net: do wii network transfers in blocks (#178)

[dborth]

a PR was recently submitted to me - dborth/libsmb2#2

the problem - net_recvfrom and net_sendto are neither bound nor gracefully handle large sizes.

looks like others have hit this before so I pulled the working fix from dacotaco (devkitPro/libogc@f4dbfca)

[Extrems]

I don't believe this memory copy is required to begin with, certainly not from/to MEM2. I'll have to check for MEM1.

[TeleTubby666]

Thanks for picking this up. I reported the original issue in
dborth/libsmb2#2 (real Wii, Samba 4, SMB2/3). I read through the diff and
have two observations:

1. Splitting into blocks fixes the IPC_ENOMEM (-22) on large requests, which
   was one of the two problems I saw. The other was a hang right after the
   TCP connect: libsmb2's read loop expects EAGAIN when no data is pending
   on a non-blocking socket, but net_read() -> net_recvfrom() with flags 0
   did not give me that. I worked around it by calling net_select() with a
   zero timeout first and returning EAGAIN if nothing is readable. This
   patch doesn't change that behaviour, so I'd expect the hang to remain.

2. In the new loop, if len > maxblocksize and the first block comes back
   full, the loop calls recvfrom again immediately. If no more data is
   pending, that call either blocks or returns -EAGAIN, and since
   ret = received overwrites the count, the data already read is lost.
   Returning the bytes read so far (when ret > 0) would avoid that. The same
   applies to net_sendto if a later block fails.

I haven't tested this PR build on hardware yet. Since dborth/libsmb2#2 was
closed in favour of this, I'd suggest keeping the EAGAIN/readability check
in mind until it's confirmed that the hang is gone with this change alone.
I'll report back with test results on hardware as soon as I'm home on
Friday.