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. */
68 struct stp *stp; /* STP or null if disabled. */
70 enum ofp_config_flags frag; /* Fragmentation handling. */
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 int stp_port_no; /* STP port number or 0 if 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 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
165 static struct xbundle ofpp_none_bundle;
167 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
168 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
169 static struct hmap xports = HMAP_INITIALIZER(&xports);
171 static bool may_receive(const struct xport *, struct xlate_ctx *);
172 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
174 static void xlate_normal(struct xlate_ctx *);
175 static void xlate_report(struct xlate_ctx *, const char *);
176 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
177 uint8_t table_id, bool may_packet_in);
178 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
179 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
180 static void output_normal(struct xlate_ctx *, const struct xbundle *,
182 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
184 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
186 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
187 static struct xbundle *xbundle_lookup(const struct ofbundle *);
188 static struct xport *xport_lookup(struct ofport_dpif *);
189 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
192 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
193 const struct mac_learning *ml, struct stp *stp,
194 const struct mbridge *mbridge,
195 const struct dpif_sflow *sflow,
196 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
197 bool forward_bpdu, bool has_in_band, bool has_netflow)
199 struct xbridge *xbridge = xbridge_lookup(ofproto);
202 xbridge = xzalloc(sizeof *xbridge);
203 xbridge->ofproto = ofproto;
205 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
206 hmap_init(&xbridge->xports);
207 list_init(&xbridge->xbundles);
210 if (xbridge->ml != ml) {
211 mac_learning_unref(xbridge->ml);
212 xbridge->ml = mac_learning_ref(ml);
215 if (xbridge->mbridge != mbridge) {
216 mbridge_unref(xbridge->mbridge);
217 xbridge->mbridge = mbridge_ref(mbridge);
220 if (xbridge->sflow != sflow) {
221 dpif_sflow_unref(xbridge->sflow);
222 xbridge->sflow = dpif_sflow_ref(sflow);
225 if (xbridge->ipfix != ipfix) {
226 dpif_ipfix_unref(xbridge->ipfix);
227 xbridge->ipfix = dpif_ipfix_ref(ipfix);
230 if (xbridge->stp != stp) {
231 stp_unref(xbridge->stp);
232 xbridge->stp = stp_ref(stp);
236 xbridge->name = xstrdup(name);
238 xbridge->forward_bpdu = forward_bpdu;
239 xbridge->has_in_band = has_in_band;
240 xbridge->has_netflow = has_netflow;
241 xbridge->frag = frag;
245 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
247 struct xbridge *xbridge = xbridge_lookup(ofproto);
248 struct xbundle *xbundle, *next_xbundle;
249 struct xport *xport, *next_xport;
255 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
256 xlate_ofport_remove(xport->ofport);
259 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
260 xlate_bundle_remove(xbundle->ofbundle);
263 hmap_remove(&xbridges, &xbridge->hmap_node);
269 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
270 const char *name, enum port_vlan_mode vlan_mode, int vlan,
271 unsigned long *trunks, bool use_priority_tags,
272 const struct bond *bond, const struct lacp *lacp,
275 struct xbundle *xbundle = xbundle_lookup(ofbundle);
278 xbundle = xzalloc(sizeof *xbundle);
279 xbundle->ofbundle = ofbundle;
280 xbundle->xbridge = xbridge_lookup(ofproto);
282 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
283 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
284 list_init(&xbundle->xports);
287 ovs_assert(xbundle->xbridge);
290 xbundle->name = xstrdup(name);
292 xbundle->vlan_mode = vlan_mode;
293 xbundle->vlan = vlan;
294 xbundle->trunks = trunks;
295 xbundle->use_priority_tags = use_priority_tags;
296 xbundle->floodable = floodable;
298 if (xbundle->bond != bond) {
299 bond_unref(xbundle->bond);
300 xbundle->bond = bond_ref(bond);
303 if (xbundle->lacp != lacp) {
304 lacp_unref(xbundle->lacp);
305 xbundle->lacp = lacp_ref(lacp);
310 xlate_bundle_remove(struct ofbundle *ofbundle)
312 struct xbundle *xbundle = xbundle_lookup(ofbundle);
313 struct xport *xport, *next;
319 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
320 list_remove(&xport->bundle_node);
321 xport->xbundle = NULL;
324 hmap_remove(&xbundles, &xbundle->hmap_node);
325 list_remove(&xbundle->list_node);
326 bond_unref(xbundle->bond);
327 lacp_unref(xbundle->lacp);
333 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
334 struct ofport_dpif *ofport, ofp_port_t ofp_port,
335 odp_port_t odp_port, const struct netdev *netdev,
336 const struct cfm *cfm, const struct bfd *bfd,
337 struct ofport_dpif *peer, int stp_port_no,
338 enum ofputil_port_config config, bool is_tunnel,
341 struct xport *xport = xport_lookup(ofport);
344 xport = xzalloc(sizeof *xport);
345 xport->ofport = ofport;
346 xport->xbridge = xbridge_lookup(ofproto);
347 xport->ofp_port = ofp_port;
349 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
350 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
351 hash_ofp_port(xport->ofp_port));
354 ovs_assert(xport->ofp_port == ofp_port);
356 xport->config = config;
357 xport->stp_port_no = stp_port_no;
358 xport->is_tunnel = is_tunnel;
359 xport->may_enable = may_enable;
360 xport->odp_port = odp_port;
362 if (xport->netdev != netdev) {
363 netdev_close(xport->netdev);
364 xport->netdev = netdev_ref(netdev);
367 if (xport->cfm != cfm) {
368 cfm_unref(xport->cfm);
369 xport->cfm = cfm_ref(cfm);
372 if (xport->bfd != bfd) {
373 bfd_unref(xport->bfd);
374 xport->bfd = bfd_ref(bfd);
378 xport->peer->peer = NULL;
380 xport->peer = peer ? xport_lookup(peer) : NULL;
382 xport->peer->peer = xport;
385 if (xport->xbundle) {
386 list_remove(&xport->bundle_node);
388 xport->xbundle = ofbundle ? xbundle_lookup(ofbundle) : NULL;
389 if (xport->xbundle) {
390 list_insert(&xport->xbundle->xports, &xport->bundle_node);
395 xlate_ofport_remove(struct ofport_dpif *ofport)
397 struct xport *xport = xport_lookup(ofport);
404 xport->peer->peer = NULL;
408 if (xport->xbundle) {
409 list_remove(&xport->bundle_node);
412 hmap_remove(&xports, &xport->hmap_node);
413 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
415 netdev_close(xport->netdev);
416 cfm_unref(xport->cfm);
417 bfd_unref(xport->bfd);
421 static struct xbridge *
422 xbridge_lookup(const struct ofproto_dpif *ofproto)
424 struct xbridge *xbridge;
426 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
428 if (xbridge->ofproto == ofproto) {
435 static struct xbundle *
436 xbundle_lookup(const struct ofbundle *ofbundle)
438 struct xbundle *xbundle;
440 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
442 if (xbundle->ofbundle == ofbundle) {
449 static struct xport *
450 xport_lookup(struct ofport_dpif *ofport)
454 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
456 if (xport->ofport == ofport) {
464 static enum stp_state
465 xport_stp_learn_state(const struct xport *xport)
467 enum stp_state stp_state = xport->xbridge->stp && xport->stp_port_no
468 ? stp_port_get_state(stp_get_port(xport->xbridge->stp,
471 return stp_learn_in_state(stp_state);
475 xport_stp_forward_state(const struct xport *xport)
477 enum stp_state stp_state = xport->xbridge->stp && xport->stp_port_no
478 ? stp_port_get_state(stp_get_port(xport->xbridge->stp,
481 return stp_forward_in_state(stp_state);
484 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
485 * were used to make the determination.*/
487 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
489 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
490 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
494 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
496 struct ofpbuf payload = *packet;
497 struct eth_header *eth = payload.data;
498 struct stp_port *sp = xport->xbridge->stp && xport->stp_port_no
499 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
502 /* Sink packets on ports that have STP disabled when the bridge has
504 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
508 /* Trim off padding on payload. */
509 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
510 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
513 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
514 stp_received_bpdu(sp, payload.data, payload.size);
518 static struct xport *
519 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
523 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
525 if (xport->ofp_port == ofp_port) {
533 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
535 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
536 return xport ? xport->odp_port : ODPP_NONE;
540 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
542 return (bundle->vlan_mode != PORT_VLAN_ACCESS
543 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
547 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
549 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
553 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
555 return xbundle != &ofpp_none_bundle
556 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
561 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
563 return xbundle != &ofpp_none_bundle
564 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
569 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
571 return xbundle != &ofpp_none_bundle
572 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
576 static struct xbundle *
577 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
578 bool warn, struct xport **in_xportp)
582 /* Find the port and bundle for the received packet. */
583 xport = get_ofp_port(xbridge, in_port);
587 if (xport && xport->xbundle) {
588 return xport->xbundle;
591 /* Special-case OFPP_NONE, which a controller may use as the ingress
592 * port for traffic that it is sourcing. */
593 if (in_port == OFPP_NONE) {
594 ofpp_none_bundle.name = "OFPP_NONE";
595 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
596 return &ofpp_none_bundle;
599 /* Odd. A few possible reasons here:
601 * - We deleted a port but there are still a few packets queued up
604 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
605 * we don't know about.
607 * - The ofproto client didn't configure the port as part of a bundle.
608 * This is particularly likely to happen if a packet was received on the
609 * port after it was created, but before the client had a chance to
610 * configure its bundle.
613 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
615 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
616 "port %"PRIu16, xbridge->name, in_port);
622 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
624 const struct xbridge *xbridge = ctx->xbridge;
625 mirror_mask_t mirrors;
626 struct xbundle *in_xbundle;
630 mirrors = ctx->xout->mirrors;
631 ctx->xout->mirrors = 0;
633 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
634 ctx->xin->packet != NULL, NULL);
638 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
640 /* Drop frames on bundles reserved for mirroring. */
641 if (xbundle_mirror_out(xbridge, in_xbundle)) {
642 if (ctx->xin->packet != NULL) {
643 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
644 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
645 "%s, which is reserved exclusively for mirroring",
646 ctx->xbridge->name, in_xbundle->name);
648 ofpbuf_clear(&ctx->xout->odp_actions);
653 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
654 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
657 vlan = input_vid_to_vlan(in_xbundle, vid);
663 /* Restore the original packet before adding the mirror actions. */
664 ctx->xin->flow = *orig_flow;
667 mirror_mask_t dup_mirrors;
668 struct ofbundle *out;
669 unsigned long *vlans;
674 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
675 &vlans, &dup_mirrors, &out, &out_vlan);
676 ovs_assert(has_mirror);
679 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
681 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
684 if (!vlan_mirrored) {
685 mirrors = zero_rightmost_1bit(mirrors);
689 mirrors &= ~dup_mirrors;
690 ctx->xout->mirrors |= dup_mirrors;
692 struct xbundle *out_xbundle = xbundle_lookup(out);
694 output_normal(ctx, out_xbundle, vlan);
696 } else if (vlan != out_vlan
697 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
698 struct xbundle *xbundle;
700 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
701 if (xbundle_includes_vlan(xbundle, out_vlan)
702 && !xbundle_mirror_out(xbridge, xbundle)) {
703 output_normal(ctx, xbundle, out_vlan);
710 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
711 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
712 * the bundle on which the packet was received, returns the VLAN to which the
715 * Both 'vid' and the return value are in the range 0...4095. */
717 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
719 switch (in_xbundle->vlan_mode) {
720 case PORT_VLAN_ACCESS:
721 return in_xbundle->vlan;
724 case PORT_VLAN_TRUNK:
727 case PORT_VLAN_NATIVE_UNTAGGED:
728 case PORT_VLAN_NATIVE_TAGGED:
729 return vid ? vid : in_xbundle->vlan;
736 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
737 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
740 * 'vid' should be the VID obtained from the 802.1Q header that was received as
741 * part of a packet (specify 0 if there was no 802.1Q header), in the range
744 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
746 /* Allow any VID on the OFPP_NONE port. */
747 if (in_xbundle == &ofpp_none_bundle) {
751 switch (in_xbundle->vlan_mode) {
752 case PORT_VLAN_ACCESS:
755 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
756 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
757 "packet received on port %s configured as VLAN "
758 "%"PRIu16" access port", vid, in_xbundle->name,
765 case PORT_VLAN_NATIVE_UNTAGGED:
766 case PORT_VLAN_NATIVE_TAGGED:
768 /* Port must always carry its native VLAN. */
772 case PORT_VLAN_TRUNK:
773 if (!xbundle_includes_vlan(in_xbundle, vid)) {
775 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
776 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
777 "received on port %s not configured for trunking "
778 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
790 /* Given 'vlan', the VLAN that a packet belongs to, and
791 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
792 * that should be included in the 802.1Q header. (If the return value is 0,
793 * then the 802.1Q header should only be included in the packet if there is a
796 * Both 'vlan' and the return value are in the range 0...4095. */
798 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
800 switch (out_xbundle->vlan_mode) {
801 case PORT_VLAN_ACCESS:
804 case PORT_VLAN_TRUNK:
805 case PORT_VLAN_NATIVE_TAGGED:
808 case PORT_VLAN_NATIVE_UNTAGGED:
809 return vlan == out_xbundle->vlan ? 0 : vlan;
817 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
820 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
822 ovs_be16 tci, old_tci;
825 vid = output_vlan_to_vid(out_xbundle, vlan);
826 if (list_is_empty(&out_xbundle->xports)) {
827 /* Partially configured bundle with no slaves. Drop the packet. */
829 } else if (!out_xbundle->bond) {
830 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
833 struct ofport_dpif *ofport;
835 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
836 &ctx->xout->wc, vid);
837 xport = ofport ? xport_lookup(ofport) : NULL;
840 /* No slaves enabled, so drop packet. */
847 if (tci || out_xbundle->use_priority_tags) {
848 tci |= *flow_tci & htons(VLAN_PCP_MASK);
850 tci |= htons(VLAN_CFI);
855 compose_output_action(ctx, xport->ofp_port);
859 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
860 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
861 * indicate this; newer upstream kernels use gratuitous ARP requests. */
863 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
865 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
869 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
870 if (!eth_addr_is_broadcast(flow->dl_dst)) {
874 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
875 if (flow->nw_proto == ARP_OP_REPLY) {
877 } else if (flow->nw_proto == ARP_OP_REQUEST) {
878 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
879 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
881 return flow->nw_src == flow->nw_dst;
888 update_learning_table(const struct xbridge *xbridge,
889 const struct flow *flow, struct flow_wildcards *wc,
890 int vlan, struct xbundle *in_xbundle)
892 struct mac_entry *mac;
894 /* Don't learn the OFPP_NONE port. */
895 if (in_xbundle == &ofpp_none_bundle) {
899 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
900 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
904 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
905 if (is_gratuitous_arp(flow, wc)) {
906 /* We don't want to learn from gratuitous ARP packets that are
907 * reflected back over bond slaves so we lock the learning table. */
908 if (!in_xbundle->bond) {
909 mac_entry_set_grat_arp_lock(mac);
910 } else if (mac_entry_is_grat_arp_locked(mac)) {
915 if (mac->port.p != in_xbundle->ofbundle) {
916 /* The log messages here could actually be useful in debugging,
917 * so keep the rate limit relatively high. */
918 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
919 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
920 "on port %s in VLAN %d",
921 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
922 in_xbundle->name, vlan);
924 mac->port.p = in_xbundle->ofbundle;
925 mac_learning_changed(xbridge->ml);
928 ovs_rwlock_unlock(&xbridge->ml->rwlock);
931 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
932 * dropped. Returns true if they may be forwarded, false if they should be
935 * 'in_port' must be the xport that corresponds to flow->in_port.
936 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
938 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
939 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
940 * checked by input_vid_is_valid().
942 * May also add tags to '*tags', although the current implementation only does
943 * so in one special case.
946 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
949 struct xbundle *in_xbundle = in_port->xbundle;
950 const struct xbridge *xbridge = ctx->xbridge;
951 struct flow *flow = &ctx->xin->flow;
953 /* Drop frames for reserved multicast addresses
954 * only if forward_bpdu option is absent. */
955 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
956 xlate_report(ctx, "packet has reserved destination MAC, dropping");
960 if (in_xbundle->bond) {
961 struct mac_entry *mac;
963 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
969 xlate_report(ctx, "bonding refused admissibility, dropping");
972 case BV_DROP_IF_MOVED:
973 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
974 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
975 if (mac && mac->port.p != in_xbundle->ofbundle &&
976 (!is_gratuitous_arp(flow, &ctx->xout->wc)
977 || mac_entry_is_grat_arp_locked(mac))) {
978 ovs_rwlock_unlock(&xbridge->ml->rwlock);
979 xlate_report(ctx, "SLB bond thinks this packet looped back, "
983 ovs_rwlock_unlock(&xbridge->ml->rwlock);
992 xlate_normal(struct xlate_ctx *ctx)
994 struct flow_wildcards *wc = &ctx->xout->wc;
995 struct flow *flow = &ctx->xin->flow;
996 struct xbundle *in_xbundle;
997 struct xport *in_port;
998 struct mac_entry *mac;
1002 ctx->xout->has_normal = true;
1004 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1005 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1006 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1008 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1009 ctx->xin->packet != NULL, &in_port);
1011 xlate_report(ctx, "no input bundle, dropping");
1015 /* Drop malformed frames. */
1016 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1017 !(flow->vlan_tci & htons(VLAN_CFI))) {
1018 if (ctx->xin->packet != NULL) {
1019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1020 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1021 "VLAN tag received on port %s",
1022 ctx->xbridge->name, in_xbundle->name);
1024 xlate_report(ctx, "partial VLAN tag, dropping");
1028 /* Drop frames on bundles reserved for mirroring. */
1029 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1030 if (ctx->xin->packet != NULL) {
1031 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1032 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1033 "%s, which is reserved exclusively for mirroring",
1034 ctx->xbridge->name, in_xbundle->name);
1036 xlate_report(ctx, "input port is mirror output port, dropping");
1041 vid = vlan_tci_to_vid(flow->vlan_tci);
1042 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1043 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1046 vlan = input_vid_to_vlan(in_xbundle, vid);
1048 /* Check other admissibility requirements. */
1049 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1053 /* Learn source MAC. */
1054 if (ctx->xin->may_learn) {
1055 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1058 /* Determine output bundle. */
1059 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1060 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1062 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
1063 if (mac_xbundle && mac_xbundle != in_xbundle) {
1064 xlate_report(ctx, "forwarding to learned port");
1065 output_normal(ctx, mac_xbundle, vlan);
1066 } else if (!mac_xbundle) {
1067 xlate_report(ctx, "learned port is unknown, dropping");
1069 xlate_report(ctx, "learned port is input port, dropping");
1072 struct xbundle *xbundle;
1074 xlate_report(ctx, "no learned MAC for destination, flooding");
1075 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1076 if (xbundle != in_xbundle
1077 && xbundle_includes_vlan(xbundle, vlan)
1078 && xbundle->floodable
1079 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1080 output_normal(ctx, xbundle, vlan);
1083 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1085 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1088 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1089 * the number of packets out of UINT32_MAX to sample. The given
1090 * cookie is passed back in the callback for each sampled packet.
1093 compose_sample_action(const struct xbridge *xbridge,
1094 struct ofpbuf *odp_actions,
1095 const struct flow *flow,
1096 const uint32_t probability,
1097 const union user_action_cookie *cookie,
1098 const size_t cookie_size)
1100 size_t sample_offset, actions_offset;
1103 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1105 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1107 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1108 cookie_offset = put_userspace_action(xbridge->ofproto, odp_actions, flow,
1109 cookie, cookie_size);
1111 nl_msg_end_nested(odp_actions, actions_offset);
1112 nl_msg_end_nested(odp_actions, sample_offset);
1113 return cookie_offset;
1117 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1118 odp_port_t odp_port, unsigned int n_outputs,
1119 union user_action_cookie *cookie)
1123 cookie->type = USER_ACTION_COOKIE_SFLOW;
1124 cookie->sflow.vlan_tci = vlan_tci;
1126 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1127 * port information") for the interpretation of cookie->output. */
1128 switch (n_outputs) {
1130 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1131 cookie->sflow.output = 0x40000000 | 256;
1135 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1137 cookie->sflow.output = ifindex;
1142 /* 0x80000000 means "multiple output ports. */
1143 cookie->sflow.output = 0x80000000 | n_outputs;
1148 /* Compose SAMPLE action for sFlow bridge sampling. */
1150 compose_sflow_action(const struct xbridge *xbridge,
1151 struct ofpbuf *odp_actions,
1152 const struct flow *flow,
1153 odp_port_t odp_port)
1155 uint32_t probability;
1156 union user_action_cookie cookie;
1158 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1162 probability = dpif_sflow_get_probability(xbridge->sflow);
1163 compose_sflow_cookie(xbridge, htons(0), odp_port,
1164 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1166 return compose_sample_action(xbridge, odp_actions, flow, probability,
1167 &cookie, sizeof cookie.sflow);
1171 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1172 uint32_t obs_domain_id, uint32_t obs_point_id,
1173 union user_action_cookie *cookie)
1175 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1176 cookie->flow_sample.probability = probability;
1177 cookie->flow_sample.collector_set_id = collector_set_id;
1178 cookie->flow_sample.obs_domain_id = obs_domain_id;
1179 cookie->flow_sample.obs_point_id = obs_point_id;
1183 compose_ipfix_cookie(union user_action_cookie *cookie)
1185 cookie->type = USER_ACTION_COOKIE_IPFIX;
1188 /* Compose SAMPLE action for IPFIX bridge sampling. */
1190 compose_ipfix_action(const struct xbridge *xbridge,
1191 struct ofpbuf *odp_actions,
1192 const struct flow *flow)
1194 uint32_t probability;
1195 union user_action_cookie cookie;
1197 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1201 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1202 compose_ipfix_cookie(&cookie);
1204 compose_sample_action(xbridge, odp_actions, flow, probability,
1205 &cookie, sizeof cookie.ipfix);
1208 /* SAMPLE action for sFlow must be first action in any given list of
1209 * actions. At this point we do not have all information required to
1210 * build it. So try to build sample action as complete as possible. */
1212 add_sflow_action(struct xlate_ctx *ctx)
1214 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1215 &ctx->xout->odp_actions,
1216 &ctx->xin->flow, ODPP_NONE);
1217 ctx->sflow_odp_port = 0;
1218 ctx->sflow_n_outputs = 0;
1221 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1222 * of actions, eventually after the SAMPLE action for sFlow. */
1224 add_ipfix_action(struct xlate_ctx *ctx)
1226 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1230 /* Fix SAMPLE action according to data collected while composing ODP actions.
1231 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1232 * USERSPACE action's user-cookie which is required for sflow. */
1234 fix_sflow_action(struct xlate_ctx *ctx)
1236 const struct flow *base = &ctx->base_flow;
1237 union user_action_cookie *cookie;
1239 if (!ctx->user_cookie_offset) {
1243 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1244 sizeof cookie->sflow);
1245 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1247 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1248 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1251 static enum slow_path_reason
1252 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1253 const struct xport *xport, const struct ofpbuf *packet)
1255 struct flow_wildcards *wc = &ctx->xout->wc;
1256 const struct xbridge *xbridge = ctx->xbridge;
1260 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1262 cfm_process_heartbeat(xport->cfm, packet);
1265 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1267 bfd_process_packet(xport->bfd, flow, packet);
1270 } else if (xport->xbundle && xport->xbundle->lacp
1271 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1273 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1276 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1278 stp_process_packet(xport, packet);
1287 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1290 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1291 struct flow_wildcards *wc = &ctx->xout->wc;
1292 struct flow *flow = &ctx->xin->flow;
1293 ovs_be16 flow_vlan_tci;
1294 uint32_t flow_skb_mark;
1295 uint8_t flow_nw_tos;
1296 odp_port_t out_port, odp_port;
1299 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1300 * before traversing a patch port. */
1301 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1304 xlate_report(ctx, "Nonexistent output port");
1306 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1307 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1309 } else if (check_stp && !xport_stp_forward_state(xport)) {
1310 xlate_report(ctx, "STP not in forwarding state, skipping output");
1314 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1315 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1320 const struct xport *peer = xport->peer;
1321 struct flow old_flow = ctx->xin->flow;
1322 enum slow_path_reason special;
1324 ctx->xbridge = peer->xbridge;
1325 flow->in_port.ofp_port = peer->ofp_port;
1326 flow->metadata = htonll(0);
1327 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1328 memset(flow->regs, 0, sizeof flow->regs);
1330 special = process_special(ctx, &ctx->xin->flow, peer,
1333 ctx->xout->slow = special;
1334 } else if (may_receive(peer, ctx)) {
1335 if (xport_stp_forward_state(peer)) {
1336 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1338 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1339 * learning action look at the packet, then drop it. */
1340 struct flow old_base_flow = ctx->base_flow;
1341 size_t old_size = ctx->xout->odp_actions.size;
1342 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1343 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1344 ctx->xout->mirrors = old_mirrors;
1345 ctx->base_flow = old_base_flow;
1346 ctx->xout->odp_actions.size = old_size;
1350 ctx->xin->flow = old_flow;
1351 ctx->xbridge = xport->xbundle->xbridge;
1353 if (ctx->xin->resubmit_stats) {
1354 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1355 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1361 flow_vlan_tci = flow->vlan_tci;
1362 flow_skb_mark = flow->skb_mark;
1363 flow_nw_tos = flow->nw_tos;
1365 if (ofproto_dpif_dscp_from_priority(xport->ofport, flow->skb_priority,
1367 wc->masks.nw_tos |= IP_ECN_MASK;
1368 flow->nw_tos &= ~IP_DSCP_MASK;
1369 flow->nw_tos |= dscp;
1372 if (xport->is_tunnel) {
1373 /* Save tunnel metadata so that changes made due to
1374 * the Logical (tunnel) Port are not visible for any further
1375 * matches, while explicit set actions on tunnel metadata are.
1377 struct flow_tnl flow_tnl = flow->tunnel;
1378 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1379 if (odp_port == ODPP_NONE) {
1380 xlate_report(ctx, "Tunneling decided against output");
1381 goto out; /* restore flow_nw_tos */
1383 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1384 xlate_report(ctx, "Not tunneling to our own address");
1385 goto out; /* restore flow_nw_tos */
1387 if (ctx->xin->resubmit_stats) {
1388 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1390 out_port = odp_port;
1391 commit_odp_tunnel_action(flow, &ctx->base_flow,
1392 &ctx->xout->odp_actions);
1393 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1395 ofp_port_t vlandev_port;
1397 odp_port = xport->odp_port;
1398 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1399 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1401 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1403 if (vlandev_port == ofp_port) {
1404 out_port = odp_port;
1406 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1407 flow->vlan_tci = htons(0);
1409 flow->skb_mark &= ~IPSEC_MARK;
1412 if (out_port != ODPP_NONE) {
1413 commit_odp_actions(flow, &ctx->base_flow,
1414 &ctx->xout->odp_actions, &ctx->xout->wc);
1415 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1418 ctx->sflow_odp_port = odp_port;
1419 ctx->sflow_n_outputs++;
1420 ctx->xout->nf_output_iface = ofp_port;
1425 flow->vlan_tci = flow_vlan_tci;
1426 flow->skb_mark = flow_skb_mark;
1427 flow->nw_tos = flow_nw_tos;
1431 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1433 compose_output_action__(ctx, ofp_port, true);
1436 /* Common rule processing in one place to avoid duplicating code. */
1437 static struct rule_dpif *
1438 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
1441 if (ctx->xin->resubmit_hook) {
1442 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1444 if (rule == NULL && may_packet_in) {
1446 * check if table configuration flags
1447 * OFPTC_TABLE_MISS_CONTROLLER, default.
1448 * OFPTC_TABLE_MISS_CONTINUE,
1449 * OFPTC_TABLE_MISS_DROP
1450 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
1452 rule = rule_dpif_miss_rule(ctx->xbridge->ofproto, &ctx->xin->flow);
1454 if (rule && ctx->xin->resubmit_stats) {
1455 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1461 xlate_table_action(struct xlate_ctx *ctx,
1462 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1464 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1465 struct rule_dpif *rule;
1466 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1467 uint8_t old_table_id = ctx->table_id;
1469 ctx->table_id = table_id;
1471 /* Look up a flow with 'in_port' as the input port. */
1472 ctx->xin->flow.in_port.ofp_port = in_port;
1473 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1474 &ctx->xin->flow, &ctx->xout->wc,
1477 /* Restore the original input port. Otherwise OFPP_NORMAL and
1478 * OFPP_IN_PORT will have surprising behavior. */
1479 ctx->xin->flow.in_port.ofp_port = old_in_port;
1481 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1484 struct rule_dpif *old_rule = ctx->rule;
1488 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1489 ctx->rule = old_rule;
1493 ctx->table_id = old_table_id;
1495 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1497 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1498 MAX_RESUBMIT_RECURSION);
1499 ctx->max_resubmit_trigger = true;
1504 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1505 const struct ofpact_resubmit *resubmit)
1510 in_port = resubmit->in_port;
1511 if (in_port == OFPP_IN_PORT) {
1512 in_port = ctx->xin->flow.in_port.ofp_port;
1515 table_id = resubmit->table_id;
1516 if (table_id == 255) {
1517 table_id = ctx->table_id;
1520 xlate_table_action(ctx, in_port, table_id, false);
1524 flood_packets(struct xlate_ctx *ctx, bool all)
1526 const struct xport *xport;
1528 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1529 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1534 compose_output_action__(ctx, xport->ofp_port, false);
1535 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1536 compose_output_action(ctx, xport->ofp_port);
1540 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1544 execute_controller_action(struct xlate_ctx *ctx, int len,
1545 enum ofp_packet_in_reason reason,
1546 uint16_t controller_id)
1548 struct ofputil_packet_in pin;
1549 struct ofpbuf *packet;
1552 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1553 ctx->xout->slow = SLOW_CONTROLLER;
1554 if (!ctx->xin->packet) {
1558 packet = ofpbuf_clone(ctx->xin->packet);
1560 key.skb_priority = 0;
1562 memset(&key.tunnel, 0, sizeof key.tunnel);
1564 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1565 &ctx->xout->odp_actions, &ctx->xout->wc);
1567 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1568 ctx->xout->odp_actions.size, NULL, NULL);
1570 pin.packet = packet->data;
1571 pin.packet_len = packet->size;
1572 pin.reason = reason;
1573 pin.controller_id = controller_id;
1574 pin.table_id = ctx->table_id;
1575 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1578 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1580 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, &pin);
1581 ofpbuf_delete(packet);
1585 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1587 struct flow_wildcards *wc = &ctx->xout->wc;
1588 struct flow *flow = &ctx->xin->flow;
1590 ovs_assert(eth_type_mpls(eth_type));
1592 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1593 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1595 if (flow->mpls_depth) {
1596 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1602 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1603 label = htonl(0x2); /* IPV6 Explicit Null. */
1605 label = htonl(0x0); /* IPV4 Explicit Null. */
1607 wc->masks.nw_tos |= IP_DSCP_MASK;
1608 wc->masks.nw_ttl = 0xff;
1609 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1610 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1611 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1612 flow->mpls_depth = 1;
1614 flow->dl_type = eth_type;
1618 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1620 struct flow_wildcards *wc = &ctx->xout->wc;
1621 struct flow *flow = &ctx->xin->flow;
1623 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1624 ovs_assert(!eth_type_mpls(eth_type));
1626 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1627 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1629 if (flow->mpls_depth) {
1631 flow->mpls_lse = htonl(0);
1632 if (!flow->mpls_depth) {
1633 flow->dl_type = eth_type;
1639 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1641 struct flow *flow = &ctx->xin->flow;
1643 if (!is_ip_any(flow)) {
1647 ctx->xout->wc.masks.nw_ttl = 0xff;
1648 if (flow->nw_ttl > 1) {
1654 for (i = 0; i < ids->n_controllers; i++) {
1655 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1659 /* Stop processing for current table. */
1665 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1667 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1671 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1676 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1678 struct flow *flow = &ctx->xin->flow;
1679 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1680 struct flow_wildcards *wc = &ctx->xout->wc;
1682 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1684 if (!eth_type_mpls(flow->dl_type)) {
1690 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1693 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1695 /* Stop processing for current table. */
1701 xlate_output_action(struct xlate_ctx *ctx,
1702 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1704 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1706 ctx->xout->nf_output_iface = NF_OUT_DROP;
1710 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1713 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1720 flood_packets(ctx, false);
1723 flood_packets(ctx, true);
1725 case OFPP_CONTROLLER:
1726 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1732 if (port != ctx->xin->flow.in_port.ofp_port) {
1733 compose_output_action(ctx, port);
1735 xlate_report(ctx, "skipping output to input port");
1740 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1741 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1742 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1743 ctx->xout->nf_output_iface = prev_nf_output_iface;
1744 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1745 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1746 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1751 xlate_output_reg_action(struct xlate_ctx *ctx,
1752 const struct ofpact_output_reg *or)
1754 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1755 if (port <= UINT16_MAX) {
1756 union mf_subvalue value;
1758 memset(&value, 0xff, sizeof value);
1759 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1760 xlate_output_action(ctx, u16_to_ofp(port),
1761 or->max_len, false);
1766 xlate_enqueue_action(struct xlate_ctx *ctx,
1767 const struct ofpact_enqueue *enqueue)
1769 ofp_port_t ofp_port = enqueue->port;
1770 uint32_t queue_id = enqueue->queue;
1771 uint32_t flow_priority, priority;
1774 /* Translate queue to priority. */
1775 error = ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1778 /* Fall back to ordinary output action. */
1779 xlate_output_action(ctx, enqueue->port, 0, false);
1783 /* Check output port. */
1784 if (ofp_port == OFPP_IN_PORT) {
1785 ofp_port = ctx->xin->flow.in_port.ofp_port;
1786 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1790 /* Add datapath actions. */
1791 flow_priority = ctx->xin->flow.skb_priority;
1792 ctx->xin->flow.skb_priority = priority;
1793 compose_output_action(ctx, ofp_port);
1794 ctx->xin->flow.skb_priority = flow_priority;
1796 /* Update NetFlow output port. */
1797 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1798 ctx->xout->nf_output_iface = ofp_port;
1799 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1800 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1805 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1807 uint32_t skb_priority;
1809 if (!ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1811 ctx->xin->flow.skb_priority = skb_priority;
1813 /* Couldn't translate queue to a priority. Nothing to do. A warning
1814 * has already been logged. */
1819 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
1821 const struct xbridge *xbridge = xbridge_;
1832 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1835 port = get_ofp_port(xbridge, ofp_port);
1836 return port ? port->may_enable : false;
1841 xlate_bundle_action(struct xlate_ctx *ctx,
1842 const struct ofpact_bundle *bundle)
1846 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1848 CONST_CAST(struct xbridge *, ctx->xbridge));
1849 if (bundle->dst.field) {
1850 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1852 xlate_output_action(ctx, port, 0, false);
1857 xlate_learn_action(struct xlate_ctx *ctx,
1858 const struct ofpact_learn *learn)
1860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1861 struct ofputil_flow_mod fm;
1862 uint64_t ofpacts_stub[1024 / 8];
1863 struct ofpbuf ofpacts;
1866 ctx->xout->has_learn = true;
1868 learn_mask(learn, &ctx->xout->wc);
1870 if (!ctx->xin->may_learn) {
1874 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1875 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1877 error = ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
1878 if (error && !VLOG_DROP_WARN(&rl)) {
1879 VLOG_WARN("learning action failed to modify flow table (%s)",
1880 ofperr_get_name(error));
1883 ofpbuf_uninit(&ofpacts);
1886 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1887 * means "infinite". */
1889 reduce_timeout(uint16_t max, uint16_t *timeout)
1891 if (max && (!*timeout || *timeout > max)) {
1897 xlate_fin_timeout(struct xlate_ctx *ctx,
1898 const struct ofpact_fin_timeout *oft)
1900 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1901 struct rule_dpif *rule = ctx->rule;
1903 if (list_is_empty(&rule->up.expirable)) {
1904 list_insert(&rule->up.ofproto->expirable, &rule->up.expirable);
1907 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1908 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1913 xlate_sample_action(struct xlate_ctx *ctx,
1914 const struct ofpact_sample *os)
1916 union user_action_cookie cookie;
1917 /* Scale the probability from 16-bit to 32-bit while representing
1918 * the same percentage. */
1919 uint32_t probability = (os->probability << 16) | os->probability;
1921 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1922 &ctx->xout->odp_actions, &ctx->xout->wc);
1924 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1925 os->obs_domain_id, os->obs_point_id, &cookie);
1926 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
1927 probability, &cookie, sizeof cookie.flow_sample);
1931 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
1933 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
1934 ? OFPUTIL_PC_NO_RECV_STP
1935 : OFPUTIL_PC_NO_RECV)) {
1939 /* Only drop packets here if both forwarding and learning are
1940 * disabled. If just learning is enabled, we need to have
1941 * OFPP_NORMAL and the learning action have a look at the packet
1942 * before we can drop it. */
1943 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
1951 tunnel_ecn_ok(struct xlate_ctx *ctx)
1953 if (is_ip_any(&ctx->base_flow)
1954 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1955 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1956 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1957 " but is not ECN capable");
1960 /* Set the ECN CE value in the tunneled packet. */
1961 ctx->xin->flow.nw_tos |= IP_ECN_CE;
1969 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
1970 struct xlate_ctx *ctx)
1972 struct flow_wildcards *wc = &ctx->xout->wc;
1973 struct flow *flow = &ctx->xin->flow;
1974 bool was_evictable = true;
1975 const struct ofpact *a;
1978 /* Don't let the rule we're working on get evicted underneath us. */
1979 was_evictable = ctx->rule->up.evictable;
1980 ctx->rule->up.evictable = false;
1983 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
1984 struct ofpact_controller *controller;
1985 const struct ofpact_metadata *metadata;
1993 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
1994 ofpact_get_OUTPUT(a)->max_len, true);
1997 case OFPACT_CONTROLLER:
1998 controller = ofpact_get_CONTROLLER(a);
1999 execute_controller_action(ctx, controller->max_len,
2001 controller->controller_id);
2004 case OFPACT_ENQUEUE:
2005 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2008 case OFPACT_SET_VLAN_VID:
2009 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2010 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2014 case OFPACT_SET_VLAN_PCP:
2015 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2017 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2021 case OFPACT_STRIP_VLAN:
2022 flow->vlan_tci = htons(0);
2025 case OFPACT_PUSH_VLAN:
2026 /* XXX 802.1AD(QinQ) */
2027 flow->vlan_tci = htons(VLAN_CFI);
2030 case OFPACT_SET_ETH_SRC:
2031 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2034 case OFPACT_SET_ETH_DST:
2035 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2038 case OFPACT_SET_IPV4_SRC:
2039 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2040 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2044 case OFPACT_SET_IPV4_DST:
2045 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2046 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2050 case OFPACT_SET_IPV4_DSCP:
2051 /* OpenFlow 1.0 only supports IPv4. */
2052 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2053 flow->nw_tos &= ~IP_DSCP_MASK;
2054 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2058 case OFPACT_SET_L4_SRC_PORT:
2059 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2060 if (is_ip_any(flow)) {
2061 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2065 case OFPACT_SET_L4_DST_PORT:
2066 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2067 if (is_ip_any(flow)) {
2068 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2072 case OFPACT_RESUBMIT:
2073 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2076 case OFPACT_SET_TUNNEL:
2077 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2080 case OFPACT_SET_QUEUE:
2081 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2084 case OFPACT_POP_QUEUE:
2085 flow->skb_priority = ctx->orig_skb_priority;
2088 case OFPACT_REG_MOVE:
2089 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2092 case OFPACT_REG_LOAD:
2093 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2096 case OFPACT_STACK_PUSH:
2097 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2101 case OFPACT_STACK_POP:
2102 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
2105 case OFPACT_PUSH_MPLS:
2106 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2109 case OFPACT_POP_MPLS:
2110 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2113 case OFPACT_SET_MPLS_TTL:
2114 if (compose_set_mpls_ttl_action(ctx,
2115 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2120 case OFPACT_DEC_MPLS_TTL:
2121 if (compose_dec_mpls_ttl_action(ctx)) {
2126 case OFPACT_DEC_TTL:
2127 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2133 /* Nothing to do. */
2136 case OFPACT_MULTIPATH:
2137 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2141 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2144 case OFPACT_OUTPUT_REG:
2145 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2149 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2156 case OFPACT_FIN_TIMEOUT:
2157 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2158 ctx->xout->has_fin_timeout = true;
2159 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2162 case OFPACT_CLEAR_ACTIONS:
2164 * Nothing to do because writa-actions is not supported for now.
2165 * When writa-actions is supported, clear-actions also must
2166 * be supported at the same time.
2170 case OFPACT_WRITE_METADATA:
2171 metadata = ofpact_get_WRITE_METADATA(a);
2172 flow->metadata &= ~metadata->mask;
2173 flow->metadata |= metadata->metadata & metadata->mask;
2177 /* Not implemented yet. */
2180 case OFPACT_GOTO_TABLE: {
2181 /* It is assumed that goto-table is the last action. */
2182 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2184 ovs_assert(ctx->table_id < ogt->table_id);
2185 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2186 ogt->table_id, true);
2191 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2198 ctx->rule->up.evictable = was_evictable;
2203 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2204 const struct flow *flow, struct rule_dpif *rule,
2205 uint8_t tcp_flags, const struct ofpbuf *packet)
2207 xin->ofproto = ofproto;
2209 xin->packet = packet;
2210 xin->may_learn = packet != NULL;
2212 xin->ofpacts = NULL;
2213 xin->ofpacts_len = 0;
2214 xin->tcp_flags = tcp_flags;
2215 xin->resubmit_hook = NULL;
2216 xin->report_hook = NULL;
2217 xin->resubmit_stats = NULL;
2221 xlate_out_uninit(struct xlate_out *xout)
2224 ofpbuf_uninit(&xout->odp_actions);
2228 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2229 * into datapath actions, using 'ctx', and discards the datapath actions. */
2231 xlate_actions_for_side_effects(struct xlate_in *xin)
2233 struct xlate_out xout;
2235 xlate_actions(xin, &xout);
2236 xlate_out_uninit(&xout);
2240 xlate_report(struct xlate_ctx *ctx, const char *s)
2242 if (ctx->xin->report_hook) {
2243 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2248 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2251 dst->slow = src->slow;
2252 dst->has_learn = src->has_learn;
2253 dst->has_normal = src->has_normal;
2254 dst->has_fin_timeout = src->has_fin_timeout;
2255 dst->nf_output_iface = src->nf_output_iface;
2256 dst->mirrors = src->mirrors;
2258 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2259 sizeof dst->odp_actions_stub);
2260 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2261 src->odp_actions.size);
2265 actions_output_to_local_port(const struct xlate_ctx *ctx)
2267 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2268 const struct nlattr *a;
2271 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2272 ctx->xout->odp_actions.size) {
2273 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2274 && nl_attr_get_odp_port(a) == local_odp_port) {
2281 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2282 * into datapath actions in 'odp_actions', using 'ctx'. */
2284 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2286 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
2287 * that in the future we always keep a copy of the original flow for
2288 * tracing purposes. */
2289 static bool hit_resubmit_limit;
2291 struct flow_wildcards *wc = &xout->wc;
2292 struct flow *flow = &xin->flow;
2294 enum slow_path_reason special;
2295 const struct ofpact *ofpacts;
2296 struct xport *in_port;
2297 struct flow orig_flow;
2298 struct xlate_ctx ctx;
2301 COVERAGE_INC(xlate_actions);
2303 /* Flow initialization rules:
2304 * - 'base_flow' must match the kernel's view of the packet at the
2305 * time that action processing starts. 'flow' represents any
2306 * transformations we wish to make through actions.
2307 * - By default 'base_flow' and 'flow' are the same since the input
2308 * packet matches the output before any actions are applied.
2309 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2310 * of the received packet as seen by the kernel. If we later output
2311 * to another device without any modifications this will cause us to
2312 * insert a new tag since the original one was stripped off by the
2314 * - Tunnel metadata as received is retained in 'flow'. This allows
2315 * tunnel metadata matching also in later tables.
2316 * Since a kernel action for setting the tunnel metadata will only be
2317 * generated with actual tunnel output, changing the tunnel metadata
2318 * values in 'flow' (such as tun_id) will only have effect with a later
2319 * tunnel output action.
2320 * - Tunnel 'base_flow' is completely cleared since that is what the
2321 * kernel does. If we wish to maintain the original values an action
2322 * needs to be generated. */
2327 ctx.xout->has_learn = false;
2328 ctx.xout->has_normal = false;
2329 ctx.xout->has_fin_timeout = false;
2330 ctx.xout->nf_output_iface = NF_OUT_DROP;
2331 ctx.xout->mirrors = 0;
2332 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2333 sizeof ctx.xout->odp_actions_stub);
2334 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2336 ctx.xbridge = xbridge_lookup(xin->ofproto);
2341 ctx.rule = xin->rule;
2343 ctx.base_flow = *flow;
2344 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2345 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2347 flow_wildcards_init_catchall(wc);
2348 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2349 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2350 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2351 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2353 if (tnl_port_should_receive(&ctx.xin->flow)) {
2354 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
2355 /* skb_mark is currently used only by tunnels but that will likely
2356 * change in the future. */
2357 memset(&wc->masks.skb_mark, 0xff, sizeof wc->masks.skb_mark);
2359 if (ctx.xbridge->has_netflow) {
2360 netflow_mask_wc(flow, wc);
2364 ctx.max_resubmit_trigger = false;
2365 ctx.orig_skb_priority = flow->skb_priority;
2370 ofpacts = xin->ofpacts;
2371 ofpacts_len = xin->ofpacts_len;
2372 } else if (xin->rule) {
2373 ofpacts = xin->rule->up.ofpacts;
2374 ofpacts_len = xin->rule->up.ofpacts_len;
2379 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2381 if (mbridge_has_mirrors(ctx.xbridge->mbridge) || hit_resubmit_limit) {
2382 /* Do this conditionally because the copy is expensive enough that it
2383 * shows up in profiles. */
2387 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2388 switch (ctx.xbridge->frag) {
2389 case OFPC_FRAG_NORMAL:
2390 /* We must pretend that transport ports are unavailable. */
2391 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2392 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2395 case OFPC_FRAG_DROP:
2398 case OFPC_FRAG_REASM:
2401 case OFPC_FRAG_NX_MATCH:
2402 /* Nothing to do. */
2405 case OFPC_INVALID_TTL_TO_CONTROLLER:
2410 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2411 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2413 ctx.xout->slow = special;
2415 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2416 size_t sample_actions_len;
2418 if (flow->in_port.ofp_port
2419 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2420 flow->in_port.ofp_port,
2422 ctx.base_flow.vlan_tci = 0;
2425 add_sflow_action(&ctx);
2426 add_ipfix_action(&ctx);
2427 sample_actions_len = ctx.xout->odp_actions.size;
2429 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
2430 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2432 /* We've let OFPP_NORMAL and the learning action look at the
2433 * packet, so drop it now if forwarding is disabled. */
2434 if (in_port && !xport_stp_forward_state(in_port)) {
2435 ctx.xout->odp_actions.size = sample_actions_len;
2439 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
2440 if (!hit_resubmit_limit) {
2441 /* We didn't record the original flow. Make sure we do from
2443 hit_resubmit_limit = true;
2444 } else if (!VLOG_DROP_ERR(&trace_rl)) {
2445 struct ds ds = DS_EMPTY_INITIALIZER;
2447 ofproto_trace(ctx.xbridge->ofproto, &orig_flow,
2448 ctx.xin->packet, &ds);
2449 VLOG_ERR("Trace triggered by excessive resubmit "
2450 "recursion:\n%s", ds_cstr(&ds));
2455 if (ctx.xbridge->has_in_band
2456 && in_band_must_output_to_local_port(flow)
2457 && !actions_output_to_local_port(&ctx)) {
2458 compose_output_action(&ctx, OFPP_LOCAL);
2461 fix_sflow_action(&ctx);
2463 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2464 add_mirror_actions(&ctx, &orig_flow);
2468 ofpbuf_uninit(&ctx.stack);
2470 /* Clear the metadata and register wildcard masks, because we won't
2471 * use non-header fields as part of the cache. */
2472 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2473 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);