[libvirt] [PATCH] udev: only report a warning if udev_enumerate_scan_devices fails

[Marc Hartmayer]

On Wed, Feb 13, 2019 at 03:56 PM +0100, John Ferlan <jferlan at redhat.com> wrote:
> On 2/13/19 7:38 AM, Marc Hartmayer wrote:
>> Even if an error is reported by `udev_enumerate_scan_devices`,
>> e.g. because a driver of a device has an bug, we can still enumerate
>> all other devices. Additionally the documentation of
>> udev_enumerate_scan_devices says that on success an integer >= 0 is
>> returned (see man udev_enumerate_scan_devices(3)).
>>
>> Reviewed-by: Bjoern Walk <bwalk at linux.ibm.com>
>> Signed-off-by: Marc Hartmayer <mhartmay at linux.ibm.com>
>> ---
>>  src/node_device/node_device_udev.c | 9 ++-------
>>  1 file changed, 2 insertions(+), 7 deletions(-)
>>
>
> Interesting - looking at many examples of udev_enumerate_scan_devices
> usage shows a lack of testing the return value and as is done here just
> using @udev_enumerate to add devices after the call.
>
> Eventually found some source code for enumerator_scan_devices_tags which
> I believe is what device_enumerator_scan_devices would call due to what
> our AddMatches does. It seems that code works until it finds an error,
> but still would return a partially enumerated list.

Yep, I’ve also looked at the source code. Unfortunately the behavior is
not documented… I’ve also looked for 'udev_enumerate_get_list_entry' and
it can handle NULL pointers.

>
> Long way of saying I think this is fine... However, now @ret = -1
> doesn't ever get changed, so the caller would still fail. So it's a nice
> way to test your other patch ;-)

Yes… I’ll send a v2.

>
>> diff --git a/src/node_device/node_device_udev.c b/src/node_device/node_device_udev.c
>> index 299f55260129..90168eb8a969 100644
>> --- a/src/node_device/node_device_udev.c
>> +++ b/src/node_device/node_device_udev.c
>> @@ -1480,13 +1480,8 @@ udevEnumerateDevices(struct udev *udev)
>>      if (udevEnumerateAddMatches(udev_enumerate) < 0)
>>          goto cleanup;
>>
>> -    ret = udev_enumerate_scan_devices(udev_enumerate);
>> -    if (ret != 0) {
>> -        virReportError(VIR_ERR_INTERNAL_ERROR,
>> -                       _("udev scan devices returned %d"),
>> -                       ret);
>> -        goto cleanup;
>> -    }
>> +    if (udev_enumerate_scan_devices(udev_enumerate) < 0)
>> +        VIR_WARN("udev scan devices failed");
>
> Either before or after this, set ret = 0... or change the default from
> -1 to 0 and only change if the AddMatches fails.

I’ll set 'ret = 0;' at the end.

>
> I think the other patch would still be necessary since if
> udevEnumerateAddMatches fails, then wouldn't the issue of setting
> threadQuit still exist?
>
>>
>>      udev_list_entry_foreach(list_entry,
>>                              udev_enumerate_get_list_entry(udev_enumerate)) {
>>
>
> BTW: Using udevProcessDeviceListEntry as the 'example' of not failing if
> an element of udev_enumerate is problematic, I think logically if we
> don't get a full list we'd be OK to continue as well.

I agree.

>
> John
>
--
Kind regards / Beste Grüße
   Marc Hartmayer

IBM Deutschland Research & Development GmbH
Vorsitzende des Aufsichtsrats: Matthias Hartmann
Geschäftsführung: Dirk Wittkopp
Sitz der Gesellschaft: Böblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294