[LU-14089] llog_reader assumes catalogs on non-ext fs is zfs

Thu, 29 Oct 2020 12:34:54 +0000

If for debugging purposes a lustre catalog has been copied out of its ldiskfs mount point (as suggested by the man page), llog_reader will print zfs-osd style paths when reading log

man page excerpt:

    debugfs -c -R 'dump CONFIGS/tfs-client /tmp/tfs-client' /dev/sda
    llog_reader /tmp/tfs-client

example

# mount -t ldiskfs /dev/mapper/testfs0--mdt0 /mnt/ldiskfs
# llog_reader /mnt/ldiskfs/changelog_users 
rec #1 type=1064553b len=64 offset 8192
Header size : 8192
Time : Thu Oct 29 11:25:58 2020
Number of records: 1
Target uuid : 
-----------------------
#01 (064)id=[0x43:0x1:0x0]:0 path=O/1/d3/67
# cp /mnt/ldiskfs/changelog_users /tmp/
# llog_reader /tmp/changelog_users 
rec #1 type=1064553b len=64 offset 8192
Header size : 8192
Time : Thu Oct 29 11:25:58 2020
Number of records: 1
Target uuid : 
-----------------------
#01 (064)id=[0x43:0x1:0x0]:0 path=oi.1/0x1:0x43:0x0

This can lead to confusing behaviour and should probably be documented, but more importantly I think the command could use a switch to manually specify what osd the catalog is meant for as currently the only way to get the correct output is to find or make an ext4/ldiskfs filesystem and copy the catalog there.

(code in lustre/utils/llog_reader.c, print_log_path and setting of is_ext directly in main)

Whamcloud Community JIRA

[LU-14089] llog_reader assumes catalogs on non-ext fs is zfs