When a host is rebooted, mediated devices disappear from sysfs. mdevctl (https://github.com/mdevctl/mdevctl) is a utility for managing and persisting these devices. It maintains a registry of mediated devices and can start and stop them by UUID. when libvirt attempts to create a new mediated device object, we currently fail if the path does not exist in sysfs. This patch tries a little bit harder by using mdevctl to attempt to activate the device. The approach is fairly naive -- it just calls 'mdevctl start -u $UUID' without checking whether this UUID is registered with mdevctl or not. See https://bugzilla.redhat.com/show_bug.cgi?id=1699274 Signed-off-by: Jonathon Jongsma <jjongsma(a)redhat.com&gt; --- NOTES: - an argument could be made that we should simply do nothing here. mdevctl does have support for automatically activating the mediated device when the parent device becomes available (via udev). So if the administrator set up the mdev to start automatically, this patch should not even be necessary. That said, I think this patch could still be useful.

- As I said above, the approach is pretty naive. I could have checked whether the device was defined in mdevctl's registry and given a different error message ("try registering this device with mdevctl") in that scenario. But I chose to keep it simple. src/util/virmdev.c | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-) diff --git a/src/util/virmdev.c b/src/util/virmdev.c index 3d5488cdae..7a2f0adedc 100644 --- a/src/util/virmdev.c +++ b/src/util/virmdev.c @@ -25,6 +25,7 @@ #include "virfile.h" #include "virstring.h" #include "viralloc.h" +#include "vircommand.h" #define VIR_FROM_THIS VIR_FROM_NONE @@ -148,9 +149,24 @@ virMediatedDeviceNew(const char *uuidstr, virMediatedDeviceModelType model) return NULL; if (!virFileExists(sysfspath)) { - virReportError(VIR_ERR_DEVICE_MISSING, - _("mediated device '%s' not found"), uuidstr); - return NULL; + bool activated = false; + /* if the mdev device path doesn't exist, we may still be able to start + * the device using mdevctl + */ + char *mdevctl = virFindFileInPath("mdevctl"); + if (mdevctl) { + const char *runargs[] = { mdevctl, "start", "-u", uuidstr, NULL }; + if (virRun(runargs, NULL) == 0) { + /* check to see if it the device exists now */ + activated = virFileExists(sysfspath); + } + } + + if (!activated) { + virReportError(VIR_ERR_DEVICE_MISSING, + _("mediated device '%s' not found"), uuidstr); + return NULL; + } } if (VIR_ALLOC(dev) < 0)

On Wed, 2019-08-14 at 16:14 +0200, Boris Fiuczynski wrote: > On 8/14/19 12:02 AM, Jonathon Jongsma wrote: >> When a host is rebooted, mediated devices disappear from >> sysfs. mdevctl >> (https://github.com/mdevctl/mdevctl) is a utility for managing and >> persisting these devices. It maintains a registry of mediated >> devices >> and can start and stop them by UUID. >> >> when libvirt attempts to create a new mediated device object, we >> currently >> fail if the path does not exist in sysfs. This patch tries a little >> bit >> harder by using mdevctl to attempt to activate the device. The >> approach >> is fairly naive -- it just calls 'mdevctl start -u $UUID' without >> checking whether this UUID is registered with mdevctl or not. >> >> See https://bugzilla.redhat.com/show_bug.cgi?id=1699274 >> >> Signed-off-by: Jonathon Jongsma <jjongsma(a)redhat.com&gt; >> --- >> NOTES: >> - an argument could be made that we should simply do nothing here. >> mdevctl does >> have support for automatically activating the mediated device >> when the parent >> device becomes available (via udev). So if the administrator set >> up the mdev >> to start automatically, this patch should not even be necessary. >> That said, I >> think this patch could still be useful. > > I would actually like to use this argument since this patch > unconditionally starts/creates a persistently defined mdev without > ever > stopping/destroying it and also not looking if it is defined as to > be > automatically started or manually started. If the mdev is specified > in > mdevctl to be started automatically I would rate this as mdevctl is > in > control of this mdevs lifecycle and libvirt should not interfere > with > it. It might be that I am over-interpreting auto and manual as start > options in mdevctl but it feels to me that libvirt and mdevctl > should > not run into a management clash of host resources. > > In addition what about a user specifiable tag in the domain xmls > mdev > definition which controls the management behavior of an mdev with > mdevctl or another tool? Thanks for the feedback. I welcome additional opinions on this. If there's a concensus that the right approach is to do nothing, I can drop this patch. Or alternately, we could simply point users toward mdevctl in the error message. For example, something like: diff --git a/src/util/virmdev.c b/src/util/virmdev.c index 3d5488cdae..70d990eace 100644 --- a/src/util/virmdev.c +++ b/src/util/virmdev.c @@ -149,7 +149,7 @@ virMediatedDeviceNew(const char *uuidstr, virMediatedDeviceModelType model) if (!virFileExists(sysfspath)) { virReportError(VIR_ERR_DEVICE_MISSING, - _("mediated device '%s' not found"), uuidstr); + _("mediated device '%s' not found. Persistent devices can be managed with 'mdevctl'."), uuidstr); return NULL; }