Comment # 14
on bug 9920
from Hadriel Kaplan
(In reply to comment #13)
> I don't know enough about SIP/SDP/RDP and the relevant state that needs
> saving
It's not just a SIP/SDP problem: H.245, ANSI A Interface, and Apple MIDI also
create the hashtable but never free it. And SDP creates it not just for SIP,
but also for Megaco, MGCP, and Cisco Session Management Protocol.
> but it sounds to me like even a wmem hash table would only sort of
> fix the leak (in that all the memory would eventually be freed, but much of
> it would be "dead" long before that). This isn't a problem with
> packet-scoped memory since it's lifetime is so short, but its something I've
> been trying to avoid with file-scoped memory.
Right now, RTP's conversation state is already kept for the duration of a
capture file. (I don't mean the stuff that isn't free'd, I mean the
wmem-managed stuff is scoped to wmem_file_scope)
Currently RTP has no concept of the lifetime of an RTP session - the
applications which call it do, but don't tell RTP about the end of a session.
I.e., SIP, H.245, MGCP, etc., know when the RTP stream is over, so they could
tell it to free up the conversation resources, including the hashtables and
all.
You are receiving this mail because:
- You are watching all bug changes.
