[LU-4389] If OST_SYNC causes inode update, client does not reflect change

• Type: Bug
• Resolution: Fixed
• Priority: Major
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.5.0, Lustre 2.6.0
• Labels: zfs
• Severity: 3
• Assignee: Zhenyu Xu
• Reporter: Nathaniel Clark

Description

This really only affects ZFS, because zfs does not update # of blocks until the blocks are flushed to disk, so after an OST_WRITE, the number of blocks only reflect the number on disk, but that number is updated after an OST_SYNC RPC (which returns the correct number of blocks, but is ignored by the client code).

Test: (on a ZFS backed setup)
1. client1$ dd if=/dev/zero of=/mnt/lustre/test oflag=sync bs=1M count=2
2. client1$ stat /mnt/lustre/test
3. client2$ stat /mnt/lustre/test OR clear locks on client1 and redo stat
4. See that the number of blocks is incorrect on client1 (client2 has correct value)