win32: avoid mixing SOCKET and file descriptor space

Until now, a win32 SOCKET handle is often cast to an int file
descriptor, as this is what other OS use for sockets. When necessary,
QEMU eventually queries whether it's a socket with the help of
fd_is_socket(). However, there is no guarantee of conflict between the
fd and SOCKET space. Such conflict would have surprising consequences,
we shouldn't mix them.

Also, it is often forgotten that SOCKET must be closed with
closesocket(), and not close().

Instead, let's make the win32 socket wrapper functions return and take a
file descriptor, and let util/ wrappers do the fd/SOCKET conversion as
necessary. A bit of adaptation is necessary in io/ as well.

Unfortunately, we can't drop closesocket() usage, despite
_open_osfhandle() documentation claiming transfer of ownership, testing
shows bad behaviour if you forget to call closesocket().

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Message-Id: <20230221124802.4103554-15-marcandre.lureau@redhat.com>

2 files changed, 192 insertions, 36 deletions

diff --git a/util/aio-win32.c b/util/aio-win32.c
index 08e8f5615d..6bded009a4 100644
--- a/util/aio-win32.c
+++ b/util/aio-win32.c
@@ -73,15 +73,18 @@ void aio_set_fd_handler(AioContext *ctx,
 {
     AioHandler *old_node;
     AioHandler *node = NULL;
+    SOCKET s;
 
     if (!fd_is_socket(fd)) {
         error_report("fd=%d is not a socket, AIO implementation is missing", fd);
         return;
     }
 
+    s = _get_osfhandle(fd);
+
     qemu_lockcnt_lock(&ctx->list_lock);
     QLIST_FOREACH(old_node, &ctx->aio_handlers, node) {
-        if (old_node->pfd.fd == fd && !old_node->deleted) {
+        if (old_node->pfd.fd == s && !old_node->deleted) {
             break;
         }
     }
@@ -92,7 +95,7 @@ void aio_set_fd_handler(AioContext *ctx,
 
         /* Alloc and insert if it's not already there */
         node = g_new0(AioHandler, 1);
-        node->pfd.fd = fd;
+        node->pfd.fd = s;
 
         node->pfd.events = 0;
         if (node->io_read) {
@@ -120,7 +123,7 @@ void aio_set_fd_handler(AioContext *ctx,
 
         QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
         event = event_notifier_get_handle(&ctx->notifier);
-        qemu_socket_select(node->pfd.fd, event, bitmask, NULL);
+        qemu_socket_select(fd, event, bitmask, NULL);
     }
     if (old_node) {
         aio_remove_fd_handler(ctx, old_node);
diff --git a/util/oslib-win32.c b/util/oslib-win32.c
index dbd32acc98..7836fb0be3 100644
--- a/util/oslib-win32.c
+++ b/util/oslib-win32.c
@@ -283,13 +283,20 @@ char *qemu_get_pid_name(pid_t pid)
 }
 
 
-bool qemu_socket_select(SOCKET s, WSAEVENT hEventObject,
+bool qemu_socket_select(int sockfd, WSAEVENT hEventObject,
                         long lNetworkEvents, Error **errp)
 {
+    SOCKET s = _get_osfhandle(sockfd);
+
     if (errp == NULL) {
         errp = &error_warn;
     }
 
+    if (s == INVALID_SOCKET) {
+        error_setg(errp, "invalid socket fd=%d", sockfd);
+        return false;
+    }
+
     if (WSAEventSelect(s, hEventObject, lNetworkEvents) != 0) {
         error_setg_win32(errp, WSAGetLastError(), "failed to WSAEventSelect()");
         return false;
@@ -298,9 +305,9 @@ bool qemu_socket_select(SOCKET s, WSAEVENT hEventObject,
     return true;
 }
 
-bool qemu_socket_unselect(SOCKET s, Error **errp)
+bool qemu_socket_unselect(int sockfd, Error **errp)
 {
-    return qemu_socket_select(s, NULL, 0, errp);
+    return qemu_socket_select(sockfd, NULL, 0, errp);
 }
 
 #undef connect
@@ -308,7 +315,13 @@ int qemu_connect_wrap(int sockfd, const struct sockaddr *addr,
                       socklen_t addrlen)
 {
     int ret;
-    ret = connect(sockfd, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = connect(s, addr, addrlen);
     if (ret < 0) {
         if (WSAGetLastError() == WSAEWOULDBLOCK) {
             errno = EINPROGRESS;
@@ -324,7 +337,13 @@ int qemu_connect_wrap(int sockfd, const struct sockaddr *addr,
 int qemu_listen_wrap(int sockfd, int backlog)
 {
     int ret;
-    ret = listen(sockfd, backlog);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = listen(s, backlog);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -337,7 +356,13 @@ int qemu_bind_wrap(int sockfd, const struct sockaddr *addr,
                    socklen_t addrlen)
 {
     int ret;
-    ret = bind(sockfd, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = bind(s, addr, addrlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -345,28 +370,108 @@ int qemu_bind_wrap(int sockfd, const struct sockaddr *addr,
 }
 
 
-#undef socket
-int qemu_socket_wrap(int domain, int type, int protocol)
+#undef closesocket
+int qemu_closesocket_wrap(int fd)
 {
     int ret;
-    ret = socket(domain, type, protocol);
+    DWORD flags = 0;
+    SOCKET s = _get_osfhandle(fd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    /*
+     * If we were to just call _close on the descriptor, it would close the
+     * HANDLE, but it wouldn't free any of the resources associated to the
+     * SOCKET, and we can't call _close after calling closesocket, because
+     * closesocket has already closed the HANDLE, and _close would attempt to
+     * close the HANDLE again, resulting in a double free. We can however
+     * protect the HANDLE from actually being closed long enough to close the
+     * file descriptor, then close the socket itself.
+     */
+    if (!GetHandleInformation((HANDLE)s, &flags)) {
+        errno = EACCES;
+        return -1;
+    }
+
+    if (!SetHandleInformation((HANDLE)s, HANDLE_FLAG_PROTECT_FROM_CLOSE, HANDLE_FLAG_PROTECT_FROM_CLOSE)) {
+        errno = EACCES;
+        return -1;
+    }
+
+    ret = close(fd);
+
+    if (!SetHandleInformation((HANDLE)s, flags, flags)) {
+        errno = EACCES;
+        return -1;
+    }
+
+    /*
+     * close() returns EBADF since we PROTECT_FROM_CLOSE the underlying handle,
+     * but the FD is actually freed
+     */
+    if (ret < 0 && errno != EBADF) {
+        return ret;
+    }
+
+    ret = closesocket(s);
     if (ret < 0) {
         errno = socket_error();
     }
+
     return ret;
 }
 
 
+#undef socket
+int qemu_socket_wrap(int domain, int type, int protocol)
+{
+    SOCKET s;
+    int fd;
+
+    s = socket(domain, type, protocol);
+    if (s == -1) {
+        errno = socket_error();
+        return -1;
+    }
+
+    fd = _open_osfhandle(s, _O_BINARY);
+    if (fd < 0) {
+        closesocket(s);
+        /* _open_osfhandle may not set errno, and closesocket() may override it */
+        errno = ENOMEM;
+    }
+
+    return fd;
+}
+
+
 #undef accept
 int qemu_accept_wrap(int sockfd, struct sockaddr *addr,
                      socklen_t *addrlen)
 {
-    int ret;
-    ret = accept(sockfd, addr, addrlen);
-    if (ret < 0) {
+    int fd;
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    s = accept(s, addr, addrlen);
+    if (s == -1) {
         errno = socket_error();
+        return -1;
     }
-    return ret;
+
+    fd = _open_osfhandle(s, _O_BINARY);
+    if (fd < 0) {
+        closesocket(s);
+        /* _open_osfhandle may not set errno, and closesocket() may override it */
+        errno = ENOMEM;
+    }
+
+    return fd;
 }
 
 
@@ -374,7 +479,13 @@ int qemu_accept_wrap(int sockfd, struct sockaddr *addr,
 int qemu_shutdown_wrap(int sockfd, int how)
 {
     int ret;
-    ret = shutdown(sockfd, how);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = shutdown(s, how);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -386,19 +497,13 @@ int qemu_shutdown_wrap(int sockfd, int how)
 int qemu_ioctlsocket_wrap(int fd, int req, void *val)
 {
     int ret;
-    ret = ioctlsocket(fd, req, val);
-    if (ret < 0) {
-        errno = socket_error();
-    }
-    return ret;
-}
+    SOCKET s = _get_osfhandle(fd);
 
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
 
-#undef closesocket
-int qemu_closesocket_wrap(int fd)
-{
-    int ret;
-    ret = closesocket(fd);
+    ret = ioctlsocket(s, req, val);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -411,7 +516,13 @@ int qemu_getsockopt_wrap(int sockfd, int level, int optname,
                          void *optval, socklen_t *optlen)
 {
     int ret;
-    ret = getsockopt(sockfd, level, optname, optval, optlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = getsockopt(s, level, optname, optval, optlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -424,7 +535,13 @@ int qemu_setsockopt_wrap(int sockfd, int level, int optname,
                          const void *optval, socklen_t optlen)
 {
     int ret;
-    ret = setsockopt(sockfd, level, optname, optval, optlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = setsockopt(s, level, optname, optval, optlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -437,7 +554,13 @@ int qemu_getpeername_wrap(int sockfd, struct sockaddr *addr,
                           socklen_t *addrlen)
 {
     int ret;
-    ret = getpeername(sockfd, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = getpeername(s, addr, addrlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -450,7 +573,13 @@ int qemu_getsockname_wrap(int sockfd, struct sockaddr *addr,
                           socklen_t *addrlen)
 {
     int ret;
-    ret = getsockname(sockfd, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = getsockname(s, addr, addrlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -462,7 +591,13 @@ int qemu_getsockname_wrap(int sockfd, struct sockaddr *addr,
 ssize_t qemu_send_wrap(int sockfd, const void *buf, size_t len, int flags)
 {
     int ret;
-    ret = send(sockfd, buf, len, flags);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = send(s, buf, len, flags);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -475,7 +610,13 @@ ssize_t qemu_sendto_wrap(int sockfd, const void *buf, size_t len, int flags,
                          const struct sockaddr *addr, socklen_t addrlen)
 {
     int ret;
-    ret = sendto(sockfd, buf, len, flags, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = sendto(s, buf, len, flags, addr, addrlen);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -487,7 +628,13 @@ ssize_t qemu_sendto_wrap(int sockfd, const void *buf, size_t len, int flags,
 ssize_t qemu_recv_wrap(int sockfd, void *buf, size_t len, int flags)
 {
     int ret;
-    ret = recv(sockfd, buf, len, flags);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = recv(s, buf, len, flags);
     if (ret < 0) {
         errno = socket_error();
     }
@@ -500,7 +647,13 @@ ssize_t qemu_recvfrom_wrap(int sockfd, void *buf, size_t len, int flags,
                            struct sockaddr *addr, socklen_t *addrlen)
 {
     int ret;
-    ret = recvfrom(sockfd, buf, len, flags, addr, addrlen);
+    SOCKET s = _get_osfhandle(sockfd);
+
+    if (s == INVALID_SOCKET) {
+        return -1;
+    }
+
+    ret = recvfrom(s, buf, len, flags, addr, addrlen);
     if (ret < 0) {
         errno = socket_error();
     }