[ovs-dev] [PATCH v2 1/2] tunneling: Avoid recirculation on datapath by computing the recirculate actions at translate time.
Jarno Rajahalme
jarno at ovn.org
Tue Jan 24 20:59:24 UTC 2017
The userspace CLONE action is now merged to OVS master, so you can rebase the patches to use that.
Jarno
> On Jan 24, 2017, at 3:49 AM, Zoltán Balogh <zoltan.balogh at ericsson.com> wrote:
>
> Hi,
>
> All right, we will rebase the code once the patch is merged.
>
> Could you have a look at the code and review it, please?
> So we could receive feedback in time.
>
> Furthermore it would be nice to get some help regarding the failing test cases.
>
> 768: tunnel_push_pop - packet_out FAILED (tunnel-push-pop.at:203)
> 2261: ovn -- 3 HVs, 1 LS, 3 lports/HV FAILED (ovn.at:1295)
> 2266: ovn -- 3 HVs, 3 LS, 3 lports/LS, 1 LR FAILED (ovn.at:2468)
> 2268: ovn -- 2 HVs, 2 LS, 1 lport/LS, 2 peer LRs FAILED (ovn.at:2979)
> 2298: ovn -- 1 LR with distributed router gateway port FAILED (ovn.at:6480)
>
> The patches are available here:
> https://patchwork.ozlabs.org/patch/717683/
> https://patchwork.ozlabs.org/patch/717681/
>
>
> Best regards,
> Zoltan
>
>
>
>> On Jan 22, 2017, at 5:50 AM, Jan Scheurich <jan.scheurich at web.de> wrote:
>> =20
>> How does the dpif-netdev CLONE action introduced here relate to the simil=
> ar action already introduced in the context of the OpenFLow CLONE action (s=
> ee for example https://mail.openvswitch.org/pipermail/ovs-dev/2017-January/=
> 327808.html).
>> =20
>
> OpenFlow level clone action was added a bit earlier, the email you refer to=
> is a dpif-netdev level clone action.
>
>> I assume these do comparable things and the native tunnel implementation =
> should re-use the new CLONE action.
>> =20
>
> This is true. There were other needs for the datapath level clone action, a=
> nd once that is merged this patch series should rebase to that.
>
> Jarno
>
>> /Jan
>> =20
>> On 2017-01-20 14:40, Zolt=E1n Balogh wrote:
>>> From: Sugesh Chandran <sugesh.chandran at intel.com>
>>> =20
>>> Openvswitch datapath recirculates packets for tunneling, i.e.
>>> the incoming packets are encapsulated at first pass. Further actions are
>>> applied on encapsulated packets on the second pass after recirculating.
>>> The proposed patch compute and append the post tunnel actions at the tim=
> e of
>>> translation itself instead of recirculating at datapath. These actions a=
> re solely
>>> depends on tunnel attributes so there is no need of datapath recirculati=
> on.
>>> By avoiding the recirculation at datapath, the patch offers upto 30%
>>> performance improvement for VxLAN tunneling in our testing.
>>> The action execution logic is also extended with new CLONE action to def=
> ine
>>> the packet cloning when the actions are combined. The lenght in the CLON=
> E
>>> action specifies the size of nested action set.
>>> =20
>>> Signed-off-by: Sugesh Chandran <sugesh.chandran at intel.com>
>>> Signed-off-by: Zolt=E1n Balogh <zoltan.balogh at ericsson.com>
>>> Co-authored-by: Zolt=E1n Balogh <zoltan.balogh at ericsson.com>
>>> ---
>>> datapath/linux/compat/include/linux/openvswitch.h | 1 +
>>> lib/dpif-netdev.c | 20 +-
>>> lib/dpif.c | 1 +
>>> lib/odp-execute.c | 57 ++++-
>>> lib/odp-util.c | 96 +++++++-
>>> lib/odp-util.h | 5 +
>>> ofproto/ofproto-dpif-sflow.c | 1 +
>>> ofproto/ofproto-dpif-xlate.c | 267 +++++++++++----=
> -------
>>> 8 files changed, 288 insertions(+), 160 deletions(-)
>>> =20
>>> diff --git a/datapath/linux/compat/include/linux/openvswitch.h b/datapat=
> h/linux/compat/include/linux/openvswitch.h
>>> index 12260d8..91849d6 100644
>>> --- a/datapath/linux/compat/include/linux/openvswitch.h
>>> +++ b/datapath/linux/compat/include/linux/openvswitch.h
>>> @@ -818,6 +818,7 @@ enum ovs_action_attr {
>>> #ifndef __KERNEL__
>>> =09OVS_ACTION_ATTR_TUNNEL_PUSH, /* struct ovs_action_push_tnl*/
>>> =09OVS_ACTION_ATTR_TUNNEL_POP, /* u32 port number. */
>>> +=09OVS_ACTION_ATTR_CLONE,
>>> #endif
>>> =09__OVS_ACTION_ATTR_MAX,=09 /* Nothing past this will be accepted
>>> =09=09=09=09 * from userspace. */
>>> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
>>> index 3901129..0b85fe4 100644
>>> --- a/lib/dpif-netdev.c
>>> +++ b/lib/dpif-netdev.c
>>> @@ -4547,24 +4547,8 @@ dp_execute_cb(void *aux_, struct dp_packet_batch =
> *packets_,
>>> case OVS_ACTION_ATTR_TUNNEL_PUSH:
>>> if (*depth < MAX_RECIRC_DEPTH) {
>>> - struct dp_packet_batch tnl_pkt;
>>> - struct dp_packet_batch *orig_packets_ =3D packets_;
>>> - int err;
>>> -
>>> - if (!may_steal) {
>>> - dp_packet_batch_clone(&tnl_pkt, packets_);
>>> - packets_ =3D &tnl_pkt;
>>> - dp_packet_batch_reset_cutlen(orig_packets_);
>>> - }
>>> -
>>> dp_packet_batch_apply_cutlen(packets_);
>>> -
>>> - err =3D push_tnl_action(pmd, a, packets_);
>>> - if (!err) {
>>> - (*depth)++;
>>> - dp_netdev_recirculate(pmd, packets_);
>>> - (*depth)--;
>>> - }
>>> + push_tnl_action(pmd, a, packets_);
>>> return;
>>> }
>>> break;
>>> @@ -4714,6 +4698,8 @@ dp_execute_cb(void *aux_, struct dp_packet_batch *=
> packets_,
>>> break;
>>> }
>>> + case OVS_ACTION_ATTR_CLONE:
>>> + break;
>>> case OVS_ACTION_ATTR_PUSH_VLAN:
>>> case OVS_ACTION_ATTR_POP_VLAN:
>>> case OVS_ACTION_ATTR_PUSH_MPLS:
>>> diff --git a/lib/dpif.c b/lib/dpif.c
>>> index 50c3382..b698da2 100644
>>> --- a/lib/dpif.c
>>> +++ b/lib/dpif.c
>>> @@ -1183,6 +1183,7 @@ dpif_execute_helper_cb(void *aux_, struct dp_packe=
> t_batch *packets_,
>>> case OVS_ACTION_ATTR_SAMPLE:
>>> case OVS_ACTION_ATTR_TRUNC:
>>> case OVS_ACTION_ATTR_UNSPEC:
>>> + case OVS_ACTION_ATTR_CLONE:
>>> case __OVS_ACTION_ATTR_MAX:
>>> OVS_NOT_REACHED();
>>> }
>>> diff --git a/lib/odp-execute.c b/lib/odp-execute.c
>>> index 73e1016..77bca94 100644
>>> --- a/lib/odp-execute.c
>>> +++ b/lib/odp-execute.c
>>> @@ -541,6 +541,7 @@ requires_datapath_assistance(const struct nlattr *a)
>>> case OVS_ACTION_ATTR_USERSPACE:
>>> case OVS_ACTION_ATTR_RECIRC:
>>> case OVS_ACTION_ATTR_CT:
>>> + case OVS_ACTION_ATTR_CLONE:
>>> return true;
>>> case OVS_ACTION_ATTR_SET:
>>> @@ -562,6 +563,29 @@ requires_datapath_assistance(const struct nlattr *a=
> )
>>> return false;
>>> }
>>> +static inline size_t
>>> +find_combine_action_end(const struct nlattr **actions_, size_t *actions=
> _len,
>>> + const struct nlattr **comb_start,
>>> + bool *comb_last_action)
>>> +{
>>> + const struct nlattr *a;
>>> + const struct nlattr *actions =3D *actions_;
>>> + unsigned int left;
>>> + bool last_action;
>>> + size_t comb_size =3D nl_attr_get_u32(actions); /* Size of combined =
> actions */
>>> + *actions_len -=3D NLA_ALIGN(actions->nla_len);
>>> + actions =3D nl_attr_next(actions);
>>> +
>>> + *comb_start =3D actions;
>>> + a =3D (void *) ((uint8_t *) actions + NLA_ALIGN(comb_size));
>>> + left =3D (*actions_len) - comb_size;
>>> + last_action =3D (left <=3D NLA_ALIGN(a->nla_len));
>>> + *actions_ =3D a;
>>> + *comb_last_action =3D last_action ? true : false;
>>> + *actions_len =3D left;
>>> + return comb_size;
>>> +}
>>> +
>>> void
>>> odp_execute_actions(void *dp, struct dp_packet_batch *batch, bool steal=
> ,
>>> const struct nlattr *actions, size_t actions_len,
>>> @@ -570,7 +594,7 @@ odp_execute_actions(void *dp, struct dp_packet_batch=
> *batch, bool steal,
>>> struct dp_packet **packets =3D batch->packets;
>>> int cnt =3D batch->count;
>>> const struct nlattr *a;
>>> - unsigned int left;
>>> + size_t left;
>>> int i;
>>> NL_ATTR_FOR_EACH_UNSAFE (a, left, actions, actions_len) {
>>> @@ -582,7 +606,35 @@ odp_execute_actions(void *dp, struct dp_packet_batc=
> h *batch, bool steal,
>>> /* Allow 'dp_execute_action' to steal the packet data i=
> f we do
>>> * not need it any more. */
>>> bool may_steal =3D steal && last_action;
>>> -
>>> + if ((enum ovs_action_attr) type =3D=3D
>>> + OVS_ACTION_ATTR_CLONE) {
>>> + size_t comb_action_len;
>>> + const struct nlattr *comb_actions =3D NULL;
>>> + bool comb_last_action =3D true;
>>> + comb_action_len =3D find_combine_action_end(&a, &le=
> ft,
>>> + &comb_actio=
> ns,
>>> + &comb_last_=
> action);
>>> + if (OVS_UNLIKELY(!comb_action_len)) {
>>> + /* No combine present, execute current action *=
> /
>>> + goto execute;
>>> + }
>>> + if (OVS_LIKELY(comb_last_action)) {
>>> + odp_execute_actions(dp, batch, 1, comb_actions,
>>> + comb_action_len, dp_execute_act=
> ion);
>>> + return;
>>> + }
>>> + else {
>>> + /* Packets to be duplicated for the sub action =
> */
>>> + struct dp_packet_batch comb_pkt_batch;
>>> + dp_packet_batch_clone(&comb_pkt_batch, batch);
>>> + dp_packet_batch_reset_cutlen(batch);
>>> + odp_execute_actions(dp, &comb_pkt_batch, steal,
>>> + comb_actions, comb_action_len,
>>> + dp_execute_action);
>>> + }
>>> + continue;
>>> + }
>>> +execute:
>>> dp_execute_action(dp, batch, a, may_steal);
>>> if (last_action) {
>>> @@ -692,6 +744,7 @@ odp_execute_actions(void *dp, struct dp_packet_batch=
> *batch, bool steal,
>>> case OVS_ACTION_ATTR_RECIRC:
>>> case OVS_ACTION_ATTR_CT:
>>> case OVS_ACTION_ATTR_UNSPEC:
>>> + case OVS_ACTION_ATTR_CLONE:
>>> case __OVS_ACTION_ATTR_MAX:
>>> OVS_NOT_REACHED();
>>> }
>>> diff --git a/lib/odp-util.c b/lib/odp-util.c
>>> index 1e70e3a..4a311d9 100644
>>> --- a/lib/odp-util.c
>>> +++ b/lib/odp-util.c
>>> @@ -121,6 +121,7 @@ odp_action_len(uint16_t type)
>>> case OVS_ACTION_ATTR_SET_MASKED: return ATTR_LEN_VARIABLE;
>>> case OVS_ACTION_ATTR_SAMPLE: return ATTR_LEN_VARIABLE;
>>> case OVS_ACTION_ATTR_CT: return ATTR_LEN_VARIABLE;
>>> + case OVS_ACTION_ATTR_CLONE: return sizeof(uint32_t);
>>> case OVS_ACTION_ATTR_UNSPEC:
>>> case __OVS_ACTION_ATTR_MAX:
>>> @@ -509,7 +510,7 @@ format_odp_tnl_push_header(struct ds *ds, struct ovs=
> _action_push_tnl *data)
>>> gnh->oam ? "oam," : "",
>>> gnh->critical ? "crit," : "",
>>> ntohl(get_16aligned_be32(&gnh->vni)) >> 8);
>>> -
>>> +
>>> if (gnh->opt_len) {
>>> ds_put_cstr(ds, ",options(");
>>> format_geneve_opts(gnh->options, NULL, gnh->opt_len * 4,
>>> @@ -865,6 +866,14 @@ format_odp_action(struct ds *ds, const struct nlatt=
> r *a)
>>> case OVS_ACTION_ATTR_CT:
>>> format_odp_conntrack_action(ds, a);
>>> break;
>>> + case OVS_ACTION_ATTR_CLONE: {
>>> + uint32_t comb_len;
>>> + comb_len =3D nl_attr_get_u32(a);
>>> + if (comb_len) {
>>> + ds_put_cstr(ds,"{");
>>> + }
>>> + break;
>>> + }
>>> case OVS_ACTION_ATTR_UNSPEC:
>>> case __OVS_ACTION_ATTR_MAX:
>>> default:
>>> @@ -873,27 +882,81 @@ format_odp_action(struct ds *ds, const struct nlat=
> tr *a)
>>> }
>>> }
>>> +static void
>>> +print_odp_actions(struct ds *ds, bool is_first_action,
>>> + const struct nlattr **action_,
>>> + size_t *action_len, size_t clone_len, b=
> ool is_clone_ctx)
>>> +{
>>> +
>>> + enum ovs_action_attr type;
>>> + const struct nlattr *action =3D *action_;
>>> + const struct nlattr *curr_action =3D action;
>>> + size_t curr_action_len;
>>> +
>>> + if (!nl_attr_is_valid(action, *action_len)) {
>>> + return;
>>> + }
>>> + curr_action_len =3D nl_attr_len_pad(action, *action_len);
>>> + type =3D nl_attr_type(action);
>>> + if(!is_first_action) {
>>> + ds_put_char(ds, ',');
>>> + }
>>> + else {
>>> + /* Next iteration onwards, its not first action anymore. Reset =
> flag
>>> + * accordingly.
>>> + */
>>> + is_first_action =3D 0;
>>> + }
>>> +
>>> + format_odp_action(ds, action);
>>> +
>>> + action =3D nl_attr_next(action);
>>> + *action_ =3D action;
>>> + *action_len -=3D curr_action_len;
>>> + if (type =3D=3D OVS_ACTION_ATTR_CLONE) {
>>> + uint32_t comb_len;
>>> + comb_len =3D nl_attr_get_u32(curr_action);
>>> + if (comb_len) {
>>> + print_odp_actions(ds, 1, action_, action_len,
>>> + comb_len, true);
>>> + ds_put_char(ds, '}');
>>> + }
>>> + else {
>>> + return;/* Error, cannot print anything. */
>>> + }
>>> + }
>>> + if (is_clone_ctx) {
>>> + if(!clone_len) {
>>> + return;
>>> + }
>>> + else {
>>> + clone_len =3D (clone_len >=3D curr_action_len) ?
>>> + clone_len - curr_action_len :
>>> + 0;
>>> + }
>>> + }
>>> + print_odp_actions(ds, is_first_action, action_, action_len, clone_l=
> en,
>>> + is_clone_ctx);
>>> +}
>>> +
>>> void
>>> format_odp_actions(struct ds *ds, const struct nlattr *actions,
>>> size_t actions_len)
>>> {
>>> if (actions_len) {
>>> const struct nlattr *a;
>>> - unsigned int left;
>>> + size_t left;
>>> - NL_ATTR_FOR_EACH (a, left, actions, actions_len) {
>>> - if (a !=3D actions) {
>>> - ds_put_char(ds, ',');
>>> - }
>>> - format_odp_action(ds, a);
>>> - }
>>> + a =3D actions;
>>> + left =3D actions_len;
>>> + print_odp_actions(ds, 1, &a, &left, 0, false);
>>> if (left) {
>>> int i;
>>> if (left =3D=3D actions_len) {
>>> ds_put_cstr(ds, "<empty>");
>>> }
>>> - ds_put_format(ds, ",***%u leftover bytes*** (", left);
>>> + ds_put_format(ds, ",***%"PRIuSIZE" leftover bytes*** (", le=
> ft);
>>> for (i =3D 0; i < left; i++) {
>>> ds_put_format(ds, "%02x", ((const uint8_t *) a)[i]);
>>> }
>>> @@ -5321,6 +5384,21 @@ odp_put_tunnel_action(const struct flow_tnl *tunn=
> el,
>>> }
>>> void
>>> +odp_put_combine_start_action(struct ofpbuf *odp_actions, uint32_t actio=
> n_len)
>>> +{
>>> + nl_msg_put_u32(odp_actions, OVS_ACTION_ATTR_CLONE, action_len);
>>> +}
>>> +
>>> +void
>>> +odp_update_combine_start_action(struct nlattr *start_combine,
>>> + uint32_t combine_size)
>>> +{
>>> + uint32_t *combine_len;
>>> + combine_len =3D (uint32_t *)(start_combine + 1);
>>> + *combine_len =3D combine_size;
>>> +}
>>> +
>>> +void
>>> odp_put_tnl_push_action(struct ofpbuf *odp_actions,
>>> struct ovs_action_push_tnl *data)
>>> {
>>> diff --git a/lib/odp-util.h b/lib/odp-util.h
>>> index 42011bc..a32dc76 100644
>>> --- a/lib/odp-util.h
>>> +++ b/lib/odp-util.h
>>> @@ -308,4 +308,9 @@ void odp_put_tunnel_action(const struct flow_tnl *tu=
> nnel,
>>> void odp_put_tnl_push_action(struct ofpbuf *odp_actions,
>>> struct ovs_action_push_tnl *data);
>>> +void odp_put_combine_start_action(struct ofpbuf *odp_actions,
>>> + uint32_t action_len);
>>> +void
>>> +odp_update_combine_start_action(struct nlattr *start_combine,
>>> + uint32_t combine_size);
>>> #endif /* odp-util.h */
>>> diff --git a/ofproto/ofproto-dpif-sflow.c b/ofproto/ofproto-dpif-sflow.c
>>> index 37992b4..8d51374 100644
>>> --- a/ofproto/ofproto-dpif-sflow.c
>>> +++ b/ofproto/ofproto-dpif-sflow.c
>>> @@ -1163,6 +1163,7 @@ dpif_sflow_read_actions(const struct flow *flow,
>>> =09}
>>> =09case OVS_ACTION_ATTR_SAMPLE:
>>> =09case OVS_ACTION_ATTR_UNSPEC:
>>> + case OVS_ACTION_ATTR_CLONE:
>>> =09case __OVS_ACTION_ATTR_MAX:
>>> =09default:
>>> =09 break;
>>> diff --git a/ofproto/ofproto-dpif-xlate.c b/ofproto/ofproto-dpif-xlate.c
>>> index 9a15ec3..ded0932 100644
>>> --- a/ofproto/ofproto-dpif-xlate.c
>>> +++ b/ofproto/ofproto-dpif-xlate.c
>>> @@ -408,6 +408,10 @@ static void xlate_action_set(struct xlate_ctx *ctx)=
> ;
>>> static void xlate_commit_actions(struct xlate_ctx *ctx);
>>> static void
>>> +group_actions(struct xlate_ctx *ctx, const struct xport *in_dev,
>>> + struct xport *out_dev);
>>> +
>>> +static void
>>> ctx_trigger_freeze(struct xlate_ctx *ctx)
>>> {
>>> ctx->exit =3D true;
>>> @@ -3029,7 +3033,23 @@ build_tunnel_send(struct xlate_ctx *ctx, const st=
> ruct xport *xport,
>>> }
>>> tnl_push_data.tnl_port =3D odp_to_u32(tunnel_odp_port);
>>> tnl_push_data.out_port =3D odp_to_u32(out_dev->odp_port);
>>> +
>>> + struct nlattr *start_combine;
>>> + uint32_t action_size =3D 0;
>>> + uint32_t start_action_size =3D 0;
>>> + uint32_t push_action_size =3D 0;
>>> + start_action_size =3D ctx->odp_actions->size;
>>> + odp_put_combine_start_action(ctx->odp_actions, 0);
>>> + action_size =3D ctx->odp_actions->size;
>>> odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
>>> + push_action_size =3D ctx->odp_actions->size;
>>> + group_actions(ctx, xport, out_dev);
>>> + start_combine =3D ofpbuf_at(ctx->odp_actions, start_action_size, (N=
> LA_HDRLEN));
>>> + if (ctx->odp_actions->size > push_action_size) {
>>> + /* Update the combine action only when combined */
>>> + odp_update_combine_start_action(start_combine,
>>> + (ctx->odp_actions->size - action_si=
> ze));
>>> + }
>>> return 0;
>>> }
>>> @@ -3071,6 +3091,119 @@ xlate_flow_is_protected(const struct xlate_ctx =
> *ctx, const struct flow *flow, co
>>> }
>>> static void
>>> +group_actions(struct xlate_ctx *ctx, const struct xport *in_dev,
>>> + struct xport *out_dev)
>>> +{
>>> + struct flow *flow =3D &ctx->xin->flow;
>>> + struct flow old_flow =3D ctx->xin->flow;
>>> + struct flow_tnl old_flow_tnl_wc =3D ctx->wc->masks.tunnel;
>>> + bool old_conntrack =3D ctx->conntracked;
>>> + bool old_was_mpls =3D ctx->was_mpls;
>>> + ovs_version_t old_version =3D ctx->xin->tables_version;
>>> + struct ofpbuf old_stack =3D ctx->stack;
>>> + union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
>>> + struct ofpbuf old_action_set =3D ctx->action_set;
>>> + uint64_t actset_stub[1024 / 8];
>>> +
>>> + ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
>>> + ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
>>> + flow->in_port.ofp_port =3D out_dev->ofp_port;
>>> + flow->metadata =3D htonll(0);
>>> + memset(&flow->tunnel, 0, sizeof flow->tunnel);
>>> + memset(&ctx->wc->masks.tunnel, 0, sizeof ctx->wc->masks.tunnel);
>>> + flow->tunnel.metadata.tab =3D ofproto_get_tun_tab(&out_dev->xbridge=
> ->ofproto->up);
>>> + ctx->wc->masks.tunnel.metadata.tab =3D flow->tunnel.metadata.tab;
>>> + memset(flow->regs, 0, sizeof flow->regs);
>>> + flow->actset_output =3D OFPP_UNSET;
>>> + ctx->conntracked =3D false;
>>> + clear_conntrack(ctx);
>>> + mirror_mask_t old_mirrors =3D ctx->mirrors;
>>> + bool independent_mirrors =3D out_dev->xbridge !=3D ctx->xbridge;
>>> + if (independent_mirrors) {
>>> + ctx->mirrors =3D 0;
>>> + }
>>> + ctx->xbridge =3D out_dev->xbridge;
>>> +
>>> + /* The bridge is now known so obtain its table version. */
>>> + ctx->xin->tables_version
>>> + =3D ofproto_dpif_get_tables_version(ctx->xbridge->ofproto=
> );
>>> +
>>> + if (!process_special(ctx, out_dev) && may_receive(out_dev, ctx)) {
>>> + if (xport_stp_forward_state(out_dev) && xport_rstp_forward_stat=
> e(out_dev)) {
>>> + xlate_table_action(ctx, flow->in_port.ofp_port, 0, true=
> , true);
>>> + if (!ctx->freezing) {
>>> + xlate_action_set(ctx);
>>> + }
>>> + if (ctx->freezing) {
>>> + finish_freezing(ctx);
>>> + }
>>> + } else {
>>> + /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL=
> and
>>> + * the learning action look at the packet, then drop it. */
>>> + struct flow old_base_flow =3D ctx->base_flow;
>>> + size_t old_size =3D ctx->odp_actions->size;
>>> + mirror_mask_t old_mirrors2 =3D ctx->mirrors;
>>> +
>>> + xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, tr=
> ue);
>>> + ctx->mirrors =3D old_mirrors2;
>>> + ctx->base_flow =3D old_base_flow;
>>> + ctx->odp_actions->size =3D old_size;
>>> +
>>> + /* Undo changes that may have been done for freezing. */
>>> + ctx_cancel_freeze(ctx);
>>> + }
>>> + }
>>> +
>>> + if (independent_mirrors) {
>>> + ctx->mirrors =3D old_mirrors;
>>> + }
>>> + ctx->xin->flow =3D old_flow;
>>> + ctx->xbridge =3D in_dev->xbridge;
>>> + ofpbuf_uninit(&ctx->action_set);
>>> + ctx->action_set =3D old_action_set;
>>> + ofpbuf_uninit(&ctx->stack);
>>> + ctx->stack =3D old_stack;
>>> +
>>> + /* Restore calling bridge's lookup version. */
>>> + ctx->xin->tables_version =3D old_version;
>>> +
>>> + /* Restore to calling bridge tunneling information */
>>> + ctx->wc->masks.tunnel =3D old_flow_tnl_wc;
>>> +
>>> + /* The out bridge popping MPLS should have no effect on the origina=
> l
>>> + * bridge. */
>>> + ctx->was_mpls =3D old_was_mpls;
>>> +
>>> + /* The out bridge's conntrack execution should have no effect on th=
> e
>>> + * original bridge. */
>>> + ctx->conntracked =3D old_conntrack;
>>> +
>>> + /* The fact that the out bridge exits (for any reason) does not mea=
> n
>>> + * that the original bridge should exit. Specifically, if the out
>>> + * bridge freezes translation, the original bridge must continue
>>> + * processing with the original, not the frozen packet! */
>>> + ctx->exit =3D false;
>>> +
>>> + /* Out bridge errors do not propagate back. */
>>> + ctx->error =3D XLATE_OK;
>>> +
>>> + if (ctx->xin->resubmit_stats) {
>>> + netdev_vport_inc_tx(in_dev->netdev, ctx->xin->resubmit_stats);
>>> + netdev_vport_inc_rx(out_dev->netdev, ctx->xin->resubmit_stats);
>>> + if (out_dev->bfd) {
>>> + bfd_account_rx(out_dev->bfd, ctx->xin->resubmit_stats);
>>> + }
>>> + }
>>> + if (ctx->xin->xcache) {
>>> + struct xc_entry *entry;
>>> + entry =3D xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
>>> + entry->dev.tx =3D netdev_ref(in_dev->netdev);
>>> + entry->dev.rx =3D netdev_ref(out_dev->netdev);
>>> + entry->dev.bfd =3D bfd_ref(out_dev->bfd);
>>> + }
>>> +}
>>> +
>>> +static void
>>> compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
>>> const struct xlate_bond_recirc *xr, bool check_=
> stp)
>>> {
>>> @@ -3133,138 +3266,8 @@ compose_output_action__(struct xlate_ctx *ctx, o=
> fp_port_t ofp_port,
>>> }
>>> if (xport->peer) {
>>> - const struct xport *peer =3D xport->peer;
>>> - struct flow old_flow =3D ctx->xin->flow;
>>> - struct flow_tnl old_flow_tnl_wc =3D ctx->wc->masks.tunnel;
>>> - bool old_conntrack =3D ctx->conntracked;
>>> - bool old_was_mpls =3D ctx->was_mpls;
>>> - ovs_version_t old_version =3D ctx->xin->tables_version;
>>> - struct ofpbuf old_stack =3D ctx->stack;
>>> - uint8_t new_stack[1024];
>>> - struct ofpbuf old_action_set =3D ctx->action_set;
>>> - struct ovs_list *old_trace =3D ctx->xin->trace;
>>> - uint64_t actset_stub[1024 / 8];
>>> -
>>> - ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
>>> - ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_st=
> ub);
>>> - flow->in_port.ofp_port =3D peer->ofp_port;
>>> - flow->metadata =3D htonll(0);
>>> - memset(&flow->tunnel, 0, sizeof flow->tunnel);
>>> - flow->tunnel.metadata.tab =3D ofproto_get_tun_tab(
>>> - &peer->xbridge->ofproto->up);
>>> - ctx->wc->masks.tunnel.metadata.tab =3D flow->tunnel.metadata.ta=
> b;
>>> - memset(flow->regs, 0, sizeof flow->regs);
>>> - flow->actset_output =3D OFPP_UNSET;
>>> - clear_conntrack(ctx);
>>> - ctx->xin->trace =3D xlate_report(ctx, OFT_BRIDGE,
>>> - "bridge(\"%s\")", peer->xbridge-=
>> name);
>>> -
>>> - /* When the patch port points to a different bridge, then the m=
> irrors
>>> - * for that bridge clearly apply independently to the packet, s=
> o we
>>> - * reset the mirror bitmap to zero and then restore it after th=
> e packet
>>> - * returns.
>>> - *
>>> - * When the patch port points to the same bridge, this is more =
> of a
>>> - * design decision: can mirrors be re-applied to the packet aft=
> er it
>>> - * re-enters the bridge, or should we treat that as doubly mirr=
> oring a
>>> - * single packet? The former may be cleaner, since it respects=
> the
>>> - * model in which a patch port is like a physical cable plugged=
> from
>>> - * one switch port to another, but the latter may be less surpr=
> ising to
>>> - * users. We take the latter choice, for now at least. (To us=
> e the
>>> - * former choice, hard-code 'independent_mirrors' to "true".) *=
> /
>>> - mirror_mask_t old_mirrors =3D ctx->mirrors;
>>> - bool independent_mirrors =3D peer->xbridge !=3D ctx->xbridge;
>>> - if (independent_mirrors) {
>>> - ctx->mirrors =3D 0;
>>> - }
>>> - ctx->xbridge =3D peer->xbridge;
>>> -
>>> - /* The bridge is now known so obtain its table version. */
>>> - ctx->xin->tables_version
>>> - =3D ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
>>> -
>>> - if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
>>> - if (xport_stp_forward_state(peer) && xport_rstp_forward_sta=
> te(peer)) {
>>> - xlate_table_action(ctx, flow->in_port.ofp_port, 0, true=
> , true);
>>> - if (!ctx->freezing) {
>>> - xlate_action_set(ctx);
>>> - }
>>> - if (ctx->freezing) {
>>> - finish_freezing(ctx);
>>> - }
>>> - } else {
>>> - /* Forwarding is disabled by STP and RSTP. Let OFPP_NO=
> RMAL and
>>> - * the learning action look at the packet, then drop it=
> . */
>>> - struct flow old_base_flow =3D ctx->base_flow;
>>> - size_t old_size =3D ctx->odp_actions->size;
>>> - mirror_mask_t old_mirrors2 =3D ctx->mirrors;
>>> -
>>> - xlate_table_action(ctx, flow->in_port.ofp_port, 0, true=
> , true);
>>> - ctx->mirrors =3D old_mirrors2;
>>> - ctx->base_flow =3D old_base_flow;
>>> - ctx->odp_actions->size =3D old_size;
>>> -
>>> - /* Undo changes that may have been done for freezing. *=
> /
>>> - ctx_cancel_freeze(ctx);
>>> - }
>>> - }
>>> -
>>> - ctx->xin->trace =3D old_trace;
>>> - if (independent_mirrors) {
>>> - ctx->mirrors =3D old_mirrors;
>>> - }
>>> - ctx->xin->flow =3D old_flow;
>>> - ctx->xbridge =3D xport->xbridge;
>>> - ofpbuf_uninit(&ctx->action_set);
>>> - ctx->action_set =3D old_action_set;
>>> - ofpbuf_uninit(&ctx->stack);
>>> - ctx->stack =3D old_stack;
>>> -
>>> - /* Restore calling bridge's lookup version. */
>>> - ctx->xin->tables_version =3D old_version;
>>> -
>>> - /* Since this packet came in on a patch port (from the perspect=
> ive of
>>> - * the peer bridge), it cannot have useful tunnel information. =
> As a
>>> - * result, any wildcards generated on that tunnel also cannot b=
> e valid.
>>> - * The tunnel wildcards must be restored to their original vers=
> ion since
>>> - * the peer bridge uses a separate tunnel metadata table and th=
> erefore
>>> - * any generated wildcards will be garbage in the context of ou=
> r
>>> - * metadata table. */
>>> - ctx->wc->masks.tunnel =3D old_flow_tnl_wc;
>>> -
>>> - /* The peer bridge popping MPLS should have no effect on the or=
> iginal
>>> - * bridge. */
>>> - ctx->was_mpls =3D old_was_mpls;
>>> -
>>> - /* The peer bridge's conntrack execution should have no effect =
> on the
>>> - * original bridge. */
>>> - ctx->conntracked =3D old_conntrack;
>>> -
>>> - /* The fact that the peer bridge exits (for any reason) does no=
> t mean
>>> - * that the original bridge should exit. Specifically, if the =
> peer
>>> - * bridge freezes translation, the original bridge must continu=
> e
>>> - * processing with the original, not the frozen packet! */
>>> - ctx->exit =3D false;
>>> -
>>> - /* Peer bridge errors do not propagate back. */
>>> - ctx->error =3D XLATE_OK;
>>> -
>>> - if (ctx->xin->resubmit_stats) {
>>> - netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats=
> );
>>> - netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats)=
> ;
>>> - if (peer->bfd) {
>>> - bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
>>> - }
>>> - }
>>> - if (ctx->xin->xcache) {
>>> - struct xc_entry *entry;
>>> -
>>> - entry =3D xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV=
> );
>>> - entry->dev.tx =3D netdev_ref(xport->netdev);
>>> - entry->dev.rx =3D netdev_ref(peer->netdev);
>>> - entry->dev.bfd =3D bfd_ref(peer->bfd);
>>> - }
>>> - return;
>>> + group_actions(ctx, xport, xport->peer);
>>> + return;
>>> }
>>> flow_vlan_tci =3D flow->vlan_tci;
>> =20
>> _______________________________________________
>> dev mailing list
>> dev at openvswitch.org
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>
> _______________________________________________
> dev mailing list
> dev at openvswitch.org
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
More information about the dev
mailing list