| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Tue, 17 Apr 2018 17:18:13 +0200
From: Ursula Braun <ubraun@...ux.ibm.com>
To: davem@...emloft.net
Cc: netdev@...r.kernel.org, linux-s390@...r.kernel.org,
schwidefsky@...ibm.com, heiko.carstens@...ibm.com,
raspl@...ux.vnet.ibm.com, ubraun@...ux.vnet.ibm.com
Subject: [PATCH net-next 2/4] net/smc: handle sockopt TCP_NODELAY
From: Ursula Braun <ubraun@...ux.vnet.ibm.com>
TCP sockopts must not interfere with the CLC handshake on the
CLC socket. Therefore, we defer some of them till the CLC
handshake has completed, like resetting TCP_NODELAY.
While touching setsockopt, the TCP_FASTOPEN sockopts are
ignored, since SMC-connection setup is based on the TCP
three-way-handshake.
Signed-off-by: Ursula Braun <ubraun@...ux.vnet.ibm.com>
---
net/smc/af_smc.c | 109 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-
net/smc/smc.h | 4 ++
2 files changed, 111 insertions(+), 2 deletions(-)
diff --git a/net/smc/af_smc.c b/net/smc/af_smc.c
index 5f8046c62d90..96f4d182f998 100644
--- a/net/smc/af_smc.c
+++ b/net/smc/af_smc.c
@@ -377,6 +377,22 @@ static void smc_link_save_peer_info(struct smc_link *link,
link->peer_mtu = clc->qp_mtu;
}
+/* deferred setsockopt's not desired during clc handshake */
+static void smc_apply_deferred_sockopts(struct smc_sock *smc)
+{
+ struct smc_sock *opt_smc = smc;
+ u8 val;
+
+ if (smc->listen_smc)
+ opt_smc = smc->listen_smc;
+ if (opt_smc->deferred_nodelay_reset) {
+ val = 0;
+ kernel_setsockopt(smc->clcsock, SOL_TCP, TCP_NODELAY, &val,
+ sizeof(val));
+ opt_smc->deferred_nodelay_reset = 0;
+ }
+}
+
/* setup for RDMA connection of client */
static int smc_connect_rdma(struct smc_sock *smc)
{
@@ -506,6 +522,7 @@ static int smc_connect_rdma(struct smc_sock *smc)
smc_tx_init(smc);
out_connected:
+ smc_apply_deferred_sockopts(smc);
smc_copy_sock_settings_to_clc(smc);
if (smc->sk.sk_state == SMC_INIT)
smc->sk.sk_state = SMC_ACTIVE;
@@ -908,6 +925,7 @@ static void smc_listen_work(struct work_struct *work)
mutex_unlock(&smc_create_lgr_pending);
out_connected:
+ smc_apply_deferred_sockopts(new_smc);
sk_refcnt_debug_inc(newsmcsk);
if (newsmcsk->sk_state == SMC_INIT)
newsmcsk->sk_state = SMC_ACTIVE;
@@ -1280,9 +1298,60 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
{
struct sock *sk = sock->sk;
struct smc_sock *smc;
+ int val;
smc = smc_sk(sk);
+ if (smc->use_fallback || level != SOL_TCP)
+ goto clcsock;
+
+ /* level SOL_TCP */
+ switch (optname) {
+ case TCP_CONGESTION:
+ case TCP_ULP:
+ /* sockopts without integer value; do not apply to SMC */
+ goto clcsock;
+ default:
+ break;
+ }
+ if (optlen < sizeof(int))
+ return -EINVAL;
+ if (get_user(val, (int __user *)optval))
+ return -EFAULT;
+
+ lock_sock(sk);
+ switch (optname) {
+ case TCP_NODELAY:
+ if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
+ release_sock(sk);
+ goto clcsock;
+ }
+ /* for the CLC-handshake TCP_NODELAY is desired;
+ * in case of fallback to TCP, a nodelay reset is
+ * triggered afterwards.
+ */
+ if (val)
+ smc->deferred_nodelay_reset = 0;
+ else
+ smc->deferred_nodelay_reset = 1;
+ break;
+ case TCP_FASTOPEN:
+ case TCP_FASTOPEN_CONNECT:
+ case TCP_FASTOPEN_KEY:
+ case TCP_FASTOPEN_NO_COOKIE:
+ /* ignore these options; 3-way handshake shouldn't be
+ * bypassed with SMC
+ */
+ break;
+ default:
+ /* apply option to the CLC socket */
+ release_sock(sk);
+ goto clcsock;
+ }
+ release_sock(sk);
+ return 0;
+
+clcsock:
/* generic setsockopts reaching us here always apply to the
* CLC socket
*/
@@ -1293,10 +1362,41 @@ static int smc_setsockopt(struct socket *sock, int level, int optname,
static int smc_getsockopt(struct socket *sock, int level, int optname,
char __user *optval, int __user *optlen)
{
+ struct sock *sk = sock->sk;
struct smc_sock *smc;
+ int val, len;
- smc = smc_sk(sock->sk);
- /* socket options apply to the CLC socket */
+ smc = smc_sk(sk);
+
+ if (smc->use_fallback || level != SOL_TCP)
+ goto clcsock;
+
+ if (get_user(len, optlen))
+ return -EFAULT;
+ len = min_t(unsigned int, len, sizeof(int));
+ if (len < 0)
+ return -EINVAL;
+
+ /* level SOL_TCP */
+ switch (optname) {
+ case TCP_NODELAY:
+ if (smc->deferred_nodelay_reset)
+ val = 0;
+ else
+ goto clcsock;
+ break;
+ default:
+ goto clcsock;
+ }
+
+ if (put_user(len, optlen))
+ return -EFAULT;
+ if (copy_to_user(optval, &val, len))
+ return -EFAULT;
+ return 0;
+
+clcsock:
+ /* socket options applying to the CLC socket */
return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
optval, optlen);
}
@@ -1387,6 +1487,7 @@ static int smc_create(struct net *net, struct socket *sock, int protocol,
int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
struct smc_sock *smc;
struct sock *sk;
+ u8 val = 1;
int rc;
rc = -ESOCKTNOSUPPORT;
@@ -1412,6 +1513,10 @@ static int smc_create(struct net *net, struct socket *sock, int protocol,
sk_common_release(sk);
goto out;
}
+ /* clc handshake should run with disabled Nagle algorithm */
+ kernel_setsockopt(smc->clcsock, SOL_TCP, TCP_NODELAY, &val,
+ sizeof(val));
+ smc->deferred_nodelay_reset = 1; /* TCP_NODELAY is not the default */
smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
diff --git a/net/smc/smc.h b/net/smc/smc.h
index e4829a2f46ba..6dfc1c90bed2 100644
--- a/net/smc/smc.h
+++ b/net/smc/smc.h
@@ -185,6 +185,10 @@ struct smc_sock { /* smc sock container */
* started, waiting for unsent
* data to be sent
*/
+ u8 deferred_nodelay_reset : 1;
+ /* defer Nagle after CLC
+ * handshake
+ */
};
static inline struct smc_sock *smc_sk(const struct sock *sk)
--
2.13.5
Powered by blists - more mailing lists