Re: [PATCH v4 04/11] osdep: implement qemu_socketpair() for win32

[Marc-André Lureau]

Hi

On Tue, Mar 7, 2023 at 6:50 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
>
> On Mon, Mar 06, 2023 at 04:27:44PM +0400, marcandre.lureau@redhat.com wrote:
> > From: Marc-André Lureau <marcandre.lureau@redhat.com>
> >
> > Manually implement a socketpair() function, using UNIX sockets and
> > simple peer credential checking.
> >
> > QEMU doesn't make much use of socketpair, beside vhost-user which is not
> > available for win32 at this point. However, I intend to use it for
> > writing some new portable tests.
> >
> > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
> > ---
> >  include/qemu/sockets.h |   2 -
> >  util/oslib-win32.c     | 110 +++++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 110 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/qemu/sockets.h b/include/qemu/sockets.h
> > index 2b0698a7c9..d935fd80da 100644
> > --- a/include/qemu/sockets.h
> > +++ b/include/qemu/sockets.h
> > @@ -15,7 +15,6 @@ int inet_aton(const char *cp, struct in_addr *ia);
> >  bool fd_is_socket(int fd);
> >  int qemu_socket(int domain, int type, int protocol);
> >
> > -#ifndef WIN32
> >  /**
> >   * qemu_socketpair:
> >   * @domain: specifies a communication domain, such as PF_UNIX
> > @@ -30,7 +29,6 @@ int qemu_socket(int domain, int type, int protocol);
> >   * Return 0 on success.
> >   */
> >  int qemu_socketpair(int domain, int type, int protocol, int sv[2]);
> > -#endif
> >
> >  int qemu_accept(int s, struct sockaddr *addr, socklen_t *addrlen);
> >  /*
> > diff --git a/util/oslib-win32.c b/util/oslib-win32.c
> > index 29a667ae3d..16f8a67f7e 100644
> > --- a/util/oslib-win32.c
> > +++ b/util/oslib-win32.c
> > @@ -310,6 +310,116 @@ bool qemu_socket_unselect(int sockfd, Error **errp)
> >      return qemu_socket_select(sockfd, NULL, 0, errp);
> >  }
> >
> > +int qemu_socketpair(int domain, int type, int protocol, int sv[2])
> > +{
> > +    struct sockaddr_un addr = {
> > +        0,
> > +    };
> > +    socklen_t socklen;
> > +    int listener = -1;
> > +    int client = -1;
> > +    int server = -1;
> > +    g_autofree char *path = NULL;
> > +    int tmpfd;
> > +    u_long arg;
> > +    int ret = -1;
> > +
> > +    g_return_val_if_fail(sv != NULL, -1);
> > +
> > +    addr.sun_family = AF_UNIX;
> > +    socklen = sizeof(addr);
> > +
> > +    tmpfd = g_file_open_tmp(NULL, &path, NULL);
> > +    if (tmpfd == -1 || !path) {
> > +        errno = EACCES;
> > +        goto out;
> > +    }
> > +
> > +    close(tmpfd);
> > +
> > +    if (strlen(path) >= sizeof(addr.sun_path)) {
> > +        errno = EINVAL;
> > +        goto out;
> > +    }
> > +
> > +    strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
> > +
> > +    listener = socket(domain, type, protocol);
> > +    if (listener == -1) {
> > +        goto out;
> > +    }
> > +
> > +    if (DeleteFile(path) == 0 && GetLastError() != ERROR_FILE_NOT_FOUND) {
> > +        errno = EACCES;
> > +        goto out;
> > +    }
> > +    g_clear_pointer(&path, g_free);
> > +
> > +    if (bind(listener, (struct sockaddr *)&addr, socklen) == -1) {
> > +        goto out;
> > +    }
> > +
> > +    if (listen(listener, 1) == -1) {
> > +        goto out;
> > +    }
> > +
> > +    client = socket(domain, type, protocol);
> > +    if (client == -1) {
> > +        goto out;
> > +    }
> > +
> > +    arg = 1;
> > +    if (ioctlsocket(client, FIONBIO, &arg) != NO_ERROR) {
> > +        goto out;
> > +    }
> > +
> > +    if (connect(client, (struct sockaddr *)&addr, socklen) == -1 &&
> > +        WSAGetLastError() != WSAEWOULDBLOCK) {
> > +        goto out;
> > +    }
> > +
> > +    server = accept(listener, NULL, NULL);
> > +    if (server == -1) {
> > +        goto out;
> > +    }
>
> In theory at this point 'client' if connect() returned WSAEWOULDBLOCK,
> then at this point it should be fully connected. I wonder if that is
> actually guaranteed though, or should we do something to validate
> there's no race condition ?
>
> > +
> > +    arg = 0;
> > +    if (ioctlsocket(client, FIONBIO, &arg) != NO_ERROR) {
> > +        goto out;
> > +    }
> > +
> > +    arg = 0;
> > +    if (ioctlsocket(client, SIO_AF_UNIX_GETPEERPID, &arg) != NO_ERROR) {
> > +        goto out;
> > +    }
>
> Maybe this will force a synchronization point ?

yeah, I guess switching back to sync and getting the peer pid,

I assume the unix socket pair to be ready at this point.

>
> Alteratively select() + getsockopt(SO_ERROR) is what we used to
> do to check for connect() completion (logic removed now but can
> be seen in b2587932582333197c88bf663785b19f441989d7)
>

That's hopefully not necessary.

thanks

>
>
> > +
> > +    if (arg != GetCurrentProcessId()) {
> > +        errno = EPERM;
> > +        goto out;
> > +    }
> > +
> > +    sv[0] = server;
> > +    server = -1;
> > +    sv[1] = client;
> > +    client = -1;
> > +    ret = 0;
> > +
> > +out:
> > +    if (listener != -1) {
> > +        close(listener);
> > +    }
> > +    if (client != -1) {
> > +        close(client);
> > +    }
> > +    if (server != -1) {
> > +        close(server);
> > +    }
> > +    if (path) {
> > +        DeleteFile(path);
> > +    }
> > +    return ret;
> > +}
> > +
> >  #undef connect
> >  int qemu_connect_wrap(int sockfd, const struct sockaddr *addr,
> >                        socklen_t addrlen)
> > --
> > 2.39.2
> >
>
> With regards,
> Daniel
> --
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
>
>


-- 
Marc-André Lureau