1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
17 #include "ofproto/ofproto-dpif-xlate.h"
23 #include "byte-order.h"
28 #include "dynamic-string.h"
31 #include "mac-learning.h"
32 #include "meta-flow.h"
33 #include "multipath.h"
34 #include "netdev-vport.h"
37 #include "odp-execute.h"
38 #include "ofp-actions.h"
39 #include "ofproto/ofproto-dpif-ipfix.h"
40 #include "ofproto/ofproto-dpif-sflow.h"
41 #include "ofproto/ofproto-dpif.h"
45 COVERAGE_DEFINE(ofproto_dpif_xlate);
47 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
49 /* Maximum depth of flow table recursion (due to resubmit actions) in a
50 * flow translation. */
51 #define MAX_RESUBMIT_RECURSION 64
55 struct xlate_out *xout;
57 struct ofproto_dpif *ofproto;
59 /* Flow at the last commit. */
60 struct flow base_flow;
62 /* Tunnel IP destination address as received. This is stored separately
63 * as the base_flow.tunnel is cleared on init to reflect the datapath
64 * behavior. Used to make sure not to send tunneled output to ourselves,
65 * which might lead to an infinite loop. This could happen easily
66 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
67 * actually set the tun_dst field. */
68 ovs_be32 orig_tunnel_ip_dst;
70 /* Stack for the push and pop actions. Each stack element is of type
71 * "union mf_subvalue". */
72 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
75 /* The rule that we are currently translating, or NULL. */
76 struct rule_dpif *rule;
78 int recurse; /* Recursion level, via xlate_table_action. */
79 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
80 uint32_t orig_skb_priority; /* Priority when packet arrived. */
81 uint8_t table_id; /* OpenFlow table ID where flow was found. */
82 uint32_t sflow_n_outputs; /* Number of output ports. */
83 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
84 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
85 bool exit; /* No further actions should be processed. */
88 /* A controller may use OFPP_NONE as the ingress port to indicate that
89 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
90 * when an input bundle is needed for validation (e.g., mirroring or
91 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
92 * any 'port' structs, so care must be taken when dealing with it. */
93 static struct ofbundle ofpp_none_bundle = {
95 .vlan_mode = PORT_VLAN_TRUNK
98 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
99 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
101 static void xlate_normal(struct xlate_ctx *);
102 static void xlate_report(struct xlate_ctx *, const char *);
103 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
104 uint8_t table_id, bool may_packet_in);
105 static bool input_vid_is_valid(uint16_t vid, struct ofbundle *, bool warn);
106 static uint16_t input_vid_to_vlan(const struct ofbundle *, uint16_t vid);
107 static void output_normal(struct xlate_ctx *, const struct ofbundle *,
109 static void compose_output_action(struct xlate_ctx *, uint16_t ofp_port);
111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
114 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
116 return (bundle->vlan_mode != PORT_VLAN_ACCESS
117 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
121 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
123 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
127 vlan_is_mirrored(const struct ofmirror *m, int vlan)
129 return !m->vlans || bitmap_is_set(m->vlans, vlan);
132 static struct ofbundle *
133 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
134 bool warn, struct ofport_dpif **in_ofportp)
136 struct ofport_dpif *ofport;
138 /* Find the port and bundle for the received packet. */
139 ofport = get_ofp_port(ofproto, in_port);
141 *in_ofportp = ofport;
143 if (ofport && ofport->bundle) {
144 return ofport->bundle;
147 /* Special-case OFPP_NONE, which a controller may use as the ingress
148 * port for traffic that it is sourcing. */
149 if (in_port == OFPP_NONE) {
150 return &ofpp_none_bundle;
153 /* Odd. A few possible reasons here:
155 * - We deleted a port but there are still a few packets queued up
158 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
159 * we don't know about.
161 * - The ofproto client didn't configure the port as part of a bundle.
162 * This is particularly likely to happen if a packet was received on the
163 * port after it was created, but before the client had a chance to
164 * configure its bundle.
167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
169 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
170 "port %"PRIu16, ofproto->up.name, in_port);
176 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
178 struct ofproto_dpif *ofproto = ctx->ofproto;
179 mirror_mask_t mirrors;
180 struct ofbundle *in_bundle;
183 const struct nlattr *a;
186 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
187 ctx->xin->packet != NULL, NULL);
191 mirrors = in_bundle->src_mirrors;
193 /* Drop frames on bundles reserved for mirroring. */
194 if (in_bundle->mirror_out) {
195 if (ctx->xin->packet != NULL) {
196 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
197 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
198 "%s, which is reserved exclusively for mirroring",
199 ctx->ofproto->up.name, in_bundle->name);
205 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
206 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
209 vlan = input_vid_to_vlan(in_bundle, vid);
211 /* Look at the output ports to check for destination selections. */
213 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
214 ctx->xout->odp_actions.size) {
215 enum ovs_action_attr type = nl_attr_type(a);
216 struct ofport_dpif *ofport;
218 if (type != OVS_ACTION_ATTR_OUTPUT) {
222 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
223 if (ofport && ofport->bundle) {
224 mirrors |= ofport->bundle->dst_mirrors;
232 /* Restore the original packet before adding the mirror actions. */
233 ctx->xin->flow = *orig_flow;
238 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
241 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
244 if (!vlan_is_mirrored(m, vlan)) {
245 mirrors = zero_rightmost_1bit(mirrors);
249 mirrors &= ~m->dup_mirrors;
250 ctx->xout->mirrors |= m->dup_mirrors;
252 output_normal(ctx, m->out, vlan);
253 } else if (vlan != m->out_vlan
254 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
255 struct ofbundle *bundle;
257 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
258 if (ofbundle_includes_vlan(bundle, m->out_vlan)
259 && !bundle->mirror_out) {
260 output_normal(ctx, bundle, m->out_vlan);
267 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
268 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
269 * the bundle on which the packet was received, returns the VLAN to which the
272 * Both 'vid' and the return value are in the range 0...4095. */
274 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
276 switch (in_bundle->vlan_mode) {
277 case PORT_VLAN_ACCESS:
278 return in_bundle->vlan;
281 case PORT_VLAN_TRUNK:
284 case PORT_VLAN_NATIVE_UNTAGGED:
285 case PORT_VLAN_NATIVE_TAGGED:
286 return vid ? vid : in_bundle->vlan;
293 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
294 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
297 * 'vid' should be the VID obtained from the 802.1Q header that was received as
298 * part of a packet (specify 0 if there was no 802.1Q header), in the range
301 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
303 /* Allow any VID on the OFPP_NONE port. */
304 if (in_bundle == &ofpp_none_bundle) {
308 switch (in_bundle->vlan_mode) {
309 case PORT_VLAN_ACCESS:
312 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
313 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
314 "packet received on port %s configured as VLAN "
315 "%"PRIu16" access port",
316 in_bundle->ofproto->up.name, vid,
317 in_bundle->name, in_bundle->vlan);
323 case PORT_VLAN_NATIVE_UNTAGGED:
324 case PORT_VLAN_NATIVE_TAGGED:
326 /* Port must always carry its native VLAN. */
330 case PORT_VLAN_TRUNK:
331 if (!ofbundle_includes_vlan(in_bundle, vid)) {
333 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
334 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
335 "received on port %s not configured for trunking "
337 in_bundle->ofproto->up.name, vid,
338 in_bundle->name, vid);
350 /* Given 'vlan', the VLAN that a packet belongs to, and
351 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
352 * that should be included in the 802.1Q header. (If the return value is 0,
353 * then the 802.1Q header should only be included in the packet if there is a
356 * Both 'vlan' and the return value are in the range 0...4095. */
358 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
360 switch (out_bundle->vlan_mode) {
361 case PORT_VLAN_ACCESS:
364 case PORT_VLAN_TRUNK:
365 case PORT_VLAN_NATIVE_TAGGED:
368 case PORT_VLAN_NATIVE_UNTAGGED:
369 return vlan == out_bundle->vlan ? 0 : vlan;
377 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
380 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
381 struct ofport_dpif *port;
383 ovs_be16 tci, old_tci;
385 vid = output_vlan_to_vid(out_bundle, vlan);
386 if (!out_bundle->bond) {
387 port = ofbundle_get_a_port(out_bundle);
389 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
390 &ctx->xout->wc, vid, &ctx->xout->tags);
392 /* No slaves enabled, so drop packet. */
399 if (tci || out_bundle->use_priority_tags) {
400 tci |= *flow_tci & htons(VLAN_PCP_MASK);
402 tci |= htons(VLAN_CFI);
407 compose_output_action(ctx, port->up.ofp_port);
411 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
412 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
413 * indicate this; newer upstream kernels use gratuitous ARP requests. */
415 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
417 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
421 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
422 if (!eth_addr_is_broadcast(flow->dl_dst)) {
426 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
427 if (flow->nw_proto == ARP_OP_REPLY) {
429 } else if (flow->nw_proto == ARP_OP_REQUEST) {
430 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
431 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
433 return flow->nw_src == flow->nw_dst;
440 update_learning_table(struct ofproto_dpif *ofproto,
441 const struct flow *flow, struct flow_wildcards *wc,
442 int vlan, struct ofbundle *in_bundle)
444 struct mac_entry *mac;
446 /* Don't learn the OFPP_NONE port. */
447 if (in_bundle == &ofpp_none_bundle) {
451 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
455 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
456 if (is_gratuitous_arp(flow, wc)) {
457 /* We don't want to learn from gratuitous ARP packets that are
458 * reflected back over bond slaves so we lock the learning table. */
459 if (!in_bundle->bond) {
460 mac_entry_set_grat_arp_lock(mac);
461 } else if (mac_entry_is_grat_arp_locked(mac)) {
466 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
467 /* The log messages here could actually be useful in debugging,
468 * so keep the rate limit relatively high. */
469 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
470 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
471 "on port %s in VLAN %d",
472 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
473 in_bundle->name, vlan);
475 mac->port.p = in_bundle;
476 mac_learning_changed(ofproto->ml, mac);
480 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
481 * dropped. Returns true if they may be forwarded, false if they should be
484 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
485 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
487 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
488 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
489 * checked by input_vid_is_valid().
491 * May also add tags to '*tags', although the current implementation only does
492 * so in one special case.
495 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
498 struct ofproto_dpif *ofproto = ctx->ofproto;
499 struct flow *flow = &ctx->xin->flow;
500 struct ofbundle *in_bundle = in_port->bundle;
502 /* Drop frames for reserved multicast addresses
503 * only if forward_bpdu option is absent. */
504 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
505 xlate_report(ctx, "packet has reserved destination MAC, dropping");
509 if (in_bundle->bond) {
510 struct mac_entry *mac;
512 switch (bond_check_admissibility(in_bundle->bond, in_port,
513 flow->dl_dst, &ctx->xout->tags)) {
518 xlate_report(ctx, "bonding refused admissibility, dropping");
521 case BV_DROP_IF_MOVED:
522 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
523 if (mac && mac->port.p != in_bundle &&
524 (!is_gratuitous_arp(flow, &ctx->xout->wc)
525 || mac_entry_is_grat_arp_locked(mac))) {
526 xlate_report(ctx, "SLB bond thinks this packet looped back, "
538 xlate_normal(struct xlate_ctx *ctx)
540 struct flow_wildcards *wc = &ctx->xout->wc;
541 struct flow *flow = &ctx->xin->flow;
542 struct ofport_dpif *in_port;
543 struct ofbundle *in_bundle;
544 struct mac_entry *mac;
548 ctx->xout->has_normal = true;
550 /* Check the dl_type, since we may check for gratuituous ARP. */
551 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
553 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
554 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
555 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
557 in_bundle = lookup_input_bundle(ctx->ofproto, flow->in_port,
558 ctx->xin->packet != NULL, &in_port);
560 xlate_report(ctx, "no input bundle, dropping");
564 /* Drop malformed frames. */
565 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
566 !(flow->vlan_tci & htons(VLAN_CFI))) {
567 if (ctx->xin->packet != NULL) {
568 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
569 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
570 "VLAN tag received on port %s",
571 ctx->ofproto->up.name, in_bundle->name);
573 xlate_report(ctx, "partial VLAN tag, dropping");
577 /* Drop frames on bundles reserved for mirroring. */
578 if (in_bundle->mirror_out) {
579 if (ctx->xin->packet != NULL) {
580 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
581 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
582 "%s, which is reserved exclusively for mirroring",
583 ctx->ofproto->up.name, in_bundle->name);
585 xlate_report(ctx, "input port is mirror output port, dropping");
590 vid = vlan_tci_to_vid(flow->vlan_tci);
591 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
592 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
595 vlan = input_vid_to_vlan(in_bundle, vid);
597 /* Check other admissibility requirements. */
598 if (in_port && !is_admissible(ctx, in_port, vlan)) {
602 /* Learn source MAC. */
603 if (ctx->xin->may_learn) {
604 update_learning_table(ctx->ofproto, flow, wc, vlan, in_bundle);
607 /* Determine output bundle. */
608 mac = mac_learning_lookup(ctx->ofproto->ml, flow->dl_dst, vlan,
611 if (mac->port.p != in_bundle) {
612 xlate_report(ctx, "forwarding to learned port");
613 output_normal(ctx, mac->port.p, vlan);
615 xlate_report(ctx, "learned port is input port, dropping");
618 struct ofbundle *bundle;
620 xlate_report(ctx, "no learned MAC for destination, flooding");
621 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
622 if (bundle != in_bundle
623 && ofbundle_includes_vlan(bundle, vlan)
625 && !bundle->mirror_out) {
626 output_normal(ctx, bundle, vlan);
629 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
633 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
634 * the number of packets out of UINT32_MAX to sample. The given
635 * cookie is passed back in the callback for each sampled packet.
638 compose_sample_action(const struct ofproto_dpif *ofproto,
639 struct ofpbuf *odp_actions,
640 const struct flow *flow,
641 const uint32_t probability,
642 const union user_action_cookie *cookie,
643 const size_t cookie_size)
645 size_t sample_offset, actions_offset;
648 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
650 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
652 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
653 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
656 nl_msg_end_nested(odp_actions, actions_offset);
657 nl_msg_end_nested(odp_actions, sample_offset);
658 return cookie_offset;
662 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
663 ovs_be16 vlan_tci, uint32_t odp_port,
664 unsigned int n_outputs, union user_action_cookie *cookie)
668 cookie->type = USER_ACTION_COOKIE_SFLOW;
669 cookie->sflow.vlan_tci = vlan_tci;
671 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
672 * port information") for the interpretation of cookie->output. */
675 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
676 cookie->sflow.output = 0x40000000 | 256;
680 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
682 cookie->sflow.output = ifindex;
687 /* 0x80000000 means "multiple output ports. */
688 cookie->sflow.output = 0x80000000 | n_outputs;
693 /* Compose SAMPLE action for sFlow bridge sampling. */
695 compose_sflow_action(const struct ofproto_dpif *ofproto,
696 struct ofpbuf *odp_actions,
697 const struct flow *flow,
700 uint32_t probability;
701 union user_action_cookie cookie;
703 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
707 probability = dpif_sflow_get_probability(ofproto->sflow);
708 compose_sflow_cookie(ofproto, htons(0), odp_port,
709 odp_port == OVSP_NONE ? 0 : 1, &cookie);
711 return compose_sample_action(ofproto, odp_actions, flow, probability,
712 &cookie, sizeof cookie.sflow);
716 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
717 uint32_t obs_domain_id, uint32_t obs_point_id,
718 union user_action_cookie *cookie)
720 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
721 cookie->flow_sample.probability = probability;
722 cookie->flow_sample.collector_set_id = collector_set_id;
723 cookie->flow_sample.obs_domain_id = obs_domain_id;
724 cookie->flow_sample.obs_point_id = obs_point_id;
728 compose_ipfix_cookie(union user_action_cookie *cookie)
730 cookie->type = USER_ACTION_COOKIE_IPFIX;
733 /* Compose SAMPLE action for IPFIX bridge sampling. */
735 compose_ipfix_action(const struct ofproto_dpif *ofproto,
736 struct ofpbuf *odp_actions,
737 const struct flow *flow)
739 uint32_t probability;
740 union user_action_cookie cookie;
742 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
746 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
747 compose_ipfix_cookie(&cookie);
749 compose_sample_action(ofproto, odp_actions, flow, probability,
750 &cookie, sizeof cookie.ipfix);
753 /* SAMPLE action for sFlow must be first action in any given list of
754 * actions. At this point we do not have all information required to
755 * build it. So try to build sample action as complete as possible. */
757 add_sflow_action(struct xlate_ctx *ctx)
759 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
760 &ctx->xout->odp_actions,
761 &ctx->xin->flow, OVSP_NONE);
762 ctx->sflow_odp_port = 0;
763 ctx->sflow_n_outputs = 0;
766 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
767 * of actions, eventually after the SAMPLE action for sFlow. */
769 add_ipfix_action(struct xlate_ctx *ctx)
771 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
775 /* Fix SAMPLE action according to data collected while composing ODP actions.
776 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
777 * USERSPACE action's user-cookie which is required for sflow. */
779 fix_sflow_action(struct xlate_ctx *ctx)
781 const struct flow *base = &ctx->base_flow;
782 union user_action_cookie *cookie;
784 if (!ctx->user_cookie_offset) {
788 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
789 sizeof cookie->sflow);
790 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
792 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
793 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
796 static enum slow_path_reason
797 process_special(struct xlate_ctx *ctx, const struct flow *flow,
798 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
800 struct ofproto_dpif *ofproto = ctx->ofproto;
801 struct flow_wildcards *wc = &ctx->xout->wc;
805 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
807 cfm_process_heartbeat(ofport->cfm, packet);
810 } else if (ofport->bfd && bfd_should_process_flow(flow, wc)) {
812 bfd_process_packet(ofport->bfd, flow, packet);
815 } else if (ofport->bundle && ofport->bundle->lacp
816 && flow->dl_type == htons(ETH_TYPE_LACP)) {
818 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
821 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
823 stp_process_packet(ofport, packet);
832 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
835 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
836 struct flow *flow = &ctx->xin->flow;
837 ovs_be16 flow_vlan_tci;
838 uint32_t flow_skb_mark;
840 uint32_t out_port, odp_port;
843 /* If 'struct flow' gets additional metadata, we'll need to zero it out
844 * before traversing a patch port. */
845 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
848 xlate_report(ctx, "Nonexistent output port");
850 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
851 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
853 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
854 xlate_report(ctx, "STP not in forwarding state, skipping output");
859 struct ofport_dpif *peer = ofport->peer;
860 struct flow old_flow = ctx->xin->flow;
861 enum slow_path_reason special;
863 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
864 flow->in_port = peer->up.ofp_port;
865 flow->metadata = htonll(0);
866 memset(&flow->tunnel, 0, sizeof flow->tunnel);
867 memset(flow->regs, 0, sizeof flow->regs);
869 special = process_special(ctx, &ctx->xin->flow, peer,
872 ctx->xout->slow = special;
873 } else if (may_receive(peer, ctx)) {
874 if (stp_forward_in_state(peer->stp_state)) {
875 xlate_table_action(ctx, flow->in_port, 0, true);
877 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
878 * learning action look at the packet, then drop it. */
879 struct flow old_base_flow = ctx->base_flow;
880 size_t old_size = ctx->xout->odp_actions.size;
881 xlate_table_action(ctx, flow->in_port, 0, true);
882 ctx->base_flow = old_base_flow;
883 ctx->xout->odp_actions.size = old_size;
887 ctx->xin->flow = old_flow;
888 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
890 if (ctx->xin->resubmit_stats) {
891 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
892 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
898 flow_vlan_tci = flow->vlan_tci;
899 flow_skb_mark = flow->skb_mark;
900 flow_nw_tos = flow->nw_tos;
902 if (ofproto_dpif_dscp_from_priority(ofport, flow->skb_priority, &dscp)) {
903 flow->nw_tos &= ~IP_DSCP_MASK;
904 flow->nw_tos |= dscp;
907 if (ofport->tnl_port) {
908 /* Save tunnel metadata so that changes made due to
909 * the Logical (tunnel) Port are not visible for any further
910 * matches, while explicit set actions on tunnel metadata are.
912 struct flow_tnl flow_tnl = flow->tunnel;
913 odp_port = tnl_port_send(ofport->tnl_port, flow, &ctx->xout->wc);
914 if (odp_port == OVSP_NONE) {
915 xlate_report(ctx, "Tunneling decided against output");
916 goto out; /* restore flow_nw_tos */
918 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
919 xlate_report(ctx, "Not tunneling to our own address");
920 goto out; /* restore flow_nw_tos */
922 if (ctx->xin->resubmit_stats) {
923 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
926 commit_odp_tunnel_action(flow, &ctx->base_flow,
927 &ctx->xout->odp_actions);
928 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
930 uint16_t vlandev_port;
931 odp_port = ofport->odp_port;
932 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
934 if (vlandev_port == ofp_port) {
937 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
938 flow->vlan_tci = htons(0);
940 flow->skb_mark &= ~IPSEC_MARK;
943 if (out_port != OVSP_NONE) {
944 commit_odp_actions(flow, &ctx->base_flow, &ctx->xout->odp_actions);
945 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
948 ctx->sflow_odp_port = odp_port;
949 ctx->sflow_n_outputs++;
950 ctx->xout->nf_output_iface = ofp_port;
955 flow->vlan_tci = flow_vlan_tci;
956 flow->skb_mark = flow_skb_mark;
957 flow->nw_tos = flow_nw_tos;
961 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
963 compose_output_action__(ctx, ofp_port, true);
967 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
969 struct ofproto_dpif *ofproto = ctx->ofproto;
970 uint8_t table_id = ctx->table_id;
972 if (table_id > 0 && table_id < N_TABLES) {
973 struct table_dpif *table = &ofproto->tables[table_id];
974 if (table->other_table) {
975 ctx->xout->tags |= (rule && rule->tag
977 : rule_calculate_tag(&ctx->xin->flow,
978 &table->other_table->mask,
984 /* Common rule processing in one place to avoid duplicating code. */
985 static struct rule_dpif *
986 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
989 if (ctx->xin->resubmit_hook) {
990 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
992 if (rule == NULL && may_packet_in) {
994 * check if table configuration flags
995 * OFPTC_TABLE_MISS_CONTROLLER, default.
996 * OFPTC_TABLE_MISS_CONTINUE,
997 * OFPTC_TABLE_MISS_DROP
998 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
1000 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
1002 if (rule && ctx->xin->resubmit_stats) {
1003 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1009 xlate_table_action(struct xlate_ctx *ctx,
1010 uint16_t in_port, uint8_t table_id, bool may_packet_in)
1012 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1013 struct rule_dpif *rule;
1014 uint16_t old_in_port = ctx->xin->flow.in_port;
1015 uint8_t old_table_id = ctx->table_id;
1017 ctx->table_id = table_id;
1019 /* Look up a flow with 'in_port' as the input port. */
1020 ctx->xin->flow.in_port = in_port;
1021 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
1022 &ctx->xout->wc, table_id);
1024 tag_the_flow(ctx, rule);
1026 /* Restore the original input port. Otherwise OFPP_NORMAL and
1027 * OFPP_IN_PORT will have surprising behavior. */
1028 ctx->xin->flow.in_port = old_in_port;
1030 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1033 struct rule_dpif *old_rule = ctx->rule;
1037 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1038 ctx->rule = old_rule;
1042 ctx->table_id = old_table_id;
1044 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1046 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1047 MAX_RESUBMIT_RECURSION);
1048 ctx->max_resubmit_trigger = true;
1053 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1054 const struct ofpact_resubmit *resubmit)
1059 in_port = resubmit->in_port;
1060 if (in_port == OFPP_IN_PORT) {
1061 in_port = ctx->xin->flow.in_port;
1064 table_id = resubmit->table_id;
1065 if (table_id == 255) {
1066 table_id = ctx->table_id;
1069 xlate_table_action(ctx, in_port, table_id, false);
1073 flood_packets(struct xlate_ctx *ctx, bool all)
1075 struct ofport_dpif *ofport;
1077 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
1078 uint16_t ofp_port = ofport->up.ofp_port;
1080 if (ofp_port == ctx->xin->flow.in_port) {
1085 compose_output_action__(ctx, ofp_port, false);
1086 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
1087 compose_output_action(ctx, ofp_port);
1091 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1095 execute_controller_action(struct xlate_ctx *ctx, int len,
1096 enum ofp_packet_in_reason reason,
1097 uint16_t controller_id)
1099 struct ofputil_packet_in pin;
1100 struct ofpbuf *packet;
1103 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1104 ctx->xout->slow = SLOW_CONTROLLER;
1105 if (!ctx->xin->packet) {
1109 packet = ofpbuf_clone(ctx->xin->packet);
1111 key.skb_priority = 0;
1113 memset(&key.tunnel, 0, sizeof key.tunnel);
1115 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1116 &ctx->xout->odp_actions);
1118 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1119 ctx->xout->odp_actions.size, NULL, NULL);
1121 pin.packet = packet->data;
1122 pin.packet_len = packet->size;
1123 pin.reason = reason;
1124 pin.controller_id = controller_id;
1125 pin.table_id = ctx->table_id;
1126 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1129 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1131 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
1132 ofpbuf_delete(packet);
1136 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1138 struct flow_wildcards *wc = &ctx->xout->wc;
1139 struct flow *flow = &ctx->xin->flow;
1141 ovs_assert(eth_type_mpls(eth_type));
1143 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1144 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1145 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1147 if (flow->mpls_depth) {
1148 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1154 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1155 label = htonl(0x2); /* IPV6 Explicit Null. */
1157 label = htonl(0x0); /* IPV4 Explicit Null. */
1159 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1160 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1161 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1162 flow->mpls_depth = 1;
1164 flow->dl_type = eth_type;
1168 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1170 struct flow_wildcards *wc = &ctx->xout->wc;
1171 struct flow *flow = &ctx->xin->flow;
1173 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1174 ovs_assert(!eth_type_mpls(eth_type));
1176 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1177 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1178 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1180 if (flow->mpls_depth) {
1182 flow->mpls_lse = htonl(0);
1183 if (!flow->mpls_depth) {
1184 flow->dl_type = eth_type;
1190 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1192 struct flow *flow = &ctx->xin->flow;
1194 if (!is_ip_any(flow)) {
1198 if (flow->nw_ttl > 1) {
1204 for (i = 0; i < ids->n_controllers; i++) {
1205 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1209 /* Stop processing for current table. */
1215 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1217 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1221 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1226 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1228 struct flow *flow = &ctx->xin->flow;
1229 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1231 if (!eth_type_mpls(flow->dl_type)) {
1237 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1240 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1242 /* Stop processing for current table. */
1248 xlate_output_action(struct xlate_ctx *ctx,
1249 uint16_t port, uint16_t max_len, bool may_packet_in)
1251 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1253 ctx->xout->nf_output_iface = NF_OUT_DROP;
1257 compose_output_action(ctx, ctx->xin->flow.in_port);
1260 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
1266 flood_packets(ctx, false);
1269 flood_packets(ctx, true);
1271 case OFPP_CONTROLLER:
1272 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1278 if (port != ctx->xin->flow.in_port) {
1279 compose_output_action(ctx, port);
1281 xlate_report(ctx, "skipping output to input port");
1286 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1287 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1288 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1289 ctx->xout->nf_output_iface = prev_nf_output_iface;
1290 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1291 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1292 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1297 xlate_output_reg_action(struct xlate_ctx *ctx,
1298 const struct ofpact_output_reg *or)
1300 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1301 if (port <= UINT16_MAX) {
1302 union mf_subvalue value;
1304 memset(&value, 0xff, sizeof value);
1305 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1306 xlate_output_action(ctx, port, or->max_len, false);
1311 xlate_enqueue_action(struct xlate_ctx *ctx,
1312 const struct ofpact_enqueue *enqueue)
1314 uint16_t ofp_port = enqueue->port;
1315 uint32_t queue_id = enqueue->queue;
1316 uint32_t flow_priority, priority;
1319 /* Translate queue to priority. */
1320 error = ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id, &priority);
1322 /* Fall back to ordinary output action. */
1323 xlate_output_action(ctx, enqueue->port, 0, false);
1327 /* Check output port. */
1328 if (ofp_port == OFPP_IN_PORT) {
1329 ofp_port = ctx->xin->flow.in_port;
1330 } else if (ofp_port == ctx->xin->flow.in_port) {
1334 /* Add datapath actions. */
1335 flow_priority = ctx->xin->flow.skb_priority;
1336 ctx->xin->flow.skb_priority = priority;
1337 compose_output_action(ctx, ofp_port);
1338 ctx->xin->flow.skb_priority = flow_priority;
1340 /* Update NetFlow output port. */
1341 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1342 ctx->xout->nf_output_iface = ofp_port;
1343 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1344 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1349 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1351 uint32_t skb_priority;
1353 if (!ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id,
1355 ctx->xin->flow.skb_priority = skb_priority;
1357 /* Couldn't translate queue to a priority. Nothing to do. A warning
1358 * has already been logged. */
1363 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
1365 struct ofproto_dpif *ofproto = ofproto_;
1366 struct ofport_dpif *port;
1376 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1379 port = get_ofp_port(ofproto, ofp_port);
1380 return port ? port->may_enable : false;
1385 xlate_bundle_action(struct xlate_ctx *ctx,
1386 const struct ofpact_bundle *bundle)
1390 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1391 slave_enabled_cb, ctx->ofproto);
1392 if (bundle->dst.field) {
1393 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
1395 xlate_output_action(ctx, port, 0, false);
1400 xlate_learn_action(struct xlate_ctx *ctx,
1401 const struct ofpact_learn *learn)
1403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1404 struct ofputil_flow_mod fm;
1405 uint64_t ofpacts_stub[1024 / 8];
1406 struct ofpbuf ofpacts;
1409 ctx->xout->has_learn = true;
1411 learn_mask(learn, &ctx->xout->wc);
1413 if (!ctx->xin->may_learn) {
1417 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1418 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1420 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
1421 if (error && !VLOG_DROP_WARN(&rl)) {
1422 VLOG_WARN("learning action failed to modify flow table (%s)",
1423 ofperr_get_name(error));
1426 ofpbuf_uninit(&ofpacts);
1429 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1430 * means "infinite". */
1432 reduce_timeout(uint16_t max, uint16_t *timeout)
1434 if (max && (!*timeout || *timeout > max)) {
1440 xlate_fin_timeout(struct xlate_ctx *ctx,
1441 const struct ofpact_fin_timeout *oft)
1443 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1444 struct rule_dpif *rule = ctx->rule;
1446 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1447 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1452 xlate_sample_action(struct xlate_ctx *ctx,
1453 const struct ofpact_sample *os)
1455 union user_action_cookie cookie;
1456 /* Scale the probability from 16-bit to 32-bit while representing
1457 * the same percentage. */
1458 uint32_t probability = (os->probability << 16) | os->probability;
1460 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1461 &ctx->xout->odp_actions);
1463 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1464 os->obs_domain_id, os->obs_point_id, &cookie);
1465 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
1466 probability, &cookie, sizeof cookie.flow_sample);
1470 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
1472 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
1474 ? OFPUTIL_PC_NO_RECV_STP
1475 : OFPUTIL_PC_NO_RECV)) {
1479 /* Only drop packets here if both forwarding and learning are
1480 * disabled. If just learning is enabled, we need to have
1481 * OFPP_NORMAL and the learning action have a look at the packet
1482 * before we can drop it. */
1483 if (!stp_forward_in_state(port->stp_state)
1484 && !stp_learn_in_state(port->stp_state)) {
1492 tunnel_ecn_ok(struct xlate_ctx *ctx)
1494 if (is_ip_any(&ctx->base_flow)
1495 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1496 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1497 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1498 " but is not ECN capable");
1501 /* Set the ECN CE value in the tunneled packet. */
1502 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1510 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1511 struct xlate_ctx *ctx)
1513 struct flow_wildcards *wc = &ctx->xout->wc;
1514 struct flow *flow = &ctx->xin->flow;
1515 bool was_evictable = true;
1516 const struct ofpact *a;
1519 /* Don't let the rule we're working on get evicted underneath us. */
1520 was_evictable = ctx->rule->up.evictable;
1521 ctx->rule->up.evictable = false;
1524 do_xlate_actions_again:
1525 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1526 struct ofpact_controller *controller;
1527 const struct ofpact_metadata *metadata;
1535 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1536 ofpact_get_OUTPUT(a)->max_len, true);
1539 case OFPACT_CONTROLLER:
1540 controller = ofpact_get_CONTROLLER(a);
1541 execute_controller_action(ctx, controller->max_len,
1543 controller->controller_id);
1546 case OFPACT_ENQUEUE:
1547 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1550 case OFPACT_SET_VLAN_VID:
1551 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1552 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1556 case OFPACT_SET_VLAN_PCP:
1557 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1559 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1563 case OFPACT_STRIP_VLAN:
1564 flow->vlan_tci = htons(0);
1567 case OFPACT_PUSH_VLAN:
1568 /* XXX 802.1AD(QinQ) */
1569 flow->vlan_tci = htons(VLAN_CFI);
1572 case OFPACT_SET_ETH_SRC:
1573 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
1576 case OFPACT_SET_ETH_DST:
1577 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
1580 case OFPACT_SET_IPV4_SRC:
1581 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1582 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1583 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1587 case OFPACT_SET_IPV4_DST:
1588 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1589 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1590 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1594 case OFPACT_SET_IPV4_DSCP:
1595 /* OpenFlow 1.0 only supports IPv4. */
1596 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1597 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1598 flow->nw_tos &= ~IP_DSCP_MASK;
1599 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1603 case OFPACT_SET_L4_SRC_PORT:
1604 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1605 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1606 if (is_ip_any(flow)) {
1607 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1611 case OFPACT_SET_L4_DST_PORT:
1612 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1613 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1614 if (is_ip_any(flow)) {
1615 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
1619 case OFPACT_RESUBMIT:
1620 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
1623 case OFPACT_SET_TUNNEL:
1624 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
1627 case OFPACT_SET_QUEUE:
1628 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
1631 case OFPACT_POP_QUEUE:
1632 memset(&wc->masks.skb_priority, 0xff,
1633 sizeof wc->masks.skb_priority);
1635 flow->skb_priority = ctx->orig_skb_priority;
1638 case OFPACT_REG_MOVE:
1639 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
1642 case OFPACT_REG_LOAD:
1643 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
1646 case OFPACT_STACK_PUSH:
1647 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
1651 case OFPACT_STACK_POP:
1652 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
1655 case OFPACT_PUSH_MPLS:
1656 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
1659 case OFPACT_POP_MPLS:
1660 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
1663 case OFPACT_SET_MPLS_TTL:
1664 if (compose_set_mpls_ttl_action(ctx,
1665 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
1670 case OFPACT_DEC_MPLS_TTL:
1671 if (compose_dec_mpls_ttl_action(ctx)) {
1676 case OFPACT_DEC_TTL:
1677 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1678 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
1684 /* Nothing to do. */
1687 case OFPACT_MULTIPATH:
1688 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
1692 ctx->ofproto->has_bundle_action = true;
1693 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
1696 case OFPACT_OUTPUT_REG:
1697 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
1701 xlate_learn_action(ctx, ofpact_get_LEARN(a));
1708 case OFPACT_FIN_TIMEOUT:
1709 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1710 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1711 ctx->xout->has_fin_timeout = true;
1712 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
1715 case OFPACT_CLEAR_ACTIONS:
1717 * Nothing to do because writa-actions is not supported for now.
1718 * When writa-actions is supported, clear-actions also must
1719 * be supported at the same time.
1723 case OFPACT_WRITE_METADATA:
1724 metadata = ofpact_get_WRITE_METADATA(a);
1725 flow->metadata &= ~metadata->mask;
1726 flow->metadata |= metadata->metadata & metadata->mask;
1729 case OFPACT_GOTO_TABLE: {
1730 /* It is assumed that goto-table is the last action. */
1731 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
1732 struct rule_dpif *rule;
1734 ovs_assert(ctx->table_id < ogt->table_id);
1736 ctx->table_id = ogt->table_id;
1738 /* Look up a flow from the new table. */
1739 rule = rule_dpif_lookup_in_table(ctx->ofproto, flow, wc,
1742 tag_the_flow(ctx, rule);
1744 rule = ctx_rule_hooks(ctx, rule, true);
1748 ctx->rule->up.evictable = was_evictable;
1751 was_evictable = rule->up.evictable;
1752 rule->up.evictable = false;
1754 /* Tail recursion removal. */
1755 ofpacts = rule->up.ofpacts;
1756 ofpacts_len = rule->up.ofpacts_len;
1757 goto do_xlate_actions_again;
1763 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
1770 ctx->rule->up.evictable = was_evictable;
1775 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
1776 const struct flow *flow, struct rule_dpif *rule,
1777 uint8_t tcp_flags, const struct ofpbuf *packet)
1779 xin->ofproto = ofproto;
1781 xin->packet = packet;
1782 xin->may_learn = packet != NULL;
1784 xin->ofpacts = NULL;
1785 xin->ofpacts_len = 0;
1786 xin->tcp_flags = tcp_flags;
1787 xin->resubmit_hook = NULL;
1788 xin->report_hook = NULL;
1789 xin->resubmit_stats = NULL;
1793 xlate_out_uninit(struct xlate_out *xout)
1796 ofpbuf_uninit(&xout->odp_actions);
1800 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
1801 * into datapath actions, using 'ctx', and discards the datapath actions. */
1803 xlate_actions_for_side_effects(struct xlate_in *xin)
1805 struct xlate_out xout;
1807 xlate_actions(xin, &xout);
1808 xlate_out_uninit(&xout);
1812 xlate_report(struct xlate_ctx *ctx, const char *s)
1814 if (ctx->xin->report_hook) {
1815 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
1820 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
1823 dst->tags = src->tags;
1824 dst->slow = src->slow;
1825 dst->has_learn = src->has_learn;
1826 dst->has_normal = src->has_normal;
1827 dst->has_fin_timeout = src->has_fin_timeout;
1828 dst->nf_output_iface = src->nf_output_iface;
1829 dst->mirrors = src->mirrors;
1831 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
1832 sizeof dst->odp_actions_stub);
1833 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
1834 src->odp_actions.size);
1838 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
1839 * into datapath actions in 'odp_actions', using 'ctx'. */
1841 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
1843 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
1844 * that in the future we always keep a copy of the original flow for
1845 * tracing purposes. */
1846 static bool hit_resubmit_limit;
1848 struct flow_wildcards *wc = &xout->wc;
1849 struct flow *flow = &xin->flow;
1851 enum slow_path_reason special;
1852 const struct ofpact *ofpacts;
1853 struct ofport_dpif *in_port;
1854 struct flow orig_flow;
1855 struct xlate_ctx ctx;
1858 COVERAGE_INC(ofproto_dpif_xlate);
1860 /* Flow initialization rules:
1861 * - 'base_flow' must match the kernel's view of the packet at the
1862 * time that action processing starts. 'flow' represents any
1863 * transformations we wish to make through actions.
1864 * - By default 'base_flow' and 'flow' are the same since the input
1865 * packet matches the output before any actions are applied.
1866 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
1867 * of the received packet as seen by the kernel. If we later output
1868 * to another device without any modifications this will cause us to
1869 * insert a new tag since the original one was stripped off by the
1871 * - Tunnel metadata as received is retained in 'flow'. This allows
1872 * tunnel metadata matching also in later tables.
1873 * Since a kernel action for setting the tunnel metadata will only be
1874 * generated with actual tunnel output, changing the tunnel metadata
1875 * values in 'flow' (such as tun_id) will only have effect with a later
1876 * tunnel output action.
1877 * - Tunnel 'base_flow' is completely cleared since that is what the
1878 * kernel does. If we wish to maintain the original values an action
1879 * needs to be generated. */
1884 ctx.ofproto = xin->ofproto;
1885 ctx.rule = xin->rule;
1887 ctx.base_flow = *flow;
1888 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
1889 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
1891 flow_wildcards_init_catchall(wc);
1892 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
1894 if (tnl_port_should_receive(&ctx.xin->flow)) {
1895 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
1898 /* Disable most wildcarding for NetFlow. */
1899 if (xin->ofproto->netflow) {
1900 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1901 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1902 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1903 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
1904 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1905 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1906 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1907 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
1908 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
1913 ctx.xout->has_learn = false;
1914 ctx.xout->has_normal = false;
1915 ctx.xout->has_fin_timeout = false;
1916 ctx.xout->nf_output_iface = NF_OUT_DROP;
1917 ctx.xout->mirrors = 0;
1919 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
1920 sizeof ctx.xout->odp_actions_stub);
1921 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
1924 ctx.max_resubmit_trigger = false;
1925 ctx.orig_skb_priority = flow->skb_priority;
1930 ofpacts = xin->ofpacts;
1931 ofpacts_len = xin->ofpacts_len;
1932 } else if (xin->rule) {
1933 ofpacts = xin->rule->up.ofpacts;
1934 ofpacts_len = xin->rule->up.ofpacts_len;
1939 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
1941 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
1942 /* Do this conditionally because the copy is expensive enough that it
1943 * shows up in profiles. */
1947 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1948 switch (ctx.ofproto->up.frag_handling) {
1949 case OFPC_FRAG_NORMAL:
1950 /* We must pretend that transport ports are unavailable. */
1951 flow->tp_src = ctx.base_flow.tp_src = htons(0);
1952 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
1955 case OFPC_FRAG_DROP:
1958 case OFPC_FRAG_REASM:
1961 case OFPC_FRAG_NX_MATCH:
1962 /* Nothing to do. */
1965 case OFPC_INVALID_TTL_TO_CONTROLLER:
1970 in_port = get_ofp_port(ctx.ofproto, flow->in_port);
1971 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
1973 ctx.xout->slow = special;
1975 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1976 size_t sample_actions_len;
1977 uint32_t local_odp_port;
1980 != vsp_realdev_to_vlandev(ctx.ofproto, flow->in_port,
1982 ctx.base_flow.vlan_tci = 0;
1985 add_sflow_action(&ctx);
1986 add_ipfix_action(&ctx);
1987 sample_actions_len = ctx.xout->odp_actions.size;
1989 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
1990 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
1992 /* We've let OFPP_NORMAL and the learning action look at the
1993 * packet, so drop it now if forwarding is disabled. */
1994 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
1995 ctx.xout->odp_actions.size = sample_actions_len;
1999 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
2000 if (!hit_resubmit_limit) {
2001 /* We didn't record the original flow. Make sure we do from
2003 hit_resubmit_limit = true;
2004 } else if (!VLOG_DROP_ERR(&trace_rl)) {
2005 struct ds ds = DS_EMPTY_INITIALIZER;
2007 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, &ds);
2008 VLOG_ERR("Trace triggered by excessive resubmit "
2009 "recursion:\n%s", ds_cstr(&ds));
2014 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
2015 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, flow,
2017 ctx.xout->odp_actions.data,
2018 ctx.xout->odp_actions.size)) {
2019 compose_output_action(&ctx, OFPP_LOCAL);
2021 if (ctx.ofproto->has_mirrors) {
2022 add_mirror_actions(&ctx, &orig_flow);
2024 fix_sflow_action(&ctx);
2027 ofpbuf_uninit(&ctx.stack);
2029 /* Clear the metadata and register wildcard masks, because we won't
2030 * use non-header fields as part of the cache. */
2031 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2032 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);