On Fri, 2010-10-22 at 12:36 +0200, Guillem Jover wrote: > Hi! > > On Sat, 2010-10-09 at 00:50:11 +0200, Giovanni Mascellani wrote: > > Il 08/10/2010 03:19, Doug Brewer ha scritto: > > > I fired up two consoles, running atftp --option multicast --get -r > > > ex.bin 10.0.0.1 > > > > I think I were wrong, the problem seems to be quite different: the > > failing sendto() sets errno to 22, which means "Invalid argument". I > > don't know where is the problem, but have not enough time to investigate > > right now. > > W/o having checked the actual code I'd say this is due to BSD kernels > requiring the socklen_t argument of some of the socket functions to be > exactly the size of the sockaddr matching the address family, and not > just bigger as in the size of sockaddr_storage. The correct size should > always be known from addrinfo for example. > > And I'd just assume this is a regression introduced by the patch for > <http://bugs.debian.org/580473>, Ben Hutchings CCed. Which just seems > to use sizeof() on struct sockaddr_storage. > > Also skimming over that patch it seems some of the new functions like > sockaddr_print_addr could be replaced by address neutral counterparts > like getnameinfo. And the sockaddr_set_port and sockaddr_get_port could > be avoided by using getaddrinfo and getnameinfo instead, but I've not > checked deeply.
Please read the code properly instead of idly speculating. I know all
about getaddrinfo() and it simply doesn't cover everything atftpd needs
to do.
I think the real problem is, as Doug Brewer said, the use of
connect(..., &{AF_UNSPEC}, ...) to disconnect a datagram socket. I
believe that is a Linux-specific feature and that there is no equivalent
under FreeBSD.
Ben.
--
Ben Hutchings
Once a job is fouled up, anything done to improve it makes it worse.
signature.asc
Description: This is a digitally signed message part
