While stress-testing private server, marshalling "aaah" causes an error: is there a limit on DBUS_TYPE_UNIX_FD values one can marshal in a single session?

[rony]

Simon,

thank you very much for your information and explanations!

Cheers,

---rony


On 25.08.2011 11:11, Simon McVittie wrote:
> On Wed, 24 Aug 2011 at 17:50:30 +0200, rony wrote:
>   
>> Is it possible that there is a limit for DBUS_TYPE_UNIX_FD values that
>> can be generated in a single run of a private DBus server?
>>     
> There is a maximum number of fds per process (all fds, including stdin, stdout,
> stderr, all open files and sockets, any fds passed to you via D-Bus, one
> fd per DBusConnection and one fd per DBusServer), and a maximum for the entire
> system.
>
> The limit is OS-dependent - on my Debian GNU/Linux laptop, the limits are
> currently 1024 per process (ulimit -n) and 394112 for the whole system
> (/proc/sys/fs/file-max).
>
> In addition, libdbus limits each message to 1024 fds by default. You can
> increase that limit with dbus_connection_set_max_message_unix_fds(), up to
> the maximum number allowed by the protocol, which is
> DBUS_MAXIMUM_MESSAGE_UNIX_FDS = 2**25.
>
>   
>> P.S.: Being totally unfamilar with UNIX_FD in this context, is there
>> somewhere a documentation/example of how they need to be applied in
>> dbus? (Noticing e.g. that a TCP/IP connection will not support UNIX_FDs.)
>>     
> They're file descriptors passed over a Unix socket (the SCM_RIGHTS mechanism,
> at least on Linux - the socket-level API might be different on other OSs), as
> documented in the unix(7) man page. The GUnixFDMessage and GUnixFDList objects
> in GIO are the same thing, and might be documented more clearly.
>
>     smcv
>
>