freebsd,linux: add recvmmsg() + sendmmsg() udp implementation

This commits adds support for recvmmsg() and sendmmsg() extensions to
recvmsg() and sendmsg() that allows the caller to receive and send
multiple message from a socket using a single system call. This has
performance benefits for some applications.

Co-authored-by: Ondřej Surý <[email protected]>
Co-authored-by: Witold Kręcicki <[email protected]>

PR-URL: libuv#2532
Reviewed-By: Ben Noordhuis <[email protected]>
Reviewed-By: Saúl Ibarra Corretgé <[email protected]>
Reviewed-By: Santiago Gimeno <[email protected]>

docs/src/udp.rst

@@ -42,6 +42,12 @@ Data types
             * any traffic, in effect "stealing" the port from the previous listener.
             */
             UV_UDP_REUSEADDR = 4
+            /*
+             * Indicates that the message was received by recvmmsg and that it's not at
+             * the beginning of the buffer allocated by alloc_cb - so the buffer provided
+             * must not be freed by the recv_cb callback.
+             */
+            UV_UDP_MMSG_CHUNK = 8
         };
 
 .. c:type:: void (*uv_udp_send_cb)(uv_udp_send_t* req, int status)
@@ -62,12 +68,13 @@ Data types
     * `buf`: :c:type:`uv_buf_t` with the received data.
     * `addr`: ``struct sockaddr*`` containing the address of the sender.
       Can be NULL. Valid for the duration of the callback only.
-    * `flags`: One or more or'ed UV_UDP_* constants. Right now only
-      ``UV_UDP_PARTIAL`` is used.
+    * `flags`: One or more or'ed UV_UDP_* constants.
 
     The callee is responsible for freeing the buffer, libuv does not reuse it.
     The buffer may be a null buffer (where `buf->base` == NULL and `buf->len` == 0)
-    on error.
+    on error. Don't free the buffer when the UV_UDP_MMSG_CHUNK flag is set.
+    The final callback receives the whole buffer (containing the first chunk)
+    with the UV_UDP_MMSG_CHUNK flag cleared.
 
     .. note::
         The receive callback will be called with `nread` == 0 and `addr` == NULL when there is

include/uv.h

@@ -605,7 +605,13 @@ enum uv_udp_flags {
    * (provided they all set the flag) but only the last one to bind will receive
    * any traffic, in effect "stealing" the port from the previous listener.
    */
-  UV_UDP_REUSEADDR = 4
+  UV_UDP_REUSEADDR = 4,
+  /*
+   * Indicates that the message was received by recvmmsg and that it's not at
+   * the beginning of the buffer allocated by alloc_cb - so the buffer provided
+   * must not be freed by the recv_cb callback.
+   */
+  UV_UDP_MMSG_CHUNK = 8
 };
 
 typedef void (*uv_udp_send_cb)(uv_udp_send_t* req, int status);

src/unix/freebsd.c

@@ -288,3 +288,28 @@ int uv_cpu_info(uv_cpu_info_t** cpu_infos, int* count) {
   uv__free(cp_times);
   return 0;
 }
+
+
+int uv__sendmmsg(int fd,
+                 struct uv__mmsghdr* mmsg,
+                 unsigned int vlen,
+                 unsigned int flags) {
+#if __FreeBSD__ >= 11
+  return sendmmsg(fd, mmsg, vlen, flags);
+#else
+  return errno = ENOSYS, -1;
+#endif
+}
+
+
+int uv__recvmmsg(int fd,
+                 struct uv__mmsghdr* mmsg,
+                 unsigned int vlen,
+                 unsigned int flags,
+                 struct timespec* timeout) {
+#if __FreeBSD__ >= 11
+  return recvmmsg(fd, mmsg, vlen, flags, timeout);
+#else
+  return errno = ENOSYS, -1;
+#endif
+}

src/unix/internal.h

@@ -31,6 +31,7 @@
 #include <fcntl.h>  /* O_CLOEXEC and O_NONBLOCK, if supported. */
 #include <stdio.h>
 #include <errno.h>
+#include <sys/socket.h>
 
 #if defined(__STRICT_ANSI__)
 # define inline __inline
@@ -327,4 +328,27 @@ int uv__getsockpeername(const uv_handle_t* handle,
                         struct sockaddr* name,
                         int* namelen);
 
+#if defined(__linux__)            ||                                      \
+    defined(__FreeBSD__)          ||                                      \
+    defined(__FreeBSD_kernel__)
+#define HAVE_MMSG 1
+struct uv__mmsghdr {
+  struct msghdr msg_hdr;
+  unsigned int msg_len;
+};
+
+int uv__recvmmsg(int fd,
+                 struct uv__mmsghdr* mmsg,
+                 unsigned int vlen,
+                 unsigned int flags,
+                 struct timespec* timeout);
+int uv__sendmmsg(int fd,
+                 struct uv__mmsghdr* mmsg,
+                 unsigned int vlen,
+                 unsigned int flags);
+#else
+#define HAVE_MMSG 0
+#endif
+
+
 #endif /* UV_UNIX_INTERNAL_H_ */

src/unix/linux-syscalls.c

@@ -126,6 +126,7 @@
 # endif
 #endif /* __NR_getrandom */
 
+struct uv__mmsghdr;
 
 int uv__sendmmsg(int fd,
                  struct uv__mmsghdr* mmsg,

src/unix/linux-syscalls.h

@@ -61,20 +61,6 @@ struct uv__statx {
   uint64_t unused1[14];
 };
 
-struct uv__mmsghdr {
-  struct msghdr msg_hdr;
-  unsigned int msg_len;
-};
-
-int uv__recvmmsg(int fd,
-                 struct uv__mmsghdr* mmsg,
-                 unsigned int vlen,
-                 unsigned int flags,
-                 struct timespec* timeout);
-int uv__sendmmsg(int fd,
-                 struct uv__mmsghdr* mmsg,
-                 unsigned int vlen,
-                 unsigned int flags);
 ssize_t uv__preadv(int fd, const struct iovec *iov, int iovcnt, int64_t offset);
 ssize_t uv__pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset);
 int uv__dup3(int oldfd, int newfd, int flags);