[PATCH net] net/sched: cls_flower: Return invalid for unknown ct_state flags rules

From: wenxu <redacted>

Reject the unknown ct_state flags of cls flower rules. This also make
the userspace like ovs to probe the ct_state flags support in the
kernel.

Signed-off-by: wenxu <redacted>
---
 include/uapi/linux/pkt_cls.h | 9 +++++++++
 net/sched/cls_flower.c       | 4 ++++
 2 files changed, 13 insertions(+)

diff --git a/include/uapi/linux/pkt_cls.h b/include/uapi/linux/pkt_cls.h
index ee95f42..1ac8ff8 100644
--- a/include/uapi/linux/pkt_cls.h
+++ b/include/uapi/linux/pkt_cls.h

@@ -591,8 +591,17 @@ enum {
 	TCA_FLOWER_KEY_CT_FLAGS_ESTABLISHED = 1 << 1, /* Part of an existing connection. */
 	TCA_FLOWER_KEY_CT_FLAGS_RELATED = 1 << 2, /* Related to an established connection. */
 	TCA_FLOWER_KEY_CT_FLAGS_TRACKED = 1 << 3, /* Conntrack has occurred. */
+
+	__TCA_FLOWER_KEY_CT_FLAGS_MAX,
 };
 
+#define TCA_FLOWER_KEY_CT_FLAGS_MAX \
+		((__TCA_FLOWER_KEY_CT_FLAGS_MAX - 1) << 1)
+#define TCA_FLOWER_KEY_CT_FLAGS_MASK \
+		(TCA_FLOWER_KEY_CT_FLAGS_MAX - 1)
+#define TCA_FLOWER_KEY_CT_FLAGS_UNKNOWN(v) \
+		((v) & (~TCA_FLOWER_KEY_CT_FLAGS_MASK))
+
 enum {
 	TCA_FLOWER_KEY_ENC_OPTS_UNSPEC,
 	TCA_FLOWER_KEY_ENC_OPTS_GENEVE, /* Nested

diff --git a/net/sched/cls_flower.c b/net/sched/cls_flower.c
index 84f9325..ebee70f 100644
--- a/net/sched/cls_flower.c
+++ b/net/sched/cls_flower.c

@@ -1403,6 +1403,10 @@ static int fl_set_key_ct(struct nlattr **tb,
 		fl_set_key_val(tb, &key->ct_state, TCA_FLOWER_KEY_CT_STATE,
 			       &mask->ct_state, TCA_FLOWER_KEY_CT_STATE_MASK,
 			       sizeof(key->ct_state));
+		if (TCA_FLOWER_KEY_CT_FLAGS_UNKNOWN(mask->ct_state)) {
+			NL_SET_ERR_MSG(extack, "invalid ct_state flags");
+			return -EINVAL;
+		}
 	}
 	if (tb[TCA_FLOWER_KEY_CT_ZONE]) {
 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {

-- 
1.8.3.1

Hi,

On Thu, Feb 04, 2021 at 12:17:24PM +0800, wenxu@ucloud.cn wrote:

From: wenxu <redacted>

Reject the unknown ct_state flags of cls flower rules. This also make
the userspace like ovs to probe the ct_state flags support in the
kernel.

That's a good start but it could also do some combination sanity
checks, like ovs does in validate_ct_state(). For example, it does:

      if (state && !(state & CS_TRACKED)) {
          ds_put_format(ds, "%s: invalid connection state: "
                        "If \"trk\" is unset, no other flags are set\n",

...

cls_flower is inconsistent on this but please use NL_SET_ERR_MSG_MOD instead.

+			return -EINVAL;
+		}
 	}
 	if (tb[TCA_FLOWER_KEY_CT_ZONE]) {
 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
-- 
1.8.3.1

在 2021/2/4 21:38, Marcelo Ricardo Leitner 写道:

Hi,

On Thu, Feb 04, 2021 at 12:17:24PM +0800, wenxu@ucloud.cn wrote:

quoted

From: wenxu <redacted>

Reject the unknown ct_state flags of cls flower rules. This also make
the userspace like ovs to probe the ct_state flags support in the
kernel.

That's a good start but it could also do some combination sanity
checks, like ovs does in validate_ct_state(). For example, it does:

      if (state && !(state & CS_TRACKED)) {
          ds_put_format(ds, "%s: invalid connection state: "
                        "If \"trk\" is unset, no other flags are set\n",

So this sanity checks maybe also need to be added in the ovs kernel modules?

The kernel datapath can work without ovs-vswitchd.

On Thu, Feb 04, 2021 at 11:50:53PM +0800, wenxu wrote:

在 2021/2/4 21:38, Marcelo Ricardo Leitner 写道:

quoted

Hi,

On Thu, Feb 04, 2021 at 12:17:24PM +0800, wenxu@ucloud.cn wrote:

quoted

From: wenxu <redacted>

Reject the unknown ct_state flags of cls flower rules. This also make
the userspace like ovs to probe the ct_state flags support in the
kernel.

That's a good start but it could also do some combination sanity
checks, like ovs does in validate_ct_state(). For example, it does:

      if (state && !(state & CS_TRACKED)) {
          ds_put_format(ds, "%s: invalid connection state: "
                        "If \"trk\" is unset, no other flags are set\n",

So this sanity checks maybe also need to be added in the ovs kernel modules?

The kernel datapath can work without ovs-vswitchd.

IMHO that would be nice, yes.

  Marcelo