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"
22 #include "byte-order.h"
26 #include "dynamic-string.h"
28 #include "mac-learning.h"
29 #include "meta-flow.h"
30 #include "multipath.h"
31 #include "netdev-vport.h"
34 #include "odp-execute.h"
35 #include "ofp-actions.h"
36 #include "ofproto/ofproto-dpif-ipfix.h"
37 #include "ofproto/ofproto-dpif-sflow.h"
38 #include "ofproto/ofproto-dpif.h"
42 COVERAGE_DEFINE(ofproto_dpif_xlate);
44 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
48 struct xlate_out *xout;
50 struct ofproto_dpif *ofproto;
52 /* Flow at the last commit. */
53 struct flow base_flow;
55 /* Tunnel IP destination address as received. This is stored separately
56 * as the base_flow.tunnel is cleared on init to reflect the datapath
57 * behavior. Used to make sure not to send tunneled output to ourselves,
58 * which might lead to an infinite loop. This could happen easily
59 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
60 * actually set the tun_dst field. */
61 ovs_be32 orig_tunnel_ip_dst;
63 /* Stack for the push and pop actions. Each stack element is of type
64 * "union mf_subvalue". */
65 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
68 /* The rule that we are currently translating, or NULL. */
69 struct rule_dpif *rule;
71 int recurse; /* Recursion level, via xlate_table_action. */
72 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
73 uint32_t orig_skb_priority; /* Priority when packet arrived. */
74 uint8_t table_id; /* OpenFlow table ID where flow was found. */
75 uint32_t sflow_n_outputs; /* Number of output ports. */
76 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
77 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
78 bool exit; /* No further actions should be processed. */
81 /* A controller may use OFPP_NONE as the ingress port to indicate that
82 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
83 * when an input bundle is needed for validation (e.g., mirroring or
84 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
85 * any 'port' structs, so care must be taken when dealing with it. */
86 static struct ofbundle ofpp_none_bundle = {
88 .vlan_mode = PORT_VLAN_TRUNK
91 static bool may_receive(const struct ofport_dpif *, struct xlate_ctx *);
92 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
94 static void xlate_normal(struct xlate_ctx *);
95 static void xlate_report(struct xlate_ctx *, const char *);
96 static void xlate_table_action(struct xlate_ctx *, uint16_t in_port,
97 uint8_t table_id, bool may_packet_in);
98 static bool input_vid_is_valid(uint16_t vid, struct ofbundle *, bool warn);
99 static uint16_t input_vid_to_vlan(const struct ofbundle *, uint16_t vid);
100 static void output_normal(struct xlate_ctx *, const struct ofbundle *,
102 static void compose_output_action(struct xlate_ctx *, uint16_t ofp_port);
104 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
107 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
109 return (bundle->vlan_mode != PORT_VLAN_ACCESS
110 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
114 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
116 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
120 vlan_is_mirrored(const struct ofmirror *m, int vlan)
122 return !m->vlans || bitmap_is_set(m->vlans, vlan);
125 static struct ofbundle *
126 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
127 bool warn, struct ofport_dpif **in_ofportp)
129 struct ofport_dpif *ofport;
131 /* Find the port and bundle for the received packet. */
132 ofport = get_ofp_port(ofproto, in_port);
134 *in_ofportp = ofport;
136 if (ofport && ofport->bundle) {
137 return ofport->bundle;
140 /* Special-case OFPP_NONE, which a controller may use as the ingress
141 * port for traffic that it is sourcing. */
142 if (in_port == OFPP_NONE) {
143 return &ofpp_none_bundle;
146 /* Odd. A few possible reasons here:
148 * - We deleted a port but there are still a few packets queued up
151 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
152 * we don't know about.
154 * - The ofproto client didn't configure the port as part of a bundle.
155 * This is particularly likely to happen if a packet was received on the
156 * port after it was created, but before the client had a chance to
157 * configure its bundle.
160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
162 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
163 "port %"PRIu16, ofproto->up.name, in_port);
169 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
171 struct ofproto_dpif *ofproto = ctx->ofproto;
172 mirror_mask_t mirrors;
173 struct ofbundle *in_bundle;
176 const struct nlattr *a;
179 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
180 ctx->xin->packet != NULL, NULL);
184 mirrors = in_bundle->src_mirrors;
186 /* Drop frames on bundles reserved for mirroring. */
187 if (in_bundle->mirror_out) {
188 if (ctx->xin->packet != NULL) {
189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
190 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
191 "%s, which is reserved exclusively for mirroring",
192 ctx->ofproto->up.name, in_bundle->name);
198 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
199 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
202 vlan = input_vid_to_vlan(in_bundle, vid);
204 /* Look at the output ports to check for destination selections. */
206 NL_ATTR_FOR_EACH (a, left, ctx->xout->odp_actions.data,
207 ctx->xout->odp_actions.size) {
208 enum ovs_action_attr type = nl_attr_type(a);
209 struct ofport_dpif *ofport;
211 if (type != OVS_ACTION_ATTR_OUTPUT) {
215 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
216 if (ofport && ofport->bundle) {
217 mirrors |= ofport->bundle->dst_mirrors;
225 /* Restore the original packet before adding the mirror actions. */
226 ctx->xin->flow = *orig_flow;
231 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
234 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
237 if (!vlan_is_mirrored(m, vlan)) {
238 mirrors = zero_rightmost_1bit(mirrors);
242 mirrors &= ~m->dup_mirrors;
243 ctx->xout->mirrors |= m->dup_mirrors;
245 output_normal(ctx, m->out, vlan);
246 } else if (vlan != m->out_vlan
247 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
248 struct ofbundle *bundle;
250 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
251 if (ofbundle_includes_vlan(bundle, m->out_vlan)
252 && !bundle->mirror_out) {
253 output_normal(ctx, bundle, m->out_vlan);
260 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
261 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
262 * the bundle on which the packet was received, returns the VLAN to which the
265 * Both 'vid' and the return value are in the range 0...4095. */
267 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
269 switch (in_bundle->vlan_mode) {
270 case PORT_VLAN_ACCESS:
271 return in_bundle->vlan;
274 case PORT_VLAN_TRUNK:
277 case PORT_VLAN_NATIVE_UNTAGGED:
278 case PORT_VLAN_NATIVE_TAGGED:
279 return vid ? vid : in_bundle->vlan;
286 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
287 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
290 * 'vid' should be the VID obtained from the 802.1Q header that was received as
291 * part of a packet (specify 0 if there was no 802.1Q header), in the range
294 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
296 /* Allow any VID on the OFPP_NONE port. */
297 if (in_bundle == &ofpp_none_bundle) {
301 switch (in_bundle->vlan_mode) {
302 case PORT_VLAN_ACCESS:
305 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
306 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
307 "packet received on port %s configured as VLAN "
308 "%"PRIu16" access port",
309 in_bundle->ofproto->up.name, vid,
310 in_bundle->name, in_bundle->vlan);
316 case PORT_VLAN_NATIVE_UNTAGGED:
317 case PORT_VLAN_NATIVE_TAGGED:
319 /* Port must always carry its native VLAN. */
323 case PORT_VLAN_TRUNK:
324 if (!ofbundle_includes_vlan(in_bundle, vid)) {
326 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
327 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
328 "received on port %s not configured for trunking "
330 in_bundle->ofproto->up.name, vid,
331 in_bundle->name, vid);
343 /* Given 'vlan', the VLAN that a packet belongs to, and
344 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
345 * that should be included in the 802.1Q header. (If the return value is 0,
346 * then the 802.1Q header should only be included in the packet if there is a
349 * Both 'vlan' and the return value are in the range 0...4095. */
351 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
353 switch (out_bundle->vlan_mode) {
354 case PORT_VLAN_ACCESS:
357 case PORT_VLAN_TRUNK:
358 case PORT_VLAN_NATIVE_TAGGED:
361 case PORT_VLAN_NATIVE_UNTAGGED:
362 return vlan == out_bundle->vlan ? 0 : vlan;
370 output_normal(struct xlate_ctx *ctx, const struct ofbundle *out_bundle,
373 struct ofport_dpif *port;
375 ovs_be16 tci, old_tci;
377 vid = output_vlan_to_vid(out_bundle, vlan);
378 if (!out_bundle->bond) {
379 port = ofbundle_get_a_port(out_bundle);
381 port = bond_choose_output_slave(out_bundle->bond, &ctx->xin->flow,
382 &ctx->xout->wc, vid, &ctx->xout->tags);
384 /* No slaves enabled, so drop packet. */
389 old_tci = ctx->xin->flow.vlan_tci;
391 if (tci || out_bundle->use_priority_tags) {
392 tci |= ctx->xin->flow.vlan_tci & htons(VLAN_PCP_MASK);
394 tci |= htons(VLAN_CFI);
397 ctx->xin->flow.vlan_tci = tci;
399 compose_output_action(ctx, port->up.ofp_port);
400 ctx->xin->flow.vlan_tci = old_tci;
403 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
404 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
405 * indicate this; newer upstream kernels use gratuitous ARP requests. */
407 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
409 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
413 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
414 if (!eth_addr_is_broadcast(flow->dl_dst)) {
418 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
419 if (flow->nw_proto == ARP_OP_REPLY) {
421 } else if (flow->nw_proto == ARP_OP_REQUEST) {
422 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
423 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
425 return flow->nw_src == flow->nw_dst;
432 update_learning_table(struct ofproto_dpif *ofproto,
433 const struct flow *flow, struct flow_wildcards *wc,
434 int vlan, struct ofbundle *in_bundle)
436 struct mac_entry *mac;
438 /* Don't learn the OFPP_NONE port. */
439 if (in_bundle == &ofpp_none_bundle) {
443 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
447 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
448 if (is_gratuitous_arp(flow, wc)) {
449 /* We don't want to learn from gratuitous ARP packets that are
450 * reflected back over bond slaves so we lock the learning table. */
451 if (!in_bundle->bond) {
452 mac_entry_set_grat_arp_lock(mac);
453 } else if (mac_entry_is_grat_arp_locked(mac)) {
458 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
459 /* The log messages here could actually be useful in debugging,
460 * so keep the rate limit relatively high. */
461 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
462 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
463 "on port %s in VLAN %d",
464 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
465 in_bundle->name, vlan);
467 mac->port.p = in_bundle;
468 tag_set_add(&ofproto->backer->revalidate_set,
469 mac_learning_changed(ofproto->ml, mac));
473 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
474 * dropped. Returns true if they may be forwarded, false if they should be
477 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
478 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
480 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
481 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
482 * checked by input_vid_is_valid().
484 * May also add tags to '*tags', although the current implementation only does
485 * so in one special case.
488 is_admissible(struct xlate_ctx *ctx, struct ofport_dpif *in_port,
491 struct ofproto_dpif *ofproto = ctx->ofproto;
492 struct flow *flow = &ctx->xin->flow;
493 struct ofbundle *in_bundle = in_port->bundle;
495 /* Drop frames for reserved multicast addresses
496 * only if forward_bpdu option is absent. */
497 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
498 xlate_report(ctx, "packet has reserved destination MAC, dropping");
502 if (in_bundle->bond) {
503 struct mac_entry *mac;
505 switch (bond_check_admissibility(in_bundle->bond, in_port,
506 flow->dl_dst, &ctx->xout->tags)) {
511 xlate_report(ctx, "bonding refused admissibility, dropping");
514 case BV_DROP_IF_MOVED:
515 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
516 if (mac && mac->port.p != in_bundle &&
517 (!is_gratuitous_arp(flow, &ctx->xout->wc)
518 || mac_entry_is_grat_arp_locked(mac))) {
519 xlate_report(ctx, "SLB bond thinks this packet looped back, "
531 xlate_normal(struct xlate_ctx *ctx)
533 struct ofport_dpif *in_port;
534 struct ofbundle *in_bundle;
535 struct mac_entry *mac;
539 ctx->xout->has_normal = true;
541 /* Check the dl_type, since we may check for gratuituous ARP. */
542 memset(&ctx->xout->wc.masks.dl_type, 0xff,
543 sizeof ctx->xout->wc.masks.dl_type);
545 memset(&ctx->xout->wc.masks.dl_src, 0xff,
546 sizeof ctx->xout->wc.masks.dl_src);
547 memset(&ctx->xout->wc.masks.dl_dst, 0xff,
548 sizeof ctx->xout->wc.masks.dl_dst);
549 memset(&ctx->xout->wc.masks.vlan_tci, 0xff,
550 sizeof ctx->xout->wc.masks.vlan_tci);
552 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->xin->flow.in_port,
553 ctx->xin->packet != NULL, &in_port);
555 xlate_report(ctx, "no input bundle, dropping");
559 /* Drop malformed frames. */
560 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_VLAN) &&
561 !(ctx->xin->flow.vlan_tci & htons(VLAN_CFI))) {
562 if (ctx->xin->packet != NULL) {
563 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
564 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
565 "VLAN tag received on port %s",
566 ctx->ofproto->up.name, in_bundle->name);
568 xlate_report(ctx, "partial VLAN tag, dropping");
572 /* Drop frames on bundles reserved for mirroring. */
573 if (in_bundle->mirror_out) {
574 if (ctx->xin->packet != NULL) {
575 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
576 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
577 "%s, which is reserved exclusively for mirroring",
578 ctx->ofproto->up.name, in_bundle->name);
580 xlate_report(ctx, "input port is mirror output port, dropping");
585 vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
586 if (!input_vid_is_valid(vid, in_bundle, ctx->xin->packet != NULL)) {
587 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
590 vlan = input_vid_to_vlan(in_bundle, vid);
592 /* Check other admissibility requirements. */
593 if (in_port && !is_admissible(ctx, in_port, vlan)) {
597 /* Learn source MAC. */
598 if (ctx->xin->may_learn) {
599 update_learning_table(ctx->ofproto, &ctx->xin->flow, &ctx->xout->wc,
603 /* Determine output bundle. */
604 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->xin->flow.dl_dst, vlan,
607 if (mac->port.p != in_bundle) {
608 xlate_report(ctx, "forwarding to learned port");
609 output_normal(ctx, mac->port.p, vlan);
611 xlate_report(ctx, "learned port is input port, dropping");
614 struct ofbundle *bundle;
616 xlate_report(ctx, "no learned MAC for destination, flooding");
617 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
618 if (bundle != in_bundle
619 && ofbundle_includes_vlan(bundle, vlan)
621 && !bundle->mirror_out) {
622 output_normal(ctx, bundle, vlan);
625 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
629 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
630 * the number of packets out of UINT32_MAX to sample. The given
631 * cookie is passed back in the callback for each sampled packet.
634 compose_sample_action(const struct ofproto_dpif *ofproto,
635 struct ofpbuf *odp_actions,
636 const struct flow *flow,
637 const uint32_t probability,
638 const union user_action_cookie *cookie,
639 const size_t cookie_size)
641 size_t sample_offset, actions_offset;
644 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
646 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
648 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
649 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
652 nl_msg_end_nested(odp_actions, actions_offset);
653 nl_msg_end_nested(odp_actions, sample_offset);
654 return cookie_offset;
658 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
659 ovs_be16 vlan_tci, uint32_t odp_port,
660 unsigned int n_outputs, union user_action_cookie *cookie)
664 cookie->type = USER_ACTION_COOKIE_SFLOW;
665 cookie->sflow.vlan_tci = vlan_tci;
667 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
668 * port information") for the interpretation of cookie->output. */
671 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
672 cookie->sflow.output = 0x40000000 | 256;
676 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
678 cookie->sflow.output = ifindex;
683 /* 0x80000000 means "multiple output ports. */
684 cookie->sflow.output = 0x80000000 | n_outputs;
689 /* Compose SAMPLE action for sFlow bridge sampling. */
691 compose_sflow_action(const struct ofproto_dpif *ofproto,
692 struct ofpbuf *odp_actions,
693 const struct flow *flow,
696 uint32_t probability;
697 union user_action_cookie cookie;
699 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
703 probability = dpif_sflow_get_probability(ofproto->sflow);
704 compose_sflow_cookie(ofproto, htons(0), odp_port,
705 odp_port == OVSP_NONE ? 0 : 1, &cookie);
707 return compose_sample_action(ofproto, odp_actions, flow, probability,
708 &cookie, sizeof cookie.sflow);
712 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
713 uint32_t obs_domain_id, uint32_t obs_point_id,
714 union user_action_cookie *cookie)
716 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
717 cookie->flow_sample.probability = probability;
718 cookie->flow_sample.collector_set_id = collector_set_id;
719 cookie->flow_sample.obs_domain_id = obs_domain_id;
720 cookie->flow_sample.obs_point_id = obs_point_id;
724 compose_ipfix_cookie(union user_action_cookie *cookie)
726 cookie->type = USER_ACTION_COOKIE_IPFIX;
729 /* Compose SAMPLE action for IPFIX bridge sampling. */
731 compose_ipfix_action(const struct ofproto_dpif *ofproto,
732 struct ofpbuf *odp_actions,
733 const struct flow *flow)
735 uint32_t probability;
736 union user_action_cookie cookie;
738 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
742 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
743 compose_ipfix_cookie(&cookie);
745 compose_sample_action(ofproto, odp_actions, flow, probability,
746 &cookie, sizeof cookie.ipfix);
749 /* SAMPLE action for sFlow must be first action in any given list of
750 * actions. At this point we do not have all information required to
751 * build it. So try to build sample action as complete as possible. */
753 add_sflow_action(struct xlate_ctx *ctx)
755 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
756 &ctx->xout->odp_actions,
757 &ctx->xin->flow, OVSP_NONE);
758 ctx->sflow_odp_port = 0;
759 ctx->sflow_n_outputs = 0;
762 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
763 * of actions, eventually after the SAMPLE action for sFlow. */
765 add_ipfix_action(struct xlate_ctx *ctx)
767 compose_ipfix_action(ctx->ofproto, &ctx->xout->odp_actions,
771 /* Fix SAMPLE action according to data collected while composing ODP actions.
772 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
773 * USERSPACE action's user-cookie which is required for sflow. */
775 fix_sflow_action(struct xlate_ctx *ctx)
777 const struct flow *base = &ctx->base_flow;
778 union user_action_cookie *cookie;
780 if (!ctx->user_cookie_offset) {
784 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
785 sizeof cookie->sflow);
786 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
788 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
789 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
793 compose_output_action__(struct xlate_ctx *ctx, uint16_t ofp_port,
796 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
797 ovs_be16 flow_vlan_tci;
798 uint32_t flow_skb_mark;
800 struct priority_to_dscp *pdscp;
801 uint32_t out_port, odp_port;
803 /* If 'struct flow' gets additional metadata, we'll need to zero it out
804 * before traversing a patch port. */
805 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
808 xlate_report(ctx, "Nonexistent output port");
810 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
811 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
813 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
814 xlate_report(ctx, "STP not in forwarding state, skipping output");
818 if (netdev_vport_is_patch(ofport->up.netdev)) {
819 struct ofport_dpif *peer = ofport_get_peer(ofport);
820 struct flow old_flow = ctx->xin->flow;
821 const struct ofproto_dpif *peer_ofproto;
822 enum slow_path_reason special;
823 struct ofport_dpif *in_port;
826 xlate_report(ctx, "Nonexistent patch port peer");
830 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
831 if (peer_ofproto->backer != ctx->ofproto->backer) {
832 xlate_report(ctx, "Patch port peer on a different datapath");
836 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
837 ctx->xin->flow.in_port = peer->up.ofp_port;
838 ctx->xin->flow.metadata = htonll(0);
839 memset(&ctx->xin->flow.tunnel, 0, sizeof ctx->xin->flow.tunnel);
840 memset(ctx->xin->flow.regs, 0, sizeof ctx->xin->flow.regs);
842 in_port = get_ofp_port(ctx->ofproto, ctx->xin->flow.in_port);
843 special = process_special(ctx->ofproto, &ctx->xin->flow, in_port,
846 ctx->xout->slow = special;
847 } else if (!in_port || may_receive(in_port, ctx)) {
848 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
849 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
851 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
852 * learning action look at the packet, then drop it. */
853 struct flow old_base_flow = ctx->base_flow;
854 size_t old_size = ctx->xout->odp_actions.size;
855 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, true);
856 ctx->base_flow = old_base_flow;
857 ctx->xout->odp_actions.size = old_size;
861 ctx->xin->flow = old_flow;
862 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
864 if (ctx->xin->resubmit_stats) {
865 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
866 netdev_vport_inc_rx(peer->up.netdev, ctx->xin->resubmit_stats);
872 flow_vlan_tci = ctx->xin->flow.vlan_tci;
873 flow_skb_mark = ctx->xin->flow.skb_mark;
874 flow_nw_tos = ctx->xin->flow.nw_tos;
876 pdscp = get_priority(ofport, ctx->xin->flow.skb_priority);
878 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
879 ctx->xin->flow.nw_tos |= pdscp->dscp;
882 if (ofport->tnl_port) {
883 /* Save tunnel metadata so that changes made due to
884 * the Logical (tunnel) Port are not visible for any further
885 * matches, while explicit set actions on tunnel metadata are.
887 struct flow_tnl flow_tnl = ctx->xin->flow.tunnel;
888 odp_port = tnl_port_send(ofport->tnl_port, &ctx->xin->flow);
889 if (odp_port == OVSP_NONE) {
890 xlate_report(ctx, "Tunneling decided against output");
891 goto out; /* restore flow_nw_tos */
893 if (ctx->xin->flow.tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
894 xlate_report(ctx, "Not tunneling to our own address");
895 goto out; /* restore flow_nw_tos */
897 if (ctx->xin->resubmit_stats) {
898 netdev_vport_inc_tx(ofport->up.netdev, ctx->xin->resubmit_stats);
901 commit_odp_tunnel_action(&ctx->xin->flow, &ctx->base_flow,
902 &ctx->xout->odp_actions);
903 ctx->xin->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
905 uint16_t vlandev_port;
906 odp_port = ofport->odp_port;
907 vlandev_port = vsp_realdev_to_vlandev(ctx->ofproto, ofp_port,
908 ctx->xin->flow.vlan_tci);
909 if (vlandev_port == ofp_port) {
912 out_port = ofp_port_to_odp_port(ctx->ofproto, vlandev_port);
913 ctx->xin->flow.vlan_tci = htons(0);
915 ctx->xin->flow.skb_mark &= ~IPSEC_MARK;
917 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
918 &ctx->xout->odp_actions);
919 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
921 ctx->sflow_odp_port = odp_port;
922 ctx->sflow_n_outputs++;
923 ctx->xout->nf_output_iface = ofp_port;
926 ctx->xin->flow.vlan_tci = flow_vlan_tci;
927 ctx->xin->flow.skb_mark = flow_skb_mark;
929 ctx->xin->flow.nw_tos = flow_nw_tos;
933 compose_output_action(struct xlate_ctx *ctx, uint16_t ofp_port)
935 compose_output_action__(ctx, ofp_port, true);
939 tag_the_flow(struct xlate_ctx *ctx, struct rule_dpif *rule)
941 struct ofproto_dpif *ofproto = ctx->ofproto;
942 uint8_t table_id = ctx->table_id;
944 if (table_id > 0 && table_id < N_TABLES) {
945 struct table_dpif *table = &ofproto->tables[table_id];
946 if (table->other_table) {
947 ctx->xout->tags |= (rule && rule->tag
949 : rule_calculate_tag(&ctx->xin->flow,
950 &table->other_table->mask,
956 /* Common rule processing in one place to avoid duplicating code. */
957 static struct rule_dpif *
958 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
961 if (ctx->xin->resubmit_hook) {
962 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
964 if (rule == NULL && may_packet_in) {
966 * check if table configuration flags
967 * OFPTC_TABLE_MISS_CONTROLLER, default.
968 * OFPTC_TABLE_MISS_CONTINUE,
969 * OFPTC_TABLE_MISS_DROP
970 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
972 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->xin->flow);
974 if (rule && ctx->xin->resubmit_stats) {
975 rule_credit_stats(rule, ctx->xin->resubmit_stats);
981 xlate_table_action(struct xlate_ctx *ctx,
982 uint16_t in_port, uint8_t table_id, bool may_packet_in)
984 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
985 struct rule_dpif *rule;
986 uint16_t old_in_port = ctx->xin->flow.in_port;
987 uint8_t old_table_id = ctx->table_id;
989 ctx->table_id = table_id;
991 /* Look up a flow with 'in_port' as the input port. */
992 ctx->xin->flow.in_port = in_port;
993 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
994 &ctx->xout->wc, table_id);
996 tag_the_flow(ctx, rule);
998 /* Restore the original input port. Otherwise OFPP_NORMAL and
999 * OFPP_IN_PORT will have surprising behavior. */
1000 ctx->xin->flow.in_port = old_in_port;
1002 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1005 struct rule_dpif *old_rule = ctx->rule;
1009 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1010 ctx->rule = old_rule;
1014 ctx->table_id = old_table_id;
1016 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1018 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1019 MAX_RESUBMIT_RECURSION);
1020 ctx->max_resubmit_trigger = true;
1025 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1026 const struct ofpact_resubmit *resubmit)
1031 in_port = resubmit->in_port;
1032 if (in_port == OFPP_IN_PORT) {
1033 in_port = ctx->xin->flow.in_port;
1036 table_id = resubmit->table_id;
1037 if (table_id == 255) {
1038 table_id = ctx->table_id;
1041 xlate_table_action(ctx, in_port, table_id, false);
1045 flood_packets(struct xlate_ctx *ctx, bool all)
1047 struct ofport_dpif *ofport;
1049 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
1050 uint16_t ofp_port = ofport->up.ofp_port;
1052 if (ofp_port == ctx->xin->flow.in_port) {
1057 compose_output_action__(ctx, ofp_port, false);
1058 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
1059 compose_output_action(ctx, ofp_port);
1063 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1067 execute_controller_action(struct xlate_ctx *ctx, int len,
1068 enum ofp_packet_in_reason reason,
1069 uint16_t controller_id)
1071 struct ofputil_packet_in pin;
1072 struct ofpbuf *packet;
1075 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1076 ctx->xout->slow = SLOW_CONTROLLER;
1077 if (!ctx->xin->packet) {
1081 packet = ofpbuf_clone(ctx->xin->packet);
1083 key.skb_priority = 0;
1085 memset(&key.tunnel, 0, sizeof key.tunnel);
1087 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1088 &ctx->xout->odp_actions);
1090 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1091 ctx->xout->odp_actions.size, NULL, NULL);
1093 pin.packet = packet->data;
1094 pin.packet_len = packet->size;
1095 pin.reason = reason;
1096 pin.controller_id = controller_id;
1097 pin.table_id = ctx->table_id;
1098 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1101 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1103 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
1104 ofpbuf_delete(packet);
1108 execute_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1110 ovs_assert(eth_type_mpls(eth_type));
1112 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1113 sizeof ctx->xout->wc.masks.dl_type);
1114 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
1115 sizeof ctx->xout->wc.masks.mpls_lse);
1116 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
1117 sizeof ctx->xout->wc.masks.mpls_depth);
1119 if (ctx->base_flow.mpls_depth) {
1120 ctx->xin->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
1121 ctx->xin->flow.mpls_depth++;
1126 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IPV6)) {
1127 label = htonl(0x2); /* IPV6 Explicit Null. */
1129 label = htonl(0x0); /* IPV4 Explicit Null. */
1131 tc = (ctx->xin->flow.nw_tos & IP_DSCP_MASK) >> 2;
1132 ttl = ctx->xin->flow.nw_ttl ? ctx->xin->flow.nw_ttl : 0x40;
1133 ctx->xin->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1134 ctx->xin->flow.mpls_depth = 1;
1136 ctx->xin->flow.dl_type = eth_type;
1140 execute_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1142 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1143 ovs_assert(!eth_type_mpls(eth_type));
1145 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1146 sizeof ctx->xout->wc.masks.dl_type);
1147 memset(&ctx->xout->wc.masks.mpls_lse, 0xff,
1148 sizeof ctx->xout->wc.masks.mpls_lse);
1149 memset(&ctx->xout->wc.masks.mpls_depth, 0xff,
1150 sizeof ctx->xout->wc.masks.mpls_depth);
1152 if (ctx->xin->flow.mpls_depth) {
1153 ctx->xin->flow.mpls_depth--;
1154 ctx->xin->flow.mpls_lse = htonl(0);
1155 if (!ctx->xin->flow.mpls_depth) {
1156 ctx->xin->flow.dl_type = eth_type;
1162 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1164 if (ctx->xin->flow.dl_type != htons(ETH_TYPE_IP) &&
1165 ctx->xin->flow.dl_type != htons(ETH_TYPE_IPV6)) {
1169 if (ctx->xin->flow.nw_ttl > 1) {
1170 ctx->xin->flow.nw_ttl--;
1175 for (i = 0; i < ids->n_controllers; i++) {
1176 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1180 /* Stop processing for current table. */
1186 execute_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1188 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1192 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1197 execute_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1199 uint8_t ttl = mpls_lse_to_ttl(ctx->xin->flow.mpls_lse);
1201 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1207 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1210 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1212 /* Stop processing for current table. */
1218 xlate_output_action(struct xlate_ctx *ctx,
1219 uint16_t port, uint16_t max_len, bool may_packet_in)
1221 uint16_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1223 ctx->xout->nf_output_iface = NF_OUT_DROP;
1227 compose_output_action(ctx, ctx->xin->flow.in_port);
1230 xlate_table_action(ctx, ctx->xin->flow.in_port, 0, may_packet_in);
1236 flood_packets(ctx, false);
1239 flood_packets(ctx, true);
1241 case OFPP_CONTROLLER:
1242 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1248 if (port != ctx->xin->flow.in_port) {
1249 compose_output_action(ctx, port);
1251 xlate_report(ctx, "skipping output to input port");
1256 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1257 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1258 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1259 ctx->xout->nf_output_iface = prev_nf_output_iface;
1260 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1261 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1262 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1267 xlate_output_reg_action(struct xlate_ctx *ctx,
1268 const struct ofpact_output_reg *or)
1270 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1271 if (port <= UINT16_MAX) {
1272 union mf_subvalue value;
1274 memset(&value, 0xff, sizeof value);
1275 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1276 xlate_output_action(ctx, port, or->max_len, false);
1281 xlate_enqueue_action(struct xlate_ctx *ctx,
1282 const struct ofpact_enqueue *enqueue)
1284 uint16_t ofp_port = enqueue->port;
1285 uint32_t queue_id = enqueue->queue;
1286 uint32_t flow_priority, priority;
1289 /* Translate queue to priority. */
1290 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
1291 queue_id, &priority);
1293 /* Fall back to ordinary output action. */
1294 xlate_output_action(ctx, enqueue->port, 0, false);
1298 /* Check output port. */
1299 if (ofp_port == OFPP_IN_PORT) {
1300 ofp_port = ctx->xin->flow.in_port;
1301 } else if (ofp_port == ctx->xin->flow.in_port) {
1305 /* Add datapath actions. */
1306 flow_priority = ctx->xin->flow.skb_priority;
1307 ctx->xin->flow.skb_priority = priority;
1308 compose_output_action(ctx, ofp_port);
1309 ctx->xin->flow.skb_priority = flow_priority;
1311 /* Update NetFlow output port. */
1312 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1313 ctx->xout->nf_output_iface = ofp_port;
1314 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1315 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1320 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1322 uint32_t skb_priority;
1324 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
1325 queue_id, &skb_priority)) {
1326 ctx->xin->flow.skb_priority = skb_priority;
1328 /* Couldn't translate queue to a priority. Nothing to do. A warning
1329 * has already been logged. */
1334 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
1336 struct ofproto_dpif *ofproto = ofproto_;
1337 struct ofport_dpif *port;
1347 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1350 port = get_ofp_port(ofproto, ofp_port);
1351 return port ? port->may_enable : false;
1356 xlate_bundle_action(struct xlate_ctx *ctx,
1357 const struct ofpact_bundle *bundle)
1361 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1362 slave_enabled_cb, ctx->ofproto);
1363 if (bundle->dst.field) {
1364 nxm_reg_load(&bundle->dst, port, &ctx->xin->flow);
1366 xlate_output_action(ctx, port, 0, false);
1371 xlate_learn_action(struct xlate_ctx *ctx,
1372 const struct ofpact_learn *learn)
1374 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1375 struct ofputil_flow_mod fm;
1376 uint64_t ofpacts_stub[1024 / 8];
1377 struct ofpbuf ofpacts;
1380 ctx->xout->has_learn = true;
1382 learn_mask(learn, &ctx->xout->wc);
1384 if (!ctx->xin->may_learn) {
1388 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1389 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1391 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
1392 if (error && !VLOG_DROP_WARN(&rl)) {
1393 VLOG_WARN("learning action failed to modify flow table (%s)",
1394 ofperr_get_name(error));
1397 ofpbuf_uninit(&ofpacts);
1400 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1401 * means "infinite". */
1403 reduce_timeout(uint16_t max, uint16_t *timeout)
1405 if (max && (!*timeout || *timeout > max)) {
1411 xlate_fin_timeout(struct xlate_ctx *ctx,
1412 const struct ofpact_fin_timeout *oft)
1414 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1415 struct rule_dpif *rule = ctx->rule;
1417 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1418 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1423 xlate_sample_action(struct xlate_ctx *ctx,
1424 const struct ofpact_sample *os)
1426 union user_action_cookie cookie;
1427 /* Scale the probability from 16-bit to 32-bit while representing
1428 * the same percentage. */
1429 uint32_t probability = (os->probability << 16) | os->probability;
1431 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1432 &ctx->xout->odp_actions);
1434 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1435 os->obs_domain_id, os->obs_point_id, &cookie);
1436 compose_sample_action(ctx->ofproto, &ctx->xout->odp_actions, &ctx->xin->flow,
1437 probability, &cookie, sizeof cookie.flow_sample);
1441 may_receive(const struct ofport_dpif *port, struct xlate_ctx *ctx)
1443 if (port->up.pp.config & (eth_addr_equals(ctx->xin->flow.dl_dst,
1445 ? OFPUTIL_PC_NO_RECV_STP
1446 : OFPUTIL_PC_NO_RECV)) {
1450 /* Only drop packets here if both forwarding and learning are
1451 * disabled. If just learning is enabled, we need to have
1452 * OFPP_NORMAL and the learning action have a look at the packet
1453 * before we can drop it. */
1454 if (!stp_forward_in_state(port->stp_state)
1455 && !stp_learn_in_state(port->stp_state)) {
1463 tunnel_ecn_ok(struct xlate_ctx *ctx)
1465 if (is_ip_any(&ctx->base_flow)
1466 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1467 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1468 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1469 " but is not ECN capable");
1472 /* Set the ECN CE value in the tunneled packet. */
1473 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1481 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1482 struct xlate_ctx *ctx)
1484 bool was_evictable = true;
1485 const struct ofpact *a;
1488 /* Don't let the rule we're working on get evicted underneath us. */
1489 was_evictable = ctx->rule->up.evictable;
1490 ctx->rule->up.evictable = false;
1493 do_xlate_actions_again:
1494 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1495 struct ofpact_controller *controller;
1496 const struct ofpact_metadata *metadata;
1504 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1505 ofpact_get_OUTPUT(a)->max_len, true);
1508 case OFPACT_CONTROLLER:
1509 controller = ofpact_get_CONTROLLER(a);
1510 execute_controller_action(ctx, controller->max_len,
1512 controller->controller_id);
1515 case OFPACT_ENQUEUE:
1516 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1519 case OFPACT_SET_VLAN_VID:
1520 ctx->xin->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
1521 ctx->xin->flow.vlan_tci |=
1522 (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1526 case OFPACT_SET_VLAN_PCP:
1527 ctx->xin->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
1528 ctx->xin->flow.vlan_tci |=
1529 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1533 case OFPACT_STRIP_VLAN:
1534 ctx->xin->flow.vlan_tci = htons(0);
1537 case OFPACT_PUSH_VLAN:
1538 /* XXX 802.1AD(QinQ) */
1539 ctx->xin->flow.vlan_tci = htons(VLAN_CFI);
1542 case OFPACT_SET_ETH_SRC:
1543 memcpy(ctx->xin->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
1547 case OFPACT_SET_ETH_DST:
1548 memcpy(ctx->xin->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
1552 case OFPACT_SET_IPV4_SRC:
1553 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1554 sizeof ctx->xout->wc.masks.dl_type);
1555 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
1556 ctx->xin->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1560 case OFPACT_SET_IPV4_DST:
1561 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1562 sizeof ctx->xout->wc.masks.dl_type);
1563 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
1564 ctx->xin->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1568 case OFPACT_SET_IPV4_DSCP:
1569 /* OpenFlow 1.0 only supports IPv4. */
1570 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1571 sizeof ctx->xout->wc.masks.dl_type);
1572 if (ctx->xin->flow.dl_type == htons(ETH_TYPE_IP)) {
1573 ctx->xin->flow.nw_tos &= ~IP_DSCP_MASK;
1574 ctx->xin->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1578 case OFPACT_SET_L4_SRC_PORT:
1579 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1580 sizeof ctx->xout->wc.masks.dl_type);
1581 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
1582 sizeof ctx->xout->wc.masks.nw_proto);
1583 if (is_ip_any(&ctx->xin->flow)) {
1584 ctx->xin->flow.tp_src =
1585 htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1589 case OFPACT_SET_L4_DST_PORT:
1590 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1591 sizeof ctx->xout->wc.masks.dl_type);
1592 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
1593 sizeof ctx->xout->wc.masks.nw_proto);
1594 if (is_ip_any(&ctx->xin->flow)) {
1595 ctx->xin->flow.tp_dst =
1596 htons(ofpact_get_SET_L4_DST_PORT(a)->port);
1600 case OFPACT_RESUBMIT:
1601 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
1604 case OFPACT_SET_TUNNEL:
1605 ctx->xin->flow.tunnel.tun_id =
1606 htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
1609 case OFPACT_SET_QUEUE:
1610 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
1613 case OFPACT_POP_QUEUE:
1614 memset(&ctx->xout->wc.masks.skb_priority, 0xff,
1615 sizeof ctx->xout->wc.masks.skb_priority);
1617 ctx->xin->flow.skb_priority = ctx->orig_skb_priority;
1620 case OFPACT_REG_MOVE:
1621 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->xin->flow,
1625 case OFPACT_REG_LOAD:
1626 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->xin->flow);
1629 case OFPACT_STACK_PUSH:
1630 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->xin->flow,
1631 &ctx->xout->wc, &ctx->stack);
1634 case OFPACT_STACK_POP:
1635 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->xin->flow,
1639 case OFPACT_PUSH_MPLS:
1640 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
1643 case OFPACT_POP_MPLS:
1644 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
1647 case OFPACT_SET_MPLS_TTL:
1648 if (execute_set_mpls_ttl_action(ctx,
1649 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
1654 case OFPACT_DEC_MPLS_TTL:
1655 if (execute_dec_mpls_ttl_action(ctx)) {
1660 case OFPACT_DEC_TTL:
1661 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1662 sizeof ctx->xout->wc.masks.dl_type);
1663 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
1669 /* Nothing to do. */
1672 case OFPACT_MULTIPATH:
1673 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->xin->flow,
1678 ctx->ofproto->has_bundle_action = true;
1679 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
1682 case OFPACT_OUTPUT_REG:
1683 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
1687 xlate_learn_action(ctx, ofpact_get_LEARN(a));
1694 case OFPACT_FIN_TIMEOUT:
1695 memset(&ctx->xout->wc.masks.dl_type, 0xff,
1696 sizeof ctx->xout->wc.masks.dl_type);
1697 memset(&ctx->xout->wc.masks.nw_proto, 0xff,
1698 sizeof ctx->xout->wc.masks.nw_proto);
1699 ctx->xout->has_fin_timeout = true;
1700 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
1703 case OFPACT_CLEAR_ACTIONS:
1705 * Nothing to do because writa-actions is not supported for now.
1706 * When writa-actions is supported, clear-actions also must
1707 * be supported at the same time.
1711 case OFPACT_WRITE_METADATA:
1712 metadata = ofpact_get_WRITE_METADATA(a);
1713 ctx->xin->flow.metadata &= ~metadata->mask;
1714 ctx->xin->flow.metadata |= metadata->metadata & metadata->mask;
1717 case OFPACT_GOTO_TABLE: {
1718 /* It is assumed that goto-table is the last action. */
1719 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
1720 struct rule_dpif *rule;
1722 ovs_assert(ctx->table_id < ogt->table_id);
1724 ctx->table_id = ogt->table_id;
1726 /* Look up a flow from the new table. */
1727 rule = rule_dpif_lookup_in_table(ctx->ofproto, &ctx->xin->flow,
1728 &ctx->xout->wc, ctx->table_id);
1730 tag_the_flow(ctx, rule);
1732 rule = ctx_rule_hooks(ctx, rule, true);
1736 ctx->rule->up.evictable = was_evictable;
1739 was_evictable = rule->up.evictable;
1740 rule->up.evictable = false;
1742 /* Tail recursion removal. */
1743 ofpacts = rule->up.ofpacts;
1744 ofpacts_len = rule->up.ofpacts_len;
1745 goto do_xlate_actions_again;
1751 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
1758 ctx->rule->up.evictable = was_evictable;
1763 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
1764 const struct flow *flow, struct rule_dpif *rule,
1765 uint8_t tcp_flags, const struct ofpbuf *packet)
1767 xin->ofproto = ofproto;
1769 xin->packet = packet;
1770 xin->may_learn = packet != NULL;
1772 xin->ofpacts = NULL;
1773 xin->ofpacts_len = 0;
1774 xin->tcp_flags = tcp_flags;
1775 xin->resubmit_hook = NULL;
1776 xin->report_hook = NULL;
1777 xin->resubmit_stats = NULL;
1781 xlate_out_uninit(struct xlate_out *xout)
1784 ofpbuf_uninit(&xout->odp_actions);
1788 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
1789 * into datapath actions, using 'ctx', and discards the datapath actions. */
1791 xlate_actions_for_side_effects(struct xlate_in *xin)
1793 struct xlate_out xout;
1795 xlate_actions(xin, &xout);
1796 xlate_out_uninit(&xout);
1800 xlate_report(struct xlate_ctx *ctx, const char *s)
1802 if (ctx->xin->report_hook) {
1803 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
1808 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
1811 dst->tags = src->tags;
1812 dst->slow = src->slow;
1813 dst->has_learn = src->has_learn;
1814 dst->has_normal = src->has_normal;
1815 dst->has_fin_timeout = src->has_fin_timeout;
1816 dst->nf_output_iface = src->nf_output_iface;
1817 dst->mirrors = src->mirrors;
1819 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
1820 sizeof dst->odp_actions_stub);
1821 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
1822 src->odp_actions.size);
1826 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
1827 * into datapath actions in 'odp_actions', using 'ctx'. */
1829 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
1831 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
1832 * that in the future we always keep a copy of the original flow for
1833 * tracing purposes. */
1834 static bool hit_resubmit_limit;
1836 enum slow_path_reason special;
1837 const struct ofpact *ofpacts;
1838 struct ofport_dpif *in_port;
1839 struct flow orig_flow;
1840 struct xlate_ctx ctx;
1843 COVERAGE_INC(ofproto_dpif_xlate);
1845 /* Flow initialization rules:
1846 * - 'base_flow' must match the kernel's view of the packet at the
1847 * time that action processing starts. 'flow' represents any
1848 * transformations we wish to make through actions.
1849 * - By default 'base_flow' and 'flow' are the same since the input
1850 * packet matches the output before any actions are applied.
1851 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
1852 * of the received packet as seen by the kernel. If we later output
1853 * to another device without any modifications this will cause us to
1854 * insert a new tag since the original one was stripped off by the
1856 * - Tunnel metadata as received is retained in 'flow'. This allows
1857 * tunnel metadata matching also in later tables.
1858 * Since a kernel action for setting the tunnel metadata will only be
1859 * generated with actual tunnel output, changing the tunnel metadata
1860 * values in 'flow' (such as tun_id) will only have effect with a later
1861 * tunnel output action.
1862 * - Tunnel 'base_flow' is completely cleared since that is what the
1863 * kernel does. If we wish to maintain the original values an action
1864 * needs to be generated. */
1869 ctx.ofproto = xin->ofproto;
1870 ctx.rule = xin->rule;
1872 ctx.base_flow = ctx.xin->flow;
1873 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
1874 ctx.orig_tunnel_ip_dst = ctx.xin->flow.tunnel.ip_dst;
1876 flow_wildcards_init_catchall(&ctx.xout->wc);
1877 memset(&ctx.xout->wc.masks.in_port, 0xff,
1878 sizeof ctx.xout->wc.masks.in_port);
1880 if (tnl_port_should_receive(&ctx.xin->flow)) {
1881 memset(&ctx.xout->wc.masks.tunnel, 0xff,
1882 sizeof ctx.xout->wc.masks.tunnel);
1885 /* Disable most wildcarding for NetFlow. */
1886 if (xin->ofproto->netflow) {
1887 memset(&ctx.xout->wc.masks.dl_src, 0xff,
1888 sizeof ctx.xout->wc.masks.dl_src);
1889 memset(&ctx.xout->wc.masks.dl_dst, 0xff,
1890 sizeof ctx.xout->wc.masks.dl_dst);
1891 memset(&ctx.xout->wc.masks.dl_type, 0xff,
1892 sizeof ctx.xout->wc.masks.dl_type);
1893 memset(&ctx.xout->wc.masks.vlan_tci, 0xff,
1894 sizeof ctx.xout->wc.masks.vlan_tci);
1895 memset(&ctx.xout->wc.masks.nw_proto, 0xff,
1896 sizeof ctx.xout->wc.masks.nw_proto);
1897 memset(&ctx.xout->wc.masks.nw_src, 0xff,
1898 sizeof ctx.xout->wc.masks.nw_src);
1899 memset(&ctx.xout->wc.masks.nw_dst, 0xff,
1900 sizeof ctx.xout->wc.masks.nw_dst);
1901 memset(&ctx.xout->wc.masks.tp_src, 0xff,
1902 sizeof ctx.xout->wc.masks.tp_src);
1903 memset(&ctx.xout->wc.masks.tp_dst, 0xff,
1904 sizeof ctx.xout->wc.masks.tp_dst);
1909 ctx.xout->has_learn = false;
1910 ctx.xout->has_normal = false;
1911 ctx.xout->has_fin_timeout = false;
1912 ctx.xout->nf_output_iface = NF_OUT_DROP;
1913 ctx.xout->mirrors = 0;
1915 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
1916 sizeof ctx.xout->odp_actions_stub);
1917 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
1920 ctx.max_resubmit_trigger = false;
1921 ctx.orig_skb_priority = ctx.xin->flow.skb_priority;
1926 ofpacts = xin->ofpacts;
1927 ofpacts_len = xin->ofpacts_len;
1928 } else if (xin->rule) {
1929 ofpacts = xin->rule->up.ofpacts;
1930 ofpacts_len = xin->rule->up.ofpacts_len;
1935 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
1937 if (ctx.ofproto->has_mirrors || hit_resubmit_limit) {
1938 /* Do this conditionally because the copy is expensive enough that it
1939 * shows up in profiles. */
1940 orig_flow = ctx.xin->flow;
1943 if (ctx.xin->flow.nw_frag & FLOW_NW_FRAG_ANY) {
1944 switch (ctx.ofproto->up.frag_handling) {
1945 case OFPC_FRAG_NORMAL:
1946 /* We must pretend that transport ports are unavailable. */
1947 ctx.xin->flow.tp_src = ctx.base_flow.tp_src = htons(0);
1948 ctx.xin->flow.tp_dst = ctx.base_flow.tp_dst = htons(0);
1951 case OFPC_FRAG_DROP:
1954 case OFPC_FRAG_REASM:
1957 case OFPC_FRAG_NX_MATCH:
1958 /* Nothing to do. */
1961 case OFPC_INVALID_TTL_TO_CONTROLLER:
1966 in_port = get_ofp_port(ctx.ofproto, ctx.xin->flow.in_port);
1967 special = process_special(ctx.ofproto, &ctx.xin->flow, in_port,
1970 ctx.xout->slow = special;
1972 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1973 size_t sample_actions_len;
1974 uint32_t local_odp_port;
1976 if (ctx.xin->flow.in_port
1977 != vsp_realdev_to_vlandev(ctx.ofproto, ctx.xin->flow.in_port,
1978 ctx.xin->flow.vlan_tci)) {
1979 ctx.base_flow.vlan_tci = 0;
1982 add_sflow_action(&ctx);
1983 add_ipfix_action(&ctx);
1984 sample_actions_len = ctx.xout->odp_actions.size;
1986 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
1987 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
1989 /* We've let OFPP_NORMAL and the learning action look at the
1990 * packet, so drop it now if forwarding is disabled. */
1991 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
1992 ctx.xout->odp_actions.size = sample_actions_len;
1996 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
1997 if (!hit_resubmit_limit) {
1998 /* We didn't record the original flow. Make sure we do from
2000 hit_resubmit_limit = true;
2001 } else if (!VLOG_DROP_ERR(&trace_rl)) {
2002 struct ds ds = DS_EMPTY_INITIALIZER;
2004 ofproto_trace(ctx.ofproto, &orig_flow, ctx.xin->packet, &ds);
2005 VLOG_ERR("Trace triggered by excessive resubmit "
2006 "recursion:\n%s", ds_cstr(&ds));
2011 local_odp_port = ofp_port_to_odp_port(ctx.ofproto, OFPP_LOCAL);
2012 if (!connmgr_must_output_local(ctx.ofproto->up.connmgr, &ctx.xin->flow,
2014 ctx.xout->odp_actions.data,
2015 ctx.xout->odp_actions.size)) {
2016 compose_output_action(&ctx, OFPP_LOCAL);
2018 if (ctx.ofproto->has_mirrors) {
2019 add_mirror_actions(&ctx, &orig_flow);
2021 fix_sflow_action(&ctx);
2024 ofpbuf_uninit(&ctx.stack);
2026 /* Clear the metadata and register wildcard masks, because we won't
2027 * use non-header fields as part of the cache. */
2028 memset(&ctx.xout->wc.masks.metadata, 0,
2029 sizeof ctx.xout->wc.masks.metadata);
2030 memset(&ctx.xout->wc.masks.regs, 0, sizeof ctx.xout->wc.masks.regs);