| < Previous by Date | Date Index | Next by Date > |
| Thread Index | Next in Thread > |
|
The following patch address several issues. The first is a
multi-thread, multi-cpu race condition with the current callid. If you have
several separate instances of dum, the same call-id can be computed on a
multicore/processor machine. To solve this, the patch adds process and thread-id
to the salt value. The process / thread id is currently implemented on Win32,
Apple, and Linux. On linux, getpid() returns a different value for each thread.
The other platforms use the respective calls for getting the process id and
thread id. I did not implement other platforms since I don’t have access
to them. To address other concerns that the call-id does not contain
enough randomness the patch increases the number of hex digits chosen to 16.
This should massively reduce the collision space of call-ids as people have
reported on the mailing list previously. SightSpeed Inc. |
Attachment:
resiprocate_thread_random_callid.diff
Description: resiprocate_thread_random_callid.diff