[WIP] Netlink bits

[polachok]

Makes it (almost) possible to create a netlink socket.
See #200
Not really because socket() doesn't accept protocol value, and netlink requires it.

[carllerche]

Cool! I'll keep my eye on this.

[polachok]

Any ideas on how to solve build failures? Haven't seen this kind before.

[carllerche]

Sorry for the delay! Feel free to ping me on IRC if I am slow :)

I believe that the error is due to using glob imports. By having use super::<something> in the netlinks mod and then in the super mod you have use netlinks::* there is a cycle and I don't think rustc is smart enough yet to figure it out.

I've definitely gotten bit by this in the past and the error messages are very confusing!

[polachok]

It's actually smart enough in nightly (and 1.4 I'm using).
I'll try to fix that for 1.1.

[polachok]

Ok, I finally managed to please 1.1 compiler, please take a look.
I'd also appreciate advice on how to go about the socket() issue (see top message).

[carllerche]

shoot... i let this slip!

Regarding the socket issue, the nix binding should expose the flag and we'll have to bump the version.

[kamalmarhubi]

indent to match the affected item?

[kamalmarhubi]

Hi @polachok! I'm going over outstanding PRs on nix. Since you got #211 in, I think this should be mergeable after a few little changes! Thanks!

[kamalmarhubi]

can we use this from libc?

http://rust-lang-nursery.github.io/libc/x86_64-unknown-linux-gnu/libc/struct.sockaddr_nl.html

[kamalmarhubi]

(sorry for the derive suggestion, should have realised this was present)

[kamalmarhubi]

I also just realised this would mean none of those impls would be present. unclear if they're required or not?

[fiveop]

libc already defines Clone and Copy for most or all of its structs (see s! macro: http://rust-lang-nursery.github.io/libc/x86_64-unknown-linux-gnu/src/libc/macros.rs.html#42). So unless PartialEq, Eq, Debug and Hash are necessary, it should be fine to use the libc version.

[polachok]

They're required by SockAddr: https://github.com/polachok/nix-rust/blob/netlink/src/sys/socket/addr.rs#L510

[fiveop]

If they make sense for the raw struct in our code, they would probably make sense for the corresponding raw struct in libc. Would you create a PR for libc?

[kamalmarhubi]

I just created rust-lang/libc#159 for that.

[kamalmarhubi]

PR was rejected, with rationale:

At this time we're not looking for higher level abstractions in libc (e.g. any other form of "Rust code" like Eq or Hash). That's in general left to higher level libraries.

[kamalmarhubi]

I think it still makes sense to use the struct from libc and implement the required traits as in:

diff --git a/src/sys/socket/addr.rs b/src/sys/socket/addr.rs
index eadd990..1f38c4d 100644
--- a/src/sys/socket/addr.rs
+++ b/src/sys/socket/addr.rs
@@ -17,16 +17,6 @@ use ::sys::socket::addr::netlink::NetlinkAddr;
  *
  */

-#[cfg(any(target_os = "linux", target_os = "android"))]
-#[derive(Copy, Clone, PartialEq, Eq, Debug, Hash)]
-#[repr(C)]
-pub struct sockaddr_nl {
-    pub nl_family: sa_family_t,
-    nl_pad: libc::c_ushort,
-    pub nl_pid: u32,
-    pub nl_groups: u32
-}
-
 #[repr(i32)]
 #[derive(Copy, Clone, PartialEq, Eq, Debug, Hash)]
 pub enum AddressFamily {
@@ -503,7 +493,7 @@ impl SockAddr {
             SockAddr::Inet(InetAddr::V6(ref addr)) => (mem::transmute(addr), mem::size_of::<libc::sockaddr_in6>() as libc::socklen_t),
             SockAddr::Unix(UnixAddr(ref addr, len)) => (mem::transmute(addr), (len + mem::size_of::<libc::sa_family_t>()) as libc::socklen_t),
             #[cfg(any(target_os = "linux", target_os = "android"))]
-            SockAddr::Netlink(NetlinkAddr(ref sa)) => (mem::transmute(sa), mem::size_of::<sockaddr_nl>() as libc::socklen_t),
+            SockAddr::Netlink(NetlinkAddr(ref sa)) => (mem::transmute(sa), mem::size_of::<libc::sockaddr_nl>() as libc::socklen_t),
         }
     }
 }
@@ -559,21 +549,41 @@ impl fmt::Display for SockAddr {

 #[cfg(any(target_os = "linux", target_os = "android"))]
 pub mod netlink {
-    use ::sys::socket::addr::{AddressFamily,sockaddr_nl};
-    use libc::sa_family_t;
-    use std::fmt;
+    use ::sys::socket::addr::{AddressFamily};
+    use libc::{sa_family_t, sockaddr_nl};
+    use std::{fmt, mem};
+    use std::hash::{Hash, Hasher};

-    #[derive(Copy, Clone, PartialEq, Eq, Debug, Hash)]
+    #[derive(Copy, Clone)]
     pub struct NetlinkAddr(pub sockaddr_nl);

+    // , PartialEq, Eq, Debug, Hash
+    impl PartialEq for NetlinkAddr {
+        fn eq(&self, other: &Self) -> bool {
+            let (inner, other) = (self.0, other.0);
+            (inner.nl_family, inner.nl_pid, inner.nl_groups) ==
+            (other.nl_family, other.nl_pid, other.nl_groups)
+        }
+    }
+
+    impl Eq for NetlinkAddr {}
+
+    impl Hash for NetlinkAddr {
+        fn hash<H: Hasher>(&self, s: &mut H) {
+            let inner = self.0;
+            (inner.nl_family, inner.nl_pid, inner.nl_groups).hash(s);
+        }
+    }
+
+
     impl NetlinkAddr {
         pub fn new(pid: u32, groups: u32) -> NetlinkAddr {
-            NetlinkAddr(sockaddr_nl {
-                nl_family: AddressFamily::Netlink as sa_family_t,
-                nl_pad: 0,
-                nl_pid: pid,
-                nl_groups: groups,
-            })
+            let mut addr: sockaddr_nl = unsafe { mem::zeroed() };
+            addr.nl_family = AddressFamily::Netlink as sa_family_t;
+            addr.nl_pid = pid;
+            addr.nl_groups = groups;
+
+            NetlinkAddr(addr)
         }

         pub fn pid(&self) -> u32 {

[kamalmarhubi]

(above patch should git apply cleanly on your PR)

[kamalmarhubi]

One last thing re using a struct from libc. Let me know what you think, then we'll get this in! :-)

[kamalmarhubi]

actually if I've written the patch, I might as well just merge this... rebased and merged as f167e8f...745c791
(ci run at 745c791: https://travis-ci.org/kamalmarhubi/nix-rust/builds/105444337)

[kamalmarhubi]

Thanks @polachok!