Sorry, for the late reply.
On Fri, Mar 24, 2017 at 3:40 PM, Michael Kerrisk (man-pages)
<mtk.manpages@xxxxxxxxx> wrote:
> Hi Dmitry,
>
> On 03/20/2017 10:58 PM, Dmitry V. Levin wrote:
>> Hi,
>>
>> On Mon, Mar 20, 2017 at 10:02:15PM +0100, Christian Brauner wrote:
>>> Hi,
>>>
>>> Thanks Dmitry. One comment
>>>
>>> On 20 Mar 2017 9:51 p.m., "Dmitry V. Levin" <ldv@xxxxxxxxxxxx> wrote:
>>>
>>> ---
>>> man3/ttyname.3 | 8 +++++++-
>>> 1 file changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/man3/ttyname.3 b/man3/ttyname.3
>>> index 14c24e7..0be50c6 100644
>>> --- a/man3/ttyname.3
>>> +++ b/man3/ttyname.3
>>> @@ -71,6 +71,11 @@ File descriptor does not refer to a terminal device.
>>> .RB ( ttyname_r ())
>>> .I buflen
>>> was too small to allow storing the pathname.
>>> +.TP
>>> +.\" glibc commit 15e9a4f378c8607c2ae1aa465436af4321db0e23
>>> +.B ENODEV
>>> +File descriptor refers to a slave pseudoterminal device
>>> +but the corresponding pathname could not be found.
>>>
>>> I think it would be good to explicitly mention that ENODEV is set in case
>>> the fd does not refer to a pts device in its namespace. Otherwise users
>>> might take this as an indication that this is a more generic error which is
>>> not the case.
>>
>> In fact, this is a more generic error than a namespace mismatch, although
>> the latter is the most likely reason.
>>
>> Imagine that the corresponding file inside /dev/pts/ is not available for
>> some reason and the stat call has failed. This situation could be
>> reproduced e.g. by bind-mounting an empty directory to /dev/pts.
>> A subsequent ttyname invocation would end up with ENODEV because the device
>> is literally not available although it's withing the same namespace.
>>
>> I don't mind if you change the description to mention the namespace case
>> as the most likely, but please do not make it the only case when ENODEV
>> can happen.
>
> I'm open on this point. If someone wants to write a suitable patch, I'll
> probably take it.
This point is in my opinion crucial. The over-mounting scenario is
still a valid case for most programs that do not actually care about
what /dev/pts/<n> is actually used to go on which is a nice side-effect
of the patch. The namespace part should be mentioned since ttyname{_r}()
explicitly does not return anything when it detects that
/proc/self/fd/<n> points to /dev/pts/<n> but /dev/pts/<n> does not exist
in the same namespace. This was the original motivation when we
wrote the patch. So users that get ENODEV from ttyname{_r}() should
be aware that resolving the symlink manually afterwards doesn't give them
a /dev/pts/<n> path valid in the current namespace.
I'll put this on my TODO list but if someone is willing to send a
patch right away
please do so. :)
Christian
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
