Re: [libvirt] [PATCH 2/3] qemu: Process the hostdev "rawio" setting
On 09/18/2014 12:50 PM, Michal Privoznik wrote:
On 10.09.2014 01:40, John Ferlan wrote:
> Mimic the "Disk" processing for 'rawio', but for a scsi_host
hostdev
> lun device.
<...snip...>
> diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c
> index b1d8a32..3544716 100644
> --- a/src/qemu/qemu_process.c
> +++ b/src/qemu/qemu_process.c
> @@ -3756,6 +3756,7 @@ int qemuProcessStart(virConnectPtr conn,
> struct qemuProcessHookData hookData;
> unsigned long cur_balloon;
> size_t i;
> + bool rawio_set = false;
> char *nodeset = NULL;
> virBitmapPtr nodemask = NULL;
> unsigned int stop_flags;
> @@ -4122,13 +4123,15 @@ int qemuProcessStart(virConnectPtr conn,
> virDomainDeviceDef dev;
> virDomainDiskDefPtr disk = vm->def->disks[i];
>
> - if (vm->def->disks[i]->rawio == 1)
> + if (vm->def->disks[i]->rawio == 1) {
> #ifdef CAP_SYS_RAWIO
> virCommandAllowCap(cmd, CAP_SYS_RAWIO);
> #else
> virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
> _("Raw I/O is not supported on this
platform"));
> #endif
> + rawio_set = true;
> + }
Interesting. So if rawio is requested we shout an error but don't fail
actually. I think we are missing 'goto cleanup' here.
See 9a2f36ec04e0436b1ba9f0c21f9be234b25ac579
I can add a goto if desired or perhaps change it to a VIR_WARN() or
something else well.
I've copied the author of that commit to get an opinion...
>
> dev.type = VIR_DOMAIN_DEVICE_DISK;
> dev.data.disk = disk;
> @@ -4139,6 +4142,21 @@ int qemuProcessStart(virConnectPtr conn,
> goto cleanup;
> }
>
> + /* If rawio not already set, check hostdevs as well */
> + if (!rawio_set) {
> + for (i = 0; i < vm->def->nhostdevs; i++) {
> + virDomainHostdevSubsysSCSIPtr scsisrc =
> + &vm->def->hostdevs[i]->source.subsys.u.scsi;
> + if (scsisrc->rawio == 1)
> +#ifdef CAP_SYS_RAWIO
> + virCommandAllowCap(cmd, CAP_SYS_RAWIO);
> +#else
> + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
> + _("Raw I/O is not supported on this
platform"));
> +#endif
And here too then.
Monkey see, monkey do[o] ;-)
John
> + }
> + }
> +