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"
32 #include "mac-learning.h"
33 #include "meta-flow.h"
34 #include "multipath.h"
35 #include "netdev-vport.h"
38 #include "odp-execute.h"
39 #include "ofp-actions.h"
40 #include "ofproto/ofproto-dpif-ipfix.h"
41 #include "ofproto/ofproto-dpif-mirror.h"
42 #include "ofproto/ofproto-dpif-sflow.h"
43 #include "ofproto/ofproto-dpif.h"
47 COVERAGE_DEFINE(ofproto_dpif_xlate);
49 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
51 /* Maximum depth of flow table recursion (due to resubmit actions) in a
52 * flow translation. */
53 #define MAX_RESUBMIT_RECURSION 64
57 struct xlate_out *xout;
59 struct ofproto_dpif *ofproto;
61 /* Flow at the last commit. */
62 struct flow base_flow;
64 /* Tunnel IP destination address as received. This is stored separately
65 * as the base_flow.tunnel is cleared on init to reflect the datapath
66 * behavior. Used to make sure not to send tunneled output to ourselves,
67 * which might lead to an infinite loop. This could happen easily
68 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
69 * actually set the tun_dst field. */
70 ovs_be32 orig_tunnel_ip_dst;
72 /* Stack for the push and pop actions. Each stack element is of type
73 * "union mf_subvalue". */
74 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
77 /* The rule that we are currently translating, or NULL. */
78 struct rule_dpif *rule;
80 int recurse; /* Recursion level, via xlate_table_action. */
81 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
82 uint32_t orig_skb_priority; /* Priority when packet arrived. */
83 uint8_t table_id; /* OpenFlow table ID where flow was found. */
84 uint32_t sflow_n_outputs; /* Number of output ports. */
85 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
86 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
87 bool exit; /* No further actions should be processed. */
90 /* A controller may use OFPP_NONE as the ingress port to indicate that
91 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
92 * when an input bundle is needed for validation (e.g., mirroring or
93 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
94 * any 'port' structs, so care must be taken when dealing with it. */
95 static struct ofbundle ofpp_none_bundle = {
97 .vlan_mode = PORT_VLAN_TRUNK
100 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
101 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
103 static void xlate_normal(struct xlate_ctx *);
104 static void xlate_report(struct xlate_ctx *, const char *);
105 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
106 uint8_t table_id, bool may_packet_in);
107 static bool input_vid_is_valid(uint16_t vid, struct ofbundle *, bool warn);
108 static uint16_t input_vid_to_vlan(const struct ofbundle *, uint16_t vid);
109 static void output_normal(struct xlate_ctx *, const struct ofbundle *,
111 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
116 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
118 return (bundle->vlan_mode != PORT_VLAN_ACCESS
119 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
123 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
125 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
128 static struct ofbundle *
129 lookup_input_bundle(const struct ofproto_dpif *ofproto, ofp_port_t in_port,
130 bool warn, struct ofport_dpif **in_ofportp)
132 struct ofport_dpif *ofport;
134 /* Find the port and bundle for the received packet. */
135 ofport = get_ofp_port(ofproto, in_port);
137 *in_ofportp = ofport;
139 if (ofport && ofport->bundle) {
140 return ofport->bundle;
143 /* Special-case OFPP_NONE, which a controller may use as the ingress
144 * port for traffic that it is sourcing. */
145 if (in_port == OFPP_NONE) {
146 return &ofpp_none_bundle;
149 /* Odd. A few possible reasons here:
151 * - We deleted a port but there are still a few packets queued up
154 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
155 * we don't know about.
157 * - The ofproto client didn't configure the port as part of a bundle.
158 * This is particularly likely to happen if a packet was received on the
159 * port after it was created, but before the client had a chance to
160 * configure its bundle.
163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
165 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
166 "port %"PRIu16, ofproto->up.name, in_port);
172 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
174 struct ofproto_dpif *ofproto = ctx->ofproto;
175 mirror_mask_t mirrors;
176 struct ofbundle *in_bundle;
180 mirrors = ctx->xout->mirrors;
181 ctx->xout->mirrors = 0;
183 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port.ofp_port,
184 ctx->xin->packet != NULL, NULL);
188 mirrors |= mirror_bundle_src(ctx->ofproto->mbridge, in_bundle);
190 /* Drop frames on bundles reserved for mirroring. */
191 if (mirror_bundle_out(ctx->ofproto->mbridge, in_bundle)) {
192 if (ctx->xin->packet != NULL) {
193 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
194 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
195 "%s, which is reserved exclusively for mirroring",
196 ctx->ofproto->up.name, in_bundle->name);
198 ofpbuf_clear(&ctx->xout->odp_actions);
203 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
204 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
207 vlan = input_vid_to_vlan(in_bundle, vid);
213 /* Restore the original packet before adding the mirror actions. */
214 ctx->xin->flow = *orig_flow;
217 mirror_mask_t dup_mirrors;
218 struct ofbundle *out;
219 unsigned long *vlans;
224 has_mirror = mirror_get(ofproto->mbridge, mirror_mask_ffs(mirrors) - 1,
225 &vlans, &dup_mirrors, &out, &out_vlan);
226 ovs_assert(has_mirror);
229 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
231 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
234 if (!vlan_mirrored) {
235 mirrors = zero_rightmost_1bit(mirrors);
239 mirrors &= ~dup_mirrors;
240 ctx->xout->mirrors |= dup_mirrors;
242 output_normal(ctx, out, vlan);
243 } else if (vlan != out_vlan
244 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
245 struct ofbundle *bundle;
247 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
248 if (ofbundle_includes_vlan(bundle, out_vlan)
249 && !mirror_bundle_out(bundle->ofproto->mbridge, bundle)) {
250 output_normal(ctx, bundle, out_vlan);
257 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
258 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
259 * the bundle on which the packet was received, returns the VLAN to which the
262 * Both 'vid' and the return value are in the range 0...4095. */
264 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
266 switch (in_bundle->vlan_mode) {
267 case PORT_VLAN_ACCESS:
268 return in_bundle->vlan;
271 case PORT_VLAN_TRUNK:
274 case PORT_VLAN_NATIVE_UNTAGGED:
275 case PORT_VLAN_NATIVE_TAGGED:
276 return vid ? vid : in_bundle->vlan;
283 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
284 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
287 * 'vid' should be the VID obtained from the 802.1Q header that was received as
288 * part of a packet (specify 0 if there was no 802.1Q header), in the range
291 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
293 /* Allow any VID on the OFPP_NONE port. */
294 if (in_bundle == &ofpp_none_bundle) {
298 switch (in_bundle->vlan_mode) {
299 case PORT_VLAN_ACCESS:
302 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
303 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
304 "packet received on port %s configured as VLAN "
305 "%"PRIu16" access port",
306 in_bundle->ofproto->up.name, vid,
307 in_bundle->name, in_bundle->vlan);
313 case PORT_VLAN_NATIVE_UNTAGGED:
314 case PORT_VLAN_NATIVE_TAGGED:
316 /* Port must always carry its native VLAN. */
320 case PORT_VLAN_TRUNK:
321 if (!ofbundle_includes_vlan(in_bundle, vid)) {
323 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
324 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
325 "received on port %s not configured for trunking "
327 in_bundle->ofproto->up.name, vid,
328 in_bundle->name, vid);
340 /* Given 'vlan', the VLAN that a packet belongs to, and
341 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
342 * that should be included in the 802.1Q header. (If the return value is 0,
343 * then the 802.1Q header should only be included in the packet if there is a
346 * Both 'vlan' and the return value are in the range 0...4095. */
348 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
350 switch (out_bundle->vlan_mode) {
351 case PORT_VLAN_ACCESS:
354 case PORT_VLAN_TRUNK:
355 case PORT_VLAN_NATIVE_TAGGED:
358 case PORT_VLAN_NATIVE_UNTAGGED:
359 return vlan == out_bundle->vlan ? 0 : vlan;
367 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
370 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
371 struct ofport_dpif *port;
373 ovs_be16 tci, old_tci;
375 vid = output_vlan_to_vid(out_bundle, vlan);
376 if (!out_bundle->bond) {
377 port = ofbundle_get_a_port(out_bundle);
379 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
380 &ctx->xout->wc, vid, &ctx->xout->tags);
382 /* No slaves enabled, so drop packet. */
389 if (tci || out_bundle->use_priority_tags) {
390 tci |= *flow_tci & htons(VLAN_PCP_MASK);
392 tci |= htons(VLAN_CFI);
397 compose_output_action(ctx, port->up.ofp_port);
401 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
402 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
403 * indicate this; newer upstream kernels use gratuitous ARP requests. */
405 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
407 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
411 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
412 if (!eth_addr_is_broadcast(flow->dl_dst)) {
416 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
417 if (flow->nw_proto == ARP_OP_REPLY) {
419 } else if (flow->nw_proto == ARP_OP_REQUEST) {
420 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
421 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
423 return flow->nw_src == flow->nw_dst;
430 update_learning_table(struct ofproto_dpif *ofproto,
431 const struct flow *flow, struct flow_wildcards *wc,
432 int vlan, struct ofbundle *in_bundle)
434 struct mac_entry *mac;
436 /* Don't learn the OFPP_NONE port. */
437 if (in_bundle == &ofpp_none_bundle) {
441 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
445 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
446 if (is_gratuitous_arp(flow, wc)) {
447 /* We don't want to learn from gratuitous ARP packets that are
448 * reflected back over bond slaves so we lock the learning table. */
449 if (!in_bundle->bond) {
450 mac_entry_set_grat_arp_lock(mac);
451 } else if (mac_entry_is_grat_arp_locked(mac)) {
456 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
457 /* The log messages here could actually be useful in debugging,
458 * so keep the rate limit relatively high. */
459 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
460 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
461 "on port %s in VLAN %d",
462 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
463 in_bundle->name, vlan);
465 mac->port.p = in_bundle;
466 mac_learning_changed(ofproto->ml, mac);
470 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
471 * dropped. Returns true if they may be forwarded, false if they should be
474 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
475 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
477 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
478 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
479 * checked by input_vid_is_valid().
481 * May also add tags to '*tags', although the current implementation only does
482 * so in one special case.
485 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
488 struct ofproto_dpif *ofproto = ctx->ofproto;
489 struct flow *flow = &ctx->xin->flow;
490 struct ofbundle *in_bundle = in_port->bundle;
492 /* Drop frames for reserved multicast addresses
493 * only if forward_bpdu option is absent. */
494 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
495 xlate_report(ctx, "packet has reserved destination MAC, dropping");
499 if (in_bundle->bond) {
500 struct mac_entry *mac;
502 switch (bond_check_admissibility(in_bundle->bond, in_port,
503 flow->dl_dst, &ctx->xout->tags)) {
508 xlate_report(ctx, "bonding refused admissibility, dropping");
511 case BV_DROP_IF_MOVED:
512 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
513 if (mac && mac->port.p != in_bundle &&
514 (!is_gratuitous_arp(flow, &ctx->xout->wc)
515 || mac_entry_is_grat_arp_locked(mac))) {
516 xlate_report(ctx, "SLB bond thinks this packet looped back, "
528 xlate_normal(struct xlate_ctx *ctx)
530 struct flow_wildcards *wc = &ctx->xout->wc;
531 struct flow *flow = &ctx->xin->flow;
532 struct ofport_dpif *in_port;
533 struct ofbundle *in_bundle;
534 struct mac_entry *mac;
538 ctx->xout->has_normal = true;
540 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
541 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
542 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
544 in_bundle = lookup_input_bundle(ctx->ofproto, flow->in_port.ofp_port,
545 ctx->xin->packet != NULL, &in_port);
547 xlate_report(ctx, "no input bundle, dropping");
551 /* Drop malformed frames. */
552 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
553 !(flow->vlan_tci & htons(VLAN_CFI))) {
554 if (ctx->xin->packet != NULL) {
555 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
556 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
557 "VLAN tag received on port %s",
558 ctx->ofproto->up.name, in_bundle->name);
560 xlate_report(ctx, "partial VLAN tag, dropping");
564 /* Drop frames on bundles reserved for mirroring. */
565 if (mirror_bundle_out(ctx->ofproto->mbridge, in_bundle)) {
566 if (ctx->xin->packet != NULL) {
567 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
568 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
569 "%s, which is reserved exclusively for mirroring",
570 ctx->ofproto->up.name, in_bundle->name);
572 xlate_report(ctx, "input port is mirror output port, dropping");
577 vid = vlan_tci_to_vid(flow->vlan_tci);
578 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
579 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
582 vlan = input_vid_to_vlan(in_bundle, vid);
584 /* Check other admissibility requirements. */
585 if (in_port && !is_admissible(ctx, in_port, vlan)) {
589 /* Learn source MAC. */
590 if (ctx->xin->may_learn) {
591 update_learning_table(ctx->ofproto, flow, wc, vlan, in_bundle);
594 /* Determine output bundle. */
595 mac = mac_learning_lookup(ctx->ofproto->ml, flow->dl_dst, vlan,
598 if (mac->port.p != in_bundle) {
599 xlate_report(ctx, "forwarding to learned port");
600 output_normal(ctx, mac->port.p, vlan);
602 xlate_report(ctx, "learned port is input port, dropping");
605 struct ofbundle *bundle;
607 xlate_report(ctx, "no learned MAC for destination, flooding");
608 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
609 if (bundle != in_bundle
610 && ofbundle_includes_vlan(bundle, vlan)
612 && !mirror_bundle_out(bundle->ofproto->mbridge, bundle)) {
613 output_normal(ctx, bundle, vlan);
616 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
620 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
621 * the number of packets out of UINT32_MAX to sample. The given
622 * cookie is passed back in the callback for each sampled packet.
625 compose_sample_action(const struct ofproto_dpif *ofproto,
626 struct ofpbuf *odp_actions,
627 const struct flow *flow,
628 const uint32_t probability,
629 const union user_action_cookie *cookie,
630 const size_t cookie_size)
632 size_t sample_offset, actions_offset;
635 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
637 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
639 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
640 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
643 nl_msg_end_nested(odp_actions, actions_offset);
644 nl_msg_end_nested(odp_actions, sample_offset);
645 return cookie_offset;
649 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
650 ovs_be16 vlan_tci, odp_port_t odp_port,
651 unsigned int n_outputs, union user_action_cookie *cookie)
655 cookie->type = USER_ACTION_COOKIE_SFLOW;
656 cookie->sflow.vlan_tci = vlan_tci;
658 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
659 * port information") for the interpretation of cookie->output. */
662 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
663 cookie->sflow.output = 0x40000000 | 256;
667 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
669 cookie->sflow.output = ifindex;
674 /* 0x80000000 means "multiple output ports. */
675 cookie->sflow.output = 0x80000000 | n_outputs;
680 /* Compose SAMPLE action for sFlow bridge sampling. */
682 compose_sflow_action(const struct ofproto_dpif *ofproto,
683 struct ofpbuf *odp_actions,
684 const struct flow *flow,
687 uint32_t probability;
688 union user_action_cookie cookie;
690 if (!ofproto->sflow || flow->in_port.ofp_port == OFPP_NONE) {
694 probability = dpif_sflow_get_probability(ofproto->sflow);
695 compose_sflow_cookie(ofproto, htons(0), odp_port,
696 odp_port == ODPP_NONE ? 0 : 1, &cookie);
698 return compose_sample_action(ofproto, odp_actions, flow, probability,
699 &cookie, sizeof cookie.sflow);
703 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
704 uint32_t obs_domain_id, uint32_t obs_point_id,
705 union user_action_cookie *cookie)
707 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
708 cookie->flow_sample.probability = probability;
709 cookie->flow_sample.collector_set_id = collector_set_id;
710 cookie->flow_sample.obs_domain_id = obs_domain_id;
711 cookie->flow_sample.obs_point_id = obs_point_id;
715 compose_ipfix_cookie(union user_action_cookie *cookie)
717 cookie->type = USER_ACTION_COOKIE_IPFIX;
720 /* Compose SAMPLE action for IPFIX bridge sampling. */
722 compose_ipfix_action(const struct ofproto_dpif *ofproto,
723 struct ofpbuf *odp_actions,
724 const struct flow *flow)
726 uint32_t probability;
727 union user_action_cookie cookie;
729 if (!ofproto->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
733 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
734 compose_ipfix_cookie(&cookie);
736 compose_sample_action(ofproto, odp_actions, flow, probability,
737 &cookie, sizeof cookie.ipfix);
740 /* SAMPLE action for sFlow must be first action in any given list of
741 * actions. At this point we do not have all information required to
742 * build it. So try to build sample action as complete as possible. */
744 add_sflow_action(struct xlate_ctx *ctx)
746 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
747 &ctx->xout->odp_actions,
748 &ctx->xin->flow, ODPP_NONE);
749 ctx->sflow_odp_port = 0;
750 ctx->sflow_n_outputs = 0;
753 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
754 * of actions, eventually after the SAMPLE action for sFlow. */
756 add_ipfix_action(struct xlate_ctx *ctx)
758 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
762 /* Fix SAMPLE action according to data collected while composing ODP actions.
763 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
764 * USERSPACE action's user-cookie which is required for sflow. */
766 fix_sflow_action(struct xlate_ctx *ctx)
768 const struct flow *base = &ctx->base_flow;
769 union user_action_cookie *cookie;
771 if (!ctx->user_cookie_offset) {
775 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
776 sizeof cookie->sflow);
777 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
779 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
780 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
783 static enum slow_path_reason
784 process_special(struct xlate_ctx *ctx, const struct flow *flow,
785 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
787 struct ofproto_dpif *ofproto = ctx->ofproto;
788 struct flow_wildcards *wc = &ctx->xout->wc;
792 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
794 cfm_process_heartbeat(ofport->cfm, packet);
797 } else if (ofport->bfd && bfd_should_process_flow(flow, wc)) {
799 bfd_process_packet(ofport->bfd, flow, packet);
802 } else if (ofport->bundle && ofport->bundle->lacp
803 && flow->dl_type == htons(ETH_TYPE_LACP)) {
805 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
808 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
810 stp_process_packet(ofport, packet);
819 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
822 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
823 struct flow_wildcards *wc = &ctx->xout->wc;
824 struct flow *flow = &ctx->xin->flow;
825 ovs_be16 flow_vlan_tci;
826 uint32_t flow_skb_mark;
828 odp_port_t out_port, odp_port;
831 /* If 'struct flow' gets additional metadata, we'll need to zero it out
832 * before traversing a patch port. */
833 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
836 xlate_report(ctx, "Nonexistent output port");
838 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
839 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
841 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
842 xlate_report(ctx, "STP not in forwarding state, skipping output");
846 if (mbridge_has_mirrors(ctx->ofproto->mbridge) && ofport->bundle) {
847 ctx->xout->mirrors |=
848 mirror_bundle_dst(ofport->bundle->ofproto->mbridge,
853 struct ofport_dpif *peer = ofport->peer;
854 struct flow old_flow = ctx->xin->flow;
855 enum slow_path_reason special;
857 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
858 flow->in_port.ofp_port = peer->up.ofp_port;
859 flow->metadata = htonll(0);
860 memset(&flow->tunnel, 0, sizeof flow->tunnel);
861 memset(flow->regs, 0, sizeof flow->regs);
863 special = process_special(ctx, &ctx->xin->flow, peer,
866 ctx->xout->slow = special;
867 } else if (may_receive(peer, ctx)) {
868 if (stp_forward_in_state(peer->stp_state)) {
869 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
871 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
872 * learning action look at the packet, then drop it. */
873 struct flow old_base_flow = ctx->base_flow;
874 size_t old_size = ctx->xout->odp_actions.size;
875 mirror_mask_t old_mirrors = ctx->xout->mirrors;
876 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
877 ctx->xout->mirrors = old_mirrors;
878 ctx->base_flow = old_base_flow;
879 ctx->xout->odp_actions.size = old_size;
883 ctx->xin->flow = old_flow;
884 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
886 if (ctx->xin->resubmit_stats) {
887 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
888 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
894 flow_vlan_tci = flow->vlan_tci;
895 flow_skb_mark = flow->skb_mark;
896 flow_nw_tos = flow->nw_tos;
898 if (ofproto_dpif_dscp_from_priority(ofport, flow->skb_priority, &dscp)) {
899 wc->masks.nw_tos |= IP_ECN_MASK;
900 flow->nw_tos &= ~IP_DSCP_MASK;
901 flow->nw_tos |= dscp;
904 if (ofport->is_tunnel) {
905 /* Save tunnel metadata so that changes made due to
906 * the Logical (tunnel) Port are not visible for any further
907 * matches, while explicit set actions on tunnel metadata are.
909 struct flow_tnl flow_tnl = flow->tunnel;
910 odp_port = tnl_port_send(ofport, flow, &ctx->xout->wc);
911 if (odp_port == ODPP_NONE) {
912 xlate_report(ctx, "Tunneling decided against output");
913 goto out; /* restore flow_nw_tos */
915 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
916 xlate_report(ctx, "Not tunneling to our own address");
917 goto out; /* restore flow_nw_tos */
919 if (ctx->xin->resubmit_stats) {
920 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
923 commit_odp_tunnel_action(flow, &ctx->base_flow,
924 &ctx->xout->odp_actions);
925 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
927 ofp_port_t vlandev_port;
929 odp_port = ofport->odp_port;
930 if (!hmap_is_empty(&ctx->ofproto->realdev_vid_map)) {
931 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
933 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
935 if (vlandev_port == ofp_port) {
938 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
939 flow->vlan_tci = htons(0);
941 flow->skb_mark &= ~IPSEC_MARK;
944 if (out_port != ODPP_NONE) {
945 commit_odp_actions(flow, &ctx->base_flow,
946 &ctx->xout->odp_actions, &ctx->xout->wc);
947 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
950 ctx->sflow_odp_port = odp_port;
951 ctx->sflow_n_outputs++;
952 ctx->xout->nf_output_iface = ofp_port;
957 flow->vlan_tci = flow_vlan_tci;
958 flow->skb_mark = flow_skb_mark;
959 flow->nw_tos = flow_nw_tos;
963 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
965 compose_output_action__(ctx, ofp_port, true);
968 /* Common rule processing in one place to avoid duplicating code. */
969 static struct rule_dpif *
970 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
973 if (ctx->xin->resubmit_hook) {
974 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
976 if (rule == NULL && may_packet_in) {
978 * check if table configuration flags
979 * OFPTC_TABLE_MISS_CONTROLLER, default.
980 * OFPTC_TABLE_MISS_CONTINUE,
981 * OFPTC_TABLE_MISS_DROP
982 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
984 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
986 if (rule && ctx->xin->resubmit_stats) {
987 rule_credit_stats(rule, ctx->xin->resubmit_stats);
993 xlate_table_action(struct xlate_ctx *ctx,
994 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
996 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
997 struct rule_dpif *rule;
998 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
999 uint8_t old_table_id = ctx->table_id;
1001 ctx->table_id = table_id;
1003 /* Look up a flow with 'in_port' as the input port. */
1004 ctx->xin->flow.in_port.ofp_port = in_port;
1005 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
1006 &ctx->xout->wc, table_id);
1008 ctx->xout->tags |= calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1009 ctx->table_id, rule);
1011 /* Restore the original input port. Otherwise OFPP_NORMAL and
1012 * OFPP_IN_PORT will have surprising behavior. */
1013 ctx->xin->flow.in_port.ofp_port = old_in_port;
1015 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1018 struct rule_dpif *old_rule = ctx->rule;
1022 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1023 ctx->rule = old_rule;
1027 ctx->table_id = old_table_id;
1029 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1031 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1032 MAX_RESUBMIT_RECURSION);
1033 ctx->max_resubmit_trigger = true;
1038 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1039 const struct ofpact_resubmit *resubmit)
1044 in_port = resubmit->in_port;
1045 if (in_port == OFPP_IN_PORT) {
1046 in_port = ctx->xin->flow.in_port.ofp_port;
1049 table_id = resubmit->table_id;
1050 if (table_id == 255) {
1051 table_id = ctx->table_id;
1054 xlate_table_action(ctx, in_port, table_id, false);
1058 flood_packets(struct xlate_ctx *ctx, bool all)
1060 struct ofport_dpif *ofport;
1062 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
1063 ofp_port_t ofp_port = ofport->up.ofp_port;
1065 if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1070 compose_output_action__(ctx, ofp_port, false);
1071 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
1072 compose_output_action(ctx, ofp_port);
1076 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1080 execute_controller_action(struct xlate_ctx *ctx, int len,
1081 enum ofp_packet_in_reason reason,
1082 uint16_t controller_id)
1084 struct ofputil_packet_in pin;
1085 struct ofpbuf *packet;
1088 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1089 ctx->xout->slow = SLOW_CONTROLLER;
1090 if (!ctx->xin->packet) {
1094 packet = ofpbuf_clone(ctx->xin->packet);
1096 key.skb_priority = 0;
1098 memset(&key.tunnel, 0, sizeof key.tunnel);
1100 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1101 &ctx->xout->odp_actions, &ctx->xout->wc);
1103 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1104 ctx->xout->odp_actions.size, NULL, NULL);
1106 pin.packet = packet->data;
1107 pin.packet_len = packet->size;
1108 pin.reason = reason;
1109 pin.controller_id = controller_id;
1110 pin.table_id = ctx->table_id;
1111 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1114 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1116 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
1117 ofpbuf_delete(packet);
1121 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1123 struct flow_wildcards *wc = &ctx->xout->wc;
1124 struct flow *flow = &ctx->xin->flow;
1126 ovs_assert(eth_type_mpls(eth_type));
1128 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1129 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1131 if (flow->mpls_depth) {
1132 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1138 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1139 label = htonl(0x2); /* IPV6 Explicit Null. */
1141 label = htonl(0x0); /* IPV4 Explicit Null. */
1143 wc->masks.nw_tos |= IP_DSCP_MASK;
1144 wc->masks.nw_ttl = 0xff;
1145 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1146 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1147 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1148 flow->mpls_depth = 1;
1150 flow->dl_type = eth_type;
1154 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1156 struct flow_wildcards *wc = &ctx->xout->wc;
1157 struct flow *flow = &ctx->xin->flow;
1159 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1160 ovs_assert(!eth_type_mpls(eth_type));
1162 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1163 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1165 if (flow->mpls_depth) {
1167 flow->mpls_lse = htonl(0);
1168 if (!flow->mpls_depth) {
1169 flow->dl_type = eth_type;
1175 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1177 struct flow *flow = &ctx->xin->flow;
1179 if (!is_ip_any(flow)) {
1183 ctx->xout->wc.masks.nw_ttl = 0xff;
1184 if (flow->nw_ttl > 1) {
1190 for (i = 0; i < ids->n_controllers; i++) {
1191 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1195 /* Stop processing for current table. */
1201 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1203 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1207 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1212 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1214 struct flow *flow = &ctx->xin->flow;
1215 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1216 struct flow_wildcards *wc = &ctx->xout->wc;
1218 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1220 if (!eth_type_mpls(flow->dl_type)) {
1226 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1229 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1231 /* Stop processing for current table. */
1237 xlate_output_action(struct xlate_ctx *ctx,
1238 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1240 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1242 ctx->xout->nf_output_iface = NF_OUT_DROP;
1246 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1249 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1256 flood_packets(ctx, false);
1259 flood_packets(ctx, true);
1261 case OFPP_CONTROLLER:
1262 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1268 if (port != ctx->xin->flow.in_port.ofp_port) {
1269 compose_output_action(ctx, port);
1271 xlate_report(ctx, "skipping output to input port");
1276 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1277 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1278 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1279 ctx->xout->nf_output_iface = prev_nf_output_iface;
1280 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1281 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1282 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1287 xlate_output_reg_action(struct xlate_ctx *ctx,
1288 const struct ofpact_output_reg *or)
1290 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1291 if (port <= UINT16_MAX) {
1292 union mf_subvalue value;
1294 memset(&value, 0xff, sizeof value);
1295 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1296 xlate_output_action(ctx, u16_to_ofp(port),
1297 or->max_len, false);
1302 xlate_enqueue_action(struct xlate_ctx *ctx,
1303 const struct ofpact_enqueue *enqueue)
1305 ofp_port_t ofp_port = enqueue->port;
1306 uint32_t queue_id = enqueue->queue;
1307 uint32_t flow_priority, priority;
1310 /* Translate queue to priority. */
1311 error = ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id, &priority);
1313 /* Fall back to ordinary output action. */
1314 xlate_output_action(ctx, enqueue->port, 0, false);
1318 /* Check output port. */
1319 if (ofp_port == OFPP_IN_PORT) {
1320 ofp_port = ctx->xin->flow.in_port.ofp_port;
1321 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1325 /* Add datapath actions. */
1326 flow_priority = ctx->xin->flow.skb_priority;
1327 ctx->xin->flow.skb_priority = priority;
1328 compose_output_action(ctx, ofp_port);
1329 ctx->xin->flow.skb_priority = flow_priority;
1331 /* Update NetFlow output port. */
1332 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1333 ctx->xout->nf_output_iface = ofp_port;
1334 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1335 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1340 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1342 uint32_t skb_priority;
1344 if (!ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id,
1346 ctx->xin->flow.skb_priority = skb_priority;
1348 /* Couldn't translate queue to a priority. Nothing to do. A warning
1349 * has already been logged. */
1354 slave_enabled_cb(ofp_port_t ofp_port, void *ofproto_)
1356 struct ofproto_dpif *ofproto = ofproto_;
1357 struct ofport_dpif *port;
1367 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1370 port = get_ofp_port(ofproto, ofp_port);
1371 return port ? port->may_enable : false;
1376 xlate_bundle_action(struct xlate_ctx *ctx,
1377 const struct ofpact_bundle *bundle)
1381 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1382 slave_enabled_cb, ctx->ofproto);
1383 if (bundle->dst.field) {
1384 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1386 xlate_output_action(ctx, port, 0, false);
1391 xlate_learn_action(struct xlate_ctx *ctx,
1392 const struct ofpact_learn *learn)
1394 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1395 struct ofputil_flow_mod fm;
1396 uint64_t ofpacts_stub[1024 / 8];
1397 struct ofpbuf ofpacts;
1400 ctx->xout->has_learn = true;
1402 learn_mask(learn, &ctx->xout->wc);
1404 if (!ctx->xin->may_learn) {
1408 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1409 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1411 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
1412 if (error && !VLOG_DROP_WARN(&rl)) {
1413 VLOG_WARN("learning action failed to modify flow table (%s)",
1414 ofperr_get_name(error));
1417 ofpbuf_uninit(&ofpacts);
1420 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1421 * means "infinite". */
1423 reduce_timeout(uint16_t max, uint16_t *timeout)
1425 if (max && (!*timeout || *timeout > max)) {
1431 xlate_fin_timeout(struct xlate_ctx *ctx,
1432 const struct ofpact_fin_timeout *oft)
1434 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1435 struct rule_dpif *rule = ctx->rule;
1437 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1438 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1443 xlate_sample_action(struct xlate_ctx *ctx,
1444 const struct ofpact_sample *os)
1446 union user_action_cookie cookie;
1447 /* Scale the probability from 16-bit to 32-bit while representing
1448 * the same percentage. */
1449 uint32_t probability = (os->probability << 16) | os->probability;
1451 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1452 &ctx->xout->odp_actions, &ctx->xout->wc);
1454 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1455 os->obs_domain_id, os->obs_point_id, &cookie);
1456 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
1457 probability, &cookie, sizeof cookie.flow_sample);
1461 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
1463 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
1465 ? OFPUTIL_PC_NO_RECV_STP
1466 : OFPUTIL_PC_NO_RECV)) {
1470 /* Only drop packets here if both forwarding and learning are
1471 * disabled. If just learning is enabled, we need to have
1472 * OFPP_NORMAL and the learning action have a look at the packet
1473 * before we can drop it. */
1474 if (!stp_forward_in_state(port->stp_state)
1475 && !stp_learn_in_state(port->stp_state)) {
1483 tunnel_ecn_ok(struct xlate_ctx *ctx)
1485 if (is_ip_any(&ctx->base_flow)
1486 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1487 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1488 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1489 " but is not ECN capable");
1492 /* Set the ECN CE value in the tunneled packet. */
1493 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1501 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1502 struct xlate_ctx *ctx)
1504 struct flow_wildcards *wc = &ctx->xout->wc;
1505 struct flow *flow = &ctx->xin->flow;
1506 bool was_evictable = true;
1507 const struct ofpact *a;
1510 /* Don't let the rule we're working on get evicted underneath us. */
1511 was_evictable = ctx->rule->up.evictable;
1512 ctx->rule->up.evictable = false;
1515 do_xlate_actions_again:
1516 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1517 struct ofpact_controller *controller;
1518 const struct ofpact_metadata *metadata;
1526 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1527 ofpact_get_OUTPUT(a)->max_len, true);
1530 case OFPACT_CONTROLLER:
1531 controller = ofpact_get_CONTROLLER(a);
1532 execute_controller_action(ctx, controller->max_len,
1534 controller->controller_id);
1537 case OFPACT_ENQUEUE:
1538 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1541 case OFPACT_SET_VLAN_VID:
1542 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1543 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1547 case OFPACT_SET_VLAN_PCP:
1548 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1550 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1554 case OFPACT_STRIP_VLAN:
1555 flow->vlan_tci = htons(0);
1558 case OFPACT_PUSH_VLAN:
1559 /* XXX 802.1AD(QinQ) */
1560 flow->vlan_tci = htons(VLAN_CFI);
1563 case OFPACT_SET_ETH_SRC:
1564 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
1567 case OFPACT_SET_ETH_DST:
1568 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
1571 case OFPACT_SET_IPV4_SRC:
1572 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1573 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1577 case OFPACT_SET_IPV4_DST:
1578 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1579 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1583 case OFPACT_SET_IPV4_DSCP:
1584 /* OpenFlow 1.0 only supports IPv4. */
1585 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1586 flow->nw_tos &= ~IP_DSCP_MASK;
1587 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1591 case OFPACT_SET_L4_SRC_PORT:
1592 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1593 if (is_ip_any(flow)) {
1594 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1598 case OFPACT_SET_L4_DST_PORT:
1599 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1600 if (is_ip_any(flow)) {
1601 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
1605 case OFPACT_RESUBMIT:
1606 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
1609 case OFPACT_SET_TUNNEL:
1610 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
1613 case OFPACT_SET_QUEUE:
1614 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
1617 case OFPACT_POP_QUEUE:
1618 flow->skb_priority = ctx->orig_skb_priority;
1621 case OFPACT_REG_MOVE:
1622 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
1625 case OFPACT_REG_LOAD:
1626 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
1629 case OFPACT_STACK_PUSH:
1630 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
1634 case OFPACT_STACK_POP:
1635 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
1638 case OFPACT_PUSH_MPLS:
1639 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
1642 case OFPACT_POP_MPLS:
1643 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
1646 case OFPACT_SET_MPLS_TTL:
1647 if (compose_set_mpls_ttl_action(ctx,
1648 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
1653 case OFPACT_DEC_MPLS_TTL:
1654 if (compose_dec_mpls_ttl_action(ctx)) {
1659 case OFPACT_DEC_TTL:
1660 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
1666 /* Nothing to do. */
1669 case OFPACT_MULTIPATH:
1670 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
1674 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
1677 case OFPACT_OUTPUT_REG:
1678 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
1682 xlate_learn_action(ctx, ofpact_get_LEARN(a));
1689 case OFPACT_FIN_TIMEOUT:
1690 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1691 ctx->xout->has_fin_timeout = true;
1692 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
1695 case OFPACT_CLEAR_ACTIONS:
1697 * Nothing to do because writa-actions is not supported for now.
1698 * When writa-actions is supported, clear-actions also must
1699 * be supported at the same time.
1703 case OFPACT_WRITE_METADATA:
1704 metadata = ofpact_get_WRITE_METADATA(a);
1705 flow->metadata &= ~metadata->mask;
1706 flow->metadata |= metadata->metadata & metadata->mask;
1710 /* Not implemented yet. */
1713 case OFPACT_GOTO_TABLE: {
1714 /* It is assumed that goto-table is the last action. */
1715 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
1716 struct rule_dpif *rule;
1718 ovs_assert(ctx->table_id < ogt->table_id);
1720 ctx->table_id = ogt->table_id;
1722 /* Look up a flow from the new table. */
1723 rule = rule_dpif_lookup_in_table(ctx->ofproto, flow, wc,
1726 ctx->xout->tags = calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1727 ctx->table_id, rule);
1729 rule = ctx_rule_hooks(ctx, rule, true);
1733 ctx->rule->up.evictable = was_evictable;
1736 was_evictable = rule->up.evictable;
1737 rule->up.evictable = false;
1739 /* Tail recursion removal. */
1740 ofpacts = rule->up.ofpacts;
1741 ofpacts_len = rule->up.ofpacts_len;
1742 goto do_xlate_actions_again;
1748 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
1755 ctx->rule->up.evictable = was_evictable;
1760 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
1761 const struct flow *flow, struct rule_dpif *rule,
1762 uint8_t tcp_flags, const struct ofpbuf *packet)
1764 xin->ofproto = ofproto;
1766 xin->packet = packet;
1767 xin->may_learn = packet != NULL;
1769 xin->ofpacts = NULL;
1770 xin->ofpacts_len = 0;
1771 xin->tcp_flags = tcp_flags;
1772 xin->resubmit_hook = NULL;
1773 xin->report_hook = NULL;
1774 xin->resubmit_stats = NULL;
1778 xlate_out_uninit(struct xlate_out *xout)
1781 ofpbuf_uninit(&xout->odp_actions);
1785 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
1786 * into datapath actions, using 'ctx', and discards the datapath actions. */
1788 xlate_actions_for_side_effects(struct xlate_in *xin)
1790 struct xlate_out xout;
1792 xlate_actions(xin, &xout);
1793 xlate_out_uninit(&xout);
1797 xlate_report(struct xlate_ctx *ctx, const char *s)
1799 if (ctx->xin->report_hook) {
1800 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
1805 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
1808 dst->tags = src->tags;
1809 dst->slow = src->slow;
1810 dst->has_learn = src->has_learn;
1811 dst->has_normal = src->has_normal;
1812 dst->has_fin_timeout = src->has_fin_timeout;
1813 dst->nf_output_iface = src->nf_output_iface;
1814 dst->mirrors = src->mirrors;
1816 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
1817 sizeof dst->odp_actions_stub);
1818 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
1819 src->odp_actions.size);
1823 actions_output_to_local_port(const struct xlate_ctx *ctx)
1825 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
1826 const struct nlattr *a;
1829 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
1830 ctx->xout->odp_actions.size) {
1831 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
1832 && nl_attr_get_odp_port(a) == local_odp_port) {
1839 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
1840 * into datapath actions in 'odp_actions', using 'ctx'. */
1842 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
1844 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
1845 * that in the future we always keep a copy of the original flow for
1846 * tracing purposes. */
1847 static bool hit_resubmit_limit;
1849 struct flow_wildcards *wc = &xout->wc;
1850 struct flow *flow = &xin->flow;
1852 enum slow_path_reason special;
1853 const struct ofpact *ofpacts;
1854 struct ofport_dpif *in_port;
1855 struct flow orig_flow;
1856 struct xlate_ctx ctx;
1859 COVERAGE_INC(ofproto_dpif_xlate);
1861 /* Flow initialization rules:
1862 * - 'base_flow' must match the kernel's view of the packet at the
1863 * time that action processing starts. 'flow' represents any
1864 * transformations we wish to make through actions.
1865 * - By default 'base_flow' and 'flow' are the same since the input
1866 * packet matches the output before any actions are applied.
1867 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
1868 * of the received packet as seen by the kernel. If we later output
1869 * to another device without any modifications this will cause us to
1870 * insert a new tag since the original one was stripped off by the
1872 * - Tunnel metadata as received is retained in 'flow'. This allows
1873 * tunnel metadata matching also in later tables.
1874 * Since a kernel action for setting the tunnel metadata will only be
1875 * generated with actual tunnel output, changing the tunnel metadata
1876 * values in 'flow' (such as tun_id) will only have effect with a later
1877 * tunnel output action.
1878 * - Tunnel 'base_flow' is completely cleared since that is what the
1879 * kernel does. If we wish to maintain the original values an action
1880 * needs to be generated. */
1885 ctx.ofproto = xin->ofproto;
1886 ctx.rule = xin->rule;
1888 ctx.base_flow = *flow;
1889 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
1890 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
1892 flow_wildcards_init_catchall(wc);
1893 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
1894 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
1895 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1896 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
1898 if (tnl_port_should_receive(&ctx.xin->flow)) {
1899 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
1901 if (xin->ofproto->netflow) {
1902 netflow_mask_wc(flow, wc);
1907 ctx.xout->has_learn = false;
1908 ctx.xout->has_normal = false;
1909 ctx.xout->has_fin_timeout = false;
1910 ctx.xout->nf_output_iface = NF_OUT_DROP;
1911 ctx.xout->mirrors = 0;
1913 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
1914 sizeof ctx.xout->odp_actions_stub);
1915 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
1918 ctx.max_resubmit_trigger = false;
1919 ctx.orig_skb_priority = flow->skb_priority;
1924 ofpacts = xin->ofpacts;
1925 ofpacts_len = xin->ofpacts_len;
1926 } else if (xin->rule) {
1927 ofpacts = xin->rule->up.ofpacts;
1928 ofpacts_len = xin->rule->up.ofpacts_len;
1933 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
1935 if (mbridge_has_mirrors(ctx.ofproto->mbridge) || hit_resubmit_limit) {
1936 /* Do this conditionally because the copy is expensive enough that it
1937 * shows up in profiles. */
1941 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1942 switch (ctx.ofproto->up.frag_handling) {
1943 case OFPC_FRAG_NORMAL:
1944 /* We must pretend that transport ports are unavailable. */
1945 flow->tp_src = ctx.base_flow.tp_src = htons(0);
1946 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
1949 case OFPC_FRAG_DROP:
1952 case OFPC_FRAG_REASM:
1955 case OFPC_FRAG_NX_MATCH:
1956 /* Nothing to do. */
1959 case OFPC_INVALID_TTL_TO_CONTROLLER:
1964 in_port = get_ofp_port(ctx.ofproto, flow->in_port.ofp_port);
1965 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
1967 ctx.xout->slow = special;
1969 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1970 size_t sample_actions_len;
1972 if (flow->in_port.ofp_port
1973 != vsp_realdev_to_vlandev(ctx.ofproto, flow->in_port.ofp_port,
1975 ctx.base_flow.vlan_tci = 0;
1978 add_sflow_action(&ctx);
1979 add_ipfix_action(&ctx);
1980 sample_actions_len = ctx.xout->odp_actions.size;
1982 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
1983 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
1985 /* We've let OFPP_NORMAL and the learning action look at the
1986 * packet, so drop it now if forwarding is disabled. */
1987 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
1988 ctx.xout->odp_actions.size = sample_actions_len;
1992 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
1993 if (!hit_resubmit_limit) {
1994 /* We didn't record the original flow. Make sure we do from
1996 hit_resubmit_limit = true;
1997 } else if (!VLOG_DROP_ERR(&trace_rl)) {
1998 struct ds ds = DS_EMPTY_INITIALIZER;
2000 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, &ds);
2001 VLOG_ERR("Trace triggered by excessive resubmit "
2002 "recursion:\n%s", ds_cstr(&ds));
2007 if (connmgr_has_in_band(ctx.ofproto->up.connmgr)
2008 && in_band_must_output_to_local_port(flow)
2009 && !actions_output_to_local_port(&ctx)) {
2010 compose_output_action(&ctx, OFPP_LOCAL);
2013 fix_sflow_action(&ctx);
2015 if (mbridge_has_mirrors(ctx.ofproto->mbridge)) {
2016 add_mirror_actions(&ctx, &orig_flow);
2020 ofpbuf_uninit(&ctx.stack);
2022 /* Clear the metadata and register wildcard masks, because we won't
2023 * use non-header fields as part of the cache. */
2024 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2025 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);