Some random comments: 1) What will be the default value for the new timeout?
2) I think this should be at least KERN_WARN(ing): + iscsi_cls_session_printk(KERN_INFO, session, + "session dev loss timed out after %d secs\n", + session->dev_loss_tmo); 3) As FC hardly ever looses packets or suffers from variable delays, it's much easier to timeout an FC device than a TCP device (IMHO) 4) With Linux an non-persistent device names, frequent remove and re-add of devices should be avoided if possible (In other operating systems the same device file may re-appear if the device comes back online) 5) I hope (i.e. I have no idea about it) that outstanding requests for a device that is going to be removed are removed cleanly before as well. Regards, Ulrich >>> Mike Christie <micha...@cs.wisc.edu> schrieb am 06.10.2010 um 10:46 in Nachricht <4cac376c.3040...@cs.wisc.edu>: > Hey, > > At storage summit this year, it was decided iscsi needs to catch up with > the other iscsi drivers, and add something like FC's dev_loss_tmo. When > this timeout expires, it will cause the iscsi layer to remove the scsi > devices and fail any IO that was queued. Upper layers like dm-multipath > (actually multipathd and dm-multipath work together) then handle hotplug > removal events from the device deletion event, by removing paths from > the dm-multipath device. > > When the session gets logged back in, we kick off a scan and re-find > devices. multipathd and dm-multipath then handle this by adding the > paths back to the multipath device. > > If you are not using dm-multipath and are mounting the FS on the iscsi > device then you basically see the same thing as before. IO errors, > followed by the FS getting mounted read only. > > This patch will also fix a problem where if a device is offlined due to > the transport going kaput scsi_internal_device_unblock() would not set > the device back to running. With this patch the device gets destroyed > and a new one gets added, so we completely bypass that problem like how > FC and SAS does (note: it does not address the problem with > iscsi_block_eh races - waiting on the fc discussion). > > This patch was made over Linus's tree but can be applied to scsi-misc or > scsi-rc-fixes. I have only done some light testing. I wanted to post > this though, so other driver developers could test it out. -- You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
