Wireshark-dev: [Wireshark-dev] Reducing memory usage by not storing reassembled packet data
From: Guy Harris <[email protected]>
Date: Thu, 10 May 2012 20:02:31 -0700
On May 10, 2012, at 1:24 AM, Anders Broman wrote:

> - In the reassembly routines If I remember correctly, I might be wrong I think we may waste memory for TCP. Fragments and reassembled fragments?
>     - It might be possible to store file pointer and length rather than the fragment data and read in that data when needed.

Possible in some cases, although when I thought about it recently I realized that would only handle some cases.

If a reassembled packet is made up from other reassembled fragments (e.g., the unlikely-in-real-life case of reassembly of TCP segments in fragmented IP datagrams, or the more-likely case of higher-level packets reassembled from lower-level packets reassembled from TCP segments), finding the underlying chunks from the capture could be a bit of work and there wouldn't be a one-to-one correspondence between chunks referred to by the reassembly data structures and chunks from the capture file.

If a reassembled packet is made up from fragments extracted from decompressed or decrypted packets, then the chunks referred to by the reassembly data structures don't correspond *at all* to chunks from the capture file.