Max wrote:
In this situation it would
be perfect if se_alloc() engine could call a callback with a user
supplied argument before it starts to deallocate anything.
The init mechanism was created before the ephemeral memory stuff even existed, and has never been updated.
Some better mechanism, such as the one you suggest, would probably be a good idea.
I've developed a patch which implements such a mechanism and would
like it to be committed to the trunk.
Briefly: it costs one additional pointer per each memory chunk
allocated. Plus it costs thee additional pointers per allocation, if
allocation is made from pool in chunk mode when destructor is used,
otherwise it costs nothing. Callback is supplied as a parameter to
allocation function.
I would gladly read you comments, critics and suggestions regarding this patch.
Would it be simpler (and more generic) to have a "cleanup"/EOF routine
similar to the init mechanism but called as soon as we reach (the final)
EOF (before the se memory is freed)?
It could also be used to indicate to things like TCP desegmentation that
we've reached EOF so TCP could attempt to desegment things marked
DESEGMENT_UNTIL_FIN even though we haven't seen a FIN (see bug 3785).
Any memory held hoping for more packets (e.g., the last fragments for
reassembly) could be freed at the end of the capture instead of waiting
for the file to be closed.
