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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
41 #include "odp-execute.h"
42 #include "ofp-actions.h"
43 #include "ofproto/ofproto-dpif-ipfix.h"
44 #include "ofproto/ofproto-dpif-mirror.h"
45 #include "ofproto/ofproto-dpif-sflow.h"
46 #include "ofproto/ofproto-dpif.h"
47 #include "ofproto/ofproto-provider.h"
51 COVERAGE_DEFINE(xlate_actions);
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
55 /* Maximum depth of flow table recursion (due to resubmit actions) in a
56 * flow translation. */
57 #define MAX_RESUBMIT_RECURSION 64
59 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
62 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
63 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
65 struct list xbundles; /* Owned xbundles. */
66 struct hmap xports; /* Indexed by ofp_port. */
68 char *name; /* Name used in log messages. */
69 struct dpif *dpif; /* Datapath interface. */
70 struct mac_learning *ml; /* Mac learning handle. */
71 struct mbridge *mbridge; /* Mirroring. */
72 struct dpif_sflow *sflow; /* SFlow handle, or null. */
73 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
74 struct stp *stp; /* STP or null if disabled. */
76 /* Special rules installed by ofproto-dpif. */
77 struct rule_dpif *miss_rule;
78 struct rule_dpif *no_packet_in_rule;
80 enum ofp_config_flags frag; /* Fragmentation handling. */
81 bool has_netflow; /* Bridge runs netflow? */
82 bool has_in_band; /* Bridge has in band control? */
83 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
87 struct hmap_node hmap_node; /* In global 'xbundles' map. */
88 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
90 struct list list_node; /* In parent 'xbridges' list. */
91 struct xbridge *xbridge; /* Parent xbridge. */
93 struct list xports; /* Contains "struct xport"s. */
95 char *name; /* Name used in log messages. */
96 struct bond *bond; /* Nonnull iff more than one port. */
97 struct lacp *lacp; /* LACP handle or null. */
99 enum port_vlan_mode vlan_mode; /* VLAN mode. */
100 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
101 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
102 * NULL if all VLANs are trunked. */
103 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
104 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
108 struct hmap_node hmap_node; /* Node in global 'xports' map. */
109 struct ofport_dpif *ofport; /* Key in global 'xports map. */
111 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
112 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
114 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
116 struct list bundle_node; /* In parent xbundle (if it exists). */
117 struct xbundle *xbundle; /* Parent xbundle or null. */
119 struct netdev *netdev; /* 'ofport''s netdev. */
121 struct xbridge *xbridge; /* Parent bridge. */
122 struct xport *peer; /* Patch port peer or null. */
124 enum ofputil_port_config config; /* OpenFlow port configuration. */
125 int stp_port_no; /* STP port number or -1 if not in use. */
127 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
129 bool may_enable; /* May be enabled in bonds. */
130 bool is_tunnel; /* Is a tunnel port. */
132 struct cfm *cfm; /* CFM handle or null. */
133 struct bfd *bfd; /* BFD handle or null. */
137 struct xlate_in *xin;
138 struct xlate_out *xout;
140 const struct xbridge *xbridge;
142 /* Flow at the last commit. */
143 struct flow base_flow;
145 /* Tunnel IP destination address as received. This is stored separately
146 * as the base_flow.tunnel is cleared on init to reflect the datapath
147 * behavior. Used to make sure not to send tunneled output to ourselves,
148 * which might lead to an infinite loop. This could happen easily
149 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
150 * actually set the tun_dst field. */
151 ovs_be32 orig_tunnel_ip_dst;
153 /* Stack for the push and pop actions. Each stack element is of type
154 * "union mf_subvalue". */
155 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
158 /* The rule that we are currently translating, or NULL. */
159 struct rule_dpif *rule;
161 int recurse; /* Recursion level, via xlate_table_action. */
162 uint32_t orig_skb_priority; /* Priority when packet arrived. */
163 uint8_t table_id; /* OpenFlow table ID where flow was found. */
164 uint32_t sflow_n_outputs; /* Number of output ports. */
165 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
166 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
167 bool exit; /* No further actions should be processed. */
170 /* A controller may use OFPP_NONE as the ingress port to indicate that
171 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
172 * when an input bundle is needed for validation (e.g., mirroring or
173 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
174 * any 'port' structs, so care must be taken when dealing with it.
175 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
176 static struct xbundle ofpp_none_bundle;
178 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
179 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
180 * traffic egressing the 'ofport' with that priority should be marked with. */
181 struct skb_priority_to_dscp {
182 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
183 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
185 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
188 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
189 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
190 static struct hmap xports = HMAP_INITIALIZER(&xports);
192 static bool may_receive(const struct xport *, struct xlate_ctx *);
193 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
195 static void xlate_normal(struct xlate_ctx *);
196 static void xlate_report(struct xlate_ctx *, const char *);
197 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
198 uint8_t table_id, bool may_packet_in);
199 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
200 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
201 static void output_normal(struct xlate_ctx *, const struct xbundle *,
203 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
205 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
206 static struct xbundle *xbundle_lookup(const struct ofbundle *);
207 static struct xport *xport_lookup(const struct ofport_dpif *);
208 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
209 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
210 uint32_t skb_priority);
211 static void clear_skb_priorities(struct xport *);
212 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
216 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
217 struct dpif *dpif, struct rule_dpif *miss_rule,
218 struct rule_dpif *no_packet_in_rule,
219 const struct mac_learning *ml, struct stp *stp,
220 const struct mbridge *mbridge,
221 const struct dpif_sflow *sflow,
222 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
223 bool forward_bpdu, bool has_in_band, bool has_netflow)
225 struct xbridge *xbridge = xbridge_lookup(ofproto);
228 xbridge = xzalloc(sizeof *xbridge);
229 xbridge->ofproto = ofproto;
231 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
232 hmap_init(&xbridge->xports);
233 list_init(&xbridge->xbundles);
236 if (xbridge->ml != ml) {
237 mac_learning_unref(xbridge->ml);
238 xbridge->ml = mac_learning_ref(ml);
241 if (xbridge->mbridge != mbridge) {
242 mbridge_unref(xbridge->mbridge);
243 xbridge->mbridge = mbridge_ref(mbridge);
246 if (xbridge->sflow != sflow) {
247 dpif_sflow_unref(xbridge->sflow);
248 xbridge->sflow = dpif_sflow_ref(sflow);
251 if (xbridge->ipfix != ipfix) {
252 dpif_ipfix_unref(xbridge->ipfix);
253 xbridge->ipfix = dpif_ipfix_ref(ipfix);
256 if (xbridge->stp != stp) {
257 stp_unref(xbridge->stp);
258 xbridge->stp = stp_ref(stp);
262 xbridge->name = xstrdup(name);
264 xbridge->dpif = dpif;
265 xbridge->forward_bpdu = forward_bpdu;
266 xbridge->has_in_band = has_in_band;
267 xbridge->has_netflow = has_netflow;
268 xbridge->frag = frag;
269 xbridge->miss_rule = miss_rule;
270 xbridge->no_packet_in_rule = no_packet_in_rule;
274 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
276 struct xbridge *xbridge = xbridge_lookup(ofproto);
277 struct xbundle *xbundle, *next_xbundle;
278 struct xport *xport, *next_xport;
284 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
285 xlate_ofport_remove(xport->ofport);
288 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
289 xlate_bundle_remove(xbundle->ofbundle);
292 hmap_remove(&xbridges, &xbridge->hmap_node);
293 mac_learning_unref(xbridge->ml);
294 mbridge_unref(xbridge->mbridge);
295 dpif_sflow_unref(xbridge->sflow);
296 dpif_ipfix_unref(xbridge->ipfix);
297 stp_unref(xbridge->stp);
298 hmap_destroy(&xbridge->xports);
304 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
305 const char *name, enum port_vlan_mode vlan_mode, int vlan,
306 unsigned long *trunks, bool use_priority_tags,
307 const struct bond *bond, const struct lacp *lacp,
310 struct xbundle *xbundle = xbundle_lookup(ofbundle);
313 xbundle = xzalloc(sizeof *xbundle);
314 xbundle->ofbundle = ofbundle;
315 xbundle->xbridge = xbridge_lookup(ofproto);
317 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
318 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
319 list_init(&xbundle->xports);
322 ovs_assert(xbundle->xbridge);
325 xbundle->name = xstrdup(name);
327 xbundle->vlan_mode = vlan_mode;
328 xbundle->vlan = vlan;
329 xbundle->trunks = trunks;
330 xbundle->use_priority_tags = use_priority_tags;
331 xbundle->floodable = floodable;
333 if (xbundle->bond != bond) {
334 bond_unref(xbundle->bond);
335 xbundle->bond = bond_ref(bond);
338 if (xbundle->lacp != lacp) {
339 lacp_unref(xbundle->lacp);
340 xbundle->lacp = lacp_ref(lacp);
345 xlate_bundle_remove(struct ofbundle *ofbundle)
347 struct xbundle *xbundle = xbundle_lookup(ofbundle);
348 struct xport *xport, *next;
354 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
355 list_remove(&xport->bundle_node);
356 xport->xbundle = NULL;
359 hmap_remove(&xbundles, &xbundle->hmap_node);
360 list_remove(&xbundle->list_node);
361 bond_unref(xbundle->bond);
362 lacp_unref(xbundle->lacp);
368 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
369 struct ofport_dpif *ofport, ofp_port_t ofp_port,
370 odp_port_t odp_port, const struct netdev *netdev,
371 const struct cfm *cfm, const struct bfd *bfd,
372 struct ofport_dpif *peer, int stp_port_no,
373 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
374 enum ofputil_port_config config, bool is_tunnel,
377 struct xport *xport = xport_lookup(ofport);
381 xport = xzalloc(sizeof *xport);
382 xport->ofport = ofport;
383 xport->xbridge = xbridge_lookup(ofproto);
384 xport->ofp_port = ofp_port;
386 hmap_init(&xport->skb_priorities);
387 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
388 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
389 hash_ofp_port(xport->ofp_port));
392 ovs_assert(xport->ofp_port == ofp_port);
394 xport->config = config;
395 xport->stp_port_no = stp_port_no;
396 xport->is_tunnel = is_tunnel;
397 xport->may_enable = may_enable;
398 xport->odp_port = odp_port;
400 if (xport->netdev != netdev) {
401 netdev_close(xport->netdev);
402 xport->netdev = netdev_ref(netdev);
405 if (xport->cfm != cfm) {
406 cfm_unref(xport->cfm);
407 xport->cfm = cfm_ref(cfm);
410 if (xport->bfd != bfd) {
411 bfd_unref(xport->bfd);
412 xport->bfd = bfd_ref(bfd);
416 xport->peer->peer = NULL;
418 xport->peer = xport_lookup(peer);
420 xport->peer->peer = xport;
423 if (xport->xbundle) {
424 list_remove(&xport->bundle_node);
426 xport->xbundle = xbundle_lookup(ofbundle);
427 if (xport->xbundle) {
428 list_insert(&xport->xbundle->xports, &xport->bundle_node);
431 clear_skb_priorities(xport);
432 for (i = 0; i < n_qdscp; i++) {
433 struct skb_priority_to_dscp *pdscp;
434 uint32_t skb_priority;
436 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
441 pdscp = xmalloc(sizeof *pdscp);
442 pdscp->skb_priority = skb_priority;
443 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
444 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
445 hash_int(pdscp->skb_priority, 0));
450 xlate_ofport_remove(struct ofport_dpif *ofport)
452 struct xport *xport = xport_lookup(ofport);
459 xport->peer->peer = NULL;
463 if (xport->xbundle) {
464 list_remove(&xport->bundle_node);
467 clear_skb_priorities(xport);
468 hmap_destroy(&xport->skb_priorities);
470 hmap_remove(&xports, &xport->hmap_node);
471 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
473 netdev_close(xport->netdev);
474 cfm_unref(xport->cfm);
475 bfd_unref(xport->bfd);
479 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
480 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
481 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
482 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
483 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
484 * 'packet' ingressed.
486 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
487 * 'flow''s in_port to OFPP_NONE.
489 * This function does post-processing on data returned from
490 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
491 * of the upcall processing logic. In particular, if the extracted in_port is
492 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
493 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
494 * a VLAN header onto 'packet' (if it is nonnull).
496 * Similarly, this function also includes some logic to help with tunnels. It
497 * may modify 'flow' as necessary to make the tunneling implementation
498 * transparent to the upcall processing logic.
500 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
501 * or some other positive errno if there are other problems. */
503 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
504 const struct nlattr *key, size_t key_len,
505 struct flow *flow, enum odp_key_fitness *fitnessp,
506 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
508 enum odp_key_fitness fitness;
509 const struct xport *xport;
512 ovs_rwlock_rdlock(&xlate_rwlock);
513 fitness = odp_flow_key_to_flow(key, key_len, flow);
514 if (fitness == ODP_FIT_ERROR) {
520 *odp_in_port = flow->in_port.odp_port;
523 xport = xport_lookup(tnl_port_should_receive(flow)
524 ? tnl_port_receive(flow)
525 : odp_port_to_ofport(backer, flow->in_port.odp_port));
527 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
532 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
534 /* Make the packet resemble the flow, so that it gets sent to
535 * an OpenFlow controller properly, so that it looks correct
536 * for sFlow, and so that flow_extract() will get the correct
537 * vlan_tci if it is called on 'packet'.
539 * The allocated space inside 'packet' probably also contains
540 * 'key', that is, both 'packet' and 'key' are probably part of
541 * a struct dpif_upcall (see the large comment on that
542 * structure definition), so pushing data on 'packet' is in
543 * general not a good idea since it could overwrite 'key' or
544 * free it as a side effect. However, it's OK in this special
545 * case because we know that 'packet' is inside a Netlink
546 * attribute: pushing 4 bytes will just overwrite the 4-byte
547 * "struct nlattr", which is fine since we don't need that
549 eth_push_vlan(packet, flow->vlan_tci);
551 /* We can't reproduce 'key' from 'flow'. */
552 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
557 *ofproto = xport->xbridge->ofproto;
564 ovs_rwlock_unlock(&xlate_rwlock);
568 static struct xbridge *
569 xbridge_lookup(const struct ofproto_dpif *ofproto)
571 struct xbridge *xbridge;
577 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
579 if (xbridge->ofproto == ofproto) {
586 static struct xbundle *
587 xbundle_lookup(const struct ofbundle *ofbundle)
589 struct xbundle *xbundle;
595 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
597 if (xbundle->ofbundle == ofbundle) {
604 static struct xport *
605 xport_lookup(const struct ofport_dpif *ofport)
613 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
615 if (xport->ofport == ofport) {
622 static struct stp_port *
623 xport_get_stp_port(const struct xport *xport)
625 return xport->xbridge->stp && xport->stp_port_no != -1
626 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
630 static enum stp_state
631 xport_stp_learn_state(const struct xport *xport)
633 struct stp_port *sp = xport_get_stp_port(xport);
634 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
638 xport_stp_forward_state(const struct xport *xport)
640 struct stp_port *sp = xport_get_stp_port(xport);
641 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
644 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
645 * were used to make the determination.*/
647 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
649 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
650 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
654 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
656 struct stp_port *sp = xport_get_stp_port(xport);
657 struct ofpbuf payload = *packet;
658 struct eth_header *eth = payload.data;
660 /* Sink packets on ports that have STP disabled when the bridge has
662 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
666 /* Trim off padding on payload. */
667 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
668 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
671 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
672 stp_received_bpdu(sp, payload.data, payload.size);
676 static struct xport *
677 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
681 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
683 if (xport->ofp_port == ofp_port) {
691 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
693 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
694 return xport ? xport->odp_port : ODPP_NONE;
698 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
700 return (bundle->vlan_mode != PORT_VLAN_ACCESS
701 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
705 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
707 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
711 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
713 return xbundle != &ofpp_none_bundle
714 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
719 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
721 return xbundle != &ofpp_none_bundle
722 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
727 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
729 return xbundle != &ofpp_none_bundle
730 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
734 static struct xbundle *
735 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
736 bool warn, struct xport **in_xportp)
740 /* Find the port and bundle for the received packet. */
741 xport = get_ofp_port(xbridge, in_port);
745 if (xport && xport->xbundle) {
746 return xport->xbundle;
749 /* Special-case OFPP_NONE, which a controller may use as the ingress
750 * port for traffic that it is sourcing. */
751 if (in_port == OFPP_NONE) {
752 ofpp_none_bundle.name = "OFPP_NONE";
753 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
754 return &ofpp_none_bundle;
757 /* Odd. A few possible reasons here:
759 * - We deleted a port but there are still a few packets queued up
762 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
763 * we don't know about.
765 * - The ofproto client didn't configure the port as part of a bundle.
766 * This is particularly likely to happen if a packet was received on the
767 * port after it was created, but before the client had a chance to
768 * configure its bundle.
771 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
773 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
774 "port %"PRIu16, xbridge->name, in_port);
780 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
782 const struct xbridge *xbridge = ctx->xbridge;
783 mirror_mask_t mirrors;
784 struct xbundle *in_xbundle;
788 mirrors = ctx->xout->mirrors;
789 ctx->xout->mirrors = 0;
791 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
792 ctx->xin->packet != NULL, NULL);
796 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
798 /* Drop frames on bundles reserved for mirroring. */
799 if (xbundle_mirror_out(xbridge, in_xbundle)) {
800 if (ctx->xin->packet != NULL) {
801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
802 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
803 "%s, which is reserved exclusively for mirroring",
804 ctx->xbridge->name, in_xbundle->name);
806 ofpbuf_clear(&ctx->xout->odp_actions);
811 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
812 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
815 vlan = input_vid_to_vlan(in_xbundle, vid);
821 /* Restore the original packet before adding the mirror actions. */
822 ctx->xin->flow = *orig_flow;
825 mirror_mask_t dup_mirrors;
826 struct ofbundle *out;
827 unsigned long *vlans;
832 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
833 &vlans, &dup_mirrors, &out, &out_vlan);
834 ovs_assert(has_mirror);
837 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
839 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
842 if (!vlan_mirrored) {
843 mirrors = zero_rightmost_1bit(mirrors);
847 mirrors &= ~dup_mirrors;
848 ctx->xout->mirrors |= dup_mirrors;
850 struct xbundle *out_xbundle = xbundle_lookup(out);
852 output_normal(ctx, out_xbundle, vlan);
854 } else if (vlan != out_vlan
855 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
856 struct xbundle *xbundle;
858 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
859 if (xbundle_includes_vlan(xbundle, out_vlan)
860 && !xbundle_mirror_out(xbridge, xbundle)) {
861 output_normal(ctx, xbundle, out_vlan);
868 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
869 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
870 * the bundle on which the packet was received, returns the VLAN to which the
873 * Both 'vid' and the return value are in the range 0...4095. */
875 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
877 switch (in_xbundle->vlan_mode) {
878 case PORT_VLAN_ACCESS:
879 return in_xbundle->vlan;
882 case PORT_VLAN_TRUNK:
885 case PORT_VLAN_NATIVE_UNTAGGED:
886 case PORT_VLAN_NATIVE_TAGGED:
887 return vid ? vid : in_xbundle->vlan;
894 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
895 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
898 * 'vid' should be the VID obtained from the 802.1Q header that was received as
899 * part of a packet (specify 0 if there was no 802.1Q header), in the range
902 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
904 /* Allow any VID on the OFPP_NONE port. */
905 if (in_xbundle == &ofpp_none_bundle) {
909 switch (in_xbundle->vlan_mode) {
910 case PORT_VLAN_ACCESS:
913 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
914 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
915 "packet received on port %s configured as VLAN "
916 "%"PRIu16" access port", vid, in_xbundle->name,
923 case PORT_VLAN_NATIVE_UNTAGGED:
924 case PORT_VLAN_NATIVE_TAGGED:
926 /* Port must always carry its native VLAN. */
930 case PORT_VLAN_TRUNK:
931 if (!xbundle_includes_vlan(in_xbundle, vid)) {
933 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
934 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
935 "received on port %s not configured for trunking "
936 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
948 /* Given 'vlan', the VLAN that a packet belongs to, and
949 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
950 * that should be included in the 802.1Q header. (If the return value is 0,
951 * then the 802.1Q header should only be included in the packet if there is a
954 * Both 'vlan' and the return value are in the range 0...4095. */
956 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
958 switch (out_xbundle->vlan_mode) {
959 case PORT_VLAN_ACCESS:
962 case PORT_VLAN_TRUNK:
963 case PORT_VLAN_NATIVE_TAGGED:
966 case PORT_VLAN_NATIVE_UNTAGGED:
967 return vlan == out_xbundle->vlan ? 0 : vlan;
975 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
978 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
980 ovs_be16 tci, old_tci;
983 vid = output_vlan_to_vid(out_xbundle, vlan);
984 if (list_is_empty(&out_xbundle->xports)) {
985 /* Partially configured bundle with no slaves. Drop the packet. */
987 } else if (!out_xbundle->bond) {
988 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
991 struct ofport_dpif *ofport;
993 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
994 &ctx->xout->wc, vid);
995 xport = xport_lookup(ofport);
998 /* No slaves enabled, so drop packet. */
1003 old_tci = *flow_tci;
1005 if (tci || out_xbundle->use_priority_tags) {
1006 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1008 tci |= htons(VLAN_CFI);
1013 compose_output_action(ctx, xport->ofp_port);
1014 *flow_tci = old_tci;
1017 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1018 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1019 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1021 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1023 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1027 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1028 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1032 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1033 if (flow->nw_proto == ARP_OP_REPLY) {
1035 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1036 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1037 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1039 return flow->nw_src == flow->nw_dst;
1045 /* Checks whether a MAC learning update is necessary for MAC learning table
1046 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1049 * Most packets processed through the MAC learning table do not actually
1050 * change it in any way. This function requires only a read lock on the MAC
1051 * learning table, so it is much cheaper in this common case.
1053 * Keep the code here synchronized with that in update_learning_table__()
1056 is_mac_learning_update_needed(const struct mac_learning *ml,
1057 const struct flow *flow,
1058 struct flow_wildcards *wc,
1059 int vlan, struct xbundle *in_xbundle)
1060 OVS_REQ_RDLOCK(ml->rwlock)
1062 struct mac_entry *mac;
1064 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1068 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1069 if (!mac || mac_entry_age(ml, mac)) {
1073 if (is_gratuitous_arp(flow, wc)) {
1074 /* We don't want to learn from gratuitous ARP packets that are
1075 * reflected back over bond slaves so we lock the learning table. */
1076 if (!in_xbundle->bond) {
1078 } else if (mac_entry_is_grat_arp_locked(mac)) {
1083 return mac->port.p != in_xbundle->ofbundle;
1087 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1088 * received on 'in_xbundle' in 'vlan'.
1090 * This code repeats all the checks in is_mac_learning_update_needed() because
1091 * the lock was released between there and here and thus the MAC learning state
1092 * could have changed.
1094 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1097 update_learning_table__(const struct xbridge *xbridge,
1098 const struct flow *flow, struct flow_wildcards *wc,
1099 int vlan, struct xbundle *in_xbundle)
1100 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1102 struct mac_entry *mac;
1104 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1108 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1109 if (is_gratuitous_arp(flow, wc)) {
1110 /* We don't want to learn from gratuitous ARP packets that are
1111 * reflected back over bond slaves so we lock the learning table. */
1112 if (!in_xbundle->bond) {
1113 mac_entry_set_grat_arp_lock(mac);
1114 } else if (mac_entry_is_grat_arp_locked(mac)) {
1119 if (mac->port.p != in_xbundle->ofbundle) {
1120 /* The log messages here could actually be useful in debugging,
1121 * so keep the rate limit relatively high. */
1122 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1124 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1125 "on port %s in VLAN %d",
1126 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1127 in_xbundle->name, vlan);
1129 mac->port.p = in_xbundle->ofbundle;
1130 mac_learning_changed(xbridge->ml);
1135 update_learning_table(const struct xbridge *xbridge,
1136 const struct flow *flow, struct flow_wildcards *wc,
1137 int vlan, struct xbundle *in_xbundle)
1141 /* Don't learn the OFPP_NONE port. */
1142 if (in_xbundle == &ofpp_none_bundle) {
1146 /* First try the common case: no change to MAC learning table. */
1147 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1148 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1150 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1153 /* Slow path: MAC learning table might need an update. */
1154 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1155 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1156 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1160 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1161 * dropped. Returns true if they may be forwarded, false if they should be
1164 * 'in_port' must be the xport that corresponds to flow->in_port.
1165 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1167 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1168 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1169 * checked by input_vid_is_valid().
1171 * May also add tags to '*tags', although the current implementation only does
1172 * so in one special case.
1175 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1178 struct xbundle *in_xbundle = in_port->xbundle;
1179 const struct xbridge *xbridge = ctx->xbridge;
1180 struct flow *flow = &ctx->xin->flow;
1182 /* Drop frames for reserved multicast addresses
1183 * only if forward_bpdu option is absent. */
1184 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1185 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1189 if (in_xbundle->bond) {
1190 struct mac_entry *mac;
1192 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1198 xlate_report(ctx, "bonding refused admissibility, dropping");
1201 case BV_DROP_IF_MOVED:
1202 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1203 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1204 if (mac && mac->port.p != in_xbundle->ofbundle &&
1205 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1206 || mac_entry_is_grat_arp_locked(mac))) {
1207 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1208 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1212 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1221 xlate_normal(struct xlate_ctx *ctx)
1223 struct flow_wildcards *wc = &ctx->xout->wc;
1224 struct flow *flow = &ctx->xin->flow;
1225 struct xbundle *in_xbundle;
1226 struct xport *in_port;
1227 struct mac_entry *mac;
1232 ctx->xout->has_normal = true;
1234 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1235 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1236 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1238 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1239 ctx->xin->packet != NULL, &in_port);
1241 xlate_report(ctx, "no input bundle, dropping");
1245 /* Drop malformed frames. */
1246 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1247 !(flow->vlan_tci & htons(VLAN_CFI))) {
1248 if (ctx->xin->packet != NULL) {
1249 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1250 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1251 "VLAN tag received on port %s",
1252 ctx->xbridge->name, in_xbundle->name);
1254 xlate_report(ctx, "partial VLAN tag, dropping");
1258 /* Drop frames on bundles reserved for mirroring. */
1259 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1260 if (ctx->xin->packet != NULL) {
1261 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1262 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1263 "%s, which is reserved exclusively for mirroring",
1264 ctx->xbridge->name, in_xbundle->name);
1266 xlate_report(ctx, "input port is mirror output port, dropping");
1271 vid = vlan_tci_to_vid(flow->vlan_tci);
1272 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1273 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1276 vlan = input_vid_to_vlan(in_xbundle, vid);
1278 /* Check other admissibility requirements. */
1279 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1283 /* Learn source MAC. */
1284 if (ctx->xin->may_learn) {
1285 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1288 /* Determine output bundle. */
1289 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1290 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1291 mac_port = mac ? mac->port.p : NULL;
1292 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1295 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1296 if (mac_xbundle && mac_xbundle != in_xbundle) {
1297 xlate_report(ctx, "forwarding to learned port");
1298 output_normal(ctx, mac_xbundle, vlan);
1299 } else if (!mac_xbundle) {
1300 xlate_report(ctx, "learned port is unknown, dropping");
1302 xlate_report(ctx, "learned port is input port, dropping");
1305 struct xbundle *xbundle;
1307 xlate_report(ctx, "no learned MAC for destination, flooding");
1308 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1309 if (xbundle != in_xbundle
1310 && xbundle_includes_vlan(xbundle, vlan)
1311 && xbundle->floodable
1312 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1313 output_normal(ctx, xbundle, vlan);
1316 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1320 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1321 * the number of packets out of UINT32_MAX to sample. The given
1322 * cookie is passed back in the callback for each sampled packet.
1325 compose_sample_action(const struct xbridge *xbridge,
1326 struct ofpbuf *odp_actions,
1327 const struct flow *flow,
1328 const uint32_t probability,
1329 const union user_action_cookie *cookie,
1330 const size_t cookie_size)
1332 size_t sample_offset, actions_offset;
1333 odp_port_t odp_port;
1337 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1339 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1341 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1343 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1344 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1345 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1347 nl_msg_end_nested(odp_actions, actions_offset);
1348 nl_msg_end_nested(odp_actions, sample_offset);
1349 return cookie_offset;
1353 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1354 odp_port_t odp_port, unsigned int n_outputs,
1355 union user_action_cookie *cookie)
1359 cookie->type = USER_ACTION_COOKIE_SFLOW;
1360 cookie->sflow.vlan_tci = vlan_tci;
1362 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1363 * port information") for the interpretation of cookie->output. */
1364 switch (n_outputs) {
1366 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1367 cookie->sflow.output = 0x40000000 | 256;
1371 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1373 cookie->sflow.output = ifindex;
1378 /* 0x80000000 means "multiple output ports. */
1379 cookie->sflow.output = 0x80000000 | n_outputs;
1384 /* Compose SAMPLE action for sFlow bridge sampling. */
1386 compose_sflow_action(const struct xbridge *xbridge,
1387 struct ofpbuf *odp_actions,
1388 const struct flow *flow,
1389 odp_port_t odp_port)
1391 uint32_t probability;
1392 union user_action_cookie cookie;
1394 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1398 probability = dpif_sflow_get_probability(xbridge->sflow);
1399 compose_sflow_cookie(xbridge, htons(0), odp_port,
1400 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1402 return compose_sample_action(xbridge, odp_actions, flow, probability,
1403 &cookie, sizeof cookie.sflow);
1407 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1408 uint32_t obs_domain_id, uint32_t obs_point_id,
1409 union user_action_cookie *cookie)
1411 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1412 cookie->flow_sample.probability = probability;
1413 cookie->flow_sample.collector_set_id = collector_set_id;
1414 cookie->flow_sample.obs_domain_id = obs_domain_id;
1415 cookie->flow_sample.obs_point_id = obs_point_id;
1419 compose_ipfix_cookie(union user_action_cookie *cookie)
1421 cookie->type = USER_ACTION_COOKIE_IPFIX;
1424 /* Compose SAMPLE action for IPFIX bridge sampling. */
1426 compose_ipfix_action(const struct xbridge *xbridge,
1427 struct ofpbuf *odp_actions,
1428 const struct flow *flow)
1430 uint32_t probability;
1431 union user_action_cookie cookie;
1433 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1437 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1438 compose_ipfix_cookie(&cookie);
1440 compose_sample_action(xbridge, odp_actions, flow, probability,
1441 &cookie, sizeof cookie.ipfix);
1444 /* SAMPLE action for sFlow must be first action in any given list of
1445 * actions. At this point we do not have all information required to
1446 * build it. So try to build sample action as complete as possible. */
1448 add_sflow_action(struct xlate_ctx *ctx)
1450 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1451 &ctx->xout->odp_actions,
1452 &ctx->xin->flow, ODPP_NONE);
1453 ctx->sflow_odp_port = 0;
1454 ctx->sflow_n_outputs = 0;
1457 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1458 * of actions, eventually after the SAMPLE action for sFlow. */
1460 add_ipfix_action(struct xlate_ctx *ctx)
1462 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1466 /* Fix SAMPLE action according to data collected while composing ODP actions.
1467 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1468 * USERSPACE action's user-cookie which is required for sflow. */
1470 fix_sflow_action(struct xlate_ctx *ctx)
1472 const struct flow *base = &ctx->base_flow;
1473 union user_action_cookie *cookie;
1475 if (!ctx->user_cookie_offset) {
1479 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1480 sizeof cookie->sflow);
1481 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1483 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1484 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1487 static enum slow_path_reason
1488 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1489 const struct xport *xport, const struct ofpbuf *packet)
1491 struct flow_wildcards *wc = &ctx->xout->wc;
1492 const struct xbridge *xbridge = ctx->xbridge;
1496 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1498 cfm_process_heartbeat(xport->cfm, packet);
1501 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1503 bfd_process_packet(xport->bfd, flow, packet);
1506 } else if (xport->xbundle && xport->xbundle->lacp
1507 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1509 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1512 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1514 stp_process_packet(xport, packet);
1523 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1526 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1527 struct flow_wildcards *wc = &ctx->xout->wc;
1528 struct flow *flow = &ctx->xin->flow;
1529 ovs_be16 flow_vlan_tci;
1530 uint32_t flow_pkt_mark;
1531 uint8_t flow_nw_tos;
1532 odp_port_t out_port, odp_port;
1535 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1536 * before traversing a patch port. */
1537 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1540 xlate_report(ctx, "Nonexistent output port");
1542 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1543 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1545 } else if (check_stp && !xport_stp_forward_state(xport)) {
1546 xlate_report(ctx, "STP not in forwarding state, skipping output");
1550 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1551 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1556 const struct xport *peer = xport->peer;
1557 struct flow old_flow = ctx->xin->flow;
1558 enum slow_path_reason special;
1560 ctx->xbridge = peer->xbridge;
1561 flow->in_port.ofp_port = peer->ofp_port;
1562 flow->metadata = htonll(0);
1563 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1564 memset(flow->regs, 0, sizeof flow->regs);
1566 special = process_special(ctx, &ctx->xin->flow, peer,
1569 ctx->xout->slow = special;
1570 } else if (may_receive(peer, ctx)) {
1571 if (xport_stp_forward_state(peer)) {
1572 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1574 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1575 * learning action look at the packet, then drop it. */
1576 struct flow old_base_flow = ctx->base_flow;
1577 size_t old_size = ctx->xout->odp_actions.size;
1578 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1579 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1580 ctx->xout->mirrors = old_mirrors;
1581 ctx->base_flow = old_base_flow;
1582 ctx->xout->odp_actions.size = old_size;
1586 ctx->xin->flow = old_flow;
1587 ctx->xbridge = xport->xbridge;
1589 if (ctx->xin->resubmit_stats) {
1590 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1591 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1597 flow_vlan_tci = flow->vlan_tci;
1598 flow_pkt_mark = flow->pkt_mark;
1599 flow_nw_tos = flow->nw_tos;
1601 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1602 wc->masks.nw_tos |= IP_ECN_MASK;
1603 flow->nw_tos &= ~IP_DSCP_MASK;
1604 flow->nw_tos |= dscp;
1607 if (xport->is_tunnel) {
1608 /* Save tunnel metadata so that changes made due to
1609 * the Logical (tunnel) Port are not visible for any further
1610 * matches, while explicit set actions on tunnel metadata are.
1612 struct flow_tnl flow_tnl = flow->tunnel;
1613 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1614 if (odp_port == ODPP_NONE) {
1615 xlate_report(ctx, "Tunneling decided against output");
1616 goto out; /* restore flow_nw_tos */
1618 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1619 xlate_report(ctx, "Not tunneling to our own address");
1620 goto out; /* restore flow_nw_tos */
1622 if (ctx->xin->resubmit_stats) {
1623 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1625 out_port = odp_port;
1626 commit_odp_tunnel_action(flow, &ctx->base_flow,
1627 &ctx->xout->odp_actions);
1628 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1630 ofp_port_t vlandev_port;
1632 odp_port = xport->odp_port;
1633 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1634 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1636 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1638 if (vlandev_port == ofp_port) {
1639 out_port = odp_port;
1641 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1642 flow->vlan_tci = htons(0);
1646 if (out_port != ODPP_NONE) {
1647 commit_odp_actions(flow, &ctx->base_flow,
1648 &ctx->xout->odp_actions, &ctx->xout->wc);
1649 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1652 ctx->sflow_odp_port = odp_port;
1653 ctx->sflow_n_outputs++;
1654 ctx->xout->nf_output_iface = ofp_port;
1659 flow->vlan_tci = flow_vlan_tci;
1660 flow->pkt_mark = flow_pkt_mark;
1661 flow->nw_tos = flow_nw_tos;
1665 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1667 compose_output_action__(ctx, ofp_port, true);
1671 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1673 struct rule_dpif *old_rule = ctx->rule;
1674 struct rule_actions *actions;
1676 if (ctx->xin->resubmit_stats) {
1677 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1682 actions = rule_dpif_get_actions(rule);
1683 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1684 rule_actions_unref(actions);
1685 ctx->rule = old_rule;
1690 xlate_table_action(struct xlate_ctx *ctx,
1691 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1693 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1694 struct rule_dpif *rule;
1695 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1696 uint8_t old_table_id = ctx->table_id;
1698 ctx->table_id = table_id;
1700 /* Look up a flow with 'in_port' as the input port. Then restore the
1701 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1702 * have surprising behavior). */
1703 ctx->xin->flow.in_port.ofp_port = in_port;
1704 rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1705 &ctx->xin->flow, &ctx->xout->wc,
1707 ctx->xin->flow.in_port.ofp_port = old_in_port;
1709 if (ctx->xin->resubmit_hook) {
1710 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1713 if (!rule && may_packet_in) {
1714 struct xport *xport;
1717 * check if table configuration flags
1718 * OFPTC_TABLE_MISS_CONTROLLER, default.
1719 * OFPTC_TABLE_MISS_CONTINUE,
1720 * OFPTC_TABLE_MISS_DROP
1721 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1722 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1723 choose_miss_rule(xport ? xport->config : 0,
1724 ctx->xbridge->miss_rule,
1725 ctx->xbridge->no_packet_in_rule, &rule);
1728 xlate_recursively(ctx, rule);
1729 rule_dpif_unref(rule);
1732 ctx->table_id = old_table_id;
1734 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1736 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1737 MAX_RESUBMIT_RECURSION);
1742 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1743 const struct ofpact_resubmit *resubmit)
1748 in_port = resubmit->in_port;
1749 if (in_port == OFPP_IN_PORT) {
1750 in_port = ctx->xin->flow.in_port.ofp_port;
1753 table_id = resubmit->table_id;
1754 if (table_id == 255) {
1755 table_id = ctx->table_id;
1758 xlate_table_action(ctx, in_port, table_id, false);
1762 flood_packets(struct xlate_ctx *ctx, bool all)
1764 const struct xport *xport;
1766 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1767 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1772 compose_output_action__(ctx, xport->ofp_port, false);
1773 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1774 compose_output_action(ctx, xport->ofp_port);
1778 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1782 execute_controller_action(struct xlate_ctx *ctx, int len,
1783 enum ofp_packet_in_reason reason,
1784 uint16_t controller_id)
1786 struct ofputil_packet_in *pin;
1787 struct ofpbuf *packet;
1790 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1791 ctx->xout->slow = SLOW_CONTROLLER;
1792 if (!ctx->xin->packet) {
1796 packet = ofpbuf_clone(ctx->xin->packet);
1798 key.skb_priority = 0;
1800 memset(&key.tunnel, 0, sizeof key.tunnel);
1802 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1803 &ctx->xout->odp_actions, &ctx->xout->wc);
1805 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1806 ctx->xout->odp_actions.size, NULL, NULL);
1808 pin = xmalloc(sizeof *pin);
1809 pin->packet_len = packet->size;
1810 pin->packet = ofpbuf_steal_data(packet);
1811 pin->reason = reason;
1812 pin->controller_id = controller_id;
1813 pin->table_id = ctx->table_id;
1814 pin->cookie = ctx->rule ? rule_dpif_get_flow_cookie(ctx->rule) : 0;
1816 pin->send_len = len;
1817 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1819 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1820 ofpbuf_delete(packet);
1824 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1826 struct flow_wildcards *wc = &ctx->xout->wc;
1827 struct flow *flow = &ctx->xin->flow;
1829 ovs_assert(eth_type_mpls(eth_type));
1831 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1832 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1834 if (flow->mpls_depth) {
1835 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1841 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1842 label = htonl(0x2); /* IPV6 Explicit Null. */
1844 label = htonl(0x0); /* IPV4 Explicit Null. */
1846 wc->masks.nw_tos |= IP_DSCP_MASK;
1847 wc->masks.nw_ttl = 0xff;
1848 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1849 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1850 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1851 flow->mpls_depth = 1;
1853 flow->dl_type = eth_type;
1857 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1859 struct flow_wildcards *wc = &ctx->xout->wc;
1860 struct flow *flow = &ctx->xin->flow;
1862 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1863 ovs_assert(!eth_type_mpls(eth_type));
1865 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1866 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1868 if (flow->mpls_depth) {
1870 flow->mpls_lse = htonl(0);
1871 if (!flow->mpls_depth) {
1872 flow->dl_type = eth_type;
1878 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1880 struct flow *flow = &ctx->xin->flow;
1882 if (!is_ip_any(flow)) {
1886 ctx->xout->wc.masks.nw_ttl = 0xff;
1887 if (flow->nw_ttl > 1) {
1893 for (i = 0; i < ids->n_controllers; i++) {
1894 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1898 /* Stop processing for current table. */
1904 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1906 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1910 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1911 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1916 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1918 struct flow *flow = &ctx->xin->flow;
1919 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1920 struct flow_wildcards *wc = &ctx->xout->wc;
1922 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1924 if (!eth_type_mpls(flow->dl_type)) {
1930 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1933 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1935 /* Stop processing for current table. */
1941 xlate_output_action(struct xlate_ctx *ctx,
1942 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1944 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1946 ctx->xout->nf_output_iface = NF_OUT_DROP;
1950 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1953 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1960 flood_packets(ctx, false);
1963 flood_packets(ctx, true);
1965 case OFPP_CONTROLLER:
1966 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1972 if (port != ctx->xin->flow.in_port.ofp_port) {
1973 compose_output_action(ctx, port);
1975 xlate_report(ctx, "skipping output to input port");
1980 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1981 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1982 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1983 ctx->xout->nf_output_iface = prev_nf_output_iface;
1984 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1985 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1986 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1991 xlate_output_reg_action(struct xlate_ctx *ctx,
1992 const struct ofpact_output_reg *or)
1994 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1995 if (port <= UINT16_MAX) {
1996 union mf_subvalue value;
1998 memset(&value, 0xff, sizeof value);
1999 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2000 xlate_output_action(ctx, u16_to_ofp(port),
2001 or->max_len, false);
2006 xlate_enqueue_action(struct xlate_ctx *ctx,
2007 const struct ofpact_enqueue *enqueue)
2009 ofp_port_t ofp_port = enqueue->port;
2010 uint32_t queue_id = enqueue->queue;
2011 uint32_t flow_priority, priority;
2014 /* Translate queue to priority. */
2015 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2017 /* Fall back to ordinary output action. */
2018 xlate_output_action(ctx, enqueue->port, 0, false);
2022 /* Check output port. */
2023 if (ofp_port == OFPP_IN_PORT) {
2024 ofp_port = ctx->xin->flow.in_port.ofp_port;
2025 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2029 /* Add datapath actions. */
2030 flow_priority = ctx->xin->flow.skb_priority;
2031 ctx->xin->flow.skb_priority = priority;
2032 compose_output_action(ctx, ofp_port);
2033 ctx->xin->flow.skb_priority = flow_priority;
2035 /* Update NetFlow output port. */
2036 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2037 ctx->xout->nf_output_iface = ofp_port;
2038 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2039 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2044 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2046 uint32_t skb_priority;
2048 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2049 ctx->xin->flow.skb_priority = skb_priority;
2051 /* Couldn't translate queue to a priority. Nothing to do. A warning
2052 * has already been logged. */
2057 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2059 const struct xbridge *xbridge = xbridge_;
2070 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2073 port = get_ofp_port(xbridge, ofp_port);
2074 return port ? port->may_enable : false;
2079 xlate_bundle_action(struct xlate_ctx *ctx,
2080 const struct ofpact_bundle *bundle)
2084 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2086 CONST_CAST(struct xbridge *, ctx->xbridge));
2087 if (bundle->dst.field) {
2088 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2091 xlate_output_action(ctx, port, 0, false);
2096 xlate_learn_action(struct xlate_ctx *ctx,
2097 const struct ofpact_learn *learn)
2099 uint64_t ofpacts_stub[1024 / 8];
2100 struct ofputil_flow_mod fm;
2101 struct ofpbuf ofpacts;
2103 ctx->xout->has_learn = true;
2105 learn_mask(learn, &ctx->xout->wc);
2107 if (!ctx->xin->may_learn) {
2111 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2112 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2113 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2114 ofpbuf_uninit(&ofpacts);
2118 xlate_fin_timeout(struct xlate_ctx *ctx,
2119 const struct ofpact_fin_timeout *oft)
2121 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2122 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2123 oft->fin_hard_timeout);
2128 xlate_sample_action(struct xlate_ctx *ctx,
2129 const struct ofpact_sample *os)
2131 union user_action_cookie cookie;
2132 /* Scale the probability from 16-bit to 32-bit while representing
2133 * the same percentage. */
2134 uint32_t probability = (os->probability << 16) | os->probability;
2136 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2137 &ctx->xout->odp_actions, &ctx->xout->wc);
2139 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2140 os->obs_domain_id, os->obs_point_id, &cookie);
2141 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2142 probability, &cookie, sizeof cookie.flow_sample);
2146 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2148 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2149 ? OFPUTIL_PC_NO_RECV_STP
2150 : OFPUTIL_PC_NO_RECV)) {
2154 /* Only drop packets here if both forwarding and learning are
2155 * disabled. If just learning is enabled, we need to have
2156 * OFPP_NORMAL and the learning action have a look at the packet
2157 * before we can drop it. */
2158 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2166 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2167 struct xlate_ctx *ctx)
2169 struct flow_wildcards *wc = &ctx->xout->wc;
2170 struct flow *flow = &ctx->xin->flow;
2171 const struct ofpact *a;
2173 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2174 struct ofpact_controller *controller;
2175 const struct ofpact_metadata *metadata;
2183 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2184 ofpact_get_OUTPUT(a)->max_len, true);
2188 /* XXX not yet implemented */
2191 case OFPACT_CONTROLLER:
2192 controller = ofpact_get_CONTROLLER(a);
2193 execute_controller_action(ctx, controller->max_len,
2195 controller->controller_id);
2198 case OFPACT_ENQUEUE:
2199 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2202 case OFPACT_SET_VLAN_VID:
2203 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2204 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2205 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2209 case OFPACT_SET_VLAN_PCP:
2210 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2211 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2213 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2217 case OFPACT_STRIP_VLAN:
2218 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2219 flow->vlan_tci = htons(0);
2222 case OFPACT_PUSH_VLAN:
2223 /* XXX 802.1AD(QinQ) */
2224 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2225 flow->vlan_tci = htons(VLAN_CFI);
2228 case OFPACT_SET_ETH_SRC:
2229 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2230 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2233 case OFPACT_SET_ETH_DST:
2234 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2235 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2238 case OFPACT_SET_IPV4_SRC:
2239 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2240 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2241 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2245 case OFPACT_SET_IPV4_DST:
2246 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2247 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2248 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2252 case OFPACT_SET_IPV4_DSCP:
2253 wc->masks.nw_tos |= IP_DSCP_MASK;
2254 /* OpenFlow 1.0 only supports IPv4. */
2255 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2256 flow->nw_tos &= ~IP_DSCP_MASK;
2257 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2261 case OFPACT_SET_L4_SRC_PORT:
2262 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2263 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2264 if (is_ip_any(flow)) {
2265 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2269 case OFPACT_SET_L4_DST_PORT:
2270 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2271 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2272 if (is_ip_any(flow)) {
2273 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2277 case OFPACT_RESUBMIT:
2278 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2281 case OFPACT_SET_TUNNEL:
2282 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2285 case OFPACT_SET_QUEUE:
2286 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2289 case OFPACT_POP_QUEUE:
2290 flow->skb_priority = ctx->orig_skb_priority;
2293 case OFPACT_REG_MOVE:
2294 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2297 case OFPACT_REG_LOAD:
2298 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2301 case OFPACT_STACK_PUSH:
2302 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2306 case OFPACT_STACK_POP:
2307 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2311 case OFPACT_PUSH_MPLS:
2312 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2315 case OFPACT_POP_MPLS:
2316 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2319 case OFPACT_SET_MPLS_TTL:
2320 if (compose_set_mpls_ttl_action(ctx,
2321 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2326 case OFPACT_DEC_MPLS_TTL:
2327 if (compose_dec_mpls_ttl_action(ctx)) {
2332 case OFPACT_DEC_TTL:
2333 wc->masks.nw_ttl = 0xff;
2334 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2340 /* Nothing to do. */
2343 case OFPACT_MULTIPATH:
2344 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2348 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2351 case OFPACT_OUTPUT_REG:
2352 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2356 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2363 case OFPACT_FIN_TIMEOUT:
2364 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2365 ctx->xout->has_fin_timeout = true;
2366 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2369 case OFPACT_CLEAR_ACTIONS:
2371 * Nothing to do because writa-actions is not supported for now.
2372 * When writa-actions is supported, clear-actions also must
2373 * be supported at the same time.
2377 case OFPACT_WRITE_METADATA:
2378 metadata = ofpact_get_WRITE_METADATA(a);
2379 flow->metadata &= ~metadata->mask;
2380 flow->metadata |= metadata->metadata & metadata->mask;
2384 /* Not implemented yet. */
2387 case OFPACT_GOTO_TABLE: {
2388 /* It is assumed that goto-table is the last action. */
2389 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2391 ovs_assert(ctx->table_id < ogt->table_id);
2392 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2393 ogt->table_id, true);
2398 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2405 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2406 const struct flow *flow, struct rule_dpif *rule,
2407 uint8_t tcp_flags, const struct ofpbuf *packet)
2409 xin->ofproto = ofproto;
2411 xin->packet = packet;
2412 xin->may_learn = packet != NULL;
2414 xin->ofpacts = NULL;
2415 xin->ofpacts_len = 0;
2416 xin->tcp_flags = tcp_flags;
2417 xin->resubmit_hook = NULL;
2418 xin->report_hook = NULL;
2419 xin->resubmit_stats = NULL;
2423 xlate_out_uninit(struct xlate_out *xout)
2426 ofpbuf_uninit(&xout->odp_actions);
2430 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2431 * into datapath actions, using 'ctx', and discards the datapath actions. */
2433 xlate_actions_for_side_effects(struct xlate_in *xin)
2435 struct xlate_out xout;
2437 xlate_actions(xin, &xout);
2438 xlate_out_uninit(&xout);
2442 xlate_report(struct xlate_ctx *ctx, const char *s)
2444 if (ctx->xin->report_hook) {
2445 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2450 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2453 dst->slow = src->slow;
2454 dst->has_learn = src->has_learn;
2455 dst->has_normal = src->has_normal;
2456 dst->has_fin_timeout = src->has_fin_timeout;
2457 dst->nf_output_iface = src->nf_output_iface;
2458 dst->mirrors = src->mirrors;
2460 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2461 sizeof dst->odp_actions_stub);
2462 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2463 src->odp_actions.size);
2466 static struct skb_priority_to_dscp *
2467 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2469 struct skb_priority_to_dscp *pdscp;
2472 hash = hash_int(skb_priority, 0);
2473 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2474 if (pdscp->skb_priority == skb_priority) {
2482 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2485 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2486 *dscp = pdscp ? pdscp->dscp : 0;
2487 return pdscp != NULL;
2491 clear_skb_priorities(struct xport *xport)
2493 struct skb_priority_to_dscp *pdscp, *next;
2495 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2496 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2502 actions_output_to_local_port(const struct xlate_ctx *ctx)
2504 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2505 const struct nlattr *a;
2508 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2509 ctx->xout->odp_actions.size) {
2510 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2511 && nl_attr_get_odp_port(a) == local_odp_port) {
2518 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2519 * into datapath actions in 'odp_actions', using 'ctx'.
2521 * The caller must take responsibility for eventually freeing 'xout', with
2522 * xlate_out_uninit(). */
2524 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2526 struct flow_wildcards *wc = &xout->wc;
2527 struct flow *flow = &xin->flow;
2529 struct rule_actions *actions = NULL;
2530 enum slow_path_reason special;
2531 const struct ofpact *ofpacts;
2532 struct xport *in_port;
2533 struct flow orig_flow;
2534 struct xlate_ctx ctx;
2538 COVERAGE_INC(xlate_actions);
2540 ovs_rwlock_rdlock(&xlate_rwlock);
2542 /* Flow initialization rules:
2543 * - 'base_flow' must match the kernel's view of the packet at the
2544 * time that action processing starts. 'flow' represents any
2545 * transformations we wish to make through actions.
2546 * - By default 'base_flow' and 'flow' are the same since the input
2547 * packet matches the output before any actions are applied.
2548 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2549 * of the received packet as seen by the kernel. If we later output
2550 * to another device without any modifications this will cause us to
2551 * insert a new tag since the original one was stripped off by the
2553 * - Tunnel metadata as received is retained in 'flow'. This allows
2554 * tunnel metadata matching also in later tables.
2555 * Since a kernel action for setting the tunnel metadata will only be
2556 * generated with actual tunnel output, changing the tunnel metadata
2557 * values in 'flow' (such as tun_id) will only have effect with a later
2558 * tunnel output action.
2559 * - Tunnel 'base_flow' is completely cleared since that is what the
2560 * kernel does. If we wish to maintain the original values an action
2561 * needs to be generated. */
2566 ctx.xout->has_learn = false;
2567 ctx.xout->has_normal = false;
2568 ctx.xout->has_fin_timeout = false;
2569 ctx.xout->nf_output_iface = NF_OUT_DROP;
2570 ctx.xout->mirrors = 0;
2571 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2572 sizeof ctx.xout->odp_actions_stub);
2573 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2575 ctx.xbridge = xbridge_lookup(xin->ofproto);
2580 ctx.rule = xin->rule;
2582 ctx.base_flow = *flow;
2583 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2584 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2586 flow_wildcards_init_catchall(wc);
2587 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2588 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2589 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2590 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2592 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
2593 if (ctx.xbridge->has_netflow) {
2594 netflow_mask_wc(flow, wc);
2598 ctx.orig_skb_priority = flow->skb_priority;
2603 ofpacts = xin->ofpacts;
2604 ofpacts_len = xin->ofpacts_len;
2605 } else if (xin->rule) {
2606 actions = rule_dpif_get_actions(xin->rule);
2607 ofpacts = actions->ofpacts;
2608 ofpacts_len = actions->ofpacts_len;
2613 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2615 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2616 /* Do this conditionally because the copy is expensive enough that it
2617 * shows up in profiles. */
2621 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2622 switch (ctx.xbridge->frag) {
2623 case OFPC_FRAG_NORMAL:
2624 /* We must pretend that transport ports are unavailable. */
2625 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2626 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2629 case OFPC_FRAG_DROP:
2632 case OFPC_FRAG_REASM:
2635 case OFPC_FRAG_NX_MATCH:
2636 /* Nothing to do. */
2639 case OFPC_INVALID_TTL_TO_CONTROLLER:
2644 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2645 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2647 ctx.xout->slow = special;
2649 size_t sample_actions_len;
2651 if (flow->in_port.ofp_port
2652 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2653 flow->in_port.ofp_port,
2655 ctx.base_flow.vlan_tci = 0;
2658 add_sflow_action(&ctx);
2659 add_ipfix_action(&ctx);
2660 sample_actions_len = ctx.xout->odp_actions.size;
2662 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
2663 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2665 /* We've let OFPP_NORMAL and the learning action look at the
2666 * packet, so drop it now if forwarding is disabled. */
2667 if (in_port && !xport_stp_forward_state(in_port)) {
2668 ctx.xout->odp_actions.size = sample_actions_len;
2672 if (ctx.xbridge->has_in_band
2673 && in_band_must_output_to_local_port(flow)
2674 && !actions_output_to_local_port(&ctx)) {
2675 compose_output_action(&ctx, OFPP_LOCAL);
2678 fix_sflow_action(&ctx);
2680 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2681 add_mirror_actions(&ctx, &orig_flow);
2685 ofpbuf_uninit(&ctx.stack);
2687 /* Clear the metadata and register wildcard masks, because we won't
2688 * use non-header fields as part of the cache. */
2689 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2690 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
2693 ovs_rwlock_unlock(&xlate_rwlock);
2695 rule_actions_unref(actions);