[lwip-users] Re: [lwip] Memory leak and queued packets in etharp.c

[leon . woestenberg]

Hello,

> > Another issue is the queueing of packets that
> > are created by the user code with netbuf_ref().
> > This creates a chained pbuf with at least one
> > one element having the attribute PBUF_ROM.
> > There's a pending danger that the data part
> > of such a buffer gets overwritten by the
> > application before the queued packet is sent.
> > Should these be copied?

> The application must only overwrite that area when it's
> certain that it got sent.How it is to be notified I don't know.
> But the point of PBUF_ROM is to tell the stack not to copy so
> I think it shouldn't.
>
Very good mentioning this, as this is an outstanding issue not
yet resolved.

As you mention, two different approaches exist:

1) We need to find a way to notify the application that the
pbuf is sent. If possible, by return code. Too ugly would be
another callback IMHO.

2) Copy PBUF_ROM pbufs only when they are queued in the stack.

PBUF_ROM merely then would mean: "the data is constant and
available for lwIP during the call to a lwIP function. If lwIP\
references it afterwards, it must copy the pauload itself"

I would strongly opt for solution 2, unless someone else comes
up with an (even :-) more elegant solution.

Regards, Leon.

[This message was sent through the lwip discussion list.]