Merge branch 'master' into socket-mss

6 files changed, 222 insertions, 2 deletions

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 8865c59f..04e37d28 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -34,8 +34,12 @@ This project adheres to [Semantic Versioning](https://semver.org/).
   (#[1503](https://github.com/nix-rust/nix/pull/1503))
 - Enabled `pwritev` and `preadv` for more operating systems.
   (#[1511](https://github.com/nix-rust/nix/pull/1511))
-- Added support for `TCP_MAXSEG` TCP Maximum Segment Size socket options
+ Added support for `TCP_MAXSEG` TCP Maximum Segment Size socket options
   (#[1292](https://github.com/nix-rust/nix/pull/1292))
+- Added `Ipv4RecvErr` and `Ipv6RecvErr` sockopts and associated control messages.
+  (#[1514](https://github.com/nix-rust/nix/pull/1514))
+- Added `AsRawFd` implementation on `PollFd`.
+  (#[1516](https://github.com/nix-rust/nix/pull/1516))
 
 ### Changed
 
diff --git a/src/poll.rs b/src/poll.rs
index e814337a..0eaf7e1d 100644
--- a/src/poll.rs
+++ b/src/poll.rs
@@ -3,7 +3,7 @@
 use crate::sys::time::TimeSpec;
 #[cfg(any(target_os = "android", target_os = "dragonfly", target_os = "freebsd", target_os = "linux"))]
 use crate::sys::signal::SigSet;
-use std::os::unix::io::RawFd;
+use std::os::unix::io::{AsRawFd, RawFd};
 
 use crate::Result;
 use crate::errno::Errno;
@@ -41,6 +41,12 @@ impl PollFd {
     }
 }
 
+impl AsRawFd for PollFd {
+    fn as_raw_fd(&self) -> RawFd {
+        self.pollfd.fd
+    }
+}
+
 libc_bitflags! {
     /// These flags define the different events that can be monitored by `poll` and `ppoll`
     pub struct PollFlags: libc::c_short {
diff --git a/src/sys/socket/mod.rs b/src/sys/socket/mod.rs
index 33585bb9..50d8b6fb 100644
--- a/src/sys/socket/mod.rs
+++ b/src/sys/socket/mod.rs
@@ -653,6 +653,13 @@ pub enum ControlMessageOwned {
     #[cfg(any(target_os = "android", target_os = "fuchsia", target_os = "linux"))]
     RxqOvfl(u32),
 
+    /// Socket error queue control messages read with the `MSG_ERRQUEUE` flag.
+    #[cfg(any(target_os = "android", target_os = "linux"))]
+    Ipv4RecvErr(libc::sock_extended_err, Option<sockaddr_in>),
+    /// Socket error queue control messages read with the `MSG_ERRQUEUE` flag.
+    #[cfg(any(target_os = "android", target_os = "linux"))]
+    Ipv6RecvErr(libc::sock_extended_err, Option<sockaddr_in6>),
+
     /// Catch-all variant for unimplemented cmsg types.
     #[doc(hidden)]
     Unknown(UnknownCmsg),
@@ -756,6 +763,16 @@ impl ControlMessageOwned {
                 let drop_counter = ptr::read_unaligned(p as *const u32);
                 ControlMessageOwned::RxqOvfl(drop_counter)
             },
+            #[cfg(any(target_os = "android", target_os = "linux"))]
+            (libc::IPPROTO_IP, libc::IP_RECVERR) => {
+                let (err, addr) = Self::recv_err_helper::<sockaddr_in>(p, len);
+                ControlMessageOwned::Ipv4RecvErr(err, addr)
+            },
+            #[cfg(any(target_os = "android", target_os = "linux"))]
+            (libc::IPPROTO_IPV6, libc::IPV6_RECVERR) => {
+                let (err, addr) = Self::recv_err_helper::<sockaddr_in6>(p, len);
+                ControlMessageOwned::Ipv6RecvErr(err, addr)
+            },
             (_, _) => {
                 let sl = slice::from_raw_parts(p, len);
                 let ucmsg = UnknownCmsg(*header, Vec::<u8>::from(sl));
@@ -763,6 +780,24 @@ impl ControlMessageOwned {
             }
         }
     }
+
+    #[cfg(any(target_os = "android", target_os = "linux"))]
+    unsafe fn recv_err_helper<T>(p: *mut libc::c_uchar, len: usize) -> (libc::sock_extended_err, Option<T>) {
+        let ee = p as *const libc::sock_extended_err;
+        let err = ptr::read_unaligned(ee);
+
+        // For errors originating on the network, SO_EE_OFFENDER(ee) points inside the p[..len]
+        // CMSG_DATA buffer.  For local errors, there is no address included in the control
+        // message, and SO_EE_OFFENDER(ee) points beyond the end of the buffer.  So, we need to
+        // validate that the address object is in-bounds before we attempt to copy it.
+        let addrp = libc::SO_EE_OFFENDER(ee) as *const T;
+
+        if addrp.offset(1) as usize - (p as usize) > len {
+            (err, None)
+        } else {
+            (err, Some(ptr::read_unaligned(addrp)))
+        }
+    }
 }
 
 /// A type-safe zero-copy wrapper around a single control message, as used wih
diff --git a/src/sys/socket/sockopt.rs b/src/sys/socket/sockopt.rs
index 67f5f71e..17dd7568 100644
--- a/src/sys/socket/sockopt.rs
+++ b/src/sys/socket/sockopt.rs
@@ -352,6 +352,10 @@ sockopt_impl!(Both, UdpGroSegment, libc::IPPROTO_UDP, libc::UDP_GRO, bool);
 #[cfg(any(target_os = "android", target_os = "fuchsia", target_os = "linux"))]
 sockopt_impl!(Both, RxqOvfl, libc::SOL_SOCKET, libc::SO_RXQ_OVFL, libc::c_int);
 sockopt_impl!(Both, Ipv6V6Only, libc::IPPROTO_IPV6, libc::IPV6_V6ONLY, bool);
+#[cfg(any(target_os = "android", target_os = "linux"))]
+sockopt_impl!(Both, Ipv4RecvErr, libc::IPPROTO_IP, libc::IP_RECVERR, bool);
+#[cfg(any(target_os = "android", target_os = "linux"))]
+sockopt_impl!(Both, Ipv6RecvErr, libc::IPPROTO_IPV6, libc::IPV6_RECVERR, bool);
 
 #[cfg(any(target_os = "android", target_os = "linux"))]
 #[derive(Copy, Clone, Debug)]
diff --git a/test/sys/test_socket.rs b/test/sys/test_socket.rs
index aceffccb..2b0bc831 100644
--- a/test/sys/test_socket.rs
+++ b/test/sys/test_socket.rs
@@ -712,6 +712,7 @@ pub fn test_af_alg_cipher() {
 #[test]
 pub fn test_af_alg_aead() {
     use libc::{ALG_OP_DECRYPT, ALG_OP_ENCRYPT};
+    use nix::fcntl::{fcntl, FcntlArg, OFlag};
     use nix::sys::uio::IoVec;
     use nix::unistd::{read, close};
     use nix::sys::socket::{socket, sendmsg, bind, accept, setsockopt,
@@ -790,6 +791,11 @@ pub fn test_af_alg_aead() {
 
     // allocate buffer for decrypted data
     let mut decrypted = vec![0u8; payload_len + (assoc_size as usize) + auth_size];
+    // Starting with kernel 4.9, the interface changed slightly such that the
+    // authentication tag memory is only needed in the output buffer for encryption
+    // and in the input buffer for decryption.
+    // Do not block on read, as we may have fewer bytes than buffer size
+    fcntl(session_socket,FcntlArg::F_SETFL(OFlag::O_NONBLOCK)).expect("fcntl non_blocking");
     let num_bytes = read(session_socket, &mut decrypted).expect("read decrypt");
 
     assert!(num_bytes >= payload_len + (assoc_size as usize));
@@ -1789,3 +1795,160 @@ fn test_recvmsg_rxq_ovfl() {
     nix::unistd::close(in_socket).unwrap();
     nix::unistd::close(out_socket).unwrap();
 }
+
+#[cfg(any(
+    target_os = "linux",
+    target_os = "android",
+))]
+mod linux_errqueue {
+    use nix::sys::socket::*;
+    use super::{FromStr, SocketAddr};
+
+    // Send a UDP datagram to a bogus destination address and observe an ICMP error (v4).
+    //
+    // Disable the test on QEMU because QEMU emulation of IP_RECVERR is broken (as documented on PR
+    // #1514).
+    #[cfg_attr(qemu, ignore)]
+    #[test]
+    fn test_recverr_v4() {
+        #[repr(u8)]
+        enum IcmpTypes {
+            DestUnreach = 3, // ICMP_DEST_UNREACH
+        }
+        #[repr(u8)]
+        enum IcmpUnreachCodes {
+            PortUnreach = 3, // ICMP_PORT_UNREACH
+        }
+
+        test_recverr_impl::<sockaddr_in, _, _>(
+            "127.0.0.1:6800",
+            AddressFamily::Inet,
+            sockopt::Ipv4RecvErr,
+            libc::SO_EE_ORIGIN_ICMP,
+            IcmpTypes::DestUnreach as u8,
+            IcmpUnreachCodes::PortUnreach as u8,
+            // Closure handles protocol-specific testing and returns generic sock_extended_err for
+            // protocol-independent test impl.
+            |cmsg| {
+                if let ControlMessageOwned::Ipv4RecvErr(ext_err, err_addr) = cmsg {
+                    if let Some(origin) = err_addr {
+                        // Validate that our network error originated from 127.0.0.1:0.
+                        assert_eq!(origin.sin_family, AddressFamily::Inet as _);
+                        assert_eq!(Ipv4Addr(origin.sin_addr), Ipv4Addr::new(127, 0, 0, 1));
+                        assert_eq!(origin.sin_port, 0);
+                    } else {
+                        panic!("Expected some error origin");
+                    }
+                    return *ext_err
+                } else {
+                    panic!("Unexpected control message {:?}", cmsg);
+                }
+            },
+        )
+    }
+
+    // Essentially the same test as v4.
+    //
+    // Disable the test on QEMU because QEMU emulation of IPV6_RECVERR is broken (as documented on
+    // PR #1514).
+    #[cfg_attr(qemu, ignore)]
+    #[test]
+    fn test_recverr_v6() {
+        #[repr(u8)]
+        enum IcmpV6Types {
+            DestUnreach = 1, // ICMPV6_DEST_UNREACH
+        }
+        #[repr(u8)]
+        enum IcmpV6UnreachCodes {
+            PortUnreach = 4, // ICMPV6_PORT_UNREACH
+        }
+
+        test_recverr_impl::<sockaddr_in6, _, _>(
+            "[::1]:6801",
+            AddressFamily::Inet6,
+            sockopt::Ipv6RecvErr,
+            libc::SO_EE_ORIGIN_ICMP6,
+            IcmpV6Types::DestUnreach as u8,
+            IcmpV6UnreachCodes::PortUnreach as u8,
+            // Closure handles protocol-specific testing and returns generic sock_extended_err for
+            // protocol-independent test impl.
+            |cmsg| {
+                if let ControlMessageOwned::Ipv6RecvErr(ext_err, err_addr) = cmsg {
+                    if let Some(origin) = err_addr {
+                        // Validate that our network error originated from localhost:0.
+                        assert_eq!(origin.sin6_family, AddressFamily::Inet6 as _);
+                        assert_eq!(
+                            Ipv6Addr(origin.sin6_addr),
+                            Ipv6Addr::from_std(&"::1".parse().unwrap()),
+                        );
+                        assert_eq!(origin.sin6_port, 0);
+                    } else {
+                        panic!("Expected some error origin");
+                    }
+                    return *ext_err
+                } else {
+                    panic!("Unexpected control message {:?}", cmsg);
+                }
+            },
+        )
+    }
+
+    fn test_recverr_impl<SA, OPT, TESTF>(sa: &str,
+                                         af: AddressFamily,
+                                         opt: OPT,
+                                         ee_origin: u8,
+                                         ee_type: u8,
+                                         ee_code: u8,
+                                         testf: TESTF)
+        where
+            OPT: SetSockOpt<Val = bool>,
+            TESTF: FnOnce(&ControlMessageOwned) -> libc::sock_extended_err,
+    {
+        use nix::errno::Errno;
+        use nix::sys::uio::IoVec;
+
+        const MESSAGE_CONTENTS: &str = "ABCDEF";
+
+        let sock_addr = {
+            let std_sa = SocketAddr::from_str(sa).unwrap();
+            let inet_addr = InetAddr::from_std(&std_sa);
+            SockAddr::new_inet(inet_addr)
+        };
+        let sock = socket(af, SockType::Datagram, SockFlag::SOCK_CLOEXEC, None).unwrap();
+        setsockopt(sock, opt, &true).unwrap();
+        if let Err(e) = sendto(sock, MESSAGE_CONTENTS.as_bytes(), &sock_addr, MsgFlags::empty()) {
+            assert_eq!(e, Errno::EADDRNOTAVAIL);
+            println!("{:?} not available, skipping test.", af);
+            return;
+        }
+
+        let mut buf = [0u8; 8];
+        let iovec = [IoVec::from_mut_slice(&mut buf)];
+        let mut cspace = cmsg_space!(libc::sock_extended_err, SA);
+
+        let msg = recvmsg(sock, &iovec, Some(&mut cspace), MsgFlags::MSG_ERRQUEUE).unwrap();
+        // The sent message / destination associated with the error is returned:
+        assert_eq!(msg.bytes, MESSAGE_CONTENTS.as_bytes().len());
+        assert_eq!(&buf[..msg.bytes], MESSAGE_CONTENTS.as_bytes());
+        // recvmsg(2): "The original destination address of the datagram that caused the error is
+        // supplied via msg_name;" however, this is not literally true.  E.g., an earlier version
+        // of this test used 0.0.0.0 (::0) as the destination address, which was mutated into
+        // 127.0.0.1 (::1).
+        assert_eq!(msg.address, Some(sock_addr));
+
+        // Check for expected control message.
+        let ext_err = match msg.cmsgs().next() {
+            Some(cmsg) => testf(&cmsg),
+            None => panic!("No control message"),
+        };
+
+        assert_eq!(ext_err.ee_errno, libc::ECONNREFUSED as u32);
+        assert_eq!(ext_err.ee_origin, ee_origin);
+        // ip(7): ee_type and ee_code are set from the type and code fields of the ICMP (ICMPv6)
+        // header.
+        assert_eq!(ext_err.ee_type, ee_type);
+        assert_eq!(ext_err.ee_code, ee_code);
+        // ip(7): ee_info contains the discovered MTU for EMSGSIZE errors.
+        assert_eq!(ext_err.ee_info, 0);
+    }
+}
diff --git a/test/test_poll.rs b/test/test_poll.rs
index 0395512b..ee89c4a0 100644
--- a/test/test_poll.rs
+++ b/test/test_poll.rs
@@ -64,3 +64,11 @@ fn test_ppoll() {
     assert_eq!(nfds, 1);
     assert!(fds[0].revents().unwrap().contains(PollFlags::POLLIN));
 }
+
+#[test]
+fn test_pollfd_fd() {
+    use std::os::unix::io::AsRawFd;
+
+    let pfd = PollFd::new(0x1234, PollFlags::empty());
+    assert_eq!(pfd.as_raw_fd(), 0x1234);
+}