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);
202 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
203 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
206 vlan = input_vid_to_vlan(in_bundle, vid);
212 /* Restore the original packet before adding the mirror actions. */
213 ctx->xin->flow = *orig_flow;
216 mirror_mask_t dup_mirrors;
217 struct ofbundle *out;
218 unsigned long *vlans;
223 has_mirror = mirror_get(ofproto->mbridge, mirror_mask_ffs(mirrors) - 1,
224 &vlans, &dup_mirrors, &out, &out_vlan);
225 ovs_assert(has_mirror);
228 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
230 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
233 if (!vlan_mirrored) {
234 mirrors = zero_rightmost_1bit(mirrors);
238 mirrors &= ~dup_mirrors;
239 ctx->xout->mirrors |= dup_mirrors;
241 output_normal(ctx, out, vlan);
242 } else if (vlan != out_vlan
243 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
244 struct ofbundle *bundle;
246 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
247 if (ofbundle_includes_vlan(bundle, out_vlan)
248 && !mirror_bundle_out(bundle->ofproto->mbridge, bundle)) {
249 output_normal(ctx, bundle, out_vlan);
256 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
257 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
258 * the bundle on which the packet was received, returns the VLAN to which the
261 * Both 'vid' and the return value are in the range 0...4095. */
263 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
265 switch (in_bundle->vlan_mode) {
266 case PORT_VLAN_ACCESS:
267 return in_bundle->vlan;
270 case PORT_VLAN_TRUNK:
273 case PORT_VLAN_NATIVE_UNTAGGED:
274 case PORT_VLAN_NATIVE_TAGGED:
275 return vid ? vid : in_bundle->vlan;
282 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
283 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
286 * 'vid' should be the VID obtained from the 802.1Q header that was received as
287 * part of a packet (specify 0 if there was no 802.1Q header), in the range
290 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
292 /* Allow any VID on the OFPP_NONE port. */
293 if (in_bundle == &ofpp_none_bundle) {
297 switch (in_bundle->vlan_mode) {
298 case PORT_VLAN_ACCESS:
301 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
302 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
303 "packet received on port %s configured as VLAN "
304 "%"PRIu16" access port",
305 in_bundle->ofproto->up.name, vid,
306 in_bundle->name, in_bundle->vlan);
312 case PORT_VLAN_NATIVE_UNTAGGED:
313 case PORT_VLAN_NATIVE_TAGGED:
315 /* Port must always carry its native VLAN. */
319 case PORT_VLAN_TRUNK:
320 if (!ofbundle_includes_vlan(in_bundle, vid)) {
322 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
323 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
324 "received on port %s not configured for trunking "
326 in_bundle->ofproto->up.name, vid,
327 in_bundle->name, vid);
339 /* Given 'vlan', the VLAN that a packet belongs to, and
340 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
341 * that should be included in the 802.1Q header. (If the return value is 0,
342 * then the 802.1Q header should only be included in the packet if there is a
345 * Both 'vlan' and the return value are in the range 0...4095. */
347 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
349 switch (out_bundle->vlan_mode) {
350 case PORT_VLAN_ACCESS:
353 case PORT_VLAN_TRUNK:
354 case PORT_VLAN_NATIVE_TAGGED:
357 case PORT_VLAN_NATIVE_UNTAGGED:
358 return vlan == out_bundle->vlan ? 0 : vlan;
366 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
369 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
370 struct ofport_dpif *port;
372 ovs_be16 tci, old_tci;
374 vid = output_vlan_to_vid(out_bundle, vlan);
375 if (!out_bundle->bond) {
376 port = ofbundle_get_a_port(out_bundle);
378 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
379 &ctx->xout->wc, vid, &ctx->xout->tags);
381 /* No slaves enabled, so drop packet. */
388 if (tci || out_bundle->use_priority_tags) {
389 tci |= *flow_tci & htons(VLAN_PCP_MASK);
391 tci |= htons(VLAN_CFI);
396 compose_output_action(ctx, port->up.ofp_port);
400 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
401 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
402 * indicate this; newer upstream kernels use gratuitous ARP requests. */
404 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
406 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
410 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
411 if (!eth_addr_is_broadcast(flow->dl_dst)) {
415 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
416 if (flow->nw_proto == ARP_OP_REPLY) {
418 } else if (flow->nw_proto == ARP_OP_REQUEST) {
419 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
420 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
422 return flow->nw_src == flow->nw_dst;
429 update_learning_table(struct ofproto_dpif *ofproto,
430 const struct flow *flow, struct flow_wildcards *wc,
431 int vlan, struct ofbundle *in_bundle)
433 struct mac_entry *mac;
435 /* Don't learn the OFPP_NONE port. */
436 if (in_bundle == &ofpp_none_bundle) {
440 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
444 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
445 if (is_gratuitous_arp(flow, wc)) {
446 /* We don't want to learn from gratuitous ARP packets that are
447 * reflected back over bond slaves so we lock the learning table. */
448 if (!in_bundle->bond) {
449 mac_entry_set_grat_arp_lock(mac);
450 } else if (mac_entry_is_grat_arp_locked(mac)) {
455 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
456 /* The log messages here could actually be useful in debugging,
457 * so keep the rate limit relatively high. */
458 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
459 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
460 "on port %s in VLAN %d",
461 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
462 in_bundle->name, vlan);
464 mac->port.p = in_bundle;
465 mac_learning_changed(ofproto->ml, mac);
469 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
470 * dropped. Returns true if they may be forwarded, false if they should be
473 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
474 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
476 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
477 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
478 * checked by input_vid_is_valid().
480 * May also add tags to '*tags', although the current implementation only does
481 * so in one special case.
484 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
487 struct ofproto_dpif *ofproto = ctx->ofproto;
488 struct flow *flow = &ctx->xin->flow;
489 struct ofbundle *in_bundle = in_port->bundle;
491 /* Drop frames for reserved multicast addresses
492 * only if forward_bpdu option is absent. */
493 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
494 xlate_report(ctx, "packet has reserved destination MAC, dropping");
498 if (in_bundle->bond) {
499 struct mac_entry *mac;
501 switch (bond_check_admissibility(in_bundle->bond, in_port,
502 flow->dl_dst, &ctx->xout->tags)) {
507 xlate_report(ctx, "bonding refused admissibility, dropping");
510 case BV_DROP_IF_MOVED:
511 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
512 if (mac && mac->port.p != in_bundle &&
513 (!is_gratuitous_arp(flow, &ctx->xout->wc)
514 || mac_entry_is_grat_arp_locked(mac))) {
515 xlate_report(ctx, "SLB bond thinks this packet looped back, "
527 xlate_normal(struct xlate_ctx *ctx)
529 struct flow_wildcards *wc = &ctx->xout->wc;
530 struct flow *flow = &ctx->xin->flow;
531 struct ofport_dpif *in_port;
532 struct ofbundle *in_bundle;
533 struct mac_entry *mac;
537 ctx->xout->has_normal = true;
539 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
540 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
541 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
543 in_bundle = lookup_input_bundle(ctx->ofproto, flow->in_port.ofp_port,
544 ctx->xin->packet != NULL, &in_port);
546 xlate_report(ctx, "no input bundle, dropping");
550 /* Drop malformed frames. */
551 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
552 !(flow->vlan_tci & htons(VLAN_CFI))) {
553 if (ctx->xin->packet != NULL) {
554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
555 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
556 "VLAN tag received on port %s",
557 ctx->ofproto->up.name, in_bundle->name);
559 xlate_report(ctx, "partial VLAN tag, dropping");
563 /* Drop frames on bundles reserved for mirroring. */
564 if (mirror_bundle_out(ctx->ofproto->mbridge, in_bundle)) {
565 if (ctx->xin->packet != NULL) {
566 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
567 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
568 "%s, which is reserved exclusively for mirroring",
569 ctx->ofproto->up.name, in_bundle->name);
571 xlate_report(ctx, "input port is mirror output port, dropping");
576 vid = vlan_tci_to_vid(flow->vlan_tci);
577 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
578 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
581 vlan = input_vid_to_vlan(in_bundle, vid);
583 /* Check other admissibility requirements. */
584 if (in_port && !is_admissible(ctx, in_port, vlan)) {
588 /* Learn source MAC. */
589 if (ctx->xin->may_learn) {
590 update_learning_table(ctx->ofproto, flow, wc, vlan, in_bundle);
593 /* Determine output bundle. */
594 mac = mac_learning_lookup(ctx->ofproto->ml, flow->dl_dst, vlan,
597 if (mac->port.p != in_bundle) {
598 xlate_report(ctx, "forwarding to learned port");
599 output_normal(ctx, mac->port.p, vlan);
601 xlate_report(ctx, "learned port is input port, dropping");
604 struct ofbundle *bundle;
606 xlate_report(ctx, "no learned MAC for destination, flooding");
607 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
608 if (bundle != in_bundle
609 && ofbundle_includes_vlan(bundle, vlan)
611 && !mirror_bundle_out(bundle->ofproto->mbridge, bundle)) {
612 output_normal(ctx, bundle, vlan);
615 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
619 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
620 * the number of packets out of UINT32_MAX to sample. The given
621 * cookie is passed back in the callback for each sampled packet.
624 compose_sample_action(const struct ofproto_dpif *ofproto,
625 struct ofpbuf *odp_actions,
626 const struct flow *flow,
627 const uint32_t probability,
628 const union user_action_cookie *cookie,
629 const size_t cookie_size)
631 size_t sample_offset, actions_offset;
634 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
636 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
638 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
639 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
642 nl_msg_end_nested(odp_actions, actions_offset);
643 nl_msg_end_nested(odp_actions, sample_offset);
644 return cookie_offset;
648 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
649 ovs_be16 vlan_tci, odp_port_t odp_port,
650 unsigned int n_outputs, union user_action_cookie *cookie)
654 cookie->type = USER_ACTION_COOKIE_SFLOW;
655 cookie->sflow.vlan_tci = vlan_tci;
657 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
658 * port information") for the interpretation of cookie->output. */
661 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
662 cookie->sflow.output = 0x40000000 | 256;
666 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
668 cookie->sflow.output = ifindex;
673 /* 0x80000000 means "multiple output ports. */
674 cookie->sflow.output = 0x80000000 | n_outputs;
679 /* Compose SAMPLE action for sFlow bridge sampling. */
681 compose_sflow_action(const struct ofproto_dpif *ofproto,
682 struct ofpbuf *odp_actions,
683 const struct flow *flow,
686 uint32_t probability;
687 union user_action_cookie cookie;
689 if (!ofproto->sflow || flow->in_port.ofp_port == OFPP_NONE) {
693 probability = dpif_sflow_get_probability(ofproto->sflow);
694 compose_sflow_cookie(ofproto, htons(0), odp_port,
695 odp_port == ODPP_NONE ? 0 : 1, &cookie);
697 return compose_sample_action(ofproto, odp_actions, flow, probability,
698 &cookie, sizeof cookie.sflow);
702 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
703 uint32_t obs_domain_id, uint32_t obs_point_id,
704 union user_action_cookie *cookie)
706 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
707 cookie->flow_sample.probability = probability;
708 cookie->flow_sample.collector_set_id = collector_set_id;
709 cookie->flow_sample.obs_domain_id = obs_domain_id;
710 cookie->flow_sample.obs_point_id = obs_point_id;
714 compose_ipfix_cookie(union user_action_cookie *cookie)
716 cookie->type = USER_ACTION_COOKIE_IPFIX;
719 /* Compose SAMPLE action for IPFIX bridge sampling. */
721 compose_ipfix_action(const struct ofproto_dpif *ofproto,
722 struct ofpbuf *odp_actions,
723 const struct flow *flow)
725 uint32_t probability;
726 union user_action_cookie cookie;
728 if (!ofproto->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
732 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
733 compose_ipfix_cookie(&cookie);
735 compose_sample_action(ofproto, odp_actions, flow, probability,
736 &cookie, sizeof cookie.ipfix);
739 /* SAMPLE action for sFlow must be first action in any given list of
740 * actions. At this point we do not have all information required to
741 * build it. So try to build sample action as complete as possible. */
743 add_sflow_action(struct xlate_ctx *ctx)
745 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
746 &ctx->xout->odp_actions,
747 &ctx->xin->flow, ODPP_NONE);
748 ctx->sflow_odp_port = 0;
749 ctx->sflow_n_outputs = 0;
752 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
753 * of actions, eventually after the SAMPLE action for sFlow. */
755 add_ipfix_action(struct xlate_ctx *ctx)
757 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
761 /* Fix SAMPLE action according to data collected while composing ODP actions.
762 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
763 * USERSPACE action's user-cookie which is required for sflow. */
765 fix_sflow_action(struct xlate_ctx *ctx)
767 const struct flow *base = &ctx->base_flow;
768 union user_action_cookie *cookie;
770 if (!ctx->user_cookie_offset) {
774 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
775 sizeof cookie->sflow);
776 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
778 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
779 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
782 static enum slow_path_reason
783 process_special(struct xlate_ctx *ctx, const struct flow *flow,
784 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
786 struct ofproto_dpif *ofproto = ctx->ofproto;
787 struct flow_wildcards *wc = &ctx->xout->wc;
791 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
793 cfm_process_heartbeat(ofport->cfm, packet);
796 } else if (ofport->bfd && bfd_should_process_flow(flow, wc)) {
798 bfd_process_packet(ofport->bfd, flow, packet);
801 } else if (ofport->bundle && ofport->bundle->lacp
802 && flow->dl_type == htons(ETH_TYPE_LACP)) {
804 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
807 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
809 stp_process_packet(ofport, packet);
818 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
821 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
822 struct flow_wildcards *wc = &ctx->xout->wc;
823 struct flow *flow = &ctx->xin->flow;
824 ovs_be16 flow_vlan_tci;
825 uint32_t flow_skb_mark;
827 odp_port_t out_port, odp_port;
830 /* If 'struct flow' gets additional metadata, we'll need to zero it out
831 * before traversing a patch port. */
832 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
835 xlate_report(ctx, "Nonexistent output port");
837 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
838 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
840 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
841 xlate_report(ctx, "STP not in forwarding state, skipping output");
845 if (mbridge_has_mirrors(ctx->ofproto->mbridge) && ofport->bundle) {
846 ctx->xout->mirrors |=
847 mirror_bundle_dst(ofport->bundle->ofproto->mbridge,
852 struct ofport_dpif *peer = ofport->peer;
853 struct flow old_flow = ctx->xin->flow;
854 enum slow_path_reason special;
856 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
857 flow->in_port.ofp_port = peer->up.ofp_port;
858 flow->metadata = htonll(0);
859 memset(&flow->tunnel, 0, sizeof flow->tunnel);
860 memset(flow->regs, 0, sizeof flow->regs);
862 special = process_special(ctx, &ctx->xin->flow, peer,
865 ctx->xout->slow = special;
866 } else if (may_receive(peer, ctx)) {
867 if (stp_forward_in_state(peer->stp_state)) {
868 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
870 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
871 * learning action look at the packet, then drop it. */
872 struct flow old_base_flow = ctx->base_flow;
873 size_t old_size = ctx->xout->odp_actions.size;
874 mirror_mask_t old_mirrors = ctx->xout->mirrors;
875 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
876 ctx->xout->mirrors = old_mirrors;
877 ctx->base_flow = old_base_flow;
878 ctx->xout->odp_actions.size = old_size;
882 ctx->xin->flow = old_flow;
883 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
885 if (ctx->xin->resubmit_stats) {
886 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
887 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
893 flow_vlan_tci = flow->vlan_tci;
894 flow_skb_mark = flow->skb_mark;
895 flow_nw_tos = flow->nw_tos;
897 if (ofproto_dpif_dscp_from_priority(ofport, flow->skb_priority, &dscp)) {
898 wc->masks.nw_tos |= IP_ECN_MASK;
899 flow->nw_tos &= ~IP_DSCP_MASK;
900 flow->nw_tos |= dscp;
903 if (ofport->is_tunnel) {
904 /* Save tunnel metadata so that changes made due to
905 * the Logical (tunnel) Port are not visible for any further
906 * matches, while explicit set actions on tunnel metadata are.
908 struct flow_tnl flow_tnl = flow->tunnel;
909 odp_port = tnl_port_send(ofport, flow, &ctx->xout->wc);
910 if (odp_port == ODPP_NONE) {
911 xlate_report(ctx, "Tunneling decided against output");
912 goto out; /* restore flow_nw_tos */
914 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
915 xlate_report(ctx, "Not tunneling to our own address");
916 goto out; /* restore flow_nw_tos */
918 if (ctx->xin->resubmit_stats) {
919 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
922 commit_odp_tunnel_action(flow, &ctx->base_flow,
923 &ctx->xout->odp_actions);
924 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
926 ofp_port_t vlandev_port;
928 odp_port = ofport->odp_port;
929 if (!hmap_is_empty(&ctx->ofproto->realdev_vid_map)) {
930 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
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 != ODPP_NONE) {
944 commit_odp_actions(flow, &ctx->base_flow,
945 &ctx->xout->odp_actions, &ctx->xout->wc);
946 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
949 ctx->sflow_odp_port = odp_port;
950 ctx->sflow_n_outputs++;
951 ctx->xout->nf_output_iface = ofp_port;
956 flow->vlan_tci = flow_vlan_tci;
957 flow->skb_mark = flow_skb_mark;
958 flow->nw_tos = flow_nw_tos;
962 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
964 compose_output_action__(ctx, ofp_port, true);
967 /* Common rule processing in one place to avoid duplicating code. */
968 static struct rule_dpif *
969 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
972 if (ctx->xin->resubmit_hook) {
973 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
975 if (rule == NULL && may_packet_in) {
977 * check if table configuration flags
978 * OFPTC_TABLE_MISS_CONTROLLER, default.
979 * OFPTC_TABLE_MISS_CONTINUE,
980 * OFPTC_TABLE_MISS_DROP
981 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
983 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
985 if (rule && ctx->xin->resubmit_stats) {
986 rule_credit_stats(rule, ctx->xin->resubmit_stats);
992 xlate_table_action(struct xlate_ctx *ctx,
993 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
995 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
996 struct rule_dpif *rule;
997 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
998 uint8_t old_table_id = ctx->table_id;
1000 ctx->table_id = table_id;
1002 /* Look up a flow with 'in_port' as the input port. */
1003 ctx->xin->flow.in_port.ofp_port = in_port;
1004 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
1005 &ctx->xout->wc, table_id);
1007 ctx->xout->tags |= calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1008 ctx->table_id, rule);
1010 /* Restore the original input port. Otherwise OFPP_NORMAL and
1011 * OFPP_IN_PORT will have surprising behavior. */
1012 ctx->xin->flow.in_port.ofp_port = old_in_port;
1014 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1017 struct rule_dpif *old_rule = ctx->rule;
1021 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1022 ctx->rule = old_rule;
1026 ctx->table_id = old_table_id;
1028 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1030 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1031 MAX_RESUBMIT_RECURSION);
1032 ctx->max_resubmit_trigger = true;
1037 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1038 const struct ofpact_resubmit *resubmit)
1043 in_port = resubmit->in_port;
1044 if (in_port == OFPP_IN_PORT) {
1045 in_port = ctx->xin->flow.in_port.ofp_port;
1048 table_id = resubmit->table_id;
1049 if (table_id == 255) {
1050 table_id = ctx->table_id;
1053 xlate_table_action(ctx, in_port, table_id, false);
1057 flood_packets(struct xlate_ctx *ctx, bool all)
1059 struct ofport_dpif *ofport;
1061 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
1062 ofp_port_t ofp_port = ofport->up.ofp_port;
1064 if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1069 compose_output_action__(ctx, ofp_port, false);
1070 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
1071 compose_output_action(ctx, ofp_port);
1075 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1079 execute_controller_action(struct xlate_ctx *ctx, int len,
1080 enum ofp_packet_in_reason reason,
1081 uint16_t controller_id)
1083 struct ofputil_packet_in pin;
1084 struct ofpbuf *packet;
1087 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1088 ctx->xout->slow = SLOW_CONTROLLER;
1089 if (!ctx->xin->packet) {
1093 packet = ofpbuf_clone(ctx->xin->packet);
1095 key.skb_priority = 0;
1097 memset(&key.tunnel, 0, sizeof key.tunnel);
1099 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1100 &ctx->xout->odp_actions, &ctx->xout->wc);
1102 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1103 ctx->xout->odp_actions.size, NULL, NULL);
1105 pin.packet = packet->data;
1106 pin.packet_len = packet->size;
1107 pin.reason = reason;
1108 pin.controller_id = controller_id;
1109 pin.table_id = ctx->table_id;
1110 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1113 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1115 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
1116 ofpbuf_delete(packet);
1120 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1122 struct flow_wildcards *wc = &ctx->xout->wc;
1123 struct flow *flow = &ctx->xin->flow;
1125 ovs_assert(eth_type_mpls(eth_type));
1127 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1128 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1130 if (flow->mpls_depth) {
1131 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1137 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1138 label = htonl(0x2); /* IPV6 Explicit Null. */
1140 label = htonl(0x0); /* IPV4 Explicit Null. */
1142 wc->masks.nw_tos |= IP_DSCP_MASK;
1143 wc->masks.nw_ttl = 0xff;
1144 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1145 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1146 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1147 flow->mpls_depth = 1;
1149 flow->dl_type = eth_type;
1153 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1155 struct flow_wildcards *wc = &ctx->xout->wc;
1156 struct flow *flow = &ctx->xin->flow;
1158 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1159 ovs_assert(!eth_type_mpls(eth_type));
1161 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1162 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1164 if (flow->mpls_depth) {
1166 flow->mpls_lse = htonl(0);
1167 if (!flow->mpls_depth) {
1168 flow->dl_type = eth_type;
1174 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1176 struct flow *flow = &ctx->xin->flow;
1178 if (!is_ip_any(flow)) {
1182 ctx->xout->wc.masks.nw_ttl = 0xff;
1183 if (flow->nw_ttl > 1) {
1189 for (i = 0; i < ids->n_controllers; i++) {
1190 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1194 /* Stop processing for current table. */
1200 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1202 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1206 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1211 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1213 struct flow *flow = &ctx->xin->flow;
1214 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1215 struct flow_wildcards *wc = &ctx->xout->wc;
1217 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1219 if (!eth_type_mpls(flow->dl_type)) {
1225 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1228 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1230 /* Stop processing for current table. */
1236 xlate_output_action(struct xlate_ctx *ctx,
1237 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1239 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1241 ctx->xout->nf_output_iface = NF_OUT_DROP;
1245 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1248 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1255 flood_packets(ctx, false);
1258 flood_packets(ctx, true);
1260 case OFPP_CONTROLLER:
1261 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1267 if (port != ctx->xin->flow.in_port.ofp_port) {
1268 compose_output_action(ctx, port);
1270 xlate_report(ctx, "skipping output to input port");
1275 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1276 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1277 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1278 ctx->xout->nf_output_iface = prev_nf_output_iface;
1279 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1280 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1281 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1286 xlate_output_reg_action(struct xlate_ctx *ctx,
1287 const struct ofpact_output_reg *or)
1289 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1290 if (port <= UINT16_MAX) {
1291 union mf_subvalue value;
1293 memset(&value, 0xff, sizeof value);
1294 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1295 xlate_output_action(ctx, u16_to_ofp(port),
1296 or->max_len, false);
1301 xlate_enqueue_action(struct xlate_ctx *ctx,
1302 const struct ofpact_enqueue *enqueue)
1304 ofp_port_t ofp_port = enqueue->port;
1305 uint32_t queue_id = enqueue->queue;
1306 uint32_t flow_priority, priority;
1309 /* Translate queue to priority. */
1310 error = ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id, &priority);
1312 /* Fall back to ordinary output action. */
1313 xlate_output_action(ctx, enqueue->port, 0, false);
1317 /* Check output port. */
1318 if (ofp_port == OFPP_IN_PORT) {
1319 ofp_port = ctx->xin->flow.in_port.ofp_port;
1320 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1324 /* Add datapath actions. */
1325 flow_priority = ctx->xin->flow.skb_priority;
1326 ctx->xin->flow.skb_priority = priority;
1327 compose_output_action(ctx, ofp_port);
1328 ctx->xin->flow.skb_priority = flow_priority;
1330 /* Update NetFlow output port. */
1331 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1332 ctx->xout->nf_output_iface = ofp_port;
1333 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1334 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1339 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1341 uint32_t skb_priority;
1343 if (!ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id,
1345 ctx->xin->flow.skb_priority = skb_priority;
1347 /* Couldn't translate queue to a priority. Nothing to do. A warning
1348 * has already been logged. */
1353 slave_enabled_cb(ofp_port_t ofp_port, void *ofproto_)
1355 struct ofproto_dpif *ofproto = ofproto_;
1356 struct ofport_dpif *port;
1366 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1369 port = get_ofp_port(ofproto, ofp_port);
1370 return port ? port->may_enable : false;
1375 xlate_bundle_action(struct xlate_ctx *ctx,
1376 const struct ofpact_bundle *bundle)
1380 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1381 slave_enabled_cb, ctx->ofproto);
1382 if (bundle->dst.field) {
1383 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1385 xlate_output_action(ctx, port, 0, false);
1390 xlate_learn_action(struct xlate_ctx *ctx,
1391 const struct ofpact_learn *learn)
1393 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1394 struct ofputil_flow_mod fm;
1395 uint64_t ofpacts_stub[1024 / 8];
1396 struct ofpbuf ofpacts;
1399 ctx->xout->has_learn = true;
1401 learn_mask(learn, &ctx->xout->wc);
1403 if (!ctx->xin->may_learn) {
1407 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1408 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1410 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
1411 if (error && !VLOG_DROP_WARN(&rl)) {
1412 VLOG_WARN("learning action failed to modify flow table (%s)",
1413 ofperr_get_name(error));
1416 ofpbuf_uninit(&ofpacts);
1419 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1420 * means "infinite". */
1422 reduce_timeout(uint16_t max, uint16_t *timeout)
1424 if (max && (!*timeout || *timeout > max)) {
1430 xlate_fin_timeout(struct xlate_ctx *ctx,
1431 const struct ofpact_fin_timeout *oft)
1433 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1434 struct rule_dpif *rule = ctx->rule;
1436 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1437 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1442 xlate_sample_action(struct xlate_ctx *ctx,
1443 const struct ofpact_sample *os)
1445 union user_action_cookie cookie;
1446 /* Scale the probability from 16-bit to 32-bit while representing
1447 * the same percentage. */
1448 uint32_t probability = (os->probability << 16) | os->probability;
1450 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1451 &ctx->xout->odp_actions, &ctx->xout->wc);
1453 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1454 os->obs_domain_id, os->obs_point_id, &cookie);
1455 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
1456 probability, &cookie, sizeof cookie.flow_sample);
1460 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
1462 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
1464 ? OFPUTIL_PC_NO_RECV_STP
1465 : OFPUTIL_PC_NO_RECV)) {
1469 /* Only drop packets here if both forwarding and learning are
1470 * disabled. If just learning is enabled, we need to have
1471 * OFPP_NORMAL and the learning action have a look at the packet
1472 * before we can drop it. */
1473 if (!stp_forward_in_state(port->stp_state)
1474 && !stp_learn_in_state(port->stp_state)) {
1482 tunnel_ecn_ok(struct xlate_ctx *ctx)
1484 if (is_ip_any(&ctx->base_flow)
1485 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1486 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1487 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1488 " but is not ECN capable");
1491 /* Set the ECN CE value in the tunneled packet. */
1492 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1500 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1501 struct xlate_ctx *ctx)
1503 struct flow_wildcards *wc = &ctx->xout->wc;
1504 struct flow *flow = &ctx->xin->flow;
1505 bool was_evictable = true;
1506 const struct ofpact *a;
1509 /* Don't let the rule we're working on get evicted underneath us. */
1510 was_evictable = ctx->rule->up.evictable;
1511 ctx->rule->up.evictable = false;
1514 do_xlate_actions_again:
1515 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1516 struct ofpact_controller *controller;
1517 const struct ofpact_metadata *metadata;
1525 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1526 ofpact_get_OUTPUT(a)->max_len, true);
1529 case OFPACT_CONTROLLER:
1530 controller = ofpact_get_CONTROLLER(a);
1531 execute_controller_action(ctx, controller->max_len,
1533 controller->controller_id);
1536 case OFPACT_ENQUEUE:
1537 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1540 case OFPACT_SET_VLAN_VID:
1541 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1542 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1546 case OFPACT_SET_VLAN_PCP:
1547 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1549 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1553 case OFPACT_STRIP_VLAN:
1554 flow->vlan_tci = htons(0);
1557 case OFPACT_PUSH_VLAN:
1558 /* XXX 802.1AD(QinQ) */
1559 flow->vlan_tci = htons(VLAN_CFI);
1562 case OFPACT_SET_ETH_SRC:
1563 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
1566 case OFPACT_SET_ETH_DST:
1567 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
1570 case OFPACT_SET_IPV4_SRC:
1571 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1572 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1576 case OFPACT_SET_IPV4_DST:
1577 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1578 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1582 case OFPACT_SET_IPV4_DSCP:
1583 /* OpenFlow 1.0 only supports IPv4. */
1584 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1585 flow->nw_tos &= ~IP_DSCP_MASK;
1586 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1590 case OFPACT_SET_L4_SRC_PORT:
1591 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1592 if (is_ip_any(flow)) {
1593 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1597 case OFPACT_SET_L4_DST_PORT:
1598 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1599 if (is_ip_any(flow)) {
1600 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
1604 case OFPACT_RESUBMIT:
1605 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
1608 case OFPACT_SET_TUNNEL:
1609 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
1612 case OFPACT_SET_QUEUE:
1613 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
1616 case OFPACT_POP_QUEUE:
1617 flow->skb_priority = ctx->orig_skb_priority;
1620 case OFPACT_REG_MOVE:
1621 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
1624 case OFPACT_REG_LOAD:
1625 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
1628 case OFPACT_STACK_PUSH:
1629 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
1633 case OFPACT_STACK_POP:
1634 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
1637 case OFPACT_PUSH_MPLS:
1638 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
1641 case OFPACT_POP_MPLS:
1642 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
1645 case OFPACT_SET_MPLS_TTL:
1646 if (compose_set_mpls_ttl_action(ctx,
1647 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
1652 case OFPACT_DEC_MPLS_TTL:
1653 if (compose_dec_mpls_ttl_action(ctx)) {
1658 case OFPACT_DEC_TTL:
1659 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
1665 /* Nothing to do. */
1668 case OFPACT_MULTIPATH:
1669 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
1673 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
1676 case OFPACT_OUTPUT_REG:
1677 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
1681 xlate_learn_action(ctx, ofpact_get_LEARN(a));
1688 case OFPACT_FIN_TIMEOUT:
1689 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1690 ctx->xout->has_fin_timeout = true;
1691 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
1694 case OFPACT_CLEAR_ACTIONS:
1696 * Nothing to do because writa-actions is not supported for now.
1697 * When writa-actions is supported, clear-actions also must
1698 * be supported at the same time.
1702 case OFPACT_WRITE_METADATA:
1703 metadata = ofpact_get_WRITE_METADATA(a);
1704 flow->metadata &= ~metadata->mask;
1705 flow->metadata |= metadata->metadata & metadata->mask;
1709 /* Not implemented yet. */
1712 case OFPACT_GOTO_TABLE: {
1713 /* It is assumed that goto-table is the last action. */
1714 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
1715 struct rule_dpif *rule;
1717 ovs_assert(ctx->table_id < ogt->table_id);
1719 ctx->table_id = ogt->table_id;
1721 /* Look up a flow from the new table. */
1722 rule = rule_dpif_lookup_in_table(ctx->ofproto, flow, wc,
1725 ctx->xout->tags = calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1726 ctx->table_id, rule);
1728 rule = ctx_rule_hooks(ctx, rule, true);
1732 ctx->rule->up.evictable = was_evictable;
1735 was_evictable = rule->up.evictable;
1736 rule->up.evictable = false;
1738 /* Tail recursion removal. */
1739 ofpacts = rule->up.ofpacts;
1740 ofpacts_len = rule->up.ofpacts_len;
1741 goto do_xlate_actions_again;
1747 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
1754 ctx->rule->up.evictable = was_evictable;
1759 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
1760 const struct flow *flow, struct rule_dpif *rule,
1761 uint8_t tcp_flags, const struct ofpbuf *packet)
1763 xin->ofproto = ofproto;
1765 xin->packet = packet;
1766 xin->may_learn = packet != NULL;
1768 xin->ofpacts = NULL;
1769 xin->ofpacts_len = 0;
1770 xin->tcp_flags = tcp_flags;
1771 xin->resubmit_hook = NULL;
1772 xin->report_hook = NULL;
1773 xin->resubmit_stats = NULL;
1777 xlate_out_uninit(struct xlate_out *xout)
1780 ofpbuf_uninit(&xout->odp_actions);
1784 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
1785 * into datapath actions, using 'ctx', and discards the datapath actions. */
1787 xlate_actions_for_side_effects(struct xlate_in *xin)
1789 struct xlate_out xout;
1791 xlate_actions(xin, &xout);
1792 xlate_out_uninit(&xout);
1796 xlate_report(struct xlate_ctx *ctx, const char *s)
1798 if (ctx->xin->report_hook) {
1799 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
1804 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
1807 dst->tags = src->tags;
1808 dst->slow = src->slow;
1809 dst->has_learn = src->has_learn;
1810 dst->has_normal = src->has_normal;
1811 dst->has_fin_timeout = src->has_fin_timeout;
1812 dst->nf_output_iface = src->nf_output_iface;
1813 dst->mirrors = src->mirrors;
1815 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
1816 sizeof dst->odp_actions_stub);
1817 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
1818 src->odp_actions.size);
1822 actions_output_to_local_port(const struct xlate_ctx *ctx)
1824 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
1825 const struct nlattr *a;
1828 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
1829 ctx->xout->odp_actions.size) {
1830 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
1831 && nl_attr_get_odp_port(a) == local_odp_port) {
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);
1893 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
1894 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1895 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
1897 if (tnl_port_should_receive(&ctx.xin->flow)) {
1898 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
1900 if (xin->ofproto->netflow) {
1901 netflow_mask_wc(flow, wc);
1906 ctx.xout->has_learn = false;
1907 ctx.xout->has_normal = false;
1908 ctx.xout->has_fin_timeout = false;
1909 ctx.xout->nf_output_iface = NF_OUT_DROP;
1910 ctx.xout->mirrors = 0;
1912 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
1913 sizeof ctx.xout->odp_actions_stub);
1914 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
1917 ctx.max_resubmit_trigger = false;
1918 ctx.orig_skb_priority = flow->skb_priority;
1923 ofpacts = xin->ofpacts;
1924 ofpacts_len = xin->ofpacts_len;
1925 } else if (xin->rule) {
1926 ofpacts = xin->rule->up.ofpacts;
1927 ofpacts_len = xin->rule->up.ofpacts_len;
1932 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
1934 if (mbridge_has_mirrors(ctx.ofproto->mbridge) || hit_resubmit_limit) {
1935 /* Do this conditionally because the copy is expensive enough that it
1936 * shows up in profiles. */
1940 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1941 switch (ctx.ofproto->up.frag_handling) {
1942 case OFPC_FRAG_NORMAL:
1943 /* We must pretend that transport ports are unavailable. */
1944 flow->tp_src = ctx.base_flow.tp_src = htons(0);
1945 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
1948 case OFPC_FRAG_DROP:
1951 case OFPC_FRAG_REASM:
1954 case OFPC_FRAG_NX_MATCH:
1955 /* Nothing to do. */
1958 case OFPC_INVALID_TTL_TO_CONTROLLER:
1963 in_port = get_ofp_port(ctx.ofproto, flow->in_port.ofp_port);
1964 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
1966 ctx.xout->slow = special;
1968 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1969 size_t sample_actions_len;
1971 if (flow->in_port.ofp_port
1972 != vsp_realdev_to_vlandev(ctx.ofproto, flow->in_port.ofp_port,
1974 ctx.base_flow.vlan_tci = 0;
1977 add_sflow_action(&ctx);
1978 add_ipfix_action(&ctx);
1979 sample_actions_len = ctx.xout->odp_actions.size;
1981 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
1982 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
1984 /* We've let OFPP_NORMAL and the learning action look at the
1985 * packet, so drop it now if forwarding is disabled. */
1986 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
1987 ctx.xout->odp_actions.size = sample_actions_len;
1991 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
1992 if (!hit_resubmit_limit) {
1993 /* We didn't record the original flow. Make sure we do from
1995 hit_resubmit_limit = true;
1996 } else if (!VLOG_DROP_ERR(&trace_rl)) {
1997 struct ds ds = DS_EMPTY_INITIALIZER;
1999 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, &ds);
2000 VLOG_ERR("Trace triggered by excessive resubmit "
2001 "recursion:\n%s", ds_cstr(&ds));
2006 if (connmgr_has_in_band(ctx.ofproto->up.connmgr)
2007 && in_band_must_output_to_local_port(flow)
2008 && !actions_output_to_local_port(&ctx)) {
2009 compose_output_action(&ctx, OFPP_LOCAL);
2011 if (mbridge_has_mirrors(ctx.ofproto->mbridge)) {
2012 add_mirror_actions(&ctx, &orig_flow);
2014 fix_sflow_action(&ctx);
2017 ofpbuf_uninit(&ctx.stack);
2019 /* Clear the metadata and register wildcard masks, because we won't
2020 * use non-header fields as part of the cache. */
2021 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2022 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);