Re: [Jack-Devel] NetOne driver in JACK2

On Fri, Apr 01, 2011 at 02:11:31PM +0200, Stéphane Letz wrote:
> Hi Torben and jack-dev community.
> 
> I received some bugs report (crash in server with "jackd -d netone"), like:
> 
> 	• NetOne driver started
> 	• AutoConfig Override !!!
> 	• MTU is set to 1400 bytes
> 	• Cannot read socket fd = 10 err = Interrupted system call
> 	• [Thread 0xb0862b70 (LWP 6381) exited]
> 	• NotifyClient fails name = exaile.py event = 18 val1 = 1 val2 = 0
> 	• netxruns... duration: 11ms
> 	• jackd: ../common/JackEngine.cpp:932: int Jack::JackEngine::PortConnect(int, jack_port_id_t, jack_port_id_t): Assertion `ref >= 0' failed.
> 
> 
> 	• jof@studio:~$ jack_netsource -c 120 -n 3 -H 192.168.1.13
> 	• notification: uint decode error
> 	• Connected :-)
> 	• netjack: at frame 000085 -> total netxruns 1  (1%) queue time= 23221
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• notification: uint decode error
> 	• ^C*** glibc detected *** jack_netsource: double free or corruption (!prev): 0x09bd4830 ***
> 	• ======= Backtrace: =========
> 	•

the uint decode error is a celt error.
That generally happens, when celt versions dont match.
(This is basically the case why i didnt bother to support later celt
 versions, adi and me came to the conclusion, that distros should ship 
 celt-0.7 until celts bitstream was frozen)

celt-0.11 is tagged with a "soft freeze" of the bitstream.
So maybe celt-0.12 is going to have a real freeze.


> 
> The last thing I did (before jack 1.9.7 release) was to synchronize NetJack1 code from JACK1 to JACK2, hoping I did that correctly (basically updating netjack.h,c  and netjack_packet,h,c)
> 
> Reviewing the code I see:
> 
>  - using  exit(1); in netjack.c instead of properly returning a proper error.

gonna try to fix that. But it only happens when malicious Packets are
sent. (its basically just some sanity checks on the autoconfig values)

In jack1 exiting at that point was sensible.
But with backend switching in place its probably bad.

> 
> - CELT : celt_mode_create should be matched with celt_mode_destroy. My feeling is that this is not done everywhere it should (especially in JackNetOneDriver.cpp)
> 
> This is certainly not the cause of the bugs, but I think this should be fixed (along the "jack_netsource: double free or corruption")n before going anywhere else.

without a valgrind dump, i dont know, where the corruption happens.
and since celt is seeing a broken bitstream, i suspect its a celt bug.
(that needs to be fixed too, but maybe it already is in later versions)

i need infos which celt versions are used, or a proper valgrind dump.

> 
> Could you please have a look? 
> 
> or if not Torben someone else?
> 
> Thanks.
> 
> Stéphane
> 
> 
> 
> 

-- 
torben Hohn

1301741418.8465_0.ltw:2,a <20110402104946.GB20002 at siel dot b>