[PATCH] udev: add tags also on bind action
aleksander at aleksander.es
Sun May 27 19:43:06 UTC 2018
> On Sun, May 27, 2018 at 6:16 PM, Bjørn Mork <bjorn at mork.no> wrote:
>> Never heard of "bind" events before. Googling them led me straight to
>> which tells me that that I haven't been paying much attention for the
>> last 5 kernel releases...
>> Except for that, this looks like a systemd bug. The Debian bug is
>> reported agains v237, so your v236 will be affected.
> Ohh, this explains everything I'm seeing indeed.
>> I will test your patch, but adding "bind" to every existing rule for
>> every device and every application cannot be right? This should be
>> handled by systemd-udev.
> Yes, should be handled by systemd-udev, definitely. This is extremely
> But in the meantime, all the logic we have in place to handle udev
> tags at device level are affected. The udev tags we set for e.g.
> tagging port types aren't affected (as there's no bind event for the
> tty device), but all the udev tags for the blacklist and greylist are
> affected currently. So I believe there's something we need to do...
> maybe adding the tag on the bind event is overkill?
> There's another possibility that I see we could use, and that would be
> not to tag the device, but tag all the ports of the device instead.
> This is already supported in git master because the "udev-less" logic
> used in OpenWRT applies device-level tags in the different ports, so
> we could do the same to avoid this... The problem is that this would
> force us to keep this in place forever I guess... and that brings me
> back to thinking just adding the rules also in bind wouldn't be that
> Note anyway, that this udev bug affects MM running in DEFAULT filter
> mode. If using the STRICT filter, as we're suggesting all
> distributions to use now, they would not be affected (as all
> port-level udev tags wouldn't be affected, and there's no
> blacklist/greylist filtering in STRICT mode).
> Well, a third option would be to hack our own port type rules,
> blacklist, greylist... without using udev at all. But I really
> wouldn't want to have to do that right now...
> Whenever we find a solution for this issue in our side, even if
> temporary, I'll suggest we release 1.8.
Others have already started to duplicate the actions in "bind" events
as well, e.g.:
Yes, ideally fixed in systemd-udev, but I still suggest to apply a
workaround ourselves for now... processing bind events as well seems
More information about the ModemManager-devel