hesela.devfast_io_fail_tmo (FC path failure)
For Linux SCSI over Fibre Channel, fast_io_fail_tmo is the interval in seconds after a remote-port problem before requesting failure of affected I/O.
The Linux v6.12 FC handler requests failure without promising that every request has finished on return. This is separate from dev_loss_tmo target-removal processing and from a DM-Multipath map's all-paths-down queue policy. [1]
The parameter name is not FC-exclusive: RHEL 9 maps DM Multipath's value to iSCSI recovery_tmo, while native NVMe-oF exposes a separate fast_io_fail_tmo alongside controller-reconnection options. Keep the transport and implementation with the setting. [2] [3]
These FC path-loss timers do not trigger if the remote port is never reported lost, even when SCSI commands hang. The manual discusses eh_deadline for that different condition and warns that its HBA reset affects all target paths on the adapter. [5]
Relations
- related
- dev_loss_tmo (transport device loss), no_path_retry, Multipath I/O, I/O timeout, Fibre Channel
Citations
- Linux v6.12: FC transport loss, fail-fast and target-removal work.
- RHEL 9: DM Multipath device-timeout overrides, section 5.7.
- libnvme v1.11: fabrics connection options, including fast_io_fail_tmo, ctrl_loss_tmo and reconnect_delay.
- multipath-tools: multipath.conf(5), eh_deadline and FC timeout trigger; pinned commit d53932bdbee02f440518ea384e63d7c4d3f66685.
Further reading
Sources & caveats
- FC semantics are scoped to Linux v6.12; iSCSI mapping is documented for RHEL 9, and the separate native NVMe option is listed in libnvme v1.11. No path-loss experiment or application-completion bound is claimed.