[LU-19245] dir migrate doesn't update a regular file LOV lmm_oi with new file id

• Type: Bug
• Resolution: Unresolved
• Priority: Medium
• Severity: 3
• Assignee: Alexander Zarochentsev
• Reporter: Alexander Zarochentsev

Description

A dir migrate doesn't change regular file LOV attribute during migrate, so LFSCK starts to complain about wrong lmm_oi:

[175833.553728] Lustre: lustre-OST0001-osd: layout LFSCK slave notifies master
[175833.555342] Lustre: lustre-MDT0001-osd: layout LFSCK master found bad lmm_oi for [0x240000402:0x4:0x0]: rc = 240
[175833.555445] Lustre: lustre-MDT0001-osd: layout LFSCK master handles notify 3 from OST 1, status 1, flags 2, flags2 1

while the lmm_oi is mostly obsoleted nowadays, it still gets set during object creation and updated in layout swap.

the below is a demonstration of the bug:

[root@rocky94 tests]# rm -rf /mnt/lustre/foo*
[root@rocky94 tests]# ../utils/lfs mkdir -i 0 /mnt/lustre/foodir
[root@rocky94 tests]# ../utils/lfs setstripe -E 1M -c 1 -E -1 -c 2 /mnt/lustre/foodir/foo
[root@rocky94 tests]# getfattr   -n 'trusted.lov' /mnt/lustre/foodir/foo 
getfattr: Removing leading '/' from absolute path names
# file: mnt/lustre/foodir/foo
trusted.lov=0s0AvWC9gAAAACAAAAAAACAAAAAAAAAAAAAAAAAAAAAAABAAAAEAAAAAAAAAAAAAAAAAAQAAAAAACAAAAAOAAAAAAAAAAAAAAAAAAAAAAAAAACAAAAAAAAAAAAEAAAAAAA//////////+4AAAAIAAAAAAAAAAAAAAAAAAAAAAAAADQC9ELAQAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAQD//wEEAMACAAAAAwAAAAAAAAAAAAAAAQAAANAL0QsBAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAACAP//

[root@rocky94 tests]# ../utils/lfs migrate -m 1  /mnt/lustre/foodir
[root@rocky94 tests]# getfattr   -n 'trusted.lov' /mnt/lustre/foodir/foo 
getfattr: Removing leading '/' from absolute path names
# file: mnt/lustre/foodir/foo
trusted.lov=0s0AvWC9gAAAACAAAAAAACAAAAAAAAAAAAAAAAAAAAAAABAAAAEAAAAAAAAAAAAAAAAAAQAAAAAACAAAAAOAAAAAAAAAAAAAAAAAAAAAAAAAACAAAAAAAAAAAAEAAAAAAA//////////+4AAAAIAAAAAAAAAAAAAAAAAAAAAAAAADQC9ELAQAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAQD//wEEAMACAAAAAwAAAAAAAAAAAAAAAQAAANAL0QsBAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAACAP//

[root@rocky94 tests]#

the LOV attribute is not changed during migration as it should be, at least that lmm_oi should contain new value, an encoded new object FID which was changed by the migration procedure.

the file 'foo' indeed has a FID from MDT1:

[root@rocky94 tests]# ../utils/lfs path2fid  /mnt/lustre/foodir/foo 
[0x240000402:0x4:0x0]
[root@rocky94 tests]# 

and two lov attributes (taken before and after migration) are identical:

[root@rocky94 tests]# cmp -l  <(echo trusted.lov=0s0AvWC9gAAAACAAAAAAACAAAAAAAAAAAAAAAAAAAAAAABAAAAEAAAAAAAAAAAAAAAAAAQAAAAAACAAAAAOAAAAAAAAAAAAAAAAAAAAAAAAAACAAAAAAAAAAAAEAAAAAAA//////////+4AAAAIAAAAAAAAAAAAAAAAAAAAAAAAADQC9ELAQAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAQD//wEEAMACAAAAAwAAAAAAAAAAAAAAAQAAANAL0QsBAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAACAP//) <(echo trusted.lov=0s0AvWC9gAAAACAAAAAAACAAAAAAAAAAAAAAAAAAAAAAABAAAAEAAAAAAAAAAAAAAAAAAQAAAAAACAAAAAOAAAAAAAAAAAAAAAAAAAAAAAAAACAAAAAAAAAAAAEAAAAAAA//////////+4AAAAIAAAAAAAAAAAAAAAAAAAAAAAAADQC9ELAQAAAAAAAAAAAAAAAAAAAAAAAAAAABAAAQD//wEEAMACAAAAAwAAAAAAAAAAAAAAAQAAANAL0QsBAAAAAAAAAAAAAAAAAAAAAAAAAAAAQAACAP//)
[root@rocky94 tests]#

Issue Links

• is related to: LU-19308 interop 2.15 vs master crash in migration mdd_xattrs_migrate_prep()) ASSERTION( tobj != ((void *)0) ) failed