netfilter: conntrack: remove offload_pickup sysctl again
authorFlorian Westphal <fw@strlen.de>
Wed, 4 Aug 2021 13:02:15 +0000 (15:02 +0200)
committerPablo Neira Ayuso <pablo@netfilter.org>
Fri, 6 Aug 2021 15:07:41 +0000 (17:07 +0200)
These two sysctls were added because the hardcoded defaults (2 minutes,
tcp, 30 seconds, udp) turned out to be too low for some setups.

They appeared in 5.14-rc1 so it should be fine to remove it again.

Marcelo convinced me that there should be no difference between a flow
that was offloaded vs. a flow that was not wrt. timeout handling.
Thus the default is changed to those for TCP established and UDP stream,
5 days and 120 seconds, respectively.

Marcelo also suggested to account for the timeout value used for the
offloading, this avoids increase beyond the value in the conntrack-sysctl
and will also instantly expire the conntrack entry with altered sysctls.

Example:
   nf_conntrack_udp_timeout_stream=60
   nf_flowtable_udp_timeout=60

This will remove offloaded udp flows after one minute, rather than two.

An earlier version of this patch also cleared the ASSURED bit to
allow nf_conntrack to evict the entry via early_drop (i.e., table full).
However, it looks like we can safely assume that connection timed out
via HW is still in established state, so this isn't needed.

Quoting Oz:
 [..] the hardware sends all packets with a set FIN flags to sw.
 [..] Connections that are aged in hardware are expected to be in the
 established state.

In case it turns out that back-to-sw-path transition can occur for
'dodgy' connections too (e.g., one side disappeared while software-path
would have been in RETRANS timeout), we can adjust this later.

Cc: Oz Shlomo <ozsh@nvidia.com>
Cc: Paul Blakey <paulb@nvidia.com>
Suggested-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Documentation/networking/nf_conntrack-sysctl.rst
include/net/netns/conntrack.h
net/netfilter/nf_conntrack_proto_tcp.c
net/netfilter/nf_conntrack_proto_udp.c
net/netfilter/nf_conntrack_standalone.c
net/netfilter/nf_flow_table_core.c

index d31ed6c..024d784 100644 (file)
@@ -191,19 +191,9 @@ nf_flowtable_tcp_timeout - INTEGER (seconds)
         TCP connections may be offloaded from nf conntrack to nf flow table.
         Once aged, the connection is returned to nf conntrack with tcp pickup timeout.
 
-nf_flowtable_tcp_pickup - INTEGER (seconds)
-        default 120
-
-        TCP connection timeout after being aged from nf flow table offload.
-
 nf_flowtable_udp_timeout - INTEGER (seconds)
         default 30
 
         Control offload timeout for udp connections.
         UDP connections may be offloaded from nf conntrack to nf flow table.
         Once aged, the connection is returned to nf conntrack with udp pickup timeout.
-
-nf_flowtable_udp_pickup - INTEGER (seconds)
-        default 30
-
-        UDP connection timeout after being aged from nf flow table offload.
index 37e5300..fefd38d 100644 (file)
@@ -30,7 +30,6 @@ struct nf_tcp_net {
        u8 tcp_ignore_invalid_rst;
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        unsigned int offload_timeout;
-       unsigned int offload_pickup;
 #endif
 };
 
@@ -44,7 +43,6 @@ struct nf_udp_net {
        unsigned int timeouts[UDP_CT_MAX];
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        unsigned int offload_timeout;
-       unsigned int offload_pickup;
 #endif
 };
 
index 3259416..af5115e 100644 (file)
@@ -1478,7 +1478,6 @@ void nf_conntrack_tcp_init_net(struct net *net)
 
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        tn->offload_timeout = 30 * HZ;
-       tn->offload_pickup = 120 * HZ;
 #endif
 }
 
index 698fee4..f8e3c0d 100644 (file)
@@ -271,7 +271,6 @@ void nf_conntrack_udp_init_net(struct net *net)
 
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        un->offload_timeout = 30 * HZ;
-       un->offload_pickup = 30 * HZ;
 #endif
 }
 
index 214d9f9..e84b499 100644 (file)
@@ -575,7 +575,6 @@ enum nf_ct_sysctl_index {
        NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_UNACK,
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_OFFLOAD,
-       NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_OFFLOAD_PICKUP,
 #endif
        NF_SYSCTL_CT_PROTO_TCP_LOOSE,
        NF_SYSCTL_CT_PROTO_TCP_LIBERAL,
@@ -585,7 +584,6 @@ enum nf_ct_sysctl_index {
        NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_STREAM,
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_OFFLOAD,
-       NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_OFFLOAD_PICKUP,
 #endif
        NF_SYSCTL_CT_PROTO_TIMEOUT_ICMP,
        NF_SYSCTL_CT_PROTO_TIMEOUT_ICMPV6,
@@ -776,12 +774,6 @@ static struct ctl_table nf_ct_sysctl_table[] = {
                .mode           = 0644,
                .proc_handler   = proc_dointvec_jiffies,
        },
-       [NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_OFFLOAD_PICKUP] = {
-               .procname       = "nf_flowtable_tcp_pickup",
-               .maxlen         = sizeof(unsigned int),
-               .mode           = 0644,
-               .proc_handler   = proc_dointvec_jiffies,
-       },
 #endif
        [NF_SYSCTL_CT_PROTO_TCP_LOOSE] = {
                .procname       = "nf_conntrack_tcp_loose",
@@ -832,12 +824,6 @@ static struct ctl_table nf_ct_sysctl_table[] = {
                .mode           = 0644,
                .proc_handler   = proc_dointvec_jiffies,
        },
-       [NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_OFFLOAD_PICKUP] = {
-               .procname       = "nf_flowtable_udp_pickup",
-               .maxlen         = sizeof(unsigned int),
-               .mode           = 0644,
-               .proc_handler   = proc_dointvec_jiffies,
-       },
 #endif
        [NF_SYSCTL_CT_PROTO_TIMEOUT_ICMP] = {
                .procname       = "nf_conntrack_icmp_timeout",
@@ -1018,7 +1004,6 @@ static void nf_conntrack_standalone_init_tcp_sysctl(struct net *net,
 
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        table[NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_OFFLOAD].data = &tn->offload_timeout;
-       table[NF_SYSCTL_CT_PROTO_TIMEOUT_TCP_OFFLOAD_PICKUP].data = &tn->offload_pickup;
 #endif
 
 }
@@ -1111,7 +1096,6 @@ static int nf_conntrack_standalone_init_sysctl(struct net *net)
        table[NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_STREAM].data = &un->timeouts[UDP_CT_REPLIED];
 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
        table[NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_OFFLOAD].data = &un->offload_timeout;
-       table[NF_SYSCTL_CT_PROTO_TIMEOUT_UDP_OFFLOAD_PICKUP].data = &un->offload_pickup;
 #endif
 
        nf_conntrack_standalone_init_tcp_sysctl(net, table);
index 551976e..8788b51 100644 (file)
@@ -183,7 +183,7 @@ static void flow_offload_fixup_ct_timeout(struct nf_conn *ct)
        const struct nf_conntrack_l4proto *l4proto;
        struct net *net = nf_ct_net(ct);
        int l4num = nf_ct_protonum(ct);
-       unsigned int timeout;
+       s32 timeout;
 
        l4proto = nf_ct_l4proto_find(l4num);
        if (!l4proto)
@@ -192,15 +192,20 @@ static void flow_offload_fixup_ct_timeout(struct nf_conn *ct)
        if (l4num == IPPROTO_TCP) {
                struct nf_tcp_net *tn = nf_tcp_pernet(net);
 
-               timeout = tn->offload_pickup;
+               timeout = tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
+               timeout -= tn->offload_timeout;
        } else if (l4num == IPPROTO_UDP) {
                struct nf_udp_net *tn = nf_udp_pernet(net);
 
-               timeout = tn->offload_pickup;
+               timeout = tn->timeouts[UDP_CT_REPLIED];
+               timeout -= tn->offload_timeout;
        } else {
                return;
        }
 
+       if (timeout < 0)
+               timeout = 0;
+
        if (nf_flow_timeout_delta(ct->timeout) > (__s32)timeout)
                ct->timeout = nfct_time_stamp + timeout;
 }