Re: [PATCH v2 1/4] hurd: Don't pass FD_CLOEXEC in CMSG_DATA

[Samuel Thibault <samuel.thibault@gnu.org>]

Sergey Bugaev, le dim. 23 avril 2023 19:05:45 +0300, a ecrit:
> It is of no concern to the receiving process whether or not the sender
> process wants to close its copy of sent file descriptor upon exec, and
> it should not influence whether or not the received file descriptor gets
> the FD_CLOEXEC flag set in the receiving process.
> 
> The latter should in fact be dependent on the MSG_CMSG_CLOEXEC flag
> being passed to the recvmsg () call, which is going to be implemented
> in the following commit.
> 
> Fixes 344e755248ce02c0f8d095d11cc49e340703d926
> "hurd: Support sending file descriptors over Unix sockets"
> 
> Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
> ---
>  sysdeps/mach/hurd/recvmsg.c | 5 +++--
>  sysdeps/mach/hurd/sendmsg.c | 7 +++++--
>  2 files changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/sysdeps/mach/hurd/recvmsg.c b/sysdeps/mach/hurd/recvmsg.c
> index 39de86f6..c08eb499 100644
> --- a/sysdeps/mach/hurd/recvmsg.c
> +++ b/sysdeps/mach/hurd/recvmsg.c
> @@ -189,7 +189,8 @@ __libc_recvmsg (int fd, struct msghdr *message, int flags)
>      if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS)
>        {
>  	/* SCM_RIGHTS support.  */
> -	/* The fd's flags are passed in the control data.  */
> +	/* The fd's flags are passed in the control data, but there currently
> +	   aren't any defined other than FD_CLOEXEC which we don't respect.  */
>  	int *fds = (int *) CMSG_DATA (cmsg);
>  	nfds = (cmsg->cmsg_len - CMSG_ALIGN (sizeof (struct cmsghdr)))
>  	       / sizeof (int);
> @@ -200,7 +201,7 @@ __libc_recvmsg (int fd, struct msghdr *message, int flags)
>  	    if (err)
>  	      goto cleanup;
>  	    fds[j] = opened_fds[newfds] = _hurd_intern_fd (newports[newfds],
> -							   fds[j], 0);
> +							   0, 0);

The two patches actually make me realize that there was a confusion here
between FD_* flags and O_* flags. _hurd_intern_fd definitely takes O_*
flags (and translates O_CLOEXEC to FD_CLOEXEC). If we are to transport
some flags, it'd probably rather be O_* flags. I'll commit that way.

>  	    if (fds[j] == -1)
>  	      {
>  		err = errno;
> diff --git a/sysdeps/mach/hurd/sendmsg.c b/sysdeps/mach/hurd/sendmsg.c
> index 5871d1d8..2f19797b 100644
> --- a/sysdeps/mach/hurd/sendmsg.c
> +++ b/sysdeps/mach/hurd/sendmsg.c
> @@ -138,8 +138,11 @@ __libc_sendmsg (int fd, const struct msghdr *message, int flags)
>  					     0, 0, 0, 0);
>  		   if (! err)
>  		     nports++;
> -		   /* We pass the flags in the control data.  */
> -		   fds[i] = descriptor->flags;
> +		   /* We pass the file descriptor flags in the control data.
> +		      The only currently defined flag is FD_CLOEXEC, which we
> +		      don't want to pass and so we mask it out, so this will
> +		      always be 0 currently.  */
> +		   fds[i] = descriptor->flags & ~FD_CLOEXEC;
>  		   err;
>  		 }));
>  
> -- 
> 2.40.0
> 

-- 
Samuel
---
Pour une évaluation indépendante, transparente et rigoureuse !
Je soutiens la Commission d'Évaluation de l'Inria.