In message <20031125184125.03D428B@coconut.itojun.org>Jun-ichiro itojun Hagino 
writes
>> Getting all the book-keeping and pcb rundowns right is more than I
>> wanted to buy into before we start the 2.0 release cycle.
>
>	we could have multiple pcbs with same (src,dst) pair, so bookkeeping
>	ip_id value on pcb does not look right to me (more work needs to be
>	done).

Exactly, that was well understood.  Why do you bring up this strawman?


I said (in the same thread you quote from) one non-strawman approach:
an entry per src/dst IPaddress pair, (on a hashtable, for non-connected
or non-pcb output needing an ip_id), with a reference-count in the
hashtable entry, a (ref-counted) pointer in a pcb, and decrement
of the refcount each time a pcb is detached from the hash entry.

That last part is the "book-keeping and pcb rundown" I referred to.