Re: [Qemu-devel] [PATCH] linux-user: Implement sendmmsg syscall

[Peter Maydell]

On 5 September 2013 14:37, Alexander Graf <address@hidden> wrote:
> Glibc when built for newer kernels assumes that the sendmmsg syscall is 
> available.
> Without it, dns resolution simply fails to work.
>
> Wrap the syscall with existing infrastructure so that we don't have a host 
> dependency
> on sendmmsg.
>
> Signed-off-by: Alexander Graf <address@hidden>
> ---
>  linux-user/syscall.c      | 29 +++++++++++++++++++++++++++++
>  linux-user/syscall_defs.h |  4 ++++
>  2 files changed, 33 insertions(+)
>
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 0272990..59cfdf4 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -1863,6 +1863,30 @@ out2:
>      return ret;
>  }
>
> +#ifdef TARGET_NR_sendmmsg
> +static abi_long do_sendmmsg(int fd, abi_ulong target_msgvec,
> +                            unsigned int vlen, unsigned int flags)
> +{
> +    struct target_mmsghdr *mmsgp;
> +    abi_ulong arg2 = target_msgvec;
> +    int i;

If we're doing this in terms of sendmsg, we need to
implement the kernel's "if CMSG_COMPAT fail EINVAL"
logic ourselves. Compare
 http://lxr.linux.no/#linux+v3.11/net/socket.c#L2175

> +
> +    if (!(mmsgp = lock_user(VERIFY_WRITE, target_msgvec,
> +                            sizeof(*mmsgp) * vlen, 1))) {
> +        return -TARGET_EFAULT;
> +    }
> +
> +    for (i = 0; i < vlen; i++) {
> +        mmsgp[i].msg_len = tswap32(do_sendrecvmsg(fd, arg2, flags, 1));

If do_sendrecvmsg reports an errno we should stop
processing the array and return it, not just
stuff it into msg_len. Compare:
  http://lxr.linux.no/#linux+v3.11/net/socket.c#L2155

> +        arg2 += sizeof(struct target_mmsghdr);
> +    }
> +
> +    unlock_user(mmsgp, target_msgvec, 0);
> +    /* XXX need to handle nonblocking case too */
> +    return vlen;
> +}
> +#endif

thanks
-- PMM