[libvirt] [PATCH v3 6/6] nodedev: Work around the uevent race by hooking up virFileWaitForAccess

Erik Skultety eskultet at redhat.com
Tue Aug 29 13:01:29 UTC 2017

On Mon, Aug 28, 2017 at 12:40:44PM -0400, John Ferlan wrote:
> On 08/24/2017 07:23 AM, Erik Skultety wrote:
> > If we find ourselves in the situation that the 'add' uevent has been
> > fired earlier than the sysfs tree for a device was created, we should
> > use the best-effort approach and give kernel some predetermined amount
> > of time, thus waiting for the attributes to be ready rather than
> > discarding the device from our device list forever. If those don't appear
> > in the given time frame, we need to move on, since libvirt can't wait
> > indefinitely.
> >
> > Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1463285
> >
> > Signed-off-by: Erik Skultety <eskultet at redhat.com>
> > ---
> >  src/node_device/node_device_udev.c | 16 +++++++++++++++-
> >  1 file changed, 15 insertions(+), 1 deletion(-)
> >
> IIRC - I pointed out to you that this is eerily familiar to something
> that happens in the vHBA code w/r/t to wwnn/wwpn files. Except that the
> files exist, but have a -1 in them which is totally bogus. Then some
> magic thing happens and the real wwnn/wwpn is placed into the file, but
> libvirt already looked and failed.  When I tried to work around this the
> decision was to let it be and call it a kernel / udev bug.
> https://www.redhat.com/archives/libvir-list/2016-June/msg02213.html
> and Daniel's answer
> https://www.redhat.com/archives/libvir-list/2016-July/msg00912.html
> Although yes, with the other changes in place one would think having a
> wait is no big deal.
> Still are you guaranteed that once the file exists that the data within
> the file is valid?  In the vHBA case it wasn't and that led to issues.

Yes, I recall you pointing me to this issue before and you're right that if the
data is bogus, we can't do much about that, except that in this case, I'm only
relying on the existence of the file/dir, because I need its name to determine
the mediated device's type, not its content, which arguably makes it a different


> I'd "use this" processing instead of the hack I proposed as well seeing
> as it doesn't seem kernel/udev fixing issues such as these is on any
> priority list /-{

Exactly ^this :(

More information about the libvir-list mailing list