Re: [PATCH v2 12/15] SUNRPC: Add RPC-with-TLS support to xprtsock.c

[Chuck Lever III <chuck.lever@oracle.com>]

> On Jul 18, 2022, at 4:10 PM, Jeff Layton <jlayton@kernel.org> wrote:
> 
> On Mon, 2022-06-06 at 10:51 -0400, Chuck Lever wrote:
>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>> ---
>> include/linux/sunrpc/xprtsock.h |    1 
>> net/sunrpc/xprtsock.c           |  243 ++++++++++++++++++++++++++++++++-------
>> 2 files changed, 201 insertions(+), 43 deletions(-)
>> 
>> diff --git a/include/linux/sunrpc/xprtsock.h b/include/linux/sunrpc/xprtsock.h
>> index e0b6009f1f69..eaf3d705f758 100644
>> --- a/include/linux/sunrpc/xprtsock.h
>> +++ b/include/linux/sunrpc/xprtsock.h
>> @@ -57,6 +57,7 @@ struct sock_xprt {
>> 	struct work_struct	error_worker;
>> 	struct work_struct	recv_worker;
>> 	struct mutex		recv_mutex;
>> +	struct completion	handshake_done;
>> 	struct sockaddr_storage	srcaddr;
>> 	unsigned short		srcport;
>> 	int			xprt_err;
>> diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
>> index a4fee00412d4..63fe97ede573 100644
>> --- a/net/sunrpc/xprtsock.c
>> +++ b/net/sunrpc/xprtsock.c
>> @@ -48,6 +48,7 @@
>> #include <net/udp.h>
>> #include <net/tcp.h>
>> #include <net/tls.h>
>> +#include <net/tlsh.h>
>> 
>> #include <linux/bvec.h>
>> #include <linux/highmem.h>
>> @@ -197,6 +198,11 @@ static struct ctl_table sunrpc_table[] = {
>>  */
>> #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
>> 
>> +/*
>> + * TLS handshake timeout.
>> + */
>> +#define XS_TLS_HANDSHAKE_TO    (20U * HZ)
>> +
>> #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
>> # undef  RPC_DEBUG_DATA
>> # define RPCDBG_FACILITY	RPCDBG_TRANS
>> @@ -1254,6 +1260,8 @@ static void xs_reset_transport(struct sock_xprt *transport)
>> 	if (atomic_read(&transport->xprt.swapper))
>> 		sk_clear_memalloc(sk);
>> 
>> +	/* TODO: Maybe send a TLS Closure alert */
>> +
>> 	kernel_sock_shutdown(sock, SHUT_RDWR);
>> 
>> 	mutex_lock(&transport->recv_mutex);
>> @@ -2424,6 +2432,147 @@ static void xs_tcp_setup_socket(struct work_struct *work)
>> 
>> #if IS_ENABLED(CONFIG_TLS)
>> 
>> +/**
>> + * xs_tls_handshake_done - TLS handshake completion handler
>> + * @data: address of xprt to wake
>> + * @status: status of handshake
>> + *
>> + */
>> +static void xs_tls_handshake_done(void *data, int status)
>> +{
>> +	struct rpc_xprt *xprt = data;
>> +	struct sock_xprt *transport =
>> +				container_of(xprt, struct sock_xprt, xprt);
>> +
>> +	transport->xprt_err = status ? -EACCES : 0;
>> +	complete(&transport->handshake_done);
>> +	xprt_put(xprt);
>> +}
>> +
>> +static int xs_tls_handshake_sync(struct rpc_xprt *xprt, unsigned int xprtsec)
>> +{
>> +	struct sock_xprt *transport =
>> +				container_of(xprt, struct sock_xprt, xprt);
>> +	int rc;
>> +
>> +	init_completion(&transport->handshake_done);
>> +	set_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state);
>> +
>> +	transport->xprt_err = -ETIMEDOUT;
>> +	switch (xprtsec) {
>> +	case RPC_XPRTSEC_TLS_X509:
>> +		rc = tls_client_hello_x509(transport->sock,
>> +					   xs_tls_handshake_done, xprt_get(xprt),
>> +					   TLSH_DEFAULT_PRIORITIES, TLSH_NO_PEERID,
>> +					   TLSH_NO_CERT);
>> +		break;
>> +	case RPC_XPRTSEC_TLS_PSK:
>> +		rc = tls_client_hello_psk(transport->sock, xs_tls_handshake_done,
>> +					  xprt_get(xprt), TLSH_DEFAULT_PRIORITIES,
>> +					  TLSH_NO_PEERID);
>> +		break;
>> +	default:
>> +		rc = -EACCES;
>> +	}
>> +	if (rc)
>> +		goto out;
>> +
>> +	rc = wait_for_completion_interruptible_timeout(&transport->handshake_done,
>> +						       XS_TLS_HANDSHAKE_TO);
> 
> Should this be interruptible or killable? I'm not sure we want to give
> up on a non-fatal signal, do we? (e.g. SIGCHLD).
> 
> Actually...it looks like this function always runs in workqueue context,
> so this should probably just be wait_for_completion...or better yet,
> consider doing this asynchronously so we don't block a workqueue thread.

This code is similar to rpcrdma_create_id(), which IIRC also runs in
a workqueue thread and has been this way for many years.

If one is wrong, the other is too. But I see recently merged code (in
drivers/nvme/rdma/host.c) that does exactly the same.

<shrug> I'm not married to the approach used in the patch, but it
doesn't seem strange to me.


>> +	if (rc < 0)
>> +		goto out;
>> +
>> +	rc = transport->xprt_err;
>> +
>> +out:
>> +	xs_stream_reset_connect(transport);
>> +	clear_bit(XPRT_SOCK_IGNORE_RECV, &transport->sock_state);
>> +	return rc;
>> +}
>> +
>> +/*
>> + * Transfer the connected socket to @dst_transport, then mark that
>> + * xprt CONNECTED.
>> + */
>> +static int xs_tls_finish_connecting(struct rpc_xprt *src_xprt,
>> +				    struct sock_xprt *dst_transport)
>> +{
>> +	struct sock_xprt *src_transport =
>> +			container_of(src_xprt, struct sock_xprt, xprt);
>> +	struct rpc_xprt *dst_xprt = &dst_transport->xprt;
>> +
>> +	if (!dst_transport->inet) {
>> +		struct socket *sock = src_transport->sock;
>> +		struct sock *sk = sock->sk;
>> +
>> +		/* Avoid temporary address, they are bad for long-lived
>> +		 * connections such as NFS mounts.
>> +		 * RFC4941, section 3.6 suggests that:
>> +		 *    Individual applications, which have specific
>> +		 *    knowledge about the normal duration of connections,
>> +		 *    MAY override this as appropriate.
>> +		 */
>> +		if (xs_addr(dst_xprt)->sa_family == PF_INET6) {
>> +			ip6_sock_set_addr_preferences(sk,
>> +				IPV6_PREFER_SRC_PUBLIC);
>> +		}
>> +
>> +		xs_tcp_set_socket_timeouts(dst_xprt, sock);
>> +		tcp_sock_set_nodelay(sk);
>> +
>> +		lock_sock(sk);
>> +
>> +		/*
>> +		 * @sk is already connected, so it now has the RPC callbacks.
>> +		 * Reach into @src_transport to save the original ones.
>> +		 */
>> +		dst_transport->old_data_ready = src_transport->old_data_ready;
>> +		dst_transport->old_state_change = src_transport->old_state_change;
>> +		dst_transport->old_write_space = src_transport->old_write_space;
>> +		dst_transport->old_error_report = src_transport->old_error_report;
>> +		sk->sk_user_data = dst_xprt;
>> +
>> +		/* socket options */
>> +		sock_reset_flag(sk, SOCK_LINGER);
>> +
>> +		xprt_clear_connected(dst_xprt);
>> +
>> +		dst_transport->sock = sock;
>> +		dst_transport->inet = sk;
>> +		dst_transport->file = src_transport->file;
>> +
>> +		release_sock(sk);
>> +
>> +		/* Reset src_transport before shutting down its clnt */
>> +		mutex_lock(&src_transport->recv_mutex);
>> +		src_transport->inet = NULL;
>> +		src_transport->sock = NULL;
>> +		src_transport->file = NULL;
>> +
>> +		xprt_clear_connected(src_xprt);
>> +		xs_sock_reset_connection_flags(src_xprt);
>> +		xs_stream_reset_connect(src_transport);
>> +		mutex_unlock(&src_transport->recv_mutex);
>> +	}
>> +
>> +	if (!xprt_bound(dst_xprt))
>> +		return -ENOTCONN;
>> +
>> +	xs_set_memalloc(dst_xprt);
>> +
>> +	if (!xprt_test_and_set_connected(dst_xprt)) {
>> +		dst_xprt->connect_cookie++;
>> +		clear_bit(XPRT_SOCK_CONNECTING, &dst_transport->sock_state);
>> +		xprt_clear_connecting(dst_xprt);
>> +
>> +		dst_xprt->stat.connect_count++;
>> +		dst_xprt->stat.connect_time += (long)jiffies -
>> +					   dst_xprt->stat.connect_start;
>> +		xs_run_error_worker(dst_transport, XPRT_SOCK_WAKE_PENDING);
>> +	}
>> +	return 0;
>> +}
>> +
>> /**
>>  * xs_tls_connect - establish a TLS session on a socket
>>  * @work: queued work item
>> @@ -2433,61 +2582,70 @@ static void xs_tls_connect(struct work_struct *work)
>> {
>> 	struct sock_xprt *transport =
>> 		container_of(work, struct sock_xprt, connect_worker.work);
>> +	struct rpc_create_args args = {
>> +		.net		= transport->xprt.xprt_net,
>> +		.protocol	= transport->xprt.prot,
>> +		.address	= (struct sockaddr *)&transport->xprt.addr,
>> +		.addrsize	= transport->xprt.addrlen,
>> +		.timeout	= transport->xprtsec_clnt->cl_timeout,
>> +		.servername	= transport->xprt.servername,
>> +		.nodename	= transport->xprtsec_clnt->cl_nodename,
>> +		.program	= transport->xprtsec_clnt->cl_program,
>> +		.prognumber	= transport->xprtsec_clnt->cl_prog,
>> +		.version	= transport->xprtsec_clnt->cl_vers,
>> +		.authflavor	= RPC_AUTH_TLS,
>> +		.cred		= transport->xprtsec_clnt->cl_cred,
>> +		.xprtsec	= RPC_XPRTSEC_NONE,
>> +	};
>> +	unsigned int pflags = current->flags;
>> 	struct rpc_clnt *clnt;
>> +	struct rpc_xprt *xprt;
>> +	int status;
>> 
>> -	clnt = transport->xprtsec_clnt;
>> -	transport->xprtsec_clnt = NULL;
>> +	if (atomic_read(&transport->xprt.swapper))
>> +		current->flags |= PF_MEMALLOC;
>> +
>> +	xs_stream_start_connect(transport);
>> +
>> +	clnt = rpc_create(&args);
>> 	if (IS_ERR(clnt))
>> 		goto out_unlock;
>> +	rcu_read_lock();
>> +	xprt = xprt_get(rcu_dereference(clnt->cl_xprt));
>> +	rcu_read_unlock();
>> 
>> -	xs_tcp_setup_socket(work);
>> +	status = xs_tls_handshake_sync(xprt, transport->xprt.xprtsec);
>> +	if (status)
>> +		goto out_close;
>> 
>> +	status = xs_tls_finish_connecting(xprt, transport);
>> +	if (status)
>> +		goto out_close;
>> +	trace_rpc_socket_connect(&transport->xprt, transport->sock, 0);
>> +
>> +	xprt_put(xprt);
>> 	rpc_shutdown_client(clnt);
>> 
>> out_unlock:
>> +	xprt_unlock_connect(&transport->xprt, transport);
>> +	current_restore_flags(pflags, PF_MEMALLOC);
>> +	transport->xprtsec_clnt = NULL;
>> 	return;
>> -}
>> 
>> -static void xs_set_xprtsec_clnt(struct rpc_clnt *clnt, struct rpc_xprt *xprt)
>> -{
>> -	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
>> -	struct rpc_create_args args = {
>> -		.net		= xprt->xprt_net,
>> -		.protocol	= xprt->prot,
>> -		.address	= (struct sockaddr *)&xprt->addr,
>> -		.addrsize	= xprt->addrlen,
>> -		.timeout	= clnt->cl_timeout,
>> -		.servername	= xprt->servername,
>> -		.nodename	= clnt->cl_nodename,
>> -		.program	= clnt->cl_program,
>> -		.prognumber	= clnt->cl_prog,
>> -		.version	= clnt->cl_vers,
>> -		.authflavor	= RPC_AUTH_TLS,
>> -		.cred		= clnt->cl_cred,
>> -		.xprtsec	= RPC_XPRTSEC_NONE,
>> -		.flags		= RPC_CLNT_CREATE_NOPING,
>> -	};
>> -
>> -	switch (xprt->xprtsec) {
>> -	case RPC_XPRTSEC_TLS_X509:
>> -	case RPC_XPRTSEC_TLS_PSK:
>> -		transport->xprtsec_clnt = rpc_create(&args);
>> -		break;
>> -	default:
>> -		transport->xprtsec_clnt = ERR_PTR(-ENOTCONN);
>> -	}
>> -}
>> -
>> -#else
>> -
>> -static void xs_set_xprtsec_clnt(struct rpc_clnt *clnt, struct rpc_xprt *xprt)
>> -{
>> -	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
>> +out_close:
>> +	xprt_put(xprt);
>> +	rpc_shutdown_client(clnt);
>> 
>> -	transport->xprtsec_clnt = ERR_PTR(-ENOTCONN);
>> +	/* xprt_force_disconnect() wakes tasks with a fixed tk_status code.
>> +	 * Wake them first here to ensure they get our tk_status code.
>> +	 */
>> +	xprt_wake_pending_tasks(&transport->xprt, status);
>> +	xs_tcp_force_close(&transport->xprt);
>> +	xprt_clear_connecting(&transport->xprt);
>> +	goto out_unlock;
>> }
>> 
>> -#endif /*CONFIG_TLS */
>> +#endif /* CONFIG_TLS */
>> 
>> /**
>>  * xs_connect - connect a socket to a remote endpoint
>> @@ -2520,8 +2678,7 @@ static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
>> 	} else
>> 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
>> 
>> -	xs_set_xprtsec_clnt(task->tk_client, xprt);
>> -
>> +	transport->xprtsec_clnt = task->tk_client;
>> 	queue_delayed_work(xprtiod_workqueue,
>> 			&transport->connect_worker,
>> 			delay);
>> 
>> 
> 
> -- 
> Jeff Layton <jlayton@kernel.org>

--
Chuck Lever