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-sflow.h"
42 #include "ofproto/ofproto-dpif.h"
46 COVERAGE_DEFINE(ofproto_dpif_xlate);
48 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
50 /* Maximum depth of flow table recursion (due to resubmit actions) in a
51 * flow translation. */
52 #define MAX_RESUBMIT_RECURSION 64
56 struct xlate_out *xout;
58 struct ofproto_dpif *ofproto;
60 /* Flow at the last commit. */
61 struct flow base_flow;
63 /* Tunnel IP destination address as received. This is stored separately
64 * as the base_flow.tunnel is cleared on init to reflect the datapath
65 * behavior. Used to make sure not to send tunneled output to ourselves,
66 * which might lead to an infinite loop. This could happen easily
67 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
68 * actually set the tun_dst field. */
69 ovs_be32 orig_tunnel_ip_dst;
71 /* Stack for the push and pop actions. Each stack element is of type
72 * "union mf_subvalue". */
73 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
76 /* The rule that we are currently translating, or NULL. */
77 struct rule_dpif *rule;
79 int recurse; /* Recursion level, via xlate_table_action. */
80 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
81 uint32_t orig_skb_priority; /* Priority when packet arrived. */
82 uint8_t table_id; /* OpenFlow table ID where flow was found. */
83 uint32_t sflow_n_outputs; /* Number of output ports. */
84 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
85 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
86 bool exit; /* No further actions should be processed. */
89 /* A controller may use OFPP_NONE as the ingress port to indicate that
90 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
91 * when an input bundle is needed for validation (e.g., mirroring or
92 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
93 * any 'port' structs, so care must be taken when dealing with it. */
94 static struct ofbundle ofpp_none_bundle = {
96 .vlan_mode = PORT_VLAN_TRUNK
99 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
100 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
102 static void xlate_normal(struct xlate_ctx *);
103 static void xlate_report(struct xlate_ctx *, const char *);
104 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
105 uint8_t table_id, bool may_packet_in);
106 static bool input_vid_is_valid(uint16_t vid, struct ofbundle *, bool warn);
107 static uint16_t input_vid_to_vlan(const struct ofbundle *, uint16_t vid);
108 static void output_normal(struct xlate_ctx *, const struct ofbundle *,
110 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
112 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
115 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
117 return (bundle->vlan_mode != PORT_VLAN_ACCESS
118 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
122 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
124 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
128 vlan_is_mirrored(const struct ofmirror *m, int vlan)
130 return !m->vlans || bitmap_is_set(m->vlans, vlan);
133 static struct ofbundle *
134 lookup_input_bundle(const struct ofproto_dpif *ofproto, ofp_port_t in_port,
135 bool warn, struct ofport_dpif **in_ofportp)
137 struct ofport_dpif *ofport;
139 /* Find the port and bundle for the received packet. */
140 ofport = get_ofp_port(ofproto, in_port);
142 *in_ofportp = ofport;
144 if (ofport && ofport->bundle) {
145 return ofport->bundle;
148 /* Special-case OFPP_NONE, which a controller may use as the ingress
149 * port for traffic that it is sourcing. */
150 if (in_port == OFPP_NONE) {
151 return &ofpp_none_bundle;
154 /* Odd. A few possible reasons here:
156 * - We deleted a port but there are still a few packets queued up
159 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
160 * we don't know about.
162 * - The ofproto client didn't configure the port as part of a bundle.
163 * This is particularly likely to happen if a packet was received on the
164 * port after it was created, but before the client had a chance to
165 * configure its bundle.
168 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
170 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
171 "port %"PRIu16, ofproto->up.name, in_port);
177 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
179 struct ofproto_dpif *ofproto = ctx->ofproto;
180 mirror_mask_t mirrors;
181 struct ofbundle *in_bundle;
185 mirrors = ctx->xout->mirrors;
186 ctx->xout->mirrors = 0;
188 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port.ofp_port,
189 ctx->xin->packet != NULL, NULL);
193 mirrors |= in_bundle->src_mirrors;
195 /* Drop frames on bundles reserved for mirroring. */
196 if (in_bundle->mirror_out) {
197 if (ctx->xin->packet != NULL) {
198 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
199 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
200 "%s, which is reserved exclusively for mirroring",
201 ctx->ofproto->up.name, in_bundle->name);
207 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
208 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
211 vlan = input_vid_to_vlan(in_bundle, vid);
217 /* Restore the original packet before adding the mirror actions. */
218 ctx->xin->flow = *orig_flow;
223 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
226 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
229 if (!vlan_is_mirrored(m, vlan)) {
230 mirrors = zero_rightmost_1bit(mirrors);
234 mirrors &= ~m->dup_mirrors;
235 ctx->xout->mirrors |= m->dup_mirrors;
237 output_normal(ctx, m->out, vlan);
238 } else if (vlan != m->out_vlan
239 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
240 struct ofbundle *bundle;
242 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
243 if (ofbundle_includes_vlan(bundle, m->out_vlan)
244 && !bundle->mirror_out) {
245 output_normal(ctx, bundle, m->out_vlan);
252 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
253 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
254 * the bundle on which the packet was received, returns the VLAN to which the
257 * Both 'vid' and the return value are in the range 0...4095. */
259 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
261 switch (in_bundle->vlan_mode) {
262 case PORT_VLAN_ACCESS:
263 return in_bundle->vlan;
266 case PORT_VLAN_TRUNK:
269 case PORT_VLAN_NATIVE_UNTAGGED:
270 case PORT_VLAN_NATIVE_TAGGED:
271 return vid ? vid : in_bundle->vlan;
278 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
279 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
282 * 'vid' should be the VID obtained from the 802.1Q header that was received as
283 * part of a packet (specify 0 if there was no 802.1Q header), in the range
286 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
288 /* Allow any VID on the OFPP_NONE port. */
289 if (in_bundle == &ofpp_none_bundle) {
293 switch (in_bundle->vlan_mode) {
294 case PORT_VLAN_ACCESS:
297 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
298 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
299 "packet received on port %s configured as VLAN "
300 "%"PRIu16" access port",
301 in_bundle->ofproto->up.name, vid,
302 in_bundle->name, in_bundle->vlan);
308 case PORT_VLAN_NATIVE_UNTAGGED:
309 case PORT_VLAN_NATIVE_TAGGED:
311 /* Port must always carry its native VLAN. */
315 case PORT_VLAN_TRUNK:
316 if (!ofbundle_includes_vlan(in_bundle, vid)) {
318 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
319 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
320 "received on port %s not configured for trunking "
322 in_bundle->ofproto->up.name, vid,
323 in_bundle->name, vid);
335 /* Given 'vlan', the VLAN that a packet belongs to, and
336 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
337 * that should be included in the 802.1Q header. (If the return value is 0,
338 * then the 802.1Q header should only be included in the packet if there is a
341 * Both 'vlan' and the return value are in the range 0...4095. */
343 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
345 switch (out_bundle->vlan_mode) {
346 case PORT_VLAN_ACCESS:
349 case PORT_VLAN_TRUNK:
350 case PORT_VLAN_NATIVE_TAGGED:
353 case PORT_VLAN_NATIVE_UNTAGGED:
354 return vlan == out_bundle->vlan ? 0 : vlan;
362 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
365 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
366 struct ofport_dpif *port;
368 ovs_be16 tci, old_tci;
370 vid = output_vlan_to_vid(out_bundle, vlan);
371 if (!out_bundle->bond) {
372 port = ofbundle_get_a_port(out_bundle);
374 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
375 &ctx->xout->wc, vid, &ctx->xout->tags);
377 /* No slaves enabled, so drop packet. */
384 if (tci || out_bundle->use_priority_tags) {
385 tci |= *flow_tci & htons(VLAN_PCP_MASK);
387 tci |= htons(VLAN_CFI);
392 compose_output_action(ctx, port->up.ofp_port);
396 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
397 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
398 * indicate this; newer upstream kernels use gratuitous ARP requests. */
400 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
402 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
406 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
407 if (!eth_addr_is_broadcast(flow->dl_dst)) {
411 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
412 if (flow->nw_proto == ARP_OP_REPLY) {
414 } else if (flow->nw_proto == ARP_OP_REQUEST) {
415 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
416 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
418 return flow->nw_src == flow->nw_dst;
425 update_learning_table(struct ofproto_dpif *ofproto,
426 const struct flow *flow, struct flow_wildcards *wc,
427 int vlan, struct ofbundle *in_bundle)
429 struct mac_entry *mac;
431 /* Don't learn the OFPP_NONE port. */
432 if (in_bundle == &ofpp_none_bundle) {
436 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
440 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
441 if (is_gratuitous_arp(flow, wc)) {
442 /* We don't want to learn from gratuitous ARP packets that are
443 * reflected back over bond slaves so we lock the learning table. */
444 if (!in_bundle->bond) {
445 mac_entry_set_grat_arp_lock(mac);
446 } else if (mac_entry_is_grat_arp_locked(mac)) {
451 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
452 /* The log messages here could actually be useful in debugging,
453 * so keep the rate limit relatively high. */
454 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
455 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
456 "on port %s in VLAN %d",
457 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
458 in_bundle->name, vlan);
460 mac->port.p = in_bundle;
461 mac_learning_changed(ofproto->ml, mac);
465 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
466 * dropped. Returns true if they may be forwarded, false if they should be
469 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
470 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
472 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
473 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
474 * checked by input_vid_is_valid().
476 * May also add tags to '*tags', although the current implementation only does
477 * so in one special case.
480 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
483 struct ofproto_dpif *ofproto = ctx->ofproto;
484 struct flow *flow = &ctx->xin->flow;
485 struct ofbundle *in_bundle = in_port->bundle;
487 /* Drop frames for reserved multicast addresses
488 * only if forward_bpdu option is absent. */
489 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
490 xlate_report(ctx, "packet has reserved destination MAC, dropping");
494 if (in_bundle->bond) {
495 struct mac_entry *mac;
497 switch (bond_check_admissibility(in_bundle->bond, in_port,
498 flow->dl_dst, &ctx->xout->tags)) {
503 xlate_report(ctx, "bonding refused admissibility, dropping");
506 case BV_DROP_IF_MOVED:
507 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
508 if (mac && mac->port.p != in_bundle &&
509 (!is_gratuitous_arp(flow, &ctx->xout->wc)
510 || mac_entry_is_grat_arp_locked(mac))) {
511 xlate_report(ctx, "SLB bond thinks this packet looped back, "
523 xlate_normal(struct xlate_ctx *ctx)
525 struct flow_wildcards *wc = &ctx->xout->wc;
526 struct flow *flow = &ctx->xin->flow;
527 struct ofport_dpif *in_port;
528 struct ofbundle *in_bundle;
529 struct mac_entry *mac;
533 ctx->xout->has_normal = true;
535 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
536 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
537 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
539 in_bundle = lookup_input_bundle(ctx->ofproto, flow->in_port.ofp_port,
540 ctx->xin->packet != NULL, &in_port);
542 xlate_report(ctx, "no input bundle, dropping");
546 /* Drop malformed frames. */
547 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
548 !(flow->vlan_tci & htons(VLAN_CFI))) {
549 if (ctx->xin->packet != NULL) {
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
551 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
552 "VLAN tag received on port %s",
553 ctx->ofproto->up.name, in_bundle->name);
555 xlate_report(ctx, "partial VLAN tag, dropping");
559 /* Drop frames on bundles reserved for mirroring. */
560 if (in_bundle->mirror_out) {
561 if (ctx->xin->packet != NULL) {
562 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
563 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
564 "%s, which is reserved exclusively for mirroring",
565 ctx->ofproto->up.name, in_bundle->name);
567 xlate_report(ctx, "input port is mirror output port, dropping");
572 vid = vlan_tci_to_vid(flow->vlan_tci);
573 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
574 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
577 vlan = input_vid_to_vlan(in_bundle, vid);
579 /* Check other admissibility requirements. */
580 if (in_port && !is_admissible(ctx, in_port, vlan)) {
584 /* Learn source MAC. */
585 if (ctx->xin->may_learn) {
586 update_learning_table(ctx->ofproto, flow, wc, vlan, in_bundle);
589 /* Determine output bundle. */
590 mac = mac_learning_lookup(ctx->ofproto->ml, flow->dl_dst, vlan,
593 if (mac->port.p != in_bundle) {
594 xlate_report(ctx, "forwarding to learned port");
595 output_normal(ctx, mac->port.p, vlan);
597 xlate_report(ctx, "learned port is input port, dropping");
600 struct ofbundle *bundle;
602 xlate_report(ctx, "no learned MAC for destination, flooding");
603 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
604 if (bundle != in_bundle
605 && ofbundle_includes_vlan(bundle, vlan)
607 && !bundle->mirror_out) {
608 output_normal(ctx, bundle, vlan);
611 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
615 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
616 * the number of packets out of UINT32_MAX to sample. The given
617 * cookie is passed back in the callback for each sampled packet.
620 compose_sample_action(const struct ofproto_dpif *ofproto,
621 struct ofpbuf *odp_actions,
622 const struct flow *flow,
623 const uint32_t probability,
624 const union user_action_cookie *cookie,
625 const size_t cookie_size)
627 size_t sample_offset, actions_offset;
630 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
632 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
634 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
635 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
638 nl_msg_end_nested(odp_actions, actions_offset);
639 nl_msg_end_nested(odp_actions, sample_offset);
640 return cookie_offset;
644 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
645 ovs_be16 vlan_tci, odp_port_t odp_port,
646 unsigned int n_outputs, union user_action_cookie *cookie)
650 cookie->type = USER_ACTION_COOKIE_SFLOW;
651 cookie->sflow.vlan_tci = vlan_tci;
653 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
654 * port information") for the interpretation of cookie->output. */
657 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
658 cookie->sflow.output = 0x40000000 | 256;
662 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
664 cookie->sflow.output = ifindex;
669 /* 0x80000000 means "multiple output ports. */
670 cookie->sflow.output = 0x80000000 | n_outputs;
675 /* Compose SAMPLE action for sFlow bridge sampling. */
677 compose_sflow_action(const struct ofproto_dpif *ofproto,
678 struct ofpbuf *odp_actions,
679 const struct flow *flow,
682 uint32_t probability;
683 union user_action_cookie cookie;
685 if (!ofproto->sflow || flow->in_port.ofp_port == OFPP_NONE) {
689 probability = dpif_sflow_get_probability(ofproto->sflow);
690 compose_sflow_cookie(ofproto, htons(0), odp_port,
691 odp_port == ODPP_NONE ? 0 : 1, &cookie);
693 return compose_sample_action(ofproto, odp_actions, flow, probability,
694 &cookie, sizeof cookie.sflow);
698 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
699 uint32_t obs_domain_id, uint32_t obs_point_id,
700 union user_action_cookie *cookie)
702 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
703 cookie->flow_sample.probability = probability;
704 cookie->flow_sample.collector_set_id = collector_set_id;
705 cookie->flow_sample.obs_domain_id = obs_domain_id;
706 cookie->flow_sample.obs_point_id = obs_point_id;
710 compose_ipfix_cookie(union user_action_cookie *cookie)
712 cookie->type = USER_ACTION_COOKIE_IPFIX;
715 /* Compose SAMPLE action for IPFIX bridge sampling. */
717 compose_ipfix_action(const struct ofproto_dpif *ofproto,
718 struct ofpbuf *odp_actions,
719 const struct flow *flow)
721 uint32_t probability;
722 union user_action_cookie cookie;
724 if (!ofproto->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
728 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
729 compose_ipfix_cookie(&cookie);
731 compose_sample_action(ofproto, odp_actions, flow, probability,
732 &cookie, sizeof cookie.ipfix);
735 /* SAMPLE action for sFlow must be first action in any given list of
736 * actions. At this point we do not have all information required to
737 * build it. So try to build sample action as complete as possible. */
739 add_sflow_action(struct xlate_ctx *ctx)
741 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
742 &ctx->xout->odp_actions,
743 &ctx->xin->flow, ODPP_NONE);
744 ctx->sflow_odp_port = 0;
745 ctx->sflow_n_outputs = 0;
748 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
749 * of actions, eventually after the SAMPLE action for sFlow. */
751 add_ipfix_action(struct xlate_ctx *ctx)
753 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
757 /* Fix SAMPLE action according to data collected while composing ODP actions.
758 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
759 * USERSPACE action's user-cookie which is required for sflow. */
761 fix_sflow_action(struct xlate_ctx *ctx)
763 const struct flow *base = &ctx->base_flow;
764 union user_action_cookie *cookie;
766 if (!ctx->user_cookie_offset) {
770 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
771 sizeof cookie->sflow);
772 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
774 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
775 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
778 static enum slow_path_reason
779 process_special(struct xlate_ctx *ctx, const struct flow *flow,
780 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
782 struct ofproto_dpif *ofproto = ctx->ofproto;
783 struct flow_wildcards *wc = &ctx->xout->wc;
787 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow, wc)) {
789 cfm_process_heartbeat(ofport->cfm, packet);
792 } else if (ofport->bfd && bfd_should_process_flow(flow, wc)) {
794 bfd_process_packet(ofport->bfd, flow, packet);
797 } else if (ofport->bundle && ofport->bundle->lacp
798 && flow->dl_type == htons(ETH_TYPE_LACP)) {
800 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
803 } else if (ofproto->stp && stp_should_process_flow(flow, wc)) {
805 stp_process_packet(ofport, packet);
814 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
817 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
818 struct flow_wildcards *wc = &ctx->xout->wc;
819 struct flow *flow = &ctx->xin->flow;
820 ovs_be16 flow_vlan_tci;
821 uint32_t flow_skb_mark;
823 odp_port_t out_port, odp_port;
826 /* If 'struct flow' gets additional metadata, we'll need to zero it out
827 * before traversing a patch port. */
828 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
831 xlate_report(ctx, "Nonexistent output port");
833 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
834 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
836 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
837 xlate_report(ctx, "STP not in forwarding state, skipping output");
841 if (ctx->ofproto->has_mirrors && ofport->bundle) {
842 ctx->xout->mirrors |= ofport->bundle->dst_mirrors;
846 struct ofport_dpif *peer = ofport->peer;
847 struct flow old_flow = ctx->xin->flow;
848 enum slow_path_reason special;
850 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
851 flow->in_port.ofp_port = peer->up.ofp_port;
852 flow->metadata = htonll(0);
853 memset(&flow->tunnel, 0, sizeof flow->tunnel);
854 memset(flow->regs, 0, sizeof flow->regs);
856 special = process_special(ctx, &ctx->xin->flow, peer,
859 ctx->xout->slow = special;
860 } else if (may_receive(peer, ctx)) {
861 if (stp_forward_in_state(peer->stp_state)) {
862 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
864 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
865 * learning action look at the packet, then drop it. */
866 struct flow old_base_flow = ctx->base_flow;
867 size_t old_size = ctx->xout->odp_actions.size;
868 mirror_mask_t old_mirrors = ctx->xout->mirrors;
869 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
870 ctx->xout->mirrors = old_mirrors;
871 ctx->base_flow = old_base_flow;
872 ctx->xout->odp_actions.size = old_size;
876 ctx->xin->flow = old_flow;
877 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
879 if (ctx->xin->resubmit_stats) {
880 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
881 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
887 flow_vlan_tci = flow->vlan_tci;
888 flow_skb_mark = flow->skb_mark;
889 flow_nw_tos = flow->nw_tos;
891 if (ofproto_dpif_dscp_from_priority(ofport, flow->skb_priority, &dscp)) {
892 wc->masks.nw_tos |= IP_ECN_MASK;
893 flow->nw_tos &= ~IP_DSCP_MASK;
894 flow->nw_tos |= dscp;
897 if (ofport->is_tunnel) {
898 /* Save tunnel metadata so that changes made due to
899 * the Logical (tunnel) Port are not visible for any further
900 * matches, while explicit set actions on tunnel metadata are.
902 struct flow_tnl flow_tnl = flow->tunnel;
903 odp_port = tnl_port_send(ofport, flow, &ctx->xout->wc);
904 if (odp_port == ODPP_NONE) {
905 xlate_report(ctx, "Tunneling decided against output");
906 goto out; /* restore flow_nw_tos */
908 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
909 xlate_report(ctx, "Not tunneling to our own address");
910 goto out; /* restore flow_nw_tos */
912 if (ctx->xin->resubmit_stats) {
913 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
916 commit_odp_tunnel_action(flow, &ctx->base_flow,
917 &ctx->xout->odp_actions);
918 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
920 ofp_port_t vlandev_port;
922 odp_port = ofport->odp_port;
923 if (!hmap_is_empty(&ctx->ofproto->realdev_vid_map)) {
924 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
926 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
928 if (vlandev_port == ofp_port) {
931 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
932 flow->vlan_tci = htons(0);
934 flow->skb_mark &= ~IPSEC_MARK;
937 if (out_port != ODPP_NONE) {
938 commit_odp_actions(flow, &ctx->base_flow,
939 &ctx->xout->odp_actions, &ctx->xout->wc);
940 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
943 ctx->sflow_odp_port = odp_port;
944 ctx->sflow_n_outputs++;
945 ctx->xout->nf_output_iface = ofp_port;
950 flow->vlan_tci = flow_vlan_tci;
951 flow->skb_mark = flow_skb_mark;
952 flow->nw_tos = flow_nw_tos;
956 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
958 compose_output_action__(ctx, ofp_port, true);
961 /* Common rule processing in one place to avoid duplicating code. */
962 static struct rule_dpif *
963 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
966 if (ctx->xin->resubmit_hook) {
967 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
969 if (rule == NULL && may_packet_in) {
971 * check if table configuration flags
972 * OFPTC_TABLE_MISS_CONTROLLER, default.
973 * OFPTC_TABLE_MISS_CONTINUE,
974 * OFPTC_TABLE_MISS_DROP
975 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
977 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
979 if (rule && ctx->xin->resubmit_stats) {
980 rule_credit_stats(rule, ctx->xin->resubmit_stats);
986 xlate_table_action(struct xlate_ctx *ctx,
987 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
989 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
990 struct rule_dpif *rule;
991 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
992 uint8_t old_table_id = ctx->table_id;
994 ctx->table_id = table_id;
996 /* Look up a flow with 'in_port' as the input port. */
997 ctx->xin->flow.in_port.ofp_port = in_port;
998 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
999 &ctx->xout->wc, table_id);
1001 ctx->xout->tags |= calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1002 ctx->table_id, rule);
1004 /* Restore the original input port. Otherwise OFPP_NORMAL and
1005 * OFPP_IN_PORT will have surprising behavior. */
1006 ctx->xin->flow.in_port.ofp_port = old_in_port;
1008 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1011 struct rule_dpif *old_rule = ctx->rule;
1015 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1016 ctx->rule = old_rule;
1020 ctx->table_id = old_table_id;
1022 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1024 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1025 MAX_RESUBMIT_RECURSION);
1026 ctx->max_resubmit_trigger = true;
1031 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1032 const struct ofpact_resubmit *resubmit)
1037 in_port = resubmit->in_port;
1038 if (in_port == OFPP_IN_PORT) {
1039 in_port = ctx->xin->flow.in_port.ofp_port;
1042 table_id = resubmit->table_id;
1043 if (table_id == 255) {
1044 table_id = ctx->table_id;
1047 xlate_table_action(ctx, in_port, table_id, false);
1051 flood_packets(struct xlate_ctx *ctx, bool all)
1053 struct ofport_dpif *ofport;
1055 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
1056 ofp_port_t ofp_port = ofport->up.ofp_port;
1058 if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1063 compose_output_action__(ctx, ofp_port, false);
1064 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
1065 compose_output_action(ctx, ofp_port);
1069 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1073 execute_controller_action(struct xlate_ctx *ctx, int len,
1074 enum ofp_packet_in_reason reason,
1075 uint16_t controller_id)
1077 struct ofputil_packet_in pin;
1078 struct ofpbuf *packet;
1081 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1082 ctx->xout->slow = SLOW_CONTROLLER;
1083 if (!ctx->xin->packet) {
1087 packet = ofpbuf_clone(ctx->xin->packet);
1089 key.skb_priority = 0;
1091 memset(&key.tunnel, 0, sizeof key.tunnel);
1093 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1094 &ctx->xout->odp_actions, &ctx->xout->wc);
1096 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1097 ctx->xout->odp_actions.size, NULL, NULL);
1099 pin.packet = packet->data;
1100 pin.packet_len = packet->size;
1101 pin.reason = reason;
1102 pin.controller_id = controller_id;
1103 pin.table_id = ctx->table_id;
1104 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1107 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1109 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
1110 ofpbuf_delete(packet);
1114 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1116 struct flow_wildcards *wc = &ctx->xout->wc;
1117 struct flow *flow = &ctx->xin->flow;
1119 ovs_assert(eth_type_mpls(eth_type));
1121 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1122 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1124 if (flow->mpls_depth) {
1125 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1131 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1132 label = htonl(0x2); /* IPV6 Explicit Null. */
1134 label = htonl(0x0); /* IPV4 Explicit Null. */
1136 wc->masks.nw_tos |= IP_DSCP_MASK;
1137 wc->masks.nw_ttl = 0xff;
1138 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1139 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1140 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1141 flow->mpls_depth = 1;
1143 flow->dl_type = eth_type;
1147 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1149 struct flow_wildcards *wc = &ctx->xout->wc;
1150 struct flow *flow = &ctx->xin->flow;
1152 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1153 ovs_assert(!eth_type_mpls(eth_type));
1155 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1156 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1158 if (flow->mpls_depth) {
1160 flow->mpls_lse = htonl(0);
1161 if (!flow->mpls_depth) {
1162 flow->dl_type = eth_type;
1168 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1170 struct flow *flow = &ctx->xin->flow;
1172 if (!is_ip_any(flow)) {
1176 ctx->xout->wc.masks.nw_ttl = 0xff;
1177 if (flow->nw_ttl > 1) {
1183 for (i = 0; i < ids->n_controllers; i++) {
1184 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1188 /* Stop processing for current table. */
1194 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1196 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1200 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1205 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1207 struct flow *flow = &ctx->xin->flow;
1208 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1209 struct flow_wildcards *wc = &ctx->xout->wc;
1211 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1213 if (!eth_type_mpls(flow->dl_type)) {
1219 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1222 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1224 /* Stop processing for current table. */
1230 xlate_output_action(struct xlate_ctx *ctx,
1231 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1233 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1235 ctx->xout->nf_output_iface = NF_OUT_DROP;
1239 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1242 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1249 flood_packets(ctx, false);
1252 flood_packets(ctx, true);
1254 case OFPP_CONTROLLER:
1255 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1261 if (port != ctx->xin->flow.in_port.ofp_port) {
1262 compose_output_action(ctx, port);
1264 xlate_report(ctx, "skipping output to input port");
1269 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1270 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1271 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1272 ctx->xout->nf_output_iface = prev_nf_output_iface;
1273 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1274 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1275 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1280 xlate_output_reg_action(struct xlate_ctx *ctx,
1281 const struct ofpact_output_reg *or)
1283 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1284 if (port <= UINT16_MAX) {
1285 union mf_subvalue value;
1287 memset(&value, 0xff, sizeof value);
1288 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1289 xlate_output_action(ctx, u16_to_ofp(port),
1290 or->max_len, false);
1295 xlate_enqueue_action(struct xlate_ctx *ctx,
1296 const struct ofpact_enqueue *enqueue)
1298 ofp_port_t ofp_port = enqueue->port;
1299 uint32_t queue_id = enqueue->queue;
1300 uint32_t flow_priority, priority;
1303 /* Translate queue to priority. */
1304 error = ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id, &priority);
1306 /* Fall back to ordinary output action. */
1307 xlate_output_action(ctx, enqueue->port, 0, false);
1311 /* Check output port. */
1312 if (ofp_port == OFPP_IN_PORT) {
1313 ofp_port = ctx->xin->flow.in_port.ofp_port;
1314 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1318 /* Add datapath actions. */
1319 flow_priority = ctx->xin->flow.skb_priority;
1320 ctx->xin->flow.skb_priority = priority;
1321 compose_output_action(ctx, ofp_port);
1322 ctx->xin->flow.skb_priority = flow_priority;
1324 /* Update NetFlow output port. */
1325 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1326 ctx->xout->nf_output_iface = ofp_port;
1327 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1328 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1333 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1335 uint32_t skb_priority;
1337 if (!ofproto_dpif_queue_to_priority(ctx->ofproto, queue_id,
1339 ctx->xin->flow.skb_priority = skb_priority;
1341 /* Couldn't translate queue to a priority. Nothing to do. A warning
1342 * has already been logged. */
1347 slave_enabled_cb(ofp_port_t ofp_port, void *ofproto_)
1349 struct ofproto_dpif *ofproto = ofproto_;
1350 struct ofport_dpif *port;
1360 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1363 port = get_ofp_port(ofproto, ofp_port);
1364 return port ? port->may_enable : false;
1369 xlate_bundle_action(struct xlate_ctx *ctx,
1370 const struct ofpact_bundle *bundle)
1374 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1375 slave_enabled_cb, ctx->ofproto);
1376 if (bundle->dst.field) {
1377 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1379 xlate_output_action(ctx, port, 0, false);
1384 xlate_learn_action(struct xlate_ctx *ctx,
1385 const struct ofpact_learn *learn)
1387 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1388 struct ofputil_flow_mod fm;
1389 uint64_t ofpacts_stub[1024 / 8];
1390 struct ofpbuf ofpacts;
1393 ctx->xout->has_learn = true;
1395 learn_mask(learn, &ctx->xout->wc);
1397 if (!ctx->xin->may_learn) {
1401 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1402 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1404 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
1405 if (error && !VLOG_DROP_WARN(&rl)) {
1406 VLOG_WARN("learning action failed to modify flow table (%s)",
1407 ofperr_get_name(error));
1410 ofpbuf_uninit(&ofpacts);
1413 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1414 * means "infinite". */
1416 reduce_timeout(uint16_t max, uint16_t *timeout)
1418 if (max && (!*timeout || *timeout > max)) {
1424 xlate_fin_timeout(struct xlate_ctx *ctx,
1425 const struct ofpact_fin_timeout *oft)
1427 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1428 struct rule_dpif *rule = ctx->rule;
1430 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1431 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1436 xlate_sample_action(struct xlate_ctx *ctx,
1437 const struct ofpact_sample *os)
1439 union user_action_cookie cookie;
1440 /* Scale the probability from 16-bit to 32-bit while representing
1441 * the same percentage. */
1442 uint32_t probability = (os->probability << 16) | os->probability;
1444 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1445 &ctx->xout->odp_actions, &ctx->xout->wc);
1447 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1448 os->obs_domain_id, os->obs_point_id, &cookie);
1449 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
1450 probability, &cookie, sizeof cookie.flow_sample);
1454 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
1456 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
1458 ? OFPUTIL_PC_NO_RECV_STP
1459 : OFPUTIL_PC_NO_RECV)) {
1463 /* Only drop packets here if both forwarding and learning are
1464 * disabled. If just learning is enabled, we need to have
1465 * OFPP_NORMAL and the learning action have a look at the packet
1466 * before we can drop it. */
1467 if (!stp_forward_in_state(port->stp_state)
1468 && !stp_learn_in_state(port->stp_state)) {
1476 tunnel_ecn_ok(struct xlate_ctx *ctx)
1478 if (is_ip_any(&ctx->base_flow)
1479 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1480 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1481 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1482 " but is not ECN capable");
1485 /* Set the ECN CE value in the tunneled packet. */
1486 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1494 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1495 struct xlate_ctx *ctx)
1497 struct flow_wildcards *wc = &ctx->xout->wc;
1498 struct flow *flow = &ctx->xin->flow;
1499 bool was_evictable = true;
1500 const struct ofpact *a;
1503 /* Don't let the rule we're working on get evicted underneath us. */
1504 was_evictable = ctx->rule->up.evictable;
1505 ctx->rule->up.evictable = false;
1508 do_xlate_actions_again:
1509 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1510 struct ofpact_controller *controller;
1511 const struct ofpact_metadata *metadata;
1519 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1520 ofpact_get_OUTPUT(a)->max_len, true);
1523 case OFPACT_CONTROLLER:
1524 controller = ofpact_get_CONTROLLER(a);
1525 execute_controller_action(ctx, controller->max_len,
1527 controller->controller_id);
1530 case OFPACT_ENQUEUE:
1531 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1534 case OFPACT_SET_VLAN_VID:
1535 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1536 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1540 case OFPACT_SET_VLAN_PCP:
1541 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1543 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1547 case OFPACT_STRIP_VLAN:
1548 flow->vlan_tci = htons(0);
1551 case OFPACT_PUSH_VLAN:
1552 /* XXX 802.1AD(QinQ) */
1553 flow->vlan_tci = htons(VLAN_CFI);
1556 case OFPACT_SET_ETH_SRC:
1557 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
1560 case OFPACT_SET_ETH_DST:
1561 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
1564 case OFPACT_SET_IPV4_SRC:
1565 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1566 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1570 case OFPACT_SET_IPV4_DST:
1571 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1572 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1576 case OFPACT_SET_IPV4_DSCP:
1577 /* OpenFlow 1.0 only supports IPv4. */
1578 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1579 flow->nw_tos &= ~IP_DSCP_MASK;
1580 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1584 case OFPACT_SET_L4_SRC_PORT:
1585 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1586 if (is_ip_any(flow)) {
1587 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1591 case OFPACT_SET_L4_DST_PORT:
1592 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1593 if (is_ip_any(flow)) {
1594 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
1598 case OFPACT_RESUBMIT:
1599 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
1602 case OFPACT_SET_TUNNEL:
1603 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
1606 case OFPACT_SET_QUEUE:
1607 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
1610 case OFPACT_POP_QUEUE:
1611 flow->skb_priority = ctx->orig_skb_priority;
1614 case OFPACT_REG_MOVE:
1615 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
1618 case OFPACT_REG_LOAD:
1619 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
1622 case OFPACT_STACK_PUSH:
1623 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
1627 case OFPACT_STACK_POP:
1628 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
1631 case OFPACT_PUSH_MPLS:
1632 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
1635 case OFPACT_POP_MPLS:
1636 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
1639 case OFPACT_SET_MPLS_TTL:
1640 if (compose_set_mpls_ttl_action(ctx,
1641 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
1646 case OFPACT_DEC_MPLS_TTL:
1647 if (compose_dec_mpls_ttl_action(ctx)) {
1652 case OFPACT_DEC_TTL:
1653 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
1659 /* Nothing to do. */
1662 case OFPACT_MULTIPATH:
1663 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
1667 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
1670 case OFPACT_OUTPUT_REG:
1671 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
1675 xlate_learn_action(ctx, ofpact_get_LEARN(a));
1682 case OFPACT_FIN_TIMEOUT:
1683 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1684 ctx->xout->has_fin_timeout = true;
1685 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
1688 case OFPACT_CLEAR_ACTIONS:
1690 * Nothing to do because writa-actions is not supported for now.
1691 * When writa-actions is supported, clear-actions also must
1692 * be supported at the same time.
1696 case OFPACT_WRITE_METADATA:
1697 metadata = ofpact_get_WRITE_METADATA(a);
1698 flow->metadata &= ~metadata->mask;
1699 flow->metadata |= metadata->metadata & metadata->mask;
1703 /* Not implemented yet. */
1706 case OFPACT_GOTO_TABLE: {
1707 /* It is assumed that goto-table is the last action. */
1708 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
1709 struct rule_dpif *rule;
1711 ovs_assert(ctx->table_id < ogt->table_id);
1713 ctx->table_id = ogt->table_id;
1715 /* Look up a flow from the new table. */
1716 rule = rule_dpif_lookup_in_table(ctx->ofproto, flow, wc,
1719 ctx->xout->tags = calculate_flow_tag(ctx->ofproto, &ctx->xin->flow,
1720 ctx->table_id, rule);
1722 rule = ctx_rule_hooks(ctx, rule, true);
1726 ctx->rule->up.evictable = was_evictable;
1729 was_evictable = rule->up.evictable;
1730 rule->up.evictable = false;
1732 /* Tail recursion removal. */
1733 ofpacts = rule->up.ofpacts;
1734 ofpacts_len = rule->up.ofpacts_len;
1735 goto do_xlate_actions_again;
1741 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
1748 ctx->rule->up.evictable = was_evictable;
1753 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
1754 const struct flow *flow, struct rule_dpif *rule,
1755 uint8_t tcp_flags, const struct ofpbuf *packet)
1757 xin->ofproto = ofproto;
1759 xin->packet = packet;
1760 xin->may_learn = packet != NULL;
1762 xin->ofpacts = NULL;
1763 xin->ofpacts_len = 0;
1764 xin->tcp_flags = tcp_flags;
1765 xin->resubmit_hook = NULL;
1766 xin->report_hook = NULL;
1767 xin->resubmit_stats = NULL;
1771 xlate_out_uninit(struct xlate_out *xout)
1774 ofpbuf_uninit(&xout->odp_actions);
1778 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
1779 * into datapath actions, using 'ctx', and discards the datapath actions. */
1781 xlate_actions_for_side_effects(struct xlate_in *xin)
1783 struct xlate_out xout;
1785 xlate_actions(xin, &xout);
1786 xlate_out_uninit(&xout);
1790 xlate_report(struct xlate_ctx *ctx, const char *s)
1792 if (ctx->xin->report_hook) {
1793 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
1798 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
1801 dst->tags = src->tags;
1802 dst->slow = src->slow;
1803 dst->has_learn = src->has_learn;
1804 dst->has_normal = src->has_normal;
1805 dst->has_fin_timeout = src->has_fin_timeout;
1806 dst->nf_output_iface = src->nf_output_iface;
1807 dst->mirrors = src->mirrors;
1809 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
1810 sizeof dst->odp_actions_stub);
1811 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
1812 src->odp_actions.size);
1816 actions_output_to_local_port(const struct xlate_ctx *ctx)
1818 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
1819 const struct nlattr *a;
1822 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
1823 ctx->xout->odp_actions.size) {
1824 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
1825 && nl_attr_get_odp_port(a) == local_odp_port) {
1832 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
1833 * into datapath actions in 'odp_actions', using 'ctx'. */
1835 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
1837 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
1838 * that in the future we always keep a copy of the original flow for
1839 * tracing purposes. */
1840 static bool hit_resubmit_limit;
1842 struct flow_wildcards *wc = &xout->wc;
1843 struct flow *flow = &xin->flow;
1845 enum slow_path_reason special;
1846 const struct ofpact *ofpacts;
1847 struct ofport_dpif *in_port;
1848 struct flow orig_flow;
1849 struct xlate_ctx ctx;
1852 COVERAGE_INC(ofproto_dpif_xlate);
1854 /* Flow initialization rules:
1855 * - 'base_flow' must match the kernel's view of the packet at the
1856 * time that action processing starts. 'flow' represents any
1857 * transformations we wish to make through actions.
1858 * - By default 'base_flow' and 'flow' are the same since the input
1859 * packet matches the output before any actions are applied.
1860 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
1861 * of the received packet as seen by the kernel. If we later output
1862 * to another device without any modifications this will cause us to
1863 * insert a new tag since the original one was stripped off by the
1865 * - Tunnel metadata as received is retained in 'flow'. This allows
1866 * tunnel metadata matching also in later tables.
1867 * Since a kernel action for setting the tunnel metadata will only be
1868 * generated with actual tunnel output, changing the tunnel metadata
1869 * values in 'flow' (such as tun_id) will only have effect with a later
1870 * tunnel output action.
1871 * - Tunnel 'base_flow' is completely cleared since that is what the
1872 * kernel does. If we wish to maintain the original values an action
1873 * needs to be generated. */
1878 ctx.ofproto = xin->ofproto;
1879 ctx.rule = xin->rule;
1881 ctx.base_flow = *flow;
1882 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
1883 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
1885 flow_wildcards_init_catchall(wc);
1886 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
1887 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
1888 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
1889 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
1891 if (tnl_port_should_receive(&ctx.xin->flow)) {
1892 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
1894 if (xin->ofproto->netflow) {
1895 netflow_mask_wc(flow, wc);
1900 ctx.xout->has_learn = false;
1901 ctx.xout->has_normal = false;
1902 ctx.xout->has_fin_timeout = false;
1903 ctx.xout->nf_output_iface = NF_OUT_DROP;
1904 ctx.xout->mirrors = 0;
1906 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
1907 sizeof ctx.xout->odp_actions_stub);
1908 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
1911 ctx.max_resubmit_trigger = false;
1912 ctx.orig_skb_priority = flow->skb_priority;
1917 ofpacts = xin->ofpacts;
1918 ofpacts_len = xin->ofpacts_len;
1919 } else if (xin->rule) {
1920 ofpacts = xin->rule->up.ofpacts;
1921 ofpacts_len = xin->rule->up.ofpacts_len;
1926 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
1928 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
1929 /* Do this conditionally because the copy is expensive enough that it
1930 * shows up in profiles. */
1934 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
1935 switch (ctx.ofproto->up.frag_handling) {
1936 case OFPC_FRAG_NORMAL:
1937 /* We must pretend that transport ports are unavailable. */
1938 flow->tp_src = ctx.base_flow.tp_src = htons(0);
1939 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
1942 case OFPC_FRAG_DROP:
1945 case OFPC_FRAG_REASM:
1948 case OFPC_FRAG_NX_MATCH:
1949 /* Nothing to do. */
1952 case OFPC_INVALID_TTL_TO_CONTROLLER:
1957 in_port = get_ofp_port(ctx.ofproto, flow->in_port.ofp_port);
1958 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
1960 ctx.xout->slow = special;
1962 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1963 size_t sample_actions_len;
1965 if (flow->in_port.ofp_port
1966 != vsp_realdev_to_vlandev(ctx.ofproto, flow->in_port.ofp_port,
1968 ctx.base_flow.vlan_tci = 0;
1971 add_sflow_action(&ctx);
1972 add_ipfix_action(&ctx);
1973 sample_actions_len = ctx.xout->odp_actions.size;
1975 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
1976 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
1978 /* We've let OFPP_NORMAL and the learning action look at the
1979 * packet, so drop it now if forwarding is disabled. */
1980 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
1981 ctx.xout->odp_actions.size = sample_actions_len;
1985 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
1986 if (!hit_resubmit_limit) {
1987 /* We didn't record the original flow. Make sure we do from
1989 hit_resubmit_limit = true;
1990 } else if (!VLOG_DROP_ERR(&trace_rl)) {
1991 struct ds ds = DS_EMPTY_INITIALIZER;
1993 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, &ds);
1994 VLOG_ERR("Trace triggered by excessive resubmit "
1995 "recursion:\n%s", ds_cstr(&ds));
2000 if (connmgr_has_in_band(ctx.ofproto->up.connmgr)
2001 && in_band_must_output_to_local_port(flow)
2002 && !actions_output_to_local_port(&ctx)) {
2003 compose_output_action(&ctx, OFPP_LOCAL);
2005 if (ctx.ofproto->has_mirrors) {
2006 add_mirror_actions(&ctx, &orig_flow);
2008 fix_sflow_action(&ctx);
2011 ofpbuf_uninit(&ctx.stack);
2013 /* Clear the metadata and register wildcard masks, because we won't
2014 * use non-header fields as part of the cache. */
2015 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2016 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);