[PATCH 0/3] net/rds: SOL_RDS socket option to explicitly select transport

[PATCH 0/3] net/rds: SOL_RDS socket option to explicitly select transport

[Sowmini Varadhan]

Today the underlying transport (TCP or IB) for a PF_RDS socket is
implicitly selected based on the local address used to bind(2) the
PF_RDS socket. This results in some non-deterministic behavior when
there are un-numbered and IPoIB interfaces sharing the same IP address.
It also places the constraint that the IB interface must have an IP
address (and thus, IPoIB) configured on it.

The non-determinism may be avoided by providing the user-space application
a socket option that allows it to explicitly select the transport
prior to bind(2).

Patch 1 of this series provides the constant definitions needed by
the application via <linux/rds.h>.

Patch 2 provides the setsockopt support, and Patch 3 provides the
getsockopt support.

Sowmini Varadhan (3):
  Declare SO_RDS_TRANSPORT and RDS_TRANS_* constants in
    uapi/linux/rds.h
  Add setsockopt support for SO_RDS_TRANSPORT
  Add setsockopt support for SO_RDS_TRANSPORT

 include/uapi/linux/rds.h |   10 ++++++++++
 net/rds/af_rds.c         |   41 +++++++++++++++++++++++++++++++++++++++++
 net/rds/bind.c           |    4 ++++
 net/rds/rds.h            |    6 +-----
 net/rds/transport.c      |   21 +++++++++++++++++++++
 5 files changed, 77 insertions(+), 5 deletions(-)

[PATCH 1/3] net/rds: Declare SO_RDS_TRANSPORT and RDS_TRANS_* constants in uapi/linux/rds.h

[Sowmini Varadhan]

User space applications that desire to explicitly select the
underlying transport for a PF_RDS socket may do so by using the
SO_RDS_TRANSPORT socket option at the SOL_RDS level before bind().
The integer argument provided to the socket option would be one
of the RDS_TRANS_* values, e.g., RDS_TRANS_TCP. This commit exports
the constant values need by such applications via <linux/rds.h>

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
---
 include/uapi/linux/rds.h |   10 ++++++++++
 net/rds/rds.h            |    5 -----
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/include/uapi/linux/rds.h b/include/uapi/linux/rds.h
index 9195095..0f9265c 100644
--- a/include/uapi/linux/rds.h
+++ b/include/uapi/linux/rds.h
@@ -38,6 +38,8 @@
 
 #define RDS_IB_ABI_VERSION		0x301
 
+#define	SOL_RDS		276
+
 /*
  * setsockopt/getsockopt for SOL_RDS
  */
@@ -48,6 +50,14 @@
 #define RDS_RECVERR			5
 #define RDS_CONG_MONITOR		6
 #define RDS_GET_MR_FOR_DEST		7
+#define SO_RDS_TRANSPORT		8
+
+/* supported values for SO_RDS_TRANSPORT */
+#define	RDS_TRANS_IB	0
+#define	RDS_TRANS_IWARP	1
+#define	RDS_TRANS_TCP	2
+#define RDS_TRANS_COUNT	3
+#define	RDS_TRANS_NONE	(~0)
 
 /*
  * Control message types for SOL_RDS.
diff --git a/net/rds/rds.h b/net/rds/rds.h
index 0d41155..76db508 100644
--- a/net/rds/rds.h
+++ b/net/rds/rds.h
@@ -408,11 +408,6 @@ struct rds_notifier {
  *                 should try hard not to block.
  */
 
-#define RDS_TRANS_IB	0
-#define RDS_TRANS_IWARP	1
-#define RDS_TRANS_TCP	2
-#define RDS_TRANS_COUNT	3
-
 struct rds_transport {
 	char			t_name[TRANSNAMSIZ];
 	struct list_head	t_item;
-- 
1.7.1

[PATCH 2/3] net/rds: Add setsockopt support for SO_RDS_TRANSPORT

[Sowmini Varadhan]

An application may deterministically attach the underlying transport for
a PF_RDS socket by invoking setsockopt(2) with the SO_RDS_TRANSPORT
option at the SOL_RDS level. The integer argument to setsockopt must be
one of the RDS_TRANS_* transport types, e.g., RDS_TRANS_TCP. The option
must be specified before invoking bind(2) on the socket, and may only
be used once on the socket. An attempt to set the option on a bound
socket, or to invoke the option after a successful SO_RDS_TRANSPORT
attachment, will return EOPNOTSUPP.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
---
 net/rds/af_rds.c    |   27 +++++++++++++++++++++++++++
 net/rds/bind.c      |    4 ++++
 net/rds/rds.h       |    1 +
 net/rds/transport.c |   21 +++++++++++++++++++++
 4 files changed, 53 insertions(+), 0 deletions(-)

diff --git a/net/rds/af_rds.c b/net/rds/af_rds.c
index 3d83641..0487744 100644
--- a/net/rds/af_rds.c
+++ b/net/rds/af_rds.c
@@ -270,6 +270,28 @@ static int rds_cong_monitor(struct rds_sock *rs, char __user *optval,
 	return ret;
 }
 
+static int rds_set_transport(struct rds_sock *rs, char __user *optval,
+			     int optlen)
+{
+	int t_type;
+
+	if (rs->rs_transport)
+		return -EOPNOTSUPP; /* previously attached to transport */
+
+	if (optlen != sizeof(int))
+		return -EINVAL;
+
+	if (copy_from_user(&t_type, (int __user *)optval, sizeof(t_type)))
+		return -EFAULT;
+
+	if (t_type < 0 || t_type >= RDS_TRANS_COUNT)
+		return -EINVAL;
+
+	rs->rs_transport = rds_trans_get(t_type);
+
+	return rs->rs_transport ? 0 : -ENOPROTOOPT;
+}
+
 static int rds_setsockopt(struct socket *sock, int level, int optname,
 			  char __user *optval, unsigned int optlen)
 {
@@ -300,6 +322,11 @@ static int rds_setsockopt(struct socket *sock, int level, int optname,
 	case RDS_CONG_MONITOR:
 		ret = rds_cong_monitor(rs, optval, optlen);
 		break;
+	case SO_RDS_TRANSPORT:
+		lock_sock(sock->sk);
+		ret = rds_set_transport(rs, optval, optlen);
+		release_sock(sock->sk);
+		break;
 	default:
 		ret = -ENOPROTOOPT;
 	}
diff --git a/net/rds/bind.c b/net/rds/bind.c
index a2e6562..4ebd29c 100644
--- a/net/rds/bind.c
+++ b/net/rds/bind.c
@@ -181,6 +181,10 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 	if (ret)
 		goto out;
 
+	if (rs->rs_transport) { /* previously bound */
+		ret = 0;
+		goto out;
+	}
 	trans = rds_trans_get_preferred(sin->sin_addr.s_addr);
 	if (!trans) {
 		ret = -EADDRNOTAVAIL;
diff --git a/net/rds/rds.h b/net/rds/rds.h
index 76db508..a33fb4a 100644
--- a/net/rds/rds.h
+++ b/net/rds/rds.h
@@ -798,6 +798,7 @@ struct rds_transport *rds_trans_get_preferred(__be32 addr);
 void rds_trans_put(struct rds_transport *trans);
 unsigned int rds_trans_stats_info_copy(struct rds_info_iterator *iter,
 				       unsigned int avail);
+struct rds_transport *rds_trans_get(int t_type);
 int rds_trans_init(void);
 void rds_trans_exit(void);
 
diff --git a/net/rds/transport.c b/net/rds/transport.c
index 7f2ac4f..8b4a6cd 100644
--- a/net/rds/transport.c
+++ b/net/rds/transport.c
@@ -101,6 +101,27 @@ struct rds_transport *rds_trans_get_preferred(__be32 addr)
 	return ret;
 }
 
+struct rds_transport *rds_trans_get(int t_type)
+{
+	struct rds_transport *ret = NULL;
+	struct rds_transport *trans;
+	unsigned int i;
+
+	down_read(&rds_trans_sem);
+	for (i = 0; i < RDS_TRANS_COUNT; i++) {
+		trans = transports[i];
+
+		if (trans && trans->t_type == t_type &&
+		    (!trans->t_owner || try_module_get(trans->t_owner))) {
+			ret = trans;
+			break;
+		}
+	}
+	up_read(&rds_trans_sem);
+
+	return ret;
+}
+
 /*
  * This returns the number of stats entries in the snapshot and only
  * copies them using the iter if there is enough space for them.  The
-- 
1.7.1

[PATCH 3/3] net/rds Add getsockopt support for SO_RDS_TRANSPORT

[Sowmini Varadhan]

The currently attached transport for a PF_RDS socket may be obtained
from user space by invoking getsockopt(2) using the SO_RDS_TRANSPORT
option at the SOL_RDS level. The integer optval returned will be one
of the RDS_TRANS_* constants defined in linux/rds.h.


Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
---
 net/rds/af_rds.c |   14 ++++++++++++++
 1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/net/rds/af_rds.c b/net/rds/af_rds.c
index 0487744..2ad9032 100644
--- a/net/rds/af_rds.c
+++ b/net/rds/af_rds.c
@@ -339,6 +339,7 @@ static int rds_getsockopt(struct socket *sock, int level, int optname,
 {
 	struct rds_sock *rs = rds_sk_to_rs(sock->sk);
 	int ret = -ENOPROTOOPT, len;
+	int trans;
 
 	if (level != SOL_RDS)
 		goto out;
@@ -364,6 +365,19 @@ static int rds_getsockopt(struct socket *sock, int level, int optname,
 		else
 			ret = 0;
 		break;
+	case SO_RDS_TRANSPORT:
+		if (len < sizeof(int)) {
+			ret = -EINVAL;
+			break;
+		}
+		trans = (rs->rs_transport ? rs->rs_transport->t_type :
+			 RDS_TRANS_NONE); /* unbound */
+		if (put_user(trans, (int __user *)optval) ||
+		    put_user(sizeof(int), optlen))
+			ret = -EFAULT;
+		else
+			ret = 0;
+		break;
 	default:
 		break;
 	}
-- 
1.7.1

Re: [PATCH 0/3] net/rds: SOL_RDS socket option to explicitly select transport

[David Miller]

From: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Date: Fri, 29 May 2015 17:28:06 -0400

> Today the underlying transport (TCP or IB) for a PF_RDS socket is
> implicitly selected based on the local address used to bind(2) the
> PF_RDS socket. This results in some non-deterministic behavior when
> there are un-numbered and IPoIB interfaces sharing the same IP address.
> It also places the constraint that the IB interface must have an IP
> address (and thus, IPoIB) configured on it.
> 
> The non-determinism may be avoided by providing the user-space application
> a socket option that allows it to explicitly select the transport
> prior to bind(2).
> 
> Patch 1 of this series provides the constant definitions needed by
> the application via <linux/rds.h>.
> 
> Patch 2 provides the setsockopt support, and Patch 3 provides the
> getsockopt support.

Seems reasonable, series applied to net-next, thanks.