RE: [lwip-users] udp checksum error

[Goldschmidt Simon]

> Oh, I assumed that the way TCP and UDP handles pbufs are the same. And
since DHCP works fine and it didnt 
> work with the socket API, your explanation fits in just nicely. But
where is this copying supposed to be done ?
> Is it supposed to be done in the socket layer, or netconn API layer?
Thanks in advance again...
 
TCP sockets copy the data to be sent into an internal buffer as the data
might have to be sent again (retransmission). Since there is already a
pbuf allocated for this, it is just allocated slightly bigger and the
headers are included -> you get one continuous buffer for one packet.

UDP sockets don't copy the data: it doesn't have to be kept for
retransmission so it is faster to send them without copying. However,
since (for socket API) there is no space for the headers in the buffer
pointer you give to udp_sendto, a buffer for the headers has to be
allocated in addition to this. These two buffers then form a linked list
of pbufs.

In your netif->linkoutput() function, UDP socket pbufs have a next
pointer != NULL:

p->payload will point to the protocol headers
p->next->payload will point to the actual data

-> p->tot_len will be != p->len for such linked lists while for TCP
pbufs, p->len will be == p->tot_len.

Have a look at the explanation in pbuf.c: you have to copy all data
until you have copying as many bytes as the first p->tot_len says it has
(in other words, the last pbuf has p->len == p->tot_len).

I assume you always copy the first pbuf in such a linked list only, but
tell the driver you have copied p->tot_len bytes and the driver fills in
the missing bytes with 0x00?

Simon