---Problem Description---
IPR driver causes multipath to fail paths/stuck IO on Medium Errors
This problem is resolved with this upstream accepted patch, scheduled for 4.11.
The detailed problem description and resolution are described in the commit message.
Clients with a dual-controller multipathed IPR configuration that eventually runs into failing disk/sectors, will experience an I/O hang once the drive reports a Medium Error, which can hang an application or even the root filesystem (whatever is doing I/O to the failing drive), potentially hanging the system.
---Steps to Reproduce---
1) Use a disk with bad sectors (or force such condition, via internal/special tools)
2) Multipath that disk
3) Run IO to the multipath device on the bad sectors
4) Both paths will be failed, and IO is stuck due to queue_if_no_path (enabled by default for IPR)
The detailed problem description and resolution are described in the commit message.
---Problem Description---
IPR driver causes multipath to fail paths/stuck IO on Medium Errors
This problem is resolved with this upstream accepted patch, scheduled for 4.11.
The detailed problem description and resolution are described in the commit message.
> scsi: ipr: do not set DID_PASSTHROUGH on CHECK CONDITION git.kernel. org/?p= linux/kernel/ git/jejb/ scsi.git; a=commit; h=785a470496d8e 0a32e3d39f37698 4eb2c98ca5b3
> http://
Please apply to 17.04 and 16.04.
The business justification for the SRU is:
Clients with a dual-controller multipathed IPR configuration that eventually runs into failing disk/sectors, will experience an I/O hang once the drive reports a Medium Error, which can hang an application or even the root filesystem (whatever is doing I/O to the failing drive), potentially hanging the system.
Thanks.
---Additional Hardware Info---
Dual (IPR) controller setup, multipath enabled
---Steps to Reproduce---
1) Use a disk with bad sectors (or force such condition, via internal/special tools)
2) Multipath that disk
3) Run IO to the multipath device on the bad sectors
4) Both paths will be failed, and IO is stuck due to queue_if_no_path (enabled by default for IPR)
The detailed problem description and resolution are described in the commit message.