Bug#760476: gnutls28 3.3.8-3

[Nikos Mavrogiannopoulos]

Thanks, it seems I am correct. Cups closes all open descriptors. I don't know what I can do in gnutls to fix that. That looks like that should be addressed in cups.

On 22 October 2014 09:40:01 CEST, "Christophe Ségui" <christophe.segui at math.univ-toulouse.fr> wrote:
>Here is the trace of cups in start phase.
>
>
>Bests
>Christophe
>On 21/10/2014 22:50, Nikos Mavrogiannopoulos wrote:
>> On Tue, 21 Oct 2014 11:58:21 +0200
>=?UTF-8?B?Q2hyaXN0b3BoZSBTw6lndWk=?=
>>
>>> read(3, 0x7fff63334f20, 16)             = -1 EINVAL (Invalid
>argument)
>>> write(2, "gnutls[2]: Failed to read /dev/u"..., 57) = 57
>>> write(2, "gnutls[3]: ASSERT: rnd.c:142\n", 29) = 29
>>> write(2, "gnutls[3]: ASSERT: rnd.c:329\n", 29) = 29
>> Thank you. The log shows that indeed reading from /dev/urandom fails.
>> Unfortunately the strace does not start from an early stage, so I
>cannot
>> see what happens with file descriptor 3 which is supposed to be
>> /dev/urandom. What I see though is that fd 3 is also used in
>epoll_ctl(),
>> and I suspect that cups has some code that closes all open
>descriptors.
>> If that's the case it will disrupt gnutls' usage of /dev/urandom the
>> way we see here.
>>
>> regards,
>> Nikos
>>

-- 
Sent fron my mobile. Please excuse my brevity.