_cycle - 1), &new_blk)))
811 * Scan beginning of log now. The last part of the physical
812 * log is good. This scan needs to verify that it doesn't find
813 * the last_half_cycle.
816 ASSERT(head_blk <= INT_MAX);
817 if ((error = xlog_find_verify_cycle(log,
818 start_blk, (int)head_blk,
819 stop_on_cycle, &new_blk)))
827 * Now we need to make sure head_blk is not pointing to a block in
828 * the middle of a log record.
830 num_scan_bblks = XLOG_REC_SHIFT(log);
831 if (head_blk >= num_scan_bblks) {
832 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
834 /* start ptr at last block ptr before head_blk */
835 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
842 ASSERT(head_blk <= INT_MAX);
843 error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
847 /* We hit the beginning of the log during our search */
848 start_blk = log_bbnum - (num_scan_bblks - head_blk);
850 ASSERT(start_blk <= INT_MAX &&
851 (xfs_daddr_t) log_bbnum-start_blk >= 0);
852 ASSERT(head_blk <= INT_MAX);
853 error = xlog_find_verify_log_record(log, start_blk,
854 &new_blk, (int)head_blk);
859 if (new_blk != log_bbnum)
866 if (head_blk == log_bbnum)
867 *return_head_blk = 0;
869 *return_head_blk = head_blk;
871 * When returning here, we have a good block number. Bad block
872 * means that during a previous crash, we didn't have a clean break
873 * from cycle number N to cycle number N-1. In this case, we need
874 * to find the first block with cycle number N-1.
882 xfs_warn(log->l_mp, "failed to find log head");
887 * Seek backwards in the log for log record headers.
889 * Given a starting log block, walk backwards until we find the provided number
890 * of records or hit the provided tail block. The return value is the number of
891 * records encountered or a negative error code. The log block and buffer
892 * pointer of the last record seen are returned in rblk and rhead respectively.
895 xlog_rseek_logrec_hdr(
897 xfs_daddr_t head_blk,
898 xfs_daddr_t tail_blk,
902 struct xlog_rec_header **rhead,
914 * Walk backwards from the head block until we hit the tail or the first
917 end_blk = head_blk > tail_blk ? tail_blk : 0;
918 for (i = (int) head_blk - 1; i >= end_blk; i--) {
919 error = xlog_bread(log, i, 1, bp, &offset);
923 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
925 *rhead = (struct xlog_rec_header *) offset;
926 if (++found == count)
932 * If we haven't hit the tail block or the log record header count,
933 * start looking again from the end of the physical log. Note that
934 * callers can pass head == tail if the tail is not yet known.
936 if (tail_blk >= head_blk && found != count) {
937 for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
938 error = xlog_bread(log, i, 1, bp, &offset);
942 if (*(__be32 *)offset ==
943 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
946 *rhead = (struct xlog_rec_header *) offset;
947 if (++found == count)
960 * Seek forward in the log for log record headers.
962 * Given head and tail blocks, walk forward from the tail block until we find
963 * the provided number of records or hit the head block. The return value is the
964 * number of records encountered or a negative error code. The log block and
965 * buffer pointer of the last record seen are returned in rblk and rhead
969 xlog_seek_logrec_hdr(
971 xfs_daddr_t head_blk,
972 xfs_daddr_t tail_blk,
976 struct xlog_rec_header **rhead,
988 * Walk forward from the tail block until we hit the head or the last
991 end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
992 for (i = (int) tail_blk; i <= end_blk; i++) {
993 error = xlog_bread(log, i, 1, bp, &offset);
997 if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
999 *rhead = (struct xlog_rec_header *) offset;
1000 if (++found == count)
1006 * If we haven't hit the head block or the log record header count,
1007 * start looking again from the start of the physical log.
1009 if (tail_blk > head_blk && found != count) {
1010 for (i = 0; i < (int) head_blk; i++) {
1011 error = xlog_bread(log, i, 1, bp, &offset);
1015 if (*(__be32 *)offset ==
1016 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
1019 *rhead = (struct xlog_rec_header *) offset;
1020 if (++found == count)
1033 * Calculate distance from head to tail (i.e., unused space in the log).
1038 xfs_daddr_t head_blk,
1039 xfs_daddr_t tail_blk)
1041 if (head_blk < tail_blk)
1042 return tail_blk - head_blk;
1044 return tail_blk + (log->l_logBBsize - head_blk);
1048 * Verify the log tail. This is particularly important when torn or incomplete
1049 * writes have been detected near the front of the log and the head has been
1050 * walked back accordingly.
1052 * We also have to handle the case where the tail was pinned and the head
1053 * blocked behind the tail right before a crash. If the tail had been pushed
1054 * immediately prior to the crash and the subsequent checkpoint was only
1055 * partially written, it's possible it overwrote the last referenced tail in the
1056 * log with garbage. This is not a coherency problem because the tail must have
1057 * been pushed before it can be overwritten, but appears as log corruption to
1058 * recovery because we have no way to know the tail was updated if the
1059 * subsequent checkpoint didn't write successfully.
1061 * Therefore, CRC check the log from tail to head. If a failure occurs and the
1062 * offending record is within max iclog bufs from the head, walk the tail
1063 * forward and retry until a valid tail is found or corruption is detected out
1064 * of the range of a possible overwrite.
1069 xfs_daddr_t head_blk,
1070 xfs_daddr_t *tail_blk,
1073 struct xlog_rec_header *thead;
1075 xfs_daddr_t first_bad;
1078 xfs_daddr_t tmp_tail;
1079 xfs_daddr_t orig_tail = *tail_blk;
1081 bp = xlog_get_bp(log, 1);
1086 * Make sure the tail points to a record (returns positive count on
1089 error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, bp,
1090 &tmp_tail, &thead, &wrapped);
1093 if (*tail_blk != tmp_tail)
1094 *tail_blk = tmp_tail;
1097 * Run a CRC check from the tail to the head. We can't just check
1098 * MAX_ICLOGS records past the tail because the tail may point to stale
1099 * blocks cleared during the search for the head/tail. These blocks are
1100 * overwritten with zero-length records and thus record count is not a
1101 * reliable indicator of the iclog state before a crash.
1104 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1105 XLOG_RECOVER_CRCPASS, &first_bad);
1106 while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1110 * Is corruption within range of the head? If so, retry from
1111 * the next record. Otherwise return an error.
1113 tail_distance = xlog_tail_distance(log, head_blk, first_bad);
1114 if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
1117 /* skip to the next record; returns positive count on success */
1118 error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2, bp,
1119 &tmp_tail, &thead, &wrapped);
1123 *tail_blk = tmp_tail;
1125 error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
1126 XLOG_RECOVER_CRCPASS, &first_bad);
1129 if (!error && *tail_blk != orig_tail)
1131 "Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1132 orig_tail, *tail_blk);
1139 * Detect and trim torn writes from the head of the log.
1141 * Storage without sector atomicity guarantees can result in torn writes in the
1142 * log in the event of a crash. Our only means to detect this scenario is via
1143 * CRC verification. While we can't always be certain that CRC verification
1144 * failure is due to a torn write vs. an unrelated corruption, we do know that
1145 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1146 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1147 * the log and treat failures in this range as torn writes as a matter of
1148 * policy. In the event of CRC failure, the head is walked back to the last good
1149 * record in the log and the tail is updated from that record and verified.
1154 xfs_daddr_t *head_blk, /* in/out: unverified head */
1155 xfs_daddr_t *tail_blk, /* out: tail block */
1157 xfs_daddr_t *rhead_blk, /* start blk of last record */
1158 struct xlog_rec_header **rhead, /* ptr to last record */
1159 bool *wrapped) /* last rec. wraps phys. log */
1161 struct xlog_rec_header *tmp_rhead;
1162 struct xfs_buf *tmp_bp;
1163 xfs_daddr_t first_bad;
1164 xfs_daddr_t tmp_rhead_blk;
1170 * Check the head of the log for torn writes. Search backwards from the
1171 * head until we hit the tail or the maximum number of log record I/Os
1172 * that could have been in flight at one time. Use a temporary buffer so
1173 * we don't trash the rhead/bp pointers from the caller.
1175 tmp_bp = xlog_get_bp(log, 1);
1178 error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1179 XLOG_MAX_ICLOGS, tmp_bp, &tmp_rhead_blk,
1180 &tmp_rhead, &tmp_wrapped);
1181 xlog_put_bp(tmp_bp);
1186 * Now run a CRC verification pass over the records starting at the
1187 * block found above to the current head. If a CRC failure occurs, the
1188 * log block of the first bad record is saved in first_bad.
1190 error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1191 XLOG_RECOVER_CRCPASS, &first_bad);
1192 if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1194 * We've hit a potential torn write. Reset the error and warn
1199 "Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1200 first_bad, *head_blk);
1203 * Get the header block and buffer pointer for the last good
1204 * record before the bad record.
1206 * Note that xlog_find_tail() clears the blocks at the new head
1207 * (i.e., the records with invalid CRC) if the cycle number
1208 * matches the the current cycle.
1210 found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1, bp,
1211 rhead_blk, rhead, wrapped);
1214 if (found == 0) /* XXX: right thing to do here? */
1218 * Reset the head block to the starting block of the first bad
1219 * log record and set the tail block based on the last good
1222 * Bail out if the updated head/tail match as this indicates
1223 * possible corruption outside of the acceptable
1224 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1226 *head_blk = first_bad;
1227 *tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1228 if (*head_blk == *tail_blk) {
1236 return xlog_verify_tail(log, *head_blk, tail_blk,
1237 be32_to_cpu((*rhead)->h_size));
1241 * Check whether the head of the log points to an unmount record. In other
1242 * words, determine whether the log is clean. If so, update the in-core state
1246 xlog_check_unmount_rec(
1248 xfs_daddr_t *head_blk,
1249 xfs_daddr_t *tail_blk,
1250 struct xlog_rec_header *rhead,
1251 xfs_daddr_t rhead_blk,
1255 struct xlog_op_header *op_head;
1256 xfs_daddr_t umount_data_blk;
1257 xfs_daddr_t after_umount_blk;
1265 * Look for unmount record. If we find it, then we know there was a
1266 * clean unmount. Since 'i' could be the last block in the physical
1267 * log, we convert to a log block before comparing to the head_blk.
1269 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1270 * below. We won't want to clear the unmount record if there is one, so
1271 * we pass the lsn of the unmount record rather than the block after it.
1273 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1274 int h_size = be32_to_cpu(rhead->h_size);
1275 int h_version = be32_to_cpu(rhead->h_version);
1277 if ((h_version & XLOG_VERSION_2) &&
1278 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1279 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1280 if (h_size % XLOG_HEADER_CYCLE_SIZE)
1288 after_umount_blk = rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len));
1289 after_umount_blk = do_mod(after_umount_blk, log->l_logBBsize);
1290 if (*head_blk == after_umount_blk &&
1291 be32_to_cpu(rhead->h_num_logops) == 1) {
1292 umount_data_blk = rhead_blk + hblks;
1293 umount_data_blk = do_mod(umount_data_blk, log->l_logBBsize);
1294 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1298 op_head = (struct xlog_op_header *)offset;
1299 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1301 * Set tail and last sync so that newly written log
1302 * records will point recovery to after the current
1305 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1306 log->l_curr_cycle, after_umount_blk);
1307 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1308 log->l_curr_cycle, after_umount_blk);
1309 *tail_blk = after_umount_blk;
1321 xfs_daddr_t head_blk,
1322 struct xlog_rec_header *rhead,
1323 xfs_daddr_t rhead_blk,
1327 * Reset log values according to the state of the log when we
1328 * crashed. In the case where head_blk == 0, we bump curr_cycle
1329 * one because the next write starts a new cycle rather than
1330 * continuing the cycle of the last good log record. At this
1331 * point we have guaranteed that all partial log records have been
1332 * accounted for. Therefore, we know that the last good log record
1333 * written was complete and ended exactly on the end boundary
1334 * of the physical log.
1336 log->l_prev_block = rhead_blk;
1337 log->l_curr_block = (int)head_blk;
1338 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1340 log->l_curr_cycle++;
1341 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1342 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
1343 xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
1344 BBTOB(log->l_curr_block));
1345 xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
1346 BBTOB(log->l_curr_block));
1350 * Find the sync block number or the tail of the log.
1352 * This will be the block number of the last record to have its
1353 * associated buffers synced to disk. Every log record header has
1354 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
1355 * to get a sync block number. The only concern is to figure out which
1356 * log record header to believe.
1358 * The following algorithm uses the log record header with the largest
1359 * lsn. The entire log record does not need to be valid. We only care
1360 * that the header is valid.
1362 * We could speed up search by using current head_blk buffer, but it is not
1368 xfs_daddr_t *head_blk,
1369 xfs_daddr_t *tail_blk)
1371 xlog_rec_header_t *rhead;
1372 char *offset = NULL;
1375 xfs_daddr_t rhead_blk;
1377 bool wrapped = false;
1381 * Find previous log record
1383 if ((error = xlog_find_head(log, head_blk)))
1385 ASSERT(*head_blk < INT_MAX);
1387 bp = xlog_get_bp(log, 1);
1390 if (*head_blk == 0) { /* special case */
1391 error = xlog_bread(log, 0, 1, bp, &offset);
1395 if (xlog_get_cycle(offset) == 0) {
1397 /* leave all other log inited values alone */
1403 * Search backwards through the log looking for the log record header
1404 * block. This wraps all the way back around to the head so something is
1405 * seriously wrong if we can't find it.
1407 error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, bp,
1408 &rhead_blk, &rhead, &wrapped);
1412 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1415 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1418 * Set the log state based on the current head record.
1420 xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1421 tail_lsn = atomic64_read(&log->l_tail_lsn);
1424 * Look for an unmount record at the head of the log. This sets the log
1425 * state to determine whether recovery is necessary.
1427 error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1428 rhead_blk, bp, &clean);
1433 * Verify the log head if the log is not clean (e.g., we have anything
1434 * but an unmount record at the head). This uses CRC verification to
1435 * detect and trim torn writes. If discovered, CRC failures are
1436 * considered torn writes and the log head is trimmed accordingly.
1438 * Note that we can only run CRC verification when the log is dirty
1439 * because there's no guarantee that the log data behind an unmount
1440 * record is compatible with the current architecture.
1443 xfs_daddr_t orig_head = *head_blk;
1445 error = xlog_verify_head(log, head_blk, tail_blk, bp,
1446 &rhead_blk, &rhead, &wrapped);
1450 /* update in-core state again if the head changed */
1451 if (*head_blk != orig_head) {
1452 xlog_set_state(log, *head_blk, rhead, rhead_blk,
1454 tail_lsn = atomic64_read(&log->l_tail_lsn);
1455 error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1456 rhead, rhead_blk, bp,
1464 * Note that the unmount was clean. If the unmount was not clean, we
1465 * need to know this to rebuild the superblock counters from the perag
1466 * headers if we have a filesystem using non-persistent counters.
1469 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1472 * Make sure that there are no blocks in front of the head
1473 * with the same cycle number as the head. This can happen
1474 * because we allow multiple outstanding log writes concurrently,
1475 * and the later writes might make it out before earlier ones.
1477 * We use the lsn from before modifying it so that we'll never
1478 * overwrite the unmount record after a clean unmount.
1480 * Do this only if we are going to recover the filesystem
1482 * NOTE: This used to say "if (!readonly)"
1483 * However on Linux, we can & do recover a read-only filesystem.
1484 * We only skip recovery if NORECOVERY is specified on mount,
1485 * in which case we would not be here.
1487 * But... if the -device- itself is readonly, just skip this.
1488 * We can't recover this device anyway, so it won't matter.
1490 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1491 error = xlog_clear_stale_blocks(log, tail_lsn);
1497 xfs_warn(log->l_mp, "failed to locate log tail");
1502 * Is the log zeroed at all?
1504 * The last binary search should be changed to perform an X block read
1505 * once X becomes small enough. You can then search linearly through
1506 * the X blocks. This will cut down on the number of reads we need to do.
1508 * If the log is partially zeroed, this routine will pass back the blkno
1509 * of the first block with cycle number 0. It won't have a complete LR
1513 * 0 => the log is completely written to
1514 * 1 => use *blk_no as the first block of the log
1515 * <0 => error has occurred
1520 xfs_daddr_t *blk_no)
1524 uint first_cycle, last_cycle;
1525 xfs_daddr_t new_blk, last_blk, start_blk;
1526 xfs_daddr_t num_scan_bblks;
1527 int error, log_bbnum = log->l_logBBsize;
1531 /* check totally zeroed log */
1532 bp = xlog_get_bp(log, 1);
1535 error = xlog_bread(log, 0, 1, bp, &offset);
1539 first_cycle = xlog_get_cycle(offset);
1540 if (first_cycle == 0) { /* completely zeroed log */
1546 /* check partially zeroed log */
1547 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1551 last_cycle = xlog_get_cycle(offset);
1552 if (last_cycle != 0) { /* log completely written to */
1555 } else if (first_cycle != 1) {
1557 * If the cycle of the last block is zero, the cycle of
1558 * the first block must be 1. If it's not, maybe we're
1559 * not looking at a log... Bail out.
1562 "Log inconsistent or not a log (last==0, first!=1)");
1567 /* we have a partially zeroed log */
1568 last_blk = log_bbnum-1;
1569 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1573 * Validate the answer. Because there is no way to guarantee that
1574 * the entire log is made up of log records which are the same size,
1575 * we scan over the defined maximum blocks. At this point, the maximum
1576 * is not chosen to mean anything special. XXXmiken
1578 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1579 ASSERT(num_scan_bblks <= INT_MAX);
1581 if (last_blk < num_scan_bblks)
1582 num_scan_bblks = last_blk;
1583 start_blk = last_blk - num_scan_bblks;
1586 * We search for any instances of cycle number 0 that occur before
1587 * our current estimate of the head. What we're trying to detect is
1588 * 1 ... | 0 | 1 | 0...
1589 * ^ binary search ends here
1591 if ((error = xlog_find_verify_cycle(log, start_blk,
1592 (int)num_scan_bblks, 0, &new_blk)))
1598 * Potentially backup over partial log record write. We don't need
1599 * to search the end of the log because we know it is zero.
1601 error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1616 * These are simple subroutines used by xlog_clear_stale_blocks() below
1617 * to initialize a buffer full of empty log record headers and write
1618 * them into the log.
1629 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1631 memset(buf, 0, BBSIZE);
1632 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1633 recp->h_cycle = cpu_to_be32(cycle);
1634 recp->h_version = cpu_to_be32(
1635 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1636 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1637 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1638 recp->h_fmt = cpu_to_be32(XLOG_FMT);
1639 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1643 xlog_write_log_records(
1654 int sectbb = log->l_sectBBsize;
1655 int end_block = start_block + blocks;
1661 * Greedily allocate a buffer big enough to handle the full
1662 * range of basic blocks to be written. If that fails, try
1663 * a smaller size. We need to be able to write at least a
1664 * log sector, or we're out of luck.
1666 bufblks = 1 << ffs(blocks);
1667 while (bufblks > log->l_logBBsize)
1669 while (!(bp = xlog_get_bp(log, bufblks))) {
1671 if (bufblks < sectbb)
1675 /* We may need to do a read at the start to fill in part of
1676 * the buffer in the starting sector not covered by the first
1679 balign = round_down(start_block, sectbb);
1680 if (balign != start_block) {
1681 error = xlog_bread_noalign(log, start_block, 1, bp);
1685 j = start_block - balign;
1688 for (i = start_block; i < end_block; i += bufblks) {
1689 int bcount, endcount;
1691 bcount = min(bufblks, end_block - start_block);
1692 endcount = bcount - j;
1694 /* We may need to do a read at the end to fill in part of
1695 * the buffer in the final sector not covered by the write.
1696 * If this is the same sector as the above read, skip it.
1698 ealign = round_down(end_block, sectbb);
1699 if (j == 0 && (start_block + endcount > ealign)) {
1700 offset = bp->b_addr + BBTOB(ealign - start_block);
1701 error = xlog_bread_offset(log, ealign, sectbb,
1708 offset = xlog_align(log, start_block, endcount, bp);
1709 for (; j < endcount; j++) {
1710 xlog_add_record(log, offset, cycle, i+j,
1711 tail_cycle, tail_block);
1714 error = xlog_bwrite(log, start_block, endcount, bp);
1717 start_block += endcount;
1727 * This routine is called to blow away any incomplete log writes out
1728 * in front of the log head. We do this so that we won't become confused
1729 * if we come up, write only a little bit more, and then crash again.
1730 * If we leave the partial log records out there, this situation could
1731 * cause us to think those partial writes are valid blocks since they
1732 * have the current cycle number. We get rid of them by overwriting them
1733 * with empty log records with the old cycle number rather than the
1736 * The tail lsn is passed in rather than taken from
1737 * the log so that we will not write over the unmount record after a
1738 * clean unmount in a 512 block log. Doing so would leave the log without
1739 * any valid log records in it until a new one was written. If we crashed
1740 * during that time we would not be able to recover.
1743 xlog_clear_stale_blocks(
1747 int tail_cycle, head_cycle;
1748 int tail_block, head_block;
1749 int tail_distance, max_distance;
1753 tail_cycle = CYCLE_LSN(tail_lsn);
1754 tail_block = BLOCK_LSN(tail_lsn);
1755 head_cycle = log->l_curr_cycle;
1756 head_block = log->l_curr_block;
1759 * Figure out the distance between the new head of the log
1760 * and the tail. We want to write over any blocks beyond the
1761 * head that we may have written just before the crash, but
1762 * we don't want to overwrite the tail of the log.
1764 if (head_cycle == tail_cycle) {
1766 * The tail is behind the head in the physical log,
1767 * so the distance from the head to the tail is the
1768 * distance from the head to the end of the log plus
1769 * the distance from the beginning of the log to the
1772 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1773 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1774 XFS_ERRLEVEL_LOW, log->l_mp);
1775 return -EFSCORRUPTED;
1777 tail_distance = tail_block + (log->l_logBBsize - head_block);
1780 * The head is behind the tail in the physical log,
1781 * so the distance from the head to the tail is just
1782 * the tail block minus the head block.
1784 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1785 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1786 XFS_ERRLEVEL_LOW, log->l_mp);
1787 return -EFSCORRUPTED;
1789 tail_distance = tail_block - head_block;
1793 * If the head is right up against the tail, we can't clear
1796 if (tail_distance <= 0) {
1797 ASSERT(tail_distance == 0);
1801 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1803 * Take the smaller of the maximum amount of outstanding I/O
1804 * we could have and the distance to the tail to clear out.
1805 * We take the smaller so that we don't overwrite the tail and
1806 * we don't waste all day writing from the head to the tail
1809 max_distance = MIN(max_distance, tail_distance);
1811 if ((head_block + max_distance) <= log->l_logBBsize) {
1813 * We can stomp all the blocks we need to without
1814 * wrapping around the end of the log. Just do it
1815 * in a single write. Use the cycle number of the
1816 * current cycle minus one so that the log will look like:
1819 error = xlog_write_log_records(log, (head_cycle - 1),
1820 head_block, max_distance, tail_cycle,
1826 * We need to wrap around the end of the physical log in
1827 * order to clear all the blocks. Do it in two separate
1828 * I/Os. The first write should be from the head to the
1829 * end of the physical log, and it should use the current
1830 * cycle number minus one just like above.
1832 distance = log->l_logBBsize - head_block;
1833 error = xlog_write_log_records(log, (head_cycle - 1),
1834 head_block, distance, tail_cycle,
1841 * Now write the blocks at the start of the physical log.
1842 * This writes the remainder of the blocks we want to clear.
1843 * It uses the current cycle number since we're now on the
1844 * same cycle as the head so that we get:
1845 * n ... n ... | n - 1 ...
1846 * ^^^^^ blocks we're writing
1848 distance = max_distance - (log->l_logBBsize - head_block);
1849 error = xlog_write_log_records(log, head_cycle, 0, distance,
1850 tail_cycle, tail_block);
1858 /******************************************************************************
1860 * Log recover routines
1862 ******************************************************************************
1866 * Sort the log items in the transaction.
1868 * The ordering constraints are defined by the inode allocation and unlink
1869 * behaviour. The rules are:
1871 * 1. Every item is only logged once in a given transaction. Hence it
1872 * represents the last logged state of the item. Hence ordering is
1873 * dependent on the order in which operations need to be performed so
1874 * required initial conditions are always met.
1876 * 2. Cancelled buffers are recorded in pass 1 in a separate table and
1877 * there's nothing to replay from them so we can simply cull them
1878 * from the transaction. However, we can't do that until after we've
1879 * replayed all the other items because they may be dependent on the
1880 * cancelled buffer and replaying the cancelled buffer can remove it
1881 * form the cancelled buffer table. Hence they have tobe done last.
1883 * 3. Inode allocation buffers must be replayed before inode items that
1884 * read the buffer and replay changes into it. For filesystems using the
1885 * ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1886 * treated the same as inode allocation buffers as they create and
1887 * initialise the buffers directly.
1889 * 4. Inode unlink buffers must be replayed after inode items are replayed.
1890 * This ensures that inodes are completely flushed to the inode buffer
1891 * in a "free" state before we remove the unlinked inode list pointer.
1893 * Hence the ordering needs to be inode allocation buffers first, inode items
1894 * second, inode unlink buffers third and cancelled buffers last.
1896 * But there's a problem with that - we can't tell an inode allocation buffer
1897 * apart from a regular buffer, so we can't separate them. We can, however,
1898 * tell an inode unlink buffer from the others, and so we can separate them out
1899 * from all the other buffers and move them to last.
1901 * Hence, 4 lists, in order from head to tail:
1902 * - buffer_list for all buffers except cancelled/inode unlink buffers
1903 * - item_list for all non-buffer items
1904 * - inode_buffer_list for inode unlink buffers
1905 * - cancel_list for the cancelled buffers
1907 * Note that we add objects to the tail of the lists so that first-to-last
1908 * ordering is preserved within the lists. Adding objects to the head of the
1909 * list means when we traverse from the head we walk them in last-to-first
1910 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1911 * but for all other items there may be specific ordering that we need to
1915 xlog_recover_reorder_trans(
1917 struct xlog_recover *trans,
1920 xlog_recover_item_t *item, *n;
1922 LIST_HEAD(sort_list);
1923 LIST_HEAD(cancel_list);
1924 LIST_HEAD(buffer_list);
1925 LIST_HEAD(inode_buffer_list);
1926 LIST_HEAD(inode_list);
1928 list_splice_init(&trans->r_itemq, &sort_list);
1929 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1930 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
1932 switch (ITEM_TYPE(item)) {
1933 case XFS_LI_ICREATE:
1934 list_move_tail(&item->ri_list, &buffer_list);
1937 if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1938 trace_xfs_log_recover_item_reorder_head(log,
1940 list_move(&item->ri_list, &cancel_list);
1943 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1944 list_move(&item->ri_list, &inode_buffer_list);
1947 list_move_tail(&item->ri_list, &buffer_list);
1951 case XFS_LI_QUOTAOFF:
1960 trace_xfs_log_recover_item_reorder_tail(log,
1962 list_move_tail(&item->ri_list, &inode_list);
1966 "%s: unrecognized type of log operation",
1970 * return the remaining items back to the transaction
1971 * item list so they can be freed in caller.
1973 if (!list_empty(&sort_list))
1974 list_splice_init(&sort_list, &trans->r_itemq);
1980 ASSERT(list_empty(&sort_list));
1981 if (!list_empty(&buffer_list))
1982 list_splice(&buffer_list, &trans->r_itemq);
1983 if (!list_empty(&inode_list))
1984 list_splice_tail(&inode_list, &trans->r_itemq);
1985 if (!list_empty(&inode_buffer_list))
1986 list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1987 if (!list_empty(&cancel_list))
1988 list_splice_tail(&cancel_list, &trans->r_itemq);
1993 * Build up the table of buf cancel records so that we don't replay
1994 * cancelled data in the second pass. For buffer records that are
1995 * not cancel records, there is nothing to do here so we just return.
1997 * If we get a cancel record which is already in the table, this indicates
1998 * that the buffer was cancelled multiple times. In order to ensure
1999 * that during pass 2 we keep the record in the table until we reach its
2000 * last occurrence in the log, we keep a reference count in the cancel
2001 * record in the table to tell us how many times we expect to see this
2002 * record during the second pass.
2005 xlog_recover_buffer_pass1(
2007 struct xlog_recover_item *item)
2009 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2010 struct list_head *bucket;
2011 struct xfs_buf_cancel *bcp;
2014 * If this isn't a cancel buffer item, then just return.
2016 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
2017 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
2022 * Insert an xfs_buf_cancel record into the hash table of them.
2023 * If there is already an identical record, bump its reference count.
2025 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
2026 list_for_each_entry(bcp, bucket, bc_list) {
2027 if (bcp->bc_blkno == buf_f->blf_blkno &&
2028 bcp->bc_len == buf_f->blf_len) {
2030 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
2035 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
2036 bcp->bc_blkno = buf_f->blf_blkno;
2037 bcp->bc_len = buf_f->blf_len;
2038 bcp->bc_refcount = 1;
2039 list_add_tail(&bcp->bc_list, bucket);
2041 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
2046 * Check to see whether the buffer being recovered has a corresponding
2047 * entry in the buffer cancel record table. If it is, return the cancel
2048 * buffer structure to the caller.
2050 STATIC struct xfs_buf_cancel *
2051 xlog_peek_buffer_cancelled(
2055 unsigned short flags)
2057 struct list_head *bucket;
2058 struct xfs_buf_cancel *bcp;
2060 if (!log->l_buf_cancel_table) {
2061 /* empty table means no cancelled buffers in the log */
2062 ASSERT(!(flags & XFS_BLF_CANCEL));
2066 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
2067 list_for_each_entry(bcp, bucket, bc_list) {
2068 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
2073 * We didn't find a corresponding entry in the table, so return 0 so
2074 * that the buffer is NOT cancelled.
2076 ASSERT(!(flags & XFS_BLF_CANCEL));
2081 * If the buffer is being cancelled then return 1 so that it will be cancelled,
2082 * otherwise return 0. If the buffer is actually a buffer cancel item
2083 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
2084 * table and remove it from the table if this is the last reference.
2086 * We remove the cancel record from the table when we encounter its last
2087 * occurrence in the log so that if the same buffer is re-used again after its
2088 * last cancellation we actually replay the changes made at that point.
2091 xlog_check_buffer_cancelled(
2095 unsigned short flags)
2097 struct xfs_buf_cancel *bcp;
2099 bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
2104 * We've go a match, so return 1 so that the recovery of this buffer
2105 * is cancelled. If this buffer is actually a buffer cancel log
2106 * item, then decrement the refcount on the one in the table and
2107 * remove it if this is the last reference.
2109 if (flags & XFS_BLF_CANCEL) {
2110 if (--bcp->bc_refcount == 0) {
2111 list_del(&bcp->bc_list);
2119 * Perform recovery for a buffer full of inodes. In these buffers, the only
2120 * data which should be recovered is that which corresponds to the
2121 * di_next_unlinked pointers in the on disk inode structures. The rest of the
2122 * data for the inodes is always logged through the inodes themselves rather
2123 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
2125 * The only time when buffers full of inodes are fully recovered is when the
2126 * buffer is full of newly allocated inodes. In this case the buffer will
2127 * not be marked as an inode buffer and so will be sent to
2128 * xlog_recover_do_reg_buffer() below during recovery.
2131 xlog_recover_do_inode_buffer(
2132 struct xfs_mount *mp,
2133 xlog_recover_item_t *item,
2135 xfs_buf_log_format_t *buf_f)
2141 int reg_buf_offset = 0;
2142 int reg_buf_bytes = 0;
2143 int next_unlinked_offset;
2145 xfs_agino_t *logged_nextp;
2146 xfs_agino_t *buffer_nextp;
2148 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
2151 * Post recovery validation only works properly on CRC enabled
2154 if (xfs_sb_version_hascrc(&mp->m_sb))
2155 bp->b_ops = &xfs_inode_buf_ops;
2157 inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
2158 for (i = 0; i < inodes_per_buf; i++) {
2159 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
2160 offsetof(xfs_dinode_t, di_next_unlinked);
2162 while (next_unlinked_offset >=
2163 (reg_buf_offset + reg_buf_bytes)) {
2165 * The next di_next_unlinked field is beyond
2166 * the current logged region. Find the next
2167 * logged region that contains or is beyond
2168 * the current di_next_unlinked field.
2171 bit = xfs_next_bit(buf_f->blf_data_map,
2172 buf_f->blf_map_size, bit);
2175 * If there are no more logged regions in the
2176 * buffer, then we're done.
2181 nbits = xfs_contig_bits(buf_f->blf_data_map,
2182 buf_f->blf_map_size, bit);
2184 reg_buf_offset = bit << XFS_BLF_SHIFT;
2185 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
2190 * If the current logged region starts after the current
2191 * di_next_unlinked field, then move on to the next
2192 * di_next_unlinked field.
2194 if (next_unlinked_offset < reg_buf_offset)
2197 ASSERT(item->ri_buf[item_index].i_addr != NULL);
2198 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
2199 ASSERT((reg_buf_offset + reg_buf_bytes) <=
2200 BBTOB(bp->b_io_length));
2203 * The current logged region contains a copy of the
2204 * current di_next_unlinked field. Extract its value
2205 * and copy it to the buffer copy.
2207 logged_nextp = item->ri_buf[item_index].i_addr +
2208 next_unlinked_offset - reg_buf_offset;
2209 if (unlikely(*logged_nextp == 0)) {
2211 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
2212 "Trying to replay bad (0) inode di_next_unlinked field.",
2214 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
2215 XFS_ERRLEVEL_LOW, mp);
2216 return -EFSCORRUPTED;
2219 buffer_nextp = xfs_buf_offset(bp, next_unlinked_offset);
2220 *buffer_nextp = *logged_nextp;
2223 * If necessary, recalculate the CRC in the on-disk inode. We
2224 * have to leave the inode in a consistent state for whoever
2227 xfs_dinode_calc_crc(mp,
2228 xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
2236 * V5 filesystems know the age of the buffer on disk being recovered. We can
2237 * have newer objects on disk than we are replaying, and so for these cases we
2238 * don't want to replay the current change as that will make the buffer contents
2239 * temporarily invalid on disk.
2241 * The magic number might not match the buffer type we are going to recover
2242 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags. Hence
2243 * extract the LSN of the existing object in the buffer based on it's current
2244 * magic number. If we don't recognise the magic number in the buffer, then
2245 * return a LSN of -1 so that the caller knows it was an unrecognised block and
2246 * so can recover the buffer.
2248 * Note: we cannot rely solely on magic number matches to determine that the
2249 * buffer has a valid LSN - we also need to verify that it belongs to this
2250 * filesystem, so we need to extract the object's LSN and compare it to that
2251 * which we read from the superblock. If the UUIDs don't match, then we've got a
2252 * stale metadata block from an old filesystem instance that we need to recover
2256 xlog_recover_get_buf_lsn(
2257 struct xfs_mount *mp,
2263 void *blk = bp->b_addr;
2267 /* v4 filesystems always recover immediately */
2268 if (!xfs_sb_version_hascrc(&mp->m_sb))
2269 goto recover_immediately;
2271 magic32 = be32_to_cpu(*(__be32 *)blk);
2273 case XFS_ABTB_CRC_MAGIC:
2274 case XFS_ABTC_CRC_MAGIC:
2275 case XFS_ABTB_MAGIC:
2276 case XFS_ABTC_MAGIC:
2277 case XFS_RMAP_CRC_MAGIC:
2278 case XFS_REFC_CRC_MAGIC:
2279 case XFS_IBT_CRC_MAGIC:
2280 case XFS_IBT_MAGIC: {
2281 struct xfs_btree_block *btb = blk;
2283 lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2284 uuid = &btb->bb_u.s.bb_uuid;
2287 case XFS_BMAP_CRC_MAGIC:
2288 case XFS_BMAP_MAGIC: {
2289 struct xfs_btree_block *btb = blk;
2291 lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2292 uuid = &btb->bb_u.l.bb_uuid;
2296 lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2297 uuid = &((struct xfs_agf *)blk)->agf_uuid;
2299 case XFS_AGFL_MAGIC:
2300 lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2301 uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2304 lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2305 uuid = &((struct xfs_agi *)blk)->agi_uuid;
2307 case XFS_SYMLINK_MAGIC:
2308 lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2309 uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2311 case XFS_DIR3_BLOCK_MAGIC:
2312 case XFS_DIR3_DATA_MAGIC:
2313 case XFS_DIR3_FREE_MAGIC:
2314 lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2315 uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2317 case XFS_ATTR3_RMT_MAGIC:
2319 * Remote attr blocks are written synchronously, rather than
2320 * being logged. That means they do not contain a valid LSN
2321 * (i.e. transactionally ordered) in them, and hence any time we
2322 * see a buffer to replay over the top of a remote attribute
2323 * block we should simply do so.
2325 goto recover_immediately;
2328 * superblock uuids are magic. We may or may not have a
2329 * sb_meta_uuid on disk, but it will be set in the in-core
2330 * superblock. We set the uuid pointer for verification
2331 * according to the superblock feature mask to ensure we check
2332 * the relevant UUID in the superblock.
2334 lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2335 if (xfs_sb_version_hasmetauuid(&mp->m_sb))
2336 uuid = &((struct xfs_dsb *)blk)->sb_meta_uuid;
2338 uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2344 if (lsn != (xfs_lsn_t)-1) {
2345 if (!uuid_equal(&mp->m_sb.sb_meta_uuid, uuid))
2346 goto recover_immediately;
2350 magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2352 case XFS_DIR3_LEAF1_MAGIC:
2353 case XFS_DIR3_LEAFN_MAGIC:
2354 case XFS_DA3_NODE_MAGIC:
2355 lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2356 uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2362 if (lsn != (xfs_lsn_t)-1) {
2363 if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2364 goto recover_immediately;
2369 * We do individual object checks on dquot and inode buffers as they
2370 * have their own individual LSN records. Also, we could have a stale
2371 * buffer here, so we have to at least recognise these buffer types.
2373 * A notd complexity here is inode unlinked list processing - it logs
2374 * the inode directly in the buffer, but we don't know which inodes have
2375 * been modified, and there is no global buffer LSN. Hence we need to
2376 * recover all inode buffer types immediately. This problem will be
2377 * fixed by logical logging of the unlinked list modifications.
2379 magic16 = be16_to_cpu(*(__be16 *)blk);
2381 case XFS_DQUOT_MAGIC:
2382 case XFS_DINODE_MAGIC:
2383 goto recover_immediately;
2388 /* unknown buffer contents, recover immediately */
2390 recover_immediately:
2391 return (xfs_lsn_t)-1;
2396 * Validate the recovered buffer is of the correct type and attach the
2397 * appropriate buffer operations to them for writeback. Magic numbers are in a
2399 * the first 16 bits of the buffer (inode buffer, dquot buffer),
2400 * the first 32 bits of the buffer (most blocks),
2401 * inside a struct xfs_da_blkinfo at the start of the buffer.
2404 xlog_recover_validate_buf_type(
2405 struct xfs_mount *mp,
2407 xfs_buf_log_format_t *buf_f,
2408 xfs_lsn_t current_lsn)
2410 struct xfs_da_blkinfo *info = bp->b_addr;
2414 char *warnmsg = NULL;
2417 * We can only do post recovery validation on items on CRC enabled
2418 * fielsystems as we need to know when the buffer was written to be able
2419 * to determine if we should have replayed the item. If we replay old
2420 * metadata over a newer buffer, then it will enter a temporarily
2421 * inconsistent state resulting in verification failures. Hence for now
2422 * just avoid the verification stage for non-crc filesystems
2424 if (!xfs_sb_version_hascrc(&mp->m_sb))
2427 magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2428 magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2429 magicda = be16_to_cpu(info->magic);
2430 switch (xfs_blft_from_flags(buf_f)) {
2431 case XFS_BLFT_BTREE_BUF:
2433 case XFS_ABTB_CRC_MAGIC:
2434 case XFS_ABTC_CRC_MAGIC:
2435 case XFS_ABTB_MAGIC:
2436 case XFS_ABTC_MAGIC:
2437 bp->b_ops = &xfs_allocbt_buf_ops;
2439 case XFS_IBT_CRC_MAGIC:
2440 case XFS_FIBT_CRC_MAGIC:
2442 case XFS_FIBT_MAGIC:
2443 bp->b_ops = &xfs_inobt_buf_ops;
2445 case XFS_BMAP_CRC_MAGIC:
2446 case XFS_BMAP_MAGIC:
2447 bp->b_ops = &xfs_bmbt_buf_ops;
2449 case XFS_RMAP_CRC_MAGIC:
2450 bp->b_ops = &xfs_rmapbt_buf_ops;
2452 case XFS_REFC_CRC_MAGIC:
2453 bp->b_ops = &xfs_refcountbt_buf_ops;
2456 warnmsg = "Bad btree block magic!";
2460 case XFS_BLFT_AGF_BUF:
2461 if (magic32 != XFS_AGF_MAGIC) {
2462 warnmsg = "Bad AGF block magic!";
2465 bp->b_ops = &xfs_agf_buf_ops;
2467 case XFS_BLFT_AGFL_BUF:
2468 if (magic32 != XFS_AGFL_MAGIC) {
2469 warnmsg = "Bad AGFL block magic!";
2472 bp->b_ops = &xfs_agfl_buf_ops;
2474 case XFS_BLFT_AGI_BUF:
2475 if (magic32 != XFS_AGI_MAGIC) {
2476 warnmsg = "Bad AGI block magic!";
2479 bp->b_ops = &xfs_agi_buf_ops;
2481 case XFS_BLFT_UDQUOT_BUF:
2482 case XFS_BLFT_PDQUOT_BUF:
2483 case XFS_BLFT_GDQUOT_BUF:
2484 #ifdef CONFIG_XFS_QUOTA
2485 if (magic16 != XFS_DQUOT_MAGIC) {
2486 warnmsg = "Bad DQUOT block magic!";
2489 bp->b_ops = &xfs_dquot_buf_ops;
2492 "Trying to recover dquots without QUOTA support built in!");
2496 case XFS_BLFT_DINO_BUF:
2497 if (magic16 != XFS_DINODE_MAGIC) {
2498 warnmsg = "Bad INODE block magic!";
2501 bp->b_ops = &xfs_inode_buf_ops;
2503 case XFS_BLFT_SYMLINK_BUF:
2504 if (magic32 != XFS_SYMLINK_MAGIC) {
2505 warnmsg = "Bad symlink block magic!";
2508 bp->b_ops = &xfs_symlink_buf_ops;
2510 case XFS_BLFT_DIR_BLOCK_BUF:
2511 if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2512 magic32 != XFS_DIR3_BLOCK_MAGIC) {
2513 warnmsg = "Bad dir block magic!";
2516 bp->b_ops = &xfs_dir3_block_buf_ops;
2518 case XFS_BLFT_DIR_DATA_BUF:
2519 if (magic32 != XFS_DIR2_DATA_MAGIC &&
2520 magic32 != XFS_DIR3_DATA_MAGIC) {
2521 warnmsg = "Bad dir data magic!";
2524 bp->b_ops = &xfs_dir3_data_buf_ops;
2526 case XFS_BLFT_DIR_FREE_BUF:
2527 if (magic32 != XFS_DIR2_FREE_MAGIC &&
2528 magic32 != XFS_DIR3_FREE_MAGIC) {
2529 warnmsg = "Bad dir3 free magic!";
2532 bp->b_ops = &xfs_dir3_free_buf_ops;
2534 case XFS_BLFT_DIR_LEAF1_BUF:
2535 if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2536 magicda != XFS_DIR3_LEAF1_MAGIC) {
2537 warnmsg = "Bad dir leaf1 magic!";
2540 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2542 case XFS_BLFT_DIR_LEAFN_BUF:
2543 if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2544 magicda != XFS_DIR3_LEAFN_MAGIC) {
2545 warnmsg = "Bad dir leafn magic!";
2548 bp->b_ops = &xfs_dir3_leafn_buf_ops;
2550 case XFS_BLFT_DA_NODE_BUF:
2551 if (magicda != XFS_DA_NODE_MAGIC &&
2552 magicda != XFS_DA3_NODE_MAGIC) {
2553 warnmsg = "Bad da node magic!";
2556 bp->b_ops = &xfs_da3_node_buf_ops;
2558 case XFS_BLFT_ATTR_LEAF_BUF:
2559 if (magicda != XFS_ATTR_LEAF_MAGIC &&
2560 magicda != XFS_ATTR3_LEAF_MAGIC) {
2561 warnmsg = "Bad attr leaf magic!";
2564 bp->b_ops = &xfs_attr3_leaf_buf_ops;
2566 case XFS_BLFT_ATTR_RMT_BUF:
2567 if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2568 warnmsg = "Bad attr remote magic!";
2571 bp->b_ops = &xfs_attr3_rmt_buf_ops;
2573 case XFS_BLFT_SB_BUF:
2574 if (magic32 != XFS_SB_MAGIC) {
2575 warnmsg = "Bad SB block magic!";
2578 bp->b_ops = &xfs_sb_buf_ops;
2580 #ifdef CONFIG_XFS_RT
2581 case XFS_BLFT_RTBITMAP_BUF:
2582 case XFS_BLFT_RTSUMMARY_BUF:
2583 /* no magic numbers for verification of RT buffers */
2584 bp->b_ops = &xfs_rtbuf_ops;
2586 #endif /* CONFIG_XFS_RT */
2588 xfs_warn(mp, "Unknown buffer type %d!",
2589 xfs_blft_from_flags(buf_f));
2594 * Nothing else to do in the case of a NULL current LSN as this means
2595 * the buffer is more recent than the change in the log and will be
2598 if (current_lsn == NULLCOMMITLSN)
2602 xfs_warn(mp, warnmsg);
2607 * We must update the metadata LSN of the buffer as it is written out to
2608 * ensure that older transactions never replay over this one and corrupt
2609 * the buffer. This can occur if log recovery is interrupted at some
2610 * point after the current transaction completes, at which point a
2611 * subsequent mount starts recovery from the beginning.
2613 * Write verifiers update the metadata LSN from log items attached to
2614 * the buffer. Therefore, initialize a bli purely to carry the LSN to
2615 * the verifier. We'll clean it up in our ->iodone() callback.
2618 struct xfs_buf_log_item *bip;
2620 ASSERT(!bp->b_iodone || bp->b_iodone == xlog_recover_iodone);
2621 bp->b_iodone = xlog_recover_iodone;
2622 xfs_buf_item_init(bp, mp);
2624 bip->bli_item.li_lsn = current_lsn;
2629 * Perform a 'normal' buffer recovery. Each logged region of the
2630 * buffer should be copied over the corresponding region in the
2631 * given buffer. The bitmap in the buf log format structure indicates
2632 * where to place the logged data.
2635 xlog_recover_do_reg_buffer(
2636 struct xfs_mount *mp,
2637 xlog_recover_item_t *item,
2639 xfs_buf_log_format_t *buf_f,
2640 xfs_lsn_t current_lsn)
2647 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2650 i = 1; /* 0 is the buf format structure */
2652 bit = xfs_next_bit(buf_f->blf_data_map,
2653 buf_f->blf_map_size, bit);
2656 nbits = xfs_contig_bits(buf_f->blf_data_map,
2657 buf_f->blf_map_size, bit);
2659 ASSERT(item->ri_buf[i].i_addr != NULL);
2660 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2661 ASSERT(BBTOB(bp->b_io_length) >=
2662 ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2665 * The dirty regions logged in the buffer, even though
2666 * contiguous, may span multiple chunks. This is because the
2667 * dirty region may span a physical page boundary in a buffer
2668 * and hence be split into two separate vectors for writing into
2669 * the log. Hence we need to trim nbits back to the length of
2670 * the current region being copied out of the log.
2672 if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2673 nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2676 * Do a sanity check if this is a dquot buffer. Just checking
2677 * the first dquot in the buffer should do. XXXThis is
2678 * probably a good thing to do for other buf types also.
2681 if (buf_f->blf_flags &
2682 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2683 if (item->ri_buf[i].i_addr == NULL) {
2685 "XFS: NULL dquot in %s.", __func__);
2688 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2690 "XFS: dquot too small (%d) in %s.",
2691 item->ri_buf[i].i_len, __func__);
2694 error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2695 -1, 0, XFS_QMOPT_DOWARN,
2696 "dquot_buf_recover");
2701 memcpy(xfs_buf_offset(bp,
2702 (uint)bit << XFS_BLF_SHIFT), /* dest */
2703 item->ri_buf[i].i_addr, /* source */
2704 nbits<<XFS_BLF_SHIFT); /* length */
2710 /* Shouldn't be any more regions */
2711 ASSERT(i == item->ri_total);
2713 xlog_recover_validate_buf_type(mp, bp, buf_f, current_lsn);
2717 * Perform a dquot buffer recovery.
2718 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2719 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2720 * Else, treat it as a regular buffer and do recovery.
2722 * Return false if the buffer was tossed and true if we recovered the buffer to
2723 * indicate to the caller if the buffer needs writing.
2726 xlog_recover_do_dquot_buffer(
2727 struct xfs_mount *mp,
2729 struct xlog_recover_item *item,
2731 struct xfs_buf_log_format *buf_f)
2735 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2738 * Filesystems are required to send in quota flags at mount time.
2744 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2745 type |= XFS_DQ_USER;
2746 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2747 type |= XFS_DQ_PROJ;
2748 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2749 type |= XFS_DQ_GROUP;
2751 * This type of quotas was turned off, so ignore this buffer
2753 if (log->l_quotaoffs_flag & type)
2756 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, NULLCOMMITLSN);
2761 * This routine replays a modification made to a buffer at runtime.
2762 * There are actually two types of buffer, regular and inode, which
2763 * are handled differently. Inode buffers are handled differently
2764 * in that we only recover a specific set of data from them, namely
2765 * the inode di_next_unlinked fields. This is because all other inode
2766 * data is actually logged via inode records and any data we replay
2767 * here which overlaps that may be stale.
2769 * When meta-data buffers are freed at run time we log a buffer item
2770 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2771 * of the buffer in the log should not be replayed at recovery time.
2772 * This is so that if the blocks covered by the buffer are reused for
2773 * file data before we crash we don't end up replaying old, freed
2774 * meta-data into a user's file.
2776 * To handle the cancellation of buffer log items, we make two passes
2777 * over the log during recovery. During the first we build a table of
2778 * those buffers which have been cancelled, and during the second we
2779 * only replay those buffers which do not have corresponding cancel
2780 * records in the table. See xlog_recover_buffer_pass[1,2] above
2781 * for more details on the implementation of the table of cancel records.
2784 xlog_recover_buffer_pass2(
2786 struct list_head *buffer_list,
2787 struct xlog_recover_item *item,
2788 xfs_lsn_t current_lsn)
2790 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
2791 xfs_mount_t *mp = log->l_mp;
2798 * In this pass we only want to recover all the buffers which have
2799 * not been cancelled and are not cancellation buffers themselves.
2801 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2802 buf_f->blf_len, buf_f->blf_flags)) {
2803 trace_xfs_log_recover_buf_cancel(log, buf_f);
2807 trace_xfs_log_recover_buf_recover(log, buf_f);
2810 if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2811 buf_flags |= XBF_UNMAPPED;
2813 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2817 error = bp->b_error;
2819 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2824 * Recover the buffer only if we get an LSN from it and it's less than
2825 * the lsn of the transaction we are replaying.
2827 * Note that we have to be extremely careful of readahead here.
2828 * Readahead does not attach verfiers to the buffers so if we don't
2829 * actually do any replay after readahead because of the LSN we found
2830 * in the buffer if more recent than that current transaction then we
2831 * need to attach the verifier directly. Failure to do so can lead to
2832 * future recovery actions (e.g. EFI and unlinked list recovery) can
2833 * operate on the buffers and they won't get the verifier attached. This
2834 * can lead to blocks on disk having the correct content but a stale
2837 * It is safe to assume these clean buffers are currently up to date.
2838 * If the buffer is dirtied by a later transaction being replayed, then
2839 * the verifier will be reset to match whatever recover turns that
2842 lsn = xlog_recover_get_buf_lsn(mp, bp);
2843 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2844 trace_xfs_log_recover_buf_skip(log, buf_f);
2845 xlog_recover_validate_buf_type(mp, bp, buf_f, NULLCOMMITLSN);
2849 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2850 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2853 } else if (buf_f->blf_flags &
2854 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2857 dirty = xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2861 xlog_recover_do_reg_buffer(mp, item, bp, buf_f, current_lsn);
2865 * Perform delayed write on the buffer. Asynchronous writes will be
2866 * slower when taking into account all the buffers to be flushed.
2868 * Also make sure that only inode buffers with good sizes stay in
2869 * the buffer cache. The kernel moves inodes in buffers of 1 block
2870 * or mp->m_inode_cluster_size bytes, whichever is bigger. The inode
2871 * buffers in the log can be a different size if the log was generated
2872 * by an older kernel using unclustered inode buffers or a newer kernel
2873 * running with a different inode cluster size. Regardless, if the
2874 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2875 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2876 * the buffer out of the buffer cache so that the buffer won't
2877 * overlap with future reads of those inodes.
2879 if (XFS_DINODE_MAGIC ==
2880 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2881 (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2882 (uint32_t)log->l_mp->m_inode_cluster_size))) {
2884 error = xfs_bwrite(bp);
2886 ASSERT(bp->b_target->bt_mount == mp);
2887 bp->b_iodone = xlog_recover_iodone;
2888 xfs_buf_delwri_queue(bp, buffer_list);
2897 * Inode fork owner changes
2899 * If we have been told that we have to reparent the inode fork, it's because an
2900 * extent swap operation on a CRC enabled filesystem has been done and we are
2901 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2904 * The complexity here is that we don't have an inode context to work with, so
2905 * after we've replayed the inode we need to instantiate one. This is where the
2908 * We are in the middle of log recovery, so we can't run transactions. That
2909 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2910 * that will result in the corresponding iput() running the inode through
2911 * xfs_inactive(). If we've just replayed an inode core that changes the link
2912 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2913 * transactions (bad!).
2915 * So, to avoid this, we instantiate an inode directly from the inode core we've
2916 * just recovered. We have the buffer still locked, and all we really need to
2917 * instantiate is the inode core and the forks being modified. We can do this
2918 * manually, then run the inode btree owner change, and then tear down the
2919 * xfs_inode without having to run any transactions at all.
2921 * Also, because we don't have a transaction context available here but need to
2922 * gather all the buffers we modify for writeback so we pass the buffer_list
2923 * instead for the operation to use.
2927 xfs_recover_inode_owner_change(
2928 struct xfs_mount *mp,
2929 struct xfs_dinode *dip,
2930 struct xfs_inode_log_format *in_f,
2931 struct list_head *buffer_list)
2933 struct xfs_inode *ip;
2936 ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2938 ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2942 /* instantiate the inode */
2943 xfs_inode_from_disk(ip, dip);
2944 ASSERT(ip->i_d.di_version >= 3);
2946 error = xfs_iformat_fork(ip, dip);
2951 if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2952 ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2953 error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2954 ip->i_ino, buffer_list);
2959 if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2960 ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2961 error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2962 ip->i_ino, buffer_list);
2973 xlog_recover_inode_pass2(
2975 struct list_head *buffer_list,
2976 struct xlog_recover_item *item,
2977 xfs_lsn_t current_lsn)
2979 xfs_inode_log_format_t *in_f;
2980 xfs_mount_t *mp = log->l_mp;
2989 struct xfs_log_dinode *ldip;
2993 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2994 in_f = item->ri_buf[0].i_addr;
2996 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2998 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
3004 * Inode buffers can be freed, look out for it,
3005 * and do not replay the inode.
3007 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
3008 in_f->ilf_len, 0)) {
3010 trace_xfs_log_recover_inode_cancel(log, in_f);
3013 trace_xfs_log_recover_inode_recover(log, in_f);
3015 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
3016 &xfs_inode_buf_ops);
3021 error = bp->b_error;
3023 xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
3026 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
3027 dip = xfs_buf_offset(bp, in_f->ilf_boffset);
3030 * Make sure the place we're flushing out to really looks
3033 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
3035 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
3036 __func__, dip, bp, in_f->ilf_ino);
3037 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
3038 XFS_ERRLEVEL_LOW, mp);
3039 error = -EFSCORRUPTED;
3042 ldip = item->ri_buf[1].i_addr;
3043 if (unlikely(ldip->di_magic != XFS_DINODE_MAGIC)) {
3045 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
3046 __func__, item, in_f->ilf_ino);
3047 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
3048 XFS_ERRLEVEL_LOW, mp);
3049 error = -EFSCORRUPTED;
3054 * If the inode has an LSN in it, recover the inode only if it's less
3055 * than the lsn of the transaction we are replaying. Note: we still
3056 * need to replay an owner change even though the inode is more recent
3057 * than the transaction as there is no guarantee that all the btree
3058 * blocks are more recent than this transaction, too.
3060 if (dip->di_version >= 3) {
3061 xfs_lsn_t lsn = be64_to_cpu(dip->di_lsn);
3063 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3064 trace_xfs_log_recover_inode_skip(log, in_f);
3066 goto out_owner_change;
3071 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
3072 * are transactional and if ordering is necessary we can determine that
3073 * more accurately by the LSN field in the V3 inode core. Don't trust
3074 * the inode versions we might be changing them here - use the
3075 * superblock flag to determine whether we need to look at di_flushiter
3076 * to skip replay when the on disk inode is newer than the log one
3078 if (!xfs_sb_version_hascrc(&mp->m_sb) &&
3079 ldip->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
3081 * Deal with the wrap case, DI_MAX_FLUSH is less
3082 * than smaller numbers
3084 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
3085 ldip->di_flushiter < (DI_MAX_FLUSH >> 1)) {
3088 trace_xfs_log_recover_inode_skip(log, in_f);
3094 /* Take the opportunity to reset the flush iteration count */
3095 ldip->di_flushiter = 0;
3097 if (unlikely(S_ISREG(ldip->di_mode))) {
3098 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3099 (ldip->di_format != XFS_DINODE_FMT_BTREE)) {
3100 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
3101 XFS_ERRLEVEL_LOW, mp, ldip);
3103 "%s: Bad regular inode log record, rec ptr 0x%p, "
3104 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3105 __func__, item, dip, bp, in_f->ilf_ino);
3106 error = -EFSCORRUPTED;
3109 } else if (unlikely(S_ISDIR(ldip->di_mode))) {
3110 if ((ldip->di_format != XFS_DINODE_FMT_EXTENTS) &&
3111 (ldip->di_format != XFS_DINODE_FMT_BTREE) &&
3112 (ldip->di_format != XFS_DINODE_FMT_LOCAL)) {
3113 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
3114 XFS_ERRLEVEL_LOW, mp, ldip);
3116 "%s: Bad dir inode log record, rec ptr 0x%p, "
3117 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
3118 __func__, item, dip, bp, in_f->ilf_ino);
3119 error = -EFSCORRUPTED;
3123 if (unlikely(ldip->di_nextents + ldip->di_anextents > ldip->di_nblocks)){
3124 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
3125 XFS_ERRLEVEL_LOW, mp, ldip);
3127 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3128 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
3129 __func__, item, dip, bp, in_f->ilf_ino,
3130 ldip->di_nextents + ldip->di_anextents,
3132 error = -EFSCORRUPTED;
3135 if (unlikely(ldip->di_forkoff > mp->m_sb.sb_inodesize)) {
3136 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
3137 XFS_ERRLEVEL_LOW, mp, ldip);
3139 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
3140 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
3141 item, dip, bp, in_f->ilf_ino, ldip->di_forkoff);
3142 error = -EFSCORRUPTED;
3145 isize = xfs_log_dinode_size(ldip->di_version);
3146 if (unlikely(item->ri_buf[1].i_len > isize)) {
3147 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
3148 XFS_ERRLEVEL_LOW, mp, ldip);
3150 "%s: Bad inode log record length %d, rec ptr 0x%p",
3151 __func__, item->ri_buf[1].i_len, item);
3152 error = -EFSCORRUPTED;
3156 /* recover the log dinode inode into the on disk inode */
3157 xfs_log_dinode_to_disk(ldip, dip);
3159 /* the rest is in on-disk format */
3160 if (item->ri_buf[1].i_len > isize) {
3161 memcpy((char *)dip + isize,
3162 item->ri_buf[1].i_addr + isize,
3163 item->ri_buf[1].i_len - isize);
3166 fields = in_f->ilf_fields;
3167 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
3169 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
3172 memcpy(XFS_DFORK_DPTR(dip),
3173 &in_f->ilf_u.ilfu_uuid,
3178 if (in_f->ilf_size == 2)
3179 goto out_owner_change;
3180 len = item->ri_buf[2].i_len;
3181 src = item->ri_buf[2].i_addr;
3182 ASSERT(in_f->ilf_size <= 4);
3183 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
3184 ASSERT(!(fields & XFS_ILOG_DFORK) ||
3185 (len == in_f->ilf_dsize));
3187 switch (fields & XFS_ILOG_DFORK) {
3188 case XFS_ILOG_DDATA:
3190 memcpy(XFS_DFORK_DPTR(dip), src, len);
3193 case XFS_ILOG_DBROOT:
3194 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
3195 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
3196 XFS_DFORK_DSIZE(dip, mp));
3201 * There are no data fork flags set.
3203 ASSERT((fields & XFS_ILOG_DFORK) == 0);
3208 * If we logged any attribute data, recover it. There may or
3209 * may not have been any other non-core data logged in this
3212 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
3213 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
3218 len = item->ri_buf[attr_index].i_len;
3219 src = item->ri_buf[attr_index].i_addr;
3220 ASSERT(len == in_f->ilf_asize);
3222 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
3223 case XFS_ILOG_ADATA:
3225 dest = XFS_DFORK_APTR(dip);
3226 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
3227 memcpy(dest, src, len);
3230 case XFS_ILOG_ABROOT:
3231 dest = XFS_DFORK_APTR(dip);
3232 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
3233 len, (xfs_bmdr_block_t*)dest,
3234 XFS_DFORK_ASIZE(dip, mp));
3238 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
3246 if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
3247 error = xfs_recover_inode_owner_change(mp, dip, in_f,
3249 /* re-generate the checksum. */
3250 xfs_dinode_calc_crc(log->l_mp, dip);
3252 ASSERT(bp->b_target->bt_mount == mp);
3253 bp->b_iodone = xlog_recover_iodone;
3254 xfs_buf_delwri_queue(bp, buffer_list);
3265 * Recover QUOTAOFF records. We simply make a note of it in the xlog
3266 * structure, so that we know not to do any dquot item or dquot buffer recovery,
3270 xlog_recover_quotaoff_pass1(
3272 struct xlog_recover_item *item)
3274 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
3278 * The logitem format's flag tells us if this was user quotaoff,
3279 * group/project quotaoff or both.
3281 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
3282 log->l_quotaoffs_flag |= XFS_DQ_USER;
3283 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
3284 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
3285 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
3286 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
3292 * Recover a dquot record
3295 xlog_recover_dquot_pass2(
3297 struct list_head *buffer_list,
3298 struct xlog_recover_item *item,
3299 xfs_lsn_t current_lsn)
3301 xfs_mount_t *mp = log->l_mp;
3303 struct xfs_disk_dquot *ddq, *recddq;
3305 xfs_dq_logformat_t *dq_f;
3310 * Filesystems are required to send in quota flags at mount time.
3312 if (mp->m_qflags == 0)
3315 recddq = item->ri_buf[1].i_addr;
3316 if (recddq == NULL) {
3317 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
3320 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
3321 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
3322 item->ri_buf[1].i_len, __func__);
3327 * This type of quotas was turned off, so ignore this record.
3329 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3331 if (log->l_quotaoffs_flag & type)
3335 * At this point we know that quota was _not_ turned off.
3336 * Since the mount flags are not indicating to us otherwise, this
3337 * must mean that quota is on, and the dquot needs to be replayed.
3338 * Remember that we may not have fully recovered the superblock yet,
3339 * so we can't do the usual trick of looking at the SB quota bits.
3341 * The other possibility, of course, is that the quota subsystem was
3342 * removed since the last mount - ENOSYS.
3344 dq_f = item->ri_buf[0].i_addr;
3346 error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3347 "xlog_recover_dquot_pass2 (log copy)");
3350 ASSERT(dq_f->qlf_len == 1);
3353 * At this point we are assuming that the dquots have been allocated
3354 * and hence the buffer has valid dquots stamped in it. It should,
3355 * therefore, pass verifier validation. If the dquot is bad, then the
3356 * we'll return an error here, so we don't need to specifically check
3357 * the dquot in the buffer after the verifier has run.
3359 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3360 XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3361 &xfs_dquot_buf_ops);
3366 ddq = xfs_buf_offset(bp, dq_f->qlf_boffset);
3369 * If the dquot has an LSN in it, recover the dquot only if it's less
3370 * than the lsn of the transaction we are replaying.
3372 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3373 struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3374 xfs_lsn_t lsn = be64_to_cpu(dqb->dd_lsn);
3376 if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3381 memcpy(ddq, recddq, item->ri_buf[1].i_len);
3382 if (xfs_sb_version_hascrc(&mp->m_sb)) {
3383 xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3387 ASSERT(dq_f->qlf_size == 2);
3388 ASSERT(bp->b_target->bt_mount == mp);
3389 bp->b_iodone = xlog_recover_iodone;
3390 xfs_buf_delwri_queue(bp, buffer_list);
3398 * This routine is called to create an in-core extent free intent
3399 * item from the efi format structure which was logged on disk.
3400 * It allocates an in-core efi, copies the extents from the format
3401 * structure into it, and adds the efi to the AIL with the given
3405 xlog_recover_efi_pass2(
3407 struct xlog_recover_item *item,
3411 struct xfs_mount *mp = log->l_mp;
3412 struct xfs_efi_log_item *efip;
3413 struct xfs_efi_log_format *efi_formatp;
3415 efi_formatp = item->ri_buf[0].i_addr;
3417 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3418 error = xfs_efi_copy_format(&item->ri_buf[0], &efip->efi_format);
3420 xfs_efi_item_free(efip);
3423 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3425 spin_lock(&log->l_ailp->xa_lock);
3427 * The EFI has two references. One for the EFD and one for EFI to ensure
3428 * it makes it into the AIL. Insert the EFI into the AIL directly and
3429 * drop the EFI reference. Note that xfs_trans_ail_update() drops the
3432 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3433 xfs_efi_release(efip);
3439 * This routine is called when an EFD format structure is found in a committed
3440 * transaction in the log. Its purpose is to cancel the corresponding EFI if it
3441 * was still in the log. To do this it searches the AIL for the EFI with an id
3442 * equal to that in the EFD format structure. If we find it we drop the EFD
3443 * reference, which removes the EFI from the AIL and frees it.
3446 xlog_recover_efd_pass2(
3448 struct xlog_recover_item *item)
3450 xfs_efd_log_format_t *efd_formatp;
3451 xfs_efi_log_item_t *efip = NULL;
3452 xfs_log_item_t *lip;
3454 struct xfs_ail_cursor cur;
3455 struct xfs_ail *ailp = log->l_ailp;
3457 efd_formatp = item->ri_buf[0].i_addr;
3458 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3459 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3460 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3461 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3462 efi_id = efd_formatp->efd_efi_id;
3465 * Search for the EFI with the id in the EFD format structure in the
3468 spin_lock(&ailp->xa_lock);
3469 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3470 while (lip != NULL) {
3471 if (lip->li_type == XFS_LI_EFI) {
3472 efip = (xfs_efi_log_item_t *)lip;
3473 if (efip->efi_format.efi_id == efi_id) {
3475 * Drop the EFD reference to the EFI. This
3476 * removes the EFI from the AIL and frees it.
3478 spin_unlock(&ailp->xa_lock);
3479 xfs_efi_release(efip);
3480 spin_lock(&ailp->xa_lock);
3484 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3487 xfs_trans_ail_cursor_done(&cur);
3488 spin_unlock(&ailp->xa_lock);
3494 * This routine is called to create an in-core extent rmap update
3495 * item from the rui format structure which was logged on disk.
3496 * It allocates an in-core rui, copies the extents from the format
3497 * structure into it, and adds the rui to the AIL with the given
3501 xlog_recover_rui_pass2(
3503 struct xlog_recover_item *item,
3507 struct xfs_mount *mp = log->l_mp;
3508 struct xfs_rui_log_item *ruip;
3509 struct xfs_rui_log_format *rui_formatp;
3511 rui_formatp = item->ri_buf[0].i_addr;
3513 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
3514 error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
3516 xfs_rui_item_free(ruip);
3519 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
3521 spin_lock(&log->l_ailp->xa_lock);
3523 * The RUI has two references. One for the RUD and one for RUI to ensure
3524 * it makes it into the AIL. Insert the RUI into the AIL directly and
3525 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3528 xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
3529 xfs_rui_release(ruip);
3535 * This routine is called when an RUD format structure is found in a committed
3536 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
3537 * was still in the log. To do this it searches the AIL for the RUI with an id
3538 * equal to that in the RUD format structure. If we find it we drop the RUD
3539 * reference, which removes the RUI from the AIL and frees it.
3542 xlog_recover_rud_pass2(
3544 struct xlog_recover_item *item)
3546 struct xfs_rud_log_format *rud_formatp;
3547 struct xfs_rui_log_item *ruip = NULL;
3548 struct xfs_log_item *lip;
3550 struct xfs_ail_cursor cur;
3551 struct xfs_ail *ailp = log->l_ailp;
3553 rud_formatp = item->ri_buf[0].i_addr;
3554 ASSERT(item->ri_buf[0].i_len == sizeof(struct xfs_rud_log_format));
3555 rui_id = rud_formatp->rud_rui_id;
3558 * Search for the RUI with the id in the RUD format structure in the
3561 spin_lock(&ailp->xa_lock);
3562 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3563 while (lip != NULL) {
3564 if (lip->li_type == XFS_LI_RUI) {
3565 ruip = (struct xfs_rui_log_item *)lip;
3566 if (ruip->rui_format.rui_id == rui_id) {
3568 * Drop the RUD reference to the RUI. This
3569 * removes the RUI from the AIL and frees it.
3571 spin_unlock(&ailp->xa_lock);
3572 xfs_rui_release(ruip);
3573 spin_lock(&ailp->xa_lock);
3577 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3580 xfs_trans_ail_cursor_done(&cur);
3581 spin_unlock(&ailp->xa_lock);
3587 * Copy an CUI format buffer from the given buf, and into the destination
3588 * CUI format structure. The CUI/CUD items were designed not to need any
3589 * special alignment handling.
3592 xfs_cui_copy_format(
3593 struct xfs_log_iovec *buf,
3594 struct xfs_cui_log_format *dst_cui_fmt)
3596 struct xfs_cui_log_format *src_cui_fmt;
3599 src_cui_fmt = buf->i_addr;
3600 len = xfs_cui_log_format_sizeof(src_cui_fmt->cui_nextents);
3602 if (buf->i_len == len) {
3603 memcpy(dst_cui_fmt, src_cui_fmt, len);
3606 return -EFSCORRUPTED;
3610 * This routine is called to create an in-core extent refcount update
3611 * item from the cui format structure which was logged on disk.
3612 * It allocates an in-core cui, copies the extents from the format
3613 * structure into it, and adds the cui to the AIL with the given
3617 xlog_recover_cui_pass2(
3619 struct xlog_recover_item *item,
3623 struct xfs_mount *mp = log->l_mp;
3624 struct xfs_cui_log_item *cuip;
3625 struct xfs_cui_log_format *cui_formatp;
3627 cui_formatp = item->ri_buf[0].i_addr;
3629 cuip = xfs_cui_init(mp, cui_formatp->cui_nextents);
3630 error = xfs_cui_copy_format(&item->ri_buf[0], &cuip->cui_format);
3632 xfs_cui_item_free(cuip);
3635 atomic_set(&cuip->cui_next_extent, cui_formatp->cui_nextents);
3637 spin_lock(&log->l_ailp->xa_lock);
3639 * The CUI has two references. One for the CUD and one for CUI to ensure
3640 * it makes it into the AIL. Insert the CUI into the AIL directly and
3641 * drop the CUI reference. Note that xfs_trans_ail_update() drops the
3644 xfs_trans_ail_update(log->l_ailp, &cuip->cui_item, lsn);
3645 xfs_cui_release(cuip);
3651 * This routine is called when an CUD format structure is found in a committed
3652 * transaction in the log. Its purpose is to cancel the corresponding CUI if it
3653 * was still in the log. To do this it searches the AIL for the CUI with an id
3654 * equal to that in the CUD format structure. If we find it we drop the CUD
3655 * reference, which removes the CUI from the AIL and frees it.
3658 xlog_recover_cud_pass2(
3660 struct xlog_recover_item *item)
3662 struct xfs_cud_log_format *cud_formatp;
3663 struct xfs_cui_log_item *cuip = NULL;
3664 struct xfs_log_item *lip;
3666 struct xfs_ail_cursor cur;
3667 struct xfs_ail *ailp = log->l_ailp;
3669 cud_formatp = item->ri_buf[0].i_addr;
3670 if (item->ri_buf[0].i_len != sizeof(struct xfs_cud_log_format))
3671 return -EFSCORRUPTED;
3672 cui_id = cud_formatp->cud_cui_id;
3675 * Search for the CUI with the id in the CUD format structure in the
3678 spin_lock(&ailp->xa_lock);
3679 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3680 while (lip != NULL) {
3681 if (lip->li_type == XFS_LI_CUI) {
3682 cuip = (struct xfs_cui_log_item *)lip;
3683 if (cuip->cui_format.cui_id == cui_id) {
3685 * Drop the CUD reference to the CUI. This
3686 * removes the CUI from the AIL and frees it.
3688 spin_unlock(&ailp->xa_lock);
3689 xfs_cui_release(cuip);
3690 spin_lock(&ailp->xa_lock);
3694 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3697 xfs_trans_ail_cursor_done(&cur);
3698 spin_unlock(&ailp->xa_lock);
3704 * Copy an BUI format buffer from the given buf, and into the destination
3705 * BUI format structure. The BUI/BUD items were designed not to need any
3706 * special alignment handling.
3709 xfs_bui_copy_format(
3710 struct xfs_log_iovec *buf,
3711 struct xfs_bui_log_format *dst_bui_fmt)
3713 struct xfs_bui_log_format *src_bui_fmt;
3716 src_bui_fmt = buf->i_addr;
3717 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
3719 if (buf->i_len == len) {
3720 memcpy(dst_bui_fmt, src_bui_fmt, len);
3723 return -EFSCORRUPTED;
3727 * This routine is called to create an in-core extent bmap update
3728 * item from the bui format structure which was logged on disk.
3729 * It allocates an in-core bui, copies the extents from the format
3730 * structure into it, and adds the bui to the AIL with the given
3734 xlog_recover_bui_pass2(
3736 struct xlog_recover_item *item,
3740 struct xfs_mount *mp = log->l_mp;
3741 struct xfs_bui_log_item *buip;
3742 struct xfs_bui_log_format *bui_formatp;
3744 bui_formatp = item->ri_buf[0].i_addr;
3746 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
3747 return -EFSCORRUPTED;
3748 buip = xfs_bui_init(mp);
3749 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
3751 xfs_bui_item_free(buip);
3754 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
3756 spin_lock(&log->l_ailp->xa_lock);
3758 * The RUI has two references. One for the RUD and one for RUI to ensure
3759 * it makes it into the AIL. Insert the RUI into the AIL directly and
3760 * drop the RUI reference. Note that xfs_trans_ail_update() drops the
3763 xfs_trans_ail_update(log->l_ailp, &buip->bui_item, lsn);
3764 xfs_bui_release(buip);
3770 * This routine is called when an BUD format structure is found in a committed
3771 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
3772 * was still in the log. To do this it searches the AIL for the BUI with an id
3773 * equal to that in the BUD format structure. If we find it we drop the BUD
3774 * reference, which removes the BUI from the AIL and frees it.
3777 xlog_recover_bud_pass2(
3779 struct xlog_recover_item *item)
3781 struct xfs_bud_log_format *bud_formatp;
3782 struct xfs_bui_log_item *buip = NULL;
3783 struct xfs_log_item *lip;
3785 struct xfs_ail_cursor cur;
3786 struct xfs_ail *ailp = log->l_ailp;
3788 bud_formatp = item->ri_buf[0].i_addr;
3789 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format))
3790 return -EFSCORRUPTED;
3791 bui_id = bud_formatp->bud_bui_id;
3794 * Search for the BUI with the id in the BUD format structure in the
3797 spin_lock(&ailp->xa_lock);
3798 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3799 while (lip != NULL) {
3800 if (lip->li_type == XFS_LI_BUI) {
3801 buip = (struct xfs_bui_log_item *)lip;
3802 if (buip->bui_format.bui_id == bui_id) {
3804 * Drop the BUD reference to the BUI. This
3805 * removes the BUI from the AIL and frees it.
3807 spin_unlock(&ailp->xa_lock);
3808 xfs_bui_release(buip);
3809 spin_lock(&ailp->xa_lock);
3813 lip = xfs_trans_ail_cursor_next(ailp, &cur);
3816 xfs_trans_ail_cursor_done(&cur);
3817 spin_unlock(&ailp->xa_lock);
3823 * This routine is called when an inode create format structure is found in a
3824 * committed transaction in the log. It's purpose is to initialise the inodes
3825 * being allocated on disk. This requires us to get inode cluster buffers that
3826 * match the range to be initialised, stamped with inode templates and written
3827 * by delayed write so that subsequent modifications will hit the cached buffer
3828 * and only need writing out at the end of recovery.
3831 xlog_recover_do_icreate_pass2(
3833 struct list_head *buffer_list,
3834 xlog_recover_item_t *item)
3836 struct xfs_mount *mp = log->l_mp;
3837 struct xfs_icreate_log *icl;
3838 xfs_agnumber_t agno;
3839 xfs_agblock_t agbno;
3842 xfs_agblock_t length;
3843 int blks_per_cluster;
3849 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3850 if (icl->icl_type != XFS_LI_ICREATE) {
3851 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3855 if (icl->icl_size != 1) {
3856 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3860 agno = be32_to_cpu(icl->icl_ag);
3861 if (agno >= mp->m_sb.sb_agcount) {
3862 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3865 agbno = be32_to_cpu(icl->icl_agbno);
3866 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3867 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3870 isize = be32_to_cpu(icl->icl_isize);
3871 if (isize != mp->m_sb.sb_inodesize) {
3872 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3875 count = be32_to_cpu(icl->icl_count);
3877 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3880 length = be32_to_cpu(icl->icl_length);
3881 if (!length || length >= mp->m_sb.sb_agblocks) {
3882 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3887 * The inode chunk is either full or sparse and we only support
3888 * m_ialloc_min_blks sized sparse allocations at this time.
3890 if (length != mp->m_ialloc_blks &&
3891 length != mp->m_ialloc_min_blks) {
3893 "%s: unsupported chunk length", __FUNCTION__);
3897 /* verify inode count is consistent with extent length */
3898 if ((count >> mp->m_sb.sb_inopblog) != length) {
3900 "%s: inconsistent inode count and chunk length",
3906 * The icreate transaction can cover multiple cluster buffers and these
3907 * buffers could have been freed and reused. Check the individual
3908 * buffers for cancellation so we don't overwrite anything written after
3911 blks_per_cluster = xfs_icluster_size_fsb(mp);
3912 bb_per_cluster = XFS_FSB_TO_BB(mp, blks_per_cluster);
3913 nbufs = length / blks_per_cluster;
3914 for (i = 0, cancel_count = 0; i < nbufs; i++) {
3917 daddr = XFS_AGB_TO_DADDR(mp, agno,
3918 agbno + i * blks_per_cluster);
3919 if (xlog_check_buffer_cancelled(log, daddr, bb_per_cluster, 0))
3924 * We currently only use icreate for a single allocation at a time. This
3925 * means we should expect either all or none of the buffers to be
3926 * cancelled. Be conservative and skip replay if at least one buffer is
3927 * cancelled, but warn the user that something is awry if the buffers
3928 * are not consistent.
3930 * XXX: This must be refined to only skip cancelled clusters once we use
3931 * icreate for multiple chunk allocations.
3933 ASSERT(!cancel_count || cancel_count == nbufs);
3935 if (cancel_count != nbufs)
3937 "WARNING: partial inode chunk cancellation, skipped icreate.");
3938 trace_xfs_log_recover_icreate_cancel(log, icl);
3942 trace_xfs_log_recover_icreate_recover(log, icl);
3943 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
3944 length, be32_to_cpu(icl->icl_gen));
3948 xlog_recover_buffer_ra_pass2(
3950 struct xlog_recover_item *item)
3952 struct xfs_buf_log_format *buf_f = item->ri_buf[0].i_addr;
3953 struct xfs_mount *mp = log->l_mp;
3955 if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3956 buf_f->blf_len, buf_f->blf_flags)) {
3960 xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3961 buf_f->blf_len, NULL);
3965 xlog_recover_inode_ra_pass2(
3967 struct xlog_recover_item *item)
3969 struct xfs_inode_log_format ilf_buf;
3970 struct xfs_inode_log_format *ilfp;
3971 struct xfs_mount *mp = log->l_mp;
3974 if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3975 ilfp = item->ri_buf[0].i_addr;
3978 memset(ilfp, 0, sizeof(*ilfp));
3979 error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3984 if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3987 xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3988 ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3992 xlog_recover_dquot_ra_pass2(
3994 struct xlog_recover_item *item)
3996 struct xfs_mount *mp = log->l_mp;
3997 struct xfs_disk_dquot *recddq;
3998 struct xfs_dq_logformat *dq_f;
4003 if (mp->m_qflags == 0)
4006 recddq = item->ri_buf[1].i_addr;
4009 if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
4012 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
4014 if (log->l_quotaoffs_flag & type)
4017 dq_f = item->ri_buf[0].i_addr;
4019 ASSERT(dq_f->qlf_len == 1);
4021 len = XFS_FSB_TO_BB(mp, dq_f->qlf_len);
4022 if (xlog_peek_buffer_cancelled(log, dq_f->qlf_blkno, len, 0))
4025 xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno, len,
4026 &xfs_dquot_buf_ra_ops);
4030 xlog_recover_ra_pass2(
4032 struct xlog_recover_item *item)
4034 switch (ITEM_TYPE(item)) {
4036 xlog_recover_buffer_ra_pass2(log, item);
4039 xlog_recover_inode_ra_pass2(log, item);
4042 xlog_recover_dquot_ra_pass2(log, item);
4046 case XFS_LI_QUOTAOFF:
4059 xlog_recover_commit_pass1(
4061 struct xlog_recover *trans,
4062 struct xlog_recover_item *item)
4064 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
4066 switch (ITEM_TYPE(item)) {
4068 return xlog_recover_buffer_pass1(log, item);
4069 case XFS_LI_QUOTAOFF:
4070 return xlog_recover_quotaoff_pass1(log, item);
4075 case XFS_LI_ICREATE:
4082 /* nothing to do in pass 1 */
4085 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4086 __func__, ITEM_TYPE(item));
4093 xlog_recover_commit_pass2(
4095 struct xlog_recover *trans,
4096 struct list_head *buffer_list,
4097 struct xlog_recover_item *item)
4099 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
4101 switch (ITEM_TYPE(item)) {
4103 return xlog_recover_buffer_pass2(log, buffer_list, item,
4106 return xlog_recover_inode_pass2(log, buffer_list, item,
4109 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
4111 return xlog_recover_efd_pass2(log, item);
4113 return xlog_recover_rui_pass2(log, item, trans->r_lsn);
4115 return xlog_recover_rud_pass2(log, item);
4117 return xlog_recover_cui_pass2(log, item, trans->r_lsn);
4119 return xlog_recover_cud_pass2(log, item);
4121 return xlog_recover_bui_pass2(log, item, trans->r_lsn);
4123 return xlog_recover_bud_pass2(log, item);
4125 return xlog_recover_dquot_pass2(log, buffer_list, item,
4127 case XFS_LI_ICREATE:
4128 return xlog_recover_do_icreate_pass2(log, buffer_list, item);
4129 case XFS_LI_QUOTAOFF:
4130 /* nothing to do in pass2 */
4133 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
4134 __func__, ITEM_TYPE(item));
4141 xlog_recover_items_pass2(
4143 struct xlog_recover *trans,
4144 struct list_head *buffer_list,
4145 struct list_head *item_list)
4147 struct xlog_recover_item *item;
4150 list_for_each_entry(item, item_list, ri_list) {
4151 error = xlog_recover_commit_pass2(log, trans,
4161 * Perform the transaction.
4163 * If the transaction modifies a buffer or inode, do it now. Otherwise,
4164 * EFIs and EFDs get queued up by adding entries into the AIL for them.
4167 xlog_recover_commit_trans(
4169 struct xlog_recover *trans,
4171 struct list_head *buffer_list)
4174 int items_queued = 0;
4175 struct xlog_recover_item *item;
4176 struct xlog_recover_item *next;
4177 LIST_HEAD (ra_list);
4178 LIST_HEAD (done_list);
4180 #define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
4182 hlist_del_init(&trans->r_list);
4184 error = xlog_recover_reorder_trans(log, trans, pass);
4188 list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
4190 case XLOG_RECOVER_PASS1:
4191 error = xlog_recover_commit_pass1(log, trans, item);
4193 case XLOG_RECOVER_PASS2:
4194 xlog_recover_ra_pass2(log, item);
4195 list_move_tail(&item->ri_list, &ra_list);
4197 if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
4198 error = xlog_recover_items_pass2(log, trans,
4199 buffer_list, &ra_list);
4200 list_splice_tail_init(&ra_list, &done_list);
4214 if (!list_empty(&ra_list)) {
4216 error = xlog_recover_items_pass2(log, trans,
4217 buffer_list, &ra_list);
4218 list_splice_tail_init(&ra_list, &done_list);
4221 if (!list_empty(&done_list))
4222 list_splice_init(&done_list, &trans->r_itemq);
4228 xlog_recover_add_item(
4229 struct list_head *head)
4231 xlog_recover_item_t *item;
4233 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
4234 INIT_LIST_HEAD(&item->ri_list);
4235 list_add_tail(&item->ri_list, head);
4239 xlog_recover_add_to_cont_trans(
4241 struct xlog_recover *trans,
4245 xlog_recover_item_t *item;
4246 char *ptr, *old_ptr;
4250 * If the transaction is empty, the header was split across this and the
4251 * previous record. Copy the rest of the header.
4253 if (list_empty(&trans->r_itemq)) {
4254 ASSERT(len <= sizeof(struct xfs_trans_header));
4255 if (len > sizeof(struct xfs_trans_header)) {
4256 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4260 xlog_recover_add_item(&trans->r_itemq);
4261 ptr = (char *)&trans->r_theader +
4262 sizeof(struct xfs_trans_header) - len;
4263 memcpy(ptr, dp, len);
4267 /* take the tail entry */
4268 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4270 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
4271 old_len = item->ri_buf[item->ri_cnt-1].i_len;
4273 ptr = kmem_realloc(old_ptr, len + old_len, KM_SLEEP);
4274 memcpy(&ptr[old_len], dp, len);
4275 item->ri_buf[item->ri_cnt-1].i_len += len;
4276 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
4277 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
4282 * The next region to add is the start of a new region. It could be
4283 * a whole region or it could be the first part of a new region. Because
4284 * of this, the assumption here is that the type and size fields of all
4285 * format structures fit into the first 32 bits of the structure.
4287 * This works because all regions must be 32 bit aligned. Therefore, we
4288 * either have both fields or we have neither field. In the case we have
4289 * neither field, the data part of the region is zero length. We only have
4290 * a log_op_header and can throw away the header since a new one will appear
4291 * later. If we have at least 4 bytes, then we can determine how many regions
4292 * will appear in the current log item.
4295 xlog_recover_add_to_trans(
4297 struct xlog_recover *trans,
4301 xfs_inode_log_format_t *in_f; /* any will do */
4302 xlog_recover_item_t *item;
4307 if (list_empty(&trans->r_itemq)) {
4308 /* we need to catch log corruptions here */
4309 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
4310 xfs_warn(log->l_mp, "%s: bad header magic number",
4316 if (len > sizeof(struct xfs_trans_header)) {
4317 xfs_warn(log->l_mp, "%s: bad header length", __func__);
4323 * The transaction header can be arbitrarily split across op
4324 * records. If we don't have the whole thing here, copy what we
4325 * do have and handle the rest in the next record.
4327 if (len == sizeof(struct xfs_trans_header))
4328 xlog_recover_add_item(&trans->r_itemq);
4329 memcpy(&trans->r_theader, dp, len);
4333 ptr = kmem_alloc(len, KM_SLEEP);
4334 memcpy(ptr, dp, len);
4335 in_f = (xfs_inode_log_format_t *)ptr;
4337 /* take the tail entry */
4338 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
4339 if (item->ri_total != 0 &&
4340 item->ri_total == item->ri_cnt) {
4341 /* tail item is in use, get a new one */
4342 xlog_recover_add_item(&trans->r_itemq);
4343 item = list_entry(trans->r_itemq.prev,
4344 xlog_recover_item_t, ri_list);
4347 if (item->ri_total == 0) { /* first region to be added */
4348 if (in_f->ilf_size == 0 ||
4349 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
4351 "bad number of regions (%d) in inode log format",
4358 item->ri_total = in_f->ilf_size;
4360 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
4363 ASSERT(item->ri_total > item->ri_cnt);
4364 /* Description region is ri_buf[0] */
4365 item->ri_buf[item->ri_cnt].i_addr = ptr;
4366 item->ri_buf[item->ri_cnt].i_len = len;
4368 trace_xfs_log_recover_item_add(log, trans, item, 0);
4373 * Free up any resources allocated by the transaction
4375 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
4378 xlog_recover_free_trans(
4379 struct xlog_recover *trans)
4381 xlog_recover_item_t *item, *n;
4384 hlist_del_init(&trans->r_list);
4386 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
4387 /* Free the regions in the item. */
4388 list_del(&item->ri_list);
4389 for (i = 0; i < item->ri_cnt; i++)
4390 kmem_free(item->ri_buf[i].i_addr);
4391 /* Free the item itself */
4392 kmem_free(item->ri_buf);
4395 /* Free the transaction recover structure */
4400 * On error or completion, trans is freed.
4403 xlog_recovery_process_trans(
4405 struct xlog_recover *trans,
4410 struct list_head *buffer_list)
4413 bool freeit = false;
4415 /* mask off ophdr transaction container flags */
4416 flags &= ~XLOG_END_TRANS;
4417 if (flags & XLOG_WAS_CONT_TRANS)
4418 flags &= ~XLOG_CONTINUE_TRANS;
4421 * Callees must not free the trans structure. We'll decide if we need to
4422 * free it or not based on the operation being done and it's result.
4425 /* expected flag values */
4427 case XLOG_CONTINUE_TRANS:
4428 error = xlog_recover_add_to_trans(log, trans, dp, len);
4430 case XLOG_WAS_CONT_TRANS:
4431 error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
4433 case XLOG_COMMIT_TRANS:
4434 error = xlog_recover_commit_trans(log, trans, pass,
4436 /* success or fail, we are now done with this transaction. */
4440 /* unexpected flag values */
4441 case XLOG_UNMOUNT_TRANS:
4442 /* just skip trans */
4443 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
4446 case XLOG_START_TRANS:
4448 xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
4453 if (error || freeit)
4454 xlog_recover_free_trans(trans);
4459 * Lookup the transaction recovery structure associated with the ID in the
4460 * current ophdr. If the transaction doesn't exist and the start flag is set in
4461 * the ophdr, then allocate a new transaction for future ID matches to find.
4462 * Either way, return what we found during the lookup - an existing transaction
4465 STATIC struct xlog_recover *
4466 xlog_recover_ophdr_to_trans(
4467 struct hlist_head rhash[],
4468 struct xlog_rec_header *rhead,
4469 struct xlog_op_header *ohead)
4471 struct xlog_recover *trans;
4473 struct hlist_head *rhp;
4475 tid = be32_to_cpu(ohead->oh_tid);
4476 rhp = &rhash[XLOG_RHASH(tid)];
4477 hlist_for_each_entry(trans, rhp, r_list) {
4478 if (trans->r_log_tid == tid)
4483 * skip over non-start transaction headers - we could be
4484 * processing slack space before the next transaction starts
4486 if (!(ohead->oh_flags & XLOG_START_TRANS))
4489 ASSERT(be32_to_cpu(ohead->oh_len) == 0);
4492 * This is a new transaction so allocate a new recovery container to
4493 * hold the recovery ops that will follow.
4495 trans = kmem_zalloc(sizeof(struct xlog_recover), KM_SLEEP);
4496 trans->r_log_tid = tid;
4497 trans->r_lsn = be64_to_cpu(rhead->h_lsn);
4498 INIT_LIST_HEAD(&trans->r_itemq);
4499 INIT_HLIST_NODE(&trans->r_list);
4500 hlist_add_head(&trans->r_list, rhp);
4503 * Nothing more to do for this ophdr. Items to be added to this new
4504 * transaction will be in subsequent ophdr containers.
4510 xlog_recover_process_ophdr(
4512 struct hlist_head rhash[],
4513 struct xlog_rec_header *rhead,
4514 struct xlog_op_header *ohead,
4518 struct list_head *buffer_list)
4520 struct xlog_recover *trans;
4524 /* Do we understand who wrote this op? */
4525 if (ohead->oh_clientid != XFS_TRANSACTION &&
4526 ohead->oh_clientid != XFS_LOG) {
4527 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
4528 __func__, ohead->oh_clientid);
4534 * Check the ophdr contains all the data it is supposed to contain.
4536 len = be32_to_cpu(ohead->oh_len);
4537 if (dp + len > end) {
4538 xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
4543 trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
4545 /* nothing to do, so skip over this ophdr */
4550 * The recovered buffer queue is drained only once we know that all
4551 * recovery items for the current LSN have been processed. This is
4554 * - Buffer write submission updates the metadata LSN of the buffer.
4555 * - Log recovery skips items with a metadata LSN >= the current LSN of
4556 * the recovery item.
4557 * - Separate recovery items against the same metadata buffer can share
4558 * a current LSN. I.e., consider that the LSN of a recovery item is
4559 * defined as the starting LSN of the first record in which its
4560 * transaction appears, that a record can hold multiple transactions,
4561 * and/or that a transaction can span multiple records.
4563 * In other words, we are allowed to submit a buffer from log recovery
4564 * once per current LSN. Otherwise, we may incorrectly skip recovery
4565 * items and cause corruption.
4567 * We don't know up front whether buffers are updated multiple times per
4568 * LSN. Therefore, track the current LSN of each commit log record as it
4569 * is processed and drain the queue when it changes. Use commit records
4570 * because they are ordered correctly by the logging code.
4572 if (log->l_recovery_lsn != trans->r_lsn &&
4573 ohead->oh_flags & XLOG_COMMIT_TRANS) {
4574 error = xfs_buf_delwri_submit(buffer_list);
4577 log->l_recovery_lsn = trans->r_lsn;
4580 return xlog_recovery_process_trans(log, trans, dp, len,
4581 ohead->oh_flags, pass, buffer_list);
4585 * There are two valid states of the r_state field. 0 indicates that the
4586 * transaction structure is in a normal state. We have either seen the
4587 * start of the transaction or the last operation we added was not a partial
4588 * operation. If the last operation we added to the transaction was a
4589 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
4591 * NOTE: skip LRs with 0 data length.
4594 xlog_recover_process_data(
4596 struct hlist_head rhash[],
4597 struct xlog_rec_header *rhead,
4600 struct list_head *buffer_list)
4602 struct xlog_op_header *ohead;
4607 end = dp + be32_to_cpu(rhead->h_len);
4608 num_logops = be32_to_cpu(rhead->h_num_logops);
4610 /* check the log format matches our own - else we can't recover */
4611 if (xlog_header_check_recover(log->l_mp, rhead))
4614 trace_xfs_log_recover_record(log, rhead, pass);
4615 while ((dp < end) && num_logops) {
4617 ohead = (struct xlog_op_header *)dp;
4618 dp += sizeof(*ohead);
4621 /* errors will abort recovery */
4622 error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
4623 dp, end, pass, buffer_list);
4627 dp += be32_to_cpu(ohead->oh_len);
4633 /* Recover the EFI if necessary. */
4635 xlog_recover_process_efi(
4636 struct xfs_mount *mp,
4637 struct xfs_ail *ailp,
4638 struct xfs_log_item *lip)
4640 struct xfs_efi_log_item *efip;
4644 * Skip EFIs that we've already processed.
4646 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4647 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags))
4650 spin_unlock(&ailp->xa_lock);
4651 error = xfs_efi_recover(mp, efip);
4652 spin_lock(&ailp->xa_lock);
4657 /* Release the EFI since we're cancelling everything. */
4659 xlog_recover_cancel_efi(
4660 struct xfs_mount *mp,
4661 struct xfs_ail *ailp,
4662 struct xfs_log_item *lip)
4664 struct xfs_efi_log_item *efip;
4666 efip = container_of(lip, struct xfs_efi_log_item, efi_item);
4668 spin_unlock(&ailp->xa_lock);
4669 xfs_efi_release(efip);
4670 spin_lock(&ailp->xa_lock);
4673 /* Recover the RUI if necessary. */
4675 xlog_recover_process_rui(
4676 struct xfs_mount *mp,
4677 struct xfs_ail *ailp,
4678 struct xfs_log_item *lip)
4680 struct xfs_rui_log_item *ruip;
4684 * Skip RUIs that we've already processed.
4686 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4687 if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
4690 spin_unlock(&ailp->xa_lock);
4691 error = xfs_rui_recover(mp, ruip);
4692 spin_lock(&ailp->xa_lock);
4697 /* Release the RUI since we're cancelling everything. */
4699 xlog_recover_cancel_rui(
4700 struct xfs_mount *mp,
4701 struct xfs_ail *ailp,
4702 struct xfs_log_item *lip)
4704 struct xfs_rui_log_item *ruip;
4706 ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
4708 spin_unlock(&ailp->xa_lock);
4709 xfs_rui_release(ruip);
4710 spin_lock(&ailp->xa_lock);
4713 /* Recover the CUI if necessary. */
4715 xlog_recover_process_cui(
4716 struct xfs_mount *mp,
4717 struct xfs_ail *ailp,
4718 struct xfs_log_item *lip,
4719 struct xfs_defer_ops *dfops)
4721 struct xfs_cui_log_item *cuip;
4725 * Skip CUIs that we've already processed.
4727 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4728 if (test_bit(XFS_CUI_RECOVERED, &cuip->cui_flags))
4731 spin_unlock(&ailp->xa_lock);
4732 error = xfs_cui_recover(mp, cuip, dfops);
4733 spin_lock(&ailp->xa_lock);
4738 /* Release the CUI since we're cancelling everything. */
4740 xlog_recover_cancel_cui(
4741 struct xfs_mount *mp,
4742 struct xfs_ail *ailp,
4743 struct xfs_log_item *lip)
4745 struct xfs_cui_log_item *cuip;
4747 cuip = container_of(lip, struct xfs_cui_log_item, cui_item);
4749 spin_unlock(&ailp->xa_lock);
4750 xfs_cui_release(cuip);
4751 spin_lock(&ailp->xa_lock);
4754 /* Recover the BUI if necessary. */
4756 xlog_recover_process_bui(
4757 struct xfs_mount *mp,
4758 struct xfs_ail *ailp,
4759 struct xfs_log_item *lip,
4760 struct xfs_defer_ops *dfops)
4762 struct xfs_bui_log_item *buip;
4766 * Skip BUIs that we've already processed.
4768 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4769 if (test_bit(XFS_BUI_RECOVERED, &buip->bui_flags))
4772 spin_unlock(&ailp->xa_lock);
4773 error = xfs_bui_recover(mp, buip, dfops);
4774 spin_lock(&ailp->xa_lock);
4779 /* Release the BUI since we're cancelling everything. */
4781 xlog_recover_cancel_bui(
4782 struct xfs_mount *mp,
4783 struct xfs_ail *ailp,
4784 struct xfs_log_item *lip)
4786 struct xfs_bui_log_item *buip;
4788 buip = container_of(lip, struct xfs_bui_log_item, bui_item);
4790 spin_unlock(&ailp->xa_lock);
4791 xfs_bui_release(buip);
4792 spin_lock(&ailp->xa_lock);
4795 /* Is this log item a deferred action intent? */
4796 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
4798 switch (lip->li_type) {
4809 /* Take all the collected deferred ops and finish them in order. */
4811 xlog_finish_defer_ops(
4812 struct xfs_mount *mp,
4813 struct xfs_defer_ops *dfops)
4815 struct xfs_trans *tp;
4821 * We're finishing the defer_ops that accumulated as a result of
4822 * recovering unfinished intent items during log recovery. We
4823 * reserve an itruncate transaction because it is the largest
4824 * permanent transaction type. Since we're the only user of the fs
4825 * right now, take 93% (15/16) of the available free blocks. Use
4826 * weird math to avoid a 64-bit division.
4828 freeblks = percpu_counter_sum(&mp->m_fdblocks);
4831 resblks = min_t(int64_t, UINT_MAX, freeblks);
4832 resblks = (resblks * 15) >> 4;
4833 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
4834 0, XFS_TRANS_RESERVE, &tp);
4838 error = xfs_defer_finish(&tp, dfops);
4842 return xfs_trans_commit(tp);
4845 xfs_trans_cancel(tp);
4850 * When this is called, all of the log intent items which did not have
4851 * corresponding log done items should be in the AIL. What we do now
4852 * is update the data structures associated with each one.
4854 * Since we process the log intent items in normal transactions, they
4855 * will be removed at some point after the commit. This prevents us
4856 * from just walking down the list processing each one. We'll use a
4857 * flag in the intent item to skip those that we've already processed
4858 * and use the AIL iteration mechanism's generation count to try to
4859 * speed this up at least a bit.
4861 * When we start, we know that the intents are the only things in the
4862 * AIL. As we process them, however, other items are added to the
4866 xlog_recover_process_intents(
4869 struct xfs_defer_ops dfops;
4870 struct xfs_ail_cursor cur;
4871 struct xfs_log_item *lip;
4872 struct xfs_ail *ailp;
4873 xfs_fsblock_t firstfsb;
4875 #if defined(DEBUG) || defined(XFS_WARN)
4880 spin_lock(&ailp->xa_lock);
4881 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4882 #if defined(DEBUG) || defined(XFS_WARN)
4883 last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
4885 xfs_defer_init(&dfops, &firstfsb);
4886 while (lip != NULL) {
4888 * We're done when we see something other than an intent.
4889 * There should be no intents left in the AIL now.
4891 if (!xlog_item_is_intent(lip)) {
4893 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4894 ASSERT(!xlog_item_is_intent(lip));
4900 * We should never see a redo item with a LSN higher than
4901 * the last transaction we found in the log at the start
4904 ASSERT(XFS_LSN_CMP(last_lsn, lip->li_lsn) >= 0);
4907 * NOTE: If your intent processing routine can create more
4908 * deferred ops, you /must/ attach them to the dfops in this
4909 * routine or else those subsequent intents will get
4910 * replayed in the wrong order!
4912 switch (lip->li_type) {
4914 error = xlog_recover_process_efi(log->l_mp, ailp, lip);
4917 error = xlog_recover_process_rui(log->l_mp, ailp, lip);
4920 error = xlog_recover_process_cui(log->l_mp, ailp, lip,
4924 error = xlog_recover_process_bui(log->l_mp, ailp, lip,
4930 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4933 xfs_trans_ail_cursor_done(&cur);
4934 spin_unlock(&ailp->xa_lock);
4936 xfs_defer_cancel(&dfops);
4938 error = xlog_finish_defer_ops(log->l_mp, &dfops);
4944 * A cancel occurs when the mount has failed and we're bailing out.
4945 * Release all pending log intent items so they don't pin the AIL.
4948 xlog_recover_cancel_intents(
4951 struct xfs_log_item *lip;
4953 struct xfs_ail_cursor cur;
4954 struct xfs_ail *ailp;
4957 spin_lock(&ailp->xa_lock);
4958 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
4959 while (lip != NULL) {
4961 * We're done when we see something other than an intent.
4962 * There should be no intents left in the AIL now.
4964 if (!xlog_item_is_intent(lip)) {
4966 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
4967 ASSERT(!xlog_item_is_intent(lip));
4972 switch (lip->li_type) {
4974 xlog_recover_cancel_efi(log->l_mp, ailp, lip);
4977 xlog_recover_cancel_rui(log->l_mp, ailp, lip);
4980 xlog_recover_cancel_cui(log->l_mp, ailp, lip);
4983 xlog_recover_cancel_bui(log->l_mp, ailp, lip);
4987 lip = xfs_trans_ail_cursor_next(ailp, &cur);
4990 xfs_trans_ail_cursor_done(&cur);
4991 spin_unlock(&ailp->xa_lock);
4996 * This routine performs a transaction to null out a bad inode pointer
4997 * in an agi unlinked inode hash bucket.
5000 xlog_recover_clear_agi_bucket(
5002 xfs_agnumber_t agno,
5011 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
5015 error = xfs_read_agi(mp, tp, agno, &agibp);
5019 agi = XFS_BUF_TO_AGI(agibp);
5020 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
5021 offset = offsetof(xfs_agi_t, agi_unlinked) +
5022 (sizeof(xfs_agino_t) * bucket);
5023 xfs_trans_log_buf(tp, agibp, offset,
5024 (offset + sizeof(xfs_agino_t) - 1));
5026 error = xfs_trans_commit(tp);
5032 xfs_trans_cancel(tp);
5034 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
5039 xlog_recover_process_one_iunlink(
5040 struct xfs_mount *mp,
5041 xfs_agnumber_t agno,
5045 struct xfs_buf *ibp;
5046 struct xfs_dinode *dip;
5047 struct xfs_inode *ip;
5051 ino = XFS_AGINO_TO_INO(mp, agno, agino);
5052 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
5057 * Get the on disk inode to find the next inode in the bucket.
5059 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
5063 xfs_iflags_clear(ip, XFS_IRECOVERY);
5064 ASSERT(VFS_I(ip)->i_nlink == 0);
5065 ASSERT(VFS_I(ip)->i_mode != 0);
5067 /* setup for the next pass */
5068 agino = be32_to_cpu(dip->di_next_unlinked);
5072 * Prevent any DMAPI event from being sent when the reference on
5073 * the inode is dropped.
5075 ip->i_d.di_dmevmask = 0;
5084 * We can't read in the inode this bucket points to, or this inode
5085 * is messed up. Just ditch this bucket of inodes. We will lose
5086 * some inodes and space, but at least we won't hang.
5088 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
5089 * clear the inode pointer in the bucket.
5091 xlog_recover_clear_agi_bucket(mp, agno, bucket);
5096 * xlog_iunlink_recover
5098 * This is called during recovery to process any inodes which
5099 * we unlinked but not freed when the system crashed. These
5100 * inodes will be on the lists in the AGI blocks. What we do
5101 * here is scan all the AGIs and fully truncate and free any
5102 * inodes found on the lists. Each inode is removed from the
5103 * lists when it has been fully truncated and is freed. The
5104 * freeing of the inode and its removal from the list must be
5108 xlog_recover_process_iunlinks(
5112 xfs_agnumber_t agno;
5123 * Prevent any DMAPI event from being sent while in this function.
5125 mp_dmevmask = mp->m_dmevmask;
5128 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5130 * Find the agi for this ag.
5132 error = xfs_read_agi(mp, NULL, agno, &agibp);
5135 * AGI is b0rked. Don't process it.
5137 * We should probably mark the filesystem as corrupt
5138 * after we've recovered all the ag's we can....
5143 * Unlock the buffer so that it can be acquired in the normal
5144 * course of the transaction to truncate and free each inode.
5145 * Because we are not racing with anyone else here for the AGI
5146 * buffer, we don't even need to hold it locked to read the
5147 * initial unlinked bucket entries out of the buffer. We keep
5148 * buffer reference though, so that it stays pinned in memory
5149 * while we need the buffer.
5151 agi = XFS_BUF_TO_AGI(agibp);
5152 xfs_buf_unlock(agibp);
5154 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
5155 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
5156 while (agino != NULLAGINO) {
5157 agino = xlog_recover_process_one_iunlink(mp,
5158 agno, agino, bucket);
5161 xfs_buf_rele(agibp);
5164 mp->m_dmevmask = mp_dmevmask;
5169 struct xlog_rec_header *rhead,
5175 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
5176 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
5177 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
5181 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5182 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
5183 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
5184 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5185 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
5186 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
5195 * CRC check, unpack and process a log record.
5198 xlog_recover_process(
5200 struct hlist_head rhash[],
5201 struct xlog_rec_header *rhead,
5204 struct list_head *buffer_list)
5207 __le32 old_crc = rhead->h_crc;
5211 crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
5214 * Nothing else to do if this is a CRC verification pass. Just return
5215 * if this a record with a non-zero crc. Unfortunately, mkfs always
5216 * sets old_crc to 0 so we must consider this valid even on v5 supers.
5217 * Otherwise, return EFSBADCRC on failure so the callers up the stack
5218 * know precisely what failed.
5220 if (pass == XLOG_RECOVER_CRCPASS) {
5221 if (old_crc && crc != old_crc)
5227 * We're in the normal recovery path. Issue a warning if and only if the
5228 * CRC in the header is non-zero. This is an advisory warning and the
5229 * zero CRC check prevents warnings from being emitted when upgrading
5230 * the kernel from one that does not add CRCs by default.
5232 if (crc != old_crc) {
5233 if (old_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
5234 xfs_alert(log->l_mp,
5235 "log record CRC mismatch: found 0x%x, expected 0x%x.",
5236 le32_to_cpu(old_crc),
5238 xfs_hex_dump(dp, 32);
5242 * If the filesystem is CRC enabled, this mismatch becomes a
5243 * fatal log corruption failure.
5245 if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
5246 return -EFSCORRUPTED;
5249 error = xlog_unpack_data(rhead, dp, log);
5253 return xlog_recover_process_data(log, rhash, rhead, dp, pass,
5258 xlog_valid_rec_header(
5260 struct xlog_rec_header *rhead,
5265 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
5266 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
5267 XFS_ERRLEVEL_LOW, log->l_mp);
5268 return -EFSCORRUPTED;
5271 (!rhead->h_version ||
5272 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
5273 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
5274 __func__, be32_to_cpu(rhead->h_version));
5278 /* LR body must have data or it wouldn't have been written */
5279 hlen = be32_to_cpu(rhead->h_len);
5280 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
5281 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
5282 XFS_ERRLEVEL_LOW, log->l_mp);
5283 return -EFSCORRUPTED;
5285 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
5286 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
5287 XFS_ERRLEVEL_LOW, log->l_mp);
5288 return -EFSCORRUPTED;
5294 * Read the log from tail to head and process the log records found.
5295 * Handle the two cases where the tail and head are in the same cycle
5296 * and where the active portion of the log wraps around the end of
5297 * the physical log separately. The pass parameter is passed through
5298 * to the routines called to process the data and is not looked at
5302 xlog_do_recovery_pass(
5304 xfs_daddr_t head_blk,
5305 xfs_daddr_t tail_blk,
5307 xfs_daddr_t *first_bad) /* out: first bad log rec */
5309 xlog_rec_header_t *rhead;
5310 xfs_daddr_t blk_no, rblk_no;
5311 xfs_daddr_t rhead_blk;
5313 xfs_buf_t *hbp, *dbp;
5314 int error = 0, h_size, h_len;
5316 int bblks, split_bblks;
5317 int hblks, split_hblks, wrapped_hblks;
5319 struct hlist_head rhash[XLOG_RHASH_SIZE];
5320 LIST_HEAD (buffer_list);
5322 ASSERT(head_blk != tail_blk);
5323 blk_no = rhead_blk = tail_blk;
5325 for (i = 0; i < XLOG_RHASH_SIZE; i++)
5326 INIT_HLIST_HEAD(&rhash[i]);
5329 * Read the header of the tail block and get the iclog buffer size from
5330 * h_size. Use this to tell how many sectors make up the log header.
5332 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
5334 * When using variable length iclogs, read first sector of
5335 * iclog header and extract the header size from it. Get a
5336 * new hbp that is the correct size.
5338 hbp = xlog_get_bp(log, 1);
5342 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
5346 rhead = (xlog_rec_header_t *)offset;
5347 error = xlog_valid_rec_header(log, rhead, tail_blk);
5352 * xfsprogs has a bug where record length is based on lsunit but
5353 * h_size (iclog size) is hardcoded to 32k. Now that we
5354 * unconditionally CRC verify the unmount record, this means the
5355 * log buffer can be too small for the record and cause an
5358 * Detect this condition here. Use lsunit for the buffer size as
5359 * long as this looks like the mkfs case. Otherwise, return an
5360 * error to avoid a buffer overrun.
5362 h_size = be32_to_cpu(rhead->h_size);
5363 h_len = be32_to_cpu(rhead->h_len);
5364 if (h_len > h_size) {
5365 if (h_len <= log->l_mp->m_logbsize &&
5366 be32_to_cpu(rhead->h_num_logops) == 1) {
5368 "invalid iclog size (%d bytes), using lsunit (%d bytes)",
5369 h_size, log->l_mp->m_logbsize);
5370 h_size = log->l_mp->m_logbsize;
5372 return -EFSCORRUPTED;
5375 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
5376 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
5377 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
5378 if (h_size % XLOG_HEADER_CYCLE_SIZE)
5381 hbp = xlog_get_bp(log, hblks);
5386 ASSERT(log->l_sectBBsize == 1);
5388 hbp = xlog_get_bp(log, 1);
5389 h_size = XLOG_BIG_RECORD_BSIZE;
5394 dbp = xlog_get_bp(log, BTOBB(h_size));
5400 memset(rhash, 0, sizeof(rhash));
5401 if (tail_blk > head_blk) {
5403 * Perform recovery around the end of the physical log.
5404 * When the head is not on the same cycle number as the tail,
5405 * we can't do a sequential recovery.
5407 while (blk_no < log->l_logBBsize) {
5409 * Check for header wrapping around physical end-of-log
5411 offset = hbp->b_addr;
5414 if (blk_no + hblks <= log->l_logBBsize) {
5415 /* Read header in one read */
5416 error = xlog_bread(log, blk_no, hblks, hbp,
5421 /* This LR is split across physical log end */
5422 if (blk_no != log->l_logBBsize) {
5423 /* some data before physical log end */
5424 ASSERT(blk_no <= INT_MAX);
5425 split_hblks = log->l_logBBsize - (int)blk_no;
5426 ASSERT(split_hblks > 0);
5427 error = xlog_bread(log, blk_no,
5435 * Note: this black magic still works with
5436 * large sector sizes (non-512) only because:
5437 * - we increased the buffer size originally
5438 * by 1 sector giving us enough extra space
5439 * for the second read;
5440 * - the log start is guaranteed to be sector
5442 * - we read the log end (LR header start)
5443 * _first_, then the log start (LR header end)
5444 * - order is important.
5446 wrapped_hblks = hblks - split_hblks;
5447 error = xlog_bread_offset(log, 0,
5449 offset + BBTOB(split_hblks));
5453 rhead = (xlog_rec_header_t *)offset;
5454 error = xlog_valid_rec_header(log, rhead,
5455 split_hblks ? blk_no : 0);
5459 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5463 * Read the log record data in multiple reads if it
5464 * wraps around the end of the log. Note that if the
5465 * header already wrapped, blk_no could point past the
5466 * end of the log. The record data is contiguous in
5469 if (blk_no + bblks <= log->l_logBBsize ||
5470 blk_no >= log->l_logBBsize) {
5471 /* mod blk_no in case the header wrapped and
5472 * pushed it beyond the end of the log */
5473 rblk_no = do_mod(blk_no, log->l_logBBsize);
5474 error = xlog_bread(log, rblk_no, bblks, dbp,
5479 /* This log record is split across the
5480 * physical end of log */
5481 offset = dbp->b_addr;
5483 if (blk_no != log->l_logBBsize) {
5484 /* some data is before the physical
5486 ASSERT(!wrapped_hblks);
5487 ASSERT(blk_no <= INT_MAX);
5489 log->l_logBBsize - (int)blk_no;
5490 ASSERT(split_bblks > 0);
5491 error = xlog_bread(log, blk_no,
5499 * Note: this black magic still works with
5500 * large sector sizes (non-512) only because:
5501 * - we increased the buffer size originally
5502 * by 1 sector giving us enough extra space
5503 * for the second read;
5504 * - the log start is guaranteed to be sector
5506 * - we read the log end (LR header start)
5507 * _first_, then the log start (LR header end)
5508 * - order is important.
5510 error = xlog_bread_offset(log, 0,
5511 bblks - split_bblks, dbp,
5512 offset + BBTOB(split_bblks));
5517 error = xlog_recover_process(log, rhash, rhead, offset,
5518 pass, &buffer_list);
5526 ASSERT(blk_no >= log->l_logBBsize);
5527 blk_no -= log->l_logBBsize;
5531 /* read first part of physical log */
5532 while (blk_no < head_blk) {
5533 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
5537 rhead = (xlog_rec_header_t *)offset;
5538 error = xlog_valid_rec_header(log, rhead, blk_no);
5542 /* blocks in data section */
5543 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
5544 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
5549 error = xlog_recover_process(log, rhash, rhead, offset, pass,
5554 blk_no += bblks + hblks;
5564 * Submit buffers that have been added from the last record processed,
5565 * regardless of error status.
5567 if (!list_empty(&buffer_list))
5568 error2 = xfs_buf_delwri_submit(&buffer_list);
5570 if (error && first_bad)
5571 *first_bad = rhead_blk;
5574 * Transactions are freed at commit time but transactions without commit
5575 * records on disk are never committed. Free any that may be left in the
5578 for (i = 0; i < XLOG_RHASH_SIZE; i++) {
5579 struct hlist_node *tmp;
5580 struct xlog_recover *trans;
5582 hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
5583 xlog_recover_free_trans(trans);
5586 return error ? error : error2;
5590 * Do the recovery of the log. We actually do this in two phases.
5591 * The two passes are necessary in order to implement the function
5592 * of cancelling a record written into the log. The first pass
5593 * determines those things which have been cancelled, and the
5594 * second pass replays log items normally except for those which
5595 * have been cancelled. The handling of the replay and cancellations
5596 * takes place in the log item type specific routines.
5598 * The table of items which have cancel records in the log is allocated
5599 * and freed at this level, since only here do we know when all of
5600 * the log recovery has been completed.
5603 xlog_do_log_recovery(
5605 xfs_daddr_t head_blk,
5606 xfs_daddr_t tail_blk)
5610 ASSERT(head_blk != tail_blk);
5613 * First do a pass to find all of the cancelled buf log items.
5614 * Store them in the buf_cancel_table for use in the second pass.
5616 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
5617 sizeof(struct list_head),
5619 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5620 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
5622 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5623 XLOG_RECOVER_PASS1, NULL);
5625 kmem_free(log->l_buf_cancel_table);
5626 log->l_buf_cancel_table = NULL;
5630 * Then do a second pass to actually recover the items in the log.
5631 * When it is complete free the table of buf cancel items.
5633 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
5634 XLOG_RECOVER_PASS2, NULL);
5639 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
5640 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
5644 kmem_free(log->l_buf_cancel_table);
5645 log->l_buf_cancel_table = NULL;
5651 * Do the actual recovery
5656 xfs_daddr_t head_blk,
5657 xfs_daddr_t tail_blk)
5659 struct xfs_mount *mp = log->l_mp;
5664 trace_xfs_log_recover(log, head_blk, tail_blk);
5667 * First replay the images in the log.
5669 error = xlog_do_log_recovery(log, head_blk, tail_blk);
5674 * If IO errors happened during recovery, bail out.
5676 if (XFS_FORCED_SHUTDOWN(mp)) {
5681 * We now update the tail_lsn since much of the recovery has completed
5682 * and there may be space available to use. If there were no extent
5683 * or iunlinks, we can free up the entire log and set the tail_lsn to
5684 * be the last_sync_lsn. This was set in xlog_find_tail to be the
5685 * lsn of the last known good LR on disk. If there are extent frees
5686 * or iunlinks they will have some entries in the AIL; so we look at
5687 * the AIL to determine how to set the tail_lsn.
5689 xlog_assign_tail_lsn(mp);
5692 * Now that we've finished replaying all buffer and inode
5693 * updates, re-read in the superblock and reverify it.
5695 bp = xfs_getsb(mp, 0);
5696 bp->b_flags &= ~(XBF_DONE | XBF_ASYNC);
5697 ASSERT(!(bp->b_flags & XBF_WRITE));
5698 bp->b_flags |= XBF_READ;
5699 bp->b_ops = &xfs_sb_buf_ops;
5701 error = xfs_buf_submit_wait(bp);
5703 if (!XFS_FORCED_SHUTDOWN(mp)) {
5704 xfs_buf_ioerror_alert(bp, __func__);
5711 /* Convert superblock from on-disk format */
5713 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
5716 /* re-initialise in-core superblock and geometry structures */
5717 xfs_reinit_percpu_counters(mp);
5718 error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
5720 xfs_warn(mp, "Failed post-recovery per-ag init: %d", error);
5723 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
5725 xlog_recover_check_summary(log);
5727 /* Normal transactions can now occur */
5728 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
5733 * Perform recovery and re-initialize some log variables in xlog_find_tail.
5735 * Return error or zero.
5741 xfs_daddr_t head_blk, tail_blk;
5744 /* find the tail of the log */
5745 error = xlog_find_tail(log, &head_blk, &tail_blk);
5750 * The superblock was read before the log was available and thus the LSN
5751 * could not be verified. Check the superblock LSN against the current
5752 * LSN now that it's known.
5754 if (xfs_sb_version_hascrc(&log->l_mp->m_sb) &&
5755 !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
5758 if (tail_blk != head_blk) {
5759 /* There used to be a comment here:
5761 * disallow recovery on read-only mounts. note -- mount
5762 * checks for ENOSPC and turns it into an intelligent
5764 * ...but this is no longer true. Now, unless you specify
5765 * NORECOVERY (in which case this function would never be
5766 * called), we just go ahead and recover. We do this all
5767 * under the vfs layer, so we can get away with it unless
5768 * the device itself is read-only, in which case we fail.
5770 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
5775 * Version 5 superblock log feature mask validation. We know the
5776 * log is dirty so check if there are any unknown log features
5777 * in what we need to recover. If there are unknown features
5778 * (e.g. unsupported transactions, then simply reject the
5779 * attempt at recovery before touching anything.
5781 if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
5782 xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
5783 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
5785 "Superblock has unknown incompatible log features (0x%x) enabled.",
5786 (log->l_mp->m_sb.sb_features_log_incompat &
5787 XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
5789 "The log can not be fully and/or safely recovered by this kernel.");
5791 "Please recover the log on a kernel that supports the unknown features.");
5796 * Delay log recovery if the debug hook is set. This is debug
5797 * instrumention to coordinate simulation of I/O failures with
5800 if (xfs_globals.log_recovery_delay) {
5801 xfs_notice(log->l_mp,
5802 "Delaying log recovery for %d seconds.",
5803 xfs_globals.log_recovery_delay);
5804 msleep(xfs_globals.log_recovery_delay * 1000);
5807 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
5808 log->l_mp->m_logname ? log->l_mp->m_logname
5811 error = xlog_do_recover(log, head_blk, tail_blk);
5812 log->l_flags |= XLOG_RECOVERY_NEEDED;
5818 * In the first part of recovery we replay inodes and buffers and build
5819 * up the list of extent free items which need to be processed. Here
5820 * we process the extent free items and clean up the on disk unlinked
5821 * inode lists. This is separated from the first part of recovery so
5822 * that the root and real-time bitmap inodes can be read in from disk in
5823 * between the two stages. This is necessary so that we can free space
5824 * in the real-time portion of the file system.
5827 xlog_recover_finish(
5831 * Now we're ready to do the transactions needed for the
5832 * rest of recovery. Start with completing all the extent
5833 * free intent records and then process the unlinked inode
5834 * lists. At this point, we essentially run in normal mode
5835 * except that we're still performing recovery actions
5836 * rather than accepting new requests.
5838 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
5840 error = xlog_recover_process_intents(log);
5842 xfs_alert(log->l_mp, "Failed to recover intents");
5847 * Sync the log to get all the intents out of the AIL.
5848 * This isn't absolutely necessary, but it helps in
5849 * case the unlink transactions would have problems
5850 * pushing the intents out of the way.
5852 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
5854 xlog_recover_process_iunlinks(log);
5856 xlog_recover_check_summary(log);
5858 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
5859 log->l_mp->m_logname ? log->l_mp->m_logname
5861 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
5863 xfs_info(log->l_mp, "Ending clean mount");
5869 xlog_recover_cancel(
5874 if (log->l_flags & XLOG_RECOVERY_NEEDED)
5875 error = xlog_recover_cancel_intents(log);
5882 * Read all of the agf and agi counters and check that they
5883 * are consistent with the superblock counters.
5886 xlog_recover_check_summary(
5893 xfs_agnumber_t agno;
5904 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
5905 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
5907 xfs_alert(mp, "%s agf read failed agno %d error %d",
5908 __func__, agno, error);
5910 agfp = XFS_BUF_TO_AGF(agfbp);
5911 freeblks += be32_to_cpu(agfp->agf_freeblks) +
5912 be32_to_cpu(agfp->agf_flcount);
5913 xfs_buf_relse(agfbp);
5916 error = xfs_read_agi(mp, NULL, agno, &agibp);
5918 xfs_alert(mp, "%s agi read failed agno %d error %d",
5919 __func__, agno, error);
5921 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
5923 itotal += be32_to_cpu(agi->agi_count);
5924 ifree += be32_to_cpu(agi->agi_freecount);
5925 xfs_buf_relse(agibp);