Re: [lwip-users] PBUF_RAM etc

[address@hidden]

Mason wrote:
> Can you explain why the pbuf struct and the payload buffer
> need to be contiguous? What problems are solved by having
> the struct and the buffer stored contiguously?

When passing the packet through the stack, the payload pointer in the 
pbuf is moved back and forth (e.g. to point at the current protocol 
header or at the application payload). This is what pbuf_header() does.

Hiding a header (moving payload 'higher' in memory address) always works 
because you can do that as long as p->len > 0. However, restoring a 
header or making room for one (which is mainly needed when sending a 
packet out) does not work with the current members of struct pbuf, as it 
doesn't have a 'max_len' field.

To understand this you have to know that when allocating a pbuf for TCP 
or UDP data, pbuf_alloc() always leaves room for all the headers (which 
are filled in later) so that all the data is in one piece (with some 
exceptions for TCP), which makes it easier and faster for DMA engines to 
send packets.

So for non-contiguous pbuf/payload combinations, we always eitehr need a 
separate pbuf for headers or we'd need a 'max_len' pointer to know how 
big the externally allocated memory is.

> There is a pointer to a buffer in the struct, it is redundant
> if the payload is at a known offset, right?
No, since the payload is not a 'total_payload' but a 
'current_payload_pointer', like I described above.

Simon