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"
50 COVERAGE_DEFINE(xlate_actions);
52 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
54 /* Maximum depth of flow table recursion (due to resubmit actions) in a
55 * flow translation. */
56 #define MAX_RESUBMIT_RECURSION 64
58 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
61 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
62 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
64 struct list xbundles; /* Owned xbundles. */
65 struct hmap xports; /* Indexed by ofp_port. */
67 char *name; /* Name used in log messages. */
68 struct dpif *dpif; /* Datapath interface. */
69 struct mac_learning *ml; /* Mac learning handle. */
70 struct mbridge *mbridge; /* Mirroring. */
71 struct dpif_sflow *sflow; /* SFlow handle, or null. */
72 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
73 struct stp *stp; /* STP or null if disabled. */
75 /* Special rules installed by ofproto-dpif. */
76 struct rule_dpif *miss_rule;
77 struct rule_dpif *no_packet_in_rule;
79 enum ofp_config_flags frag; /* Fragmentation handling. */
80 bool has_netflow; /* Bridge runs netflow? */
81 bool has_in_band; /* Bridge has in band control? */
82 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
86 struct hmap_node hmap_node; /* In global 'xbundles' map. */
87 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
89 struct list list_node; /* In parent 'xbridges' list. */
90 struct xbridge *xbridge; /* Parent xbridge. */
92 struct list xports; /* Contains "struct xport"s. */
94 char *name; /* Name used in log messages. */
95 struct bond *bond; /* Nonnull iff more than one port. */
96 struct lacp *lacp; /* LACP handle or null. */
98 enum port_vlan_mode vlan_mode; /* VLAN mode. */
99 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
100 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
101 * NULL if all VLANs are trunked. */
102 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
103 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
107 struct hmap_node hmap_node; /* Node in global 'xports' map. */
108 struct ofport_dpif *ofport; /* Key in global 'xports map. */
110 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
111 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
113 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
115 struct list bundle_node; /* In parent xbundle (if it exists). */
116 struct xbundle *xbundle; /* Parent xbundle or null. */
118 struct netdev *netdev; /* 'ofport''s netdev. */
120 struct xbridge *xbridge; /* Parent bridge. */
121 struct xport *peer; /* Patch port peer or null. */
123 enum ofputil_port_config config; /* OpenFlow port configuration. */
124 int stp_port_no; /* STP port number or 0 if not in use. */
126 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
128 bool may_enable; /* May be enabled in bonds. */
129 bool is_tunnel; /* Is a tunnel port. */
131 struct cfm *cfm; /* CFM handle or null. */
132 struct bfd *bfd; /* BFD handle or null. */
136 struct xlate_in *xin;
137 struct xlate_out *xout;
139 const struct xbridge *xbridge;
141 /* Flow at the last commit. */
142 struct flow base_flow;
144 /* Tunnel IP destination address as received. This is stored separately
145 * as the base_flow.tunnel is cleared on init to reflect the datapath
146 * behavior. Used to make sure not to send tunneled output to ourselves,
147 * which might lead to an infinite loop. This could happen easily
148 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
149 * actually set the tun_dst field. */
150 ovs_be32 orig_tunnel_ip_dst;
152 /* Stack for the push and pop actions. Each stack element is of type
153 * "union mf_subvalue". */
154 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
157 /* The rule that we are currently translating, or NULL. */
158 struct rule_dpif *rule;
160 int recurse; /* Recursion level, via xlate_table_action. */
161 uint32_t orig_skb_priority; /* Priority when packet arrived. */
162 uint8_t table_id; /* OpenFlow table ID where flow was found. */
163 uint32_t sflow_n_outputs; /* Number of output ports. */
164 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
165 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
166 bool exit; /* No further actions should be processed. */
169 /* A controller may use OFPP_NONE as the ingress port to indicate that
170 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
171 * when an input bundle is needed for validation (e.g., mirroring or
172 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
173 * any 'port' structs, so care must be taken when dealing with it.
174 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
175 static struct xbundle ofpp_none_bundle;
177 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
178 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
179 * traffic egressing the 'ofport' with that priority should be marked with. */
180 struct skb_priority_to_dscp {
181 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
182 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
184 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
187 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
188 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
189 static struct hmap xports = HMAP_INITIALIZER(&xports);
191 static bool may_receive(const struct xport *, struct xlate_ctx *);
192 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
194 static void xlate_normal(struct xlate_ctx *);
195 static void xlate_report(struct xlate_ctx *, const char *);
196 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
197 uint8_t table_id, bool may_packet_in);
198 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
199 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
200 static void output_normal(struct xlate_ctx *, const struct xbundle *,
202 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
204 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
206 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
207 static struct xbundle *xbundle_lookup(const struct ofbundle *);
208 static struct xport *xport_lookup(const struct ofport_dpif *);
209 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
210 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
211 uint32_t skb_priority);
212 static void clear_skb_priorities(struct xport *);
213 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
217 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
218 struct dpif *dpif, struct rule_dpif *miss_rule,
219 struct rule_dpif *no_packet_in_rule,
220 const struct mac_learning *ml, struct stp *stp,
221 const struct mbridge *mbridge,
222 const struct dpif_sflow *sflow,
223 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
224 bool forward_bpdu, bool has_in_band, bool has_netflow)
226 struct xbridge *xbridge = xbridge_lookup(ofproto);
229 xbridge = xzalloc(sizeof *xbridge);
230 xbridge->ofproto = ofproto;
232 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
233 hmap_init(&xbridge->xports);
234 list_init(&xbridge->xbundles);
237 if (xbridge->ml != ml) {
238 mac_learning_unref(xbridge->ml);
239 xbridge->ml = mac_learning_ref(ml);
242 if (xbridge->mbridge != mbridge) {
243 mbridge_unref(xbridge->mbridge);
244 xbridge->mbridge = mbridge_ref(mbridge);
247 if (xbridge->sflow != sflow) {
248 dpif_sflow_unref(xbridge->sflow);
249 xbridge->sflow = dpif_sflow_ref(sflow);
252 if (xbridge->ipfix != ipfix) {
253 dpif_ipfix_unref(xbridge->ipfix);
254 xbridge->ipfix = dpif_ipfix_ref(ipfix);
257 if (xbridge->stp != stp) {
258 stp_unref(xbridge->stp);
259 xbridge->stp = stp_ref(stp);
263 xbridge->name = xstrdup(name);
265 xbridge->dpif = dpif;
266 xbridge->forward_bpdu = forward_bpdu;
267 xbridge->has_in_band = has_in_band;
268 xbridge->has_netflow = has_netflow;
269 xbridge->frag = frag;
270 xbridge->miss_rule = miss_rule;
271 xbridge->no_packet_in_rule = no_packet_in_rule;
275 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
277 struct xbridge *xbridge = xbridge_lookup(ofproto);
278 struct xbundle *xbundle, *next_xbundle;
279 struct xport *xport, *next_xport;
285 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
286 xlate_ofport_remove(xport->ofport);
289 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
290 xlate_bundle_remove(xbundle->ofbundle);
293 hmap_remove(&xbridges, &xbridge->hmap_node);
299 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
300 const char *name, enum port_vlan_mode vlan_mode, int vlan,
301 unsigned long *trunks, bool use_priority_tags,
302 const struct bond *bond, const struct lacp *lacp,
305 struct xbundle *xbundle = xbundle_lookup(ofbundle);
308 xbundle = xzalloc(sizeof *xbundle);
309 xbundle->ofbundle = ofbundle;
310 xbundle->xbridge = xbridge_lookup(ofproto);
312 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
313 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
314 list_init(&xbundle->xports);
317 ovs_assert(xbundle->xbridge);
320 xbundle->name = xstrdup(name);
322 xbundle->vlan_mode = vlan_mode;
323 xbundle->vlan = vlan;
324 xbundle->trunks = trunks;
325 xbundle->use_priority_tags = use_priority_tags;
326 xbundle->floodable = floodable;
328 if (xbundle->bond != bond) {
329 bond_unref(xbundle->bond);
330 xbundle->bond = bond_ref(bond);
333 if (xbundle->lacp != lacp) {
334 lacp_unref(xbundle->lacp);
335 xbundle->lacp = lacp_ref(lacp);
340 xlate_bundle_remove(struct ofbundle *ofbundle)
342 struct xbundle *xbundle = xbundle_lookup(ofbundle);
343 struct xport *xport, *next;
349 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
350 list_remove(&xport->bundle_node);
351 xport->xbundle = NULL;
354 hmap_remove(&xbundles, &xbundle->hmap_node);
355 list_remove(&xbundle->list_node);
356 bond_unref(xbundle->bond);
357 lacp_unref(xbundle->lacp);
363 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
364 struct ofport_dpif *ofport, ofp_port_t ofp_port,
365 odp_port_t odp_port, const struct netdev *netdev,
366 const struct cfm *cfm, const struct bfd *bfd,
367 struct ofport_dpif *peer, int stp_port_no,
368 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
369 enum ofputil_port_config config, bool is_tunnel,
372 struct xport *xport = xport_lookup(ofport);
376 xport = xzalloc(sizeof *xport);
377 xport->ofport = ofport;
378 xport->xbridge = xbridge_lookup(ofproto);
379 xport->ofp_port = ofp_port;
381 hmap_init(&xport->skb_priorities);
382 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
383 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
384 hash_ofp_port(xport->ofp_port));
387 ovs_assert(xport->ofp_port == ofp_port);
389 xport->config = config;
390 xport->stp_port_no = stp_port_no;
391 xport->is_tunnel = is_tunnel;
392 xport->may_enable = may_enable;
393 xport->odp_port = odp_port;
395 if (xport->netdev != netdev) {
396 netdev_close(xport->netdev);
397 xport->netdev = netdev_ref(netdev);
400 if (xport->cfm != cfm) {
401 cfm_unref(xport->cfm);
402 xport->cfm = cfm_ref(cfm);
405 if (xport->bfd != bfd) {
406 bfd_unref(xport->bfd);
407 xport->bfd = bfd_ref(bfd);
411 xport->peer->peer = NULL;
413 xport->peer = xport_lookup(peer);
415 xport->peer->peer = xport;
418 if (xport->xbundle) {
419 list_remove(&xport->bundle_node);
421 xport->xbundle = xbundle_lookup(ofbundle);
422 if (xport->xbundle) {
423 list_insert(&xport->xbundle->xports, &xport->bundle_node);
426 clear_skb_priorities(xport);
427 for (i = 0; i < n_qdscp; i++) {
428 struct skb_priority_to_dscp *pdscp;
429 uint32_t skb_priority;
431 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
436 pdscp = xmalloc(sizeof *pdscp);
437 pdscp->skb_priority = skb_priority;
438 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
439 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
440 hash_int(pdscp->skb_priority, 0));
445 xlate_ofport_remove(struct ofport_dpif *ofport)
447 struct xport *xport = xport_lookup(ofport);
454 xport->peer->peer = NULL;
458 if (xport->xbundle) {
459 list_remove(&xport->bundle_node);
462 clear_skb_priorities(xport);
463 hmap_destroy(&xport->skb_priorities);
465 hmap_remove(&xports, &xport->hmap_node);
466 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
468 netdev_close(xport->netdev);
469 cfm_unref(xport->cfm);
470 bfd_unref(xport->bfd);
474 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
475 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
476 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
477 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
478 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
479 * 'packet' ingressed.
481 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
482 * 'flow''s in_port to OFPP_NONE.
484 * This function does post-processing on data returned from
485 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
486 * of the upcall processing logic. In particular, if the extracted in_port is
487 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
488 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
489 * a VLAN header onto 'packet' (if it is nonnull).
491 * Similarly, this function also includes some logic to help with tunnels. It
492 * may modify 'flow' as necessary to make the tunneling implementation
493 * transparent to the upcall processing logic.
495 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
496 * or some other positive errno if there are other problems. */
498 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
499 const struct nlattr *key, size_t key_len,
500 struct flow *flow, enum odp_key_fitness *fitnessp,
501 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
503 enum odp_key_fitness fitness;
504 const struct xport *xport;
507 ovs_rwlock_rdlock(&xlate_rwlock);
508 fitness = odp_flow_key_to_flow(key, key_len, flow);
509 if (fitness == ODP_FIT_ERROR) {
515 *odp_in_port = flow->in_port.odp_port;
518 xport = xport_lookup(tnl_port_should_receive(flow)
519 ? tnl_port_receive(flow)
520 : odp_port_to_ofport(backer, flow->in_port.odp_port));
522 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
527 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
529 /* Make the packet resemble the flow, so that it gets sent to
530 * an OpenFlow controller properly, so that it looks correct
531 * for sFlow, and so that flow_extract() will get the correct
532 * vlan_tci if it is called on 'packet'.
534 * The allocated space inside 'packet' probably also contains
535 * 'key', that is, both 'packet' and 'key' are probably part of
536 * a struct dpif_upcall (see the large comment on that
537 * structure definition), so pushing data on 'packet' is in
538 * general not a good idea since it could overwrite 'key' or
539 * free it as a side effect. However, it's OK in this special
540 * case because we know that 'packet' is inside a Netlink
541 * attribute: pushing 4 bytes will just overwrite the 4-byte
542 * "struct nlattr", which is fine since we don't need that
544 eth_push_vlan(packet, flow->vlan_tci);
546 /* We can't reproduce 'key' from 'flow'. */
547 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
552 *ofproto = xport->xbridge->ofproto;
559 ovs_rwlock_unlock(&xlate_rwlock);
563 static struct xbridge *
564 xbridge_lookup(const struct ofproto_dpif *ofproto)
566 struct xbridge *xbridge;
572 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
574 if (xbridge->ofproto == ofproto) {
581 static struct xbundle *
582 xbundle_lookup(const struct ofbundle *ofbundle)
584 struct xbundle *xbundle;
590 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
592 if (xbundle->ofbundle == ofbundle) {
599 static struct xport *
600 xport_lookup(const struct ofport_dpif *ofport)
608 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
610 if (xport->ofport == ofport) {
617 static struct stp_port *
618 xport_get_stp_port(const struct xport *xport)
620 return xport->xbridge->stp && xport->stp_port_no
621 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
625 static enum stp_state
626 xport_stp_learn_state(const struct xport *xport)
628 struct stp_port *sp = xport_get_stp_port(xport);
629 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
633 xport_stp_forward_state(const struct xport *xport)
635 struct stp_port *sp = xport_get_stp_port(xport);
636 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
639 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
640 * were used to make the determination.*/
642 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
644 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
645 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
649 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
651 struct stp_port *sp = xport_get_stp_port(xport);
652 struct ofpbuf payload = *packet;
653 struct eth_header *eth = payload.data;
655 /* Sink packets on ports that have STP disabled when the bridge has
657 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
661 /* Trim off padding on payload. */
662 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
663 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
666 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
667 stp_received_bpdu(sp, payload.data, payload.size);
671 static struct xport *
672 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
676 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
678 if (xport->ofp_port == ofp_port) {
686 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
688 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
689 return xport ? xport->odp_port : ODPP_NONE;
693 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
695 return (bundle->vlan_mode != PORT_VLAN_ACCESS
696 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
700 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
702 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
706 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
708 return xbundle != &ofpp_none_bundle
709 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
714 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
716 return xbundle != &ofpp_none_bundle
717 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
722 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
724 return xbundle != &ofpp_none_bundle
725 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
729 static struct xbundle *
730 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
731 bool warn, struct xport **in_xportp)
735 /* Find the port and bundle for the received packet. */
736 xport = get_ofp_port(xbridge, in_port);
740 if (xport && xport->xbundle) {
741 return xport->xbundle;
744 /* Special-case OFPP_NONE, which a controller may use as the ingress
745 * port for traffic that it is sourcing. */
746 if (in_port == OFPP_NONE) {
747 ofpp_none_bundle.name = "OFPP_NONE";
748 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
749 return &ofpp_none_bundle;
752 /* Odd. A few possible reasons here:
754 * - We deleted a port but there are still a few packets queued up
757 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
758 * we don't know about.
760 * - The ofproto client didn't configure the port as part of a bundle.
761 * This is particularly likely to happen if a packet was received on the
762 * port after it was created, but before the client had a chance to
763 * configure its bundle.
766 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
768 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
769 "port %"PRIu16, xbridge->name, in_port);
775 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
777 const struct xbridge *xbridge = ctx->xbridge;
778 mirror_mask_t mirrors;
779 struct xbundle *in_xbundle;
783 mirrors = ctx->xout->mirrors;
784 ctx->xout->mirrors = 0;
786 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
787 ctx->xin->packet != NULL, NULL);
791 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
793 /* Drop frames on bundles reserved for mirroring. */
794 if (xbundle_mirror_out(xbridge, in_xbundle)) {
795 if (ctx->xin->packet != NULL) {
796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
797 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
798 "%s, which is reserved exclusively for mirroring",
799 ctx->xbridge->name, in_xbundle->name);
801 ofpbuf_clear(&ctx->xout->odp_actions);
806 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
807 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
810 vlan = input_vid_to_vlan(in_xbundle, vid);
816 /* Restore the original packet before adding the mirror actions. */
817 ctx->xin->flow = *orig_flow;
820 mirror_mask_t dup_mirrors;
821 struct ofbundle *out;
822 unsigned long *vlans;
827 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
828 &vlans, &dup_mirrors, &out, &out_vlan);
829 ovs_assert(has_mirror);
832 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
834 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
837 if (!vlan_mirrored) {
838 mirrors = zero_rightmost_1bit(mirrors);
842 mirrors &= ~dup_mirrors;
843 ctx->xout->mirrors |= dup_mirrors;
845 struct xbundle *out_xbundle = xbundle_lookup(out);
847 output_normal(ctx, out_xbundle, vlan);
849 } else if (vlan != out_vlan
850 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
851 struct xbundle *xbundle;
853 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
854 if (xbundle_includes_vlan(xbundle, out_vlan)
855 && !xbundle_mirror_out(xbridge, xbundle)) {
856 output_normal(ctx, xbundle, out_vlan);
863 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
864 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
865 * the bundle on which the packet was received, returns the VLAN to which the
868 * Both 'vid' and the return value are in the range 0...4095. */
870 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
872 switch (in_xbundle->vlan_mode) {
873 case PORT_VLAN_ACCESS:
874 return in_xbundle->vlan;
877 case PORT_VLAN_TRUNK:
880 case PORT_VLAN_NATIVE_UNTAGGED:
881 case PORT_VLAN_NATIVE_TAGGED:
882 return vid ? vid : in_xbundle->vlan;
889 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
890 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
893 * 'vid' should be the VID obtained from the 802.1Q header that was received as
894 * part of a packet (specify 0 if there was no 802.1Q header), in the range
897 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
899 /* Allow any VID on the OFPP_NONE port. */
900 if (in_xbundle == &ofpp_none_bundle) {
904 switch (in_xbundle->vlan_mode) {
905 case PORT_VLAN_ACCESS:
908 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
909 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
910 "packet received on port %s configured as VLAN "
911 "%"PRIu16" access port", vid, in_xbundle->name,
918 case PORT_VLAN_NATIVE_UNTAGGED:
919 case PORT_VLAN_NATIVE_TAGGED:
921 /* Port must always carry its native VLAN. */
925 case PORT_VLAN_TRUNK:
926 if (!xbundle_includes_vlan(in_xbundle, vid)) {
928 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
929 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
930 "received on port %s not configured for trunking "
931 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
943 /* Given 'vlan', the VLAN that a packet belongs to, and
944 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
945 * that should be included in the 802.1Q header. (If the return value is 0,
946 * then the 802.1Q header should only be included in the packet if there is a
949 * Both 'vlan' and the return value are in the range 0...4095. */
951 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
953 switch (out_xbundle->vlan_mode) {
954 case PORT_VLAN_ACCESS:
957 case PORT_VLAN_TRUNK:
958 case PORT_VLAN_NATIVE_TAGGED:
961 case PORT_VLAN_NATIVE_UNTAGGED:
962 return vlan == out_xbundle->vlan ? 0 : vlan;
970 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
973 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
975 ovs_be16 tci, old_tci;
978 vid = output_vlan_to_vid(out_xbundle, vlan);
979 if (list_is_empty(&out_xbundle->xports)) {
980 /* Partially configured bundle with no slaves. Drop the packet. */
982 } else if (!out_xbundle->bond) {
983 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
986 struct ofport_dpif *ofport;
988 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
989 &ctx->xout->wc, vid);
990 xport = xport_lookup(ofport);
993 /* No slaves enabled, so drop packet. */
1000 if (tci || out_xbundle->use_priority_tags) {
1001 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1003 tci |= htons(VLAN_CFI);
1008 compose_output_action(ctx, xport->ofp_port);
1009 *flow_tci = old_tci;
1012 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1013 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1014 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1016 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1018 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1022 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1023 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1027 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1028 if (flow->nw_proto == ARP_OP_REPLY) {
1030 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1031 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1032 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1034 return flow->nw_src == flow->nw_dst;
1040 /* Checks whether a MAC learning update is necessary for MAC learning table
1041 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1044 * Most packets processed through the MAC learning table do not actually
1045 * change it in any way. This function requires only a read lock on the MAC
1046 * learning table, so it is much cheaper in this common case.
1048 * Keep the code here synchronized with that in update_learning_table__()
1051 is_mac_learning_update_needed(const struct mac_learning *ml,
1052 const struct flow *flow,
1053 struct flow_wildcards *wc,
1054 int vlan, struct xbundle *in_xbundle)
1055 OVS_REQ_RDLOCK(ml->rwlock)
1057 struct mac_entry *mac;
1059 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1063 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1064 if (!mac || mac_entry_age(ml, mac)) {
1068 if (is_gratuitous_arp(flow, wc)) {
1069 /* We don't want to learn from gratuitous ARP packets that are
1070 * reflected back over bond slaves so we lock the learning table. */
1071 if (!in_xbundle->bond) {
1073 } else if (mac_entry_is_grat_arp_locked(mac)) {
1078 return mac->port.p != in_xbundle->ofbundle;
1082 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1083 * received on 'in_xbundle' in 'vlan'.
1085 * This code repeats all the checks in is_mac_learning_update_needed() because
1086 * the lock was released between there and here and thus the MAC learning state
1087 * could have changed.
1089 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1092 update_learning_table__(const struct xbridge *xbridge,
1093 const struct flow *flow, struct flow_wildcards *wc,
1094 int vlan, struct xbundle *in_xbundle)
1095 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1097 struct mac_entry *mac;
1099 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1103 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1104 if (is_gratuitous_arp(flow, wc)) {
1105 /* We don't want to learn from gratuitous ARP packets that are
1106 * reflected back over bond slaves so we lock the learning table. */
1107 if (!in_xbundle->bond) {
1108 mac_entry_set_grat_arp_lock(mac);
1109 } else if (mac_entry_is_grat_arp_locked(mac)) {
1114 if (mac->port.p != in_xbundle->ofbundle) {
1115 /* The log messages here could actually be useful in debugging,
1116 * so keep the rate limit relatively high. */
1117 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1119 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1120 "on port %s in VLAN %d",
1121 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1122 in_xbundle->name, vlan);
1124 mac->port.p = in_xbundle->ofbundle;
1125 mac_learning_changed(xbridge->ml);
1130 update_learning_table(const struct xbridge *xbridge,
1131 const struct flow *flow, struct flow_wildcards *wc,
1132 int vlan, struct xbundle *in_xbundle)
1136 /* Don't learn the OFPP_NONE port. */
1137 if (in_xbundle == &ofpp_none_bundle) {
1141 /* First try the common case: no change to MAC learning table. */
1142 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1143 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1145 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1148 /* Slow path: MAC learning table might need an update. */
1149 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1150 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1151 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1155 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1156 * dropped. Returns true if they may be forwarded, false if they should be
1159 * 'in_port' must be the xport that corresponds to flow->in_port.
1160 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1162 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1163 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1164 * checked by input_vid_is_valid().
1166 * May also add tags to '*tags', although the current implementation only does
1167 * so in one special case.
1170 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1173 struct xbundle *in_xbundle = in_port->xbundle;
1174 const struct xbridge *xbridge = ctx->xbridge;
1175 struct flow *flow = &ctx->xin->flow;
1177 /* Drop frames for reserved multicast addresses
1178 * only if forward_bpdu option is absent. */
1179 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1180 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1184 if (in_xbundle->bond) {
1185 struct mac_entry *mac;
1187 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1193 xlate_report(ctx, "bonding refused admissibility, dropping");
1196 case BV_DROP_IF_MOVED:
1197 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1198 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1199 if (mac && mac->port.p != in_xbundle->ofbundle &&
1200 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1201 || mac_entry_is_grat_arp_locked(mac))) {
1202 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1203 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1207 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1216 xlate_normal(struct xlate_ctx *ctx)
1218 struct flow_wildcards *wc = &ctx->xout->wc;
1219 struct flow *flow = &ctx->xin->flow;
1220 struct xbundle *in_xbundle;
1221 struct xport *in_port;
1222 struct mac_entry *mac;
1226 ctx->xout->has_normal = true;
1228 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1229 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1230 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1232 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1233 ctx->xin->packet != NULL, &in_port);
1235 xlate_report(ctx, "no input bundle, dropping");
1239 /* Drop malformed frames. */
1240 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1241 !(flow->vlan_tci & htons(VLAN_CFI))) {
1242 if (ctx->xin->packet != NULL) {
1243 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1244 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1245 "VLAN tag received on port %s",
1246 ctx->xbridge->name, in_xbundle->name);
1248 xlate_report(ctx, "partial VLAN tag, dropping");
1252 /* Drop frames on bundles reserved for mirroring. */
1253 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1254 if (ctx->xin->packet != NULL) {
1255 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1256 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1257 "%s, which is reserved exclusively for mirroring",
1258 ctx->xbridge->name, in_xbundle->name);
1260 xlate_report(ctx, "input port is mirror output port, dropping");
1265 vid = vlan_tci_to_vid(flow->vlan_tci);
1266 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1267 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1270 vlan = input_vid_to_vlan(in_xbundle, vid);
1272 /* Check other admissibility requirements. */
1273 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1277 /* Learn source MAC. */
1278 if (ctx->xin->may_learn) {
1279 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1282 /* Determine output bundle. */
1283 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1284 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1286 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
1287 if (mac_xbundle && mac_xbundle != in_xbundle) {
1288 xlate_report(ctx, "forwarding to learned port");
1289 output_normal(ctx, mac_xbundle, vlan);
1290 } else if (!mac_xbundle) {
1291 xlate_report(ctx, "learned port is unknown, dropping");
1293 xlate_report(ctx, "learned port is input port, dropping");
1296 struct xbundle *xbundle;
1298 xlate_report(ctx, "no learned MAC for destination, flooding");
1299 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1300 if (xbundle != in_xbundle
1301 && xbundle_includes_vlan(xbundle, vlan)
1302 && xbundle->floodable
1303 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1304 output_normal(ctx, xbundle, vlan);
1307 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1309 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1312 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1313 * the number of packets out of UINT32_MAX to sample. The given
1314 * cookie is passed back in the callback for each sampled packet.
1317 compose_sample_action(const struct xbridge *xbridge,
1318 struct ofpbuf *odp_actions,
1319 const struct flow *flow,
1320 const uint32_t probability,
1321 const union user_action_cookie *cookie,
1322 const size_t cookie_size)
1324 size_t sample_offset, actions_offset;
1325 odp_port_t odp_port;
1329 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1331 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1333 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1335 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1336 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1337 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1339 nl_msg_end_nested(odp_actions, actions_offset);
1340 nl_msg_end_nested(odp_actions, sample_offset);
1341 return cookie_offset;
1345 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1346 odp_port_t odp_port, unsigned int n_outputs,
1347 union user_action_cookie *cookie)
1351 cookie->type = USER_ACTION_COOKIE_SFLOW;
1352 cookie->sflow.vlan_tci = vlan_tci;
1354 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1355 * port information") for the interpretation of cookie->output. */
1356 switch (n_outputs) {
1358 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1359 cookie->sflow.output = 0x40000000 | 256;
1363 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1365 cookie->sflow.output = ifindex;
1370 /* 0x80000000 means "multiple output ports. */
1371 cookie->sflow.output = 0x80000000 | n_outputs;
1376 /* Compose SAMPLE action for sFlow bridge sampling. */
1378 compose_sflow_action(const struct xbridge *xbridge,
1379 struct ofpbuf *odp_actions,
1380 const struct flow *flow,
1381 odp_port_t odp_port)
1383 uint32_t probability;
1384 union user_action_cookie cookie;
1386 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1390 probability = dpif_sflow_get_probability(xbridge->sflow);
1391 compose_sflow_cookie(xbridge, htons(0), odp_port,
1392 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1394 return compose_sample_action(xbridge, odp_actions, flow, probability,
1395 &cookie, sizeof cookie.sflow);
1399 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1400 uint32_t obs_domain_id, uint32_t obs_point_id,
1401 union user_action_cookie *cookie)
1403 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1404 cookie->flow_sample.probability = probability;
1405 cookie->flow_sample.collector_set_id = collector_set_id;
1406 cookie->flow_sample.obs_domain_id = obs_domain_id;
1407 cookie->flow_sample.obs_point_id = obs_point_id;
1411 compose_ipfix_cookie(union user_action_cookie *cookie)
1413 cookie->type = USER_ACTION_COOKIE_IPFIX;
1416 /* Compose SAMPLE action for IPFIX bridge sampling. */
1418 compose_ipfix_action(const struct xbridge *xbridge,
1419 struct ofpbuf *odp_actions,
1420 const struct flow *flow)
1422 uint32_t probability;
1423 union user_action_cookie cookie;
1425 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1429 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1430 compose_ipfix_cookie(&cookie);
1432 compose_sample_action(xbridge, odp_actions, flow, probability,
1433 &cookie, sizeof cookie.ipfix);
1436 /* SAMPLE action for sFlow must be first action in any given list of
1437 * actions. At this point we do not have all information required to
1438 * build it. So try to build sample action as complete as possible. */
1440 add_sflow_action(struct xlate_ctx *ctx)
1442 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1443 &ctx->xout->odp_actions,
1444 &ctx->xin->flow, ODPP_NONE);
1445 ctx->sflow_odp_port = 0;
1446 ctx->sflow_n_outputs = 0;
1449 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1450 * of actions, eventually after the SAMPLE action for sFlow. */
1452 add_ipfix_action(struct xlate_ctx *ctx)
1454 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1458 /* Fix SAMPLE action according to data collected while composing ODP actions.
1459 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1460 * USERSPACE action's user-cookie which is required for sflow. */
1462 fix_sflow_action(struct xlate_ctx *ctx)
1464 const struct flow *base = &ctx->base_flow;
1465 union user_action_cookie *cookie;
1467 if (!ctx->user_cookie_offset) {
1471 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1472 sizeof cookie->sflow);
1473 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1475 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1476 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1479 static enum slow_path_reason
1480 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1481 const struct xport *xport, const struct ofpbuf *packet)
1483 struct flow_wildcards *wc = &ctx->xout->wc;
1484 const struct xbridge *xbridge = ctx->xbridge;
1488 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1490 cfm_process_heartbeat(xport->cfm, packet);
1493 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1495 bfd_process_packet(xport->bfd, flow, packet);
1498 } else if (xport->xbundle && xport->xbundle->lacp
1499 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1501 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1504 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1506 stp_process_packet(xport, packet);
1515 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1518 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1519 struct flow_wildcards *wc = &ctx->xout->wc;
1520 struct flow *flow = &ctx->xin->flow;
1521 ovs_be16 flow_vlan_tci;
1522 uint32_t flow_skb_mark;
1523 uint8_t flow_nw_tos;
1524 odp_port_t out_port, odp_port;
1527 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1528 * before traversing a patch port. */
1529 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1532 xlate_report(ctx, "Nonexistent output port");
1534 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1535 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1537 } else if (check_stp && !xport_stp_forward_state(xport)) {
1538 xlate_report(ctx, "STP not in forwarding state, skipping output");
1542 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1543 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1548 const struct xport *peer = xport->peer;
1549 struct flow old_flow = ctx->xin->flow;
1550 enum slow_path_reason special;
1552 ctx->xbridge = peer->xbridge;
1553 flow->in_port.ofp_port = peer->ofp_port;
1554 flow->metadata = htonll(0);
1555 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1556 memset(flow->regs, 0, sizeof flow->regs);
1558 special = process_special(ctx, &ctx->xin->flow, peer,
1561 ctx->xout->slow = special;
1562 } else if (may_receive(peer, ctx)) {
1563 if (xport_stp_forward_state(peer)) {
1564 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1566 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1567 * learning action look at the packet, then drop it. */
1568 struct flow old_base_flow = ctx->base_flow;
1569 size_t old_size = ctx->xout->odp_actions.size;
1570 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1571 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1572 ctx->xout->mirrors = old_mirrors;
1573 ctx->base_flow = old_base_flow;
1574 ctx->xout->odp_actions.size = old_size;
1578 ctx->xin->flow = old_flow;
1579 ctx->xbridge = xport->xbridge;
1581 if (ctx->xin->resubmit_stats) {
1582 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1583 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1589 flow_vlan_tci = flow->vlan_tci;
1590 flow_skb_mark = flow->skb_mark;
1591 flow_nw_tos = flow->nw_tos;
1593 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1594 wc->masks.nw_tos |= IP_ECN_MASK;
1595 flow->nw_tos &= ~IP_DSCP_MASK;
1596 flow->nw_tos |= dscp;
1599 if (xport->is_tunnel) {
1600 /* Save tunnel metadata so that changes made due to
1601 * the Logical (tunnel) Port are not visible for any further
1602 * matches, while explicit set actions on tunnel metadata are.
1604 struct flow_tnl flow_tnl = flow->tunnel;
1605 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1606 if (odp_port == ODPP_NONE) {
1607 xlate_report(ctx, "Tunneling decided against output");
1608 goto out; /* restore flow_nw_tos */
1610 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1611 xlate_report(ctx, "Not tunneling to our own address");
1612 goto out; /* restore flow_nw_tos */
1614 if (ctx->xin->resubmit_stats) {
1615 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1617 out_port = odp_port;
1618 commit_odp_tunnel_action(flow, &ctx->base_flow,
1619 &ctx->xout->odp_actions);
1620 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1622 ofp_port_t vlandev_port;
1624 odp_port = xport->odp_port;
1625 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1626 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1628 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1630 if (vlandev_port == ofp_port) {
1631 out_port = odp_port;
1633 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1634 flow->vlan_tci = htons(0);
1636 flow->skb_mark &= ~IPSEC_MARK;
1639 if (out_port != ODPP_NONE) {
1640 commit_odp_actions(flow, &ctx->base_flow,
1641 &ctx->xout->odp_actions, &ctx->xout->wc);
1642 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1645 ctx->sflow_odp_port = odp_port;
1646 ctx->sflow_n_outputs++;
1647 ctx->xout->nf_output_iface = ofp_port;
1652 flow->vlan_tci = flow_vlan_tci;
1653 flow->skb_mark = flow_skb_mark;
1654 flow->nw_tos = flow_nw_tos;
1658 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1660 compose_output_action__(ctx, ofp_port, true);
1663 /* Common rule processing in one place to avoid duplicating code. */
1664 static struct rule_dpif *
1665 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
1668 if (ctx->xin->resubmit_hook) {
1669 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1671 if (rule == NULL && may_packet_in) {
1672 struct xport *xport;
1675 * check if table configuration flags
1676 * OFPTC_TABLE_MISS_CONTROLLER, default.
1677 * OFPTC_TABLE_MISS_CONTINUE,
1678 * OFPTC_TABLE_MISS_DROP
1679 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1680 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1681 rule = choose_miss_rule(xport ? xport->config : 0,
1682 ctx->xbridge->miss_rule,
1683 ctx->xbridge->no_packet_in_rule);
1685 if (rule && ctx->xin->resubmit_stats) {
1686 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1692 xlate_table_action(struct xlate_ctx *ctx,
1693 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1695 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1696 struct rule_dpif *rule;
1697 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1698 uint8_t old_table_id = ctx->table_id;
1700 ctx->table_id = table_id;
1702 /* Look up a flow with 'in_port' as the input port. */
1703 ctx->xin->flow.in_port.ofp_port = in_port;
1704 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1705 &ctx->xin->flow, &ctx->xout->wc,
1708 /* Restore the original input port. Otherwise OFPP_NORMAL and
1709 * OFPP_IN_PORT will have surprising behavior. */
1710 ctx->xin->flow.in_port.ofp_port = old_in_port;
1712 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1715 struct rule_dpif *old_rule = ctx->rule;
1719 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1720 ctx->rule = old_rule;
1724 ctx->table_id = old_table_id;
1726 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1728 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1729 MAX_RESUBMIT_RECURSION);
1734 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1735 const struct ofpact_resubmit *resubmit)
1740 in_port = resubmit->in_port;
1741 if (in_port == OFPP_IN_PORT) {
1742 in_port = ctx->xin->flow.in_port.ofp_port;
1745 table_id = resubmit->table_id;
1746 if (table_id == 255) {
1747 table_id = ctx->table_id;
1750 xlate_table_action(ctx, in_port, table_id, false);
1754 flood_packets(struct xlate_ctx *ctx, bool all)
1756 const struct xport *xport;
1758 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1759 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1764 compose_output_action__(ctx, xport->ofp_port, false);
1765 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1766 compose_output_action(ctx, xport->ofp_port);
1770 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1774 execute_controller_action(struct xlate_ctx *ctx, int len,
1775 enum ofp_packet_in_reason reason,
1776 uint16_t controller_id)
1778 struct ofputil_packet_in *pin;
1779 struct ofpbuf *packet;
1782 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1783 ctx->xout->slow = SLOW_CONTROLLER;
1784 if (!ctx->xin->packet) {
1788 packet = ofpbuf_clone(ctx->xin->packet);
1790 key.skb_priority = 0;
1792 memset(&key.tunnel, 0, sizeof key.tunnel);
1794 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1795 &ctx->xout->odp_actions, &ctx->xout->wc);
1797 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1798 ctx->xout->odp_actions.size, NULL, NULL);
1800 pin = xmalloc(sizeof *pin);
1801 pin->packet_len = packet->size;
1802 pin->packet = ofpbuf_steal_data(packet);
1803 pin->reason = reason;
1804 pin->controller_id = controller_id;
1805 pin->table_id = ctx->table_id;
1806 pin->cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1808 pin->send_len = len;
1809 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1811 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1812 ofpbuf_delete(packet);
1816 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1818 struct flow_wildcards *wc = &ctx->xout->wc;
1819 struct flow *flow = &ctx->xin->flow;
1821 ovs_assert(eth_type_mpls(eth_type));
1823 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1824 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1826 if (flow->mpls_depth) {
1827 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1833 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1834 label = htonl(0x2); /* IPV6 Explicit Null. */
1836 label = htonl(0x0); /* IPV4 Explicit Null. */
1838 wc->masks.nw_tos |= IP_DSCP_MASK;
1839 wc->masks.nw_ttl = 0xff;
1840 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1841 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1842 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1843 flow->mpls_depth = 1;
1845 flow->dl_type = eth_type;
1849 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1851 struct flow_wildcards *wc = &ctx->xout->wc;
1852 struct flow *flow = &ctx->xin->flow;
1854 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1855 ovs_assert(!eth_type_mpls(eth_type));
1857 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1858 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1860 if (flow->mpls_depth) {
1862 flow->mpls_lse = htonl(0);
1863 if (!flow->mpls_depth) {
1864 flow->dl_type = eth_type;
1870 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1872 struct flow *flow = &ctx->xin->flow;
1874 if (!is_ip_any(flow)) {
1878 ctx->xout->wc.masks.nw_ttl = 0xff;
1879 if (flow->nw_ttl > 1) {
1885 for (i = 0; i < ids->n_controllers; i++) {
1886 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1890 /* Stop processing for current table. */
1896 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1898 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1902 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1903 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1908 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1910 struct flow *flow = &ctx->xin->flow;
1911 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1912 struct flow_wildcards *wc = &ctx->xout->wc;
1914 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1916 if (!eth_type_mpls(flow->dl_type)) {
1922 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1925 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1927 /* Stop processing for current table. */
1933 xlate_output_action(struct xlate_ctx *ctx,
1934 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1936 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1938 ctx->xout->nf_output_iface = NF_OUT_DROP;
1942 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1945 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1952 flood_packets(ctx, false);
1955 flood_packets(ctx, true);
1957 case OFPP_CONTROLLER:
1958 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1964 if (port != ctx->xin->flow.in_port.ofp_port) {
1965 compose_output_action(ctx, port);
1967 xlate_report(ctx, "skipping output to input port");
1972 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1973 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1974 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1975 ctx->xout->nf_output_iface = prev_nf_output_iface;
1976 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1977 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1978 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1983 xlate_output_reg_action(struct xlate_ctx *ctx,
1984 const struct ofpact_output_reg *or)
1986 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1987 if (port <= UINT16_MAX) {
1988 union mf_subvalue value;
1990 memset(&value, 0xff, sizeof value);
1991 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1992 xlate_output_action(ctx, u16_to_ofp(port),
1993 or->max_len, false);
1998 xlate_enqueue_action(struct xlate_ctx *ctx,
1999 const struct ofpact_enqueue *enqueue)
2001 ofp_port_t ofp_port = enqueue->port;
2002 uint32_t queue_id = enqueue->queue;
2003 uint32_t flow_priority, priority;
2006 /* Translate queue to priority. */
2007 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2009 /* Fall back to ordinary output action. */
2010 xlate_output_action(ctx, enqueue->port, 0, false);
2014 /* Check output port. */
2015 if (ofp_port == OFPP_IN_PORT) {
2016 ofp_port = ctx->xin->flow.in_port.ofp_port;
2017 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2021 /* Add datapath actions. */
2022 flow_priority = ctx->xin->flow.skb_priority;
2023 ctx->xin->flow.skb_priority = priority;
2024 compose_output_action(ctx, ofp_port);
2025 ctx->xin->flow.skb_priority = flow_priority;
2027 /* Update NetFlow output port. */
2028 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2029 ctx->xout->nf_output_iface = ofp_port;
2030 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2031 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2036 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2038 uint32_t skb_priority;
2040 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2041 ctx->xin->flow.skb_priority = skb_priority;
2043 /* Couldn't translate queue to a priority. Nothing to do. A warning
2044 * has already been logged. */
2049 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2051 const struct xbridge *xbridge = xbridge_;
2062 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2065 port = get_ofp_port(xbridge, ofp_port);
2066 return port ? port->may_enable : false;
2071 xlate_bundle_action(struct xlate_ctx *ctx,
2072 const struct ofpact_bundle *bundle)
2076 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2078 CONST_CAST(struct xbridge *, ctx->xbridge));
2079 if (bundle->dst.field) {
2080 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2083 xlate_output_action(ctx, port, 0, false);
2088 xlate_learn_action(struct xlate_ctx *ctx,
2089 const struct ofpact_learn *learn)
2091 struct ofputil_flow_mod *fm;
2092 struct ofpbuf ofpacts;
2094 ctx->xout->has_learn = true;
2096 learn_mask(learn, &ctx->xout->wc);
2098 if (!ctx->xin->may_learn) {
2102 fm = xmalloc(sizeof *fm);
2103 ofpbuf_init(&ofpacts, 0);
2104 learn_execute(learn, &ctx->xin->flow, fm, &ofpacts);
2106 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
2109 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
2110 * means "infinite". */
2112 reduce_timeout(uint16_t max, uint16_t *timeout)
2114 if (max && (!*timeout || *timeout > max)) {
2120 xlate_fin_timeout(struct xlate_ctx *ctx,
2121 const struct ofpact_fin_timeout *oft)
2123 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2124 struct rule_dpif *rule = ctx->rule;
2126 ovs_mutex_lock(&rule->up.ofproto->expirable_mutex);
2127 if (list_is_empty(&rule->up.expirable)) {
2128 list_insert(&rule->up.ofproto->expirable, &rule->up.expirable);
2130 ovs_mutex_unlock(&rule->up.ofproto->expirable_mutex);
2132 ovs_mutex_lock(&rule->up.timeout_mutex);
2133 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
2134 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
2135 ovs_mutex_unlock(&rule->up.timeout_mutex);
2140 xlate_sample_action(struct xlate_ctx *ctx,
2141 const struct ofpact_sample *os)
2143 union user_action_cookie cookie;
2144 /* Scale the probability from 16-bit to 32-bit while representing
2145 * the same percentage. */
2146 uint32_t probability = (os->probability << 16) | os->probability;
2148 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2149 &ctx->xout->odp_actions, &ctx->xout->wc);
2151 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2152 os->obs_domain_id, os->obs_point_id, &cookie);
2153 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2154 probability, &cookie, sizeof cookie.flow_sample);
2158 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2160 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2161 ? OFPUTIL_PC_NO_RECV_STP
2162 : OFPUTIL_PC_NO_RECV)) {
2166 /* Only drop packets here if both forwarding and learning are
2167 * disabled. If just learning is enabled, we need to have
2168 * OFPP_NORMAL and the learning action have a look at the packet
2169 * before we can drop it. */
2170 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2178 tunnel_ecn_ok(struct xlate_ctx *ctx)
2180 if (is_ip_any(&ctx->base_flow)
2181 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
2182 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
2183 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
2184 " but is not ECN capable");
2187 /* Set the ECN CE value in the tunneled packet. */
2188 ctx->xin->flow.nw_tos |= IP_ECN_CE;
2196 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2197 struct xlate_ctx *ctx)
2199 struct flow_wildcards *wc = &ctx->xout->wc;
2200 struct flow *flow = &ctx->xin->flow;
2201 bool was_evictable = true;
2202 const struct ofpact *a;
2205 /* Don't let the rule we're working on get evicted underneath us. */
2206 was_evictable = ctx->rule->up.evictable;
2207 ctx->rule->up.evictable = false;
2210 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2211 struct ofpact_controller *controller;
2212 const struct ofpact_metadata *metadata;
2220 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2221 ofpact_get_OUTPUT(a)->max_len, true);
2224 case OFPACT_CONTROLLER:
2225 controller = ofpact_get_CONTROLLER(a);
2226 execute_controller_action(ctx, controller->max_len,
2228 controller->controller_id);
2231 case OFPACT_ENQUEUE:
2232 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2235 case OFPACT_SET_VLAN_VID:
2236 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2237 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2238 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2242 case OFPACT_SET_VLAN_PCP:
2243 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2244 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2246 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2250 case OFPACT_STRIP_VLAN:
2251 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2252 flow->vlan_tci = htons(0);
2255 case OFPACT_PUSH_VLAN:
2256 /* XXX 802.1AD(QinQ) */
2257 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2258 flow->vlan_tci = htons(VLAN_CFI);
2261 case OFPACT_SET_ETH_SRC:
2262 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2263 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2266 case OFPACT_SET_ETH_DST:
2267 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2268 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2271 case OFPACT_SET_IPV4_SRC:
2272 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2273 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2274 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2278 case OFPACT_SET_IPV4_DST:
2279 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2280 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2281 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2285 case OFPACT_SET_IPV4_DSCP:
2286 wc->masks.nw_tos |= IP_DSCP_MASK;
2287 /* OpenFlow 1.0 only supports IPv4. */
2288 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2289 flow->nw_tos &= ~IP_DSCP_MASK;
2290 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2294 case OFPACT_SET_L4_SRC_PORT:
2295 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2296 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2297 if (is_ip_any(flow)) {
2298 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2302 case OFPACT_SET_L4_DST_PORT:
2303 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2304 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2305 if (is_ip_any(flow)) {
2306 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2310 case OFPACT_RESUBMIT:
2311 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2314 case OFPACT_SET_TUNNEL:
2315 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2318 case OFPACT_SET_QUEUE:
2319 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2322 case OFPACT_POP_QUEUE:
2323 flow->skb_priority = ctx->orig_skb_priority;
2326 case OFPACT_REG_MOVE:
2327 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2330 case OFPACT_REG_LOAD:
2331 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2334 case OFPACT_STACK_PUSH:
2335 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2339 case OFPACT_STACK_POP:
2340 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2344 case OFPACT_PUSH_MPLS:
2345 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2348 case OFPACT_POP_MPLS:
2349 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2352 case OFPACT_SET_MPLS_TTL:
2353 if (compose_set_mpls_ttl_action(ctx,
2354 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2359 case OFPACT_DEC_MPLS_TTL:
2360 if (compose_dec_mpls_ttl_action(ctx)) {
2365 case OFPACT_DEC_TTL:
2366 wc->masks.nw_ttl = 0xff;
2367 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2373 /* Nothing to do. */
2376 case OFPACT_MULTIPATH:
2377 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2381 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2384 case OFPACT_OUTPUT_REG:
2385 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2389 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2396 case OFPACT_FIN_TIMEOUT:
2397 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2398 ctx->xout->has_fin_timeout = true;
2399 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2402 case OFPACT_CLEAR_ACTIONS:
2404 * Nothing to do because writa-actions is not supported for now.
2405 * When writa-actions is supported, clear-actions also must
2406 * be supported at the same time.
2410 case OFPACT_WRITE_METADATA:
2411 metadata = ofpact_get_WRITE_METADATA(a);
2412 flow->metadata &= ~metadata->mask;
2413 flow->metadata |= metadata->metadata & metadata->mask;
2417 /* Not implemented yet. */
2420 case OFPACT_GOTO_TABLE: {
2421 /* It is assumed that goto-table is the last action. */
2422 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2424 ovs_assert(ctx->table_id < ogt->table_id);
2425 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2426 ogt->table_id, true);
2431 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2438 ctx->rule->up.evictable = was_evictable;
2443 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2444 const struct flow *flow, struct rule_dpif *rule,
2445 uint8_t tcp_flags, const struct ofpbuf *packet)
2447 xin->ofproto = ofproto;
2449 xin->packet = packet;
2450 xin->may_learn = packet != NULL;
2452 xin->ofpacts = NULL;
2453 xin->ofpacts_len = 0;
2454 xin->tcp_flags = tcp_flags;
2455 xin->resubmit_hook = NULL;
2456 xin->report_hook = NULL;
2457 xin->resubmit_stats = NULL;
2461 xlate_out_uninit(struct xlate_out *xout)
2464 ofpbuf_uninit(&xout->odp_actions);
2468 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2469 * into datapath actions, using 'ctx', and discards the datapath actions. */
2471 xlate_actions_for_side_effects(struct xlate_in *xin)
2473 struct xlate_out xout;
2475 xlate_actions(xin, &xout);
2476 xlate_out_uninit(&xout);
2480 xlate_report(struct xlate_ctx *ctx, const char *s)
2482 if (ctx->xin->report_hook) {
2483 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2488 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2491 dst->slow = src->slow;
2492 dst->has_learn = src->has_learn;
2493 dst->has_normal = src->has_normal;
2494 dst->has_fin_timeout = src->has_fin_timeout;
2495 dst->nf_output_iface = src->nf_output_iface;
2496 dst->mirrors = src->mirrors;
2498 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2499 sizeof dst->odp_actions_stub);
2500 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2501 src->odp_actions.size);
2504 static struct skb_priority_to_dscp *
2505 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2507 struct skb_priority_to_dscp *pdscp;
2510 hash = hash_int(skb_priority, 0);
2511 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2512 if (pdscp->skb_priority == skb_priority) {
2520 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2523 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2524 *dscp = pdscp ? pdscp->dscp : 0;
2525 return pdscp != NULL;
2529 clear_skb_priorities(struct xport *xport)
2531 struct skb_priority_to_dscp *pdscp, *next;
2533 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2534 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2540 actions_output_to_local_port(const struct xlate_ctx *ctx)
2542 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2543 const struct nlattr *a;
2546 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2547 ctx->xout->odp_actions.size) {
2548 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2549 && nl_attr_get_odp_port(a) == local_odp_port) {
2556 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2557 * into datapath actions in 'odp_actions', using 'ctx'. */
2559 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2561 struct flow_wildcards *wc = &xout->wc;
2562 struct flow *flow = &xin->flow;
2564 enum slow_path_reason special;
2565 const struct ofpact *ofpacts;
2566 struct xport *in_port;
2567 struct flow orig_flow;
2568 struct xlate_ctx ctx;
2571 COVERAGE_INC(xlate_actions);
2573 ovs_rwlock_rdlock(&xlate_rwlock);
2575 /* Flow initialization rules:
2576 * - 'base_flow' must match the kernel's view of the packet at the
2577 * time that action processing starts. 'flow' represents any
2578 * transformations we wish to make through actions.
2579 * - By default 'base_flow' and 'flow' are the same since the input
2580 * packet matches the output before any actions are applied.
2581 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2582 * of the received packet as seen by the kernel. If we later output
2583 * to another device without any modifications this will cause us to
2584 * insert a new tag since the original one was stripped off by the
2586 * - Tunnel metadata as received is retained in 'flow'. This allows
2587 * tunnel metadata matching also in later tables.
2588 * Since a kernel action for setting the tunnel metadata will only be
2589 * generated with actual tunnel output, changing the tunnel metadata
2590 * values in 'flow' (such as tun_id) will only have effect with a later
2591 * tunnel output action.
2592 * - Tunnel 'base_flow' is completely cleared since that is what the
2593 * kernel does. If we wish to maintain the original values an action
2594 * needs to be generated. */
2599 ctx.xout->has_learn = false;
2600 ctx.xout->has_normal = false;
2601 ctx.xout->has_fin_timeout = false;
2602 ctx.xout->nf_output_iface = NF_OUT_DROP;
2603 ctx.xout->mirrors = 0;
2604 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2605 sizeof ctx.xout->odp_actions_stub);
2606 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2608 ctx.xbridge = xbridge_lookup(xin->ofproto);
2613 ctx.rule = xin->rule;
2615 ctx.base_flow = *flow;
2616 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2617 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2619 flow_wildcards_init_catchall(wc);
2620 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2621 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2622 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2623 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2625 if (tnl_port_should_receive(&ctx.xin->flow)) {
2626 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
2627 /* skb_mark is currently used only by tunnels but that will likely
2628 * change in the future. */
2629 memset(&wc->masks.skb_mark, 0xff, sizeof wc->masks.skb_mark);
2631 if (ctx.xbridge->has_netflow) {
2632 netflow_mask_wc(flow, wc);
2636 ctx.orig_skb_priority = flow->skb_priority;
2641 ofpacts = xin->ofpacts;
2642 ofpacts_len = xin->ofpacts_len;
2643 } else if (xin->rule) {
2644 ofpacts = xin->rule->up.ofpacts;
2645 ofpacts_len = xin->rule->up.ofpacts_len;
2650 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2652 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2653 /* Do this conditionally because the copy is expensive enough that it
2654 * shows up in profiles. */
2658 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2659 switch (ctx.xbridge->frag) {
2660 case OFPC_FRAG_NORMAL:
2661 /* We must pretend that transport ports are unavailable. */
2662 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2663 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2666 case OFPC_FRAG_DROP:
2669 case OFPC_FRAG_REASM:
2672 case OFPC_FRAG_NX_MATCH:
2673 /* Nothing to do. */
2676 case OFPC_INVALID_TTL_TO_CONTROLLER:
2681 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2682 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2684 ctx.xout->slow = special;
2686 size_t sample_actions_len;
2688 if (flow->in_port.ofp_port
2689 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2690 flow->in_port.ofp_port,
2692 ctx.base_flow.vlan_tci = 0;
2695 add_sflow_action(&ctx);
2696 add_ipfix_action(&ctx);
2697 sample_actions_len = ctx.xout->odp_actions.size;
2699 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
2700 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2702 /* We've let OFPP_NORMAL and the learning action look at the
2703 * packet, so drop it now if forwarding is disabled. */
2704 if (in_port && !xport_stp_forward_state(in_port)) {
2705 ctx.xout->odp_actions.size = sample_actions_len;
2709 if (ctx.xbridge->has_in_band
2710 && in_band_must_output_to_local_port(flow)
2711 && !actions_output_to_local_port(&ctx)) {
2712 compose_output_action(&ctx, OFPP_LOCAL);
2715 fix_sflow_action(&ctx);
2717 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2718 add_mirror_actions(&ctx, &orig_flow);
2722 ofpbuf_uninit(&ctx.stack);
2724 /* Clear the metadata and register wildcard masks, because we won't
2725 * use non-header fields as part of the cache. */
2726 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2727 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
2730 ovs_rwlock_unlock(&xlate_rwlock);