udevadm triggers wrong event type for failed events

udevadm triggers wrong event type for failed events

[Andrey Borzenkov]

[-- Attachment #1: Type: text/plain, Size: 1164 bytes --]

I was puzzled why my bluetoothd was not started on boot. It turned out 
the following:

- bluetoothd requires D-Bus so attempt to start it early (may be, as 
early as initrd) fails

- event is saved as failed event; but only device path is preserved, no 
information about which event type failed

- later udevadm trigger --type=failed is run; but by default this 
triggers "change" event instead of "add". There is no "change" action 
for bluetooth subsystem (nor do I think it normally emits such event) so 
nothing happens.

This change was introduced relatively recently in

commit 236fae6cf1a619a92174efdf84cd7d91e7d4348d
Author: Kay Sievers <kay.sievers@vrfy.org>
Date:   Mon Apr 12 17:56:32 2010 +0200

    udevadm: trigger - switch default action from "add" to "change"


Now, I could of course just go and add --action=add to udev-post 
initscript. The question is - is it correct to assume that every failed 
event is "add"? Should not event type be preserved for later correct re-
trigger?

BTW both init/udev-retry.service from udev GIT and udev-post from Fedora 
have the same problem (Mandriva is using Fedora version).

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: udevadm triggers wrong event type for failed events

[Kay Sievers]

On Sat, Aug 14, 2010 at 18:16, Andrey Borzenkov <arvidjaar@mail.ru> wrote:
> I was puzzled why my bluetoothd was not started on boot. It turned out
> the following:
>
> - bluetoothd requires D-Bus so attempt to start it early (may be, as
> early as initrd) fails
>
> - event is saved as failed event; but only device path is preserved, no
> information about which event type failed
>
> - later udevadm trigger --typeúiled is run; but by default this
> triggers "change" event instead of "add". There is no "change" action
> for bluetooth subsystem (nor do I think it normally emits such event) so
> nothing happens.
>
> This change was introduced relatively recently in
>
> commit 236fae6cf1a619a92174efdf84cd7d91e7d4348d
> Author: Kay Sievers <kay.sievers@vrfy.org>
> Date:   Mon Apr 12 17:56:32 2010 +0200
>
>    udevadm: trigger - switch default action from "add" to "change"
>
>
> Now, I could of course just go and add --action­d to udev-post
> initscript. The question is - is it correct to assume that every failed
> event is "add"? Should not event type be preserved for later correct re-
> trigger?

It should match what we do in the initial init scripts, so adding
--action­d seems right. The defaults seems fine, which are always
'change'.

Kay