[lxc-devel] [PATCH] RFC: commands: handle epipe

[Dwight Engen]

On Tue, 25 Mar 2014 15:50:06 -0500
Serge Hallyn <serge.hallyn at ubuntu.com> wrote:

> If we start a lxc_wait on a container while it is exiting, it is
> possible that we open the command socket, then the monitor closes
> all its mainloop sockets and exit, then we send our credentials.
> Then we get killed by SIGPIPE.

Hey Serge, this is an interesting condition. I'm a bit confused where
the race is, looking at lxc_wait it looks like we get the monitor open
before opening the command socket with lxc_getstate, and the monitor
shouldn't exit while we are still a client of it.

> Handle that case, recognizing that if we get sigpipe then the
> container is (now) stopped.

This makes sense in general, so I agree with the change.
 
> Signed-off-by: Serge Hallyn <serge.hallyn at ubuntu.com>
> ---
>  src/lxc/af_unix.c  |  4 ++--
>  src/lxc/commands.c | 11 ++++++++++-
>  2 files changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/src/lxc/af_unix.c b/src/lxc/af_unix.c
> index a2de73e..46d8e50 100644
> --- a/src/lxc/af_unix.c
> +++ b/src/lxc/af_unix.c
> @@ -158,7 +158,7 @@ int lxc_abstract_unix_send_fd(int fd, int sendfd,
> void *data, size_t size) msg.msg_iov = &iov;
>  	msg.msg_iovlen = 1;
>  
> -	return sendmsg(fd, &msg, 0);
> +	return sendmsg(fd, &msg, MSG_NOSIGNAL);
>  }
>  
>  int lxc_abstract_unix_recv_fd(int fd, int *recvfd, void *data,
> size_t size) @@ -230,7 +230,7 @@ int
> lxc_abstract_unix_send_credential(int fd, void *data, size_t size)
> msg.msg_iov = &iov; msg.msg_iovlen = 1;
>  
> -	return sendmsg(fd, &msg, 0);
> +	return sendmsg(fd, &msg, MSG_NOSIGNAL);
>  }
>  
>  int lxc_abstract_unix_rcv_credential(int fd, void *data, size_t size)
> diff --git a/src/lxc/commands.c b/src/lxc/commands.c
> index 6b46c2c..b71274c 100644
> --- a/src/lxc/commands.c
> +++ b/src/lxc/commands.c
> @@ -263,6 +263,8 @@ static int lxc_cmd(const char *name, struct
> lxc_cmd_rr *cmd, int *stopped, 
>  	ret = lxc_abstract_unix_send_credential(sock, &cmd->req,
> sizeof(cmd->req)); if (ret != sizeof(cmd->req)) {
> +		if (errno == EPIPE)
> +			goto epipe;
>  		SYSERROR("command %s failed to send req to '@%s' %d",
>  			 lxc_cmd_str(cmd->req.cmd), offset, ret);
>  		if (ret >=0)
> @@ -271,8 +273,10 @@ static int lxc_cmd(const char *name, struct
> lxc_cmd_rr *cmd, int *stopped, }
>  
>  	if (cmd->req.datalen > 0) {
> -		ret = send(sock, cmd->req.data, cmd->req.datalen, 0);
> +		ret = send(sock, cmd->req.data, cmd->req.datalen,
> MSG_NOSIGNAL); if (ret != cmd->req.datalen) {
> +			if (errno == EPIPE)
> +				goto epipe;
>  			SYSERROR("command %s failed to send request
> data to '@%s' %d", lxc_cmd_str(cmd->req.cmd), offset, ret);
>  			if (ret >=0)
> @@ -289,6 +293,11 @@ out:
>  		cmd->rsp.ret = sock;
>  
>  	return ret;
> +
> +epipe:
> +	close(sock);
> +	*stopped = 1;
> +	return 0;
>  }
>  
>  int lxc_try_cmd(const char *name, const char *lxcpath)