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
59 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
60 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
62 struct list xbundles; /* Owned xbundles. */
63 struct hmap xports; /* Indexed by ofp_port. */
65 char *name; /* Name used in log messages. */
66 struct dpif *dpif; /* Datapath interface. */
67 struct mac_learning *ml; /* Mac learning handle. */
68 struct mbridge *mbridge; /* Mirroring. */
69 struct dpif_sflow *sflow; /* SFlow handle, or null. */
70 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
71 struct stp *stp; /* STP or null if disabled. */
73 /* Special rules installed by ofproto-dpif. */
74 struct rule_dpif *miss_rule;
75 struct rule_dpif *no_packet_in_rule;
77 enum ofp_config_flags frag; /* Fragmentation handling. */
78 bool has_netflow; /* Bridge runs netflow? */
79 bool has_in_band; /* Bridge has in band control? */
80 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
84 struct hmap_node hmap_node; /* In global 'xbundles' map. */
85 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
87 struct list list_node; /* In parent 'xbridges' list. */
88 struct xbridge *xbridge; /* Parent xbridge. */
90 struct list xports; /* Contains "struct xport"s. */
92 char *name; /* Name used in log messages. */
93 struct bond *bond; /* Nonnull iff more than one port. */
94 struct lacp *lacp; /* LACP handle or null. */
96 enum port_vlan_mode vlan_mode; /* VLAN mode. */
97 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
98 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
99 * NULL if all VLANs are trunked. */
100 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
101 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
105 struct hmap_node hmap_node; /* Node in global 'xports' map. */
106 struct ofport_dpif *ofport; /* Key in global 'xports map. */
108 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
109 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
111 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
113 struct list bundle_node; /* In parent xbundle (if it exists). */
114 struct xbundle *xbundle; /* Parent xbundle or null. */
116 struct netdev *netdev; /* 'ofport''s netdev. */
118 struct xbridge *xbridge; /* Parent bridge. */
119 struct xport *peer; /* Patch port peer or null. */
121 enum ofputil_port_config config; /* OpenFlow port configuration. */
122 int stp_port_no; /* STP port number or 0 if not in use. */
124 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
126 bool may_enable; /* May be enabled in bonds. */
127 bool is_tunnel; /* Is a tunnel port. */
129 struct cfm *cfm; /* CFM handle or null. */
130 struct bfd *bfd; /* BFD handle or null. */
134 struct xlate_in *xin;
135 struct xlate_out *xout;
137 const struct xbridge *xbridge;
139 /* Flow at the last commit. */
140 struct flow base_flow;
142 /* Tunnel IP destination address as received. This is stored separately
143 * as the base_flow.tunnel is cleared on init to reflect the datapath
144 * behavior. Used to make sure not to send tunneled output to ourselves,
145 * which might lead to an infinite loop. This could happen easily
146 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
147 * actually set the tun_dst field. */
148 ovs_be32 orig_tunnel_ip_dst;
150 /* Stack for the push and pop actions. Each stack element is of type
151 * "union mf_subvalue". */
152 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
155 /* The rule that we are currently translating, or NULL. */
156 struct rule_dpif *rule;
158 int recurse; /* Recursion level, via xlate_table_action. */
159 uint32_t orig_skb_priority; /* Priority when packet arrived. */
160 uint8_t table_id; /* OpenFlow table ID where flow was found. */
161 uint32_t sflow_n_outputs; /* Number of output ports. */
162 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
163 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
164 bool exit; /* No further actions should be processed. */
167 /* A controller may use OFPP_NONE as the ingress port to indicate that
168 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
169 * when an input bundle is needed for validation (e.g., mirroring or
170 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
171 * any 'port' structs, so care must be taken when dealing with it.
172 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
173 static struct xbundle ofpp_none_bundle;
175 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
176 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
177 * traffic egressing the 'ofport' with that priority should be marked with. */
178 struct skb_priority_to_dscp {
179 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
180 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
182 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
185 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
186 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
187 static struct hmap xports = HMAP_INITIALIZER(&xports);
189 static bool may_receive(const struct xport *, struct xlate_ctx *);
190 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
192 static void xlate_normal(struct xlate_ctx *);
193 static void xlate_report(struct xlate_ctx *, const char *);
194 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
195 uint8_t table_id, bool may_packet_in);
196 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
197 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
198 static void output_normal(struct xlate_ctx *, const struct xbundle *,
200 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
202 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
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 fitness = odp_flow_key_to_flow(key, key_len, flow);
506 if (fitness == ODP_FIT_ERROR) {
512 *odp_in_port = flow->in_port.odp_port;
515 xport = xport_lookup(tnl_port_should_receive(flow)
516 ? tnl_port_receive(flow)
517 : odp_port_to_ofport(backer, flow->in_port.odp_port));
519 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
524 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
526 /* Make the packet resemble the flow, so that it gets sent to
527 * an OpenFlow controller properly, so that it looks correct
528 * for sFlow, and so that flow_extract() will get the correct
529 * vlan_tci if it is called on 'packet'.
531 * The allocated space inside 'packet' probably also contains
532 * 'key', that is, both 'packet' and 'key' are probably part of
533 * a struct dpif_upcall (see the large comment on that
534 * structure definition), so pushing data on 'packet' is in
535 * general not a good idea since it could overwrite 'key' or
536 * free it as a side effect. However, it's OK in this special
537 * case because we know that 'packet' is inside a Netlink
538 * attribute: pushing 4 bytes will just overwrite the 4-byte
539 * "struct nlattr", which is fine since we don't need that
541 eth_push_vlan(packet, flow->vlan_tci);
543 /* We can't reproduce 'key' from 'flow'. */
544 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
549 *ofproto = xport->xbridge->ofproto;
559 static struct xbridge *
560 xbridge_lookup(const struct ofproto_dpif *ofproto)
562 struct xbridge *xbridge;
568 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
570 if (xbridge->ofproto == ofproto) {
577 static struct xbundle *
578 xbundle_lookup(const struct ofbundle *ofbundle)
580 struct xbundle *xbundle;
586 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
588 if (xbundle->ofbundle == ofbundle) {
595 static struct xport *
596 xport_lookup(const struct ofport_dpif *ofport)
604 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
606 if (xport->ofport == ofport) {
613 static struct stp_port *
614 xport_get_stp_port(const struct xport *xport)
616 return xport->xbridge->stp && xport->stp_port_no
617 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
621 static enum stp_state
622 xport_stp_learn_state(const struct xport *xport)
624 struct stp_port *sp = xport_get_stp_port(xport);
625 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
629 xport_stp_forward_state(const struct xport *xport)
631 struct stp_port *sp = xport_get_stp_port(xport);
632 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
635 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
636 * were used to make the determination.*/
638 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
640 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
641 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
645 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
647 struct stp_port *sp = xport_get_stp_port(xport);
648 struct ofpbuf payload = *packet;
649 struct eth_header *eth = payload.data;
651 /* Sink packets on ports that have STP disabled when the bridge has
653 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
657 /* Trim off padding on payload. */
658 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
659 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
662 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
663 stp_received_bpdu(sp, payload.data, payload.size);
667 static struct xport *
668 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
672 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
674 if (xport->ofp_port == ofp_port) {
682 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
684 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
685 return xport ? xport->odp_port : ODPP_NONE;
689 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
691 return (bundle->vlan_mode != PORT_VLAN_ACCESS
692 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
696 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
698 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
702 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
704 return xbundle != &ofpp_none_bundle
705 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
710 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
712 return xbundle != &ofpp_none_bundle
713 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
718 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
720 return xbundle != &ofpp_none_bundle
721 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
725 static struct xbundle *
726 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
727 bool warn, struct xport **in_xportp)
731 /* Find the port and bundle for the received packet. */
732 xport = get_ofp_port(xbridge, in_port);
736 if (xport && xport->xbundle) {
737 return xport->xbundle;
740 /* Special-case OFPP_NONE, which a controller may use as the ingress
741 * port for traffic that it is sourcing. */
742 if (in_port == OFPP_NONE) {
743 ofpp_none_bundle.name = "OFPP_NONE";
744 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
745 return &ofpp_none_bundle;
748 /* Odd. A few possible reasons here:
750 * - We deleted a port but there are still a few packets queued up
753 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
754 * we don't know about.
756 * - The ofproto client didn't configure the port as part of a bundle.
757 * This is particularly likely to happen if a packet was received on the
758 * port after it was created, but before the client had a chance to
759 * configure its bundle.
762 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
764 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
765 "port %"PRIu16, xbridge->name, in_port);
771 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
773 const struct xbridge *xbridge = ctx->xbridge;
774 mirror_mask_t mirrors;
775 struct xbundle *in_xbundle;
779 mirrors = ctx->xout->mirrors;
780 ctx->xout->mirrors = 0;
782 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
783 ctx->xin->packet != NULL, NULL);
787 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
789 /* Drop frames on bundles reserved for mirroring. */
790 if (xbundle_mirror_out(xbridge, in_xbundle)) {
791 if (ctx->xin->packet != NULL) {
792 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
793 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
794 "%s, which is reserved exclusively for mirroring",
795 ctx->xbridge->name, in_xbundle->name);
797 ofpbuf_clear(&ctx->xout->odp_actions);
802 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
803 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
806 vlan = input_vid_to_vlan(in_xbundle, vid);
812 /* Restore the original packet before adding the mirror actions. */
813 ctx->xin->flow = *orig_flow;
816 mirror_mask_t dup_mirrors;
817 struct ofbundle *out;
818 unsigned long *vlans;
823 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
824 &vlans, &dup_mirrors, &out, &out_vlan);
825 ovs_assert(has_mirror);
828 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
830 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
833 if (!vlan_mirrored) {
834 mirrors = zero_rightmost_1bit(mirrors);
838 mirrors &= ~dup_mirrors;
839 ctx->xout->mirrors |= dup_mirrors;
841 struct xbundle *out_xbundle = xbundle_lookup(out);
843 output_normal(ctx, out_xbundle, vlan);
845 } else if (vlan != out_vlan
846 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
847 struct xbundle *xbundle;
849 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
850 if (xbundle_includes_vlan(xbundle, out_vlan)
851 && !xbundle_mirror_out(xbridge, xbundle)) {
852 output_normal(ctx, xbundle, out_vlan);
859 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
860 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
861 * the bundle on which the packet was received, returns the VLAN to which the
864 * Both 'vid' and the return value are in the range 0...4095. */
866 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
868 switch (in_xbundle->vlan_mode) {
869 case PORT_VLAN_ACCESS:
870 return in_xbundle->vlan;
873 case PORT_VLAN_TRUNK:
876 case PORT_VLAN_NATIVE_UNTAGGED:
877 case PORT_VLAN_NATIVE_TAGGED:
878 return vid ? vid : in_xbundle->vlan;
885 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
886 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
889 * 'vid' should be the VID obtained from the 802.1Q header that was received as
890 * part of a packet (specify 0 if there was no 802.1Q header), in the range
893 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
895 /* Allow any VID on the OFPP_NONE port. */
896 if (in_xbundle == &ofpp_none_bundle) {
900 switch (in_xbundle->vlan_mode) {
901 case PORT_VLAN_ACCESS:
904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
905 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
906 "packet received on port %s configured as VLAN "
907 "%"PRIu16" access port", vid, in_xbundle->name,
914 case PORT_VLAN_NATIVE_UNTAGGED:
915 case PORT_VLAN_NATIVE_TAGGED:
917 /* Port must always carry its native VLAN. */
921 case PORT_VLAN_TRUNK:
922 if (!xbundle_includes_vlan(in_xbundle, vid)) {
924 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
925 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
926 "received on port %s not configured for trunking "
927 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
939 /* Given 'vlan', the VLAN that a packet belongs to, and
940 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
941 * that should be included in the 802.1Q header. (If the return value is 0,
942 * then the 802.1Q header should only be included in the packet if there is a
945 * Both 'vlan' and the return value are in the range 0...4095. */
947 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
949 switch (out_xbundle->vlan_mode) {
950 case PORT_VLAN_ACCESS:
953 case PORT_VLAN_TRUNK:
954 case PORT_VLAN_NATIVE_TAGGED:
957 case PORT_VLAN_NATIVE_UNTAGGED:
958 return vlan == out_xbundle->vlan ? 0 : vlan;
966 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
969 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
971 ovs_be16 tci, old_tci;
974 vid = output_vlan_to_vid(out_xbundle, vlan);
975 if (list_is_empty(&out_xbundle->xports)) {
976 /* Partially configured bundle with no slaves. Drop the packet. */
978 } else if (!out_xbundle->bond) {
979 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
982 struct ofport_dpif *ofport;
984 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
985 &ctx->xout->wc, vid);
986 xport = xport_lookup(ofport);
989 /* No slaves enabled, so drop packet. */
996 if (tci || out_xbundle->use_priority_tags) {
997 tci |= *flow_tci & htons(VLAN_PCP_MASK);
999 tci |= htons(VLAN_CFI);
1004 compose_output_action(ctx, xport->ofp_port);
1005 *flow_tci = old_tci;
1008 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1009 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1010 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1012 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1014 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1018 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1019 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1023 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1024 if (flow->nw_proto == ARP_OP_REPLY) {
1026 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1027 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1028 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1030 return flow->nw_src == flow->nw_dst;
1037 update_learning_table(const struct xbridge *xbridge,
1038 const struct flow *flow, struct flow_wildcards *wc,
1039 int vlan, struct xbundle *in_xbundle)
1041 struct mac_entry *mac;
1043 /* Don't learn the OFPP_NONE port. */
1044 if (in_xbundle == &ofpp_none_bundle) {
1048 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1049 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1053 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1054 if (is_gratuitous_arp(flow, wc)) {
1055 /* We don't want to learn from gratuitous ARP packets that are
1056 * reflected back over bond slaves so we lock the learning table. */
1057 if (!in_xbundle->bond) {
1058 mac_entry_set_grat_arp_lock(mac);
1059 } else if (mac_entry_is_grat_arp_locked(mac)) {
1064 if (mac->port.p != in_xbundle->ofbundle) {
1065 /* The log messages here could actually be useful in debugging,
1066 * so keep the rate limit relatively high. */
1067 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1068 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1069 "on port %s in VLAN %d",
1070 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1071 in_xbundle->name, vlan);
1073 mac->port.p = in_xbundle->ofbundle;
1074 mac_learning_changed(xbridge->ml);
1077 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1080 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1081 * dropped. Returns true if they may be forwarded, false if they should be
1084 * 'in_port' must be the xport that corresponds to flow->in_port.
1085 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1087 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1088 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1089 * checked by input_vid_is_valid().
1091 * May also add tags to '*tags', although the current implementation only does
1092 * so in one special case.
1095 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1098 struct xbundle *in_xbundle = in_port->xbundle;
1099 const struct xbridge *xbridge = ctx->xbridge;
1100 struct flow *flow = &ctx->xin->flow;
1102 /* Drop frames for reserved multicast addresses
1103 * only if forward_bpdu option is absent. */
1104 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1105 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1109 if (in_xbundle->bond) {
1110 struct mac_entry *mac;
1112 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1118 xlate_report(ctx, "bonding refused admissibility, dropping");
1121 case BV_DROP_IF_MOVED:
1122 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1123 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1124 if (mac && mac->port.p != in_xbundle->ofbundle &&
1125 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1126 || mac_entry_is_grat_arp_locked(mac))) {
1127 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1128 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1132 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1141 xlate_normal(struct xlate_ctx *ctx)
1143 struct flow_wildcards *wc = &ctx->xout->wc;
1144 struct flow *flow = &ctx->xin->flow;
1145 struct xbundle *in_xbundle;
1146 struct xport *in_port;
1147 struct mac_entry *mac;
1151 ctx->xout->has_normal = true;
1153 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1154 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1155 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1157 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1158 ctx->xin->packet != NULL, &in_port);
1160 xlate_report(ctx, "no input bundle, dropping");
1164 /* Drop malformed frames. */
1165 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1166 !(flow->vlan_tci & htons(VLAN_CFI))) {
1167 if (ctx->xin->packet != NULL) {
1168 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1169 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1170 "VLAN tag received on port %s",
1171 ctx->xbridge->name, in_xbundle->name);
1173 xlate_report(ctx, "partial VLAN tag, dropping");
1177 /* Drop frames on bundles reserved for mirroring. */
1178 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1179 if (ctx->xin->packet != NULL) {
1180 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1181 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1182 "%s, which is reserved exclusively for mirroring",
1183 ctx->xbridge->name, in_xbundle->name);
1185 xlate_report(ctx, "input port is mirror output port, dropping");
1190 vid = vlan_tci_to_vid(flow->vlan_tci);
1191 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1192 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1195 vlan = input_vid_to_vlan(in_xbundle, vid);
1197 /* Check other admissibility requirements. */
1198 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1202 /* Learn source MAC. */
1203 if (ctx->xin->may_learn) {
1204 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1207 /* Determine output bundle. */
1208 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1209 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1211 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
1212 if (mac_xbundle && mac_xbundle != in_xbundle) {
1213 xlate_report(ctx, "forwarding to learned port");
1214 output_normal(ctx, mac_xbundle, vlan);
1215 } else if (!mac_xbundle) {
1216 xlate_report(ctx, "learned port is unknown, dropping");
1218 xlate_report(ctx, "learned port is input port, dropping");
1221 struct xbundle *xbundle;
1223 xlate_report(ctx, "no learned MAC for destination, flooding");
1224 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1225 if (xbundle != in_xbundle
1226 && xbundle_includes_vlan(xbundle, vlan)
1227 && xbundle->floodable
1228 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1229 output_normal(ctx, xbundle, vlan);
1232 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1234 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1237 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1238 * the number of packets out of UINT32_MAX to sample. The given
1239 * cookie is passed back in the callback for each sampled packet.
1242 compose_sample_action(const struct xbridge *xbridge,
1243 struct ofpbuf *odp_actions,
1244 const struct flow *flow,
1245 const uint32_t probability,
1246 const union user_action_cookie *cookie,
1247 const size_t cookie_size)
1249 size_t sample_offset, actions_offset;
1250 odp_port_t odp_port;
1254 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1256 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1258 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1260 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1261 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1262 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1264 nl_msg_end_nested(odp_actions, actions_offset);
1265 nl_msg_end_nested(odp_actions, sample_offset);
1266 return cookie_offset;
1270 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1271 odp_port_t odp_port, unsigned int n_outputs,
1272 union user_action_cookie *cookie)
1276 cookie->type = USER_ACTION_COOKIE_SFLOW;
1277 cookie->sflow.vlan_tci = vlan_tci;
1279 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1280 * port information") for the interpretation of cookie->output. */
1281 switch (n_outputs) {
1283 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1284 cookie->sflow.output = 0x40000000 | 256;
1288 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1290 cookie->sflow.output = ifindex;
1295 /* 0x80000000 means "multiple output ports. */
1296 cookie->sflow.output = 0x80000000 | n_outputs;
1301 /* Compose SAMPLE action for sFlow bridge sampling. */
1303 compose_sflow_action(const struct xbridge *xbridge,
1304 struct ofpbuf *odp_actions,
1305 const struct flow *flow,
1306 odp_port_t odp_port)
1308 uint32_t probability;
1309 union user_action_cookie cookie;
1311 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1315 probability = dpif_sflow_get_probability(xbridge->sflow);
1316 compose_sflow_cookie(xbridge, htons(0), odp_port,
1317 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1319 return compose_sample_action(xbridge, odp_actions, flow, probability,
1320 &cookie, sizeof cookie.sflow);
1324 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1325 uint32_t obs_domain_id, uint32_t obs_point_id,
1326 union user_action_cookie *cookie)
1328 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1329 cookie->flow_sample.probability = probability;
1330 cookie->flow_sample.collector_set_id = collector_set_id;
1331 cookie->flow_sample.obs_domain_id = obs_domain_id;
1332 cookie->flow_sample.obs_point_id = obs_point_id;
1336 compose_ipfix_cookie(union user_action_cookie *cookie)
1338 cookie->type = USER_ACTION_COOKIE_IPFIX;
1341 /* Compose SAMPLE action for IPFIX bridge sampling. */
1343 compose_ipfix_action(const struct xbridge *xbridge,
1344 struct ofpbuf *odp_actions,
1345 const struct flow *flow)
1347 uint32_t probability;
1348 union user_action_cookie cookie;
1350 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1354 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1355 compose_ipfix_cookie(&cookie);
1357 compose_sample_action(xbridge, odp_actions, flow, probability,
1358 &cookie, sizeof cookie.ipfix);
1361 /* SAMPLE action for sFlow must be first action in any given list of
1362 * actions. At this point we do not have all information required to
1363 * build it. So try to build sample action as complete as possible. */
1365 add_sflow_action(struct xlate_ctx *ctx)
1367 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1368 &ctx->xout->odp_actions,
1369 &ctx->xin->flow, ODPP_NONE);
1370 ctx->sflow_odp_port = 0;
1371 ctx->sflow_n_outputs = 0;
1374 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1375 * of actions, eventually after the SAMPLE action for sFlow. */
1377 add_ipfix_action(struct xlate_ctx *ctx)
1379 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1383 /* Fix SAMPLE action according to data collected while composing ODP actions.
1384 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1385 * USERSPACE action's user-cookie which is required for sflow. */
1387 fix_sflow_action(struct xlate_ctx *ctx)
1389 const struct flow *base = &ctx->base_flow;
1390 union user_action_cookie *cookie;
1392 if (!ctx->user_cookie_offset) {
1396 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1397 sizeof cookie->sflow);
1398 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1400 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1401 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1404 static enum slow_path_reason
1405 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1406 const struct xport *xport, const struct ofpbuf *packet)
1408 struct flow_wildcards *wc = &ctx->xout->wc;
1409 const struct xbridge *xbridge = ctx->xbridge;
1413 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1415 cfm_process_heartbeat(xport->cfm, packet);
1418 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1420 bfd_process_packet(xport->bfd, flow, packet);
1423 } else if (xport->xbundle && xport->xbundle->lacp
1424 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1426 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1429 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1431 stp_process_packet(xport, packet);
1440 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1443 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1444 struct flow_wildcards *wc = &ctx->xout->wc;
1445 struct flow *flow = &ctx->xin->flow;
1446 ovs_be16 flow_vlan_tci;
1447 uint32_t flow_skb_mark;
1448 uint8_t flow_nw_tos;
1449 odp_port_t out_port, odp_port;
1452 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1453 * before traversing a patch port. */
1454 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1457 xlate_report(ctx, "Nonexistent output port");
1459 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1460 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1462 } else if (check_stp && !xport_stp_forward_state(xport)) {
1463 xlate_report(ctx, "STP not in forwarding state, skipping output");
1467 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1468 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1473 const struct xport *peer = xport->peer;
1474 struct flow old_flow = ctx->xin->flow;
1475 enum slow_path_reason special;
1477 ctx->xbridge = peer->xbridge;
1478 flow->in_port.ofp_port = peer->ofp_port;
1479 flow->metadata = htonll(0);
1480 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1481 memset(flow->regs, 0, sizeof flow->regs);
1483 special = process_special(ctx, &ctx->xin->flow, peer,
1486 ctx->xout->slow = special;
1487 } else if (may_receive(peer, ctx)) {
1488 if (xport_stp_forward_state(peer)) {
1489 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1491 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1492 * learning action look at the packet, then drop it. */
1493 struct flow old_base_flow = ctx->base_flow;
1494 size_t old_size = ctx->xout->odp_actions.size;
1495 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1496 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1497 ctx->xout->mirrors = old_mirrors;
1498 ctx->base_flow = old_base_flow;
1499 ctx->xout->odp_actions.size = old_size;
1503 ctx->xin->flow = old_flow;
1504 ctx->xbridge = xport->xbridge;
1506 if (ctx->xin->resubmit_stats) {
1507 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1508 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1514 flow_vlan_tci = flow->vlan_tci;
1515 flow_skb_mark = flow->skb_mark;
1516 flow_nw_tos = flow->nw_tos;
1518 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1519 wc->masks.nw_tos |= IP_ECN_MASK;
1520 flow->nw_tos &= ~IP_DSCP_MASK;
1521 flow->nw_tos |= dscp;
1524 if (xport->is_tunnel) {
1525 /* Save tunnel metadata so that changes made due to
1526 * the Logical (tunnel) Port are not visible for any further
1527 * matches, while explicit set actions on tunnel metadata are.
1529 struct flow_tnl flow_tnl = flow->tunnel;
1530 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1531 if (odp_port == ODPP_NONE) {
1532 xlate_report(ctx, "Tunneling decided against output");
1533 goto out; /* restore flow_nw_tos */
1535 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1536 xlate_report(ctx, "Not tunneling to our own address");
1537 goto out; /* restore flow_nw_tos */
1539 if (ctx->xin->resubmit_stats) {
1540 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1542 out_port = odp_port;
1543 commit_odp_tunnel_action(flow, &ctx->base_flow,
1544 &ctx->xout->odp_actions);
1545 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1547 ofp_port_t vlandev_port;
1549 odp_port = xport->odp_port;
1550 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1551 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1553 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1555 if (vlandev_port == ofp_port) {
1556 out_port = odp_port;
1558 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1559 flow->vlan_tci = htons(0);
1561 flow->skb_mark &= ~IPSEC_MARK;
1564 if (out_port != ODPP_NONE) {
1565 commit_odp_actions(flow, &ctx->base_flow,
1566 &ctx->xout->odp_actions, &ctx->xout->wc);
1567 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1570 ctx->sflow_odp_port = odp_port;
1571 ctx->sflow_n_outputs++;
1572 ctx->xout->nf_output_iface = ofp_port;
1577 flow->vlan_tci = flow_vlan_tci;
1578 flow->skb_mark = flow_skb_mark;
1579 flow->nw_tos = flow_nw_tos;
1583 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1585 compose_output_action__(ctx, ofp_port, true);
1588 /* Common rule processing in one place to avoid duplicating code. */
1589 static struct rule_dpif *
1590 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
1593 if (ctx->xin->resubmit_hook) {
1594 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1596 if (rule == NULL && may_packet_in) {
1597 struct xport *xport;
1600 * check if table configuration flags
1601 * OFPTC_TABLE_MISS_CONTROLLER, default.
1602 * OFPTC_TABLE_MISS_CONTINUE,
1603 * OFPTC_TABLE_MISS_DROP
1604 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1605 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1606 rule = choose_miss_rule(xport ? xport->config : 0,
1607 ctx->xbridge->miss_rule,
1608 ctx->xbridge->no_packet_in_rule);
1610 if (rule && ctx->xin->resubmit_stats) {
1611 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1617 xlate_table_action(struct xlate_ctx *ctx,
1618 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1620 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1621 struct rule_dpif *rule;
1622 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1623 uint8_t old_table_id = ctx->table_id;
1625 ctx->table_id = table_id;
1627 /* Look up a flow with 'in_port' as the input port. */
1628 ctx->xin->flow.in_port.ofp_port = in_port;
1629 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1630 &ctx->xin->flow, &ctx->xout->wc,
1633 /* Restore the original input port. Otherwise OFPP_NORMAL and
1634 * OFPP_IN_PORT will have surprising behavior. */
1635 ctx->xin->flow.in_port.ofp_port = old_in_port;
1637 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1640 struct rule_dpif *old_rule = ctx->rule;
1644 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1645 ctx->rule = old_rule;
1649 ctx->table_id = old_table_id;
1651 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1653 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1654 MAX_RESUBMIT_RECURSION);
1659 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1660 const struct ofpact_resubmit *resubmit)
1665 in_port = resubmit->in_port;
1666 if (in_port == OFPP_IN_PORT) {
1667 in_port = ctx->xin->flow.in_port.ofp_port;
1670 table_id = resubmit->table_id;
1671 if (table_id == 255) {
1672 table_id = ctx->table_id;
1675 xlate_table_action(ctx, in_port, table_id, false);
1679 flood_packets(struct xlate_ctx *ctx, bool all)
1681 const struct xport *xport;
1683 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1684 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1689 compose_output_action__(ctx, xport->ofp_port, false);
1690 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1691 compose_output_action(ctx, xport->ofp_port);
1695 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1699 execute_controller_action(struct xlate_ctx *ctx, int len,
1700 enum ofp_packet_in_reason reason,
1701 uint16_t controller_id)
1703 struct ofputil_packet_in *pin;
1704 struct ofpbuf *packet;
1707 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1708 ctx->xout->slow = SLOW_CONTROLLER;
1709 if (!ctx->xin->packet) {
1713 packet = ofpbuf_clone(ctx->xin->packet);
1715 key.skb_priority = 0;
1717 memset(&key.tunnel, 0, sizeof key.tunnel);
1719 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1720 &ctx->xout->odp_actions, &ctx->xout->wc);
1722 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1723 ctx->xout->odp_actions.size, NULL, NULL);
1725 pin = xmalloc(sizeof *pin);
1726 pin->packet_len = packet->size;
1727 pin->packet = ofpbuf_steal_data(packet);
1728 pin->reason = reason;
1729 pin->controller_id = controller_id;
1730 pin->table_id = ctx->table_id;
1731 pin->cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1733 pin->send_len = len;
1734 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1736 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1737 ofpbuf_delete(packet);
1741 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1743 struct flow_wildcards *wc = &ctx->xout->wc;
1744 struct flow *flow = &ctx->xin->flow;
1746 ovs_assert(eth_type_mpls(eth_type));
1748 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1749 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1751 if (flow->mpls_depth) {
1752 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1758 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1759 label = htonl(0x2); /* IPV6 Explicit Null. */
1761 label = htonl(0x0); /* IPV4 Explicit Null. */
1763 wc->masks.nw_tos |= IP_DSCP_MASK;
1764 wc->masks.nw_ttl = 0xff;
1765 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1766 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1767 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1768 flow->mpls_depth = 1;
1770 flow->dl_type = eth_type;
1774 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1776 struct flow_wildcards *wc = &ctx->xout->wc;
1777 struct flow *flow = &ctx->xin->flow;
1779 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1780 ovs_assert(!eth_type_mpls(eth_type));
1782 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1783 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1785 if (flow->mpls_depth) {
1787 flow->mpls_lse = htonl(0);
1788 if (!flow->mpls_depth) {
1789 flow->dl_type = eth_type;
1795 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1797 struct flow *flow = &ctx->xin->flow;
1799 if (!is_ip_any(flow)) {
1803 ctx->xout->wc.masks.nw_ttl = 0xff;
1804 if (flow->nw_ttl > 1) {
1810 for (i = 0; i < ids->n_controllers; i++) {
1811 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1815 /* Stop processing for current table. */
1821 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1823 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1827 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1828 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1833 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1835 struct flow *flow = &ctx->xin->flow;
1836 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1837 struct flow_wildcards *wc = &ctx->xout->wc;
1839 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1841 if (!eth_type_mpls(flow->dl_type)) {
1847 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1850 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1852 /* Stop processing for current table. */
1858 xlate_output_action(struct xlate_ctx *ctx,
1859 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1861 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1863 ctx->xout->nf_output_iface = NF_OUT_DROP;
1867 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1870 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1877 flood_packets(ctx, false);
1880 flood_packets(ctx, true);
1882 case OFPP_CONTROLLER:
1883 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1889 if (port != ctx->xin->flow.in_port.ofp_port) {
1890 compose_output_action(ctx, port);
1892 xlate_report(ctx, "skipping output to input port");
1897 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1898 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1899 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1900 ctx->xout->nf_output_iface = prev_nf_output_iface;
1901 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1902 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1903 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1908 xlate_output_reg_action(struct xlate_ctx *ctx,
1909 const struct ofpact_output_reg *or)
1911 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1912 if (port <= UINT16_MAX) {
1913 union mf_subvalue value;
1915 memset(&value, 0xff, sizeof value);
1916 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1917 xlate_output_action(ctx, u16_to_ofp(port),
1918 or->max_len, false);
1923 xlate_enqueue_action(struct xlate_ctx *ctx,
1924 const struct ofpact_enqueue *enqueue)
1926 ofp_port_t ofp_port = enqueue->port;
1927 uint32_t queue_id = enqueue->queue;
1928 uint32_t flow_priority, priority;
1931 /* Translate queue to priority. */
1932 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
1934 /* Fall back to ordinary output action. */
1935 xlate_output_action(ctx, enqueue->port, 0, false);
1939 /* Check output port. */
1940 if (ofp_port == OFPP_IN_PORT) {
1941 ofp_port = ctx->xin->flow.in_port.ofp_port;
1942 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1946 /* Add datapath actions. */
1947 flow_priority = ctx->xin->flow.skb_priority;
1948 ctx->xin->flow.skb_priority = priority;
1949 compose_output_action(ctx, ofp_port);
1950 ctx->xin->flow.skb_priority = flow_priority;
1952 /* Update NetFlow output port. */
1953 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1954 ctx->xout->nf_output_iface = ofp_port;
1955 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1956 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1961 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1963 uint32_t skb_priority;
1965 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
1966 ctx->xin->flow.skb_priority = skb_priority;
1968 /* Couldn't translate queue to a priority. Nothing to do. A warning
1969 * has already been logged. */
1974 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
1976 const struct xbridge *xbridge = xbridge_;
1987 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1990 port = get_ofp_port(xbridge, ofp_port);
1991 return port ? port->may_enable : false;
1996 xlate_bundle_action(struct xlate_ctx *ctx,
1997 const struct ofpact_bundle *bundle)
2001 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2003 CONST_CAST(struct xbridge *, ctx->xbridge));
2004 if (bundle->dst.field) {
2005 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2008 xlate_output_action(ctx, port, 0, false);
2013 xlate_learn_action(struct xlate_ctx *ctx,
2014 const struct ofpact_learn *learn)
2016 struct ofputil_flow_mod *fm;
2017 struct ofpbuf ofpacts;
2019 ctx->xout->has_learn = true;
2021 learn_mask(learn, &ctx->xout->wc);
2023 if (!ctx->xin->may_learn) {
2027 fm = xmalloc(sizeof *fm);
2028 ofpbuf_init(&ofpacts, 0);
2029 learn_execute(learn, &ctx->xin->flow, fm, &ofpacts);
2031 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
2034 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
2035 * means "infinite". */
2037 reduce_timeout(uint16_t max, uint16_t *timeout)
2039 if (max && (!*timeout || *timeout > max)) {
2045 xlate_fin_timeout(struct xlate_ctx *ctx,
2046 const struct ofpact_fin_timeout *oft)
2048 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2049 struct rule_dpif *rule = ctx->rule;
2051 ovs_mutex_lock(&rule->up.ofproto->expirable_mutex);
2052 if (list_is_empty(&rule->up.expirable)) {
2053 list_insert(&rule->up.ofproto->expirable, &rule->up.expirable);
2055 ovs_mutex_unlock(&rule->up.ofproto->expirable_mutex);
2057 ovs_mutex_lock(&rule->up.timeout_mutex);
2058 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
2059 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
2060 ovs_mutex_unlock(&rule->up.timeout_mutex);
2065 xlate_sample_action(struct xlate_ctx *ctx,
2066 const struct ofpact_sample *os)
2068 union user_action_cookie cookie;
2069 /* Scale the probability from 16-bit to 32-bit while representing
2070 * the same percentage. */
2071 uint32_t probability = (os->probability << 16) | os->probability;
2073 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2074 &ctx->xout->odp_actions, &ctx->xout->wc);
2076 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2077 os->obs_domain_id, os->obs_point_id, &cookie);
2078 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2079 probability, &cookie, sizeof cookie.flow_sample);
2083 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2085 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2086 ? OFPUTIL_PC_NO_RECV_STP
2087 : OFPUTIL_PC_NO_RECV)) {
2091 /* Only drop packets here if both forwarding and learning are
2092 * disabled. If just learning is enabled, we need to have
2093 * OFPP_NORMAL and the learning action have a look at the packet
2094 * before we can drop it. */
2095 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2103 tunnel_ecn_ok(struct xlate_ctx *ctx)
2105 if (is_ip_any(&ctx->base_flow)
2106 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
2107 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
2108 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
2109 " but is not ECN capable");
2112 /* Set the ECN CE value in the tunneled packet. */
2113 ctx->xin->flow.nw_tos |= IP_ECN_CE;
2121 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2122 struct xlate_ctx *ctx)
2124 struct flow_wildcards *wc = &ctx->xout->wc;
2125 struct flow *flow = &ctx->xin->flow;
2126 bool was_evictable = true;
2127 const struct ofpact *a;
2130 /* Don't let the rule we're working on get evicted underneath us. */
2131 was_evictable = ctx->rule->up.evictable;
2132 ctx->rule->up.evictable = false;
2135 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2136 struct ofpact_controller *controller;
2137 const struct ofpact_metadata *metadata;
2145 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2146 ofpact_get_OUTPUT(a)->max_len, true);
2149 case OFPACT_CONTROLLER:
2150 controller = ofpact_get_CONTROLLER(a);
2151 execute_controller_action(ctx, controller->max_len,
2153 controller->controller_id);
2156 case OFPACT_ENQUEUE:
2157 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2160 case OFPACT_SET_VLAN_VID:
2161 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2162 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2163 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2167 case OFPACT_SET_VLAN_PCP:
2168 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2169 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2171 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2175 case OFPACT_STRIP_VLAN:
2176 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2177 flow->vlan_tci = htons(0);
2180 case OFPACT_PUSH_VLAN:
2181 /* XXX 802.1AD(QinQ) */
2182 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2183 flow->vlan_tci = htons(VLAN_CFI);
2186 case OFPACT_SET_ETH_SRC:
2187 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2188 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2191 case OFPACT_SET_ETH_DST:
2192 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2193 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2196 case OFPACT_SET_IPV4_SRC:
2197 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2198 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2199 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2203 case OFPACT_SET_IPV4_DST:
2204 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2205 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2206 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2210 case OFPACT_SET_IPV4_DSCP:
2211 wc->masks.nw_tos |= IP_DSCP_MASK;
2212 /* OpenFlow 1.0 only supports IPv4. */
2213 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2214 flow->nw_tos &= ~IP_DSCP_MASK;
2215 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2219 case OFPACT_SET_L4_SRC_PORT:
2220 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2221 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2222 if (is_ip_any(flow)) {
2223 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2227 case OFPACT_SET_L4_DST_PORT:
2228 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2229 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2230 if (is_ip_any(flow)) {
2231 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2235 case OFPACT_RESUBMIT:
2236 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2239 case OFPACT_SET_TUNNEL:
2240 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2243 case OFPACT_SET_QUEUE:
2244 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2247 case OFPACT_POP_QUEUE:
2248 flow->skb_priority = ctx->orig_skb_priority;
2251 case OFPACT_REG_MOVE:
2252 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2255 case OFPACT_REG_LOAD:
2256 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2259 case OFPACT_STACK_PUSH:
2260 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2264 case OFPACT_STACK_POP:
2265 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2269 case OFPACT_PUSH_MPLS:
2270 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2273 case OFPACT_POP_MPLS:
2274 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2277 case OFPACT_SET_MPLS_TTL:
2278 if (compose_set_mpls_ttl_action(ctx,
2279 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2284 case OFPACT_DEC_MPLS_TTL:
2285 if (compose_dec_mpls_ttl_action(ctx)) {
2290 case OFPACT_DEC_TTL:
2291 wc->masks.nw_ttl = 0xff;
2292 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2298 /* Nothing to do. */
2301 case OFPACT_MULTIPATH:
2302 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2306 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2309 case OFPACT_OUTPUT_REG:
2310 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2314 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2321 case OFPACT_FIN_TIMEOUT:
2322 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2323 ctx->xout->has_fin_timeout = true;
2324 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2327 case OFPACT_CLEAR_ACTIONS:
2329 * Nothing to do because writa-actions is not supported for now.
2330 * When writa-actions is supported, clear-actions also must
2331 * be supported at the same time.
2335 case OFPACT_WRITE_METADATA:
2336 metadata = ofpact_get_WRITE_METADATA(a);
2337 flow->metadata &= ~metadata->mask;
2338 flow->metadata |= metadata->metadata & metadata->mask;
2342 /* Not implemented yet. */
2345 case OFPACT_GOTO_TABLE: {
2346 /* It is assumed that goto-table is the last action. */
2347 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2349 ovs_assert(ctx->table_id < ogt->table_id);
2350 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2351 ogt->table_id, true);
2356 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2363 ctx->rule->up.evictable = was_evictable;
2368 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2369 const struct flow *flow, struct rule_dpif *rule,
2370 uint8_t tcp_flags, const struct ofpbuf *packet)
2372 xin->ofproto = ofproto;
2374 xin->packet = packet;
2375 xin->may_learn = packet != NULL;
2377 xin->ofpacts = NULL;
2378 xin->ofpacts_len = 0;
2379 xin->tcp_flags = tcp_flags;
2380 xin->resubmit_hook = NULL;
2381 xin->report_hook = NULL;
2382 xin->resubmit_stats = NULL;
2386 xlate_out_uninit(struct xlate_out *xout)
2389 ofpbuf_uninit(&xout->odp_actions);
2393 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2394 * into datapath actions, using 'ctx', and discards the datapath actions. */
2396 xlate_actions_for_side_effects(struct xlate_in *xin)
2398 struct xlate_out xout;
2400 xlate_actions(xin, &xout);
2401 xlate_out_uninit(&xout);
2405 xlate_report(struct xlate_ctx *ctx, const char *s)
2407 if (ctx->xin->report_hook) {
2408 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2413 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2416 dst->slow = src->slow;
2417 dst->has_learn = src->has_learn;
2418 dst->has_normal = src->has_normal;
2419 dst->has_fin_timeout = src->has_fin_timeout;
2420 dst->nf_output_iface = src->nf_output_iface;
2421 dst->mirrors = src->mirrors;
2423 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2424 sizeof dst->odp_actions_stub);
2425 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2426 src->odp_actions.size);
2429 static struct skb_priority_to_dscp *
2430 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2432 struct skb_priority_to_dscp *pdscp;
2435 hash = hash_int(skb_priority, 0);
2436 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2437 if (pdscp->skb_priority == skb_priority) {
2445 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2448 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2449 *dscp = pdscp ? pdscp->dscp : 0;
2450 return pdscp != NULL;
2454 clear_skb_priorities(struct xport *xport)
2456 struct skb_priority_to_dscp *pdscp, *next;
2458 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2459 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2465 actions_output_to_local_port(const struct xlate_ctx *ctx)
2467 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2468 const struct nlattr *a;
2471 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2472 ctx->xout->odp_actions.size) {
2473 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2474 && nl_attr_get_odp_port(a) == local_odp_port) {
2481 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2482 * into datapath actions in 'odp_actions', using 'ctx'. */
2484 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2486 struct flow_wildcards *wc = &xout->wc;
2487 struct flow *flow = &xin->flow;
2489 enum slow_path_reason special;
2490 const struct ofpact *ofpacts;
2491 struct xport *in_port;
2492 struct flow orig_flow;
2493 struct xlate_ctx ctx;
2496 COVERAGE_INC(xlate_actions);
2498 /* Flow initialization rules:
2499 * - 'base_flow' must match the kernel's view of the packet at the
2500 * time that action processing starts. 'flow' represents any
2501 * transformations we wish to make through actions.
2502 * - By default 'base_flow' and 'flow' are the same since the input
2503 * packet matches the output before any actions are applied.
2504 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2505 * of the received packet as seen by the kernel. If we later output
2506 * to another device without any modifications this will cause us to
2507 * insert a new tag since the original one was stripped off by the
2509 * - Tunnel metadata as received is retained in 'flow'. This allows
2510 * tunnel metadata matching also in later tables.
2511 * Since a kernel action for setting the tunnel metadata will only be
2512 * generated with actual tunnel output, changing the tunnel metadata
2513 * values in 'flow' (such as tun_id) will only have effect with a later
2514 * tunnel output action.
2515 * - Tunnel 'base_flow' is completely cleared since that is what the
2516 * kernel does. If we wish to maintain the original values an action
2517 * needs to be generated. */
2522 ctx.xout->has_learn = false;
2523 ctx.xout->has_normal = false;
2524 ctx.xout->has_fin_timeout = false;
2525 ctx.xout->nf_output_iface = NF_OUT_DROP;
2526 ctx.xout->mirrors = 0;
2527 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2528 sizeof ctx.xout->odp_actions_stub);
2529 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2531 ctx.xbridge = xbridge_lookup(xin->ofproto);
2536 ctx.rule = xin->rule;
2538 ctx.base_flow = *flow;
2539 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2540 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2542 flow_wildcards_init_catchall(wc);
2543 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2544 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2545 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2546 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2548 if (tnl_port_should_receive(&ctx.xin->flow)) {
2549 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
2550 /* skb_mark is currently used only by tunnels but that will likely
2551 * change in the future. */
2552 memset(&wc->masks.skb_mark, 0xff, sizeof wc->masks.skb_mark);
2554 if (ctx.xbridge->has_netflow) {
2555 netflow_mask_wc(flow, wc);
2559 ctx.orig_skb_priority = flow->skb_priority;
2564 ofpacts = xin->ofpacts;
2565 ofpacts_len = xin->ofpacts_len;
2566 } else if (xin->rule) {
2567 ofpacts = xin->rule->up.ofpacts;
2568 ofpacts_len = xin->rule->up.ofpacts_len;
2573 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2575 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2576 /* Do this conditionally because the copy is expensive enough that it
2577 * shows up in profiles. */
2581 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2582 switch (ctx.xbridge->frag) {
2583 case OFPC_FRAG_NORMAL:
2584 /* We must pretend that transport ports are unavailable. */
2585 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2586 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2589 case OFPC_FRAG_DROP:
2592 case OFPC_FRAG_REASM:
2595 case OFPC_FRAG_NX_MATCH:
2596 /* Nothing to do. */
2599 case OFPC_INVALID_TTL_TO_CONTROLLER:
2604 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2605 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2607 ctx.xout->slow = special;
2609 size_t sample_actions_len;
2611 if (flow->in_port.ofp_port
2612 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2613 flow->in_port.ofp_port,
2615 ctx.base_flow.vlan_tci = 0;
2618 add_sflow_action(&ctx);
2619 add_ipfix_action(&ctx);
2620 sample_actions_len = ctx.xout->odp_actions.size;
2622 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
2623 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2625 /* We've let OFPP_NORMAL and the learning action look at the
2626 * packet, so drop it now if forwarding is disabled. */
2627 if (in_port && !xport_stp_forward_state(in_port)) {
2628 ctx.xout->odp_actions.size = sample_actions_len;
2632 if (ctx.xbridge->has_in_band
2633 && in_band_must_output_to_local_port(flow)
2634 && !actions_output_to_local_port(&ctx)) {
2635 compose_output_action(&ctx, OFPP_LOCAL);
2638 fix_sflow_action(&ctx);
2640 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2641 add_mirror_actions(&ctx, &orig_flow);
2645 ofpbuf_uninit(&ctx.stack);
2647 /* Clear the metadata and register wildcard masks, because we won't
2648 * use non-header fields as part of the cache. */
2649 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2650 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);