1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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)
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);
645 xport_stp_listen_state(const struct xport *xport)
647 struct stp_port *sp = xport_get_stp_port(xport);
648 return stp_listen_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
651 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
652 * were used to make the determination.*/
654 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
656 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
657 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
661 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
663 struct stp_port *sp = xport_get_stp_port(xport);
664 struct ofpbuf payload = *packet;
665 struct eth_header *eth = payload.data;
667 /* Sink packets on ports that have STP disabled when the bridge has
669 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
673 /* Trim off padding on payload. */
674 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
675 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
678 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
679 stp_received_bpdu(sp, payload.data, payload.size);
683 static struct xport *
684 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
688 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
690 if (xport->ofp_port == ofp_port) {
698 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
700 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
701 return xport ? xport->odp_port : ODPP_NONE;
705 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
707 return (bundle->vlan_mode != PORT_VLAN_ACCESS
708 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
712 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
714 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
718 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
720 return xbundle != &ofpp_none_bundle
721 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
726 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
728 return xbundle != &ofpp_none_bundle
729 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
734 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
736 return xbundle != &ofpp_none_bundle
737 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
741 static struct xbundle *
742 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
743 bool warn, struct xport **in_xportp)
747 /* Find the port and bundle for the received packet. */
748 xport = get_ofp_port(xbridge, in_port);
752 if (xport && xport->xbundle) {
753 return xport->xbundle;
756 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
757 * which a controller may use as the ingress port for traffic that
759 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
760 ofpp_none_bundle.name = "OFPP_NONE";
761 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
762 return &ofpp_none_bundle;
765 /* Odd. A few possible reasons here:
767 * - We deleted a port but there are still a few packets queued up
770 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
771 * we don't know about.
773 * - The ofproto client didn't configure the port as part of a bundle.
774 * This is particularly likely to happen if a packet was received on the
775 * port after it was created, but before the client had a chance to
776 * configure its bundle.
779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
781 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
782 "port %"PRIu16, xbridge->name, in_port);
788 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
790 const struct xbridge *xbridge = ctx->xbridge;
791 mirror_mask_t mirrors;
792 struct xbundle *in_xbundle;
796 mirrors = ctx->xout->mirrors;
797 ctx->xout->mirrors = 0;
799 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
800 ctx->xin->packet != NULL, NULL);
804 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
806 /* Drop frames on bundles reserved for mirroring. */
807 if (xbundle_mirror_out(xbridge, in_xbundle)) {
808 if (ctx->xin->packet != NULL) {
809 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
810 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
811 "%s, which is reserved exclusively for mirroring",
812 ctx->xbridge->name, in_xbundle->name);
814 ofpbuf_clear(&ctx->xout->odp_actions);
819 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
820 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
823 vlan = input_vid_to_vlan(in_xbundle, vid);
829 /* Restore the original packet before adding the mirror actions. */
830 ctx->xin->flow = *orig_flow;
833 mirror_mask_t dup_mirrors;
834 struct ofbundle *out;
835 unsigned long *vlans;
840 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
841 &vlans, &dup_mirrors, &out, &out_vlan);
842 ovs_assert(has_mirror);
845 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
847 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
850 if (!vlan_mirrored) {
851 mirrors = zero_rightmost_1bit(mirrors);
855 mirrors &= ~dup_mirrors;
856 ctx->xout->mirrors |= dup_mirrors;
858 struct xbundle *out_xbundle = xbundle_lookup(out);
860 output_normal(ctx, out_xbundle, vlan);
862 } else if (vlan != out_vlan
863 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
864 struct xbundle *xbundle;
866 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
867 if (xbundle_includes_vlan(xbundle, out_vlan)
868 && !xbundle_mirror_out(xbridge, xbundle)) {
869 output_normal(ctx, xbundle, out_vlan);
876 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
877 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
878 * the bundle on which the packet was received, returns the VLAN to which the
881 * Both 'vid' and the return value are in the range 0...4095. */
883 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
885 switch (in_xbundle->vlan_mode) {
886 case PORT_VLAN_ACCESS:
887 return in_xbundle->vlan;
890 case PORT_VLAN_TRUNK:
893 case PORT_VLAN_NATIVE_UNTAGGED:
894 case PORT_VLAN_NATIVE_TAGGED:
895 return vid ? vid : in_xbundle->vlan;
902 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
903 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
906 * 'vid' should be the VID obtained from the 802.1Q header that was received as
907 * part of a packet (specify 0 if there was no 802.1Q header), in the range
910 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
912 /* Allow any VID on the OFPP_NONE port. */
913 if (in_xbundle == &ofpp_none_bundle) {
917 switch (in_xbundle->vlan_mode) {
918 case PORT_VLAN_ACCESS:
921 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
922 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
923 "packet received on port %s configured as VLAN "
924 "%"PRIu16" access port", vid, in_xbundle->name,
931 case PORT_VLAN_NATIVE_UNTAGGED:
932 case PORT_VLAN_NATIVE_TAGGED:
934 /* Port must always carry its native VLAN. */
938 case PORT_VLAN_TRUNK:
939 if (!xbundle_includes_vlan(in_xbundle, vid)) {
941 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
942 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
943 "received on port %s not configured for trunking "
944 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
956 /* Given 'vlan', the VLAN that a packet belongs to, and
957 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
958 * that should be included in the 802.1Q header. (If the return value is 0,
959 * then the 802.1Q header should only be included in the packet if there is a
962 * Both 'vlan' and the return value are in the range 0...4095. */
964 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
966 switch (out_xbundle->vlan_mode) {
967 case PORT_VLAN_ACCESS:
970 case PORT_VLAN_TRUNK:
971 case PORT_VLAN_NATIVE_TAGGED:
974 case PORT_VLAN_NATIVE_UNTAGGED:
975 return vlan == out_xbundle->vlan ? 0 : vlan;
983 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
986 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
988 ovs_be16 tci, old_tci;
991 vid = output_vlan_to_vid(out_xbundle, vlan);
992 if (list_is_empty(&out_xbundle->xports)) {
993 /* Partially configured bundle with no slaves. Drop the packet. */
995 } else if (!out_xbundle->bond) {
996 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
999 struct ofport_dpif *ofport;
1001 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1002 &ctx->xout->wc, vid);
1003 xport = xport_lookup(ofport);
1006 /* No slaves enabled, so drop packet. */
1011 old_tci = *flow_tci;
1013 if (tci || out_xbundle->use_priority_tags) {
1014 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1016 tci |= htons(VLAN_CFI);
1021 compose_output_action(ctx, xport->ofp_port);
1022 *flow_tci = old_tci;
1025 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1026 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1027 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1029 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1031 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1035 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1036 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1040 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1041 if (flow->nw_proto == ARP_OP_REPLY) {
1043 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1044 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1045 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1047 return flow->nw_src == flow->nw_dst;
1053 /* Checks whether a MAC learning update is necessary for MAC learning table
1054 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1057 * Most packets processed through the MAC learning table do not actually
1058 * change it in any way. This function requires only a read lock on the MAC
1059 * learning table, so it is much cheaper in this common case.
1061 * Keep the code here synchronized with that in update_learning_table__()
1064 is_mac_learning_update_needed(const struct mac_learning *ml,
1065 const struct flow *flow,
1066 struct flow_wildcards *wc,
1067 int vlan, struct xbundle *in_xbundle)
1068 OVS_REQ_RDLOCK(ml->rwlock)
1070 struct mac_entry *mac;
1072 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1076 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1077 if (!mac || mac_entry_age(ml, mac)) {
1081 if (is_gratuitous_arp(flow, wc)) {
1082 /* We don't want to learn from gratuitous ARP packets that are
1083 * reflected back over bond slaves so we lock the learning table. */
1084 if (!in_xbundle->bond) {
1086 } else if (mac_entry_is_grat_arp_locked(mac)) {
1091 return mac->port.p != in_xbundle->ofbundle;
1095 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1096 * received on 'in_xbundle' in 'vlan'.
1098 * This code repeats all the checks in is_mac_learning_update_needed() because
1099 * the lock was released between there and here and thus the MAC learning state
1100 * could have changed.
1102 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1105 update_learning_table__(const struct xbridge *xbridge,
1106 const struct flow *flow, struct flow_wildcards *wc,
1107 int vlan, struct xbundle *in_xbundle)
1108 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1110 struct mac_entry *mac;
1112 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1116 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1117 if (is_gratuitous_arp(flow, wc)) {
1118 /* We don't want to learn from gratuitous ARP packets that are
1119 * reflected back over bond slaves so we lock the learning table. */
1120 if (!in_xbundle->bond) {
1121 mac_entry_set_grat_arp_lock(mac);
1122 } else if (mac_entry_is_grat_arp_locked(mac)) {
1127 if (mac->port.p != in_xbundle->ofbundle) {
1128 /* The log messages here could actually be useful in debugging,
1129 * so keep the rate limit relatively high. */
1130 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1132 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1133 "on port %s in VLAN %d",
1134 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1135 in_xbundle->name, vlan);
1137 mac->port.p = in_xbundle->ofbundle;
1138 mac_learning_changed(xbridge->ml);
1143 update_learning_table(const struct xbridge *xbridge,
1144 const struct flow *flow, struct flow_wildcards *wc,
1145 int vlan, struct xbundle *in_xbundle)
1149 /* Don't learn the OFPP_NONE port. */
1150 if (in_xbundle == &ofpp_none_bundle) {
1154 /* First try the common case: no change to MAC learning table. */
1155 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1156 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1158 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1161 /* Slow path: MAC learning table might need an update. */
1162 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1163 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1164 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1168 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1169 * dropped. Returns true if they may be forwarded, false if they should be
1172 * 'in_port' must be the xport that corresponds to flow->in_port.
1173 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1175 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1176 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1177 * checked by input_vid_is_valid().
1179 * May also add tags to '*tags', although the current implementation only does
1180 * so in one special case.
1183 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1186 struct xbundle *in_xbundle = in_port->xbundle;
1187 const struct xbridge *xbridge = ctx->xbridge;
1188 struct flow *flow = &ctx->xin->flow;
1190 /* Drop frames for reserved multicast addresses
1191 * only if forward_bpdu option is absent. */
1192 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1193 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1197 if (in_xbundle->bond) {
1198 struct mac_entry *mac;
1200 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1206 xlate_report(ctx, "bonding refused admissibility, dropping");
1209 case BV_DROP_IF_MOVED:
1210 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1211 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1212 if (mac && mac->port.p != in_xbundle->ofbundle &&
1213 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1214 || mac_entry_is_grat_arp_locked(mac))) {
1215 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1216 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1220 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1229 xlate_normal(struct xlate_ctx *ctx)
1231 struct flow_wildcards *wc = &ctx->xout->wc;
1232 struct flow *flow = &ctx->xin->flow;
1233 struct xbundle *in_xbundle;
1234 struct xport *in_port;
1235 struct mac_entry *mac;
1240 ctx->xout->has_normal = true;
1242 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1243 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1244 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1246 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1247 ctx->xin->packet != NULL, &in_port);
1249 xlate_report(ctx, "no input bundle, dropping");
1253 /* Drop malformed frames. */
1254 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1255 !(flow->vlan_tci & htons(VLAN_CFI))) {
1256 if (ctx->xin->packet != NULL) {
1257 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1258 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1259 "VLAN tag received on port %s",
1260 ctx->xbridge->name, in_xbundle->name);
1262 xlate_report(ctx, "partial VLAN tag, dropping");
1266 /* Drop frames on bundles reserved for mirroring. */
1267 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1268 if (ctx->xin->packet != NULL) {
1269 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1270 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1271 "%s, which is reserved exclusively for mirroring",
1272 ctx->xbridge->name, in_xbundle->name);
1274 xlate_report(ctx, "input port is mirror output port, dropping");
1279 vid = vlan_tci_to_vid(flow->vlan_tci);
1280 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1281 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1284 vlan = input_vid_to_vlan(in_xbundle, vid);
1286 /* Check other admissibility requirements. */
1287 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1291 /* Learn source MAC. */
1292 if (ctx->xin->may_learn) {
1293 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1296 /* Determine output bundle. */
1297 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1298 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1299 mac_port = mac ? mac->port.p : NULL;
1300 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1303 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1304 if (mac_xbundle && mac_xbundle != in_xbundle) {
1305 xlate_report(ctx, "forwarding to learned port");
1306 output_normal(ctx, mac_xbundle, vlan);
1307 } else if (!mac_xbundle) {
1308 xlate_report(ctx, "learned port is unknown, dropping");
1310 xlate_report(ctx, "learned port is input port, dropping");
1313 struct xbundle *xbundle;
1315 xlate_report(ctx, "no learned MAC for destination, flooding");
1316 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1317 if (xbundle != in_xbundle
1318 && xbundle_includes_vlan(xbundle, vlan)
1319 && xbundle->floodable
1320 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1321 output_normal(ctx, xbundle, vlan);
1324 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1328 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1329 * the number of packets out of UINT32_MAX to sample. The given
1330 * cookie is passed back in the callback for each sampled packet.
1333 compose_sample_action(const struct xbridge *xbridge,
1334 struct ofpbuf *odp_actions,
1335 const struct flow *flow,
1336 const uint32_t probability,
1337 const union user_action_cookie *cookie,
1338 const size_t cookie_size)
1340 size_t sample_offset, actions_offset;
1341 odp_port_t odp_port;
1345 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1347 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1349 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1351 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1352 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1353 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1355 nl_msg_end_nested(odp_actions, actions_offset);
1356 nl_msg_end_nested(odp_actions, sample_offset);
1357 return cookie_offset;
1361 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1362 odp_port_t odp_port, unsigned int n_outputs,
1363 union user_action_cookie *cookie)
1367 cookie->type = USER_ACTION_COOKIE_SFLOW;
1368 cookie->sflow.vlan_tci = vlan_tci;
1370 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1371 * port information") for the interpretation of cookie->output. */
1372 switch (n_outputs) {
1374 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1375 cookie->sflow.output = 0x40000000 | 256;
1379 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1381 cookie->sflow.output = ifindex;
1386 /* 0x80000000 means "multiple output ports. */
1387 cookie->sflow.output = 0x80000000 | n_outputs;
1392 /* Compose SAMPLE action for sFlow bridge sampling. */
1394 compose_sflow_action(const struct xbridge *xbridge,
1395 struct ofpbuf *odp_actions,
1396 const struct flow *flow,
1397 odp_port_t odp_port)
1399 uint32_t probability;
1400 union user_action_cookie cookie;
1402 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1406 probability = dpif_sflow_get_probability(xbridge->sflow);
1407 compose_sflow_cookie(xbridge, htons(0), odp_port,
1408 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1410 return compose_sample_action(xbridge, odp_actions, flow, probability,
1411 &cookie, sizeof cookie.sflow);
1415 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1416 uint32_t obs_domain_id, uint32_t obs_point_id,
1417 union user_action_cookie *cookie)
1419 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1420 cookie->flow_sample.probability = probability;
1421 cookie->flow_sample.collector_set_id = collector_set_id;
1422 cookie->flow_sample.obs_domain_id = obs_domain_id;
1423 cookie->flow_sample.obs_point_id = obs_point_id;
1427 compose_ipfix_cookie(union user_action_cookie *cookie)
1429 cookie->type = USER_ACTION_COOKIE_IPFIX;
1432 /* Compose SAMPLE action for IPFIX bridge sampling. */
1434 compose_ipfix_action(const struct xbridge *xbridge,
1435 struct ofpbuf *odp_actions,
1436 const struct flow *flow)
1438 uint32_t probability;
1439 union user_action_cookie cookie;
1441 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1445 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1446 compose_ipfix_cookie(&cookie);
1448 compose_sample_action(xbridge, odp_actions, flow, probability,
1449 &cookie, sizeof cookie.ipfix);
1452 /* SAMPLE action for sFlow must be first action in any given list of
1453 * actions. At this point we do not have all information required to
1454 * build it. So try to build sample action as complete as possible. */
1456 add_sflow_action(struct xlate_ctx *ctx)
1458 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1459 &ctx->xout->odp_actions,
1460 &ctx->xin->flow, ODPP_NONE);
1461 ctx->sflow_odp_port = 0;
1462 ctx->sflow_n_outputs = 0;
1465 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1466 * of actions, eventually after the SAMPLE action for sFlow. */
1468 add_ipfix_action(struct xlate_ctx *ctx)
1470 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1474 /* Fix SAMPLE action according to data collected while composing ODP actions.
1475 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1476 * USERSPACE action's user-cookie which is required for sflow. */
1478 fix_sflow_action(struct xlate_ctx *ctx)
1480 const struct flow *base = &ctx->base_flow;
1481 union user_action_cookie *cookie;
1483 if (!ctx->user_cookie_offset) {
1487 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1488 sizeof cookie->sflow);
1489 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1491 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1492 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1495 static enum slow_path_reason
1496 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1497 const struct xport *xport, const struct ofpbuf *packet)
1499 struct flow_wildcards *wc = &ctx->xout->wc;
1500 const struct xbridge *xbridge = ctx->xbridge;
1504 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1506 cfm_process_heartbeat(xport->cfm, packet);
1509 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1511 bfd_process_packet(xport->bfd, flow, packet);
1514 } else if (xport->xbundle && xport->xbundle->lacp
1515 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1517 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1520 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1522 stp_process_packet(xport, packet);
1531 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1534 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1535 struct flow_wildcards *wc = &ctx->xout->wc;
1536 struct flow *flow = &ctx->xin->flow;
1537 ovs_be16 flow_vlan_tci;
1538 uint32_t flow_pkt_mark;
1539 uint8_t flow_nw_tos;
1540 odp_port_t out_port, odp_port;
1543 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1544 * before traversing a patch port. */
1545 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1548 xlate_report(ctx, "Nonexistent output port");
1550 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1551 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1553 } else if (check_stp) {
1554 if (eth_addr_equals(ctx->base_flow.dl_dst, eth_addr_stp)) {
1555 if (!xport_stp_listen_state(xport)) {
1556 xlate_report(ctx, "STP not in listening state, "
1557 "skipping bpdu output");
1560 } else if (!xport_stp_forward_state(xport)) {
1561 xlate_report(ctx, "STP not in forwarding state, "
1567 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1568 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1573 const struct xport *peer = xport->peer;
1574 struct flow old_flow = ctx->xin->flow;
1575 enum slow_path_reason special;
1577 ctx->xbridge = peer->xbridge;
1578 flow->in_port.ofp_port = peer->ofp_port;
1579 flow->metadata = htonll(0);
1580 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1581 memset(flow->regs, 0, sizeof flow->regs);
1583 special = process_special(ctx, &ctx->xin->flow, peer,
1586 ctx->xout->slow = special;
1587 } else if (may_receive(peer, ctx)) {
1588 if (xport_stp_forward_state(peer)) {
1589 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1591 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1592 * learning action look at the packet, then drop it. */
1593 struct flow old_base_flow = ctx->base_flow;
1594 size_t old_size = ctx->xout->odp_actions.size;
1595 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1596 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1597 ctx->xout->mirrors = old_mirrors;
1598 ctx->base_flow = old_base_flow;
1599 ctx->xout->odp_actions.size = old_size;
1603 ctx->xin->flow = old_flow;
1604 ctx->xbridge = xport->xbridge;
1606 if (ctx->xin->resubmit_stats) {
1607 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1608 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1614 flow_vlan_tci = flow->vlan_tci;
1615 flow_pkt_mark = flow->pkt_mark;
1616 flow_nw_tos = flow->nw_tos;
1618 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1619 wc->masks.nw_tos |= IP_ECN_MASK;
1620 flow->nw_tos &= ~IP_DSCP_MASK;
1621 flow->nw_tos |= dscp;
1624 if (xport->is_tunnel) {
1625 /* Save tunnel metadata so that changes made due to
1626 * the Logical (tunnel) Port are not visible for any further
1627 * matches, while explicit set actions on tunnel metadata are.
1629 struct flow_tnl flow_tnl = flow->tunnel;
1630 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1631 if (odp_port == ODPP_NONE) {
1632 xlate_report(ctx, "Tunneling decided against output");
1633 goto out; /* restore flow_nw_tos */
1635 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1636 xlate_report(ctx, "Not tunneling to our own address");
1637 goto out; /* restore flow_nw_tos */
1639 if (ctx->xin->resubmit_stats) {
1640 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1642 out_port = odp_port;
1643 commit_odp_tunnel_action(flow, &ctx->base_flow,
1644 &ctx->xout->odp_actions);
1645 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1647 ofp_port_t vlandev_port;
1649 odp_port = xport->odp_port;
1650 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1651 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1653 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1655 if (vlandev_port == ofp_port) {
1656 out_port = odp_port;
1658 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1659 flow->vlan_tci = htons(0);
1663 if (out_port != ODPP_NONE) {
1664 commit_odp_actions(flow, &ctx->base_flow,
1665 &ctx->xout->odp_actions, &ctx->xout->wc);
1666 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1669 ctx->sflow_odp_port = odp_port;
1670 ctx->sflow_n_outputs++;
1671 ctx->xout->nf_output_iface = ofp_port;
1676 flow->vlan_tci = flow_vlan_tci;
1677 flow->pkt_mark = flow_pkt_mark;
1678 flow->nw_tos = flow_nw_tos;
1682 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1684 compose_output_action__(ctx, ofp_port, true);
1688 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1690 struct rule_dpif *old_rule = ctx->rule;
1691 struct rule_actions *actions;
1693 if (ctx->xin->resubmit_stats) {
1694 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1699 actions = rule_dpif_get_actions(rule);
1700 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1701 rule_actions_unref(actions);
1702 ctx->rule = old_rule;
1707 xlate_table_action(struct xlate_ctx *ctx,
1708 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1710 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1711 struct rule_dpif *rule;
1712 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1713 uint8_t old_table_id = ctx->table_id;
1715 ctx->table_id = table_id;
1717 /* Look up a flow with 'in_port' as the input port. Then restore the
1718 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1719 * have surprising behavior). */
1720 ctx->xin->flow.in_port.ofp_port = in_port;
1721 rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1722 &ctx->xin->flow, &ctx->xout->wc,
1724 ctx->xin->flow.in_port.ofp_port = old_in_port;
1726 if (ctx->xin->resubmit_hook) {
1727 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1730 if (!rule && may_packet_in) {
1731 struct xport *xport;
1734 * check if table configuration flags
1735 * OFPTC_TABLE_MISS_CONTROLLER, default.
1736 * OFPTC_TABLE_MISS_CONTINUE,
1737 * OFPTC_TABLE_MISS_DROP
1738 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1739 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1740 choose_miss_rule(xport ? xport->config : 0,
1741 ctx->xbridge->miss_rule,
1742 ctx->xbridge->no_packet_in_rule, &rule);
1745 xlate_recursively(ctx, rule);
1746 rule_dpif_unref(rule);
1749 ctx->table_id = old_table_id;
1751 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1753 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1754 MAX_RESUBMIT_RECURSION);
1759 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1760 const struct ofpact_resubmit *resubmit)
1765 in_port = resubmit->in_port;
1766 if (in_port == OFPP_IN_PORT) {
1767 in_port = ctx->xin->flow.in_port.ofp_port;
1770 table_id = resubmit->table_id;
1771 if (table_id == 255) {
1772 table_id = ctx->table_id;
1775 xlate_table_action(ctx, in_port, table_id, false);
1779 flood_packets(struct xlate_ctx *ctx, bool all)
1781 const struct xport *xport;
1783 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1784 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1789 compose_output_action__(ctx, xport->ofp_port, false);
1790 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1791 compose_output_action(ctx, xport->ofp_port);
1795 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1799 execute_controller_action(struct xlate_ctx *ctx, int len,
1800 enum ofp_packet_in_reason reason,
1801 uint16_t controller_id)
1803 struct ofputil_packet_in *pin;
1804 struct ofpbuf *packet;
1807 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1808 ctx->xout->slow = SLOW_CONTROLLER;
1809 if (!ctx->xin->packet) {
1813 packet = ofpbuf_clone(ctx->xin->packet);
1815 key.skb_priority = 0;
1817 memset(&key.tunnel, 0, sizeof key.tunnel);
1819 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1820 &ctx->xout->odp_actions, &ctx->xout->wc);
1822 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1823 ctx->xout->odp_actions.size, NULL, NULL);
1825 pin = xmalloc(sizeof *pin);
1826 pin->packet_len = packet->size;
1827 pin->packet = ofpbuf_steal_data(packet);
1828 pin->reason = reason;
1829 pin->controller_id = controller_id;
1830 pin->table_id = ctx->table_id;
1831 pin->cookie = ctx->rule ? rule_dpif_get_flow_cookie(ctx->rule) : 0;
1833 pin->send_len = len;
1834 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1836 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1837 ofpbuf_delete(packet);
1841 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1843 struct flow_wildcards *wc = &ctx->xout->wc;
1844 struct flow *flow = &ctx->xin->flow;
1846 ovs_assert(eth_type_mpls(eth_type));
1848 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1849 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1851 if (flow->mpls_depth) {
1852 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1858 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1859 label = htonl(0x2); /* IPV6 Explicit Null. */
1861 label = htonl(0x0); /* IPV4 Explicit Null. */
1863 wc->masks.nw_tos |= IP_DSCP_MASK;
1864 wc->masks.nw_ttl = 0xff;
1865 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1866 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1867 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1868 flow->mpls_depth = 1;
1870 flow->dl_type = eth_type;
1874 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1876 struct flow_wildcards *wc = &ctx->xout->wc;
1877 struct flow *flow = &ctx->xin->flow;
1879 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1880 ovs_assert(!eth_type_mpls(eth_type));
1882 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1883 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1885 if (flow->mpls_depth) {
1887 flow->mpls_lse = htonl(0);
1888 if (!flow->mpls_depth) {
1889 flow->dl_type = eth_type;
1895 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1897 struct flow *flow = &ctx->xin->flow;
1899 if (!is_ip_any(flow)) {
1903 ctx->xout->wc.masks.nw_ttl = 0xff;
1904 if (flow->nw_ttl > 1) {
1910 for (i = 0; i < ids->n_controllers; i++) {
1911 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1915 /* Stop processing for current table. */
1921 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1923 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1927 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1928 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1933 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1935 struct flow *flow = &ctx->xin->flow;
1936 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1937 struct flow_wildcards *wc = &ctx->xout->wc;
1939 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1941 if (!eth_type_mpls(flow->dl_type)) {
1947 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1950 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1952 /* Stop processing for current table. */
1958 xlate_output_action(struct xlate_ctx *ctx,
1959 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1961 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1963 ctx->xout->nf_output_iface = NF_OUT_DROP;
1967 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1970 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1977 flood_packets(ctx, false);
1980 flood_packets(ctx, true);
1982 case OFPP_CONTROLLER:
1983 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1989 if (port != ctx->xin->flow.in_port.ofp_port) {
1990 compose_output_action(ctx, port);
1992 xlate_report(ctx, "skipping output to input port");
1997 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1998 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1999 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2000 ctx->xout->nf_output_iface = prev_nf_output_iface;
2001 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2002 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2003 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2008 xlate_output_reg_action(struct xlate_ctx *ctx,
2009 const struct ofpact_output_reg *or)
2011 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2012 if (port <= UINT16_MAX) {
2013 union mf_subvalue value;
2015 memset(&value, 0xff, sizeof value);
2016 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2017 xlate_output_action(ctx, u16_to_ofp(port),
2018 or->max_len, false);
2023 xlate_enqueue_action(struct xlate_ctx *ctx,
2024 const struct ofpact_enqueue *enqueue)
2026 ofp_port_t ofp_port = enqueue->port;
2027 uint32_t queue_id = enqueue->queue;
2028 uint32_t flow_priority, priority;
2031 /* Translate queue to priority. */
2032 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2034 /* Fall back to ordinary output action. */
2035 xlate_output_action(ctx, enqueue->port, 0, false);
2039 /* Check output port. */
2040 if (ofp_port == OFPP_IN_PORT) {
2041 ofp_port = ctx->xin->flow.in_port.ofp_port;
2042 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2046 /* Add datapath actions. */
2047 flow_priority = ctx->xin->flow.skb_priority;
2048 ctx->xin->flow.skb_priority = priority;
2049 compose_output_action(ctx, ofp_port);
2050 ctx->xin->flow.skb_priority = flow_priority;
2052 /* Update NetFlow output port. */
2053 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2054 ctx->xout->nf_output_iface = ofp_port;
2055 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2056 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2061 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2063 uint32_t skb_priority;
2065 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2066 ctx->xin->flow.skb_priority = skb_priority;
2068 /* Couldn't translate queue to a priority. Nothing to do. A warning
2069 * has already been logged. */
2074 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2076 const struct xbridge *xbridge = xbridge_;
2087 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2090 port = get_ofp_port(xbridge, ofp_port);
2091 return port ? port->may_enable : false;
2096 xlate_bundle_action(struct xlate_ctx *ctx,
2097 const struct ofpact_bundle *bundle)
2101 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2103 CONST_CAST(struct xbridge *, ctx->xbridge));
2104 if (bundle->dst.field) {
2105 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2108 xlate_output_action(ctx, port, 0, false);
2113 xlate_learn_action(struct xlate_ctx *ctx,
2114 const struct ofpact_learn *learn)
2116 uint64_t ofpacts_stub[1024 / 8];
2117 struct ofputil_flow_mod fm;
2118 struct ofpbuf ofpacts;
2120 ctx->xout->has_learn = true;
2122 learn_mask(learn, &ctx->xout->wc);
2124 if (!ctx->xin->may_learn) {
2128 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2129 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2130 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2131 ofpbuf_uninit(&ofpacts);
2135 xlate_fin_timeout(struct xlate_ctx *ctx,
2136 const struct ofpact_fin_timeout *oft)
2138 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2139 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2140 oft->fin_hard_timeout);
2145 xlate_sample_action(struct xlate_ctx *ctx,
2146 const struct ofpact_sample *os)
2148 union user_action_cookie cookie;
2149 /* Scale the probability from 16-bit to 32-bit while representing
2150 * the same percentage. */
2151 uint32_t probability = (os->probability << 16) | os->probability;
2153 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2154 &ctx->xout->odp_actions, &ctx->xout->wc);
2156 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2157 os->obs_domain_id, os->obs_point_id, &cookie);
2158 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2159 probability, &cookie, sizeof cookie.flow_sample);
2163 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2165 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2166 ? OFPUTIL_PC_NO_RECV_STP
2167 : OFPUTIL_PC_NO_RECV)) {
2171 /* Only drop packets here if both forwarding and learning are
2172 * disabled. If just learning is enabled, we need to have
2173 * OFPP_NORMAL and the learning action have a look at the packet
2174 * before we can drop it. */
2175 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2183 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2184 struct xlate_ctx *ctx)
2186 struct flow_wildcards *wc = &ctx->xout->wc;
2187 struct flow *flow = &ctx->xin->flow;
2188 const struct ofpact *a;
2190 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2191 struct ofpact_controller *controller;
2192 const struct ofpact_metadata *metadata;
2200 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2201 ofpact_get_OUTPUT(a)->max_len, true);
2204 case OFPACT_CONTROLLER:
2205 controller = ofpact_get_CONTROLLER(a);
2206 execute_controller_action(ctx, controller->max_len,
2208 controller->controller_id);
2211 case OFPACT_ENQUEUE:
2212 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2215 case OFPACT_SET_VLAN_VID:
2216 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2217 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2218 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2222 case OFPACT_SET_VLAN_PCP:
2223 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2224 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2226 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2230 case OFPACT_STRIP_VLAN:
2231 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2232 flow->vlan_tci = htons(0);
2235 case OFPACT_PUSH_VLAN:
2236 /* XXX 802.1AD(QinQ) */
2237 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2238 flow->vlan_tci = htons(VLAN_CFI);
2241 case OFPACT_SET_ETH_SRC:
2242 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2243 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2246 case OFPACT_SET_ETH_DST:
2247 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2248 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2251 case OFPACT_SET_IPV4_SRC:
2252 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2253 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2254 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2258 case OFPACT_SET_IPV4_DST:
2259 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2260 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2261 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2265 case OFPACT_SET_IPV4_DSCP:
2266 wc->masks.nw_tos |= IP_DSCP_MASK;
2267 /* OpenFlow 1.0 only supports IPv4. */
2268 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2269 flow->nw_tos &= ~IP_DSCP_MASK;
2270 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2274 case OFPACT_SET_L4_SRC_PORT:
2275 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2276 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2277 if (is_ip_any(flow)) {
2278 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2282 case OFPACT_SET_L4_DST_PORT:
2283 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2284 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2285 if (is_ip_any(flow)) {
2286 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2290 case OFPACT_RESUBMIT:
2291 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2294 case OFPACT_SET_TUNNEL:
2295 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2298 case OFPACT_SET_QUEUE:
2299 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2302 case OFPACT_POP_QUEUE:
2303 flow->skb_priority = ctx->orig_skb_priority;
2306 case OFPACT_REG_MOVE:
2307 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2310 case OFPACT_REG_LOAD:
2311 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2314 case OFPACT_STACK_PUSH:
2315 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2319 case OFPACT_STACK_POP:
2320 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2324 case OFPACT_PUSH_MPLS:
2325 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2328 case OFPACT_POP_MPLS:
2329 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2332 case OFPACT_SET_MPLS_TTL:
2333 if (compose_set_mpls_ttl_action(ctx,
2334 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2339 case OFPACT_DEC_MPLS_TTL:
2340 if (compose_dec_mpls_ttl_action(ctx)) {
2345 case OFPACT_DEC_TTL:
2346 wc->masks.nw_ttl = 0xff;
2347 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2353 /* Nothing to do. */
2356 case OFPACT_MULTIPATH:
2357 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2361 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2364 case OFPACT_OUTPUT_REG:
2365 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2369 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2376 case OFPACT_FIN_TIMEOUT:
2377 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2378 ctx->xout->has_fin_timeout = true;
2379 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2382 case OFPACT_CLEAR_ACTIONS:
2384 * Nothing to do because writa-actions is not supported for now.
2385 * When writa-actions is supported, clear-actions also must
2386 * be supported at the same time.
2390 case OFPACT_WRITE_METADATA:
2391 metadata = ofpact_get_WRITE_METADATA(a);
2392 flow->metadata &= ~metadata->mask;
2393 flow->metadata |= metadata->metadata & metadata->mask;
2397 /* Not implemented yet. */
2400 case OFPACT_GOTO_TABLE: {
2401 /* It is assumed that goto-table is the last action. */
2402 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2404 ovs_assert(ctx->table_id < ogt->table_id);
2405 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2406 ogt->table_id, true);
2411 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2418 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2419 const struct flow *flow, struct rule_dpif *rule,
2420 uint8_t tcp_flags, const struct ofpbuf *packet)
2422 xin->ofproto = ofproto;
2424 xin->packet = packet;
2425 xin->may_learn = packet != NULL;
2427 xin->ofpacts = NULL;
2428 xin->ofpacts_len = 0;
2429 xin->tcp_flags = tcp_flags;
2430 xin->resubmit_hook = NULL;
2431 xin->report_hook = NULL;
2432 xin->resubmit_stats = NULL;
2436 xlate_out_uninit(struct xlate_out *xout)
2439 ofpbuf_uninit(&xout->odp_actions);
2443 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2444 * into datapath actions, using 'ctx', and discards the datapath actions. */
2446 xlate_actions_for_side_effects(struct xlate_in *xin)
2448 struct xlate_out xout;
2450 xlate_actions(xin, &xout);
2451 xlate_out_uninit(&xout);
2455 xlate_report(struct xlate_ctx *ctx, const char *s)
2457 if (ctx->xin->report_hook) {
2458 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2463 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2466 dst->slow = src->slow;
2467 dst->has_learn = src->has_learn;
2468 dst->has_normal = src->has_normal;
2469 dst->has_fin_timeout = src->has_fin_timeout;
2470 dst->nf_output_iface = src->nf_output_iface;
2471 dst->mirrors = src->mirrors;
2473 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2474 sizeof dst->odp_actions_stub);
2475 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2476 src->odp_actions.size);
2479 static struct skb_priority_to_dscp *
2480 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2482 struct skb_priority_to_dscp *pdscp;
2485 hash = hash_int(skb_priority, 0);
2486 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2487 if (pdscp->skb_priority == skb_priority) {
2495 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2498 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2499 *dscp = pdscp ? pdscp->dscp : 0;
2500 return pdscp != NULL;
2504 clear_skb_priorities(struct xport *xport)
2506 struct skb_priority_to_dscp *pdscp, *next;
2508 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2509 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2515 actions_output_to_local_port(const struct xlate_ctx *ctx)
2517 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2518 const struct nlattr *a;
2521 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2522 ctx->xout->odp_actions.size) {
2523 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2524 && nl_attr_get_odp_port(a) == local_odp_port) {
2531 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2532 * into datapath actions in 'odp_actions', using 'ctx'.
2534 * The caller must take responsibility for eventually freeing 'xout', with
2535 * xlate_out_uninit(). */
2537 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2539 struct flow_wildcards *wc = &xout->wc;
2540 struct flow *flow = &xin->flow;
2541 struct rule_dpif *rule = NULL;
2543 struct rule_actions *actions = NULL;
2544 enum slow_path_reason special;
2545 const struct ofpact *ofpacts;
2546 struct xport *in_port;
2547 struct flow orig_flow;
2548 struct xlate_ctx ctx;
2553 COVERAGE_INC(xlate_actions);
2555 ovs_rwlock_rdlock(&xlate_rwlock);
2557 /* Flow initialization rules:
2558 * - 'base_flow' must match the kernel's view of the packet at the
2559 * time that action processing starts. 'flow' represents any
2560 * transformations we wish to make through actions.
2561 * - By default 'base_flow' and 'flow' are the same since the input
2562 * packet matches the output before any actions are applied.
2563 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2564 * of the received packet as seen by the kernel. If we later output
2565 * to another device without any modifications this will cause us to
2566 * insert a new tag since the original one was stripped off by the
2568 * - Tunnel metadata as received is retained in 'flow'. This allows
2569 * tunnel metadata matching also in later tables.
2570 * Since a kernel action for setting the tunnel metadata will only be
2571 * generated with actual tunnel output, changing the tunnel metadata
2572 * values in 'flow' (such as tun_id) will only have effect with a later
2573 * tunnel output action.
2574 * - Tunnel 'base_flow' is completely cleared since that is what the
2575 * kernel does. If we wish to maintain the original values an action
2576 * needs to be generated. */
2581 ctx.xout->has_learn = false;
2582 ctx.xout->has_normal = false;
2583 ctx.xout->has_fin_timeout = false;
2584 ctx.xout->nf_output_iface = NF_OUT_DROP;
2585 ctx.xout->mirrors = 0;
2586 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2587 sizeof ctx.xout->odp_actions_stub);
2588 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2590 ctx.xbridge = xbridge_lookup(xin->ofproto);
2595 ctx.rule = xin->rule;
2597 ctx.base_flow = *flow;
2598 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2599 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2601 flow_wildcards_init_catchall(wc);
2602 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2603 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2604 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2605 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2606 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
2608 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
2609 if (ctx.xbridge->has_netflow) {
2610 netflow_mask_wc(flow, wc);
2614 ctx.orig_skb_priority = flow->skb_priority;
2618 if (!xin->ofpacts && !ctx.rule) {
2619 rule_dpif_lookup(ctx.xbridge->ofproto, flow, wc, &rule);
2620 if (ctx.xin->resubmit_stats) {
2621 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
2625 xout->fail_open = ctx.rule && rule_dpif_fail_open(ctx.rule);
2628 ofpacts = xin->ofpacts;
2629 ofpacts_len = xin->ofpacts_len;
2630 } else if (ctx.rule) {
2631 actions = rule_dpif_get_actions(ctx.rule);
2632 ofpacts = actions->ofpacts;
2633 ofpacts_len = actions->ofpacts_len;
2638 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2640 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2641 /* Do this conditionally because the copy is expensive enough that it
2642 * shows up in profiles. */
2646 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2647 switch (ctx.xbridge->frag) {
2648 case OFPC_FRAG_NORMAL:
2649 /* We must pretend that transport ports are unavailable. */
2650 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2651 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2654 case OFPC_FRAG_DROP:
2657 case OFPC_FRAG_REASM:
2660 case OFPC_FRAG_NX_MATCH:
2661 /* Nothing to do. */
2664 case OFPC_INVALID_TTL_TO_CONTROLLER:
2669 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2670 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2672 ctx.xout->slow = special;
2674 size_t sample_actions_len;
2676 if (flow->in_port.ofp_port
2677 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2678 flow->in_port.ofp_port,
2680 ctx.base_flow.vlan_tci = 0;
2683 add_sflow_action(&ctx);
2684 add_ipfix_action(&ctx);
2685 sample_actions_len = ctx.xout->odp_actions.size;
2687 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
2688 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2690 /* We've let OFPP_NORMAL and the learning action look at the
2691 * packet, so drop it now if forwarding is disabled. */
2692 if (in_port && !xport_stp_forward_state(in_port)) {
2693 ctx.xout->odp_actions.size = sample_actions_len;
2697 if (ctx.xbridge->has_in_band
2698 && in_band_must_output_to_local_port(flow)
2699 && !actions_output_to_local_port(&ctx)) {
2700 compose_output_action(&ctx, OFPP_LOCAL);
2703 fix_sflow_action(&ctx);
2705 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2706 add_mirror_actions(&ctx, &orig_flow);
2710 ofpbuf_uninit(&ctx.stack);
2712 /* Clear the metadata and register wildcard masks, because we won't
2713 * use non-header fields as part of the cache. */
2714 flow_wildcards_clear_non_packet_fields(wc);
2716 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
2717 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
2718 * these fields. The datapath interface, on the other hand, represents
2719 * them with just 8 bits each. This means that if the high 8 bits of the
2720 * masks for these fields somehow become set, then they will get chopped
2721 * off by a round trip through the datapath, and revalidation will spot
2722 * that as an inconsistency and delete the flow. Avoid the problem here by
2723 * making sure that only the low 8 bits of either field can be unwildcarded
2727 wc->masks.tp_src &= htons(UINT8_MAX);
2728 wc->masks.tp_dst &= htons(UINT8_MAX);
2732 ovs_rwlock_unlock(&xlate_rwlock);
2734 rule_dpif_unref(rule);
2735 rule_actions_unref(actions);