On Thu, Jul 17, 2014 at 07:33:03PM +0000, Alin Serdean wrote: > Add defines needed to compile netlink-socket.c and netlink.c. > > Add wrappers for the functions sendmsg and recvmsg. They can also be added to > another file instead of defining them inlined. > > Add a wrapper and the functionality behind it for syconf. > > Add the newly created files to the noinst_HEADERS in windows/automake.mk > > Signed-off-by: Alin Gabriel Serdean <aserd...@cloudbasesolutions.com>
include/windows/linux/rtnetlink.h in this file is an edited version of include/linux/rtnetlink.h from the Linux source tree, so it should include a GPLv2 license notice. As is, it only has a Cloudbase copyright and Apache license, which makes it look as if Cloudbase were the original author. It's fine to apply a Cloudbase copyright to it, but it should be in addition to the existing copyright, not (at first glance) instead of it. The recvmsg() emulation is not complete because it only uses the first iovec in the buffer. I don't see a good way to use multiple iovecs on Windows--maybe ReadFileGather but that is always asynchronous so it might be expensive--so we might have to read into a large on-stack buffer and then copy. The recvmsg() emulation has really weird error behavior. It always sets errno to EAGAIN, whether the operation succeeds or fails. It is odd to use a function _set_errno() to set errno, because the ANSI C standard (even C89) says that errno is modifiable through a normal assignment, e.g. "errno = EAGAIN". Doesn't that work on Windows? sendmsg() is supposed to return the number of bytes sent. recvmsg() and sendmsg() should return ssize_t (a signed type), not unsigned int. I am not sure that it is adequate in general for sendmsg() to send in several segments, because that could split a single datagram into two datagrams. However, I see that the current use of sendmsg() in the tree uses it in a way such that that doesn't cause problems. I wonder whether we should avoid sendmsg() on windows entirely, even in emulation. I think that we should avoid using AF_NETLINK, MSG_DONTWAIT, and SO_RCVBUFFORCE on Windows, since it doesn't have any of them, rather than defining them. In fact it looks like the code does avoid using them already. Thanks, Ben. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
