[nos-bbs] Analysis of recent frequent netrom related crashes

[Maiko (Personal)]

Okay, last one for now, and learning as I go ...

Perhaps I need to set the NR4TDISC a lot lower (default) ?

    jnos> netrom tdisc
    NR4 redundancy timer (sec): 120

Experiences anyone ? But still, even with a smaller timeout value,
there is a 'risk' of a crash, making me think the current way of
doing a circuit table lookup and reusing entries, seems not be
the brightest way of doing it ? thinking a 'rewrite', ugh, no.

Maiko / VE4KLM

On 2022-11-13 10:47 a.m., Maiko (Personal) wrote:
> I am guessing (hopefully this shows up in my debugs) ...
> 
> IF the local side requests a netrom layer 4 disconnect, then JNOS
> should probably free the callback there and then, instead of waiting
> for the final disconnect (which may not get to us). I figure it would
> not hurt to remove it at that point, since effectively it is done.
> 
> I could put in a timer based garbage collection, but I think it's
> best to get rid of the callback data ASAP or else it will crash.
> 
> Anyways ...
> 
> Maiko / VE4KLM
> 
> On 2022-11-13 10:37 a.m., Maiko (Personal) wrote:
>> Good morning,
>>
>> Slightly technical post ...
>>
>> This has been driving me nuts the past few months, it just seems
>> to have started, perhaps because I took on a new netrom neighbour
>> or two, I just don't know, but I think I know the reasons for all
>> the crashes. After a few days of inserting some very heavy debugs
>> into the code, this is where I am at this morning :
>>
>> JNOS keeps a table of netrom callbacks, the default is 20. When a
>> new connection happens, it gets put into the table, and when it's
>> done with, it is supposed to be removed from the table. However,
>> this removal is ONLY DONE when the state of the connection becomes
>> disconnected. What is happening, is that it appears the entry in
>> the table for a specific connection looks valid, but in fact it
>> has disappeared, but JNOS did not remove it, so crash !!!
>>
>> What this suggests to me is that I did not get the final NETROM
>> disconnected, so JNOS still thinks the callback data is valid, but
>> in fact it is not, the memory has disappeared, so what happens is
>> you get every few days a crash in the nr4subr.c functions, like :
>>
>>    Program received signal SIGSEGV, Segmentation fault.
>>    0x000000000047fdd9 in match_n4circ (index=23, id=71, user=0x2081457
>>    "\236\226d\240\212\234b\236\226d\240\212\234b", node=0x208145e
>>    "\236\226d\240\212\234b") at nr4subr.c:138
>>     138  if ((int)(cb->yournum) == index && (int)(cb->yourid) == id
>>
>> AND
>>
>>    Program received signal SIGSEGV, Segmentation fault.
>>    0x00007f96411f9780 in __memcmp_avx2_movbe () from /lib64/libc.so.6
>>    (gdb) where
>>    #0  0x00007f96411f9780 in __memcmp_avx2_movbe () from /lib64/libc.so.6
>>    #1  0x0000000000482727 in nrresetlinks (rp=0x22c5550) at nr3.c:1441
>>    #2  0x000000000047ca22 in doobsotick () at nrcmd.c:1316
>>
>> It is very consistent, so I am running into cases where I am not getting
>> the final netrom layer 4 disconnect, so the callback remains, but JNOS
>> needs to loop through the whole circuit table to find valid ones to 
>> match up with, and this invalid one just happens to still be in the
>> table and kablewee :]
>>
>> Anyways, I hope to have a fix of sorts for this 'soon', very 
>> frustrating. But again, why has this suddenly started happening
>> at the frequency it has for the past 3 months, possibly more ?
>>
>> Jack, this is probably what you are experiencing as well.
>>
>> Maiko / VE4KLM
>>