fix: support TCP keepalive socket options

expected/wasm32-wasi-eh/predefined-macros.txt

@@ -6227,6 +6227,9 @@
 #define __WASI_SOCK_OPTION_SEND_BUF_SIZE (UINT8_C(16))
 #define __WASI_SOCK_OPTION_SEND_LOWAT (UINT8_C(18))
 #define __WASI_SOCK_OPTION_SEND_TIMEOUT (UINT8_C(20))
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
 #define __WASI_SOCK_OPTION_TTL (UINT8_C(23))
 #define __WASI_SOCK_OPTION_TYPE (UINT8_C(25))
 #define __WASI_SOCK_PROTO_AH (UINT16_C(51))

expected/wasm32-wasi-ehpic/predefined-macros.txt

@@ -6236,6 +6236,9 @@
 #define __WASI_SOCK_OPTION_SEND_BUF_SIZE (UINT8_C(16))
 #define __WASI_SOCK_OPTION_SEND_LOWAT (UINT8_C(18))
 #define __WASI_SOCK_OPTION_SEND_TIMEOUT (UINT8_C(20))
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
 #define __WASI_SOCK_OPTION_TTL (UINT8_C(23))
 #define __WASI_SOCK_OPTION_TYPE (UINT8_C(25))
 #define __WASI_SOCK_PROTO_AH (UINT16_C(51))

expected/wasm32-wasi-threads/predefined-macros.txt

@@ -6018,6 +6018,9 @@
 #define __WASI_SOCK_OPTION_SEND_BUF_SIZE (UINT8_C(16))
 #define __WASI_SOCK_OPTION_SEND_LOWAT (UINT8_C(18))
 #define __WASI_SOCK_OPTION_SEND_TIMEOUT (UINT8_C(20))
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
 #define __WASI_SOCK_OPTION_TTL (UINT8_C(23))
 #define __WASI_SOCK_OPTION_TYPE (UINT8_C(25))
 #define __WASI_SOCK_PROTO_AH (UINT16_C(51))

expected/wasm32-wasi/predefined-macros.txt

@@ -6235,6 +6235,9 @@
 #define __WASI_SOCK_OPTION_SEND_BUF_SIZE (UINT8_C(16))
 #define __WASI_SOCK_OPTION_SEND_LOWAT (UINT8_C(18))
 #define __WASI_SOCK_OPTION_SEND_TIMEOUT (UINT8_C(20))
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
 #define __WASI_SOCK_OPTION_TTL (UINT8_C(23))
 #define __WASI_SOCK_OPTION_TYPE (UINT8_C(25))
 #define __WASI_SOCK_PROTO_AH (UINT16_C(51))

expected/wasm64-wasi/predefined-macros.txt

@@ -6083,6 +6083,9 @@
 #define __WASI_SOCK_OPTION_SEND_BUF_SIZE (UINT8_C(16))
 #define __WASI_SOCK_OPTION_SEND_LOWAT (UINT8_C(18))
 #define __WASI_SOCK_OPTION_SEND_TIMEOUT (UINT8_C(20))
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
 #define __WASI_SOCK_OPTION_TTL (UINT8_C(23))
 #define __WASI_SOCK_OPTION_TYPE (UINT8_C(25))
 #define __WASI_SOCK_PROTO_AH (UINT16_C(51))

libc-bottom-half/cloudlibc/src/libc/sys/socket/getsockopt.c

@@ -11,9 +11,15 @@ int getsockopt(int socket, int level, int option_name,
     level = SOL_SOCKET;
     option_name = SO_ONLYV6;
   }
-  if (level == IPPROTO_TCP && option_name == TCP_NODELAY) {
+  if (level == IPPROTO_TCP) {
+    switch (option_name) {
+      case TCP_NODELAY: option_name = SO_NODELAY; break;
+      case TCP_KEEPIDLE: option_name = __WASI_SOCK_OPTION_TCP_KEEP_IDLE; break;
+      case TCP_KEEPINTVL: option_name = __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL; break;
+      case TCP_KEEPCNT: option_name = __WASI_SOCK_OPTION_TCP_KEEP_COUNT; break;
+      default: errno = ENOPROTOOPT; return -1;
+    }
     level = SOL_SOCKET;
-    option_name = SO_NODELAY;
   }
 
   if (level != SOL_SOCKET) {
@@ -97,16 +103,25 @@ int getsockopt(int socket, int level, int option_name,
     case SO_RCVBUF:
     case SO_SNDBUF:
     case SO_TTL:
-    case SO_MCASTTTLV4: {
+    case SO_MCASTTTLV4:
+    case __WASI_SOCK_OPTION_TCP_KEEP_IDLE:
+    case __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL:
+    case __WASI_SOCK_OPTION_TCP_KEEP_COUNT: {
+      if (option_len == NULL || (*option_len != 0 && option_value == NULL)) {
+        errno = EFAULT;
+        return -1;
+      }
       __wasi_filesize_t fs;
       __wasi_errno_t error = __wasi_sock_get_opt_size(socket, option_name, &fs);
       if (error != 0) {
         errno = error;
         return -1;
       }
       socklen_t len = fs;
-      memcpy(option_value, &len, *option_len < sizeof(socklen_t ) ? *option_len : sizeof(socklen_t ));
-      *option_len = sizeof(socklen_t );
+      if (*option_len > sizeof(len))
+        *option_len = sizeof(len);
+      if (*option_len != 0)
+        memcpy(option_value, &len, *option_len);
       return 0;
     }
     case SO_PROTOCOL: {

libc-bottom-half/cloudlibc/src/libc/sys/socket/setsockopt.c

@@ -11,9 +11,15 @@ int setsockopt(int socket, int level, int option_name, const void *restrict opti
     level = SOL_SOCKET;
     option_name = SO_ONLYV6;
   }
-  if (level == IPPROTO_TCP && option_name == TCP_NODELAY) {
+  if (level == IPPROTO_TCP) {
+    switch (option_name) {
+      case TCP_NODELAY: option_name = SO_NODELAY; break;
+      case TCP_KEEPIDLE: option_name = __WASI_SOCK_OPTION_TCP_KEEP_IDLE; break;
+      case TCP_KEEPINTVL: option_name = __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL; break;
+      case TCP_KEEPCNT: option_name = __WASI_SOCK_OPTION_TCP_KEEP_COUNT; break;
+      default: errno = ENOSYS; return -1;
+    }
     level = SOL_SOCKET;
-    option_name = SO_NODELAY;
   }
 
   if(level!=SOL_SOCKET) {
@@ -86,19 +92,24 @@ int setsockopt(int socket, int level, int option_name, const void *restrict opti
       return 0;
     }
     case SO_RCVBUF:
-	case SO_SNDBUF:
-	case SO_TTL:
-	case SO_MCASTTTLV4: {
-	  __wasi_filesize_t fs;
-	  if (option_len >= sizeof(socklen_t)) {
-		socklen_t *len = (socklen_t *)option_value;
-		fs = *len;
-	  } else {
-		errno = EINVAL;
-    	return -1;
-	  }
-
-      __wasi_errno_t error = __wasi_sock_set_opt_size(socket, option_name, fs);
+    case SO_SNDBUF:
+    case SO_TTL:
+    case SO_MCASTTTLV4:
+    case __WASI_SOCK_OPTION_TCP_KEEP_IDLE:
+    case __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL:
+    case __WASI_SOCK_OPTION_TCP_KEEP_COUNT: {
+      socklen_t value;
+      if (option_len < sizeof(value)) {
+        errno = EINVAL;
+        return -1;
+      }
+      if (option_value == NULL) {
+        errno = EFAULT;
+        return -1;
+      }
+      memcpy(&value, option_value, sizeof(value));
+
+      __wasi_errno_t error = __wasi_sock_set_opt_size(socket, option_name, value);
       if (error != 0) {
         errno = error;
         return -1;

libc-bottom-half/headers/public/wasi/api_wasix.h

@@ -2529,6 +2529,24 @@ typedef uint8_t __wasi_sock_option_t;
  */
 #define __WASI_SOCK_OPTION_PROTO (UINT8_C(26))
 
+/**
+ * TCP idle seconds before sending keepalive probes (positive integer).
+ * Read and written using sock_get_opt_size and sock_set_opt_size.
+ */
+#define __WASI_SOCK_OPTION_TCP_KEEP_IDLE (UINT8_C(27))
+
+/**
+ * TCP seconds between keepalive probes (positive integer).
+ * Read and written using sock_get_opt_size and sock_set_opt_size.
+ */
+#define __WASI_SOCK_OPTION_TCP_KEEP_INTERVAL (UINT8_C(28))
+
+/**
+ * TCP unanswered keepalive probes before closing the connection (positive integer).
+ * Read and written using sock_get_opt_size and sock_set_opt_size.
+ */
+#define __WASI_SOCK_OPTION_TCP_KEEP_COUNT (UINT8_C(29))
+
 _Static_assert(sizeof(__wasi_sock_option_t) == 1, "witx calculated size");
 _Static_assert(_Alignof(__wasi_sock_option_t) == 1, "witx calculated align");
 

test/wasix/c_tcp_keepalive/CMakeLists.txt

@@ -0,0 +1,4 @@
+cmake_minimum_required(VERSION 3.5.0)
+project(c_tcp_keepalive C)
+
+add_executable(main main.c)

test/wasix/c_tcp_keepalive/main.c

@@ -0,0 +1,124 @@
+#include <assert.h>
+#include <errno.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/socket.h>
+#include <unistd.h>
+#include <wasi/api.h>
+
+_Static_assert(__WASI_SOCK_OPTION_TCP_KEEP_IDLE == 27, "keep idle ABI");
+_Static_assert(__WASI_SOCK_OPTION_TCP_KEEP_INTERVAL == 28, "keep interval ABI");
+_Static_assert(__WASI_SOCK_OPTION_TCP_KEEP_COUNT == 29, "keep count ABI");
+
+static const int options[] = {TCP_KEEPIDLE, TCP_KEEPINTVL, TCP_KEEPCNT};
+static const int values[] = {31, 7, 4};
+
+static void check_options(int fd) {
+  for (unsigned i = 0; i < sizeof(options) / sizeof(options[0]); i++) {
+    int value = 0;
+    socklen_t len = sizeof(value);
+    assert(getsockopt(fd, IPPROTO_TCP, options[i], &value, &len) == 0);
+    assert(len == sizeof(value));
+    assert(value == values[i]);
+  }
+}
+
+static void check_validation(int fd) {
+  for (unsigned i = 0; i < sizeof(options) / sizeof(options[0]); i++) {
+    int value = 0;
+    errno = 0;
+    assert(setsockopt(fd, IPPROTO_TCP, options[i], &value, sizeof(value)) == -1);
+    assert(errno == EINVAL);
+    value = -1;
+    errno = 0;
+    assert(setsockopt(fd, IPPROTO_TCP, options[i], &value, sizeof(value)) == -1);
+    assert(errno == EINVAL);
+    value = 1;
+    errno = 0;
+    assert(setsockopt(fd, IPPROTO_TCP, options[i], &value, sizeof(value) - 1) == -1);
+    assert(errno == EINVAL);
+    errno = 0;
+    assert(setsockopt(fd, IPPROTO_TCP, options[i], NULL, sizeof(value)) == -1);
+    assert(errno == EFAULT);
+  }
+  check_options(fd);
+}
+
+static void check_output_lengths(int fd, int level, int option, int expected) {
+  unsigned char output[sizeof(int) + 1];
+  memset(output, 0xa5, sizeof(output));
+  socklen_t len = 1;
+  assert(getsockopt(fd, level, option, output, &len) == 0);
+  assert(len == 1);
+  assert(output[0] == ((const unsigned char *)&expected)[0]);
+  assert(output[1] == 0xa5);
+
+  len = sizeof(output);
+  assert(getsockopt(fd, level, option, output, &len) == 0);
+  assert(len == sizeof(int));
+  assert(memcmp(output, &expected, sizeof(int)) == 0);
+  assert(output[sizeof(int)] == 0xa5);
+
+  len = 0;
+  assert(getsockopt(fd, level, option, NULL, &len) == 0);
+  assert(len == 0);
+  len = sizeof(int);
+  errno = 0;
+  assert(getsockopt(fd, level, option, NULL, &len) == -1);
+  assert(errno == EFAULT);
+  errno = 0;
+  assert(getsockopt(fd, level, option, output, NULL) == -1);
+  assert(errno == EFAULT);
+}
+
+int main(void) {
+  int fd = socket(AF_INET, SOCK_STREAM, 0);
+  assert(fd >= 0);
+  int enabled = 1;
+  assert(setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &enabled, sizeof(enabled)) == 0);
+  for (unsigned i = 0; i < sizeof(options) / sizeof(options[0]); i++)
+    assert(setsockopt(fd, IPPROTO_TCP, options[i], &values[i], sizeof(values[i])) == 0);
+  check_options(fd);
+  check_validation(fd);
+  check_output_lengths(fd, IPPROTO_TCP, TCP_KEEPIDLE, values[0]);
+  int buffer_size = 4096;
+  assert(setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &buffer_size, sizeof(buffer_size)) == 0);
+  check_output_lengths(fd, SOL_SOCKET, SO_RCVBUF, buffer_size);
+
+  int server = socket(AF_INET, SOCK_STREAM, 0);
+  assert(server >= 0);
+  struct sockaddr_in address = {0};
+  address.sin_family = AF_INET;
+  address.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
+  assert(bind(server, (struct sockaddr *)&address, sizeof(address)) == 0);
+  assert(listen(server, 1) == 0);
+  socklen_t len = sizeof(address);
+  assert(getsockname(server, (struct sockaddr *)&address, &len) == 0);
+  assert(connect(fd, (struct sockaddr *)&address, sizeof(address)) == 0);
+  int peer = accept(server, NULL, NULL);
+  assert(peer >= 0);
+  check_options(fd);
+  check_validation(fd);
+
+  int updated = 9;
+  assert(setsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &updated, sizeof(updated)) == 0);
+  int actual = 0;
+  len = sizeof(actual);
+  assert(getsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &actual, &len) == 0);
+  assert(actual == updated);
+
+  assert(close(peer) == 0);
+  assert(close(server) == 0);
+  assert(close(fd) == 0);
+  errno = 0;
+  assert(setsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &updated, sizeof(updated)) == -1);
+  assert(errno == EBADF);
+  errno = 0;
+  len = sizeof(actual);
+  assert(getsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &actual, &len) == -1);
+  assert(errno == EBADF);
+  puts("TCP keepalive options passed");
+  return 0;
+}

test/wasix/c_tcp_keepalive/test.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+wasmer run --net ./main