4 files changed, 309 insertions, 9 deletions

diff --git a/tests/runner.c b/tests/runner.c
index 10b6625..db62246 100644
--- a/tests/runner.c
+++ b/tests/runner.c
@@ -22,6 +22,7 @@ static const struct module modules[] = {{"scan_urlencoded_query", test_scan_urle
                                         {"ot_scan", test_scan},
                                         {"ot_buf", test_buf},
                                         {"ot_ip", test_ip},
+                                        {"ot_socket", test_socket},
                                         {"ot_accesslist", test_accesslist},
                                         {NULL, NULL}};
 
diff --git a/tests/t_scan_urlencoded_query.c b/tests/t_scan_urlencoded_query.c
index 6fa34e0..f660ea3 100644
--- a/tests/t_scan_urlencoded_query.c
+++ b/tests/t_scan_urlencoded_query.c
@@ -295,14 +295,6 @@ static void case_fmt_hex_is_lower_case(void) {
   CHECK_STR(dest, "%00%ab%ff");
 }
 
-static void case_fmt_length_only(void) {
-  const uint8_t src[] = {'a', 0xff, 'b'};
-
-  test_case("a null destination reports the length without writing");
-  CHECK_INT(fmt_urlencoded(NULL, src, sizeof(src)), 5);
-  CHECK_INT(fmt_urlencoded(NULL, src, 0), 0);
-}
-
 static void case_fmt_zero_length(void) {
   char dest[4];
 
@@ -374,7 +366,6 @@ void test_scan_urlencoded_query(void) {
   case_fmt_passes_safe_bytes();
   case_fmt_escapes_the_rest();
   case_fmt_hex_is_lower_case();
-  case_fmt_length_only();
   case_fmt_zero_length();
   case_fmt_worst_case_fits();
   case_fmt_decoder_disagrees_on_dollar();
diff --git a/tests/t_socket.c b/tests/t_socket.c
new file mode 100644
index 0000000..bbc9e5e
--- /dev/null
+++ b/tests/t_socket.c
@@ -0,0 +1,307 @@
+/* This software was written by Dirk Engling <erdgeist@erdgeist.org>
+   It is considered beerware. Prost. Skol. Cheers or whatever.
+
+   $id$ */
+
+/* System */
+#include <errno.h>
+#include <fcntl.h>
+#include <netinet/in.h>
+#include <poll.h>
+#include <string.h>
+#include <sys/socket.h>
+#include <unistd.h>
+
+/* Opentracker */
+#include "ot_ip.h"
+#include "ot_socket.h"
+#include "test.h"
+
+/* These cases talk to the loopback interface rather than to a mock, because
+   what is worth pinning down here is the behaviour of the socket options the
+   rest of the tracker relies on, and only the kernel can report that. */
+
+static const char ip_any[16] = {0};
+static const char ip_loopback[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+static const char ip_v4_loopback[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xff, 0xff, 127, 0, 0, 1};
+/* RFC 3849 reserves 2001:db8::/32 for documentation, so no host carries it */
+static const char ip_unassigned[16] = {0x20, 0x01, 0x0d, 0xb8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1};
+
+/* What a socket ended up bound to, which is not always what was asked for: a
+   FreeBSD jail, for one, rewrites a loopback bind to the address it was given.
+   A datagram a socket sends to itself must come back carrying this, so the
+   cases below compare against it rather than against the address they passed
+   to the bind. */
+static void bound_address(int s, char ip[16], uint16_t *port) {
+  struct sockaddr_storage ss;
+  socklen_t               len = sizeof(ss);
+
+  memset(&ss, 0, sizeof(ss));
+  memset(ip, 0, 16);
+  *port = 0;
+  if (getsockname(s, (struct sockaddr *)&ss, &len) == -1)
+    return;
+
+  /* Spelled out here rather than borrowed from ot_socket.c, so that a wrong
+     idea about how a kernel reports an address cannot hide in both at once */
+  if (ss.ss_family == AF_INET) {
+    const struct sockaddr_in *sa = (const struct sockaddr_in *)&ss;
+
+    memcpy(ip, ip_v4mapped_prefix, sizeof(ip_v4mapped_prefix));
+    memcpy(ip + sizeof(ip_v4mapped_prefix), &sa->sin_addr, 4);
+    *port = ntohs(sa->sin_port);
+  } else {
+    const struct sockaddr_in6 *sa = (const struct sockaddr_in6 *)&ss;
+
+    memcpy(ip, &sa->sin6_addr, 16);
+    *port = ntohs(sa->sin6_port);
+  }
+}
+
+static int is_nonblocking(int s) {
+  int flags = fcntl(s, F_GETFL, 0);
+  return flags != -1 && (flags & O_NONBLOCK) != 0;
+}
+
+static int sock_flag(int s, int level, int option) {
+  int       value = -1;
+  socklen_t len   = sizeof(value);
+
+  if (getsockopt(s, level, option, &value, &len) == -1)
+    return -1;
+  return value != 0;
+}
+
+/* Wait for one event, so a test never blocks forever on a lost packet */
+static int wait_for(int s, short events) {
+  struct pollfd pfd;
+
+  pfd.fd      = s;
+  pfd.events  = events;
+  pfd.revents = 0;
+  return poll(&pfd, 1, 2000) == 1;
+}
+
+/* ------------------------------------------------------------ invariants */
+
+/* Asked for the wildcard address, because a socket bound to one specific
+   IPv6 address cannot carry IPv4 traffic anyway and Linux turns IPV6_V6ONLY
+   back on in that case. The tracker's own default is the wildcard. */
+static void case_socket_options(void) {
+  int udp, tcp;
+
+  test_case("a bound udp socket is non blocking and serves both families");
+  udp = socket_udp_bind(ip_any, 0);
+  CHECK(udp != -1);
+  CHECK_INT(is_nonblocking(udp), 1);
+  CHECK_INT(sock_flag(udp, IPPROTO_IPV6, IPV6_V6ONLY), 0);
+  close(udp);
+
+  test_case("a listening tcp socket is non blocking and serves both families");
+  tcp = socket_tcp_listen(ip_any, 0);
+  CHECK(tcp != -1);
+  CHECK_INT(is_nonblocking(tcp), 1);
+  CHECK_INT(sock_flag(tcp, IPPROTO_IPV6, IPV6_V6ONLY), 0);
+  close(tcp);
+}
+
+#ifdef SO_REUSEPORT
+static void case_reuseport_is_opt_in(void) {
+  int s;
+
+  test_case("reuseport stays off unless the operator asks for it");
+  g_socket_reuseport = 0;
+  s                  = socket_udp_bind(ip_loopback, 0);
+  CHECK(s != -1);
+  CHECK_INT(sock_flag(s, SOL_SOCKET, SO_REUSEPORT), 0);
+  close(s);
+
+  test_case("reuseport reaches the socket when it is asked for");
+  g_socket_reuseport = 1;
+  s                  = socket_udp_bind(ip_loopback, 0);
+  CHECK(s != -1);
+  CHECK_INT(sock_flag(s, SOL_SOCKET, SO_REUSEPORT), 1);
+  close(s);
+
+  g_socket_reuseport = 0;
+}
+#endif
+
+/* ------------------------------------------------------------------- udp */
+
+static void case_udp_roundtrip(void) {
+  char     buf[32], ip[16], own_ip[16];
+  uint16_t port = 0, own_port = 0;
+  int      s;
+
+  test_case("a datagram comes back with the sender's address and port");
+  s = socket_udp_bind(ip_loopback, 0);
+  CHECK(s != -1);
+  bound_address(s, own_ip, &own_port);
+  CHECK(own_port != 0);
+  CHECK_INT(socket_send(s, "opentracker", 11, own_ip, own_port), 11);
+  CHECK_INT(wait_for(s, POLLIN), 1);
+  CHECK_INT(socket_recv(s, buf, sizeof(buf), ip, &port), 11);
+  CHECK_MEM(buf, "opentracker", 11);
+  CHECK_MEM(ip, own_ip, 16);
+  CHECK_INT(port, own_port);
+  close(s);
+}
+
+/* An IPv4 peer must arrive in the v4 mapped form opentracker stores, however
+   the kernel chose to report it. Linux fills a v4 mapped sockaddr_in6 here
+   and the BSDs fill a plain sockaddr_in, so this is the case that tells the
+   two apart. */
+static void case_udp_v4mapped(void) {
+  char     buf[32], ip[16], receiver_ip[16], sender_ip[16];
+  uint16_t port = 0, receiver_port = 0, sender_port = 0;
+  int      receiver, sender;
+
+  test_case("an IPv4 sender is reported in its v4 mapped form");
+  receiver = socket_udp_bind(ip_v4_loopback, 0);
+  CHECK(receiver != -1);
+  sender = socket_udp_bind(ip_v4_loopback, 0);
+  CHECK(sender != -1);
+  bound_address(receiver, receiver_ip, &receiver_port);
+  bound_address(sender, sender_ip, &sender_port);
+  CHECK_INT(ip_is_v4mapped(sender_ip), 1);
+
+  CHECK_INT(socket_send(sender, "v4", 2, receiver_ip, receiver_port), 2);
+  CHECK_INT(wait_for(receiver, POLLIN), 1);
+  CHECK_INT(socket_recv(receiver, buf, sizeof(buf), ip, &port), 2);
+  CHECK_INT(ip_is_v4mapped(ip), 1);
+  CHECK_MEM(ip, sender_ip, 16);
+  CHECK_INT(port, sender_port);
+  close(sender);
+  close(receiver);
+}
+
+/* ------------------------------------------------------------------- tcp */
+
+static void case_tcp_roundtrip(void) {
+  char     ip[16], listener_ip[16], client_ip[16];
+  uint16_t port = 0, listener_port = 0, client_port = 0;
+  int      listener, client, accepted;
+
+  test_case("a connection is accepted and carries the caller's address");
+  listener = socket_tcp_listen(ip_loopback, 0);
+  CHECK(listener != -1);
+  bound_address(listener, listener_ip, &listener_port);
+  CHECK(listener_port != 0);
+
+  client = socket_tcp_connect(ip_loopback, 0, listener_ip, listener_port);
+  CHECK(client != -1);
+  CHECK_INT(wait_for(client, POLLOUT), 1);
+  CHECK_INT(socket_is_connected(client), 1);
+  bound_address(client, client_ip, &client_port);
+
+  CHECK_INT(wait_for(listener, POLLIN), 1);
+  accepted = socket_accept(listener, ip, &port);
+  CHECK(accepted != -1);
+  CHECK_MEM(ip, client_ip, 16);
+  CHECK_INT(port, client_port);
+
+  test_case("the accepted socket is handed over non blocking");
+  CHECK_INT(is_nonblocking(accepted), 1);
+
+  close(accepted);
+  close(client);
+  close(listener);
+}
+
+static void case_connect_refused(void) {
+  char     listener_ip[16];
+  int      listener, client;
+  uint16_t port;
+
+  test_case("a refused connection is reported by socket_is_connected");
+  /* Take a port, then drop it, so nothing is listening on it any more */
+  listener = socket_tcp_listen(ip_loopback, 0);
+  CHECK(listener != -1);
+  bound_address(listener, listener_ip, &port);
+  close(listener);
+
+  client = socket_tcp_connect(ip_loopback, 0, listener_ip, port);
+  CHECK(client != -1);
+  CHECK_INT(wait_for(client, POLLOUT), 1);
+  CHECK_INT(socket_is_connected(client), 0);
+  CHECK_INT(errno, ECONNREFUSED);
+  close(client);
+}
+
+/* --------------------------------------------------------- failed binds */
+
+/* The lowest free descriptor is handed out first, so a number that stays put
+   across a failing call shows the call did not keep a socket behind. */
+static int next_descriptor(void) {
+  int fd = dup(0);
+
+  if (fd != -1)
+    close(fd);
+  return fd;
+}
+
+static void case_bind_failure(void) {
+  int before;
+
+  test_case("binding an address no interface carries fails");
+  before = next_descriptor();
+  CHECK(before != -1);
+  CHECK_INT(socket_tcp_listen(ip_unassigned, 0), -1);
+  CHECK_INT(socket_udp_bind(ip_unassigned, 0), -1);
+
+  test_case("a failed bind leaves no socket behind");
+  CHECK_INT(next_descriptor(), before);
+}
+
+#ifdef WANT_SYNC_LIVE
+/* Reading an IPv4 multicast option back: Linux answers with an int, the BSDs
+   with a single byte, so ask how much was written. */
+static int mcast_option(int s, int option) {
+  int       value = 0;
+  socklen_t len   = sizeof(value);
+
+  if (getsockopt(s, IPPROTO_IP, option, &value, &len) == -1)
+    return -1;
+  if (len == 1)
+    return *(unsigned char *)&value;
+  return value;
+}
+
+static void case_mcast_sender(void) {
+  int s;
+
+  test_case("the sending socket keeps its traffic on the local network");
+  s = socket_mcast_bind(ip_v4_loopback, 0, 0);
+  CHECK(s != -1);
+  CHECK_INT(mcast_option(s, IP_MULTICAST_TTL), 1);
+
+  test_case("and does not hear itself unless it is asked to");
+  CHECK_INT(mcast_option(s, IP_MULTICAST_LOOP), 0);
+
+  test_case("whoever sends from it is serving clients too, so it cannot block");
+  CHECK_INT(is_nonblocking(s), 1);
+  close(s);
+
+  test_case("the proxy asks to hear itself and gets it");
+  s = socket_mcast_bind(ip_v4_loopback, 0, 1);
+  CHECK(s != -1);
+  CHECK_INT(mcast_option(s, IP_MULTICAST_LOOP), 1);
+  close(s);
+}
+#endif
+
+void test_socket(void) {
+  case_socket_options();
+#ifdef SO_REUSEPORT
+  case_reuseport_is_opt_in();
+#endif
+  case_udp_roundtrip();
+  case_udp_v4mapped();
+  case_tcp_roundtrip();
+  case_connect_refused();
+  case_bind_failure();
+#ifdef WANT_SYNC_LIVE
+  case_mcast_sender();
+#endif
+}
diff --git a/tests/test.h b/tests/test.h
index 7449b73..d79c4d2 100644
--- a/tests/test.h
+++ b/tests/test.h
@@ -85,5 +85,6 @@ void test_accesslist(void);
 void test_scan(void);
 void test_buf(void);
 void test_ip(void);
+void test_socket(void);
 
 #endif