[LU-8405] MGT should check servicenode nid before accepting mount

• Status: Open
• Project: Lustre
• Affects Version/s: Lustre 2.7.0
• Type: Bug
• Priority: Major
• Reporter: Gary Hagensen (Inactive)
• Assignee: Hongchao Zhang
• Resolution: Unresolved
• Severity: 3

Description

The MGS will just accept a mount request even if the servicenode arguments are invalid nids. This results in the OST going to "partial connect: state when the MDT tried to use the bad nids and also the mount being put into the replay log, so that the partial attach never goes away. It survives a reboot, lustre_rmmod and remount of the MGT and MDT.It gets replayed.

The MGS should check all the failover nids (even a simple lctl ping) before accepting the mount, Allowing it to be accepted causes too many problems downstream.

Comments

[Hongchao Zhang]

the problem here is target will register itself without any nid in

mti->mti_nids (struct mgs_target_info) 

if the "--servicenode" param is given
the incorrect nodes, and the MGS will use the first nid in mti->mti_nids (will be zero!) to create the config log entry of "setup", which cause this problem.

the patch http://review.whamcloud.com/19933/ from LU-8091 can fix the problem.