#18 Kernel crash with freshly loaded 2.5.66 kernel

[Axelle Apvrille]

Unfortunately, the problem still occurs... but, by
chance, only in very special cases:

* you must reboot the machine
* load DSI
* straight after, do a PrintPolicy
==> kernel crash....

Kernel crash does NOT occur
* if something else than PrintPolicy has been called
before in DSM
* if you call something else (SetDebugLevel, or
UpdatePolicy), then unload DSM, then reload DSM.

[David Gordon]

Logged In: YES
user_id=644239

Kernel crashes much more ofter than mentioned in bug.

Changed internal_sk_buff_alloc_security(struct sk_buff *skb,
int gfp_mask):

-sksec = (sk_buff_security_t *) get_sk_buff_memory(skb);
+sksec = kmalloc(sizeof (sk_buff_security_t), gfp_mask);

However, variable gfp_mask is only available to
sk_buff_alloc_security hook, not to other sk_buff_* hooks.

Also, kernel no longer crashes consistently, but it still
crashes after a certain time.

I suggest to postpone this bug in favor of developing
digital signatures. Kernel 2.5.66 DSM module seems somewhat
more stable now.

[David Gordon]

Logged In: YES
user_id=644239

Changes have not proved very useful. Crashes seem to occur
either randomly or under stress (such as benchmarks).
Problem is related to sk_buffers.

[Axelle Apvrille]

Logged In: YES
user_id=204539

Kernel crash STILL occurs with the current CVS for 2.5.66
To reproduce the error, you can just try to launch LMBench
on it...

-- Axelle.