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"
33 #include "mac-learning.h"
34 #include "meta-flow.h"
35 #include "multipath.h"
36 #include "netdev-vport.h"
39 #include "odp-execute.h"
40 #include "ofp-actions.h"
41 #include "ofproto/ofproto-dpif-ipfix.h"
42 #include "ofproto/ofproto-dpif-mirror.h"
43 #include "ofproto/ofproto-dpif-sflow.h"
44 #include "ofproto/ofproto-dpif.h"
48 COVERAGE_DEFINE(xlate_actions);
50 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
52 /* Maximum depth of flow table recursion (due to resubmit actions) in a
53 * flow translation. */
54 #define MAX_RESUBMIT_RECURSION 64
57 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
58 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
60 struct list xbundles; /* Owned xbundles. */
61 struct hmap xports; /* Indexed by ofp_port. */
63 char *name; /* Name used in log messages. */
64 struct mac_learning *ml; /* Mac learning handle. */
65 struct mbridge *mbridge; /* Mirroring. */
66 struct dpif_sflow *sflow; /* SFlow handle, or null. */
67 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
69 enum ofp_config_flags frag; /* Fragmentation handling. */
70 bool has_stp; /* Bridge runs stp? */
71 bool has_netflow; /* Bridge runs netflow? */
72 bool has_in_band; /* Bridge has in band control? */
73 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
77 struct hmap_node hmap_node; /* In global 'xbundles' map. */
78 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
80 struct list list_node; /* In parent 'xbridges' list. */
81 struct xbridge *xbridge; /* Parent xbridge. */
83 struct list xports; /* Contains "struct xport"s. */
85 char *name; /* Name used in log messages. */
86 struct bond *bond; /* Nonnull iff more than one port. */
87 struct lacp *lacp; /* LACP handle or null. */
89 enum port_vlan_mode vlan_mode; /* VLAN mode. */
90 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
91 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
92 * NULL if all VLANs are trunked. */
93 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
94 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
98 struct hmap_node hmap_node; /* Node in global 'xports' map. */
99 struct ofport_dpif *ofport; /* Key in global 'xports map. */
101 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
102 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
104 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
106 struct list bundle_node; /* In parent xbundle (if it exists). */
107 struct xbundle *xbundle; /* Parent xbundle or null. */
109 struct netdev *netdev; /* 'ofport''s netdev. */
111 struct xbridge *xbridge; /* Parent bridge. */
112 struct xport *peer; /* Patch port peer or null. */
114 enum ofputil_port_config config; /* OpenFlow port configuration. */
115 enum stp_state stp_state; /* STP_DISABLED if STP not in use. */
117 bool may_enable; /* May be enabled in bonds. */
118 bool is_tunnel; /* Is a tunnel port. */
120 struct cfm *cfm; /* CFM handle or null. */
121 struct bfd *bfd; /* BFD handle or null. */
125 struct xlate_in *xin;
126 struct xlate_out *xout;
128 const struct xbridge *xbridge;
130 /* Flow at the last commit. */
131 struct flow base_flow;
133 /* Tunnel IP destination address as received. This is stored separately
134 * as the base_flow.tunnel is cleared on init to reflect the datapath
135 * behavior. Used to make sure not to send tunneled output to ourselves,
136 * which might lead to an infinite loop. This could happen easily
137 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
138 * actually set the tun_dst field. */
139 ovs_be32 orig_tunnel_ip_dst;
141 /* Stack for the push and pop actions. Each stack element is of type
142 * "union mf_subvalue". */
143 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
146 /* The rule that we are currently translating, or NULL. */
147 struct rule_dpif *rule;
149 int recurse; /* Recursion level, via xlate_table_action. */
150 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
151 uint32_t orig_skb_priority; /* Priority when packet arrived. */
152 uint8_t table_id; /* OpenFlow table ID where flow was found. */
153 uint32_t sflow_n_outputs; /* Number of output ports. */
154 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
155 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
156 bool exit; /* No further actions should be processed. */
159 /* A controller may use OFPP_NONE as the ingress port to indicate that
160 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
161 * when an input bundle is needed for validation (e.g., mirroring or
162 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
163 * any 'port' structs, so care must be taken when dealing with it. */
164 static struct xbundle ofpp_none_bundle = {
166 .vlan_mode = PORT_VLAN_TRUNK
169 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
170 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
171 static struct hmap xports = HMAP_INITIALIZER(&xports);
173 static bool may_receive(const struct xport *, struct xlate_ctx *);
174 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
176 static void xlate_normal(struct xlate_ctx *);
177 static void xlate_report(struct xlate_ctx *, const char *);
178 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
179 uint8_t table_id, bool may_packet_in);
180 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
181 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
182 static void output_normal(struct xlate_ctx *, const struct xbundle *,
184 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
186 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
188 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
189 static struct xbundle *xbundle_lookup(const struct ofbundle *);
190 static struct xport *xport_lookup(struct ofport_dpif *);
191 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
194 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
195 const struct mac_learning *ml, const struct mbridge *mbridge,
196 const struct dpif_sflow *sflow,
197 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
198 bool forward_bpdu, bool has_in_band, bool has_netflow,
201 struct xbridge *xbridge = xbridge_lookup(ofproto);
204 xbridge = xzalloc(sizeof *xbridge);
205 xbridge->ofproto = ofproto;
207 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
208 hmap_init(&xbridge->xports);
209 list_init(&xbridge->xbundles);
212 if (xbridge->ml != ml) {
213 mac_learning_unref(xbridge->ml);
214 xbridge->ml = mac_learning_ref(ml);
217 if (xbridge->mbridge != mbridge) {
218 mbridge_unref(xbridge->mbridge);
219 xbridge->mbridge = mbridge_ref(mbridge);
222 if (xbridge->sflow != sflow) {
223 dpif_sflow_unref(xbridge->sflow);
224 xbridge->sflow = dpif_sflow_ref(sflow);
227 if (xbridge->ipfix != ipfix) {
228 dpif_ipfix_unref(xbridge->ipfix);
229 xbridge->ipfix = dpif_ipfix_ref(ipfix);
233 xbridge->name = xstrdup(name);
235 xbridge->forward_bpdu = forward_bpdu;
236 xbridge->has_in_band = has_in_band;
237 xbridge->has_netflow = has_netflow;
238 xbridge->has_stp = has_stp;
239 xbridge->frag = frag;
243 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
245 struct xbridge *xbridge = xbridge_lookup(ofproto);
246 struct xbundle *xbundle, *next_xbundle;
247 struct xport *xport, *next_xport;
253 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
254 xlate_ofport_remove(xport->ofport);
257 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
258 xlate_bundle_remove(xbundle->ofbundle);
261 hmap_remove(&xbridges, &xbridge->hmap_node);
267 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
268 const char *name, enum port_vlan_mode vlan_mode, int vlan,
269 unsigned long *trunks, bool use_priority_tags,
270 const struct bond *bond, const struct lacp *lacp,
273 struct xbundle *xbundle = xbundle_lookup(ofbundle);
276 xbundle = xzalloc(sizeof *xbundle);
277 xbundle->ofbundle = ofbundle;
278 xbundle->xbridge = xbridge_lookup(ofproto);
280 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
281 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
282 list_init(&xbundle->xports);
285 ovs_assert(xbundle->xbridge);
288 xbundle->name = xstrdup(name);
290 xbundle->vlan_mode = vlan_mode;
291 xbundle->vlan = vlan;
292 xbundle->trunks = trunks;
293 xbundle->use_priority_tags = use_priority_tags;
294 xbundle->floodable = floodable;
296 if (xbundle->bond != bond) {
297 bond_unref(xbundle->bond);
298 xbundle->bond = bond_ref(bond);
301 if (xbundle->lacp != lacp) {
302 lacp_unref(xbundle->lacp);
303 xbundle->lacp = lacp_ref(lacp);
308 xlate_bundle_remove(struct ofbundle *ofbundle)
310 struct xbundle *xbundle = xbundle_lookup(ofbundle);
311 struct xport *xport, *next;
317 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
318 list_remove(&xport->bundle_node);
319 xport->xbundle = NULL;
322 hmap_remove(&xbundles, &xbundle->hmap_node);
323 list_remove(&xbundle->list_node);
324 bond_unref(xbundle->bond);
325 lacp_unref(xbundle->lacp);
331 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
332 struct ofport_dpif *ofport, ofp_port_t ofp_port,
333 odp_port_t odp_port, const struct netdev *netdev,
334 const struct cfm *cfm, const struct bfd *bfd,
335 struct ofport_dpif *peer, enum ofputil_port_config config,
336 enum stp_state stp_state, bool is_tunnel, bool may_enable)
338 struct xport *xport = xport_lookup(ofport);
341 xport = xzalloc(sizeof *xport);
342 xport->ofport = ofport;
343 xport->xbridge = xbridge_lookup(ofproto);
344 xport->ofp_port = ofp_port;
346 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
347 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
348 hash_ofp_port(xport->ofp_port));
351 ovs_assert(xport->ofp_port == ofp_port);
353 xport->config = config;
354 xport->stp_state = stp_state;
355 xport->is_tunnel = is_tunnel;
356 xport->may_enable = may_enable;
357 xport->odp_port = odp_port;
359 if (xport->netdev != netdev) {
360 netdev_close(xport->netdev);
361 xport->netdev = netdev_ref(netdev);
364 if (xport->cfm != cfm) {
365 cfm_unref(xport->cfm);
366 xport->cfm = cfm_ref(cfm);
369 if (xport->bfd != bfd) {
370 bfd_unref(xport->bfd);
371 xport->bfd = bfd_ref(bfd);
375 xport->peer->peer = NULL;
377 xport->peer = peer ? xport_lookup(peer) : NULL;
379 xport->peer->peer = xport;
382 if (xport->xbundle) {
383 list_remove(&xport->bundle_node);
385 xport->xbundle = ofbundle ? xbundle_lookup(ofbundle) : NULL;
386 if (xport->xbundle) {
387 list_insert(&xport->xbundle->xports, &xport->bundle_node);
392 xlate_ofport_remove(struct ofport_dpif *ofport)
394 struct xport *xport = xport_lookup(ofport);
401 xport->peer->peer = NULL;
405 list_remove(&xport->bundle_node);
406 hmap_remove(&xports, &xport->hmap_node);
407 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
409 netdev_close(xport->netdev);
410 cfm_unref(xport->cfm);
411 bfd_unref(xport->bfd);
415 static struct xbridge *
416 xbridge_lookup(const struct ofproto_dpif *ofproto)
418 struct xbridge *xbridge;
420 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
422 if (xbridge->ofproto == ofproto) {
429 static struct xbundle *
430 xbundle_lookup(const struct ofbundle *ofbundle)
432 struct xbundle *xbundle;
434 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
436 if (xbundle->ofbundle == ofbundle) {
443 static struct xport *
444 xport_lookup(struct ofport_dpif *ofport)
448 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
450 if (xport->ofport == ofport) {
457 static struct xport *
458 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
462 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
464 if (xport->ofp_port == ofp_port) {
472 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
474 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
475 return xport ? xport->odp_port : ODPP_NONE;
479 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
481 return (bundle->vlan_mode != PORT_VLAN_ACCESS
482 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
486 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
488 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
492 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
494 return xbundle != &ofpp_none_bundle
495 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
500 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
502 return xbundle != &ofpp_none_bundle
503 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
508 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
510 return xbundle != &ofpp_none_bundle
511 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
515 static struct xbundle *
516 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
517 bool warn, struct xport **in_xportp)
521 /* Find the port and bundle for the received packet. */
522 xport = get_ofp_port(xbridge, in_port);
526 if (xport && xport->xbundle) {
527 return xport->xbundle;
530 /* Special-case OFPP_NONE, which a controller may use as the ingress
531 * port for traffic that it is sourcing. */
532 if (in_port == OFPP_NONE) {
533 return &ofpp_none_bundle;
536 /* Odd. A few possible reasons here:
538 * - We deleted a port but there are still a few packets queued up
541 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
542 * we don't know about.
544 * - The ofproto client didn't configure the port as part of a bundle.
545 * This is particularly likely to happen if a packet was received on the
546 * port after it was created, but before the client had a chance to
547 * configure its bundle.
550 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
552 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
553 "port %"PRIu16, xbridge->name, in_port);
559 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
561 const struct xbridge *xbridge = ctx->xbridge;
562 mirror_mask_t mirrors;
563 struct xbundle *in_xbundle;
567 mirrors = ctx->xout->mirrors;
568 ctx->xout->mirrors = 0;
570 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
571 ctx->xin->packet != NULL, NULL);
575 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
577 /* Drop frames on bundles reserved for mirroring. */
578 if (xbundle_mirror_out(xbridge, in_xbundle)) {
579 if (ctx->xin->packet != NULL) {
580 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
581 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
582 "%s, which is reserved exclusively for mirroring",
583 ctx->xbridge->name, in_xbundle->name);
585 ofpbuf_clear(&ctx->xout->odp_actions);
590 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
591 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
594 vlan = input_vid_to_vlan(in_xbundle, vid);
600 /* Restore the original packet before adding the mirror actions. */
601 ctx->xin->flow = *orig_flow;
604 mirror_mask_t dup_mirrors;
605 struct ofbundle *out;
606 unsigned long *vlans;
611 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
612 &vlans, &dup_mirrors, &out, &out_vlan);
613 ovs_assert(has_mirror);
616 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
618 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
621 if (!vlan_mirrored) {
622 mirrors = zero_rightmost_1bit(mirrors);
626 mirrors &= ~dup_mirrors;
627 ctx->xout->mirrors |= dup_mirrors;
629 struct xbundle *out_xbundle = xbundle_lookup(out);
631 output_normal(ctx, out_xbundle, vlan);
633 } else if (vlan != out_vlan
634 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
635 struct xbundle *xbundle;
637 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
638 if (xbundle_includes_vlan(xbundle, out_vlan)
639 && !xbundle_mirror_out(xbridge, xbundle)) {
640 output_normal(ctx, xbundle, out_vlan);
647 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
648 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
649 * the bundle on which the packet was received, returns the VLAN to which the
652 * Both 'vid' and the return value are in the range 0...4095. */
654 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
656 switch (in_xbundle->vlan_mode) {
657 case PORT_VLAN_ACCESS:
658 return in_xbundle->vlan;
661 case PORT_VLAN_TRUNK:
664 case PORT_VLAN_NATIVE_UNTAGGED:
665 case PORT_VLAN_NATIVE_TAGGED:
666 return vid ? vid : in_xbundle->vlan;
673 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
674 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
677 * 'vid' should be the VID obtained from the 802.1Q header that was received as
678 * part of a packet (specify 0 if there was no 802.1Q header), in the range
681 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
683 /* Allow any VID on the OFPP_NONE port. */
684 if (in_xbundle == &ofpp_none_bundle) {
688 switch (in_xbundle->vlan_mode) {
689 case PORT_VLAN_ACCESS:
692 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
693 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
694 "packet received on port %s configured as VLAN "
695 "%"PRIu16" access port", vid, in_xbundle->name,
702 case PORT_VLAN_NATIVE_UNTAGGED:
703 case PORT_VLAN_NATIVE_TAGGED:
705 /* Port must always carry its native VLAN. */
709 case PORT_VLAN_TRUNK:
710 if (!xbundle_includes_vlan(in_xbundle, vid)) {
712 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
713 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
714 "received on port %s not configured for trunking "
715 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
727 /* Given 'vlan', the VLAN that a packet belongs to, and
728 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
729 * that should be included in the 802.1Q header. (If the return value is 0,
730 * then the 802.1Q header should only be included in the packet if there is a
733 * Both 'vlan' and the return value are in the range 0...4095. */
735 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
737 switch (out_xbundle->vlan_mode) {
738 case PORT_VLAN_ACCESS:
741 case PORT_VLAN_TRUNK:
742 case PORT_VLAN_NATIVE_TAGGED:
745 case PORT_VLAN_NATIVE_UNTAGGED:
746 return vlan == out_xbundle->vlan ? 0 : vlan;
754 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
757 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
759 ovs_be16 tci, old_tci;
762 vid = output_vlan_to_vid(out_xbundle, vlan);
763 if (list_is_empty(&out_xbundle->xports)) {
764 /* Partially configured bundle with no slaves. Drop the packet. */
766 } else if (!out_xbundle->bond) {
767 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
770 struct ofport_dpif *ofport;
772 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
775 xport = ofport ? xport_lookup(ofport) : NULL;
778 /* No slaves enabled, so drop packet. */
785 if (tci || out_xbundle->use_priority_tags) {
786 tci |= *flow_tci & htons(VLAN_PCP_MASK);
788 tci |= htons(VLAN_CFI);
793 compose_output_action(ctx, xport->ofp_port);
797 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
798 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
799 * indicate this; newer upstream kernels use gratuitous ARP requests. */
801 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
803 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
807 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
808 if (!eth_addr_is_broadcast(flow->dl_dst)) {
812 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
813 if (flow->nw_proto == ARP_OP_REPLY) {
815 } else if (flow->nw_proto == ARP_OP_REQUEST) {
816 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
817 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
819 return flow->nw_src == flow->nw_dst;
826 update_learning_table(const struct xbridge *xbridge,
827 const struct flow *flow, struct flow_wildcards *wc,
828 int vlan, struct xbundle *in_xbundle)
830 struct mac_entry *mac;
832 /* Don't learn the OFPP_NONE port. */
833 if (in_xbundle == &ofpp_none_bundle) {
837 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
841 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
842 if (is_gratuitous_arp(flow, wc)) {
843 /* We don't want to learn from gratuitous ARP packets that are
844 * reflected back over bond slaves so we lock the learning table. */
845 if (!in_xbundle->bond) {
846 mac_entry_set_grat_arp_lock(mac);
847 } else if (mac_entry_is_grat_arp_locked(mac)) {
852 if (mac_entry_is_new(mac) || mac->port.p != in_xbundle->ofbundle) {
853 /* The log messages here could actually be useful in debugging,
854 * so keep the rate limit relatively high. */
855 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
856 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
857 "on port %s in VLAN %d",
858 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
859 in_xbundle->name, vlan);
861 mac->port.p = in_xbundle->ofbundle;
862 mac_learning_changed(xbridge->ml, mac);
866 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
867 * dropped. Returns true if they may be forwarded, false if they should be
870 * 'in_port' must be the xport that corresponds to flow->in_port.
871 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
873 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
874 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
875 * checked by input_vid_is_valid().
877 * May also add tags to '*tags', although the current implementation only does
878 * so in one special case.
881 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
884 struct xbundle *in_xbundle = in_port->xbundle;
885 const struct xbridge *xbridge = ctx->xbridge;
886 struct flow *flow = &ctx->xin->flow;
888 /* Drop frames for reserved multicast addresses
889 * only if forward_bpdu option is absent. */
890 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
891 xlate_report(ctx, "packet has reserved destination MAC, dropping");
895 if (in_xbundle->bond) {
896 struct mac_entry *mac;
898 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
899 flow->dl_dst, &ctx->xout->tags)) {
904 xlate_report(ctx, "bonding refused admissibility, dropping");
907 case BV_DROP_IF_MOVED:
908 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan, NULL);
909 if (mac && mac->port.p != in_xbundle->ofbundle &&
910 (!is_gratuitous_arp(flow, &ctx->xout->wc)
911 || mac_entry_is_grat_arp_locked(mac))) {
912 xlate_report(ctx, "SLB bond thinks this packet looped back, "
924 xlate_normal(struct xlate_ctx *ctx)
926 struct flow_wildcards *wc = &ctx->xout->wc;
927 struct flow *flow = &ctx->xin->flow;
928 struct xbundle *in_xbundle;
929 struct xport *in_port;
930 struct mac_entry *mac;
934 ctx->xout->has_normal = true;
936 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
937 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
938 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
940 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
941 ctx->xin->packet != NULL, &in_port);
943 xlate_report(ctx, "no input bundle, dropping");
947 /* Drop malformed frames. */
948 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
949 !(flow->vlan_tci & htons(VLAN_CFI))) {
950 if (ctx->xin->packet != NULL) {
951 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
952 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
953 "VLAN tag received on port %s",
954 ctx->xbridge->name, in_xbundle->name);
956 xlate_report(ctx, "partial VLAN tag, dropping");
960 /* Drop frames on bundles reserved for mirroring. */
961 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
962 if (ctx->xin->packet != NULL) {
963 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
964 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
965 "%s, which is reserved exclusively for mirroring",
966 ctx->xbridge->name, in_xbundle->name);
968 xlate_report(ctx, "input port is mirror output port, dropping");
973 vid = vlan_tci_to_vid(flow->vlan_tci);
974 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
975 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
978 vlan = input_vid_to_vlan(in_xbundle, vid);
980 /* Check other admissibility requirements. */
981 if (in_port && !is_admissible(ctx, in_port, vlan)) {
985 /* Learn source MAC. */
986 if (ctx->xin->may_learn) {
987 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
990 /* Determine output bundle. */
991 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan,
994 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
995 if (mac_xbundle && mac_xbundle != in_xbundle) {
996 xlate_report(ctx, "forwarding to learned port");
997 output_normal(ctx, mac_xbundle, vlan);
998 } else if (!mac_xbundle) {
999 xlate_report(ctx, "learned port is unknown, dropping");
1001 xlate_report(ctx, "learned port is input port, dropping");
1004 struct xbundle *xbundle;
1006 xlate_report(ctx, "no learned MAC for destination, flooding");
1007 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1008 if (xbundle != in_xbundle
1009 && xbundle_includes_vlan(xbundle, vlan)
1010 && xbundle->floodable
1011 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1012 output_normal(ctx, xbundle, vlan);
1015 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1019 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1020 * the number of packets out of UINT32_MAX to sample. The given
1021 * cookie is passed back in the callback for each sampled packet.
1024 compose_sample_action(const struct xbridge *xbridge,
1025 struct ofpbuf *odp_actions,
1026 const struct flow *flow,
1027 const uint32_t probability,
1028 const union user_action_cookie *cookie,
1029 const size_t cookie_size)
1031 size_t sample_offset, actions_offset;
1034 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1036 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1038 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1039 cookie_offset = put_userspace_action(xbridge->ofproto, odp_actions, flow,
1040 cookie, cookie_size);
1042 nl_msg_end_nested(odp_actions, actions_offset);
1043 nl_msg_end_nested(odp_actions, sample_offset);
1044 return cookie_offset;
1048 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1049 odp_port_t odp_port, unsigned int n_outputs,
1050 union user_action_cookie *cookie)
1054 cookie->type = USER_ACTION_COOKIE_SFLOW;
1055 cookie->sflow.vlan_tci = vlan_tci;
1057 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1058 * port information") for the interpretation of cookie->output. */
1059 switch (n_outputs) {
1061 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1062 cookie->sflow.output = 0x40000000 | 256;
1066 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1068 cookie->sflow.output = ifindex;
1073 /* 0x80000000 means "multiple output ports. */
1074 cookie->sflow.output = 0x80000000 | n_outputs;
1079 /* Compose SAMPLE action for sFlow bridge sampling. */
1081 compose_sflow_action(const struct xbridge *xbridge,
1082 struct ofpbuf *odp_actions,
1083 const struct flow *flow,
1084 odp_port_t odp_port)
1086 uint32_t probability;
1087 union user_action_cookie cookie;
1089 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1093 probability = dpif_sflow_get_probability(xbridge->sflow);
1094 compose_sflow_cookie(xbridge, htons(0), odp_port,
1095 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1097 return compose_sample_action(xbridge, odp_actions, flow, probability,
1098 &cookie, sizeof cookie.sflow);
1102 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1103 uint32_t obs_domain_id, uint32_t obs_point_id,
1104 union user_action_cookie *cookie)
1106 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1107 cookie->flow_sample.probability = probability;
1108 cookie->flow_sample.collector_set_id = collector_set_id;
1109 cookie->flow_sample.obs_domain_id = obs_domain_id;
1110 cookie->flow_sample.obs_point_id = obs_point_id;
1114 compose_ipfix_cookie(union user_action_cookie *cookie)
1116 cookie->type = USER_ACTION_COOKIE_IPFIX;
1119 /* Compose SAMPLE action for IPFIX bridge sampling. */
1121 compose_ipfix_action(const struct xbridge *xbridge,
1122 struct ofpbuf *odp_actions,
1123 const struct flow *flow)
1125 uint32_t probability;
1126 union user_action_cookie cookie;
1128 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1132 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1133 compose_ipfix_cookie(&cookie);
1135 compose_sample_action(xbridge, odp_actions, flow, probability,
1136 &cookie, sizeof cookie.ipfix);
1139 /* SAMPLE action for sFlow must be first action in any given list of
1140 * actions. At this point we do not have all information required to
1141 * build it. So try to build sample action as complete as possible. */
1143 add_sflow_action(struct xlate_ctx *ctx)
1145 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1146 &ctx->xout->odp_actions,
1147 &ctx->xin->flow, ODPP_NONE);
1148 ctx->sflow_odp_port = 0;
1149 ctx->sflow_n_outputs = 0;
1152 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1153 * of actions, eventually after the SAMPLE action for sFlow. */
1155 add_ipfix_action(struct xlate_ctx *ctx)
1157 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1161 /* Fix SAMPLE action according to data collected while composing ODP actions.
1162 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1163 * USERSPACE action's user-cookie which is required for sflow. */
1165 fix_sflow_action(struct xlate_ctx *ctx)
1167 const struct flow *base = &ctx->base_flow;
1168 union user_action_cookie *cookie;
1170 if (!ctx->user_cookie_offset) {
1174 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1175 sizeof cookie->sflow);
1176 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1178 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1179 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1182 static enum slow_path_reason
1183 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1184 const struct xport *xport, const struct ofpbuf *packet)
1186 struct flow_wildcards *wc = &ctx->xout->wc;
1187 const struct xbridge *xbridge = ctx->xbridge;
1191 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1193 cfm_process_heartbeat(xport->cfm, packet);
1196 } else if (xport->bfd && bfd_should_process_flow(flow, wc)) {
1198 bfd_process_packet(xport->bfd, flow, packet);
1201 } else if (xport->xbundle && xport->xbundle->lacp
1202 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1204 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1207 } else if (xbridge->has_stp && stp_should_process_flow(flow, wc)) {
1209 stp_process_packet(xport->ofport, packet);
1218 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1221 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1222 struct flow_wildcards *wc = &ctx->xout->wc;
1223 struct flow *flow = &ctx->xin->flow;
1224 ovs_be16 flow_vlan_tci;
1225 uint32_t flow_skb_mark;
1226 uint8_t flow_nw_tos;
1227 odp_port_t out_port, odp_port;
1230 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1231 * before traversing a patch port. */
1232 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1235 xlate_report(ctx, "Nonexistent output port");
1237 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1238 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1240 } else if (check_stp && !stp_forward_in_state(xport->stp_state)) {
1241 xlate_report(ctx, "STP not in forwarding state, skipping output");
1245 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1246 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1251 const struct xport *peer = xport->peer;
1252 struct flow old_flow = ctx->xin->flow;
1253 enum slow_path_reason special;
1255 ctx->xbridge = peer->xbridge;
1256 flow->in_port.ofp_port = peer->ofp_port;
1257 flow->metadata = htonll(0);
1258 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1259 memset(flow->regs, 0, sizeof flow->regs);
1261 special = process_special(ctx, &ctx->xin->flow, peer,
1264 ctx->xout->slow = special;
1265 } else if (may_receive(peer, ctx)) {
1266 if (stp_forward_in_state(peer->stp_state)) {
1267 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1269 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1270 * learning action look at the packet, then drop it. */
1271 struct flow old_base_flow = ctx->base_flow;
1272 size_t old_size = ctx->xout->odp_actions.size;
1273 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1274 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1275 ctx->xout->mirrors = old_mirrors;
1276 ctx->base_flow = old_base_flow;
1277 ctx->xout->odp_actions.size = old_size;
1281 ctx->xin->flow = old_flow;
1282 ctx->xbridge = xport->xbundle->xbridge;
1284 if (ctx->xin->resubmit_stats) {
1285 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1286 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1292 flow_vlan_tci = flow->vlan_tci;
1293 flow_skb_mark = flow->skb_mark;
1294 flow_nw_tos = flow->nw_tos;
1296 if (ofproto_dpif_dscp_from_priority(xport->ofport, flow->skb_priority,
1298 wc->masks.nw_tos |= IP_ECN_MASK;
1299 flow->nw_tos &= ~IP_DSCP_MASK;
1300 flow->nw_tos |= dscp;
1303 if (xport->is_tunnel) {
1304 /* Save tunnel metadata so that changes made due to
1305 * the Logical (tunnel) Port are not visible for any further
1306 * matches, while explicit set actions on tunnel metadata are.
1308 struct flow_tnl flow_tnl = flow->tunnel;
1309 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1310 if (odp_port == ODPP_NONE) {
1311 xlate_report(ctx, "Tunneling decided against output");
1312 goto out; /* restore flow_nw_tos */
1314 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1315 xlate_report(ctx, "Not tunneling to our own address");
1316 goto out; /* restore flow_nw_tos */
1318 if (ctx->xin->resubmit_stats) {
1319 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1321 out_port = odp_port;
1322 commit_odp_tunnel_action(flow, &ctx->base_flow,
1323 &ctx->xout->odp_actions);
1324 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1326 ofp_port_t vlandev_port;
1328 odp_port = xport->odp_port;
1329 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1330 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1332 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1334 if (vlandev_port == ofp_port) {
1335 out_port = odp_port;
1337 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1338 flow->vlan_tci = htons(0);
1340 flow->skb_mark &= ~IPSEC_MARK;
1343 if (out_port != ODPP_NONE) {
1344 commit_odp_actions(flow, &ctx->base_flow,
1345 &ctx->xout->odp_actions, &ctx->xout->wc);
1346 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1349 ctx->sflow_odp_port = odp_port;
1350 ctx->sflow_n_outputs++;
1351 ctx->xout->nf_output_iface = ofp_port;
1356 flow->vlan_tci = flow_vlan_tci;
1357 flow->skb_mark = flow_skb_mark;
1358 flow->nw_tos = flow_nw_tos;
1362 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1364 compose_output_action__(ctx, ofp_port, true);
1367 /* Common rule processing in one place to avoid duplicating code. */
1368 static struct rule_dpif *
1369 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
1372 if (ctx->xin->resubmit_hook) {
1373 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1375 if (rule == NULL && may_packet_in) {
1377 * check if table configuration flags
1378 * OFPTC_TABLE_MISS_CONTROLLER, default.
1379 * OFPTC_TABLE_MISS_CONTINUE,
1380 * OFPTC_TABLE_MISS_DROP
1381 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
1383 rule = rule_dpif_miss_rule(ctx->xbridge->ofproto, &ctx->xin->flow);
1385 if (rule && ctx->xin->resubmit_stats) {
1386 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1392 xlate_table_action(struct xlate_ctx *ctx,
1393 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1395 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1396 struct rule_dpif *rule;
1397 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1398 uint8_t old_table_id = ctx->table_id;
1400 ctx->table_id = table_id;
1402 /* Look up a flow with 'in_port' as the input port. */
1403 ctx->xin->flow.in_port.ofp_port = in_port;
1404 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1405 &ctx->xin->flow, &ctx->xout->wc,
1408 ctx->xout->tags |= calculate_flow_tag(ctx->xbridge->ofproto,
1409 &ctx->xin->flow, ctx->table_id,
1412 /* Restore the original input port. Otherwise OFPP_NORMAL and
1413 * OFPP_IN_PORT will have surprising behavior. */
1414 ctx->xin->flow.in_port.ofp_port = old_in_port;
1416 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1419 struct rule_dpif *old_rule = ctx->rule;
1423 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1424 ctx->rule = old_rule;
1428 ctx->table_id = old_table_id;
1430 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1432 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1433 MAX_RESUBMIT_RECURSION);
1434 ctx->max_resubmit_trigger = true;
1439 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1440 const struct ofpact_resubmit *resubmit)
1445 in_port = resubmit->in_port;
1446 if (in_port == OFPP_IN_PORT) {
1447 in_port = ctx->xin->flow.in_port.ofp_port;
1450 table_id = resubmit->table_id;
1451 if (table_id == 255) {
1452 table_id = ctx->table_id;
1455 xlate_table_action(ctx, in_port, table_id, false);
1459 flood_packets(struct xlate_ctx *ctx, bool all)
1461 const struct xport *xport;
1463 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1464 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1469 compose_output_action__(ctx, xport->ofp_port, false);
1470 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1471 compose_output_action(ctx, xport->ofp_port);
1475 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1479 execute_controller_action(struct xlate_ctx *ctx, int len,
1480 enum ofp_packet_in_reason reason,
1481 uint16_t controller_id)
1483 struct ofputil_packet_in pin;
1484 struct ofpbuf *packet;
1487 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1488 ctx->xout->slow = SLOW_CONTROLLER;
1489 if (!ctx->xin->packet) {
1493 packet = ofpbuf_clone(ctx->xin->packet);
1495 key.skb_priority = 0;
1497 memset(&key.tunnel, 0, sizeof key.tunnel);
1499 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1500 &ctx->xout->odp_actions, &ctx->xout->wc);
1502 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1503 ctx->xout->odp_actions.size, NULL, NULL);
1505 pin.packet = packet->data;
1506 pin.packet_len = packet->size;
1507 pin.reason = reason;
1508 pin.controller_id = controller_id;
1509 pin.table_id = ctx->table_id;
1510 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1513 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1515 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, &pin);
1516 ofpbuf_delete(packet);
1520 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1522 struct flow_wildcards *wc = &ctx->xout->wc;
1523 struct flow *flow = &ctx->xin->flow;
1525 ovs_assert(eth_type_mpls(eth_type));
1527 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1528 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1530 if (flow->mpls_depth) {
1531 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1537 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1538 label = htonl(0x2); /* IPV6 Explicit Null. */
1540 label = htonl(0x0); /* IPV4 Explicit Null. */
1542 wc->masks.nw_tos |= IP_DSCP_MASK;
1543 wc->masks.nw_ttl = 0xff;
1544 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1545 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1546 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1547 flow->mpls_depth = 1;
1549 flow->dl_type = eth_type;
1553 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1555 struct flow_wildcards *wc = &ctx->xout->wc;
1556 struct flow *flow = &ctx->xin->flow;
1558 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1559 ovs_assert(!eth_type_mpls(eth_type));
1561 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1562 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1564 if (flow->mpls_depth) {
1566 flow->mpls_lse = htonl(0);
1567 if (!flow->mpls_depth) {
1568 flow->dl_type = eth_type;
1574 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1576 struct flow *flow = &ctx->xin->flow;
1578 if (!is_ip_any(flow)) {
1582 ctx->xout->wc.masks.nw_ttl = 0xff;
1583 if (flow->nw_ttl > 1) {
1589 for (i = 0; i < ids->n_controllers; i++) {
1590 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1594 /* Stop processing for current table. */
1600 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1602 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1606 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1611 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1613 struct flow *flow = &ctx->xin->flow;
1614 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1615 struct flow_wildcards *wc = &ctx->xout->wc;
1617 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1619 if (!eth_type_mpls(flow->dl_type)) {
1625 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1628 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1630 /* Stop processing for current table. */
1636 xlate_output_action(struct xlate_ctx *ctx,
1637 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1639 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1641 ctx->xout->nf_output_iface = NF_OUT_DROP;
1645 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1648 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1655 flood_packets(ctx, false);
1658 flood_packets(ctx, true);
1660 case OFPP_CONTROLLER:
1661 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1667 if (port != ctx->xin->flow.in_port.ofp_port) {
1668 compose_output_action(ctx, port);
1670 xlate_report(ctx, "skipping output to input port");
1675 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1676 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1677 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1678 ctx->xout->nf_output_iface = prev_nf_output_iface;
1679 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1680 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1681 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1686 xlate_output_reg_action(struct xlate_ctx *ctx,
1687 const struct ofpact_output_reg *or)
1689 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1690 if (port <= UINT16_MAX) {
1691 union mf_subvalue value;
1693 memset(&value, 0xff, sizeof value);
1694 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1695 xlate_output_action(ctx, u16_to_ofp(port),
1696 or->max_len, false);
1701 xlate_enqueue_action(struct xlate_ctx *ctx,
1702 const struct ofpact_enqueue *enqueue)
1704 ofp_port_t ofp_port = enqueue->port;
1705 uint32_t queue_id = enqueue->queue;
1706 uint32_t flow_priority, priority;
1709 /* Translate queue to priority. */
1710 error = ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1713 /* Fall back to ordinary output action. */
1714 xlate_output_action(ctx, enqueue->port, 0, false);
1718 /* Check output port. */
1719 if (ofp_port == OFPP_IN_PORT) {
1720 ofp_port = ctx->xin->flow.in_port.ofp_port;
1721 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1725 /* Add datapath actions. */
1726 flow_priority = ctx->xin->flow.skb_priority;
1727 ctx->xin->flow.skb_priority = priority;
1728 compose_output_action(ctx, ofp_port);
1729 ctx->xin->flow.skb_priority = flow_priority;
1731 /* Update NetFlow output port. */
1732 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1733 ctx->xout->nf_output_iface = ofp_port;
1734 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1735 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1740 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1742 uint32_t skb_priority;
1744 if (!ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1746 ctx->xin->flow.skb_priority = skb_priority;
1748 /* Couldn't translate queue to a priority. Nothing to do. A warning
1749 * has already been logged. */
1754 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
1756 const struct xbridge *xbridge = xbridge_;
1767 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1770 port = get_ofp_port(xbridge, ofp_port);
1771 return port ? port->may_enable : false;
1776 xlate_bundle_action(struct xlate_ctx *ctx,
1777 const struct ofpact_bundle *bundle)
1781 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1783 CONST_CAST(struct xbridge *, ctx->xbridge));
1784 if (bundle->dst.field) {
1785 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1787 xlate_output_action(ctx, port, 0, false);
1792 xlate_learn_action(struct xlate_ctx *ctx,
1793 const struct ofpact_learn *learn)
1795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1796 struct ofputil_flow_mod fm;
1797 uint64_t ofpacts_stub[1024 / 8];
1798 struct ofpbuf ofpacts;
1801 ctx->xout->has_learn = true;
1803 learn_mask(learn, &ctx->xout->wc);
1805 if (!ctx->xin->may_learn) {
1809 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1810 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1812 error = ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
1813 if (error && !VLOG_DROP_WARN(&rl)) {
1814 VLOG_WARN("learning action failed to modify flow table (%s)",
1815 ofperr_get_name(error));
1818 ofpbuf_uninit(&ofpacts);
1821 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1822 * means "infinite". */
1824 reduce_timeout(uint16_t max, uint16_t *timeout)
1826 if (max && (!*timeout || *timeout > max)) {
1832 xlate_fin_timeout(struct xlate_ctx *ctx,
1833 const struct ofpact_fin_timeout *oft)
1835 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1836 struct rule_dpif *rule = ctx->rule;
1838 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1839 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1844 xlate_sample_action(struct xlate_ctx *ctx,
1845 const struct ofpact_sample *os)
1847 union user_action_cookie cookie;
1848 /* Scale the probability from 16-bit to 32-bit while representing
1849 * the same percentage. */
1850 uint32_t probability = (os->probability << 16) | os->probability;
1852 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1853 &ctx->xout->odp_actions, &ctx->xout->wc);
1855 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1856 os->obs_domain_id, os->obs_point_id, &cookie);
1857 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
1858 probability, &cookie, sizeof cookie.flow_sample);
1862 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
1864 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
1865 ? OFPUTIL_PC_NO_RECV_STP
1866 : OFPUTIL_PC_NO_RECV)) {
1870 /* Only drop packets here if both forwarding and learning are
1871 * disabled. If just learning is enabled, we need to have
1872 * OFPP_NORMAL and the learning action have a look at the packet
1873 * before we can drop it. */
1874 if (!stp_forward_in_state(xport->stp_state)
1875 && !stp_learn_in_state(xport->stp_state)) {
1883 tunnel_ecn_ok(struct xlate_ctx *ctx)
1885 if (is_ip_any(&ctx->base_flow)
1886 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1887 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1888 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1889 " but is not ECN capable");
1892 /* Set the ECN CE value in the tunneled packet. */
1893 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1901 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1902 struct xlate_ctx *ctx)
1904 struct flow_wildcards *wc = &ctx->xout->wc;
1905 struct flow *flow = &ctx->xin->flow;
1906 bool was_evictable = true;
1907 const struct ofpact *a;
1910 /* Don't let the rule we're working on get evicted underneath us. */
1911 was_evictable = ctx->rule->up.evictable;
1912 ctx->rule->up.evictable = false;
1915 do_xlate_actions_again:
1916 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1917 struct ofpact_controller *controller;
1918 const struct ofpact_metadata *metadata;
1926 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1927 ofpact_get_OUTPUT(a)->max_len, true);
1930 case OFPACT_CONTROLLER:
1931 controller = ofpact_get_CONTROLLER(a);
1932 execute_controller_action(ctx, controller->max_len,
1934 controller->controller_id);
1937 case OFPACT_ENQUEUE:
1938 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
1941 case OFPACT_SET_VLAN_VID:
1942 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
1943 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
1947 case OFPACT_SET_VLAN_PCP:
1948 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
1950 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
1954 case OFPACT_STRIP_VLAN:
1955 flow->vlan_tci = htons(0);
1958 case OFPACT_PUSH_VLAN:
1959 /* XXX 802.1AD(QinQ) */
1960 flow->vlan_tci = htons(VLAN_CFI);
1963 case OFPACT_SET_ETH_SRC:
1964 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
1967 case OFPACT_SET_ETH_DST:
1968 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
1971 case OFPACT_SET_IPV4_SRC:
1972 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1973 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
1977 case OFPACT_SET_IPV4_DST:
1978 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1979 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
1983 case OFPACT_SET_IPV4_DSCP:
1984 /* OpenFlow 1.0 only supports IPv4. */
1985 if (flow->dl_type == htons(ETH_TYPE_IP)) {
1986 flow->nw_tos &= ~IP_DSCP_MASK;
1987 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
1991 case OFPACT_SET_L4_SRC_PORT:
1992 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1993 if (is_ip_any(flow)) {
1994 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
1998 case OFPACT_SET_L4_DST_PORT:
1999 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2000 if (is_ip_any(flow)) {
2001 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2005 case OFPACT_RESUBMIT:
2006 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2009 case OFPACT_SET_TUNNEL:
2010 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2013 case OFPACT_SET_QUEUE:
2014 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2017 case OFPACT_POP_QUEUE:
2018 flow->skb_priority = ctx->orig_skb_priority;
2021 case OFPACT_REG_MOVE:
2022 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2025 case OFPACT_REG_LOAD:
2026 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2029 case OFPACT_STACK_PUSH:
2030 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2034 case OFPACT_STACK_POP:
2035 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
2038 case OFPACT_PUSH_MPLS:
2039 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2042 case OFPACT_POP_MPLS:
2043 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2046 case OFPACT_SET_MPLS_TTL:
2047 if (compose_set_mpls_ttl_action(ctx,
2048 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2053 case OFPACT_DEC_MPLS_TTL:
2054 if (compose_dec_mpls_ttl_action(ctx)) {
2059 case OFPACT_DEC_TTL:
2060 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2066 /* Nothing to do. */
2069 case OFPACT_MULTIPATH:
2070 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2074 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2077 case OFPACT_OUTPUT_REG:
2078 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2082 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2089 case OFPACT_FIN_TIMEOUT:
2090 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2091 ctx->xout->has_fin_timeout = true;
2092 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2095 case OFPACT_CLEAR_ACTIONS:
2097 * Nothing to do because writa-actions is not supported for now.
2098 * When writa-actions is supported, clear-actions also must
2099 * be supported at the same time.
2103 case OFPACT_WRITE_METADATA:
2104 metadata = ofpact_get_WRITE_METADATA(a);
2105 flow->metadata &= ~metadata->mask;
2106 flow->metadata |= metadata->metadata & metadata->mask;
2110 /* Not implemented yet. */
2113 case OFPACT_GOTO_TABLE: {
2114 /* It is assumed that goto-table is the last action. */
2115 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2116 struct rule_dpif *rule;
2118 ovs_assert(ctx->table_id < ogt->table_id);
2120 ctx->table_id = ogt->table_id;
2122 /* Look up a flow from the new table. */
2123 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto, flow, wc,
2126 ctx->xout->tags |= calculate_flow_tag(ctx->xbridge->ofproto,
2128 ctx->table_id, rule);
2130 rule = ctx_rule_hooks(ctx, rule, true);
2134 ctx->rule->up.evictable = was_evictable;
2137 was_evictable = rule->up.evictable;
2138 rule->up.evictable = false;
2140 /* Tail recursion removal. */
2141 ofpacts = rule->up.ofpacts;
2142 ofpacts_len = rule->up.ofpacts_len;
2143 goto do_xlate_actions_again;
2149 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2156 ctx->rule->up.evictable = was_evictable;
2161 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2162 const struct flow *flow, struct rule_dpif *rule,
2163 uint8_t tcp_flags, const struct ofpbuf *packet)
2165 xin->ofproto = ofproto;
2167 xin->packet = packet;
2168 xin->may_learn = packet != NULL;
2170 xin->ofpacts = NULL;
2171 xin->ofpacts_len = 0;
2172 xin->tcp_flags = tcp_flags;
2173 xin->resubmit_hook = NULL;
2174 xin->report_hook = NULL;
2175 xin->resubmit_stats = NULL;
2179 xlate_out_uninit(struct xlate_out *xout)
2182 ofpbuf_uninit(&xout->odp_actions);
2186 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2187 * into datapath actions, using 'ctx', and discards the datapath actions. */
2189 xlate_actions_for_side_effects(struct xlate_in *xin)
2191 struct xlate_out xout;
2193 xlate_actions(xin, &xout);
2194 xlate_out_uninit(&xout);
2198 xlate_report(struct xlate_ctx *ctx, const char *s)
2200 if (ctx->xin->report_hook) {
2201 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2206 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2209 dst->tags = src->tags;
2210 dst->slow = src->slow;
2211 dst->has_learn = src->has_learn;
2212 dst->has_normal = src->has_normal;
2213 dst->has_fin_timeout = src->has_fin_timeout;
2214 dst->nf_output_iface = src->nf_output_iface;
2215 dst->mirrors = src->mirrors;
2217 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2218 sizeof dst->odp_actions_stub);
2219 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2220 src->odp_actions.size);
2224 actions_output_to_local_port(const struct xlate_ctx *ctx)
2226 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2227 const struct nlattr *a;
2230 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2231 ctx->xout->odp_actions.size) {
2232 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2233 && nl_attr_get_odp_port(a) == local_odp_port) {
2240 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2241 * into datapath actions in 'odp_actions', using 'ctx'. */
2243 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2245 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
2246 * that in the future we always keep a copy of the original flow for
2247 * tracing purposes. */
2248 static bool hit_resubmit_limit;
2250 struct flow_wildcards *wc = &xout->wc;
2251 struct flow *flow = &xin->flow;
2253 enum slow_path_reason special;
2254 const struct ofpact *ofpacts;
2255 struct xport *in_port;
2256 struct flow orig_flow;
2257 struct xlate_ctx ctx;
2260 COVERAGE_INC(xlate_actions);
2262 /* Flow initialization rules:
2263 * - 'base_flow' must match the kernel's view of the packet at the
2264 * time that action processing starts. 'flow' represents any
2265 * transformations we wish to make through actions.
2266 * - By default 'base_flow' and 'flow' are the same since the input
2267 * packet matches the output before any actions are applied.
2268 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2269 * of the received packet as seen by the kernel. If we later output
2270 * to another device without any modifications this will cause us to
2271 * insert a new tag since the original one was stripped off by the
2273 * - Tunnel metadata as received is retained in 'flow'. This allows
2274 * tunnel metadata matching also in later tables.
2275 * Since a kernel action for setting the tunnel metadata will only be
2276 * generated with actual tunnel output, changing the tunnel metadata
2277 * values in 'flow' (such as tun_id) will only have effect with a later
2278 * tunnel output action.
2279 * - Tunnel 'base_flow' is completely cleared since that is what the
2280 * kernel does. If we wish to maintain the original values an action
2281 * needs to be generated. */
2287 ctx.xout->has_learn = false;
2288 ctx.xout->has_normal = false;
2289 ctx.xout->has_fin_timeout = false;
2290 ctx.xout->nf_output_iface = NF_OUT_DROP;
2291 ctx.xout->mirrors = 0;
2292 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2293 sizeof ctx.xout->odp_actions_stub);
2294 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2296 ctx.xbridge = xbridge_lookup(xin->ofproto);
2301 ctx.rule = xin->rule;
2303 ctx.base_flow = *flow;
2304 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2305 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2307 flow_wildcards_init_catchall(wc);
2308 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2309 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2310 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2311 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2313 if (tnl_port_should_receive(&ctx.xin->flow)) {
2314 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
2316 if (ctx.xbridge->has_netflow) {
2317 netflow_mask_wc(flow, wc);
2321 ctx.max_resubmit_trigger = false;
2322 ctx.orig_skb_priority = flow->skb_priority;
2327 ofpacts = xin->ofpacts;
2328 ofpacts_len = xin->ofpacts_len;
2329 } else if (xin->rule) {
2330 ofpacts = xin->rule->up.ofpacts;
2331 ofpacts_len = xin->rule->up.ofpacts_len;
2336 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2338 if (mbridge_has_mirrors(ctx.xbridge->mbridge) || hit_resubmit_limit) {
2339 /* Do this conditionally because the copy is expensive enough that it
2340 * shows up in profiles. */
2344 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2345 switch (ctx.xbridge->frag) {
2346 case OFPC_FRAG_NORMAL:
2347 /* We must pretend that transport ports are unavailable. */
2348 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2349 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2352 case OFPC_FRAG_DROP:
2355 case OFPC_FRAG_REASM:
2358 case OFPC_FRAG_NX_MATCH:
2359 /* Nothing to do. */
2362 case OFPC_INVALID_TTL_TO_CONTROLLER:
2367 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2368 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2370 ctx.xout->slow = special;
2372 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2373 size_t sample_actions_len;
2375 if (flow->in_port.ofp_port
2376 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2377 flow->in_port.ofp_port,
2379 ctx.base_flow.vlan_tci = 0;
2382 add_sflow_action(&ctx);
2383 add_ipfix_action(&ctx);
2384 sample_actions_len = ctx.xout->odp_actions.size;
2386 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
2387 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2389 /* We've let OFPP_NORMAL and the learning action look at the
2390 * packet, so drop it now if forwarding is disabled. */
2391 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
2392 ctx.xout->odp_actions.size = sample_actions_len;
2396 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
2397 if (!hit_resubmit_limit) {
2398 /* We didn't record the original flow. Make sure we do from
2400 hit_resubmit_limit = true;
2401 } else if (!VLOG_DROP_ERR(&trace_rl)) {
2402 struct ds ds = DS_EMPTY_INITIALIZER;
2404 ofproto_trace(ctx.xbridge->ofproto, &orig_flow,
2405 ctx.xin->packet, &ds);
2406 VLOG_ERR("Trace triggered by excessive resubmit "
2407 "recursion:\n%s", ds_cstr(&ds));
2412 if (ctx.xbridge->has_in_band
2413 && in_band_must_output_to_local_port(flow)
2414 && !actions_output_to_local_port(&ctx)) {
2415 compose_output_action(&ctx, OFPP_LOCAL);
2418 fix_sflow_action(&ctx);
2420 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2421 add_mirror_actions(&ctx, &orig_flow);
2425 ofpbuf_uninit(&ctx.stack);
2427 /* Clear the metadata and register wildcard masks, because we won't
2428 * use non-header fields as part of the cache. */
2429 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2430 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);