Hey Paolo, On Thu, 2014-10-09 at 12:49 +0200, Paolo Bonzini wrote: > Il 09/10/2014 10:49, Paolo Bonzini ha scritto: > > > > It does not happen if you close QEMU with SIGTERM, ctrl-c, or with the > > "quit" command, because no attempt is done to bring down the VM data > > structures (or free memory, or close file descriptors) in case of a > > fatal exit. The kernel should do that for us. > > ... and in the case of vhost-scsi, doesn't it do that when > vhost_scsi_release calls vhost_scsi_clear_endpoint? > Thanks for the extra clarifications here. The SIGTERM, ctrl-c and "quit" command cases happen as you describe, and invoke vhost_scsi_release() -> vhost_scsi_clear_endpoint() to drop the endpoint reference. AFAICT, it's the SIGKILL case that is problematic both with and without this patch. With the patch, the configfs dependency on the vhost-scsi endpoint group is left in place, thus preventing the group (and underlying target_core_mod) from being removed until a VHOST_SCSI_CLEAR_ENDPOINT with the same wwpn is called to drop the original reference. Without the patch, the group can still be removed at any time, but any subsequent VHOST_SCSI_SET_ENDPOINT attempts with the original wwpn will fail after SIGKILL, because the original reference is still in place. So I held off on pushing this patch to -rc1 for the moment, but even with the above limitation preventing group shutdown after SIGKILL, I think it's still better to obtain the configfs dependency to prevent the removal of endpoints while there are active references. That said, I'm still unsure how to address the SIGKILL case, and what's the most sane way to drop dead references after it happens, and how vhost-scsi should be differentiating between dead and active references. Any ideas..? --nab -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
