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 xbridge *xbridge_lookup(const struct ofproto_dpif *);
205 static struct xbundle *xbundle_lookup(const struct ofbundle *);
206 static struct xport *xport_lookup(const struct ofport_dpif *);
207 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
208 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
209 uint32_t skb_priority);
210 static void clear_skb_priorities(struct xport *);
211 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
215 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
216 struct dpif *dpif, struct rule_dpif *miss_rule,
217 struct rule_dpif *no_packet_in_rule,
218 const struct mac_learning *ml, struct stp *stp,
219 const struct mbridge *mbridge,
220 const struct dpif_sflow *sflow,
221 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
222 bool forward_bpdu, bool has_in_band, bool has_netflow)
224 struct xbridge *xbridge = xbridge_lookup(ofproto);
227 xbridge = xzalloc(sizeof *xbridge);
228 xbridge->ofproto = ofproto;
230 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
231 hmap_init(&xbridge->xports);
232 list_init(&xbridge->xbundles);
235 if (xbridge->ml != ml) {
236 mac_learning_unref(xbridge->ml);
237 xbridge->ml = mac_learning_ref(ml);
240 if (xbridge->mbridge != mbridge) {
241 mbridge_unref(xbridge->mbridge);
242 xbridge->mbridge = mbridge_ref(mbridge);
245 if (xbridge->sflow != sflow) {
246 dpif_sflow_unref(xbridge->sflow);
247 xbridge->sflow = dpif_sflow_ref(sflow);
250 if (xbridge->ipfix != ipfix) {
251 dpif_ipfix_unref(xbridge->ipfix);
252 xbridge->ipfix = dpif_ipfix_ref(ipfix);
255 if (xbridge->stp != stp) {
256 stp_unref(xbridge->stp);
257 xbridge->stp = stp_ref(stp);
261 xbridge->name = xstrdup(name);
263 xbridge->dpif = dpif;
264 xbridge->forward_bpdu = forward_bpdu;
265 xbridge->has_in_band = has_in_band;
266 xbridge->has_netflow = has_netflow;
267 xbridge->frag = frag;
268 xbridge->miss_rule = miss_rule;
269 xbridge->no_packet_in_rule = no_packet_in_rule;
273 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
275 struct xbridge *xbridge = xbridge_lookup(ofproto);
276 struct xbundle *xbundle, *next_xbundle;
277 struct xport *xport, *next_xport;
283 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
284 xlate_ofport_remove(xport->ofport);
287 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
288 xlate_bundle_remove(xbundle->ofbundle);
291 hmap_remove(&xbridges, &xbridge->hmap_node);
297 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
298 const char *name, enum port_vlan_mode vlan_mode, int vlan,
299 unsigned long *trunks, bool use_priority_tags,
300 const struct bond *bond, const struct lacp *lacp,
303 struct xbundle *xbundle = xbundle_lookup(ofbundle);
306 xbundle = xzalloc(sizeof *xbundle);
307 xbundle->ofbundle = ofbundle;
308 xbundle->xbridge = xbridge_lookup(ofproto);
310 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
311 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
312 list_init(&xbundle->xports);
315 ovs_assert(xbundle->xbridge);
318 xbundle->name = xstrdup(name);
320 xbundle->vlan_mode = vlan_mode;
321 xbundle->vlan = vlan;
322 xbundle->trunks = trunks;
323 xbundle->use_priority_tags = use_priority_tags;
324 xbundle->floodable = floodable;
326 if (xbundle->bond != bond) {
327 bond_unref(xbundle->bond);
328 xbundle->bond = bond_ref(bond);
331 if (xbundle->lacp != lacp) {
332 lacp_unref(xbundle->lacp);
333 xbundle->lacp = lacp_ref(lacp);
338 xlate_bundle_remove(struct ofbundle *ofbundle)
340 struct xbundle *xbundle = xbundle_lookup(ofbundle);
341 struct xport *xport, *next;
347 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
348 list_remove(&xport->bundle_node);
349 xport->xbundle = NULL;
352 hmap_remove(&xbundles, &xbundle->hmap_node);
353 list_remove(&xbundle->list_node);
354 bond_unref(xbundle->bond);
355 lacp_unref(xbundle->lacp);
361 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
362 struct ofport_dpif *ofport, ofp_port_t ofp_port,
363 odp_port_t odp_port, const struct netdev *netdev,
364 const struct cfm *cfm, const struct bfd *bfd,
365 struct ofport_dpif *peer, int stp_port_no,
366 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
367 enum ofputil_port_config config, bool is_tunnel,
370 struct xport *xport = xport_lookup(ofport);
374 xport = xzalloc(sizeof *xport);
375 xport->ofport = ofport;
376 xport->xbridge = xbridge_lookup(ofproto);
377 xport->ofp_port = ofp_port;
379 hmap_init(&xport->skb_priorities);
380 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
381 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
382 hash_ofp_port(xport->ofp_port));
385 ovs_assert(xport->ofp_port == ofp_port);
387 xport->config = config;
388 xport->stp_port_no = stp_port_no;
389 xport->is_tunnel = is_tunnel;
390 xport->may_enable = may_enable;
391 xport->odp_port = odp_port;
393 if (xport->netdev != netdev) {
394 netdev_close(xport->netdev);
395 xport->netdev = netdev_ref(netdev);
398 if (xport->cfm != cfm) {
399 cfm_unref(xport->cfm);
400 xport->cfm = cfm_ref(cfm);
403 if (xport->bfd != bfd) {
404 bfd_unref(xport->bfd);
405 xport->bfd = bfd_ref(bfd);
409 xport->peer->peer = NULL;
411 xport->peer = xport_lookup(peer);
413 xport->peer->peer = xport;
416 if (xport->xbundle) {
417 list_remove(&xport->bundle_node);
419 xport->xbundle = xbundle_lookup(ofbundle);
420 if (xport->xbundle) {
421 list_insert(&xport->xbundle->xports, &xport->bundle_node);
424 clear_skb_priorities(xport);
425 for (i = 0; i < n_qdscp; i++) {
426 struct skb_priority_to_dscp *pdscp;
427 uint32_t skb_priority;
429 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
434 pdscp = xmalloc(sizeof *pdscp);
435 pdscp->skb_priority = skb_priority;
436 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
437 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
438 hash_int(pdscp->skb_priority, 0));
443 xlate_ofport_remove(struct ofport_dpif *ofport)
445 struct xport *xport = xport_lookup(ofport);
452 xport->peer->peer = NULL;
456 if (xport->xbundle) {
457 list_remove(&xport->bundle_node);
460 clear_skb_priorities(xport);
461 hmap_destroy(&xport->skb_priorities);
463 hmap_remove(&xports, &xport->hmap_node);
464 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
466 netdev_close(xport->netdev);
467 cfm_unref(xport->cfm);
468 bfd_unref(xport->bfd);
472 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
473 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
474 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
475 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
476 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
477 * 'packet' ingressed.
479 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
480 * 'flow''s in_port to OFPP_NONE.
482 * This function does post-processing on data returned from
483 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
484 * of the upcall processing logic. In particular, if the extracted in_port is
485 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
486 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
487 * a VLAN header onto 'packet' (if it is nonnull).
489 * Similarly, this function also includes some logic to help with tunnels. It
490 * may modify 'flow' as necessary to make the tunneling implementation
491 * transparent to the upcall processing logic.
493 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
494 * or some other positive errno if there are other problems. */
496 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
497 const struct nlattr *key, size_t key_len,
498 struct flow *flow, enum odp_key_fitness *fitnessp,
499 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
501 enum odp_key_fitness fitness;
502 const struct xport *xport;
505 ovs_rwlock_rdlock(&xlate_rwlock);
506 fitness = odp_flow_key_to_flow(key, key_len, flow);
507 if (fitness == ODP_FIT_ERROR) {
513 *odp_in_port = flow->in_port.odp_port;
516 xport = xport_lookup(tnl_port_should_receive(flow)
517 ? tnl_port_receive(flow)
518 : odp_port_to_ofport(backer, flow->in_port.odp_port));
520 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
525 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
527 /* Make the packet resemble the flow, so that it gets sent to
528 * an OpenFlow controller properly, so that it looks correct
529 * for sFlow, and so that flow_extract() will get the correct
530 * vlan_tci if it is called on 'packet'.
532 * The allocated space inside 'packet' probably also contains
533 * 'key', that is, both 'packet' and 'key' are probably part of
534 * a struct dpif_upcall (see the large comment on that
535 * structure definition), so pushing data on 'packet' is in
536 * general not a good idea since it could overwrite 'key' or
537 * free it as a side effect. However, it's OK in this special
538 * case because we know that 'packet' is inside a Netlink
539 * attribute: pushing 4 bytes will just overwrite the 4-byte
540 * "struct nlattr", which is fine since we don't need that
542 eth_push_vlan(packet, flow->vlan_tci);
544 /* We can't reproduce 'key' from 'flow'. */
545 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
550 *ofproto = xport->xbridge->ofproto;
557 ovs_rwlock_unlock(&xlate_rwlock);
561 static struct xbridge *
562 xbridge_lookup(const struct ofproto_dpif *ofproto)
564 struct xbridge *xbridge;
570 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
572 if (xbridge->ofproto == ofproto) {
579 static struct xbundle *
580 xbundle_lookup(const struct ofbundle *ofbundle)
582 struct xbundle *xbundle;
588 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
590 if (xbundle->ofbundle == ofbundle) {
597 static struct xport *
598 xport_lookup(const struct ofport_dpif *ofport)
606 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
608 if (xport->ofport == ofport) {
615 static struct stp_port *
616 xport_get_stp_port(const struct xport *xport)
618 return xport->xbridge->stp && xport->stp_port_no
619 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
623 static enum stp_state
624 xport_stp_learn_state(const struct xport *xport)
626 struct stp_port *sp = xport_get_stp_port(xport);
627 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
631 xport_stp_forward_state(const struct xport *xport)
633 struct stp_port *sp = xport_get_stp_port(xport);
634 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
637 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
638 * were used to make the determination.*/
640 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
642 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
643 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
647 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
649 struct stp_port *sp = xport_get_stp_port(xport);
650 struct ofpbuf payload = *packet;
651 struct eth_header *eth = payload.data;
653 /* Sink packets on ports that have STP disabled when the bridge has
655 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
659 /* Trim off padding on payload. */
660 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
661 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
664 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
665 stp_received_bpdu(sp, payload.data, payload.size);
669 static struct xport *
670 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
674 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
676 if (xport->ofp_port == ofp_port) {
684 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
686 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
687 return xport ? xport->odp_port : ODPP_NONE;
691 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
693 return (bundle->vlan_mode != PORT_VLAN_ACCESS
694 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
698 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
700 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
704 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
706 return xbundle != &ofpp_none_bundle
707 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
712 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
714 return xbundle != &ofpp_none_bundle
715 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
720 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
722 return xbundle != &ofpp_none_bundle
723 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
727 static struct xbundle *
728 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
729 bool warn, struct xport **in_xportp)
733 /* Find the port and bundle for the received packet. */
734 xport = get_ofp_port(xbridge, in_port);
738 if (xport && xport->xbundle) {
739 return xport->xbundle;
742 /* Special-case OFPP_NONE, which a controller may use as the ingress
743 * port for traffic that it is sourcing. */
744 if (in_port == OFPP_NONE) {
745 ofpp_none_bundle.name = "OFPP_NONE";
746 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
747 return &ofpp_none_bundle;
750 /* Odd. A few possible reasons here:
752 * - We deleted a port but there are still a few packets queued up
755 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
756 * we don't know about.
758 * - The ofproto client didn't configure the port as part of a bundle.
759 * This is particularly likely to happen if a packet was received on the
760 * port after it was created, but before the client had a chance to
761 * configure its bundle.
764 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
766 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
767 "port %"PRIu16, xbridge->name, in_port);
773 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
775 const struct xbridge *xbridge = ctx->xbridge;
776 mirror_mask_t mirrors;
777 struct xbundle *in_xbundle;
781 mirrors = ctx->xout->mirrors;
782 ctx->xout->mirrors = 0;
784 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
785 ctx->xin->packet != NULL, NULL);
789 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
791 /* Drop frames on bundles reserved for mirroring. */
792 if (xbundle_mirror_out(xbridge, in_xbundle)) {
793 if (ctx->xin->packet != NULL) {
794 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
795 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
796 "%s, which is reserved exclusively for mirroring",
797 ctx->xbridge->name, in_xbundle->name);
799 ofpbuf_clear(&ctx->xout->odp_actions);
804 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
805 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
808 vlan = input_vid_to_vlan(in_xbundle, vid);
814 /* Restore the original packet before adding the mirror actions. */
815 ctx->xin->flow = *orig_flow;
818 mirror_mask_t dup_mirrors;
819 struct ofbundle *out;
820 unsigned long *vlans;
825 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
826 &vlans, &dup_mirrors, &out, &out_vlan);
827 ovs_assert(has_mirror);
830 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
832 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
835 if (!vlan_mirrored) {
836 mirrors = zero_rightmost_1bit(mirrors);
840 mirrors &= ~dup_mirrors;
841 ctx->xout->mirrors |= dup_mirrors;
843 struct xbundle *out_xbundle = xbundle_lookup(out);
845 output_normal(ctx, out_xbundle, vlan);
847 } else if (vlan != out_vlan
848 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
849 struct xbundle *xbundle;
851 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
852 if (xbundle_includes_vlan(xbundle, out_vlan)
853 && !xbundle_mirror_out(xbridge, xbundle)) {
854 output_normal(ctx, xbundle, out_vlan);
861 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
862 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
863 * the bundle on which the packet was received, returns the VLAN to which the
866 * Both 'vid' and the return value are in the range 0...4095. */
868 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
870 switch (in_xbundle->vlan_mode) {
871 case PORT_VLAN_ACCESS:
872 return in_xbundle->vlan;
875 case PORT_VLAN_TRUNK:
878 case PORT_VLAN_NATIVE_UNTAGGED:
879 case PORT_VLAN_NATIVE_TAGGED:
880 return vid ? vid : in_xbundle->vlan;
887 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
888 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
891 * 'vid' should be the VID obtained from the 802.1Q header that was received as
892 * part of a packet (specify 0 if there was no 802.1Q header), in the range
895 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
897 /* Allow any VID on the OFPP_NONE port. */
898 if (in_xbundle == &ofpp_none_bundle) {
902 switch (in_xbundle->vlan_mode) {
903 case PORT_VLAN_ACCESS:
906 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
907 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
908 "packet received on port %s configured as VLAN "
909 "%"PRIu16" access port", vid, in_xbundle->name,
916 case PORT_VLAN_NATIVE_UNTAGGED:
917 case PORT_VLAN_NATIVE_TAGGED:
919 /* Port must always carry its native VLAN. */
923 case PORT_VLAN_TRUNK:
924 if (!xbundle_includes_vlan(in_xbundle, vid)) {
926 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
927 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
928 "received on port %s not configured for trunking "
929 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
941 /* Given 'vlan', the VLAN that a packet belongs to, and
942 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
943 * that should be included in the 802.1Q header. (If the return value is 0,
944 * then the 802.1Q header should only be included in the packet if there is a
947 * Both 'vlan' and the return value are in the range 0...4095. */
949 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
951 switch (out_xbundle->vlan_mode) {
952 case PORT_VLAN_ACCESS:
955 case PORT_VLAN_TRUNK:
956 case PORT_VLAN_NATIVE_TAGGED:
959 case PORT_VLAN_NATIVE_UNTAGGED:
960 return vlan == out_xbundle->vlan ? 0 : vlan;
968 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
971 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
973 ovs_be16 tci, old_tci;
976 vid = output_vlan_to_vid(out_xbundle, vlan);
977 if (list_is_empty(&out_xbundle->xports)) {
978 /* Partially configured bundle with no slaves. Drop the packet. */
980 } else if (!out_xbundle->bond) {
981 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
984 struct ofport_dpif *ofport;
986 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
987 &ctx->xout->wc, vid);
988 xport = xport_lookup(ofport);
991 /* No slaves enabled, so drop packet. */
998 if (tci || out_xbundle->use_priority_tags) {
999 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1001 tci |= htons(VLAN_CFI);
1006 compose_output_action(ctx, xport->ofp_port);
1007 *flow_tci = old_tci;
1010 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1011 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1012 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1014 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1016 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1020 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1021 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1025 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1026 if (flow->nw_proto == ARP_OP_REPLY) {
1028 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1029 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1030 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1032 return flow->nw_src == flow->nw_dst;
1038 /* Checks whether a MAC learning update is necessary for MAC learning table
1039 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1042 * Most packets processed through the MAC learning table do not actually
1043 * change it in any way. This function requires only a read lock on the MAC
1044 * learning table, so it is much cheaper in this common case.
1046 * Keep the code here synchronized with that in update_learning_table__()
1049 is_mac_learning_update_needed(const struct mac_learning *ml,
1050 const struct flow *flow,
1051 struct flow_wildcards *wc,
1052 int vlan, struct xbundle *in_xbundle)
1053 OVS_REQ_RDLOCK(ml->rwlock)
1055 struct mac_entry *mac;
1057 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1061 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1062 if (!mac || mac_entry_age(ml, mac)) {
1066 if (is_gratuitous_arp(flow, wc)) {
1067 /* We don't want to learn from gratuitous ARP packets that are
1068 * reflected back over bond slaves so we lock the learning table. */
1069 if (!in_xbundle->bond) {
1071 } else if (mac_entry_is_grat_arp_locked(mac)) {
1076 return mac->port.p != in_xbundle->ofbundle;
1080 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1081 * received on 'in_xbundle' in 'vlan'.
1083 * This code repeats all the checks in is_mac_learning_update_needed() because
1084 * the lock was released between there and here and thus the MAC learning state
1085 * could have changed.
1087 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1090 update_learning_table__(const struct xbridge *xbridge,
1091 const struct flow *flow, struct flow_wildcards *wc,
1092 int vlan, struct xbundle *in_xbundle)
1093 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1095 struct mac_entry *mac;
1097 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1101 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1102 if (is_gratuitous_arp(flow, wc)) {
1103 /* We don't want to learn from gratuitous ARP packets that are
1104 * reflected back over bond slaves so we lock the learning table. */
1105 if (!in_xbundle->bond) {
1106 mac_entry_set_grat_arp_lock(mac);
1107 } else if (mac_entry_is_grat_arp_locked(mac)) {
1112 if (mac->port.p != in_xbundle->ofbundle) {
1113 /* The log messages here could actually be useful in debugging,
1114 * so keep the rate limit relatively high. */
1115 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1117 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1118 "on port %s in VLAN %d",
1119 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1120 in_xbundle->name, vlan);
1122 mac->port.p = in_xbundle->ofbundle;
1123 mac_learning_changed(xbridge->ml);
1128 update_learning_table(const struct xbridge *xbridge,
1129 const struct flow *flow, struct flow_wildcards *wc,
1130 int vlan, struct xbundle *in_xbundle)
1134 /* Don't learn the OFPP_NONE port. */
1135 if (in_xbundle == &ofpp_none_bundle) {
1139 /* First try the common case: no change to MAC learning table. */
1140 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1141 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1143 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1146 /* Slow path: MAC learning table might need an update. */
1147 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1148 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1149 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1153 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1154 * dropped. Returns true if they may be forwarded, false if they should be
1157 * 'in_port' must be the xport that corresponds to flow->in_port.
1158 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1160 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1161 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1162 * checked by input_vid_is_valid().
1164 * May also add tags to '*tags', although the current implementation only does
1165 * so in one special case.
1168 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1171 struct xbundle *in_xbundle = in_port->xbundle;
1172 const struct xbridge *xbridge = ctx->xbridge;
1173 struct flow *flow = &ctx->xin->flow;
1175 /* Drop frames for reserved multicast addresses
1176 * only if forward_bpdu option is absent. */
1177 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1178 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1182 if (in_xbundle->bond) {
1183 struct mac_entry *mac;
1185 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1191 xlate_report(ctx, "bonding refused admissibility, dropping");
1194 case BV_DROP_IF_MOVED:
1195 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1196 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1197 if (mac && mac->port.p != in_xbundle->ofbundle &&
1198 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1199 || mac_entry_is_grat_arp_locked(mac))) {
1200 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1201 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1205 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1214 xlate_normal(struct xlate_ctx *ctx)
1216 struct flow_wildcards *wc = &ctx->xout->wc;
1217 struct flow *flow = &ctx->xin->flow;
1218 struct xbundle *in_xbundle;
1219 struct xport *in_port;
1220 struct mac_entry *mac;
1224 ctx->xout->has_normal = true;
1226 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1227 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1228 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1230 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1231 ctx->xin->packet != NULL, &in_port);
1233 xlate_report(ctx, "no input bundle, dropping");
1237 /* Drop malformed frames. */
1238 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1239 !(flow->vlan_tci & htons(VLAN_CFI))) {
1240 if (ctx->xin->packet != NULL) {
1241 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1242 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1243 "VLAN tag received on port %s",
1244 ctx->xbridge->name, in_xbundle->name);
1246 xlate_report(ctx, "partial VLAN tag, dropping");
1250 /* Drop frames on bundles reserved for mirroring. */
1251 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1252 if (ctx->xin->packet != NULL) {
1253 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1254 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1255 "%s, which is reserved exclusively for mirroring",
1256 ctx->xbridge->name, in_xbundle->name);
1258 xlate_report(ctx, "input port is mirror output port, dropping");
1263 vid = vlan_tci_to_vid(flow->vlan_tci);
1264 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1265 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1268 vlan = input_vid_to_vlan(in_xbundle, vid);
1270 /* Check other admissibility requirements. */
1271 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1275 /* Learn source MAC. */
1276 if (ctx->xin->may_learn) {
1277 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1280 /* Determine output bundle. */
1281 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1282 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1284 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
1285 if (mac_xbundle && mac_xbundle != in_xbundle) {
1286 xlate_report(ctx, "forwarding to learned port");
1287 output_normal(ctx, mac_xbundle, vlan);
1288 } else if (!mac_xbundle) {
1289 xlate_report(ctx, "learned port is unknown, dropping");
1291 xlate_report(ctx, "learned port is input port, dropping");
1294 struct xbundle *xbundle;
1296 xlate_report(ctx, "no learned MAC for destination, flooding");
1297 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1298 if (xbundle != in_xbundle
1299 && xbundle_includes_vlan(xbundle, vlan)
1300 && xbundle->floodable
1301 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1302 output_normal(ctx, xbundle, vlan);
1305 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1307 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1310 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1311 * the number of packets out of UINT32_MAX to sample. The given
1312 * cookie is passed back in the callback for each sampled packet.
1315 compose_sample_action(const struct xbridge *xbridge,
1316 struct ofpbuf *odp_actions,
1317 const struct flow *flow,
1318 const uint32_t probability,
1319 const union user_action_cookie *cookie,
1320 const size_t cookie_size)
1322 size_t sample_offset, actions_offset;
1323 odp_port_t odp_port;
1327 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1329 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1331 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1333 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1334 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1335 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1337 nl_msg_end_nested(odp_actions, actions_offset);
1338 nl_msg_end_nested(odp_actions, sample_offset);
1339 return cookie_offset;
1343 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1344 odp_port_t odp_port, unsigned int n_outputs,
1345 union user_action_cookie *cookie)
1349 cookie->type = USER_ACTION_COOKIE_SFLOW;
1350 cookie->sflow.vlan_tci = vlan_tci;
1352 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1353 * port information") for the interpretation of cookie->output. */
1354 switch (n_outputs) {
1356 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1357 cookie->sflow.output = 0x40000000 | 256;
1361 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1363 cookie->sflow.output = ifindex;
1368 /* 0x80000000 means "multiple output ports. */
1369 cookie->sflow.output = 0x80000000 | n_outputs;
1374 /* Compose SAMPLE action for sFlow bridge sampling. */
1376 compose_sflow_action(const struct xbridge *xbridge,
1377 struct ofpbuf *odp_actions,
1378 const struct flow *flow,
1379 odp_port_t odp_port)
1381 uint32_t probability;
1382 union user_action_cookie cookie;
1384 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1388 probability = dpif_sflow_get_probability(xbridge->sflow);
1389 compose_sflow_cookie(xbridge, htons(0), odp_port,
1390 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1392 return compose_sample_action(xbridge, odp_actions, flow, probability,
1393 &cookie, sizeof cookie.sflow);
1397 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1398 uint32_t obs_domain_id, uint32_t obs_point_id,
1399 union user_action_cookie *cookie)
1401 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1402 cookie->flow_sample.probability = probability;
1403 cookie->flow_sample.collector_set_id = collector_set_id;
1404 cookie->flow_sample.obs_domain_id = obs_domain_id;
1405 cookie->flow_sample.obs_point_id = obs_point_id;
1409 compose_ipfix_cookie(union user_action_cookie *cookie)
1411 cookie->type = USER_ACTION_COOKIE_IPFIX;
1414 /* Compose SAMPLE action for IPFIX bridge sampling. */
1416 compose_ipfix_action(const struct xbridge *xbridge,
1417 struct ofpbuf *odp_actions,
1418 const struct flow *flow)
1420 uint32_t probability;
1421 union user_action_cookie cookie;
1423 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1427 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1428 compose_ipfix_cookie(&cookie);
1430 compose_sample_action(xbridge, odp_actions, flow, probability,
1431 &cookie, sizeof cookie.ipfix);
1434 /* SAMPLE action for sFlow must be first action in any given list of
1435 * actions. At this point we do not have all information required to
1436 * build it. So try to build sample action as complete as possible. */
1438 add_sflow_action(struct xlate_ctx *ctx)
1440 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1441 &ctx->xout->odp_actions,
1442 &ctx->xin->flow, ODPP_NONE);
1443 ctx->sflow_odp_port = 0;
1444 ctx->sflow_n_outputs = 0;
1447 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1448 * of actions, eventually after the SAMPLE action for sFlow. */
1450 add_ipfix_action(struct xlate_ctx *ctx)
1452 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1456 /* Fix SAMPLE action according to data collected while composing ODP actions.
1457 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1458 * USERSPACE action's user-cookie which is required for sflow. */
1460 fix_sflow_action(struct xlate_ctx *ctx)
1462 const struct flow *base = &ctx->base_flow;
1463 union user_action_cookie *cookie;
1465 if (!ctx->user_cookie_offset) {
1469 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1470 sizeof cookie->sflow);
1471 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1473 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1474 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1477 static enum slow_path_reason
1478 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1479 const struct xport *xport, const struct ofpbuf *packet)
1481 struct flow_wildcards *wc = &ctx->xout->wc;
1482 const struct xbridge *xbridge = ctx->xbridge;
1486 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1488 cfm_process_heartbeat(xport->cfm, packet);
1491 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1493 bfd_process_packet(xport->bfd, flow, packet);
1496 } else if (xport->xbundle && xport->xbundle->lacp
1497 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1499 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1502 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1504 stp_process_packet(xport, packet);
1513 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1516 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1517 struct flow_wildcards *wc = &ctx->xout->wc;
1518 struct flow *flow = &ctx->xin->flow;
1519 ovs_be16 flow_vlan_tci;
1520 uint32_t flow_pkt_mark;
1521 uint8_t flow_nw_tos;
1522 odp_port_t out_port, odp_port;
1525 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1526 * before traversing a patch port. */
1527 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1530 xlate_report(ctx, "Nonexistent output port");
1532 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1533 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1535 } else if (check_stp && !xport_stp_forward_state(xport)) {
1536 xlate_report(ctx, "STP not in forwarding state, skipping output");
1540 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1541 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1546 const struct xport *peer = xport->peer;
1547 struct flow old_flow = ctx->xin->flow;
1548 enum slow_path_reason special;
1550 ctx->xbridge = peer->xbridge;
1551 flow->in_port.ofp_port = peer->ofp_port;
1552 flow->metadata = htonll(0);
1553 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1554 memset(flow->regs, 0, sizeof flow->regs);
1556 special = process_special(ctx, &ctx->xin->flow, peer,
1559 ctx->xout->slow = special;
1560 } else if (may_receive(peer, ctx)) {
1561 if (xport_stp_forward_state(peer)) {
1562 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1564 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1565 * learning action look at the packet, then drop it. */
1566 struct flow old_base_flow = ctx->base_flow;
1567 size_t old_size = ctx->xout->odp_actions.size;
1568 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1569 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1570 ctx->xout->mirrors = old_mirrors;
1571 ctx->base_flow = old_base_flow;
1572 ctx->xout->odp_actions.size = old_size;
1576 ctx->xin->flow = old_flow;
1577 ctx->xbridge = xport->xbridge;
1579 if (ctx->xin->resubmit_stats) {
1580 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1581 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1587 flow_vlan_tci = flow->vlan_tci;
1588 flow_pkt_mark = flow->pkt_mark;
1589 flow_nw_tos = flow->nw_tos;
1591 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1592 wc->masks.nw_tos |= IP_ECN_MASK;
1593 flow->nw_tos &= ~IP_DSCP_MASK;
1594 flow->nw_tos |= dscp;
1597 if (xport->is_tunnel) {
1598 /* Save tunnel metadata so that changes made due to
1599 * the Logical (tunnel) Port are not visible for any further
1600 * matches, while explicit set actions on tunnel metadata are.
1602 struct flow_tnl flow_tnl = flow->tunnel;
1603 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1604 if (odp_port == ODPP_NONE) {
1605 xlate_report(ctx, "Tunneling decided against output");
1606 goto out; /* restore flow_nw_tos */
1608 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1609 xlate_report(ctx, "Not tunneling to our own address");
1610 goto out; /* restore flow_nw_tos */
1612 if (ctx->xin->resubmit_stats) {
1613 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1615 out_port = odp_port;
1616 commit_odp_tunnel_action(flow, &ctx->base_flow,
1617 &ctx->xout->odp_actions);
1618 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1620 ofp_port_t vlandev_port;
1622 odp_port = xport->odp_port;
1623 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1624 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1626 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1628 if (vlandev_port == ofp_port) {
1629 out_port = odp_port;
1631 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1632 flow->vlan_tci = htons(0);
1636 if (out_port != ODPP_NONE) {
1637 commit_odp_actions(flow, &ctx->base_flow,
1638 &ctx->xout->odp_actions, &ctx->xout->wc);
1639 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1642 ctx->sflow_odp_port = odp_port;
1643 ctx->sflow_n_outputs++;
1644 ctx->xout->nf_output_iface = ofp_port;
1649 flow->vlan_tci = flow_vlan_tci;
1650 flow->pkt_mark = flow_pkt_mark;
1651 flow->nw_tos = flow_nw_tos;
1655 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1657 compose_output_action__(ctx, ofp_port, true);
1661 xlate_table_action(struct xlate_ctx *ctx,
1662 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1664 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1665 struct rule_dpif *rule;
1666 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1667 uint8_t old_table_id = ctx->table_id;
1669 ctx->table_id = table_id;
1671 /* Look up a flow with 'in_port' as the input port. */
1672 ctx->xin->flow.in_port.ofp_port = in_port;
1673 rule_dpif_lookup_in_table(ctx->xbridge->ofproto, &ctx->xin->flow,
1674 &ctx->xout->wc, table_id, &rule);
1676 /* Restore the original input port. Otherwise OFPP_NORMAL and
1677 * OFPP_IN_PORT will have surprising behavior. */
1678 ctx->xin->flow.in_port.ofp_port = old_in_port;
1680 if (ctx->xin->resubmit_hook) {
1681 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1684 if (rule == NULL && may_packet_in) {
1685 struct xport *xport;
1687 /* Makes clang's thread safety analysis happy. */
1691 * check if table configuration flags
1692 * OFPTC_TABLE_MISS_CONTROLLER, default.
1693 * OFPTC_TABLE_MISS_CONTINUE,
1694 * OFPTC_TABLE_MISS_DROP
1695 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1696 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1697 rule = choose_miss_rule(xport ? xport->config : 0,
1698 ctx->xbridge->miss_rule,
1699 ctx->xbridge->no_packet_in_rule);
1700 ovs_rwlock_rdlock(&rule->up.evict);
1703 if (rule && ctx->xin->resubmit_stats) {
1704 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1708 struct rule_dpif *old_rule = ctx->rule;
1712 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1713 ctx->rule = old_rule;
1718 ctx->table_id = old_table_id;
1720 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1722 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1723 MAX_RESUBMIT_RECURSION);
1728 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1729 const struct ofpact_resubmit *resubmit)
1734 in_port = resubmit->in_port;
1735 if (in_port == OFPP_IN_PORT) {
1736 in_port = ctx->xin->flow.in_port.ofp_port;
1739 table_id = resubmit->table_id;
1740 if (table_id == 255) {
1741 table_id = ctx->table_id;
1744 xlate_table_action(ctx, in_port, table_id, false);
1748 flood_packets(struct xlate_ctx *ctx, bool all)
1750 const struct xport *xport;
1752 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1753 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1758 compose_output_action__(ctx, xport->ofp_port, false);
1759 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1760 compose_output_action(ctx, xport->ofp_port);
1764 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1768 execute_controller_action(struct xlate_ctx *ctx, int len,
1769 enum ofp_packet_in_reason reason,
1770 uint16_t controller_id)
1772 struct ofputil_packet_in *pin;
1773 struct ofpbuf *packet;
1776 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1777 ctx->xout->slow = SLOW_CONTROLLER;
1778 if (!ctx->xin->packet) {
1782 packet = ofpbuf_clone(ctx->xin->packet);
1784 key.skb_priority = 0;
1786 memset(&key.tunnel, 0, sizeof key.tunnel);
1788 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1789 &ctx->xout->odp_actions, &ctx->xout->wc);
1791 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1792 ctx->xout->odp_actions.size, NULL, NULL);
1794 pin = xmalloc(sizeof *pin);
1795 pin->packet_len = packet->size;
1796 pin->packet = ofpbuf_steal_data(packet);
1797 pin->reason = reason;
1798 pin->controller_id = controller_id;
1799 pin->table_id = ctx->table_id;
1800 pin->cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1802 pin->send_len = len;
1803 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1805 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1806 ofpbuf_delete(packet);
1810 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1812 struct flow_wildcards *wc = &ctx->xout->wc;
1813 struct flow *flow = &ctx->xin->flow;
1815 ovs_assert(eth_type_mpls(eth_type));
1817 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1818 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1820 if (flow->mpls_depth) {
1821 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1827 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1828 label = htonl(0x2); /* IPV6 Explicit Null. */
1830 label = htonl(0x0); /* IPV4 Explicit Null. */
1832 wc->masks.nw_tos |= IP_DSCP_MASK;
1833 wc->masks.nw_ttl = 0xff;
1834 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1835 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1836 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1837 flow->mpls_depth = 1;
1839 flow->dl_type = eth_type;
1843 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1845 struct flow_wildcards *wc = &ctx->xout->wc;
1846 struct flow *flow = &ctx->xin->flow;
1848 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1849 ovs_assert(!eth_type_mpls(eth_type));
1851 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1852 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1854 if (flow->mpls_depth) {
1856 flow->mpls_lse = htonl(0);
1857 if (!flow->mpls_depth) {
1858 flow->dl_type = eth_type;
1864 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1866 struct flow *flow = &ctx->xin->flow;
1868 if (!is_ip_any(flow)) {
1872 ctx->xout->wc.masks.nw_ttl = 0xff;
1873 if (flow->nw_ttl > 1) {
1879 for (i = 0; i < ids->n_controllers; i++) {
1880 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1884 /* Stop processing for current table. */
1890 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1892 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1896 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1897 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1902 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1904 struct flow *flow = &ctx->xin->flow;
1905 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1906 struct flow_wildcards *wc = &ctx->xout->wc;
1908 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1910 if (!eth_type_mpls(flow->dl_type)) {
1916 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1919 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1921 /* Stop processing for current table. */
1927 xlate_output_action(struct xlate_ctx *ctx,
1928 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1930 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1932 ctx->xout->nf_output_iface = NF_OUT_DROP;
1936 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1939 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1946 flood_packets(ctx, false);
1949 flood_packets(ctx, true);
1951 case OFPP_CONTROLLER:
1952 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1958 if (port != ctx->xin->flow.in_port.ofp_port) {
1959 compose_output_action(ctx, port);
1961 xlate_report(ctx, "skipping output to input port");
1966 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1967 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1968 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1969 ctx->xout->nf_output_iface = prev_nf_output_iface;
1970 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1971 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1972 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1977 xlate_output_reg_action(struct xlate_ctx *ctx,
1978 const struct ofpact_output_reg *or)
1980 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1981 if (port <= UINT16_MAX) {
1982 union mf_subvalue value;
1984 memset(&value, 0xff, sizeof value);
1985 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1986 xlate_output_action(ctx, u16_to_ofp(port),
1987 or->max_len, false);
1992 xlate_enqueue_action(struct xlate_ctx *ctx,
1993 const struct ofpact_enqueue *enqueue)
1995 ofp_port_t ofp_port = enqueue->port;
1996 uint32_t queue_id = enqueue->queue;
1997 uint32_t flow_priority, priority;
2000 /* Translate queue to priority. */
2001 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2003 /* Fall back to ordinary output action. */
2004 xlate_output_action(ctx, enqueue->port, 0, false);
2008 /* Check output port. */
2009 if (ofp_port == OFPP_IN_PORT) {
2010 ofp_port = ctx->xin->flow.in_port.ofp_port;
2011 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2015 /* Add datapath actions. */
2016 flow_priority = ctx->xin->flow.skb_priority;
2017 ctx->xin->flow.skb_priority = priority;
2018 compose_output_action(ctx, ofp_port);
2019 ctx->xin->flow.skb_priority = flow_priority;
2021 /* Update NetFlow output port. */
2022 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2023 ctx->xout->nf_output_iface = ofp_port;
2024 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2025 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2030 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2032 uint32_t skb_priority;
2034 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2035 ctx->xin->flow.skb_priority = skb_priority;
2037 /* Couldn't translate queue to a priority. Nothing to do. A warning
2038 * has already been logged. */
2043 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2045 const struct xbridge *xbridge = xbridge_;
2056 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2059 port = get_ofp_port(xbridge, ofp_port);
2060 return port ? port->may_enable : false;
2065 xlate_bundle_action(struct xlate_ctx *ctx,
2066 const struct ofpact_bundle *bundle)
2070 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2072 CONST_CAST(struct xbridge *, ctx->xbridge));
2073 if (bundle->dst.field) {
2074 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2077 xlate_output_action(ctx, port, 0, false);
2082 xlate_learn_action(struct xlate_ctx *ctx,
2083 const struct ofpact_learn *learn)
2085 struct ofputil_flow_mod *fm;
2086 struct ofpbuf ofpacts;
2088 ctx->xout->has_learn = true;
2090 learn_mask(learn, &ctx->xout->wc);
2092 if (!ctx->xin->may_learn) {
2096 fm = xmalloc(sizeof *fm);
2097 ofpbuf_init(&ofpacts, 0);
2098 learn_execute(learn, &ctx->xin->flow, fm, &ofpacts);
2100 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
2103 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
2104 * means "infinite". */
2106 reduce_timeout(uint16_t max, uint16_t *timeout)
2108 if (max && (!*timeout || *timeout > max)) {
2114 xlate_fin_timeout(struct xlate_ctx *ctx,
2115 const struct ofpact_fin_timeout *oft)
2117 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2118 struct rule_dpif *rule = ctx->rule;
2120 ovs_mutex_lock(&rule->up.ofproto->expirable_mutex);
2121 if (list_is_empty(&rule->up.expirable)) {
2122 list_insert(&rule->up.ofproto->expirable, &rule->up.expirable);
2124 ovs_mutex_unlock(&rule->up.ofproto->expirable_mutex);
2126 ovs_mutex_lock(&rule->up.timeout_mutex);
2127 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
2128 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
2129 ovs_mutex_unlock(&rule->up.timeout_mutex);
2134 xlate_sample_action(struct xlate_ctx *ctx,
2135 const struct ofpact_sample *os)
2137 union user_action_cookie cookie;
2138 /* Scale the probability from 16-bit to 32-bit while representing
2139 * the same percentage. */
2140 uint32_t probability = (os->probability << 16) | os->probability;
2142 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2143 &ctx->xout->odp_actions, &ctx->xout->wc);
2145 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2146 os->obs_domain_id, os->obs_point_id, &cookie);
2147 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2148 probability, &cookie, sizeof cookie.flow_sample);
2152 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2154 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2155 ? OFPUTIL_PC_NO_RECV_STP
2156 : OFPUTIL_PC_NO_RECV)) {
2160 /* Only drop packets here if both forwarding and learning are
2161 * disabled. If just learning is enabled, we need to have
2162 * OFPP_NORMAL and the learning action have a look at the packet
2163 * before we can drop it. */
2164 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2172 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2173 struct xlate_ctx *ctx)
2175 struct flow_wildcards *wc = &ctx->xout->wc;
2176 struct flow *flow = &ctx->xin->flow;
2177 const struct ofpact *a;
2179 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2180 struct ofpact_controller *controller;
2181 const struct ofpact_metadata *metadata;
2189 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2190 ofpact_get_OUTPUT(a)->max_len, true);
2193 case OFPACT_CONTROLLER:
2194 controller = ofpact_get_CONTROLLER(a);
2195 execute_controller_action(ctx, controller->max_len,
2197 controller->controller_id);
2200 case OFPACT_ENQUEUE:
2201 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2204 case OFPACT_SET_VLAN_VID:
2205 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2206 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2207 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2211 case OFPACT_SET_VLAN_PCP:
2212 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2213 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2215 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2219 case OFPACT_STRIP_VLAN:
2220 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2221 flow->vlan_tci = htons(0);
2224 case OFPACT_PUSH_VLAN:
2225 /* XXX 802.1AD(QinQ) */
2226 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2227 flow->vlan_tci = htons(VLAN_CFI);
2230 case OFPACT_SET_ETH_SRC:
2231 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2232 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2235 case OFPACT_SET_ETH_DST:
2236 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2237 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2240 case OFPACT_SET_IPV4_SRC:
2241 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2242 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2243 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2247 case OFPACT_SET_IPV4_DST:
2248 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2249 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2250 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2254 case OFPACT_SET_IPV4_DSCP:
2255 wc->masks.nw_tos |= IP_DSCP_MASK;
2256 /* OpenFlow 1.0 only supports IPv4. */
2257 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2258 flow->nw_tos &= ~IP_DSCP_MASK;
2259 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2263 case OFPACT_SET_L4_SRC_PORT:
2264 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2265 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2266 if (is_ip_any(flow)) {
2267 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2271 case OFPACT_SET_L4_DST_PORT:
2272 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2273 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2274 if (is_ip_any(flow)) {
2275 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2279 case OFPACT_RESUBMIT:
2280 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2283 case OFPACT_SET_TUNNEL:
2284 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2287 case OFPACT_SET_QUEUE:
2288 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2291 case OFPACT_POP_QUEUE:
2292 flow->skb_priority = ctx->orig_skb_priority;
2295 case OFPACT_REG_MOVE:
2296 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2299 case OFPACT_REG_LOAD:
2300 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2303 case OFPACT_STACK_PUSH:
2304 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2308 case OFPACT_STACK_POP:
2309 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2313 case OFPACT_PUSH_MPLS:
2314 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2317 case OFPACT_POP_MPLS:
2318 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2321 case OFPACT_SET_MPLS_TTL:
2322 if (compose_set_mpls_ttl_action(ctx,
2323 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2328 case OFPACT_DEC_MPLS_TTL:
2329 if (compose_dec_mpls_ttl_action(ctx)) {
2334 case OFPACT_DEC_TTL:
2335 wc->masks.nw_ttl = 0xff;
2336 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2342 /* Nothing to do. */
2345 case OFPACT_MULTIPATH:
2346 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2350 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2353 case OFPACT_OUTPUT_REG:
2354 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2358 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2365 case OFPACT_FIN_TIMEOUT:
2366 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2367 ctx->xout->has_fin_timeout = true;
2368 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2371 case OFPACT_CLEAR_ACTIONS:
2373 * Nothing to do because writa-actions is not supported for now.
2374 * When writa-actions is supported, clear-actions also must
2375 * be supported at the same time.
2379 case OFPACT_WRITE_METADATA:
2380 metadata = ofpact_get_WRITE_METADATA(a);
2381 flow->metadata &= ~metadata->mask;
2382 flow->metadata |= metadata->metadata & metadata->mask;
2386 /* Not implemented yet. */
2389 case OFPACT_GOTO_TABLE: {
2390 /* It is assumed that goto-table is the last action. */
2391 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2393 ovs_assert(ctx->table_id < ogt->table_id);
2394 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2395 ogt->table_id, true);
2400 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2407 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2408 const struct flow *flow, struct rule_dpif *rule,
2409 uint8_t tcp_flags, const struct ofpbuf *packet)
2411 xin->ofproto = ofproto;
2413 xin->packet = packet;
2414 xin->may_learn = packet != NULL;
2416 xin->ofpacts = NULL;
2417 xin->ofpacts_len = 0;
2418 xin->tcp_flags = tcp_flags;
2419 xin->resubmit_hook = NULL;
2420 xin->report_hook = NULL;
2421 xin->resubmit_stats = NULL;
2425 xlate_out_uninit(struct xlate_out *xout)
2428 ofpbuf_uninit(&xout->odp_actions);
2432 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2433 * into datapath actions, using 'ctx', and discards the datapath actions. */
2435 xlate_actions_for_side_effects(struct xlate_in *xin)
2437 struct xlate_out xout;
2439 xlate_actions(xin, &xout);
2440 xlate_out_uninit(&xout);
2444 xlate_report(struct xlate_ctx *ctx, const char *s)
2446 if (ctx->xin->report_hook) {
2447 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2452 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2455 dst->slow = src->slow;
2456 dst->has_learn = src->has_learn;
2457 dst->has_normal = src->has_normal;
2458 dst->has_fin_timeout = src->has_fin_timeout;
2459 dst->nf_output_iface = src->nf_output_iface;
2460 dst->mirrors = src->mirrors;
2462 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2463 sizeof dst->odp_actions_stub);
2464 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2465 src->odp_actions.size);
2468 static struct skb_priority_to_dscp *
2469 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2471 struct skb_priority_to_dscp *pdscp;
2474 hash = hash_int(skb_priority, 0);
2475 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2476 if (pdscp->skb_priority == skb_priority) {
2484 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2487 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2488 *dscp = pdscp ? pdscp->dscp : 0;
2489 return pdscp != NULL;
2493 clear_skb_priorities(struct xport *xport)
2495 struct skb_priority_to_dscp *pdscp, *next;
2497 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2498 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2504 actions_output_to_local_port(const struct xlate_ctx *ctx)
2506 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2507 const struct nlattr *a;
2510 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2511 ctx->xout->odp_actions.size) {
2512 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2513 && nl_attr_get_odp_port(a) == local_odp_port) {
2520 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2521 * into datapath actions in 'odp_actions', using 'ctx'. */
2523 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2525 struct flow_wildcards *wc = &xout->wc;
2526 struct flow *flow = &xin->flow;
2528 enum slow_path_reason special;
2529 const struct ofpact *ofpacts;
2530 struct xport *in_port;
2531 struct flow orig_flow;
2532 struct xlate_ctx ctx;
2536 COVERAGE_INC(xlate_actions);
2538 ovs_rwlock_rdlock(&xlate_rwlock);
2540 /* Flow initialization rules:
2541 * - 'base_flow' must match the kernel's view of the packet at the
2542 * time that action processing starts. 'flow' represents any
2543 * transformations we wish to make through actions.
2544 * - By default 'base_flow' and 'flow' are the same since the input
2545 * packet matches the output before any actions are applied.
2546 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2547 * of the received packet as seen by the kernel. If we later output
2548 * to another device without any modifications this will cause us to
2549 * insert a new tag since the original one was stripped off by the
2551 * - Tunnel metadata as received is retained in 'flow'. This allows
2552 * tunnel metadata matching also in later tables.
2553 * Since a kernel action for setting the tunnel metadata will only be
2554 * generated with actual tunnel output, changing the tunnel metadata
2555 * values in 'flow' (such as tun_id) will only have effect with a later
2556 * tunnel output action.
2557 * - Tunnel 'base_flow' is completely cleared since that is what the
2558 * kernel does. If we wish to maintain the original values an action
2559 * needs to be generated. */
2564 ctx.xout->has_learn = false;
2565 ctx.xout->has_normal = false;
2566 ctx.xout->has_fin_timeout = false;
2567 ctx.xout->nf_output_iface = NF_OUT_DROP;
2568 ctx.xout->mirrors = 0;
2569 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2570 sizeof ctx.xout->odp_actions_stub);
2571 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2573 ctx.xbridge = xbridge_lookup(xin->ofproto);
2578 ctx.rule = xin->rule;
2580 ctx.base_flow = *flow;
2581 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2582 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2584 flow_wildcards_init_catchall(wc);
2585 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2586 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2587 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2588 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2590 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
2591 if (ctx.xbridge->has_netflow) {
2592 netflow_mask_wc(flow, wc);
2596 ctx.orig_skb_priority = flow->skb_priority;
2601 ofpacts = xin->ofpacts;
2602 ofpacts_len = xin->ofpacts_len;
2603 } else if (xin->rule) {
2604 ofpacts = xin->rule->up.ofpacts;
2605 ofpacts_len = xin->rule->up.ofpacts_len;
2610 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2612 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2613 /* Do this conditionally because the copy is expensive enough that it
2614 * shows up in profiles. */
2618 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2619 switch (ctx.xbridge->frag) {
2620 case OFPC_FRAG_NORMAL:
2621 /* We must pretend that transport ports are unavailable. */
2622 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2623 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2626 case OFPC_FRAG_DROP:
2629 case OFPC_FRAG_REASM:
2632 case OFPC_FRAG_NX_MATCH:
2633 /* Nothing to do. */
2636 case OFPC_INVALID_TTL_TO_CONTROLLER:
2641 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2642 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2644 ctx.xout->slow = special;
2646 size_t sample_actions_len;
2648 if (flow->in_port.ofp_port
2649 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2650 flow->in_port.ofp_port,
2652 ctx.base_flow.vlan_tci = 0;
2655 add_sflow_action(&ctx);
2656 add_ipfix_action(&ctx);
2657 sample_actions_len = ctx.xout->odp_actions.size;
2659 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
2660 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2662 /* We've let OFPP_NORMAL and the learning action look at the
2663 * packet, so drop it now if forwarding is disabled. */
2664 if (in_port && !xport_stp_forward_state(in_port)) {
2665 ctx.xout->odp_actions.size = sample_actions_len;
2669 if (ctx.xbridge->has_in_band
2670 && in_band_must_output_to_local_port(flow)
2671 && !actions_output_to_local_port(&ctx)) {
2672 compose_output_action(&ctx, OFPP_LOCAL);
2675 fix_sflow_action(&ctx);
2677 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2678 add_mirror_actions(&ctx, &orig_flow);
2682 ofpbuf_uninit(&ctx.stack);
2684 /* Clear the metadata and register wildcard masks, because we won't
2685 * use non-header fields as part of the cache. */
2686 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2687 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
2690 ovs_rwlock_unlock(&xlate_rwlock);