2017-06-09 04:38:15

by Haishuang Yan

Subject: [PATCH] netfilter: ctnetlink: move CTA_TIMEOUT case to outside

When cda[CTA_TIMEOUT] is zero, ctnetlink_new_conntrack will
free allocated ct and return, so move it to outside to optimize
this situation.

Signed-off-by: Haishuang Yan <[email protected]>
---
net/netfilter/nf_conntrack_netlink.c | 5 +----
1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/net/netfilter/nf_conntrack_netlink.c b/net/netfilter/nf_conntrack_netlink.c
index a8be9b7..d1e6b1c 100644
--- a/net/netfilter/nf_conntrack_netlink.c
+++ b/net/netfilter/nf_conntrack_netlink.c
@@ -1768,9 +1768,6 @@ static int change_seq_adj(struct nf_ct_seqadj *seq,
if (IS_ERR(ct))
return ERR_PTR(-ENOMEM);

- if (!cda[CTA_TIMEOUT])
- goto err1;
-
ct->timeout = nfct_time_stamp + ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ;

rcu_read_lock();
@@ -1944,7 +1941,7 @@ static int ctnetlink_new_conntrack(struct net *net, struct sock *ctnl,
if (nlh->nlmsg_flags & NLM_F_CREATE) {
enum ip_conntrack_events events;

- if (!cda[CTA_TUPLE_ORIG] || !cda[CTA_TUPLE_REPLY])
+ if (!cda[CTA_TUPLE_ORIG] || !cda[CTA_TUPLE_REPLY] || !cda[CTA_TIMEOUT])
return -EINVAL;
if (otuple.dst.protonum != rtuple.dst.protonum)
return -EINVAL;
--
1.8.3.1

2017-06-29 16:51:32

by Pablo Neira Ayuso

Subject: Re: [PATCH] netfilter: ctnetlink: move CTA_TIMEOUT case to outside

On Fri, Jun 09, 2017 at 12:37:47PM +0800, Haishuang Yan wrote:
> When cda[CTA_TIMEOUT] is zero, ctnetlink_new_conntrack will
> free allocated ct and return, so move it to outside to optimize
> this situation.
>
> Signed-off-by: Haishuang Yan <[email protected]>
> ---
> net/netfilter/nf_conntrack_netlink.c | 5 +----
> 1 file changed, 1 insertion(+), 4 deletions(-)
>
> diff --git a/net/netfilter/nf_conntrack_netlink.c b/net/netfilter/nf_conntrack_netlink.c
> index a8be9b7..d1e6b1c 100644
> --- a/net/netfilter/nf_conntrack_netlink.c
> +++ b/net/netfilter/nf_conntrack_netlink.c
> @@ -1768,9 +1768,6 @@ static int change_seq_adj(struct nf_ct_seqadj *seq,
> if (IS_ERR(ct))
> return ERR_PTR(-ENOMEM);
>
> - if (!cda[CTA_TIMEOUT])
> - goto err1;

Actually, I think we would make ctnetlink a better place if we just
relax this. I mean, I would like to see how a patch to use the default
timeout based on the protocol state looks like.

ctnetlink is overly pendantic, in asking things that we can probably
infer, just in case the user doesn't specify this.

> -
> ct->timeout = nfct_time_stamp + ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ;
>
> rcu_read_lock();
> @@ -1944,7 +1941,7 @@ static int ctnetlink_new_conntrack(struct net *net, struct sock *ctnl,
> if (nlh->nlmsg_flags & NLM_F_CREATE) {
> enum ip_conntrack_events events;
>
> - if (!cda[CTA_TUPLE_ORIG] || !cda[CTA_TUPLE_REPLY])
> + if (!cda[CTA_TUPLE_ORIG] || !cda[CTA_TUPLE_REPLY] || !cda[CTA_TIMEOUT])
> return -EINVAL;
> if (otuple.dst.protonum != rtuple.dst.protonum)
> return -EINVAL;
> --
> 1.8.3.1
>
>
>