[ovs-dev,v2,4/9] ct-dpif, dpif-netlink: Add conntrack timeout policy support
diff mbox series

Message ID 1564697253-37992-5-git-send-email-yihung.wei@gmail.com
State New
Headers show
Series
  • Support zone-based conntrack timeout policy
Related show

Commit Message

Yi-Hung Wei Aug. 1, 2019, 10:07 p.m. UTC
This patch first defines the dpif interface for a datapath to support
adding, deleting, getting and dumping conntrack timeout policy.
The timeout policy is identified by a 4 bytes unsigned integer in
datapath, and it currently support timeout for TCP, UDP, and ICMP
protocols.

Moreover, this patch provides the implemetation for Linux kernel
datapath in dpif-netlink.

In Linux kernel, the timeout policy is maintained per L3/L4 protocol,
and it is identified by 32 bytes null terminated string.  On the other
hand, in vswitchd, the timeout policy is a generic one that consists of
all the supported L4 protocols.  Therefore, one of the main task in
dpif-netlink is to break down the generic timeout policy into 6
sub policies (ipv4 tcp, udp, icmp, and ipv6 tcp, udp, icmp),
and push down the configuration using the netlink API in
netlink-conntrack.c.

This patch also adds missing symbols in the windows datapath so
that the build on windows can pass.

Appveyor CI:
* https://ci.appveyor.com/project/YiHungWei/ovs/builds/26387754

Signed-off-by: Yi-Hung Wei <yihung.wei@gmail.com>
---
 datapath-windows/include/OvsDpInterfaceCtExt.h     | 114 +++++
 datapath-windows/ovsext/Netlink/NetlinkProto.h     |   1 +
 include/windows/automake.mk                        |   1 +
 .../windows/linux/netfilter/nfnetlink_cttimeout.h  |   0
 lib/ct-dpif.c                                      | 104 +++++
 lib/ct-dpif.h                                      |  56 +++
 lib/dpif-netdev.c                                  |   6 +
 lib/dpif-netlink.c                                 | 462 +++++++++++++++++++++
 lib/dpif-netlink.h                                 |   1 +
 lib/dpif-provider.h                                |  38 ++
 lib/netlink-conntrack.c                            | 363 ++++++++++++++++
 lib/netlink-conntrack.h                            |  29 ++
 lib/netlink-protocol.h                             |   1 +
 13 files changed, 1176 insertions(+)
 create mode 100644 include/windows/linux/netfilter/nfnetlink_cttimeout.h

Comments

0-day Robot Aug. 1, 2019, 11:08 p.m. UTC | #1
Bleep bloop.  Greetings Yi-Hung Wei, I am a robot and I have tried out your patch.
Thanks for your contribution.

I encountered some error that I wasn't expecting.  See the details below.


checkpatch:
ERROR: Inappropriate bracing around statement
#543 FILE: lib/dpif-netlink.c:3154:
    if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) {          \

Lines checked: 1401, Warnings: 0, Errors: 1


Please check this out.  If you feel there has been an error, please email aconole@bytheb.org

Thanks,
0-day Robot
Darrell Ball Aug. 6, 2019, 1:23 a.m. UTC | #2
Thanks for the patch

I am going to avoid commenting on style or code conciseness in the
interests of time

On Thu, Aug 1, 2019 at 3:10 PM Yi-Hung Wei <yihung.wei@gmail.com> wrote:

> This patch first defines the dpif interface for a datapath to support
> adding, deleting, getting and dumping conntrack timeout policy.
> The timeout policy is identified by a 4 bytes unsigned integer in
> datapath, and it currently support timeout for TCP, UDP, and ICMP
> protocols.
>

Using a 4 integer to identify a timeout policy is needed to address a Linux
restriction.
However, in general, policies can be identified by generic string or UUIDs.
We add that flexibility later, however.


>
> Moreover, this patch provides the implemetation for Linux kernel
> datapath in dpif-netlink.
>
> In Linux kernel, the timeout policy is maintained per L3/L4 protocol,
> and it is identified by 32 bytes null terminated string.  On the other
> hand, in vswitchd, the timeout policy is a generic one that consists of
> all the supported L4 protocols.  Therefore, one of the main task in
> dpif-netlink is to break down the generic timeout policy into 6
> sub policies (ipv4 tcp, udp, icmp, and ipv6 tcp, udp, icmp),
> and push down the configuration using the netlink API in
> netlink-conntrack.c.
>
> This patch also adds missing symbols in the windows datapath so
> that the build on windows can pass.
>
> Appveyor CI:
> * https://ci.appveyor.com/project/YiHungWei/ovs/builds/26387754
>
> Signed-off-by: Yi-Hung Wei <yihung.wei@gmail.com>
> ---
>  datapath-windows/include/OvsDpInterfaceCtExt.h     | 114 +++++
>  datapath-windows/ovsext/Netlink/NetlinkProto.h     |   1 +
>  include/windows/automake.mk                        |   1 +
>  .../windows/linux/netfilter/nfnetlink_cttimeout.h  |   0
>  lib/ct-dpif.c                                      | 104 +++++
>  lib/ct-dpif.h                                      |  56 +++
>  lib/dpif-netdev.c                                  |   6 +
>  lib/dpif-netlink.c                                 | 462
> +++++++++++++++++++++
>  lib/dpif-netlink.h                                 |   1 +
>  lib/dpif-provider.h                                |  38 ++
>  lib/netlink-conntrack.c                            | 363 ++++++++++++++++
>  lib/netlink-conntrack.h                            |  29 ++
>  lib/netlink-protocol.h                             |   1 +
>  13 files changed, 1176 insertions(+)
>  create mode 100644 include/windows/linux/netfilter/nfnetlink_cttimeout.h
>
> diff --git a/datapath-windows/include/OvsDpInterfaceCtExt.h
> b/datapath-windows/include/OvsDpInterfaceCtExt.h
> index 3b947782e90c..4379855bb8dd 100644
> --- a/datapath-windows/include/OvsDpInterfaceCtExt.h
> +++ b/datapath-windows/include/OvsDpInterfaceCtExt.h
> @@ -421,4 +421,118 @@ struct nf_ct_tcp_flags {
>      UINT8 mask;
>  };
>
> +/* File: nfnetlink_cttimeout.h */
> +enum ctnl_timeout_msg_types {
> +    IPCTNL_MSG_TIMEOUT_NEW,
> +    IPCTNL_MSG_TIMEOUT_GET,
> +    IPCTNL_MSG_TIMEOUT_DELETE,
> +    IPCTNL_MSG_TIMEOUT_DEFAULT_SET,
> +    IPCTNL_MSG_TIMEOUT_DEFAULT_GET,
> +
> +    IPCTNL_MSG_TIMEOUT_MAX
> +};
> +
> +enum ctattr_timeout {
> +    CTA_TIMEOUT_UNSPEC,
> +    CTA_TIMEOUT_NAME,
> +    CTA_TIMEOUT_L3PROTO,
> +    CTA_TIMEOUT_L4PROTO,
> +    CTA_TIMEOUT_DATA,
> +    CTA_TIMEOUT_USE,
> +    __CTA_TIMEOUT_MAX
> +};
> +#define CTA_TIMEOUT_MAX (__CTA_TIMEOUT_MAX - 1)
> +
> +enum ctattr_timeout_generic {
> +    CTA_TIMEOUT_GENERIC_UNSPEC,
> +    CTA_TIMEOUT_GENERIC_TIMEOUT,
> +    __CTA_TIMEOUT_GENERIC_MAX
> +};
> +#define CTA_TIMEOUT_GENERIC_MAX (__CTA_TIMEOUT_GENERIC_MAX - 1)
> +
> +enum ctattr_timeout_tcp {
> +    CTA_TIMEOUT_TCP_UNSPEC,
> +    CTA_TIMEOUT_TCP_SYN_SENT,
> +    CTA_TIMEOUT_TCP_SYN_RECV,
> +    CTA_TIMEOUT_TCP_ESTABLISHED,
> +    CTA_TIMEOUT_TCP_FIN_WAIT,
> +    CTA_TIMEOUT_TCP_CLOSE_WAIT,
> +    CTA_TIMEOUT_TCP_LAST_ACK,
> +    CTA_TIMEOUT_TCP_TIME_WAIT,
> +    CTA_TIMEOUT_TCP_CLOSE,
> +    CTA_TIMEOUT_TCP_SYN_SENT2,
> +    CTA_TIMEOUT_TCP_RETRANS,
> +    CTA_TIMEOUT_TCP_UNACK,
> +    __CTA_TIMEOUT_TCP_MAX
> +};
> +#define CTA_TIMEOUT_TCP_MAX (__CTA_TIMEOUT_TCP_MAX - 1)
> +
> +enum ctattr_timeout_udp {
> +    CTA_TIMEOUT_UDP_UNSPEC,
> +    CTA_TIMEOUT_UDP_UNREPLIED,
> +    CTA_TIMEOUT_UDP_REPLIED,
> +    __CTA_TIMEOUT_UDP_MAX
> +};
> +#define CTA_TIMEOUT_UDP_MAX (__CTA_TIMEOUT_UDP_MAX - 1)
> +
> +enum ctattr_timeout_udplite {
> +    CTA_TIMEOUT_UDPLITE_UNSPEC,
> +    CTA_TIMEOUT_UDPLITE_UNREPLIED,
> +    CTA_TIMEOUT_UDPLITE_REPLIED,
> +    __CTA_TIMEOUT_UDPLITE_MAX
> +};
> +#define CTA_TIMEOUT_UDPLITE_MAX (__CTA_TIMEOUT_UDPLITE_MAX - 1)
> +
> +enum ctattr_timeout_icmp {
> +    CTA_TIMEOUT_ICMP_UNSPEC,
> +    CTA_TIMEOUT_ICMP_TIMEOUT,
> +    __CTA_TIMEOUT_ICMP_MAX
> +};
> +#define CTA_TIMEOUT_ICMP_MAX (__CTA_TIMEOUT_ICMP_MAX - 1)
> +
> +enum ctattr_timeout_dccp {
> +    CTA_TIMEOUT_DCCP_UNSPEC,
> +    CTA_TIMEOUT_DCCP_REQUEST,
> +    CTA_TIMEOUT_DCCP_RESPOND,
> +    CTA_TIMEOUT_DCCP_PARTOPEN,
> +    CTA_TIMEOUT_DCCP_OPEN,
> +    CTA_TIMEOUT_DCCP_CLOSEREQ,
> +    CTA_TIMEOUT_DCCP_CLOSING,
> +    CTA_TIMEOUT_DCCP_TIMEWAIT,
> +    __CTA_TIMEOUT_DCCP_MAX
> +};
> +#define CTA_TIMEOUT_DCCP_MAX (__CTA_TIMEOUT_DCCP_MAX - 1)
> +
> +enum ctattr_timeout_sctp {
> +    CTA_TIMEOUT_SCTP_UNSPEC,
> +    CTA_TIMEOUT_SCTP_CLOSED,
> +    CTA_TIMEOUT_SCTP_COOKIE_WAIT,
> +    CTA_TIMEOUT_SCTP_COOKIE_ECHOED,
> +    CTA_TIMEOUT_SCTP_ESTABLISHED,
> +    CTA_TIMEOUT_SCTP_SHUTDOWN_SENT,
> +    CTA_TIMEOUT_SCTP_SHUTDOWN_RECD,
> +    CTA_TIMEOUT_SCTP_SHUTDOWN_ACK_SENT,
> +    CTA_TIMEOUT_SCTP_HEARTBEAT_SENT,
> +    CTA_TIMEOUT_SCTP_HEARTBEAT_ACKED,
> +    __CTA_TIMEOUT_SCTP_MAX
> +};
> +#define CTA_TIMEOUT_SCTP_MAX (__CTA_TIMEOUT_SCTP_MAX - 1)
> +
> +enum ctattr_timeout_icmpv6 {
> +    CTA_TIMEOUT_ICMPV6_UNSPEC,
> +    CTA_TIMEOUT_ICMPV6_TIMEOUT,
> +    __CTA_TIMEOUT_ICMPV6_MAX
> +};
> +#define CTA_TIMEOUT_ICMPV6_MAX (__CTA_TIMEOUT_ICMPV6_MAX - 1)
> +
> +enum ctattr_timeout_gre {
> +    CTA_TIMEOUT_GRE_UNSPEC,
> +    CTA_TIMEOUT_GRE_UNREPLIED,
> +    CTA_TIMEOUT_GRE_REPLIED,
> +    __CTA_TIMEOUT_GRE_MAX
> +};
> +#define CTA_TIMEOUT_GRE_MAX (__CTA_TIMEOUT_GRE_MAX - 1)
> +
> +#define CTNL_TIMEOUT_NAME_MAX  32
> +
>  #endif /* __OVS_DP_INTERFACE_CT_EXT_H_ */
> diff --git a/datapath-windows/ovsext/Netlink/NetlinkProto.h
> b/datapath-windows/ovsext/Netlink/NetlinkProto.h
> index 59b56565c1dc..db1fa2bacae8 100644
> --- a/datapath-windows/ovsext/Netlink/NetlinkProto.h
> +++ b/datapath-windows/ovsext/Netlink/NetlinkProto.h
> @@ -51,6 +51,7 @@
>  #define NLM_F_ECHO              0x008
>
>  #define NLM_F_ROOT              0x100
> +#define NLM_F_REPLACE           0x100
>

I think a comment is needed to say 'ROOT' is used for 'GET' and 'REPLACE'
is for 'NEW'
because you are adding same id


>  #define NLM_F_MATCH             0x200
>  #define NLM_F_EXCL              0x200
>  #define NLM_F_ATOMIC            0x400
> diff --git a/include/windows/automake.mk b/include/windows/automake.mk
> index 382627b51787..883bbbf5d97c 100644
> --- a/include/windows/automake.mk
> +++ b/include/windows/automake.mk
> @@ -15,6 +15,7 @@ noinst_HEADERS += \
>         include/windows/linux/netfilter/nf_conntrack_tcp.h \
>         include/windows/linux/netfilter/nfnetlink.h \
>         include/windows/linux/netfilter/nfnetlink_conntrack.h \
> +       include/windows/linux/netfilter/nfnetlink_cttimeout.h \
>         include/windows/linux/pkt_sched.h \
>         include/windows/linux/types.h \
>         include/windows/net/if.h \
> diff --git a/include/windows/linux/netfilter/nfnetlink_cttimeout.h
> b/include/windows/linux/netfilter/nfnetlink_cttimeout.h
> new file mode 100644
> index 000000000000..e69de29bb2d1
> diff --git a/lib/ct-dpif.c b/lib/ct-dpif.c
> index 6ea7feb0ee35..7f9ce0a561f7 100644
> --- a/lib/ct-dpif.c
> +++ b/lib/ct-dpif.c
> @@ -760,3 +760,107 @@ ct_dpif_format_zone_limits(uint32_t default_limit,
>          ds_put_format(ds, ",count=%"PRIu32, zone_limit->count);
>      }
>  }
>

Since this is Linux specific, ideally, should be in dpif-netlink
But, it can be purged later, as well.


> +
> +static const char *const ct_dpif_tp_attr_string[] = {
> +#define CT_DPIF_TP_TCP_ATTR(ATTR) \
> +    [CT_DPIF_TP_ATTR_TCP_##ATTR] = "TCP_"#ATTR,
> +    CT_DPIF_TP_TCP_ATTRS
> +#undef CT_DPIF_TP_TCP_ATTR
> +#define CT_DPIF_TP_UDP_ATTR(ATTR) \
> +    [CT_DPIF_TP_ATTR_UDP_##ATTR] = "UDP_"#ATTR,
> +    CT_DPIF_TP_UDP_ATTRS
> +#undef CT_DPIF_TP_UDP_ATTR
> +#define CT_DPIF_TP_ICMP_ATTR(ATTR) \
> +    [CT_DPIF_TP_ATTR_ICMP_##ATTR] = "ICMP_"#ATTR,
> +    CT_DPIF_TP_ICMP_ATTRS
> +#undef CT_DPIF_TP_ICMP_ATTR
> +};
> +
> +static bool
> +ct_dpif_set_timeout_policy_attr(struct ct_dpif_timeout_policy *tp,
> +                                uint32_t attr, uint32_t value)
> +{
> +    if (tp->present & (1 << attr) && tp->attrs[attr] == value) {
> +        return false;
> +    }
> +    tp->attrs[attr] = value;
> +    tp->present |= 1 << attr;
> +    return true;
> +}
> +
> +/* Sets a timeout value identified by '*name' to 'value'.
> + * Returns true if the attribute is changed */
> +bool
> +ct_dpif_set_timeout_policy_attr_by_name(struct ct_dpif_timeout_policy *tp,
> +                                        const char *name, uint32_t value)
> +{
> +    uint32_t i;
> +
> +    for (i = 0; i < CT_DPIF_TP_ATTR_MAX; ++i) {
> +        if (!strcasecmp(name, ct_dpif_tp_attr_string[i])) {
> +            return ct_dpif_set_timeout_policy_attr(tp, i, value);
> +        }
> +    }
> +    return false;
> +}
> +
> +bool
> +ct_dpif_timeout_policy_support_ipproto(uint8_t ipproto)
> +{
> +    if (ipproto == IPPROTO_TCP || ipproto == IPPROTO_UDP ||
> +        ipproto == IPPROTO_ICMP || ipproto == IPPROTO_ICMPV6) {
> +        return true;
> +    }
> +    return false;
> +}
> +
> +int
> +ct_dpif_set_timeout_policy(struct dpif *dpif,
> +                           const struct ct_dpif_timeout_policy *tp)
> +{
> +    return (dpif->dpif_class->ct_set_timeout_policy
> +            ? dpif->dpif_class->ct_set_timeout_policy(dpif, tp)
> +            : EOPNOTSUPP);
> +}
> +
> +int
> +ct_dpif_del_timeout_policy(struct dpif *dpif, uint32_t tp_id)
> +{
> +    return (dpif->dpif_class->ct_del_timeout_policy
> +            ? dpif->dpif_class->ct_del_timeout_policy(dpif, tp_id)
> +            : EOPNOTSUPP);
> +}
> +
> +int
> +ct_dpif_get_timeout_policy(struct dpif *dpif, uint32_t tp_id,
> +                           struct ct_dpif_timeout_policy *tp)
> +{
> +    return (dpif->dpif_class->ct_get_timeout_policy
> +            ? dpif->dpif_class->ct_get_timeout_policy(
> +                dpif, tp_id, tp) : EOPNOTSUPP);
> +}
> +
> +int
> +ct_dpif_timeout_policy_dump_start(struct dpif *dpif, void **statep)
> +{
> +    return (dpif->dpif_class->ct_timeout_policy_dump_start
> +            ? dpif->dpif_class->ct_timeout_policy_dump_start(dpif, statep)
> +            : EOPNOTSUPP);
> +}
> +
> +int
> +ct_dpif_timeout_policy_dump_next(struct dpif *dpif, void *state,
> +                                 struct ct_dpif_timeout_policy *tp)
> +{
> +    return (dpif->dpif_class->ct_timeout_policy_dump_next
> +            ? dpif->dpif_class->ct_timeout_policy_dump_next(dpif, state,
> tp)
> +            : EOPNOTSUPP);
> +}
> +
> +int
> +ct_dpif_timeout_policy_dump_done(struct dpif *dpif, void *state)
> +{
> +    return (dpif->dpif_class->ct_timeout_policy_dump_done
> +            ? dpif->dpif_class->ct_timeout_policy_dump_done(dpif, state)
> +            : EOPNOTSUPP);
> +}
> diff --git a/lib/ct-dpif.h b/lib/ct-dpif.h
> index 2f4906817946..8dacb1c7c253 100644
> --- a/lib/ct-dpif.h
> +++ b/lib/ct-dpif.h
> @@ -225,6 +225,50 @@ struct ct_dpif_zone_limit {
>      struct ovs_list node;
>  };
>
> +#define CT_DPIF_TP_TCP_ATTRS \
> +    CT_DPIF_TP_TCP_ATTR(SYN_SENT) \
> +    CT_DPIF_TP_TCP_ATTR(SYN_RECV) \
> +    CT_DPIF_TP_TCP_ATTR(ESTABLISHED) \
> +    CT_DPIF_TP_TCP_ATTR(FIN_WAIT) \
> +    CT_DPIF_TP_TCP_ATTR(CLOSE_WAIT) \
> +    CT_DPIF_TP_TCP_ATTR(LAST_ACK) \
> +    CT_DPIF_TP_TCP_ATTR(TIME_WAIT) \
> +    CT_DPIF_TP_TCP_ATTR(CLOSE) \
> +    CT_DPIF_TP_TCP_ATTR(SYN_SENT2) \
> +    CT_DPIF_TP_TCP_ATTR(RETRANSMIT) \
> +    CT_DPIF_TP_TCP_ATTR(UNACK)
> +
> +#define CT_DPIF_TP_UDP_ATTRS \
> +    CT_DPIF_TP_UDP_ATTR(FIRST) \
> +    CT_DPIF_TP_UDP_ATTR(SINGLE) \
> +    CT_DPIF_TP_UDP_ATTR(MULTIPLE)
> +
> +#define CT_DPIF_TP_ICMP_ATTRS \
> +    CT_DPIF_TP_ICMP_ATTR(FIRST) \
> +    CT_DPIF_TP_ICMP_ATTR(REPLY)
> +
> +enum OVS_PACKED_ENUM ct_dpif_tp_attr {
> +#define CT_DPIF_TP_TCP_ATTR(ATTR) CT_DPIF_TP_ATTR_TCP_##ATTR,
> +    CT_DPIF_TP_TCP_ATTRS
> +#undef CT_DPIF_TP_TCP_ATTR
> +#define CT_DPIF_TP_UDP_ATTR(ATTR) CT_DPIF_TP_ATTR_UDP_##ATTR,
> +    CT_DPIF_TP_UDP_ATTRS
> +#undef CT_DPIF_TP_UDP_ATTR
> +#define CT_DPIF_TP_ICMP_ATTR(ATTR) CT_DPIF_TP_ATTR_ICMP_##ATTR,
> +    CT_DPIF_TP_ICMP_ATTRS
> +#undef CT_DPIF_TP_ICMP_ATTR
> +    CT_DPIF_TP_ATTR_MAX
> +};
> +
> +struct ct_dpif_timeout_policy {
> +    uint32_t    id;         /* An unique identifier of a timeout policy in
> +                             * the datapath. */
>

This is very restrictive to handle Linux case; can be enhanced later though.
I commented on this elsewhere as well


> +    uint32_t    present;    /* If a timeout attribute is present set the
> +                             * corresponding bit. */
> +    uint32_t    attrs[CT_DPIF_TP_ATTR_MAX];     /* An array that specifies
> +                                                 * timeout attribute
> values */
> +};
> +
>  int ct_dpif_dump_start(struct dpif *, struct ct_dpif_dump_state **,
>                         const uint16_t *zone, int *);
>  int ct_dpif_dump_next(struct ct_dpif_dump_state *, struct ct_dpif_entry
> *);
> @@ -262,5 +306,17 @@ bool ct_dpif_parse_zone_limit_tuple(const char *s,
> uint16_t *pzone,
>                                      uint32_t *plimit, struct ds *);
>  void ct_dpif_format_zone_limits(uint32_t default_limit,
>                                  const struct ovs_list *, struct ds *);
> +bool ct_dpif_set_timeout_policy_attr_by_name(struct
> ct_dpif_timeout_policy *tp,
> +                                             const char *key, uint32_t
> value);
> +bool ct_dpif_timeout_policy_support_ipproto(uint8_t ipproto);
> +int ct_dpif_set_timeout_policy(struct dpif *dpif,
> +                               const struct ct_dpif_timeout_policy *tp);
> +int ct_dpif_get_timeout_policy(struct dpif *dpif, uint32_t tp_id,
> +                               struct ct_dpif_timeout_policy *tp);
> +int ct_dpif_del_timeout_policy(struct dpif *dpif, uint32_t tp_id);
> +int ct_dpif_timeout_policy_dump_start(struct dpif *dpif, void **statep);
> +int ct_dpif_timeout_policy_dump_next(struct dpif *dpif, void *state,
> +                                     struct ct_dpif_timeout_policy *tp);
> +int ct_dpif_timeout_policy_dump_done(struct dpif *dpif, void *state);
>
>  #endif /* CT_DPIF_H */
> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index d0a1c58adace..2079e368fb52 100644
> --- a/lib/dpif-netdev.c
> +++ b/lib/dpif-netdev.c
> @@ -7529,6 +7529,12 @@ const struct dpif_class dpif_netdev_class = {
>      NULL,                       /* ct_set_limits */
>      NULL,                       /* ct_get_limits */
>      NULL,                       /* ct_del_limits */
> +    NULL,                       /* ct_set_timeout_policy */
> +    NULL,                       /* ct_get_timeout_policy */
> +    NULL,                       /* ct_del_timeout_policy */
> +    NULL,                       /* ct_timeout_policy_dump_start */
> +    NULL,                       /* ct_timeout_policy_dump_next */
> +    NULL,                       /* ct_timeout_policy_dump_done */
>      dpif_netdev_ipf_set_enabled,
>      dpif_netdev_ipf_set_min_frag,
>      dpif_netdev_ipf_set_max_nfrags,
> diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
> index 7bc71d6d19d7..b859508f718a 100644
> --- a/lib/dpif-netlink.c
> +++ b/lib/dpif-netlink.c
> @@ -50,6 +50,7 @@
>  #include "odp-util.h"
>  #include "openvswitch/dynamic-string.h"
>  #include "openvswitch/flow.h"
> +#include "openvswitch/hmap.h"
>  #include "openvswitch/match.h"
>  #include "openvswitch/ofpbuf.h"
>  #include "openvswitch/poll-loop.h"
> @@ -3023,6 +3024,461 @@ dpif_netlink_ct_del_limits(struct dpif *dpif
> OVS_UNUSED,
>      ofpbuf_delete(request);
>      return err;
>  }
> +
> +#define NL_TP_NAME_PREFIX "ovs_tp_"
> +
> +struct dpif_netlink_timeout_policy_protocol {
> +    uint16_t    l3num;
> +    uint8_t     l4num;
> +};
> +
> +enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol {
> +    DPIF_NL_TP_AF_INET_TCP,
> +    DPIF_NL_TP_AF_INET_UDP,
> +    DPIF_NL_TP_AF_INET_ICMP,
> +    DPIF_NL_TP_AF_INET6_TCP,
> +    DPIF_NL_TP_AF_INET6_UDP,
> +    DPIF_NL_TP_AF_INET6_ICMPV6,
> +    DPIF_NL_TP_MAX
> +};
> +
> +#define DPIF_NL_ALL_TP 0x3F
> +
> +static struct dpif_netlink_timeout_policy_protocol tp_protos[] = {
> +    [DPIF_NL_TP_AF_INET_TCP] = { .l3num = AF_INET, .l4num = IPPROTO_TCP },
> +    [DPIF_NL_TP_AF_INET_UDP] = { .l3num = AF_INET, .l4num = IPPROTO_UDP },
> +    [DPIF_NL_TP_AF_INET_ICMP] = { .l3num = AF_INET, .l4num = IPPROTO_ICMP
> },
> +    [DPIF_NL_TP_AF_INET6_TCP] = { .l3num = AF_INET6, .l4num = IPPROTO_TCP
> },
> +    [DPIF_NL_TP_AF_INET6_UDP] = { .l3num = AF_INET6, .l4num = IPPROTO_UDP
> },
> +    [DPIF_NL_TP_AF_INET6_ICMPV6] = { .l3num = AF_INET6,
> +                                     .l4num = IPPROTO_ICMPV6 },
> +};
> +
> +static void
> +dpif_netlink_format_tp_name(uint32_t id, uint16_t l3num, uint8_t l4num,
> +                            struct ds *tp_name)
> +{
> +    ds_clear(tp_name);
> +    ds_put_format(tp_name, "%s%"PRIu32"_", NL_TP_NAME_PREFIX, id);
> +    ct_dpif_format_ipproto(tp_name, l4num);
> +
> +    if (l3num == AF_INET) {
> +        ds_put_cstr(tp_name, "4");
> +    } else if (l3num == AF_INET6 && l4num != IPPROTO_ICMPV6) {
>

I did not trace ICMP6 but I'll assume it is correct
I don't really like special casing this, if it can be avoided; I did not
check since it is minor
Needs comment, especially since it looks odd


> +        ds_put_cstr(tp_name, "6");
> +    }
> +
> +    ovs_assert(tp_name->length < CTNL_TIMEOUT_NAME_MAX);
> +}
> +
> +#define CT_DPIF_NL_TP_TCP_MAPPINGS                              \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT, SYN_SENT)         \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_RECV, SYN_RECV)         \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, ESTABLISHED, ESTABLISHED)   \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, FIN_WAIT, FIN_WAIT)         \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE_WAIT, CLOSE_WAIT)     \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, LAST_ACK, LAST_ACK)         \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, TIME_WAIT, TIME_WAIT)       \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE, CLOSE)               \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT2, SYN_SENT2)       \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, RETRANSMIT, RETRANS)        \
> +    CT_DPIF_NL_TP_MAPPING(TCP, TCP, UNACK, UNACK)
> +
> +#define CT_DPIF_NL_TP_UDP_MAPPINGS                              \
> +    CT_DPIF_NL_TP_MAPPING(UDP, UDP, SINGLE, UNREPLIED)          \
> +    CT_DPIF_NL_TP_MAPPING(UDP, UDP, MULTIPLE, REPLIED)
> +
> +#define CT_DPIF_NL_TP_ICMP_MAPPINGS                             \
> +    CT_DPIF_NL_TP_MAPPING(ICMP, ICMP, FIRST, TIMEOUT)
> +
> +#define CT_DPIF_NL_TP_ICMPV6_MAPPINGS                           \
> +    CT_DPIF_NL_TP_MAPPING(ICMP, ICMPV6, FIRST, TIMEOUT)
> +
> +
> +#define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2)     \
> +if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) {  \
> +    nl_tp->present |= 1 << CTA_TIMEOUT_##PROTO2##_##ATTR2;      \
> +    nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2] =              \
> +        tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1];          \
> +}
> +
> +static void
> +dpif_netlink_get_nl_tp_tcp_attrs(const struct ct_dpif_timeout_policy *tp,
> +                                 struct nl_ct_timeout_policy *nl_tp)
> +{
> +    CT_DPIF_NL_TP_TCP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_get_nl_tp_udp_attrs(const struct ct_dpif_timeout_policy *tp,
> +                                 struct nl_ct_timeout_policy *nl_tp)
> +{
> +    CT_DPIF_NL_TP_UDP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_get_nl_tp_icmp_attrs(const struct ct_dpif_timeout_policy *tp,
> +                                  struct nl_ct_timeout_policy *nl_tp)
> +{
> +    CT_DPIF_NL_TP_ICMP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_get_nl_tp_icmpv6_attrs(const struct ct_dpif_timeout_policy
> *tp,
> +                                    struct nl_ct_timeout_policy *nl_tp)
> +{
> +    CT_DPIF_NL_TP_ICMPV6_MAPPINGS
> +}
> +
> +#undef CT_DPIF_NL_TP_MAPPING
> +
> +static void
> +dpif_netlink_get_nl_tp_attrs(const struct ct_dpif_timeout_policy *tp,
> +                             uint8_t l4num, struct nl_ct_timeout_policy
> *nl_tp)
> +{
> +    nl_tp->present = 0;
> +
> +    if (l4num == IPPROTO_TCP) {
> +        dpif_netlink_get_nl_tp_tcp_attrs(tp, nl_tp);
> +    } else if (l4num == IPPROTO_UDP) {
> +        dpif_netlink_get_nl_tp_udp_attrs(tp, nl_tp);
> +    } else if (l4num == IPPROTO_ICMP) {
> +        dpif_netlink_get_nl_tp_icmp_attrs(tp, nl_tp);
> +    } else if (l4num == IPPROTO_ICMPV6) {
> +        dpif_netlink_get_nl_tp_icmpv6_attrs(tp, nl_tp);
> +    }
> +}
> +
>

I think the dual duty of checking and also for setting to confusing
I would break it up into 2
At the very least, needs some comments


> +#define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2)
>    \
> +if (nl_tp->present & (1 << CTA_TIMEOUT_##PROTO2##_##ATTR2)) {
>    \
> +    if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) {
>   \
> +        if (tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] !=
>   \
> +            nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]) {
>    \
> +            VLOG_WARN_RL(&error_rl, "Inconsistent timeout policy %s "
>    \
> +                         "attribute %s=%"PRIu32" while %s=%"PRIu32,
>    \
> +                         nl_tp->name, "CTA_TIMEOUT_"#PROTO2"_"#ATTR2,
>    \
> +                         nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2],
>   \
> +                         "CT_DPIF_TP_ATTR_"#PROTO1"_"#ATTR1,
>   \
> +                         tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]);
>   \
> +        }
>    \
> +    } else {
>   \
> +        tp->present |= 1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1;
>    \
> +        tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] =
>    \
> +            nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2];
>    \
> +    }
>    \
> +}
> +
> +static void
> +dpif_netlink_set_ct_dpif_tp_tcp_attrs(const struct nl_ct_timeout_policy
> *nl_tp,
> +                                      struct ct_dpif_timeout_policy *tp)
> +{
> +    CT_DPIF_NL_TP_TCP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_set_ct_dpif_tp_udp_attrs(const struct nl_ct_timeout_policy
> *nl_tp,
> +                                      struct ct_dpif_timeout_policy *tp)
> +{
> +    CT_DPIF_NL_TP_UDP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_set_ct_dpif_tp_icmp_attrs(
> +    const struct nl_ct_timeout_policy *nl_tp,
> +    struct ct_dpif_timeout_policy *tp)
> +{
> +    CT_DPIF_NL_TP_ICMP_MAPPINGS
> +}
> +
> +static void
> +dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(
> +    const struct nl_ct_timeout_policy *nl_tp,
> +    struct ct_dpif_timeout_policy *tp)
> +{
> +    CT_DPIF_NL_TP_ICMPV6_MAPPINGS
> +}
> +
> +#undef CT_DPIF_NL_TP_MAPPING
> +
> +static void
> +dpif_netlink_set_ct_dpif_tp_attrs(const struct nl_ct_timeout_policy
> *nl_tp,
> +                               struct ct_dpif_timeout_policy *tp)
> +{
> +    if (nl_tp->l4num == IPPROTO_TCP) {
> +        dpif_netlink_set_ct_dpif_tp_tcp_attrs(nl_tp, tp);
> +    } else if (nl_tp->l4num == IPPROTO_UDP) {
> +        dpif_netlink_set_ct_dpif_tp_udp_attrs(nl_tp, tp);
> +    } else if (nl_tp->l4num == IPPROTO_ICMP) {
> +        dpif_netlink_set_ct_dpif_tp_icmp_attrs(nl_tp, tp);
> +    } else if (nl_tp->l4num == IPPROTO_ICMPV6) {
> +        dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(nl_tp, tp);
> +    }
> +}
> +
> +#ifdef _WIN32
> +static int
> +dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   const struct ct_dpif_timeout_policy
> *tp)
> +{
> +    return EOPNOTSUPP;
> +}
> +
> +static int
> +dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   uint32_t tp_id,
> +                                   struct ct_dpif_timeout_policy *tp)
> +{
> +    return EOPNOTSUPP;
> +}
> +
> +static int
> +dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   uint32_t tp_id)
> +{
> +    return EOPNOTSUPP;
> +}
> +
> +static int
> +dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED,
> +                                          void **statep)
> +{
> +    return EOPNOTSUPP;
> +}
> +
> +static int
> +dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
> +                                         void *state,
> +                                         struct ct_dpif_timeout_policy
> **tp)
> +{
> +    return EOPNOTSUPP;
> +}
> +
> +static int
> +dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED,
> +                                         void *state)
> +{
> +    return EOPNOTSUPP;
> +}
> +#else
>

Might be good to add some comments for benefits of others


> +static int
> +dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   const struct ct_dpif_timeout_policy
> *tp)
> +{
> +    struct nl_ct_timeout_policy nl_tp;
> +    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
> +    int i, err = 0;
> +
> +    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
> +        dpif_netlink_format_tp_name(tp->id, tp_protos[i].l3num,
> +                                    tp_protos[i].l4num, &nl_tp_name);
> +        ovs_strlcpy(nl_tp.name, ds_cstr(&nl_tp_name), sizeof nl_tp.name);
> +        nl_tp.l3num = tp_protos[i].l3num;
> +        nl_tp.l4num = tp_protos[i].l4num;
> +        dpif_netlink_get_nl_tp_attrs(tp, tp_protos[i].l4num, &nl_tp);
> +        err = nl_ct_set_timeout_policy(&nl_tp);
> +        if (err) {
> +            VLOG_WARN_RL(&error_rl, "failed to add timeout policy %s
> (%s)",
> +                         nl_tp.name, ovs_strerror(err));
> +            goto out;
> +        }
> +    }
> +
> +out:
> +    ds_destroy(&nl_tp_name);
> +    return  err;
> +}
> +
> +static int
> +dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   uint32_t tp_id,
> +                                   struct ct_dpif_timeout_policy *tp)
> +{
> +    struct nl_ct_timeout_policy nl_tp;
> +    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
> +    int i, err = 0;
> +
> +    tp->id = tp_id;
> +    tp->present = 0;
> +    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
> +        dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num,
> +                                    tp_protos[i].l4num, &nl_tp_name);
> +        err = nl_ct_get_timeout_policy(ds_cstr(&nl_tp_name), &nl_tp);
> +
> +        if (err) {
> +            VLOG_WARN_RL(&error_rl, "failed to get timeout policy %s
> (%s)",
> +                         nl_tp.name, ovs_strerror(err));
> +            goto out;
> +        }
> +        dpif_netlink_set_ct_dpif_tp_attrs(&nl_tp, tp);
> +    }
> +
> +out:
> +    ds_destroy(&nl_tp_name);
> +    return  err;
> +}
> +
> +/* Returns 0 if all the sub timeout policies are deleted or
> + * not exist in the kernel. */
> +static int
> +dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   uint32_t tp_id)
> +{
> +    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
> +    int i, err = 0;
> +
> +    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
> +        dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num,
> +                                    tp_protos[i].l4num, &nl_tp_name);
> +        err = nl_ct_del_timeout_policy(ds_cstr(&nl_tp_name));
> +        if (err == ENOENT) {
> +            err = 0;
> +        }
> +        if (err) {
> +            VLOG_WARN_RL(&error_rl, "failed to delete timeout policy %s
> (%s)",
> +                         ds_cstr(&nl_tp_name), ovs_strerror(err));
> +            goto out;
> +        }
> +    }
> +
> +out:
> +    ds_destroy(&nl_tp_name);
> +    return err;
> +}
> +
> +struct dpif_netlink_ct_timeout_policy_dump_state {
> +    struct nl_ct_timeout_policy_dump_state *nl_dump_state;
> +    struct hmap tp_dump_map;
> +};
> +
> +struct dpif_netlink_tp_dump_node {
> +    struct      hmap_node hmap_node;      /* node in tp_dump_map. */
> +    struct      ct_dpif_timeout_policy *tp;
> +    uint32_t    present;
> +};
> +
> +static struct dpif_netlink_tp_dump_node *
> +get_dpif_netlink_tp_dump_node_by_tp_id(uint32_t tp_id,
> +                                       struct hmap *tp_dump_map)
> +{
> +    struct dpif_netlink_tp_dump_node *tp_dump_node;
> +
> +    HMAP_FOR_EACH_WITH_HASH (tp_dump_node, hmap_node, hash_int(tp_id, 0),
> +                            tp_dump_map) {
> +        if (tp_dump_node->tp->id == tp_id) {
> +            return tp_dump_node;
> +        }
> +    }
> +    return NULL;
> +}
> +
> +static void
> +update_dpif_netlink_tp_dump_node(
> +    const struct nl_ct_timeout_policy *nl_tp,
> +    struct dpif_netlink_tp_dump_node *tp_dump_node)
> +{
> +    int i;
> +
> +    dpif_netlink_set_ct_dpif_tp_attrs(nl_tp, tp_dump_node->tp);
> +    for (i = 0; i < DPIF_NL_TP_MAX; ++i) {
> +        if (nl_tp->l3num == tp_protos[i].l3num &&
> +            nl_tp->l4num == tp_protos[i].l4num) {
> +            tp_dump_node->present |= 1 << i;
> +            break;
> +        }
> +    }
> +}
> +
> +static int
> +dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED,
> +                                          void **statep)
> +{
> +    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state;
> +    int err;
> +
> +    *statep = dump_state = xzalloc(sizeof *dump_state);
> +    err = nl_ct_timeout_policy_dump_start(&dump_state->nl_dump_state);
> +    if (err) {
> +        free(dump_state);
> +        return err;
> +    }
> +    hmap_init(&dump_state->tp_dump_map);
> +    return 0;
> +}
> +
>

I think this function could really benefit from comments


> +static int
> +dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
> +                                         void *state,
> +                                         struct ct_dpif_timeout_policy
> *tp)
> +{
> +    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
> +    struct dpif_netlink_tp_dump_node *tp_dump_node;
> +    struct nl_ct_timeout_policy nl_tp;
> +    uint32_t tp_id;
> +    int err;
> +
> +    do {
> +        err =  nl_ct_timeout_policy_dump_next(dump_state->nl_dump_state,
> +                                              &nl_tp);
> +        if (err) {
> +            break;
> +        }
> +
> +        if (!ovs_scan(nl_tp.name, NL_TP_NAME_PREFIX"%"PRIu32, &tp_id)) {
> +            continue;
> +        }
> +
> +        tp_dump_node = get_dpif_netlink_tp_dump_node_by_tp_id(
> +                            tp_id, &dump_state->tp_dump_map);
> +        if (!tp_dump_node) {
> +            tp_dump_node = xzalloc(sizeof *tp_dump_node);
> +            tp_dump_node->tp = xzalloc(sizeof *tp_dump_node->tp);
> +            tp_dump_node->tp->id = tp_id;
> +            hmap_insert(&dump_state->tp_dump_map,
> &tp_dump_node->hmap_node,
> +                        hash_int(tp_id, 0));
> +        }
> +
> +        update_dpif_netlink_tp_dump_node(&nl_tp, tp_dump_node);
> +        if (tp_dump_node->present == DPIF_NL_ALL_TP) {
> +            hmap_remove(&dump_state->tp_dump_map,
> &tp_dump_node->hmap_node);


Can you check above and just not insert it ?


> +            *tp = *tp_dump_node->tp;
> +            free(tp_dump_node->tp);
> +            free(tp_dump_node);
> +            break;
> +        }
> +    } while (true);
> +
> +    /* Dump the incomplete timeout policy. */
> +    if (err == EOF) {
> +        if (!hmap_is_empty(&dump_state->tp_dump_map)) {
> +            struct hmap_node *hmap_node =
> hmap_first(&dump_state->tp_dump_map);
> +
> +            hmap_remove(&dump_state->tp_dump_map, hmap_node);
> +            tp_dump_node = CONTAINER_OF(hmap_node,
> +                            struct dpif_netlink_tp_dump_node, hmap_node);
> +            *tp = *tp_dump_node->tp;
> +            free(tp_dump_node->tp);
> +            free(tp_dump_node);
> +            return 0;
> +        }
> +    }
> +
> +    return err;
> +}
> +
> +static int
> +dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED,
> +                                         void *state)
> +{
> +    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
> +    int err;
> +
> +    err = nl_ct_timeout_policy_dump_done(dump_state->nl_dump_state);
> +    hmap_destroy(&dump_state->tp_dump_map);
> +    free(dump_state);
> +    return err;
> +}
> +#endif
> +
>
>  /* Meters */
>
> @@ -3429,6 +3885,12 @@ const struct dpif_class dpif_netlink_class = {
>      dpif_netlink_ct_set_limits,
>      dpif_netlink_ct_get_limits,
>      dpif_netlink_ct_del_limits,
> +    dpif_netlink_ct_set_timeout_policy,
> +    dpif_netlink_ct_get_timeout_policy,
> +    dpif_netlink_ct_del_timeout_policy,
> +    dpif_netlink_ct_timeout_policy_dump_start,
> +    dpif_netlink_ct_timeout_policy_dump_next,
> +    dpif_netlink_ct_timeout_policy_dump_done,
>      NULL,                       /* ipf_set_enabled */
>      NULL,                       /* ipf_set_min_frag */
>      NULL,                       /* ipf_set_max_nfrags */
> diff --git a/lib/dpif-netlink.h b/lib/dpif-netlink.h
> index 0a9628088275..1e753994d747 100644
> --- a/lib/dpif-netlink.h
> +++ b/lib/dpif-netlink.h
> @@ -23,6 +23,7 @@
>  #include "odp-netlink.h"
>
>  #include "flow.h"
> +#include "netlink-conntrack.h"
>
>  struct ofpbuf;
>
> diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h
> index 12898b9e3c6d..79a2314500cf 100644
> --- a/lib/dpif-provider.h
> +++ b/lib/dpif-provider.h
> @@ -80,6 +80,7 @@ dpif_flow_dump_thread_init(struct dpif_flow_dump_thread
> *thread,
>  struct ct_dpif_dump_state;
>  struct ct_dpif_entry;
>  struct ct_dpif_tuple;
> +struct ct_dpif_timeout_policy;
>
>  /* 'dpif_ipf_proto_status' and 'dpif_ipf_status' are presently in
>   * sync with 'ipf_proto_status' and 'ipf_status', but more
> @@ -498,6 +499,43 @@ struct dpif_class {
>       * list of 'struct ct_dpif_zone_limit' entries. */
>      int (*ct_del_limits)(struct dpif *, const struct ovs_list
> *zone_limits);
>
> +    /* Connection tracking timeout policy */
> +
> +    /* A connection tracking timeout policy contains a list of timeout
> +     * attributes that specifies timeout values on various connection
> states.
> +     * In a datapath, the timeout policy is identified by a 4 bytes
> unsigned
> +     * integer, and the unsupported timeout attributes are ignored.
> +     * When a connection is committed it can be associated with a timeout
> +     * policy, or it defaults to the datapath's default timeout policy. */
> +
> +    /* Sets timeout policy '*tp' into the datapath. */
> +    int (*ct_set_timeout_policy)(struct dpif *,
> +                                 const struct ct_dpif_timeout_policy *tp);
> +    /* Gets a timeout policy specified by tp_id and stores it into
> '*tp'.*/
> +    int (*ct_get_timeout_policy)(struct dpif *, uint32_t tp_id,
> +                                 struct ct_dpif_timeout_policy *tp);
> +    /* Deletes a timeout policy identified by 'tp_id'. */
> +    int (*ct_del_timeout_policy)(struct dpif *, uint32_t tp_id);
> +
> +    /* Conntrack timeout policy dumping interface.
> +     *
> +     * These functions provide a datapath-agnostic dumping interface
> +     * to the conntrack timeout policy provided by the datapaths.
> +     *
> +     * ct_timeout_policy_dump_start() should put in '*statep' a pointer to
> +     * a newly allocated structure that will be passed by the caller to
> +     * ct_timeout_policy_dump_next() and ct_timeout_policy_dump_done().
> +     *
> +     * ct_timeout_policy_dump_next() fills a timeout policy into '*tp',
> +     * and prepares to dump the next one on a subsequent invocation.
> +     *
> +     * ct_timeout_policy_dump_done() should perform any cleanup necessary
> +     * (including deallocating the 'state' structure, if applicable). */
> +    int (*ct_timeout_policy_dump_start)(struct dpif *, void **statep);
> +    int (*ct_timeout_policy_dump_next)(struct dpif *, void *state,
> +                                       struct ct_dpif_timeout_policy *tp);
> +    int (*ct_timeout_policy_dump_done)(struct dpif *, void *state);
> +
>      /* IP Fragmentation. */
>
>      /* Disables or enables conntrack fragment reassembly.  The default
> diff --git a/lib/netlink-conntrack.c b/lib/netlink-conntrack.c
> index 7631ba5d5d31..9bc0ddb66248 100644
> --- a/lib/netlink-conntrack.c
> +++ b/lib/netlink-conntrack.c
> @@ -840,6 +840,369 @@ nl_ct_parse_helper(struct nlattr *nla, struct
> ct_dpif_helper *helper)
>      return parsed;
>  }
>
> +static int nl_ct_timeout_policy_max_attr[] = {
> +    [IPPROTO_TCP] = CTA_TIMEOUT_TCP_MAX,
> +    [IPPROTO_UDP] = CTA_TIMEOUT_UDP_MAX,
> +    [IPPROTO_ICMP] = CTA_TIMEOUT_ICMP_MAX,
> +    [IPPROTO_ICMPV6] = CTA_TIMEOUT_ICMPV6_MAX
> +};
> +
> +static void
> +nl_ct_set_timeout_policy_attr(struct nl_ct_timeout_policy *nl_tp,
> +                              uint32_t attr, uint32_t val)
> +{
> +    nl_tp->present |= 1 << attr;
> +    nl_tp->attrs[attr] = val;
> +}
> +
> +static int
> +nl_ct_parse_tcp_timeout_policy_data(struct nlattr *nla,
> +                                    struct nl_ct_timeout_policy *nl_tp)
> +{
> +    static const struct nl_policy policy[] = {
> +        [CTA_TIMEOUT_TCP_SYN_SENT] =    { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_SYN_RECV] =    { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_FIN_WAIT] =    { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_CLOSE_WAIT] =  { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_LAST_ACK] =    { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_TIME_WAIT] =   { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_CLOSE] =       { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_SYN_SENT2] =   { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_RETRANS] =     { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_TCP_UNACK] =       { .type = NL_A_BE32,
> +                                          .optional = false },
> +    };
> +    struct nlattr *attrs[ARRAY_SIZE(policy)];
> +    int i;
> +
> +    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
> +        VLOG_ERR_RL(&rl, "Could not parse nested tcp timeout options. "
> +                    "Possibly incompatible Linux kernel version.");
> +        return EINVAL;
> +    }
> +
> +    for (i = CTA_TIMEOUT_TCP_SYN_SENT; i <= CTA_TIMEOUT_TCP_UNACK; i++) {
> +        nl_ct_set_timeout_policy_attr(nl_tp, i,
> +                                      ntohl(nl_attr_get_be32(attrs[i])));
> +    }
> +    return 0;
> +}
> +
> +static int
> +nl_ct_parse_udp_timeout_policy_data(struct nlattr *nla,
> +                                    struct nl_ct_timeout_policy *nl_tp)
> +{
> +    static const struct nl_policy policy[] = {
> +        [CTA_TIMEOUT_UDP_UNREPLIED] =   { .type = NL_A_BE32,
> +                                          .optional = false },
> +        [CTA_TIMEOUT_UDP_REPLIED] =     { .type = NL_A_BE32,
> +                                          .optional = false },
> +    };
> +    struct nlattr *attrs[ARRAY_SIZE(policy)];
> +    int i;
> +
> +    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
> +        VLOG_ERR_RL(&rl, "Could not parse nested tcp timeout options. "
> +                    "Possibly incompatible Linux kernel version.");
> +        return EINVAL;
> +    }
> +
> +    for (i = CTA_TIMEOUT_UDP_UNREPLIED; i <= CTA_TIMEOUT_UDP_REPLIED;
> i++) {
> +        nl_ct_set_timeout_policy_attr(nl_tp, i,
> +                                      ntohl(nl_attr_get_be32(attrs[i])));
> +    }
> +    return 0;
> +}
> +
> +static int
> +nl_ct_parse_icmp_timeout_policy_data(struct nlattr *nla,
> +                                     struct nl_ct_timeout_policy *nl_tp)
> +{
> +    static const struct nl_policy policy[] = {
> +        [CTA_TIMEOUT_ICMP_TIMEOUT] =   { .type = NL_A_BE32,
> +                                         .optional = false },
> +    };
> +    struct nlattr *attrs[ARRAY_SIZE(policy)];
> +
> +    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
> +        VLOG_ERR_RL(&rl, "Could not parse nested icmp timeout options. "
> +                    "Possibly incompatible Linux kernel version.");
> +        return EINVAL;
> +    }
> +
> +    nl_ct_set_timeout_policy_attr(
> +        nl_tp, CTA_TIMEOUT_ICMP_TIMEOUT,
> +        ntohl(nl_attr_get_be32(attrs[CTA_TIMEOUT_ICMP_TIMEOUT])));
> +    return 0;
> +}
> +
> +static int
> +nl_ct_parse_icmpv6_timeout_policy_data(struct nlattr *nla,
> +                                       struct nl_ct_timeout_policy *nl_tp)
> +{
> +    static const struct nl_policy policy[] = {
> +        [CTA_TIMEOUT_ICMPV6_TIMEOUT] =   { .type = NL_A_BE32,
> +                                           .optional = false },
> +    };
> +    struct nlattr *attrs[ARRAY_SIZE(policy)];
> +
> +    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
> +        VLOG_ERR_RL(&rl, "Could not parse nested icmpv6 timeout options. "
> +                    "Possibly incompatible Linux kernel version.");
> +        return EINVAL;
> +    }
> +
> +    nl_ct_set_timeout_policy_attr(
> +        nl_tp, CTA_TIMEOUT_ICMPV6_TIMEOUT,
> +        ntohl(nl_attr_get_be32(attrs[CTA_TIMEOUT_ICMPV6_TIMEOUT])));
> +    return 0;
> +}
> +
> +static int
> +nl_ct_parse_timeout_policy_data(struct nlattr *nla,
> +                                struct nl_ct_timeout_policy *nl_tp)
> +{
> +    switch (nl_tp->l4num) {
> +        case IPPROTO_TCP:
> +            return nl_ct_parse_tcp_timeout_policy_data(nla, nl_tp);
> +        case IPPROTO_UDP:
> +            return nl_ct_parse_udp_timeout_policy_data(nla, nl_tp);
> +        case IPPROTO_ICMP:
> +            return nl_ct_parse_icmp_timeout_policy_data(nla, nl_tp);
> +        case IPPROTO_ICMPV6:
> +            return nl_ct_parse_icmpv6_timeout_policy_data(nla, nl_tp);
> +        default:
> +            return EINVAL;
> +    }
> +}
> +
> +static int
> +nl_ct_timeout_policy_from_ofpbuf(struct ofpbuf *buf,
> +                                 struct nl_ct_timeout_policy *nl_tp,
> +                                 bool default_tp)
> +{
> +    static const struct nl_policy policy[] = {
> +        [CTA_TIMEOUT_NAME] =    { .type = NL_A_STRING, .optional = false
> },
> +        [CTA_TIMEOUT_L3PROTO] = { .type = NL_A_BE16, .optional = false },
> +        [CTA_TIMEOUT_L4PROTO] = { .type = NL_A_U8, .optional = false },
> +        [CTA_TIMEOUT_DATA] =    { .type = NL_A_NESTED, .optional = false }
> +    };
> +    static const struct nl_policy policy_default_tp[] = {
> +        [CTA_TIMEOUT_L3PROTO] = { .type = NL_A_BE16, .optional = false },
> +        [CTA_TIMEOUT_L4PROTO] = { .type = NL_A_U8, .optional = false },
> +        [CTA_TIMEOUT_DATA] =    { .type = NL_A_NESTED, .optional = false }
> +    };
> +
> +    struct nlattr *attrs[ARRAY_SIZE(policy)];
> +    struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size);
> +    struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
> +    struct nfgenmsg *nfmsg = ofpbuf_try_pull(&b, sizeof *nfmsg);
> +    int err;
> +
> +    if (!nlmsg || !nfmsg
> +        || NFNL_SUBSYS_ID(nlmsg->nlmsg_type) !=
> NFNL_SUBSYS_CTNETLINK_TIMEOUT
> +        || nfmsg->version != NFNETLINK_V0
> +        || !nl_policy_parse(&b, 0, default_tp ? policy_default_tp :
> policy,
> +                            attrs, default_tp ?
> ARRAY_SIZE(policy_default_tp) :
> +                                                ARRAY_SIZE(policy))) {
> +        return EINVAL;
> +    }
> +
> +    if (!default_tp) {
> +        ovs_strlcpy(nl_tp->name,
> nl_attr_get_string(attrs[CTA_TIMEOUT_NAME]),
> +                    sizeof nl_tp->name);
> +    }
> +    nl_tp->l3num = ntohs(nl_attr_get_be16(attrs[CTA_TIMEOUT_L3PROTO]));
> +    nl_tp->l4num = nl_attr_get_u8(attrs[CTA_TIMEOUT_L4PROTO]);
> +    nl_tp->present = 0;
> +
> +    err = nl_ct_parse_timeout_policy_data(attrs[CTA_TIMEOUT_DATA], nl_tp);
> +    return err;
> +}
> +
> +int
> +nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
> +{
> +    struct ofpbuf buf;
> +    size_t offset;
> +    int i, err;
> +
> +    ofpbuf_init(&buf, 512);
> +    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_NEW, NLM_F_REQUEST |
> NLM_F_CREATE
> +                        | NLM_F_ACK | NLM_F_REPLACE);
> +
> +    nl_msg_put_string(&buf, CTA_TIMEOUT_NAME, nl_tp->name);
> +    nl_msg_put_be16(&buf, CTA_TIMEOUT_L3PROTO, htons(nl_tp->l3num));
> +    nl_msg_put_u8(&buf, CTA_TIMEOUT_L4PROTO, nl_tp->l4num);
> +
> +    offset = nl_msg_start_nested(&buf, CTA_TIMEOUT_DATA);
> +    for (i = 1; i <= nl_ct_timeout_policy_max_attr[nl_tp->l4num]; ++i) {
> +        if (nl_tp->present & 1 << i) {
> +            nl_msg_put_be32(&buf, i, htonl(nl_tp->attrs[i]));
> +        }
> +    }
> +    nl_msg_end_nested(&buf, offset);
> +
> +    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
> +    ofpbuf_uninit(&buf);
> +    return err;
> +}
> +
> +int
> +nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
>

Remove all the default APIs form here and .c file, as discussed


> +{
> +    struct ofpbuf buf;
> +    size_t offset;
> +    int i, err;
> +
> +    ofpbuf_init(&buf, 512);
> +    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_DEFAULT_SET, NLM_F_REQUEST
> +                        | NLM_F_ACK | NLM_F_REPLACE);
> +
> +    nl_msg_put_be16(&buf, CTA_TIMEOUT_L3PROTO, htons(nl_tp->l3num));
> +    nl_msg_put_u8(&buf, CTA_TIMEOUT_L4PROTO, nl_tp->l4num);
> +
> +    offset = nl_msg_start_nested(&buf, CTA_TIMEOUT_DATA);
> +    for (i = 1; i <= nl_ct_timeout_policy_max_attr[nl_tp->l4num]; ++i) {
> +        if (nl_tp->present & 1 << i) {
> +            nl_msg_put_be32(&buf, i, htonl(nl_tp->attrs[i]));
> +        }
> +    }
> +    nl_msg_end_nested(&buf, offset);
> +
> +    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
> +    ofpbuf_uninit(&buf);
> +    return err;
> +}
> +
> +int
> +nl_ct_get_timeout_policy(const char *tp_name,
> +                         struct nl_ct_timeout_policy *nl_tp)
> +{
> +    struct ofpbuf request, *reply;
> +    int err;
> +
> +    ofpbuf_init(&request, 512);
> +    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC,
> NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_GET, NLM_F_REQUEST |
> NLM_F_ACK);
> +    nl_msg_put_string(&request, CTA_TIMEOUT_NAME, tp_name);
> +    err = nl_transact(NETLINK_NETFILTER, &request, &reply);
> +    if (err) {
> +        goto out;
> +    }
> +
> +    err = nl_ct_timeout_policy_from_ofpbuf(reply, nl_tp, false);
> +
> +out:
> +    ofpbuf_uninit(&request);
> +    ofpbuf_delete(reply);
> +    return err;
> +}
> +
> +int
> +nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
> +                                 struct nl_ct_timeout_policy *nl_tp)
>

Remove all the default APIs form here and .c file


> +{
> +    struct ofpbuf request, *reply;
> +    int err;
> +
> +    ofpbuf_init(&request, 512);
> +    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC,
> NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_DEFAULT_GET,
> +                        NLM_F_REQUEST | NLM_F_ACK);
> +
> +    nl_msg_put_be16(&request, CTA_TIMEOUT_L3PROTO, htons(l3num));
> +    nl_msg_put_u8(&request, CTA_TIMEOUT_L4PROTO, l4num);
> +    err = nl_transact(NETLINK_NETFILTER, &request, &reply);
> +    if (err) {
> +        goto out;
> +    }
> +
> +    err = nl_ct_timeout_policy_from_ofpbuf(reply, nl_tp, true);
> +
> +out:
> +    ofpbuf_uninit(&request);
> +    ofpbuf_delete(reply);
> +    return err;
> +}
> +
> +int
> +nl_ct_del_timeout_policy(const char *tp_name)
> +{
> +    struct ofpbuf buf;
> +    int err;
> +
> +    ofpbuf_init(&buf, 64);
> +    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_DELETE, NLM_F_REQUEST |
> NLM_F_ACK);
> +
> +    nl_msg_put_string(&buf, CTA_TIMEOUT_NAME, tp_name);
> +    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
> +    ofpbuf_uninit(&buf);
> +    return err;
> +}
> +
> +struct nl_ct_timeout_policy_dump_state {
> +    struct nl_dump dump;
> +    struct ofpbuf buf;
> +};
> +
> +int
> +nl_ct_timeout_policy_dump_start(
> +    struct nl_ct_timeout_policy_dump_state **statep)
> +{
> +    struct ofpbuf request;
> +    struct nl_ct_timeout_policy_dump_state *state;
> +
> +    *statep = state = xzalloc(sizeof *state);
> +    ofpbuf_init(&request, 512);
> +    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC,
> NFNL_SUBSYS_CTNETLINK_TIMEOUT,
> +                        IPCTNL_MSG_TIMEOUT_GET,
> +                        NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP);
> +
> +    nl_dump_start(&state->dump, NETLINK_NETFILTER, &request);
> +    ofpbuf_uninit(&request);
> +    ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE);
> +    return 0;
> +}
> +
> +int
> +nl_ct_timeout_policy_dump_next(struct nl_ct_timeout_policy_dump_state
> *state,
> +                               struct nl_ct_timeout_policy *nl_tp)
> +{
> +    struct ofpbuf reply;
> +    int err;
> +
> +    if (!nl_dump_next(&state->dump, &reply, &state->buf)) {
> +        return EOF;
> +    }
> +    err = nl_ct_timeout_policy_from_ofpbuf(&reply, nl_tp, false);
> +    ofpbuf_uninit(&reply);
> +    return err;
> +}
> +
> +int
> +nl_ct_timeout_policy_dump_done(struct nl_ct_timeout_policy_dump_state
> *state)
> +{
> +    int err  = nl_dump_done(&state->dump);
> +    ofpbuf_uninit(&state->buf);
> +    free(state);
> +    return err;
> +}
> +
>  /* Translate netlink entry status flags to CT_DPIF_TCP status flags. */
>  static uint32_t
>  ips_status_to_dpif_flags(uint32_t status)
> diff --git a/lib/netlink-conntrack.h b/lib/netlink-conntrack.h
> index 8b536fd65ba8..ae6e428e0929 100644
> --- a/lib/netlink-conntrack.h
> +++ b/lib/netlink-conntrack.h
> @@ -17,9 +17,12 @@
>  #ifndef NETLINK_CONNTRACK_H
>  #define NETLINK_CONNTRACK_H
>
> +#include <linux/netfilter/nfnetlink_cttimeout.h>
> +
>  #include "byte-order.h"
>  #include "compiler.h"
>  #include "ct-dpif.h"
> +#include "netlink-socket.h"
>  #include "openvswitch/dynamic-string.h"
>  #include "openvswitch/hmap.h"
>  #include "openvswitch/ofpbuf.h"
> @@ -33,7 +36,18 @@ enum nl_ct_event_type {
>      NL_CT_EVENT_DELETE = 1 << 2,
>  };
>
> +#define NL_CT_TIMEOUT_POLICY_MAX_ATTR (CTA_TIMEOUT_TCP_MAX + 1)
> +
> +struct nl_ct_timeout_policy {
> +    char        name[CTNL_TIMEOUT_NAME_MAX];
> +    uint16_t    l3num;
> +    uint8_t     l4num;
> +    uint32_t    attrs[NL_CT_TIMEOUT_POLICY_MAX_ATTR];
> +    uint32_t    present;
> +};
> +
>  struct nl_ct_dump_state;
> +struct nl_ct_timeout_policy_dump_state;
>
>  int nl_ct_dump_start(struct nl_ct_dump_state **, const uint16_t *zone,
>          int *ptot_bkts);
> @@ -44,6 +58,21 @@ int nl_ct_flush(void);
>  int nl_ct_flush_zone(uint16_t zone);
>  int nl_ct_flush_tuple(const struct ct_dpif_tuple *, uint16_t zone);
>
> +int nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp);
> +int nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy
> *nl_tp);
>

Remove all the default APIs form here and .c file, as discussed


> +int nl_ct_get_timeout_policy(const char *tp_name,
> +                             struct nl_ct_timeout_policy *nl_tp);
> +int nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
> +                                     struct nl_ct_timeout_policy *nl_tp);
>

Remove all the default APIs form here and .c file



> +int nl_ct_del_timeout_policy(const char *tp_name);
> +int nl_ct_timeout_policy_dump_start(
> +    struct nl_ct_timeout_policy_dump_state **statep);
> +int nl_ct_timeout_policy_dump_next(
> +    struct nl_ct_timeout_policy_dump_state *state,
> +    struct nl_ct_timeout_policy *nl_tp);
> +int nl_ct_timeout_policy_dump_done(
> +    struct nl_ct_timeout_policy_dump_state *state);
> +
>  bool nl_ct_parse_entry(struct ofpbuf *, struct ct_dpif_entry *,
>                         enum nl_ct_event_type *);
>  void nl_ct_format_event_entry(const struct ct_dpif_entry *,
> diff --git a/lib/netlink-protocol.h b/lib/netlink-protocol.h
> index c0617dfad21f..bf631b1a14d0 100644
> --- a/lib/netlink-protocol.h
> +++ b/lib/netlink-protocol.h
> @@ -48,6 +48,7 @@
>  #define NLM_F_ECHO              0x008
>
>  #define NLM_F_ROOT              0x100
> +#define NLM_F_REPLACE           0x100
>  #define NLM_F_MATCH             0x200
>  #define NLM_F_EXCL              0x200
>  #define NLM_F_ATOMIC            0x400
> --
> 2.7.4
>
> _______________________________________________
> dev mailing list
> dev@openvswitch.org
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>
Justin Pettit Aug. 6, 2019, 5:27 a.m. UTC | #3
> On Aug 1, 2019, at 3:07 PM, Yi-Hung Wei <yihung.wei@gmail.com> wrote:
> 
> This patch also adds missing symbols in the windows datapath so
> that the build on windows can pass.

Alin, can you do a quick sanity-check on the Windows part?  Thanks.

Yi-Hung, it looks good to me.  Just mostly some minor stuff below.

> diff --git a/datapath-windows/ovsext/Netlink/NetlinkProto.h b/datapath-windows/ovsext/Netlink/NetlinkProto.h
> index 59b56565c1dc..db1fa2bacae8 100644
> --- a/datapath-windows/ovsext/Netlink/NetlinkProto.h
> +++ b/datapath-windows/ovsext/Netlink/NetlinkProto.h
> @@ -51,6 +51,7 @@
> #define NLM_F_ECHO              0x008
> 
> #define NLM_F_ROOT              0x100
> +#define NLM_F_REPLACE           0x100
> #define NLM_F_MATCH             0x200
> #define NLM_F_EXCL              0x200
> #define NLM_F_ATOMIC            0x400

Below, I mention reorganizing the similar Linux header file based on request type.  I'd recommend that here, too.

> diff --git a/lib/ct-dpif.h b/lib/ct-dpif.h
> index 2f4906817946..8dacb1c7c253 100644
> --- a/lib/ct-dpif.h
> +++ b/lib/ct-dpif.h
> ...
> +struct ct_dpif_timeout_policy {
> +    uint32_t    id;         /* An unique identifier of a timeout policy in
> +                             * the datapath. */
> +    uint32_t    present;    /* If a timeout attribute is present set the
> +                             * corresponding bit. */

This is a minor nit, but I think this would be clearer if it were less abstract.  For example:

    uint32_t    id;         /* Unique identifier for the timeout policy in
                             * the datapath. */
    uint32_t    present;    /* If a timeout attribute is present set the
                             * corresponding CT_DPIF_TP_ATTR_* mapping bit. */

> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index d0a1c58adace..2079e368fb52 100644
> --- a/lib/dpif-netdev.c
> +++ b/lib/dpif-netdev.c
> @@ -7529,6 +7529,12 @@ const struct dpif_class dpif_netdev_class = {
>     NULL,                       /* ct_set_limits */
>     NULL,                       /* ct_get_limits */
>     NULL,                       /* ct_del_limits */
> +    NULL,                       /* ct_set_timeout_policy */
> +    NULL,                       /* ct_get_timeout_policy */
> +    NULL,                       /* ct_del_timeout_policy */
> +    NULL,                       /* ct_timeout_policy_dump_start */
> +    NULL,                       /* ct_timeout_policy_dump_next */
> +    NULL,                       /* ct_timeout_policy_dump_done */

Is there a plan to add timeout policies to the userspace datapath?  It's probably worth adding a line to "Documentation/faq/releases.rst".

> diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
> index 7bc71d6d19d7..b859508f718a 100644
> --- a/lib/dpif-netlink.c
> +++ b/lib/dpif-netlink.c
> 
> +enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol {
> +    DPIF_NL_TP_AF_INET_TCP,
> +    DPIF_NL_TP_AF_INET_UDP,
> +    DPIF_NL_TP_AF_INET_ICMP,
> +    DPIF_NL_TP_AF_INET6_TCP,
> +    DPIF_NL_TP_AF_INET6_UDP,
> +    DPIF_NL_TP_AF_INET6_ICMPV6,
> +    DPIF_NL_TP_MAX
> +};
> +
> +#define DPIF_NL_ALL_TP 0x3F

Not that it will be changing all that much, but I think it's always nice to auto-generate these sorts of values:

	#define DPIF_NL_ALL_TP ((1UL << DPIF_NL_TP_MAX) - 1)

> +static int
> +dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
> +                                   const struct ct_dpif_timeout_policy *tp)
> +{
> ...
> +out:
> +    ds_destroy(&nl_tp_name);
> +    return  err;

There's an extra space before 'err'.

> +struct dpif_netlink_tp_dump_node {
> +    struct      hmap_node hmap_node;      /* node in tp_dump_map. */
> +    struct      ct_dpif_timeout_policy *tp;
> +    uint32_t    present;

I think having this called 'present' is a little confusing since 'tp' has a different member called 'present'.  How about something like "l3_l4_present"?

> +static int
> +dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
> +                                         void *state,
> +                                         struct ct_dpif_timeout_policy *tp)
> +{
> ...
> +    struct nl_ct_timeout_policy nl_tp;
> +    uint32_t tp_id;

These two variables could be declared in a tighter scope.

> +static int
> +dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED,
> +                                         void *state)
> +{
> +    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
> +    int err;
> +
> +    err = nl_ct_timeout_policy_dump_done(dump_state->nl_dump_state);
> +    hmap_destroy(&dump_state->tp_dump_map);
> +    free(dump_state);
> +    return err;
> +}

If someone calls "_dump_done()" before they've called "_dump_next()" until it returned EOF, it seems like we could leak nodes in the dump state.  Should those be cleaned up?  If not, it should probably at least be documented in this function and the prototype description.  (I didn't propose a patch here, since I didn't know how you want to handle it.)

> diff --git a/lib/dpif-netlink.h b/lib/dpif-netlink.h
> index 0a9628088275..1e753994d747 100644
> --- a/lib/dpif-netlink.h
> +++ b/lib/dpif-netlink.h
> @@ -23,6 +23,7 @@
> #include "odp-netlink.h"
> 
> #include "flow.h"
> +#include "netlink-conntrack.h"

Why is this included?  I was able to build fine without it.

> diff --git a/lib/netlink-conntrack.c b/lib/netlink-conntrack.c
> index 7631ba5d5d31..9bc0ddb66248 100644
> --- a/lib/netlink-conntrack.c
> +++ b/lib/netlink-conntrack.c
> ...
> +int
> +nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
> +{
> ...
> +}

I don't think this function is needed anymore.

> +int
> +nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
> +                                 struct nl_ct_timeout_policy *nl_tp)
> +{
> ...
> +}

Same with this.

> diff --git a/lib/netlink-conntrack.h b/lib/netlink-conntrack.h
> index 8b536fd65ba8..ae6e428e0929 100644
> --- a/lib/netlink-conntrack.h
> +++ b/lib/netlink-conntrack.h
> @@ -17,9 +17,12 @@
> ...
> +#include "netlink-socket.h"

Is this needed?  It seems like it builds fine without it.

> diff --git a/lib/netlink-protocol.h b/lib/netlink-protocol.h
> index c0617dfad21f..bf631b1a14d0 100644
> --- a/lib/netlink-protocol.h
> +++ b/lib/netlink-protocol.h
> @@ -48,6 +48,7 @@
> #define NLM_F_ECHO              0x008
> 
> #define NLM_F_ROOT              0x100
> +#define NLM_F_REPLACE           0x100

I know it's not new to your changes, but it looks odd to have two different definitions with the same value.  I think it would be clearer if they were grouped and labeled with the request type.

Thanks,

--Justin


-=-=-=-=-=-=-=-=-=-=-

diff --git a/Documentation/faq/releases.rst b/Documentation/faq/releases.rst
index 8daa23bb2d0c..985e20067dba 100644
--- a/Documentation/faq/releases.rst
+++ b/Documentation/faq/releases.rst
@@ -110,8 +110,9 @@ Q: Are all features available with all datapaths?
     ========================== ============== ============== ========= =======
     Connection tracking             4.3            YES          YES      YES
     Conntrack Fragment Reass.       4.3            YES          YES      YES
+    Conntrack Timeout Policies      4.3            YES          NO       NO
+    Conntrack Zone Limit            4.18           YES          NO       YES
     NAT                             4.6            YES          YES      YES
-    Conntrack zone limit            4.18           YES          NO       YES
     Tunnel - LISP                   NO             YES          NO       NO
     Tunnel - STT                    NO             YES          NO       YES
     Tunnel - GRE                    3.11           YES          YES      YES
diff --git a/datapath-windows/ovsext/Netlink/NetlinkProto.h b/datapath-windows/ovsext/Netlink/NetlinkProto.h
index db1fa2bacae8..b32f6f7fb114 100644
--- a/datapath-windows/ovsext/Netlink/NetlinkProto.h
+++ b/datapath-windows/ovsext/Netlink/NetlinkProto.h
@@ -50,14 +50,17 @@
 #define NLM_F_ACK               0x004
 #define NLM_F_ECHO              0x008
 
+/* GET request flag.*/
 #define NLM_F_ROOT              0x100
-#define NLM_F_REPLACE           0x100
 #define NLM_F_MATCH             0x200
-#define NLM_F_EXCL              0x200
 #define NLM_F_ATOMIC            0x400
-#define NLM_F_CREATE            0x400
 #define NLM_F_DUMP              (NLM_F_ROOT | NLM_F_MATCH)
 
+/* NEW request flags. */
+#define NLM_F_REPLACE           0x100
+#define NLM_F_EXCL              0x200
+#define NLM_F_CREATE            0x400
+
 /* nlmsg_type values. */
 #define NLMSG_NOOP              1
 #define NLMSG_ERROR             2
diff --git a/lib/ct-dpif.h b/lib/ct-dpif.h
index 8dacb1c7c253..aabd6962f2c0 100644
--- a/lib/ct-dpif.h
+++ b/lib/ct-dpif.h
@@ -261,10 +261,10 @@ enum OVS_PACKED_ENUM ct_dpif_tp_attr {
 };
 
 struct ct_dpif_timeout_policy {
-    uint32_t    id;         /* An unique identifier of a timeout policy in
+    uint32_t    id;         /* Unique identifier for the timeout policy in
                              * the datapath. */
     uint32_t    present;    /* If a timeout attribute is present set the
-                             * corresponding bit. */
+                             * corresponding CT_DPIF_TP_ATTR_* mapping bit. */
     uint32_t    attrs[CT_DPIF_TP_ATTR_MAX];     /* An array that specifies
                                                  * timeout attribute values */
 };
diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
index b859508f718a..08073a6f6a56 100644
--- a/lib/dpif-netlink.c
+++ b/lib/dpif-netlink.c
@@ -3042,7 +3042,8 @@ enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol {
     DPIF_NL_TP_MAX
 };
 
-#define DPIF_NL_ALL_TP 0x3F
+#define DPIF_NL_ALL_TP ((1UL << DPIF_NL_TP_MAX) - 1)
+
 
 static struct dpif_netlink_timeout_policy_protocol tp_protos[] = {
     [DPIF_NL_TP_AF_INET_TCP] = { .l3num = AF_INET, .l4num = IPPROTO_TCP },
@@ -3202,7 +3203,7 @@ dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(
 
 static void
 dpif_netlink_set_ct_dpif_tp_attrs(const struct nl_ct_timeout_policy *nl_tp,
-                               struct ct_dpif_timeout_policy *tp)
+                                  struct ct_dpif_timeout_policy *tp)
 {
     if (nl_tp->l4num == IPPROTO_TCP) {
         dpif_netlink_set_ct_dpif_tp_tcp_attrs(nl_tp, tp);
@@ -3285,7 +3286,7 @@ dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
 
 out:
     ds_destroy(&nl_tp_name);
-    return  err;
+    return err;
 }
 
 static int
@@ -3314,7 +3315,7 @@ dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED,
 
 out:
     ds_destroy(&nl_tp_name);
-    return  err;
+    return err;
 }
 
 /* Returns 0 if all the sub timeout policies are deleted or
@@ -3353,7 +3354,7 @@ struct dpif_netlink_ct_timeout_policy_dump_state {
 struct dpif_netlink_tp_dump_node {
     struct      hmap_node hmap_node;      /* node in tp_dump_map. */
     struct      ct_dpif_timeout_policy *tp;
-    uint32_t    present;
+    uint32_t    l3_l4_present;
 };
 
 static struct dpif_netlink_tp_dump_node *
@@ -3363,7 +3364,7 @@ get_dpif_netlink_tp_dump_node_by_tp_id(uint32_t tp_id,
     struct dpif_netlink_tp_dump_node *tp_dump_node;
 
     HMAP_FOR_EACH_WITH_HASH (tp_dump_node, hmap_node, hash_int(tp_id, 0),
-                            tp_dump_map) {
+                             tp_dump_map) {
         if (tp_dump_node->tp->id == tp_id) {
             return tp_dump_node;
         }
@@ -3382,7 +3383,7 @@ update_dpif_netlink_tp_dump_node(
     for (i = 0; i < DPIF_NL_TP_MAX; ++i) {
         if (nl_tp->l3num == tp_protos[i].l3num &&
             nl_tp->l4num == tp_protos[i].l4num) {
-            tp_dump_node->present |= 1 << i;
+            tp_dump_node->l3_l4_present |= 1 << i;
             break;
         }
     }
@@ -3412,11 +3413,12 @@ dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
 {
     struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
     struct dpif_netlink_tp_dump_node *tp_dump_node;
-    struct nl_ct_timeout_policy nl_tp;
-    uint32_t tp_id;
     int err;
 
     do {
+        struct nl_ct_timeout_policy nl_tp;
+        uint32_t tp_id;
+
         err =  nl_ct_timeout_policy_dump_next(dump_state->nl_dump_state,
                                               &nl_tp);
         if (err) {
@@ -3438,7 +3440,7 @@ dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
         }
 
         update_dpif_netlink_tp_dump_node(&nl_tp, tp_dump_node);
-        if (tp_dump_node->present == DPIF_NL_ALL_TP) {
+        if (tp_dump_node->l3_l4_present == DPIF_NL_ALL_TP) {
             hmap_remove(&dump_state->tp_dump_map, &tp_dump_node->hmap_node);
             *tp = *tp_dump_node->tp;
             free(tp_dump_node->tp);
diff --git a/lib/dpif-netlink.h b/lib/dpif-netlink.h
index 1e753994d747..d2689029c3e0 100644
--- a/lib/dpif-netlink.h
+++ b/lib/dpif-netlink.h
@@ -20,7 +20,6 @@
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
-#include "odp-netlink.h"
 
 #include "flow.h"
 #include "netlink-conntrack.h"
diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h
index 79a2314500cf..fb56d91eafe9 100644
--- a/lib/dpif-provider.h
+++ b/lib/dpif-provider.h
@@ -502,16 +502,16 @@ struct dpif_class {
     /* Connection tracking timeout policy */
 
     /* A connection tracking timeout policy contains a list of timeout
-     * attributes that specifies timeout values on various connection states.
-     * In a datapath, the timeout policy is identified by a 4 bytes unsigned
-     * integer, and the unsupported timeout attributes are ignored.
-     * When a connection is committed it can be associated with a timeout
+     * attributes that specify timeout values on various connection states.
+     * In a datapath, the timeout policy is identified by a 4-byte unsigned
+     * integer.  Unsupported timeout attributes are ignored.  When a
+     * connection is committed it can be associated with a timeout
      * policy, or it defaults to the datapath's default timeout policy. */
 
     /* Sets timeout policy '*tp' into the datapath. */
     int (*ct_set_timeout_policy)(struct dpif *,
                                  const struct ct_dpif_timeout_policy *tp);
-    /* Gets a timeout policy specified by tp_id and stores it into '*tp'.*/
+    /* Gets a timeout policy specified by tp_id and stores it into '*tp'. */
     int (*ct_get_timeout_policy)(struct dpif *, uint32_t tp_id,
                                  struct ct_dpif_timeout_policy *tp);
     /* Deletes a timeout policy identified by 'tp_id'. */
diff --git a/lib/netlink-conntrack.c b/lib/netlink-conntrack.c
index 9bc0ddb66248..828e4a5a84c1 100644
--- a/lib/netlink-conntrack.c
+++ b/lib/netlink-conntrack.c
@@ -1060,34 +1060,6 @@ nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
     return err;
 }
 
-int
-nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
-{
-    struct ofpbuf buf;
-    size_t offset;
-    int i, err;
-
-    ofpbuf_init(&buf, 512);
-    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
-                        IPCTNL_MSG_TIMEOUT_DEFAULT_SET, NLM_F_REQUEST
-                        | NLM_F_ACK | NLM_F_REPLACE);
-
-    nl_msg_put_be16(&buf, CTA_TIMEOUT_L3PROTO, htons(nl_tp->l3num));
-    nl_msg_put_u8(&buf, CTA_TIMEOUT_L4PROTO, nl_tp->l4num);
-
-    offset = nl_msg_start_nested(&buf, CTA_TIMEOUT_DATA);
-    for (i = 1; i <= nl_ct_timeout_policy_max_attr[nl_tp->l4num]; ++i) {
-        if (nl_tp->present & 1 << i) {
-            nl_msg_put_be32(&buf, i, htonl(nl_tp->attrs[i]));
-        }
-    }
-    nl_msg_end_nested(&buf, offset);
-
-    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
-    ofpbuf_uninit(&buf);
-    return err;
-}
-
 int
 nl_ct_get_timeout_policy(const char *tp_name,
                          struct nl_ct_timeout_policy *nl_tp)
@@ -1112,33 +1084,6 @@ out:
     return err;
 }
 
-int
-nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
-                                 struct nl_ct_timeout_policy *nl_tp)
-{
-    struct ofpbuf request, *reply;
-    int err;
-
-    ofpbuf_init(&request, 512);
-    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
-                        IPCTNL_MSG_TIMEOUT_DEFAULT_GET,
-                        NLM_F_REQUEST | NLM_F_ACK);
-
-    nl_msg_put_be16(&request, CTA_TIMEOUT_L3PROTO, htons(l3num));
-    nl_msg_put_u8(&request, CTA_TIMEOUT_L4PROTO, l4num);
-    err = nl_transact(NETLINK_NETFILTER, &request, &reply);
-    if (err) {
-        goto out;
-    }
-
-    err = nl_ct_timeout_policy_from_ofpbuf(reply, nl_tp, true);
-
-out:
-    ofpbuf_uninit(&request);
-    ofpbuf_delete(reply);
-    return err;
-}
-
 int
 nl_ct_del_timeout_policy(const char *tp_name)
 {
diff --git a/lib/netlink-conntrack.h b/lib/netlink-conntrack.h
index ae6e428e0929..81c74549bd16 100644
--- a/lib/netlink-conntrack.h
+++ b/lib/netlink-conntrack.h
@@ -22,7 +22,6 @@
 #include "byte-order.h"
 #include "compiler.h"
 #include "ct-dpif.h"
-#include "netlink-socket.h"
 #include "openvswitch/dynamic-string.h"
 #include "openvswitch/hmap.h"
 #include "openvswitch/ofpbuf.h"
@@ -50,7 +49,7 @@ struct nl_ct_dump_state;
 struct nl_ct_timeout_policy_dump_state;
 
 int nl_ct_dump_start(struct nl_ct_dump_state **, const uint16_t *zone,
-        int *ptot_bkts);
+                     int *ptot_bkts);
 int nl_ct_dump_next(struct nl_ct_dump_state *, struct ct_dpif_entry *);
 int nl_ct_dump_done(struct nl_ct_dump_state *);
 
@@ -59,11 +58,8 @@ int nl_ct_flush_zone(uint16_t zone);
 int nl_ct_flush_tuple(const struct ct_dpif_tuple *, uint16_t zone);
 
 int nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp);
-int nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp);
 int nl_ct_get_timeout_policy(const char *tp_name,
                              struct nl_ct_timeout_policy *nl_tp);
-int nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
-                                     struct nl_ct_timeout_policy *nl_tp);
 int nl_ct_del_timeout_policy(const char *tp_name);
 int nl_ct_timeout_policy_dump_start(
     struct nl_ct_timeout_policy_dump_state **statep);
diff --git a/lib/netlink-protocol.h b/lib/netlink-protocol.h
index bf631b1a14d0..ceded7915ef8 100644
--- a/lib/netlink-protocol.h
+++ b/lib/netlink-protocol.h
@@ -47,14 +47,17 @@
 #define NLM_F_ACK               0x004
 #define NLM_F_ECHO              0x008
 
+/* GET request flag.*/
 #define NLM_F_ROOT              0x100
-#define NLM_F_REPLACE           0x100
 #define NLM_F_MATCH             0x200
-#define NLM_F_EXCL              0x200
 #define NLM_F_ATOMIC            0x400
-#define NLM_F_CREATE            0x400
 #define NLM_F_DUMP              (NLM_F_ROOT | NLM_F_MATCH)
 
+/* NEW request flags. */
+#define NLM_F_REPLACE           0x100
+#define NLM_F_EXCL              0x200
+#define NLM_F_CREATE            0x400
+
 /* nlmsg_type values. */
 #define NLMSG_NOOP              1
 #define NLMSG_ERROR             2
Alin Gabriel Serdean Aug. 13, 2019, 11:27 a.m. UTC | #4
> -----Original Message-----
> From: ovs-dev-bounces@openvswitch.org <ovs-dev-
> bounces@openvswitch.org> On Behalf Of Justin Pettit
> Sent: Tuesday, August 6, 2019 8:28 AM
> To: Yi-Hung Wei <yihung.wei@gmail.com>
> Cc: ovs dev <dev@openvswitch.org>; Alin Gabriel Serdean
> <aserdean@ovn.org>
> Subject: Re: [ovs-dev] [PATCH v2 4/9] ct-dpif, dpif-netlink: Add conntrack
> timeout policy support
> 
> 
> > On Aug 1, 2019, at 3:07 PM, Yi-Hung Wei <yihung.wei@gmail.com> wrote:
> >
> > This patch also adds missing symbols in the windows datapath so that
> > the build on windows can pass.
> 
> Alin, can you do a quick sanity-check on the Windows part?  Thanks.
[Alin] Sorry for the delay. I replied here:
https://patchwork.ozlabs.org/patch/1145968/#2236957

Patch
diff mbox series

diff --git a/datapath-windows/include/OvsDpInterfaceCtExt.h b/datapath-windows/include/OvsDpInterfaceCtExt.h
index 3b947782e90c..4379855bb8dd 100644
--- a/datapath-windows/include/OvsDpInterfaceCtExt.h
+++ b/datapath-windows/include/OvsDpInterfaceCtExt.h
@@ -421,4 +421,118 @@  struct nf_ct_tcp_flags {
     UINT8 mask;
 };
 
+/* File: nfnetlink_cttimeout.h */
+enum ctnl_timeout_msg_types {
+    IPCTNL_MSG_TIMEOUT_NEW,
+    IPCTNL_MSG_TIMEOUT_GET,
+    IPCTNL_MSG_TIMEOUT_DELETE,
+    IPCTNL_MSG_TIMEOUT_DEFAULT_SET,
+    IPCTNL_MSG_TIMEOUT_DEFAULT_GET,
+
+    IPCTNL_MSG_TIMEOUT_MAX
+};
+
+enum ctattr_timeout {
+    CTA_TIMEOUT_UNSPEC,
+    CTA_TIMEOUT_NAME,
+    CTA_TIMEOUT_L3PROTO,
+    CTA_TIMEOUT_L4PROTO,
+    CTA_TIMEOUT_DATA,
+    CTA_TIMEOUT_USE,
+    __CTA_TIMEOUT_MAX
+};
+#define CTA_TIMEOUT_MAX (__CTA_TIMEOUT_MAX - 1)
+
+enum ctattr_timeout_generic {
+    CTA_TIMEOUT_GENERIC_UNSPEC,
+    CTA_TIMEOUT_GENERIC_TIMEOUT,
+    __CTA_TIMEOUT_GENERIC_MAX
+};
+#define CTA_TIMEOUT_GENERIC_MAX (__CTA_TIMEOUT_GENERIC_MAX - 1)
+
+enum ctattr_timeout_tcp {
+    CTA_TIMEOUT_TCP_UNSPEC,
+    CTA_TIMEOUT_TCP_SYN_SENT,
+    CTA_TIMEOUT_TCP_SYN_RECV,
+    CTA_TIMEOUT_TCP_ESTABLISHED,
+    CTA_TIMEOUT_TCP_FIN_WAIT,
+    CTA_TIMEOUT_TCP_CLOSE_WAIT,
+    CTA_TIMEOUT_TCP_LAST_ACK,
+    CTA_TIMEOUT_TCP_TIME_WAIT,
+    CTA_TIMEOUT_TCP_CLOSE,
+    CTA_TIMEOUT_TCP_SYN_SENT2,
+    CTA_TIMEOUT_TCP_RETRANS,
+    CTA_TIMEOUT_TCP_UNACK,
+    __CTA_TIMEOUT_TCP_MAX
+};
+#define CTA_TIMEOUT_TCP_MAX (__CTA_TIMEOUT_TCP_MAX - 1)
+
+enum ctattr_timeout_udp {
+    CTA_TIMEOUT_UDP_UNSPEC,
+    CTA_TIMEOUT_UDP_UNREPLIED,
+    CTA_TIMEOUT_UDP_REPLIED,
+    __CTA_TIMEOUT_UDP_MAX
+};
+#define CTA_TIMEOUT_UDP_MAX (__CTA_TIMEOUT_UDP_MAX - 1)
+
+enum ctattr_timeout_udplite {
+    CTA_TIMEOUT_UDPLITE_UNSPEC,
+    CTA_TIMEOUT_UDPLITE_UNREPLIED,
+    CTA_TIMEOUT_UDPLITE_REPLIED,
+    __CTA_TIMEOUT_UDPLITE_MAX
+};
+#define CTA_TIMEOUT_UDPLITE_MAX (__CTA_TIMEOUT_UDPLITE_MAX - 1)
+
+enum ctattr_timeout_icmp {
+    CTA_TIMEOUT_ICMP_UNSPEC,
+    CTA_TIMEOUT_ICMP_TIMEOUT,
+    __CTA_TIMEOUT_ICMP_MAX
+};
+#define CTA_TIMEOUT_ICMP_MAX (__CTA_TIMEOUT_ICMP_MAX - 1)
+
+enum ctattr_timeout_dccp {
+    CTA_TIMEOUT_DCCP_UNSPEC,
+    CTA_TIMEOUT_DCCP_REQUEST,
+    CTA_TIMEOUT_DCCP_RESPOND,
+    CTA_TIMEOUT_DCCP_PARTOPEN,
+    CTA_TIMEOUT_DCCP_OPEN,
+    CTA_TIMEOUT_DCCP_CLOSEREQ,
+    CTA_TIMEOUT_DCCP_CLOSING,
+    CTA_TIMEOUT_DCCP_TIMEWAIT,
+    __CTA_TIMEOUT_DCCP_MAX
+};
+#define CTA_TIMEOUT_DCCP_MAX (__CTA_TIMEOUT_DCCP_MAX - 1)
+
+enum ctattr_timeout_sctp {
+    CTA_TIMEOUT_SCTP_UNSPEC,
+    CTA_TIMEOUT_SCTP_CLOSED,
+    CTA_TIMEOUT_SCTP_COOKIE_WAIT,
+    CTA_TIMEOUT_SCTP_COOKIE_ECHOED,
+    CTA_TIMEOUT_SCTP_ESTABLISHED,
+    CTA_TIMEOUT_SCTP_SHUTDOWN_SENT,
+    CTA_TIMEOUT_SCTP_SHUTDOWN_RECD,
+    CTA_TIMEOUT_SCTP_SHUTDOWN_ACK_SENT,
+    CTA_TIMEOUT_SCTP_HEARTBEAT_SENT,
+    CTA_TIMEOUT_SCTP_HEARTBEAT_ACKED,
+    __CTA_TIMEOUT_SCTP_MAX
+};
+#define CTA_TIMEOUT_SCTP_MAX (__CTA_TIMEOUT_SCTP_MAX - 1)
+
+enum ctattr_timeout_icmpv6 {
+    CTA_TIMEOUT_ICMPV6_UNSPEC,
+    CTA_TIMEOUT_ICMPV6_TIMEOUT,
+    __CTA_TIMEOUT_ICMPV6_MAX
+};
+#define CTA_TIMEOUT_ICMPV6_MAX (__CTA_TIMEOUT_ICMPV6_MAX - 1)
+
+enum ctattr_timeout_gre {
+    CTA_TIMEOUT_GRE_UNSPEC,
+    CTA_TIMEOUT_GRE_UNREPLIED,
+    CTA_TIMEOUT_GRE_REPLIED,
+    __CTA_TIMEOUT_GRE_MAX
+};
+#define CTA_TIMEOUT_GRE_MAX (__CTA_TIMEOUT_GRE_MAX - 1)
+
+#define CTNL_TIMEOUT_NAME_MAX	32
+
 #endif /* __OVS_DP_INTERFACE_CT_EXT_H_ */
diff --git a/datapath-windows/ovsext/Netlink/NetlinkProto.h b/datapath-windows/ovsext/Netlink/NetlinkProto.h
index 59b56565c1dc..db1fa2bacae8 100644
--- a/datapath-windows/ovsext/Netlink/NetlinkProto.h
+++ b/datapath-windows/ovsext/Netlink/NetlinkProto.h
@@ -51,6 +51,7 @@ 
 #define NLM_F_ECHO              0x008
 
 #define NLM_F_ROOT              0x100
+#define NLM_F_REPLACE           0x100
 #define NLM_F_MATCH             0x200
 #define NLM_F_EXCL              0x200
 #define NLM_F_ATOMIC            0x400
diff --git a/include/windows/automake.mk b/include/windows/automake.mk
index 382627b51787..883bbbf5d97c 100644
--- a/include/windows/automake.mk
+++ b/include/windows/automake.mk
@@ -15,6 +15,7 @@  noinst_HEADERS += \
 	include/windows/linux/netfilter/nf_conntrack_tcp.h \
 	include/windows/linux/netfilter/nfnetlink.h \
 	include/windows/linux/netfilter/nfnetlink_conntrack.h \
+	include/windows/linux/netfilter/nfnetlink_cttimeout.h \
 	include/windows/linux/pkt_sched.h \
 	include/windows/linux/types.h \
 	include/windows/net/if.h \
diff --git a/include/windows/linux/netfilter/nfnetlink_cttimeout.h b/include/windows/linux/netfilter/nfnetlink_cttimeout.h
new file mode 100644
index 000000000000..e69de29bb2d1
diff --git a/lib/ct-dpif.c b/lib/ct-dpif.c
index 6ea7feb0ee35..7f9ce0a561f7 100644
--- a/lib/ct-dpif.c
+++ b/lib/ct-dpif.c
@@ -760,3 +760,107 @@  ct_dpif_format_zone_limits(uint32_t default_limit,
         ds_put_format(ds, ",count=%"PRIu32, zone_limit->count);
     }
 }
+
+static const char *const ct_dpif_tp_attr_string[] = {
+#define CT_DPIF_TP_TCP_ATTR(ATTR) \
+    [CT_DPIF_TP_ATTR_TCP_##ATTR] = "TCP_"#ATTR,
+    CT_DPIF_TP_TCP_ATTRS
+#undef CT_DPIF_TP_TCP_ATTR
+#define CT_DPIF_TP_UDP_ATTR(ATTR) \
+    [CT_DPIF_TP_ATTR_UDP_##ATTR] = "UDP_"#ATTR,
+    CT_DPIF_TP_UDP_ATTRS
+#undef CT_DPIF_TP_UDP_ATTR
+#define CT_DPIF_TP_ICMP_ATTR(ATTR) \
+    [CT_DPIF_TP_ATTR_ICMP_##ATTR] = "ICMP_"#ATTR,
+    CT_DPIF_TP_ICMP_ATTRS
+#undef CT_DPIF_TP_ICMP_ATTR
+};
+
+static bool
+ct_dpif_set_timeout_policy_attr(struct ct_dpif_timeout_policy *tp,
+                                uint32_t attr, uint32_t value)
+{
+    if (tp->present & (1 << attr) && tp->attrs[attr] == value) {
+        return false;
+    }
+    tp->attrs[attr] = value;
+    tp->present |= 1 << attr;
+    return true;
+}
+
+/* Sets a timeout value identified by '*name' to 'value'.
+ * Returns true if the attribute is changed */
+bool
+ct_dpif_set_timeout_policy_attr_by_name(struct ct_dpif_timeout_policy *tp,
+                                        const char *name, uint32_t value)
+{
+    uint32_t i;
+
+    for (i = 0; i < CT_DPIF_TP_ATTR_MAX; ++i) {
+        if (!strcasecmp(name, ct_dpif_tp_attr_string[i])) {
+            return ct_dpif_set_timeout_policy_attr(tp, i, value);
+        }
+    }
+    return false;
+}
+
+bool
+ct_dpif_timeout_policy_support_ipproto(uint8_t ipproto)
+{
+    if (ipproto == IPPROTO_TCP || ipproto == IPPROTO_UDP ||
+        ipproto == IPPROTO_ICMP || ipproto == IPPROTO_ICMPV6) {
+        return true;
+    }
+    return false;
+}
+
+int
+ct_dpif_set_timeout_policy(struct dpif *dpif,
+                           const struct ct_dpif_timeout_policy *tp)
+{
+    return (dpif->dpif_class->ct_set_timeout_policy
+            ? dpif->dpif_class->ct_set_timeout_policy(dpif, tp)
+            : EOPNOTSUPP);
+}
+
+int
+ct_dpif_del_timeout_policy(struct dpif *dpif, uint32_t tp_id)
+{
+    return (dpif->dpif_class->ct_del_timeout_policy
+            ? dpif->dpif_class->ct_del_timeout_policy(dpif, tp_id)
+            : EOPNOTSUPP);
+}
+
+int
+ct_dpif_get_timeout_policy(struct dpif *dpif, uint32_t tp_id,
+                           struct ct_dpif_timeout_policy *tp)
+{
+    return (dpif->dpif_class->ct_get_timeout_policy
+            ? dpif->dpif_class->ct_get_timeout_policy(
+                dpif, tp_id, tp) : EOPNOTSUPP);
+}
+
+int
+ct_dpif_timeout_policy_dump_start(struct dpif *dpif, void **statep)
+{
+    return (dpif->dpif_class->ct_timeout_policy_dump_start
+            ? dpif->dpif_class->ct_timeout_policy_dump_start(dpif, statep)
+            : EOPNOTSUPP);
+}
+
+int
+ct_dpif_timeout_policy_dump_next(struct dpif *dpif, void *state,
+                                 struct ct_dpif_timeout_policy *tp)
+{
+    return (dpif->dpif_class->ct_timeout_policy_dump_next
+            ? dpif->dpif_class->ct_timeout_policy_dump_next(dpif, state, tp)
+            : EOPNOTSUPP);
+}
+
+int
+ct_dpif_timeout_policy_dump_done(struct dpif *dpif, void *state)
+{
+    return (dpif->dpif_class->ct_timeout_policy_dump_done
+            ? dpif->dpif_class->ct_timeout_policy_dump_done(dpif, state)
+            : EOPNOTSUPP);
+}
diff --git a/lib/ct-dpif.h b/lib/ct-dpif.h
index 2f4906817946..8dacb1c7c253 100644
--- a/lib/ct-dpif.h
+++ b/lib/ct-dpif.h
@@ -225,6 +225,50 @@  struct ct_dpif_zone_limit {
     struct ovs_list node;
 };
 
+#define CT_DPIF_TP_TCP_ATTRS \
+    CT_DPIF_TP_TCP_ATTR(SYN_SENT) \
+    CT_DPIF_TP_TCP_ATTR(SYN_RECV) \
+    CT_DPIF_TP_TCP_ATTR(ESTABLISHED) \
+    CT_DPIF_TP_TCP_ATTR(FIN_WAIT) \
+    CT_DPIF_TP_TCP_ATTR(CLOSE_WAIT) \
+    CT_DPIF_TP_TCP_ATTR(LAST_ACK) \
+    CT_DPIF_TP_TCP_ATTR(TIME_WAIT) \
+    CT_DPIF_TP_TCP_ATTR(CLOSE) \
+    CT_DPIF_TP_TCP_ATTR(SYN_SENT2) \
+    CT_DPIF_TP_TCP_ATTR(RETRANSMIT) \
+    CT_DPIF_TP_TCP_ATTR(UNACK)
+
+#define CT_DPIF_TP_UDP_ATTRS \
+    CT_DPIF_TP_UDP_ATTR(FIRST) \
+    CT_DPIF_TP_UDP_ATTR(SINGLE) \
+    CT_DPIF_TP_UDP_ATTR(MULTIPLE)
+
+#define CT_DPIF_TP_ICMP_ATTRS \
+    CT_DPIF_TP_ICMP_ATTR(FIRST) \
+    CT_DPIF_TP_ICMP_ATTR(REPLY)
+
+enum OVS_PACKED_ENUM ct_dpif_tp_attr {
+#define CT_DPIF_TP_TCP_ATTR(ATTR) CT_DPIF_TP_ATTR_TCP_##ATTR,
+    CT_DPIF_TP_TCP_ATTRS
+#undef CT_DPIF_TP_TCP_ATTR
+#define CT_DPIF_TP_UDP_ATTR(ATTR) CT_DPIF_TP_ATTR_UDP_##ATTR,
+    CT_DPIF_TP_UDP_ATTRS
+#undef CT_DPIF_TP_UDP_ATTR
+#define CT_DPIF_TP_ICMP_ATTR(ATTR) CT_DPIF_TP_ATTR_ICMP_##ATTR,
+    CT_DPIF_TP_ICMP_ATTRS
+#undef CT_DPIF_TP_ICMP_ATTR
+    CT_DPIF_TP_ATTR_MAX
+};
+
+struct ct_dpif_timeout_policy {
+    uint32_t    id;         /* An unique identifier of a timeout policy in
+                             * the datapath. */
+    uint32_t    present;    /* If a timeout attribute is present set the
+                             * corresponding bit. */
+    uint32_t    attrs[CT_DPIF_TP_ATTR_MAX];     /* An array that specifies
+                                                 * timeout attribute values */
+};
+
 int ct_dpif_dump_start(struct dpif *, struct ct_dpif_dump_state **,
                        const uint16_t *zone, int *);
 int ct_dpif_dump_next(struct ct_dpif_dump_state *, struct ct_dpif_entry *);
@@ -262,5 +306,17 @@  bool ct_dpif_parse_zone_limit_tuple(const char *s, uint16_t *pzone,
                                     uint32_t *plimit, struct ds *);
 void ct_dpif_format_zone_limits(uint32_t default_limit,
                                 const struct ovs_list *, struct ds *);
+bool ct_dpif_set_timeout_policy_attr_by_name(struct ct_dpif_timeout_policy *tp,
+                                             const char *key, uint32_t value);
+bool ct_dpif_timeout_policy_support_ipproto(uint8_t ipproto);
+int ct_dpif_set_timeout_policy(struct dpif *dpif,
+                               const struct ct_dpif_timeout_policy *tp);
+int ct_dpif_get_timeout_policy(struct dpif *dpif, uint32_t tp_id,
+                               struct ct_dpif_timeout_policy *tp);
+int ct_dpif_del_timeout_policy(struct dpif *dpif, uint32_t tp_id);
+int ct_dpif_timeout_policy_dump_start(struct dpif *dpif, void **statep);
+int ct_dpif_timeout_policy_dump_next(struct dpif *dpif, void *state,
+                                     struct ct_dpif_timeout_policy *tp);
+int ct_dpif_timeout_policy_dump_done(struct dpif *dpif, void *state);
 
 #endif /* CT_DPIF_H */
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index d0a1c58adace..2079e368fb52 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -7529,6 +7529,12 @@  const struct dpif_class dpif_netdev_class = {
     NULL,                       /* ct_set_limits */
     NULL,                       /* ct_get_limits */
     NULL,                       /* ct_del_limits */
+    NULL,                       /* ct_set_timeout_policy */
+    NULL,                       /* ct_get_timeout_policy */
+    NULL,                       /* ct_del_timeout_policy */
+    NULL,                       /* ct_timeout_policy_dump_start */
+    NULL,                       /* ct_timeout_policy_dump_next */
+    NULL,                       /* ct_timeout_policy_dump_done */
     dpif_netdev_ipf_set_enabled,
     dpif_netdev_ipf_set_min_frag,
     dpif_netdev_ipf_set_max_nfrags,
diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
index 7bc71d6d19d7..b859508f718a 100644
--- a/lib/dpif-netlink.c
+++ b/lib/dpif-netlink.c
@@ -50,6 +50,7 @@ 
 #include "odp-util.h"
 #include "openvswitch/dynamic-string.h"
 #include "openvswitch/flow.h"
+#include "openvswitch/hmap.h"
 #include "openvswitch/match.h"
 #include "openvswitch/ofpbuf.h"
 #include "openvswitch/poll-loop.h"
@@ -3023,6 +3024,461 @@  dpif_netlink_ct_del_limits(struct dpif *dpif OVS_UNUSED,
     ofpbuf_delete(request);
     return err;
 }
+
+#define NL_TP_NAME_PREFIX "ovs_tp_"
+
+struct dpif_netlink_timeout_policy_protocol {
+    uint16_t    l3num;
+    uint8_t     l4num;
+};
+
+enum OVS_PACKED_ENUM dpif_netlink_support_timeout_policy_protocol {
+    DPIF_NL_TP_AF_INET_TCP,
+    DPIF_NL_TP_AF_INET_UDP,
+    DPIF_NL_TP_AF_INET_ICMP,
+    DPIF_NL_TP_AF_INET6_TCP,
+    DPIF_NL_TP_AF_INET6_UDP,
+    DPIF_NL_TP_AF_INET6_ICMPV6,
+    DPIF_NL_TP_MAX
+};
+
+#define DPIF_NL_ALL_TP 0x3F
+
+static struct dpif_netlink_timeout_policy_protocol tp_protos[] = {
+    [DPIF_NL_TP_AF_INET_TCP] = { .l3num = AF_INET, .l4num = IPPROTO_TCP },
+    [DPIF_NL_TP_AF_INET_UDP] = { .l3num = AF_INET, .l4num = IPPROTO_UDP },
+    [DPIF_NL_TP_AF_INET_ICMP] = { .l3num = AF_INET, .l4num = IPPROTO_ICMP },
+    [DPIF_NL_TP_AF_INET6_TCP] = { .l3num = AF_INET6, .l4num = IPPROTO_TCP },
+    [DPIF_NL_TP_AF_INET6_UDP] = { .l3num = AF_INET6, .l4num = IPPROTO_UDP },
+    [DPIF_NL_TP_AF_INET6_ICMPV6] = { .l3num = AF_INET6,
+                                     .l4num = IPPROTO_ICMPV6 },
+};
+
+static void
+dpif_netlink_format_tp_name(uint32_t id, uint16_t l3num, uint8_t l4num,
+                            struct ds *tp_name)
+{
+    ds_clear(tp_name);
+    ds_put_format(tp_name, "%s%"PRIu32"_", NL_TP_NAME_PREFIX, id);
+    ct_dpif_format_ipproto(tp_name, l4num);
+
+    if (l3num == AF_INET) {
+        ds_put_cstr(tp_name, "4");
+    } else if (l3num == AF_INET6 && l4num != IPPROTO_ICMPV6) {
+        ds_put_cstr(tp_name, "6");
+    }
+
+    ovs_assert(tp_name->length < CTNL_TIMEOUT_NAME_MAX);
+}
+
+#define CT_DPIF_NL_TP_TCP_MAPPINGS                              \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT, SYN_SENT)         \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_RECV, SYN_RECV)         \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, ESTABLISHED, ESTABLISHED)   \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, FIN_WAIT, FIN_WAIT)         \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE_WAIT, CLOSE_WAIT)     \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, LAST_ACK, LAST_ACK)         \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, TIME_WAIT, TIME_WAIT)       \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, CLOSE, CLOSE)               \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, SYN_SENT2, SYN_SENT2)       \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, RETRANSMIT, RETRANS)        \
+    CT_DPIF_NL_TP_MAPPING(TCP, TCP, UNACK, UNACK)
+
+#define CT_DPIF_NL_TP_UDP_MAPPINGS                              \
+    CT_DPIF_NL_TP_MAPPING(UDP, UDP, SINGLE, UNREPLIED)          \
+    CT_DPIF_NL_TP_MAPPING(UDP, UDP, MULTIPLE, REPLIED)
+
+#define CT_DPIF_NL_TP_ICMP_MAPPINGS                             \
+    CT_DPIF_NL_TP_MAPPING(ICMP, ICMP, FIRST, TIMEOUT)
+
+#define CT_DPIF_NL_TP_ICMPV6_MAPPINGS                           \
+    CT_DPIF_NL_TP_MAPPING(ICMP, ICMPV6, FIRST, TIMEOUT)
+
+
+#define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2)     \
+if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) {  \
+    nl_tp->present |= 1 << CTA_TIMEOUT_##PROTO2##_##ATTR2;      \
+    nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2] =              \
+        tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1];          \
+}
+
+static void
+dpif_netlink_get_nl_tp_tcp_attrs(const struct ct_dpif_timeout_policy *tp,
+                                 struct nl_ct_timeout_policy *nl_tp)
+{
+    CT_DPIF_NL_TP_TCP_MAPPINGS
+}
+
+static void
+dpif_netlink_get_nl_tp_udp_attrs(const struct ct_dpif_timeout_policy *tp,
+                                 struct nl_ct_timeout_policy *nl_tp)
+{
+    CT_DPIF_NL_TP_UDP_MAPPINGS
+}
+
+static void
+dpif_netlink_get_nl_tp_icmp_attrs(const struct ct_dpif_timeout_policy *tp,
+                                  struct nl_ct_timeout_policy *nl_tp)
+{
+    CT_DPIF_NL_TP_ICMP_MAPPINGS
+}
+
+static void
+dpif_netlink_get_nl_tp_icmpv6_attrs(const struct ct_dpif_timeout_policy *tp,
+                                    struct nl_ct_timeout_policy *nl_tp)
+{
+    CT_DPIF_NL_TP_ICMPV6_MAPPINGS
+}
+
+#undef CT_DPIF_NL_TP_MAPPING
+
+static void
+dpif_netlink_get_nl_tp_attrs(const struct ct_dpif_timeout_policy *tp,
+                             uint8_t l4num, struct nl_ct_timeout_policy *nl_tp)
+{
+    nl_tp->present = 0;
+
+    if (l4num == IPPROTO_TCP) {
+        dpif_netlink_get_nl_tp_tcp_attrs(tp, nl_tp);
+    } else if (l4num == IPPROTO_UDP) {
+        dpif_netlink_get_nl_tp_udp_attrs(tp, nl_tp);
+    } else if (l4num == IPPROTO_ICMP) {
+        dpif_netlink_get_nl_tp_icmp_attrs(tp, nl_tp);
+    } else if (l4num == IPPROTO_ICMPV6) {
+        dpif_netlink_get_nl_tp_icmpv6_attrs(tp, nl_tp);
+    }
+}
+
+#define CT_DPIF_NL_TP_MAPPING(PROTO1, PROTO2, ATTR1, ATTR2)                 \
+if (nl_tp->present & (1 << CTA_TIMEOUT_##PROTO2##_##ATTR2)) {               \
+    if (tp->present & (1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1)) {          \
+        if (tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] !=                \
+            nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2]) {                 \
+            VLOG_WARN_RL(&error_rl, "Inconsistent timeout policy %s "       \
+                         "attribute %s=%"PRIu32" while %s=%"PRIu32,         \
+                         nl_tp->name, "CTA_TIMEOUT_"#PROTO2"_"#ATTR2,       \
+                         nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2],      \
+                         "CT_DPIF_TP_ATTR_"#PROTO1"_"#ATTR1,                \
+                         tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1]);    \
+        }                                                                   \
+    } else {                                                                \
+        tp->present |= 1 << CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1;             \
+        tp->attrs[CT_DPIF_TP_ATTR_##PROTO1##_##ATTR1] =                     \
+            nl_tp->attrs[CTA_TIMEOUT_##PROTO2##_##ATTR2];                   \
+    }                                                                       \
+}
+
+static void
+dpif_netlink_set_ct_dpif_tp_tcp_attrs(const struct nl_ct_timeout_policy *nl_tp,
+                                      struct ct_dpif_timeout_policy *tp)
+{
+    CT_DPIF_NL_TP_TCP_MAPPINGS
+}
+
+static void
+dpif_netlink_set_ct_dpif_tp_udp_attrs(const struct nl_ct_timeout_policy *nl_tp,
+                                      struct ct_dpif_timeout_policy *tp)
+{
+    CT_DPIF_NL_TP_UDP_MAPPINGS
+}
+
+static void
+dpif_netlink_set_ct_dpif_tp_icmp_attrs(
+    const struct nl_ct_timeout_policy *nl_tp,
+    struct ct_dpif_timeout_policy *tp)
+{
+    CT_DPIF_NL_TP_ICMP_MAPPINGS
+}
+
+static void
+dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(
+    const struct nl_ct_timeout_policy *nl_tp,
+    struct ct_dpif_timeout_policy *tp)
+{
+    CT_DPIF_NL_TP_ICMPV6_MAPPINGS
+}
+
+#undef CT_DPIF_NL_TP_MAPPING
+
+static void
+dpif_netlink_set_ct_dpif_tp_attrs(const struct nl_ct_timeout_policy *nl_tp,
+                               struct ct_dpif_timeout_policy *tp)
+{
+    if (nl_tp->l4num == IPPROTO_TCP) {
+        dpif_netlink_set_ct_dpif_tp_tcp_attrs(nl_tp, tp);
+    } else if (nl_tp->l4num == IPPROTO_UDP) {
+        dpif_netlink_set_ct_dpif_tp_udp_attrs(nl_tp, tp);
+    } else if (nl_tp->l4num == IPPROTO_ICMP) {
+        dpif_netlink_set_ct_dpif_tp_icmp_attrs(nl_tp, tp);
+    } else if (nl_tp->l4num == IPPROTO_ICMPV6) {
+        dpif_netlink_set_ct_dpif_tp_icmpv6_attrs(nl_tp, tp);
+    }
+}
+
+#ifdef _WIN32
+static int
+dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   const struct ct_dpif_timeout_policy *tp)
+{
+    return EOPNOTSUPP;
+}
+
+static int
+dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   uint32_t tp_id,
+                                   struct ct_dpif_timeout_policy *tp)
+{
+    return EOPNOTSUPP;
+}
+
+static int
+dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   uint32_t tp_id)
+{
+    return EOPNOTSUPP;
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED,
+                                          void **statep)
+{
+    return EOPNOTSUPP;
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
+                                         void *state,
+                                         struct ct_dpif_timeout_policy **tp)
+{
+    return EOPNOTSUPP;
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED,
+                                         void *state)
+{
+    return EOPNOTSUPP;
+}
+#else
+static int
+dpif_netlink_ct_set_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   const struct ct_dpif_timeout_policy *tp)
+{
+    struct nl_ct_timeout_policy nl_tp;
+    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
+    int i, err = 0;
+
+    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
+        dpif_netlink_format_tp_name(tp->id, tp_protos[i].l3num,
+                                    tp_protos[i].l4num, &nl_tp_name);
+        ovs_strlcpy(nl_tp.name, ds_cstr(&nl_tp_name), sizeof nl_tp.name);
+        nl_tp.l3num = tp_protos[i].l3num;
+        nl_tp.l4num = tp_protos[i].l4num;
+        dpif_netlink_get_nl_tp_attrs(tp, tp_protos[i].l4num, &nl_tp);
+        err = nl_ct_set_timeout_policy(&nl_tp);
+        if (err) {
+            VLOG_WARN_RL(&error_rl, "failed to add timeout policy %s (%s)",
+                         nl_tp.name, ovs_strerror(err));
+            goto out;
+        }
+    }
+
+out:
+    ds_destroy(&nl_tp_name);
+    return  err;
+}
+
+static int
+dpif_netlink_ct_get_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   uint32_t tp_id,
+                                   struct ct_dpif_timeout_policy *tp)
+{
+    struct nl_ct_timeout_policy nl_tp;
+    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
+    int i, err = 0;
+
+    tp->id = tp_id;
+    tp->present = 0;
+    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
+        dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num,
+                                    tp_protos[i].l4num, &nl_tp_name);
+        err = nl_ct_get_timeout_policy(ds_cstr(&nl_tp_name), &nl_tp);
+
+        if (err) {
+            VLOG_WARN_RL(&error_rl, "failed to get timeout policy %s (%s)",
+                         nl_tp.name, ovs_strerror(err));
+            goto out;
+        }
+        dpif_netlink_set_ct_dpif_tp_attrs(&nl_tp, tp);
+    }
+
+out:
+    ds_destroy(&nl_tp_name);
+    return  err;
+}
+
+/* Returns 0 if all the sub timeout policies are deleted or
+ * not exist in the kernel. */
+static int
+dpif_netlink_ct_del_timeout_policy(struct dpif *dpif OVS_UNUSED,
+                                   uint32_t tp_id)
+{
+    struct ds nl_tp_name = DS_EMPTY_INITIALIZER;
+    int i, err = 0;
+
+    for (i = 0; i < ARRAY_SIZE(tp_protos); ++i) {
+        dpif_netlink_format_tp_name(tp_id, tp_protos[i].l3num,
+                                    tp_protos[i].l4num, &nl_tp_name);
+        err = nl_ct_del_timeout_policy(ds_cstr(&nl_tp_name));
+        if (err == ENOENT) {
+            err = 0;
+        }
+        if (err) {
+            VLOG_WARN_RL(&error_rl, "failed to delete timeout policy %s (%s)",
+                         ds_cstr(&nl_tp_name), ovs_strerror(err));
+            goto out;
+        }
+    }
+
+out:
+    ds_destroy(&nl_tp_name);
+    return err;
+}
+
+struct dpif_netlink_ct_timeout_policy_dump_state {
+    struct nl_ct_timeout_policy_dump_state *nl_dump_state;
+    struct hmap tp_dump_map;
+};
+
+struct dpif_netlink_tp_dump_node {
+    struct      hmap_node hmap_node;      /* node in tp_dump_map. */
+    struct      ct_dpif_timeout_policy *tp;
+    uint32_t    present;
+};
+
+static struct dpif_netlink_tp_dump_node *
+get_dpif_netlink_tp_dump_node_by_tp_id(uint32_t tp_id,
+                                       struct hmap *tp_dump_map)
+{
+    struct dpif_netlink_tp_dump_node *tp_dump_node;
+
+    HMAP_FOR_EACH_WITH_HASH (tp_dump_node, hmap_node, hash_int(tp_id, 0),
+                            tp_dump_map) {
+        if (tp_dump_node->tp->id == tp_id) {
+            return tp_dump_node;
+        }
+    }
+    return NULL;
+}
+
+static void
+update_dpif_netlink_tp_dump_node(
+    const struct nl_ct_timeout_policy *nl_tp,
+    struct dpif_netlink_tp_dump_node *tp_dump_node)
+{
+    int i;
+
+    dpif_netlink_set_ct_dpif_tp_attrs(nl_tp, tp_dump_node->tp);
+    for (i = 0; i < DPIF_NL_TP_MAX; ++i) {
+        if (nl_tp->l3num == tp_protos[i].l3num &&
+            nl_tp->l4num == tp_protos[i].l4num) {
+            tp_dump_node->present |= 1 << i;
+            break;
+        }
+    }
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_start(struct dpif *dpif OVS_UNUSED,
+                                          void **statep)
+{
+    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state;
+    int err;
+
+    *statep = dump_state = xzalloc(sizeof *dump_state);
+    err = nl_ct_timeout_policy_dump_start(&dump_state->nl_dump_state);
+    if (err) {
+        free(dump_state);
+        return err;
+    }
+    hmap_init(&dump_state->tp_dump_map);
+    return 0;
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_next(struct dpif *dpif OVS_UNUSED,
+                                         void *state,
+                                         struct ct_dpif_timeout_policy *tp)
+{
+    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
+    struct dpif_netlink_tp_dump_node *tp_dump_node;
+    struct nl_ct_timeout_policy nl_tp;
+    uint32_t tp_id;
+    int err;
+
+    do {
+        err =  nl_ct_timeout_policy_dump_next(dump_state->nl_dump_state,
+                                              &nl_tp);
+        if (err) {
+            break;
+        }
+
+        if (!ovs_scan(nl_tp.name, NL_TP_NAME_PREFIX"%"PRIu32, &tp_id)) {
+            continue;
+        }
+
+        tp_dump_node = get_dpif_netlink_tp_dump_node_by_tp_id(
+                            tp_id, &dump_state->tp_dump_map);
+        if (!tp_dump_node) {
+            tp_dump_node = xzalloc(sizeof *tp_dump_node);
+            tp_dump_node->tp = xzalloc(sizeof *tp_dump_node->tp);
+            tp_dump_node->tp->id = tp_id;
+            hmap_insert(&dump_state->tp_dump_map, &tp_dump_node->hmap_node,
+                        hash_int(tp_id, 0));
+        }
+
+        update_dpif_netlink_tp_dump_node(&nl_tp, tp_dump_node);
+        if (tp_dump_node->present == DPIF_NL_ALL_TP) {
+            hmap_remove(&dump_state->tp_dump_map, &tp_dump_node->hmap_node);
+            *tp = *tp_dump_node->tp;
+            free(tp_dump_node->tp);
+            free(tp_dump_node);
+            break;
+        }
+    } while (true);
+
+    /* Dump the incomplete timeout policy. */
+    if (err == EOF) {
+        if (!hmap_is_empty(&dump_state->tp_dump_map)) {
+            struct hmap_node *hmap_node = hmap_first(&dump_state->tp_dump_map);
+
+            hmap_remove(&dump_state->tp_dump_map, hmap_node);
+            tp_dump_node = CONTAINER_OF(hmap_node,
+                            struct dpif_netlink_tp_dump_node, hmap_node);
+            *tp = *tp_dump_node->tp;
+            free(tp_dump_node->tp);
+            free(tp_dump_node);
+            return 0;
+        }
+    }
+
+    return err;
+}
+
+static int
+dpif_netlink_ct_timeout_policy_dump_done(struct dpif *dpif OVS_UNUSED,
+                                         void *state)
+{
+    struct dpif_netlink_ct_timeout_policy_dump_state *dump_state = state;
+    int err;
+
+    err = nl_ct_timeout_policy_dump_done(dump_state->nl_dump_state);
+    hmap_destroy(&dump_state->tp_dump_map);
+    free(dump_state);
+    return err;
+}
+#endif
+
 
 /* Meters */
 
@@ -3429,6 +3885,12 @@  const struct dpif_class dpif_netlink_class = {
     dpif_netlink_ct_set_limits,
     dpif_netlink_ct_get_limits,
     dpif_netlink_ct_del_limits,
+    dpif_netlink_ct_set_timeout_policy,
+    dpif_netlink_ct_get_timeout_policy,
+    dpif_netlink_ct_del_timeout_policy,
+    dpif_netlink_ct_timeout_policy_dump_start,
+    dpif_netlink_ct_timeout_policy_dump_next,
+    dpif_netlink_ct_timeout_policy_dump_done,
     NULL,                       /* ipf_set_enabled */
     NULL,                       /* ipf_set_min_frag */
     NULL,                       /* ipf_set_max_nfrags */
diff --git a/lib/dpif-netlink.h b/lib/dpif-netlink.h
index 0a9628088275..1e753994d747 100644
--- a/lib/dpif-netlink.h
+++ b/lib/dpif-netlink.h
@@ -23,6 +23,7 @@ 
 #include "odp-netlink.h"
 
 #include "flow.h"
+#include "netlink-conntrack.h"
 
 struct ofpbuf;
 
diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h
index 12898b9e3c6d..79a2314500cf 100644
--- a/lib/dpif-provider.h
+++ b/lib/dpif-provider.h
@@ -80,6 +80,7 @@  dpif_flow_dump_thread_init(struct dpif_flow_dump_thread *thread,
 struct ct_dpif_dump_state;
 struct ct_dpif_entry;
 struct ct_dpif_tuple;
+struct ct_dpif_timeout_policy;
 
 /* 'dpif_ipf_proto_status' and 'dpif_ipf_status' are presently in
  * sync with 'ipf_proto_status' and 'ipf_status', but more
@@ -498,6 +499,43 @@  struct dpif_class {
      * list of 'struct ct_dpif_zone_limit' entries. */
     int (*ct_del_limits)(struct dpif *, const struct ovs_list *zone_limits);
 
+    /* Connection tracking timeout policy */
+
+    /* A connection tracking timeout policy contains a list of timeout
+     * attributes that specifies timeout values on various connection states.
+     * In a datapath, the timeout policy is identified by a 4 bytes unsigned
+     * integer, and the unsupported timeout attributes are ignored.
+     * When a connection is committed it can be associated with a timeout
+     * policy, or it defaults to the datapath's default timeout policy. */
+
+    /* Sets timeout policy '*tp' into the datapath. */
+    int (*ct_set_timeout_policy)(struct dpif *,
+                                 const struct ct_dpif_timeout_policy *tp);
+    /* Gets a timeout policy specified by tp_id and stores it into '*tp'.*/
+    int (*ct_get_timeout_policy)(struct dpif *, uint32_t tp_id,
+                                 struct ct_dpif_timeout_policy *tp);
+    /* Deletes a timeout policy identified by 'tp_id'. */
+    int (*ct_del_timeout_policy)(struct dpif *, uint32_t tp_id);
+
+    /* Conntrack timeout policy dumping interface.
+     *
+     * These functions provide a datapath-agnostic dumping interface
+     * to the conntrack timeout policy provided by the datapaths.
+     *
+     * ct_timeout_policy_dump_start() should put in '*statep' a pointer to
+     * a newly allocated structure that will be passed by the caller to
+     * ct_timeout_policy_dump_next() and ct_timeout_policy_dump_done().
+     *
+     * ct_timeout_policy_dump_next() fills a timeout policy into '*tp',
+     * and prepares to dump the next one on a subsequent invocation.
+     *
+     * ct_timeout_policy_dump_done() should perform any cleanup necessary
+     * (including deallocating the 'state' structure, if applicable). */
+    int (*ct_timeout_policy_dump_start)(struct dpif *, void **statep);
+    int (*ct_timeout_policy_dump_next)(struct dpif *, void *state,
+                                       struct ct_dpif_timeout_policy *tp);
+    int (*ct_timeout_policy_dump_done)(struct dpif *, void *state);
+
     /* IP Fragmentation. */
 
     /* Disables or enables conntrack fragment reassembly.  The default
diff --git a/lib/netlink-conntrack.c b/lib/netlink-conntrack.c
index 7631ba5d5d31..9bc0ddb66248 100644
--- a/lib/netlink-conntrack.c
+++ b/lib/netlink-conntrack.c
@@ -840,6 +840,369 @@  nl_ct_parse_helper(struct nlattr *nla, struct ct_dpif_helper *helper)
     return parsed;
 }
 
+static int nl_ct_timeout_policy_max_attr[] = {
+    [IPPROTO_TCP] = CTA_TIMEOUT_TCP_MAX,
+    [IPPROTO_UDP] = CTA_TIMEOUT_UDP_MAX,
+    [IPPROTO_ICMP] = CTA_TIMEOUT_ICMP_MAX,
+    [IPPROTO_ICMPV6] = CTA_TIMEOUT_ICMPV6_MAX
+};
+
+static void
+nl_ct_set_timeout_policy_attr(struct nl_ct_timeout_policy *nl_tp,
+                              uint32_t attr, uint32_t val)
+{
+    nl_tp->present |= 1 << attr;
+    nl_tp->attrs[attr] = val;
+}
+
+static int
+nl_ct_parse_tcp_timeout_policy_data(struct nlattr *nla,
+                                    struct nl_ct_timeout_policy *nl_tp)
+{
+    static const struct nl_policy policy[] = {
+        [CTA_TIMEOUT_TCP_SYN_SENT] =    { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_SYN_RECV] =    { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_FIN_WAIT] =    { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_CLOSE_WAIT] =  { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_LAST_ACK] =    { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_TIME_WAIT] =   { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_CLOSE] =       { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_SYN_SENT2] =   { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_RETRANS] =     { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_TCP_UNACK] =       { .type = NL_A_BE32,
+                                          .optional = false },
+    };
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+    int i;
+
+    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
+        VLOG_ERR_RL(&rl, "Could not parse nested tcp timeout options. "
+                    "Possibly incompatible Linux kernel version.");
+        return EINVAL;
+    }
+
+    for (i = CTA_TIMEOUT_TCP_SYN_SENT; i <= CTA_TIMEOUT_TCP_UNACK; i++) {
+        nl_ct_set_timeout_policy_attr(nl_tp, i,
+                                      ntohl(nl_attr_get_be32(attrs[i])));
+    }
+    return 0;
+}
+
+static int
+nl_ct_parse_udp_timeout_policy_data(struct nlattr *nla,
+                                    struct nl_ct_timeout_policy *nl_tp)
+{
+    static const struct nl_policy policy[] = {
+        [CTA_TIMEOUT_UDP_UNREPLIED] =   { .type = NL_A_BE32,
+                                          .optional = false },
+        [CTA_TIMEOUT_UDP_REPLIED] =     { .type = NL_A_BE32,
+                                          .optional = false },
+    };
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+    int i;
+
+    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
+        VLOG_ERR_RL(&rl, "Could not parse nested tcp timeout options. "
+                    "Possibly incompatible Linux kernel version.");
+        return EINVAL;
+    }
+
+    for (i = CTA_TIMEOUT_UDP_UNREPLIED; i <= CTA_TIMEOUT_UDP_REPLIED; i++) {
+        nl_ct_set_timeout_policy_attr(nl_tp, i,
+                                      ntohl(nl_attr_get_be32(attrs[i])));
+    }
+    return 0;
+}
+
+static int
+nl_ct_parse_icmp_timeout_policy_data(struct nlattr *nla,
+                                     struct nl_ct_timeout_policy *nl_tp)
+{
+    static const struct nl_policy policy[] = {
+        [CTA_TIMEOUT_ICMP_TIMEOUT] =   { .type = NL_A_BE32,
+                                         .optional = false },
+    };
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+
+    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
+        VLOG_ERR_RL(&rl, "Could not parse nested icmp timeout options. "
+                    "Possibly incompatible Linux kernel version.");
+        return EINVAL;
+    }
+
+    nl_ct_set_timeout_policy_attr(
+        nl_tp, CTA_TIMEOUT_ICMP_TIMEOUT,
+        ntohl(nl_attr_get_be32(attrs[CTA_TIMEOUT_ICMP_TIMEOUT])));
+    return 0;
+}
+
+static int
+nl_ct_parse_icmpv6_timeout_policy_data(struct nlattr *nla,
+                                       struct nl_ct_timeout_policy *nl_tp)
+{
+    static const struct nl_policy policy[] = {
+        [CTA_TIMEOUT_ICMPV6_TIMEOUT] =   { .type = NL_A_BE32,
+                                           .optional = false },
+    };
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+
+    if (!nl_parse_nested(nla, policy, attrs, ARRAY_SIZE(policy))) {
+        VLOG_ERR_RL(&rl, "Could not parse nested icmpv6 timeout options. "
+                    "Possibly incompatible Linux kernel version.");
+        return EINVAL;
+    }
+
+    nl_ct_set_timeout_policy_attr(
+        nl_tp, CTA_TIMEOUT_ICMPV6_TIMEOUT,
+        ntohl(nl_attr_get_be32(attrs[CTA_TIMEOUT_ICMPV6_TIMEOUT])));
+    return 0;
+}
+
+static int
+nl_ct_parse_timeout_policy_data(struct nlattr *nla,
+                                struct nl_ct_timeout_policy *nl_tp)
+{
+    switch (nl_tp->l4num) {
+        case IPPROTO_TCP:
+            return nl_ct_parse_tcp_timeout_policy_data(nla, nl_tp);
+        case IPPROTO_UDP:
+            return nl_ct_parse_udp_timeout_policy_data(nla, nl_tp);
+        case IPPROTO_ICMP:
+            return nl_ct_parse_icmp_timeout_policy_data(nla, nl_tp);
+        case IPPROTO_ICMPV6:
+            return nl_ct_parse_icmpv6_timeout_policy_data(nla, nl_tp);
+        default:
+            return EINVAL;
+    }
+}
+
+static int
+nl_ct_timeout_policy_from_ofpbuf(struct ofpbuf *buf,
+                                 struct nl_ct_timeout_policy *nl_tp,
+                                 bool default_tp)
+{
+    static const struct nl_policy policy[] = {
+        [CTA_TIMEOUT_NAME] =    { .type = NL_A_STRING, .optional = false },
+        [CTA_TIMEOUT_L3PROTO] = { .type = NL_A_BE16, .optional = false },
+        [CTA_TIMEOUT_L4PROTO] = { .type = NL_A_U8, .optional = false },
+        [CTA_TIMEOUT_DATA] =    { .type = NL_A_NESTED, .optional = false }
+    };
+    static const struct nl_policy policy_default_tp[] = {
+        [CTA_TIMEOUT_L3PROTO] = { .type = NL_A_BE16, .optional = false },
+        [CTA_TIMEOUT_L4PROTO] = { .type = NL_A_U8, .optional = false },
+        [CTA_TIMEOUT_DATA] =    { .type = NL_A_NESTED, .optional = false }
+    };
+
+    struct nlattr *attrs[ARRAY_SIZE(policy)];
+    struct ofpbuf b = ofpbuf_const_initializer(buf->data, buf->size);
+    struct nlmsghdr *nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+    struct nfgenmsg *nfmsg = ofpbuf_try_pull(&b, sizeof *nfmsg);
+    int err;
+
+    if (!nlmsg || !nfmsg
+        || NFNL_SUBSYS_ID(nlmsg->nlmsg_type) != NFNL_SUBSYS_CTNETLINK_TIMEOUT
+        || nfmsg->version != NFNETLINK_V0
+        || !nl_policy_parse(&b, 0, default_tp ? policy_default_tp : policy,
+                            attrs, default_tp ? ARRAY_SIZE(policy_default_tp) :
+                                                ARRAY_SIZE(policy))) {
+        return EINVAL;
+    }
+
+    if (!default_tp) {
+        ovs_strlcpy(nl_tp->name, nl_attr_get_string(attrs[CTA_TIMEOUT_NAME]),
+                    sizeof nl_tp->name);
+    }
+    nl_tp->l3num = ntohs(nl_attr_get_be16(attrs[CTA_TIMEOUT_L3PROTO]));
+    nl_tp->l4num = nl_attr_get_u8(attrs[CTA_TIMEOUT_L4PROTO]);
+    nl_tp->present = 0;
+
+    err = nl_ct_parse_timeout_policy_data(attrs[CTA_TIMEOUT_DATA], nl_tp);
+    return err;
+}
+
+int
+nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
+{
+    struct ofpbuf buf;
+    size_t offset;
+    int i, err;
+
+    ofpbuf_init(&buf, 512);
+    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_NEW, NLM_F_REQUEST | NLM_F_CREATE
+                        | NLM_F_ACK | NLM_F_REPLACE);
+
+    nl_msg_put_string(&buf, CTA_TIMEOUT_NAME, nl_tp->name);
+    nl_msg_put_be16(&buf, CTA_TIMEOUT_L3PROTO, htons(nl_tp->l3num));
+    nl_msg_put_u8(&buf, CTA_TIMEOUT_L4PROTO, nl_tp->l4num);
+
+    offset = nl_msg_start_nested(&buf, CTA_TIMEOUT_DATA);
+    for (i = 1; i <= nl_ct_timeout_policy_max_attr[nl_tp->l4num]; ++i) {
+        if (nl_tp->present & 1 << i) {
+            nl_msg_put_be32(&buf, i, htonl(nl_tp->attrs[i]));
+        }
+    }
+    nl_msg_end_nested(&buf, offset);
+
+    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
+    ofpbuf_uninit(&buf);
+    return err;
+}
+
+int
+nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp)
+{
+    struct ofpbuf buf;
+    size_t offset;
+    int i, err;
+
+    ofpbuf_init(&buf, 512);
+    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_DEFAULT_SET, NLM_F_REQUEST
+                        | NLM_F_ACK | NLM_F_REPLACE);
+
+    nl_msg_put_be16(&buf, CTA_TIMEOUT_L3PROTO, htons(nl_tp->l3num));
+    nl_msg_put_u8(&buf, CTA_TIMEOUT_L4PROTO, nl_tp->l4num);
+
+    offset = nl_msg_start_nested(&buf, CTA_TIMEOUT_DATA);
+    for (i = 1; i <= nl_ct_timeout_policy_max_attr[nl_tp->l4num]; ++i) {
+        if (nl_tp->present & 1 << i) {
+            nl_msg_put_be32(&buf, i, htonl(nl_tp->attrs[i]));
+        }
+    }
+    nl_msg_end_nested(&buf, offset);
+
+    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
+    ofpbuf_uninit(&buf);
+    return err;
+}
+
+int
+nl_ct_get_timeout_policy(const char *tp_name,
+                         struct nl_ct_timeout_policy *nl_tp)
+{
+    struct ofpbuf request, *reply;
+    int err;
+
+    ofpbuf_init(&request, 512);
+    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_GET, NLM_F_REQUEST | NLM_F_ACK);
+    nl_msg_put_string(&request, CTA_TIMEOUT_NAME, tp_name);
+    err = nl_transact(NETLINK_NETFILTER, &request, &reply);
+    if (err) {
+        goto out;
+    }
+
+    err = nl_ct_timeout_policy_from_ofpbuf(reply, nl_tp, false);
+
+out:
+    ofpbuf_uninit(&request);
+    ofpbuf_delete(reply);
+    return err;
+}
+
+int
+nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
+                                 struct nl_ct_timeout_policy *nl_tp)
+{
+    struct ofpbuf request, *reply;
+    int err;
+
+    ofpbuf_init(&request, 512);
+    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_DEFAULT_GET,
+                        NLM_F_REQUEST | NLM_F_ACK);
+
+    nl_msg_put_be16(&request, CTA_TIMEOUT_L3PROTO, htons(l3num));
+    nl_msg_put_u8(&request, CTA_TIMEOUT_L4PROTO, l4num);
+    err = nl_transact(NETLINK_NETFILTER, &request, &reply);
+    if (err) {
+        goto out;
+    }
+
+    err = nl_ct_timeout_policy_from_ofpbuf(reply, nl_tp, true);
+
+out:
+    ofpbuf_uninit(&request);
+    ofpbuf_delete(reply);
+    return err;
+}
+
+int
+nl_ct_del_timeout_policy(const char *tp_name)
+{
+    struct ofpbuf buf;
+    int err;
+
+    ofpbuf_init(&buf, 64);
+    nl_msg_put_nfgenmsg(&buf, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_DELETE, NLM_F_REQUEST | NLM_F_ACK);
+
+    nl_msg_put_string(&buf, CTA_TIMEOUT_NAME, tp_name);
+    err = nl_transact(NETLINK_NETFILTER, &buf, NULL);
+    ofpbuf_uninit(&buf);
+    return err;
+}
+
+struct nl_ct_timeout_policy_dump_state {
+    struct nl_dump dump;
+    struct ofpbuf buf;
+};
+
+int
+nl_ct_timeout_policy_dump_start(
+    struct nl_ct_timeout_policy_dump_state **statep)
+{
+    struct ofpbuf request;
+    struct nl_ct_timeout_policy_dump_state *state;
+
+    *statep = state = xzalloc(sizeof *state);
+    ofpbuf_init(&request, 512);
+    nl_msg_put_nfgenmsg(&request, 0, AF_UNSPEC, NFNL_SUBSYS_CTNETLINK_TIMEOUT,
+                        IPCTNL_MSG_TIMEOUT_GET,
+                        NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP);
+
+    nl_dump_start(&state->dump, NETLINK_NETFILTER, &request);
+    ofpbuf_uninit(&request);
+    ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE);
+    return 0;
+}
+
+int
+nl_ct_timeout_policy_dump_next(struct nl_ct_timeout_policy_dump_state *state,
+                               struct nl_ct_timeout_policy *nl_tp)
+{
+    struct ofpbuf reply;
+    int err;
+
+    if (!nl_dump_next(&state->dump, &reply, &state->buf)) {
+        return EOF;
+    }
+    err = nl_ct_timeout_policy_from_ofpbuf(&reply, nl_tp, false);
+    ofpbuf_uninit(&reply);
+    return err;
+}
+
+int
+nl_ct_timeout_policy_dump_done(struct nl_ct_timeout_policy_dump_state *state)
+{
+    int err  = nl_dump_done(&state->dump);
+    ofpbuf_uninit(&state->buf);
+    free(state);
+    return err;
+}
+
 /* Translate netlink entry status flags to CT_DPIF_TCP status flags. */
 static uint32_t
 ips_status_to_dpif_flags(uint32_t status)
diff --git a/lib/netlink-conntrack.h b/lib/netlink-conntrack.h
index 8b536fd65ba8..ae6e428e0929 100644
--- a/lib/netlink-conntrack.h
+++ b/lib/netlink-conntrack.h
@@ -17,9 +17,12 @@ 
 #ifndef NETLINK_CONNTRACK_H
 #define NETLINK_CONNTRACK_H
 
+#include <linux/netfilter/nfnetlink_cttimeout.h>
+
 #include "byte-order.h"
 #include "compiler.h"
 #include "ct-dpif.h"
+#include "netlink-socket.h"
 #include "openvswitch/dynamic-string.h"
 #include "openvswitch/hmap.h"
 #include "openvswitch/ofpbuf.h"
@@ -33,7 +36,18 @@  enum nl_ct_event_type {
     NL_CT_EVENT_DELETE = 1 << 2,
 };
 
+#define NL_CT_TIMEOUT_POLICY_MAX_ATTR (CTA_TIMEOUT_TCP_MAX + 1)
+
+struct nl_ct_timeout_policy {
+    char        name[CTNL_TIMEOUT_NAME_MAX];
+    uint16_t    l3num;
+    uint8_t     l4num;
+    uint32_t    attrs[NL_CT_TIMEOUT_POLICY_MAX_ATTR];
+    uint32_t    present;
+};
+
 struct nl_ct_dump_state;
+struct nl_ct_timeout_policy_dump_state;
 
 int nl_ct_dump_start(struct nl_ct_dump_state **, const uint16_t *zone,
         int *ptot_bkts);
@@ -44,6 +58,21 @@  int nl_ct_flush(void);
 int nl_ct_flush_zone(uint16_t zone);
 int nl_ct_flush_tuple(const struct ct_dpif_tuple *, uint16_t zone);
 
+int nl_ct_set_timeout_policy(const struct nl_ct_timeout_policy *nl_tp);
+int nl_ct_set_default_timeout_policy(const struct nl_ct_timeout_policy *nl_tp);
+int nl_ct_get_timeout_policy(const char *tp_name,
+                             struct nl_ct_timeout_policy *nl_tp);
+int nl_ct_get_default_timeout_policy(uint16_t l3num, uint8_t l4num,
+                                     struct nl_ct_timeout_policy *nl_tp);
+int nl_ct_del_timeout_policy(const char *tp_name);
+int nl_ct_timeout_policy_dump_start(
+    struct nl_ct_timeout_policy_dump_state **statep);
+int nl_ct_timeout_policy_dump_next(
+    struct nl_ct_timeout_policy_dump_state *state,
+    struct nl_ct_timeout_policy *nl_tp);
+int nl_ct_timeout_policy_dump_done(
+    struct nl_ct_timeout_policy_dump_state *state);
+
 bool nl_ct_parse_entry(struct ofpbuf *, struct ct_dpif_entry *,
                        enum nl_ct_event_type *);
 void nl_ct_format_event_entry(const struct ct_dpif_entry *,
diff --git a/lib/netlink-protocol.h b/lib/netlink-protocol.h
index c0617dfad21f..bf631b1a14d0 100644
--- a/lib/netlink-protocol.h
+++ b/lib/netlink-protocol.h
@@ -48,6 +48,7 @@ 
 #define NLM_F_ECHO              0x008
 
 #define NLM_F_ROOT              0x100
+#define NLM_F_REPLACE           0x100
 #define NLM_F_MATCH             0x200
 #define NLM_F_EXCL              0x200
 #define NLM_F_ATOMIC            0x400