1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
17 #include "ofproto/ofproto-dpif-xlate.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
41 #include "odp-execute.h"
42 #include "ofp-actions.h"
43 #include "ofproto/ofproto-dpif-ipfix.h"
44 #include "ofproto/ofproto-dpif-mirror.h"
45 #include "ofproto/ofproto-dpif-sflow.h"
46 #include "ofproto/ofproto-dpif.h"
47 #include "ofproto/ofproto-provider.h"
51 COVERAGE_DEFINE(xlate_actions);
53 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
55 /* Maximum depth of flow table recursion (due to resubmit actions) in a
56 * flow translation. */
57 #define MAX_RESUBMIT_RECURSION 64
59 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
62 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
63 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
65 struct list xbundles; /* Owned xbundles. */
66 struct hmap xports; /* Indexed by ofp_port. */
68 char *name; /* Name used in log messages. */
69 struct dpif *dpif; /* Datapath interface. */
70 struct mac_learning *ml; /* Mac learning handle. */
71 struct mbridge *mbridge; /* Mirroring. */
72 struct dpif_sflow *sflow; /* SFlow handle, or null. */
73 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
74 struct stp *stp; /* STP or null if disabled. */
76 /* Special rules installed by ofproto-dpif. */
77 struct rule_dpif *miss_rule;
78 struct rule_dpif *no_packet_in_rule;
80 enum ofp_config_flags frag; /* Fragmentation handling. */
81 bool has_netflow; /* Bridge runs netflow? */
82 bool has_in_band; /* Bridge has in band control? */
83 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
87 struct hmap_node hmap_node; /* In global 'xbundles' map. */
88 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
90 struct list list_node; /* In parent 'xbridges' list. */
91 struct xbridge *xbridge; /* Parent xbridge. */
93 struct list xports; /* Contains "struct xport"s. */
95 char *name; /* Name used in log messages. */
96 struct bond *bond; /* Nonnull iff more than one port. */
97 struct lacp *lacp; /* LACP handle or null. */
99 enum port_vlan_mode vlan_mode; /* VLAN mode. */
100 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
101 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
102 * NULL if all VLANs are trunked. */
103 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
104 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
108 struct hmap_node hmap_node; /* Node in global 'xports' map. */
109 struct ofport_dpif *ofport; /* Key in global 'xports map. */
111 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
112 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
114 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
116 struct list bundle_node; /* In parent xbundle (if it exists). */
117 struct xbundle *xbundle; /* Parent xbundle or null. */
119 struct netdev *netdev; /* 'ofport''s netdev. */
121 struct xbridge *xbridge; /* Parent bridge. */
122 struct xport *peer; /* Patch port peer or null. */
124 enum ofputil_port_config config; /* OpenFlow port configuration. */
125 int stp_port_no; /* STP port number or -1 if not in use. */
127 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
129 bool may_enable; /* May be enabled in bonds. */
130 bool is_tunnel; /* Is a tunnel port. */
132 struct cfm *cfm; /* CFM handle or null. */
133 struct bfd *bfd; /* BFD handle or null. */
137 struct xlate_in *xin;
138 struct xlate_out *xout;
140 const struct xbridge *xbridge;
142 /* Flow at the last commit. */
143 struct flow base_flow;
145 /* Tunnel IP destination address as received. This is stored separately
146 * as the base_flow.tunnel is cleared on init to reflect the datapath
147 * behavior. Used to make sure not to send tunneled output to ourselves,
148 * which might lead to an infinite loop. This could happen easily
149 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
150 * actually set the tun_dst field. */
151 ovs_be32 orig_tunnel_ip_dst;
153 /* Stack for the push and pop actions. Each stack element is of type
154 * "union mf_subvalue". */
155 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
158 /* The rule that we are currently translating, or NULL. */
159 struct rule_dpif *rule;
161 int mpls_depth_delta; /* Delta of the mpls stack depth since
162 * actions were last committed.
163 * Must be between -1 and 1 inclusive. */
164 ovs_be32 pre_push_mpls_lse; /* Used to record the top-most MPLS LSE
165 * prior to an mpls_push so that it may be
166 * used for a subsequent mpls_pop. */
168 int recurse; /* Recursion level, via xlate_table_action. */
169 uint32_t orig_skb_priority; /* Priority when packet arrived. */
170 uint8_t table_id; /* OpenFlow table ID where flow was found. */
171 uint32_t sflow_n_outputs; /* Number of output ports. */
172 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
173 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
174 bool exit; /* No further actions should be processed. */
177 /* A controller may use OFPP_NONE as the ingress port to indicate that
178 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
179 * when an input bundle is needed for validation (e.g., mirroring or
180 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
181 * any 'port' structs, so care must be taken when dealing with it.
182 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
183 static struct xbundle ofpp_none_bundle;
185 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
186 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
187 * traffic egressing the 'ofport' with that priority should be marked with. */
188 struct skb_priority_to_dscp {
189 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
190 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
192 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
195 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
196 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
197 static struct hmap xports = HMAP_INITIALIZER(&xports);
199 static bool may_receive(const struct xport *, struct xlate_ctx *);
200 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
202 static void xlate_normal(struct xlate_ctx *);
203 static void xlate_report(struct xlate_ctx *, const char *);
204 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
205 uint8_t table_id, bool may_packet_in);
206 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
207 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
208 static void output_normal(struct xlate_ctx *, const struct xbundle *,
210 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
212 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
213 static struct xbundle *xbundle_lookup(const struct ofbundle *);
214 static struct xport *xport_lookup(const struct ofport_dpif *);
215 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
216 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
217 uint32_t skb_priority);
218 static void clear_skb_priorities(struct xport *);
219 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
223 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
224 struct dpif *dpif, struct rule_dpif *miss_rule,
225 struct rule_dpif *no_packet_in_rule,
226 const struct mac_learning *ml, struct stp *stp,
227 const struct mbridge *mbridge,
228 const struct dpif_sflow *sflow,
229 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
230 bool forward_bpdu, bool has_in_band, bool has_netflow)
232 struct xbridge *xbridge = xbridge_lookup(ofproto);
235 xbridge = xzalloc(sizeof *xbridge);
236 xbridge->ofproto = ofproto;
238 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
239 hmap_init(&xbridge->xports);
240 list_init(&xbridge->xbundles);
243 if (xbridge->ml != ml) {
244 mac_learning_unref(xbridge->ml);
245 xbridge->ml = mac_learning_ref(ml);
248 if (xbridge->mbridge != mbridge) {
249 mbridge_unref(xbridge->mbridge);
250 xbridge->mbridge = mbridge_ref(mbridge);
253 if (xbridge->sflow != sflow) {
254 dpif_sflow_unref(xbridge->sflow);
255 xbridge->sflow = dpif_sflow_ref(sflow);
258 if (xbridge->ipfix != ipfix) {
259 dpif_ipfix_unref(xbridge->ipfix);
260 xbridge->ipfix = dpif_ipfix_ref(ipfix);
263 if (xbridge->stp != stp) {
264 stp_unref(xbridge->stp);
265 xbridge->stp = stp_ref(stp);
269 xbridge->name = xstrdup(name);
271 xbridge->dpif = dpif;
272 xbridge->forward_bpdu = forward_bpdu;
273 xbridge->has_in_band = has_in_band;
274 xbridge->has_netflow = has_netflow;
275 xbridge->frag = frag;
276 xbridge->miss_rule = miss_rule;
277 xbridge->no_packet_in_rule = no_packet_in_rule;
281 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
283 struct xbridge *xbridge = xbridge_lookup(ofproto);
284 struct xbundle *xbundle, *next_xbundle;
285 struct xport *xport, *next_xport;
291 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
292 xlate_ofport_remove(xport->ofport);
295 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
296 xlate_bundle_remove(xbundle->ofbundle);
299 hmap_remove(&xbridges, &xbridge->hmap_node);
300 mac_learning_unref(xbridge->ml);
301 mbridge_unref(xbridge->mbridge);
302 dpif_sflow_unref(xbridge->sflow);
303 dpif_ipfix_unref(xbridge->ipfix);
304 stp_unref(xbridge->stp);
305 hmap_destroy(&xbridge->xports);
311 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
312 const char *name, enum port_vlan_mode vlan_mode, int vlan,
313 unsigned long *trunks, bool use_priority_tags,
314 const struct bond *bond, const struct lacp *lacp,
317 struct xbundle *xbundle = xbundle_lookup(ofbundle);
320 xbundle = xzalloc(sizeof *xbundle);
321 xbundle->ofbundle = ofbundle;
322 xbundle->xbridge = xbridge_lookup(ofproto);
324 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
325 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
326 list_init(&xbundle->xports);
329 ovs_assert(xbundle->xbridge);
332 xbundle->name = xstrdup(name);
334 xbundle->vlan_mode = vlan_mode;
335 xbundle->vlan = vlan;
336 xbundle->trunks = trunks;
337 xbundle->use_priority_tags = use_priority_tags;
338 xbundle->floodable = floodable;
340 if (xbundle->bond != bond) {
341 bond_unref(xbundle->bond);
342 xbundle->bond = bond_ref(bond);
345 if (xbundle->lacp != lacp) {
346 lacp_unref(xbundle->lacp);
347 xbundle->lacp = lacp_ref(lacp);
352 xlate_bundle_remove(struct ofbundle *ofbundle)
354 struct xbundle *xbundle = xbundle_lookup(ofbundle);
355 struct xport *xport, *next;
361 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
362 list_remove(&xport->bundle_node);
363 xport->xbundle = NULL;
366 hmap_remove(&xbundles, &xbundle->hmap_node);
367 list_remove(&xbundle->list_node);
368 bond_unref(xbundle->bond);
369 lacp_unref(xbundle->lacp);
375 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
376 struct ofport_dpif *ofport, ofp_port_t ofp_port,
377 odp_port_t odp_port, const struct netdev *netdev,
378 const struct cfm *cfm, const struct bfd *bfd,
379 struct ofport_dpif *peer, int stp_port_no,
380 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
381 enum ofputil_port_config config, bool is_tunnel,
384 struct xport *xport = xport_lookup(ofport);
388 xport = xzalloc(sizeof *xport);
389 xport->ofport = ofport;
390 xport->xbridge = xbridge_lookup(ofproto);
391 xport->ofp_port = ofp_port;
393 hmap_init(&xport->skb_priorities);
394 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
395 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
396 hash_ofp_port(xport->ofp_port));
399 ovs_assert(xport->ofp_port == ofp_port);
401 xport->config = config;
402 xport->stp_port_no = stp_port_no;
403 xport->is_tunnel = is_tunnel;
404 xport->may_enable = may_enable;
405 xport->odp_port = odp_port;
407 if (xport->netdev != netdev) {
408 netdev_close(xport->netdev);
409 xport->netdev = netdev_ref(netdev);
412 if (xport->cfm != cfm) {
413 cfm_unref(xport->cfm);
414 xport->cfm = cfm_ref(cfm);
417 if (xport->bfd != bfd) {
418 bfd_unref(xport->bfd);
419 xport->bfd = bfd_ref(bfd);
423 xport->peer->peer = NULL;
425 xport->peer = xport_lookup(peer);
427 xport->peer->peer = xport;
430 if (xport->xbundle) {
431 list_remove(&xport->bundle_node);
433 xport->xbundle = xbundle_lookup(ofbundle);
434 if (xport->xbundle) {
435 list_insert(&xport->xbundle->xports, &xport->bundle_node);
438 clear_skb_priorities(xport);
439 for (i = 0; i < n_qdscp; i++) {
440 struct skb_priority_to_dscp *pdscp;
441 uint32_t skb_priority;
443 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
448 pdscp = xmalloc(sizeof *pdscp);
449 pdscp->skb_priority = skb_priority;
450 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
451 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
452 hash_int(pdscp->skb_priority, 0));
457 xlate_ofport_remove(struct ofport_dpif *ofport)
459 struct xport *xport = xport_lookup(ofport);
466 xport->peer->peer = NULL;
470 if (xport->xbundle) {
471 list_remove(&xport->bundle_node);
474 clear_skb_priorities(xport);
475 hmap_destroy(&xport->skb_priorities);
477 hmap_remove(&xports, &xport->hmap_node);
478 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
480 netdev_close(xport->netdev);
481 cfm_unref(xport->cfm);
482 bfd_unref(xport->bfd);
486 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
487 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
488 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
489 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
490 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
491 * 'packet' ingressed.
493 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
494 * 'flow''s in_port to OFPP_NONE.
496 * This function does post-processing on data returned from
497 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
498 * of the upcall processing logic. In particular, if the extracted in_port is
499 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
500 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
501 * a VLAN header onto 'packet' (if it is nonnull).
503 * Similarly, this function also includes some logic to help with tunnels. It
504 * may modify 'flow' as necessary to make the tunneling implementation
505 * transparent to the upcall processing logic.
507 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
508 * or some other positive errno if there are other problems. */
510 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
511 const struct nlattr *key, size_t key_len,
512 struct flow *flow, enum odp_key_fitness *fitnessp,
513 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
515 enum odp_key_fitness fitness;
516 const struct xport *xport;
519 ovs_rwlock_rdlock(&xlate_rwlock);
520 fitness = odp_flow_key_to_flow(key, key_len, flow);
521 if (fitness == ODP_FIT_ERROR) {
527 *odp_in_port = flow->in_port.odp_port;
530 xport = xport_lookup(tnl_port_should_receive(flow)
531 ? tnl_port_receive(flow)
532 : odp_port_to_ofport(backer, flow->in_port.odp_port));
534 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
539 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
541 /* Make the packet resemble the flow, so that it gets sent to
542 * an OpenFlow controller properly, so that it looks correct
543 * for sFlow, and so that flow_extract() will get the correct
544 * vlan_tci if it is called on 'packet'.
546 * The allocated space inside 'packet' probably also contains
547 * 'key', that is, both 'packet' and 'key' are probably part of
548 * a struct dpif_upcall (see the large comment on that
549 * structure definition), so pushing data on 'packet' is in
550 * general not a good idea since it could overwrite 'key' or
551 * free it as a side effect. However, it's OK in this special
552 * case because we know that 'packet' is inside a Netlink
553 * attribute: pushing 4 bytes will just overwrite the 4-byte
554 * "struct nlattr", which is fine since we don't need that
556 eth_push_vlan(packet, flow->vlan_tci);
558 /* We can't reproduce 'key' from 'flow'. */
559 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
564 *ofproto = xport->xbridge->ofproto;
571 ovs_rwlock_unlock(&xlate_rwlock);
575 static struct xbridge *
576 xbridge_lookup(const struct ofproto_dpif *ofproto)
578 struct xbridge *xbridge;
584 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
586 if (xbridge->ofproto == ofproto) {
593 static struct xbundle *
594 xbundle_lookup(const struct ofbundle *ofbundle)
596 struct xbundle *xbundle;
602 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
604 if (xbundle->ofbundle == ofbundle) {
611 static struct xport *
612 xport_lookup(const struct ofport_dpif *ofport)
620 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
622 if (xport->ofport == ofport) {
629 static struct stp_port *
630 xport_get_stp_port(const struct xport *xport)
632 return xport->xbridge->stp && xport->stp_port_no != -1
633 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
637 static enum stp_state
638 xport_stp_learn_state(const struct xport *xport)
640 struct stp_port *sp = xport_get_stp_port(xport);
641 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
645 xport_stp_forward_state(const struct xport *xport)
647 struct stp_port *sp = xport_get_stp_port(xport);
648 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
651 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
652 * were used to make the determination.*/
654 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
656 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
657 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
661 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
663 struct stp_port *sp = xport_get_stp_port(xport);
664 struct ofpbuf payload = *packet;
665 struct eth_header *eth = payload.data;
667 /* Sink packets on ports that have STP disabled when the bridge has
669 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
673 /* Trim off padding on payload. */
674 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
675 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
678 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
679 stp_received_bpdu(sp, payload.data, payload.size);
683 static struct xport *
684 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
688 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
690 if (xport->ofp_port == ofp_port) {
698 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
700 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
701 return xport ? xport->odp_port : ODPP_NONE;
705 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
707 return (bundle->vlan_mode != PORT_VLAN_ACCESS
708 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
712 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
714 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
718 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
720 return xbundle != &ofpp_none_bundle
721 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
726 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
728 return xbundle != &ofpp_none_bundle
729 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
734 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
736 return xbundle != &ofpp_none_bundle
737 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
741 static struct xbundle *
742 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
743 bool warn, struct xport **in_xportp)
747 /* Find the port and bundle for the received packet. */
748 xport = get_ofp_port(xbridge, in_port);
752 if (xport && xport->xbundle) {
753 return xport->xbundle;
756 /* Special-case OFPP_NONE, which a controller may use as the ingress
757 * port for traffic that it is sourcing. */
758 if (in_port == OFPP_NONE) {
759 ofpp_none_bundle.name = "OFPP_NONE";
760 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
761 return &ofpp_none_bundle;
764 /* Odd. A few possible reasons here:
766 * - We deleted a port but there are still a few packets queued up
769 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
770 * we don't know about.
772 * - The ofproto client didn't configure the port as part of a bundle.
773 * This is particularly likely to happen if a packet was received on the
774 * port after it was created, but before the client had a chance to
775 * configure its bundle.
778 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
780 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
781 "port %"PRIu16, xbridge->name, in_port);
787 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
789 const struct xbridge *xbridge = ctx->xbridge;
790 mirror_mask_t mirrors;
791 struct xbundle *in_xbundle;
795 mirrors = ctx->xout->mirrors;
796 ctx->xout->mirrors = 0;
798 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
799 ctx->xin->packet != NULL, NULL);
803 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
805 /* Drop frames on bundles reserved for mirroring. */
806 if (xbundle_mirror_out(xbridge, in_xbundle)) {
807 if (ctx->xin->packet != NULL) {
808 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
809 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
810 "%s, which is reserved exclusively for mirroring",
811 ctx->xbridge->name, in_xbundle->name);
813 ofpbuf_clear(&ctx->xout->odp_actions);
818 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
819 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
822 vlan = input_vid_to_vlan(in_xbundle, vid);
828 /* Restore the original packet before adding the mirror actions. */
829 ctx->xin->flow = *orig_flow;
832 mirror_mask_t dup_mirrors;
833 struct ofbundle *out;
834 unsigned long *vlans;
839 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
840 &vlans, &dup_mirrors, &out, &out_vlan);
841 ovs_assert(has_mirror);
844 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
846 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
849 if (!vlan_mirrored) {
850 mirrors = zero_rightmost_1bit(mirrors);
854 mirrors &= ~dup_mirrors;
855 ctx->xout->mirrors |= dup_mirrors;
857 struct xbundle *out_xbundle = xbundle_lookup(out);
859 output_normal(ctx, out_xbundle, vlan);
861 } else if (vlan != out_vlan
862 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
863 struct xbundle *xbundle;
865 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
866 if (xbundle_includes_vlan(xbundle, out_vlan)
867 && !xbundle_mirror_out(xbridge, xbundle)) {
868 output_normal(ctx, xbundle, out_vlan);
875 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
876 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
877 * the bundle on which the packet was received, returns the VLAN to which the
880 * Both 'vid' and the return value are in the range 0...4095. */
882 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
884 switch (in_xbundle->vlan_mode) {
885 case PORT_VLAN_ACCESS:
886 return in_xbundle->vlan;
889 case PORT_VLAN_TRUNK:
892 case PORT_VLAN_NATIVE_UNTAGGED:
893 case PORT_VLAN_NATIVE_TAGGED:
894 return vid ? vid : in_xbundle->vlan;
901 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
902 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
905 * 'vid' should be the VID obtained from the 802.1Q header that was received as
906 * part of a packet (specify 0 if there was no 802.1Q header), in the range
909 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
911 /* Allow any VID on the OFPP_NONE port. */
912 if (in_xbundle == &ofpp_none_bundle) {
916 switch (in_xbundle->vlan_mode) {
917 case PORT_VLAN_ACCESS:
920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
921 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
922 "packet received on port %s configured as VLAN "
923 "%"PRIu16" access port", vid, in_xbundle->name,
930 case PORT_VLAN_NATIVE_UNTAGGED:
931 case PORT_VLAN_NATIVE_TAGGED:
933 /* Port must always carry its native VLAN. */
937 case PORT_VLAN_TRUNK:
938 if (!xbundle_includes_vlan(in_xbundle, vid)) {
940 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
941 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
942 "received on port %s not configured for trunking "
943 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
955 /* Given 'vlan', the VLAN that a packet belongs to, and
956 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
957 * that should be included in the 802.1Q header. (If the return value is 0,
958 * then the 802.1Q header should only be included in the packet if there is a
961 * Both 'vlan' and the return value are in the range 0...4095. */
963 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
965 switch (out_xbundle->vlan_mode) {
966 case PORT_VLAN_ACCESS:
969 case PORT_VLAN_TRUNK:
970 case PORT_VLAN_NATIVE_TAGGED:
973 case PORT_VLAN_NATIVE_UNTAGGED:
974 return vlan == out_xbundle->vlan ? 0 : vlan;
982 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
985 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
987 ovs_be16 tci, old_tci;
990 vid = output_vlan_to_vid(out_xbundle, vlan);
991 if (list_is_empty(&out_xbundle->xports)) {
992 /* Partially configured bundle with no slaves. Drop the packet. */
994 } else if (!out_xbundle->bond) {
995 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
998 struct ofport_dpif *ofport;
1000 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1001 &ctx->xout->wc, vid);
1002 xport = xport_lookup(ofport);
1005 /* No slaves enabled, so drop packet. */
1010 old_tci = *flow_tci;
1012 if (tci || out_xbundle->use_priority_tags) {
1013 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1015 tci |= htons(VLAN_CFI);
1020 compose_output_action(ctx, xport->ofp_port);
1021 *flow_tci = old_tci;
1024 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1025 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1026 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1028 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1030 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1034 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1035 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1039 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1040 if (flow->nw_proto == ARP_OP_REPLY) {
1042 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1043 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1044 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1046 return flow->nw_src == flow->nw_dst;
1052 /* Checks whether a MAC learning update is necessary for MAC learning table
1053 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1056 * Most packets processed through the MAC learning table do not actually
1057 * change it in any way. This function requires only a read lock on the MAC
1058 * learning table, so it is much cheaper in this common case.
1060 * Keep the code here synchronized with that in update_learning_table__()
1063 is_mac_learning_update_needed(const struct mac_learning *ml,
1064 const struct flow *flow,
1065 struct flow_wildcards *wc,
1066 int vlan, struct xbundle *in_xbundle)
1067 OVS_REQ_RDLOCK(ml->rwlock)
1069 struct mac_entry *mac;
1071 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1075 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1076 if (!mac || mac_entry_age(ml, mac)) {
1080 if (is_gratuitous_arp(flow, wc)) {
1081 /* We don't want to learn from gratuitous ARP packets that are
1082 * reflected back over bond slaves so we lock the learning table. */
1083 if (!in_xbundle->bond) {
1085 } else if (mac_entry_is_grat_arp_locked(mac)) {
1090 return mac->port.p != in_xbundle->ofbundle;
1094 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1095 * received on 'in_xbundle' in 'vlan'.
1097 * This code repeats all the checks in is_mac_learning_update_needed() because
1098 * the lock was released between there and here and thus the MAC learning state
1099 * could have changed.
1101 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1104 update_learning_table__(const struct xbridge *xbridge,
1105 const struct flow *flow, struct flow_wildcards *wc,
1106 int vlan, struct xbundle *in_xbundle)
1107 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1109 struct mac_entry *mac;
1111 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1115 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1116 if (is_gratuitous_arp(flow, wc)) {
1117 /* We don't want to learn from gratuitous ARP packets that are
1118 * reflected back over bond slaves so we lock the learning table. */
1119 if (!in_xbundle->bond) {
1120 mac_entry_set_grat_arp_lock(mac);
1121 } else if (mac_entry_is_grat_arp_locked(mac)) {
1126 if (mac->port.p != in_xbundle->ofbundle) {
1127 /* The log messages here could actually be useful in debugging,
1128 * so keep the rate limit relatively high. */
1129 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1131 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1132 "on port %s in VLAN %d",
1133 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1134 in_xbundle->name, vlan);
1136 mac->port.p = in_xbundle->ofbundle;
1137 mac_learning_changed(xbridge->ml);
1142 update_learning_table(const struct xbridge *xbridge,
1143 const struct flow *flow, struct flow_wildcards *wc,
1144 int vlan, struct xbundle *in_xbundle)
1148 /* Don't learn the OFPP_NONE port. */
1149 if (in_xbundle == &ofpp_none_bundle) {
1153 /* First try the common case: no change to MAC learning table. */
1154 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1155 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1157 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1160 /* Slow path: MAC learning table might need an update. */
1161 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1162 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1163 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1167 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1168 * dropped. Returns true if they may be forwarded, false if they should be
1171 * 'in_port' must be the xport that corresponds to flow->in_port.
1172 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1174 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1175 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1176 * checked by input_vid_is_valid().
1178 * May also add tags to '*tags', although the current implementation only does
1179 * so in one special case.
1182 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1185 struct xbundle *in_xbundle = in_port->xbundle;
1186 const struct xbridge *xbridge = ctx->xbridge;
1187 struct flow *flow = &ctx->xin->flow;
1189 /* Drop frames for reserved multicast addresses
1190 * only if forward_bpdu option is absent. */
1191 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1192 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1196 if (in_xbundle->bond) {
1197 struct mac_entry *mac;
1199 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1205 xlate_report(ctx, "bonding refused admissibility, dropping");
1208 case BV_DROP_IF_MOVED:
1209 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1210 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1211 if (mac && mac->port.p != in_xbundle->ofbundle &&
1212 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1213 || mac_entry_is_grat_arp_locked(mac))) {
1214 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1215 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1219 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1228 xlate_normal(struct xlate_ctx *ctx)
1230 struct flow_wildcards *wc = &ctx->xout->wc;
1231 struct flow *flow = &ctx->xin->flow;
1232 struct xbundle *in_xbundle;
1233 struct xport *in_port;
1234 struct mac_entry *mac;
1239 ctx->xout->has_normal = true;
1241 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1242 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1243 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1245 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1246 ctx->xin->packet != NULL, &in_port);
1248 xlate_report(ctx, "no input bundle, dropping");
1252 /* Drop malformed frames. */
1253 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1254 !(flow->vlan_tci & htons(VLAN_CFI))) {
1255 if (ctx->xin->packet != NULL) {
1256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1257 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1258 "VLAN tag received on port %s",
1259 ctx->xbridge->name, in_xbundle->name);
1261 xlate_report(ctx, "partial VLAN tag, dropping");
1265 /* Drop frames on bundles reserved for mirroring. */
1266 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1267 if (ctx->xin->packet != NULL) {
1268 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1269 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1270 "%s, which is reserved exclusively for mirroring",
1271 ctx->xbridge->name, in_xbundle->name);
1273 xlate_report(ctx, "input port is mirror output port, dropping");
1278 vid = vlan_tci_to_vid(flow->vlan_tci);
1279 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1280 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1283 vlan = input_vid_to_vlan(in_xbundle, vid);
1285 /* Check other admissibility requirements. */
1286 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1290 /* Learn source MAC. */
1291 if (ctx->xin->may_learn) {
1292 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1295 /* Determine output bundle. */
1296 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1297 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1298 mac_port = mac ? mac->port.p : NULL;
1299 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1302 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1303 if (mac_xbundle && mac_xbundle != in_xbundle) {
1304 xlate_report(ctx, "forwarding to learned port");
1305 output_normal(ctx, mac_xbundle, vlan);
1306 } else if (!mac_xbundle) {
1307 xlate_report(ctx, "learned port is unknown, dropping");
1309 xlate_report(ctx, "learned port is input port, dropping");
1312 struct xbundle *xbundle;
1314 xlate_report(ctx, "no learned MAC for destination, flooding");
1315 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1316 if (xbundle != in_xbundle
1317 && xbundle_includes_vlan(xbundle, vlan)
1318 && xbundle->floodable
1319 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1320 output_normal(ctx, xbundle, vlan);
1323 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1327 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1328 * the number of packets out of UINT32_MAX to sample. The given
1329 * cookie is passed back in the callback for each sampled packet.
1332 compose_sample_action(const struct xbridge *xbridge,
1333 struct ofpbuf *odp_actions,
1334 const struct flow *flow,
1335 const uint32_t probability,
1336 const union user_action_cookie *cookie,
1337 const size_t cookie_size)
1339 size_t sample_offset, actions_offset;
1340 odp_port_t odp_port;
1344 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1346 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1348 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1350 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1351 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1352 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1354 nl_msg_end_nested(odp_actions, actions_offset);
1355 nl_msg_end_nested(odp_actions, sample_offset);
1356 return cookie_offset;
1360 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1361 odp_port_t odp_port, unsigned int n_outputs,
1362 union user_action_cookie *cookie)
1366 cookie->type = USER_ACTION_COOKIE_SFLOW;
1367 cookie->sflow.vlan_tci = vlan_tci;
1369 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1370 * port information") for the interpretation of cookie->output. */
1371 switch (n_outputs) {
1373 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1374 cookie->sflow.output = 0x40000000 | 256;
1378 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1380 cookie->sflow.output = ifindex;
1385 /* 0x80000000 means "multiple output ports. */
1386 cookie->sflow.output = 0x80000000 | n_outputs;
1391 /* Compose SAMPLE action for sFlow bridge sampling. */
1393 compose_sflow_action(const struct xbridge *xbridge,
1394 struct ofpbuf *odp_actions,
1395 const struct flow *flow,
1396 odp_port_t odp_port)
1398 uint32_t probability;
1399 union user_action_cookie cookie;
1401 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1405 probability = dpif_sflow_get_probability(xbridge->sflow);
1406 compose_sflow_cookie(xbridge, htons(0), odp_port,
1407 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1409 return compose_sample_action(xbridge, odp_actions, flow, probability,
1410 &cookie, sizeof cookie.sflow);
1414 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1415 uint32_t obs_domain_id, uint32_t obs_point_id,
1416 union user_action_cookie *cookie)
1418 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1419 cookie->flow_sample.probability = probability;
1420 cookie->flow_sample.collector_set_id = collector_set_id;
1421 cookie->flow_sample.obs_domain_id = obs_domain_id;
1422 cookie->flow_sample.obs_point_id = obs_point_id;
1426 compose_ipfix_cookie(union user_action_cookie *cookie)
1428 cookie->type = USER_ACTION_COOKIE_IPFIX;
1431 /* Compose SAMPLE action for IPFIX bridge sampling. */
1433 compose_ipfix_action(const struct xbridge *xbridge,
1434 struct ofpbuf *odp_actions,
1435 const struct flow *flow)
1437 uint32_t probability;
1438 union user_action_cookie cookie;
1440 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1444 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1445 compose_ipfix_cookie(&cookie);
1447 compose_sample_action(xbridge, odp_actions, flow, probability,
1448 &cookie, sizeof cookie.ipfix);
1451 /* SAMPLE action for sFlow must be first action in any given list of
1452 * actions. At this point we do not have all information required to
1453 * build it. So try to build sample action as complete as possible. */
1455 add_sflow_action(struct xlate_ctx *ctx)
1457 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1458 &ctx->xout->odp_actions,
1459 &ctx->xin->flow, ODPP_NONE);
1460 ctx->sflow_odp_port = 0;
1461 ctx->sflow_n_outputs = 0;
1464 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1465 * of actions, eventually after the SAMPLE action for sFlow. */
1467 add_ipfix_action(struct xlate_ctx *ctx)
1469 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1473 /* Fix SAMPLE action according to data collected while composing ODP actions.
1474 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1475 * USERSPACE action's user-cookie which is required for sflow. */
1477 fix_sflow_action(struct xlate_ctx *ctx)
1479 const struct flow *base = &ctx->base_flow;
1480 union user_action_cookie *cookie;
1482 if (!ctx->user_cookie_offset) {
1486 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1487 sizeof cookie->sflow);
1488 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1490 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1491 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1494 static enum slow_path_reason
1495 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1496 const struct xport *xport, const struct ofpbuf *packet)
1498 struct flow_wildcards *wc = &ctx->xout->wc;
1499 const struct xbridge *xbridge = ctx->xbridge;
1503 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1505 cfm_process_heartbeat(xport->cfm, packet);
1508 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1510 bfd_process_packet(xport->bfd, flow, packet);
1513 } else if (xport->xbundle && xport->xbundle->lacp
1514 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1516 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1519 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1521 stp_process_packet(xport, packet);
1530 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1533 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1534 struct flow_wildcards *wc = &ctx->xout->wc;
1535 struct flow *flow = &ctx->xin->flow;
1536 ovs_be16 flow_vlan_tci;
1537 uint32_t flow_pkt_mark;
1538 uint8_t flow_nw_tos;
1539 odp_port_t out_port, odp_port;
1542 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1543 * before traversing a patch port. */
1544 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 21);
1547 xlate_report(ctx, "Nonexistent output port");
1549 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1550 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1552 } else if (check_stp && !xport_stp_forward_state(xport)) {
1553 xlate_report(ctx, "STP not in forwarding state, skipping output");
1557 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1558 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1563 const struct xport *peer = xport->peer;
1564 struct flow old_flow = ctx->xin->flow;
1565 enum slow_path_reason special;
1567 ctx->xbridge = peer->xbridge;
1568 flow->in_port.ofp_port = peer->ofp_port;
1569 flow->metadata = htonll(0);
1570 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1571 memset(flow->regs, 0, sizeof flow->regs);
1573 special = process_special(ctx, &ctx->xin->flow, peer,
1576 ctx->xout->slow = special;
1577 } else if (may_receive(peer, ctx)) {
1578 if (xport_stp_forward_state(peer)) {
1579 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1581 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1582 * learning action look at the packet, then drop it. */
1583 struct flow old_base_flow = ctx->base_flow;
1584 size_t old_size = ctx->xout->odp_actions.size;
1585 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1586 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1587 ctx->xout->mirrors = old_mirrors;
1588 ctx->base_flow = old_base_flow;
1589 ctx->xout->odp_actions.size = old_size;
1593 ctx->xin->flow = old_flow;
1594 ctx->xbridge = xport->xbridge;
1596 if (ctx->xin->resubmit_stats) {
1597 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1598 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1604 flow_vlan_tci = flow->vlan_tci;
1605 flow_pkt_mark = flow->pkt_mark;
1606 flow_nw_tos = flow->nw_tos;
1608 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1609 wc->masks.nw_tos |= IP_ECN_MASK;
1610 flow->nw_tos &= ~IP_DSCP_MASK;
1611 flow->nw_tos |= dscp;
1614 if (xport->is_tunnel) {
1615 /* Save tunnel metadata so that changes made due to
1616 * the Logical (tunnel) Port are not visible for any further
1617 * matches, while explicit set actions on tunnel metadata are.
1619 struct flow_tnl flow_tnl = flow->tunnel;
1620 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1621 if (odp_port == ODPP_NONE) {
1622 xlate_report(ctx, "Tunneling decided against output");
1623 goto out; /* restore flow_nw_tos */
1625 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1626 xlate_report(ctx, "Not tunneling to our own address");
1627 goto out; /* restore flow_nw_tos */
1629 if (ctx->xin->resubmit_stats) {
1630 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1632 out_port = odp_port;
1633 commit_odp_tunnel_action(flow, &ctx->base_flow,
1634 &ctx->xout->odp_actions);
1635 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1637 ofp_port_t vlandev_port;
1639 odp_port = xport->odp_port;
1640 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1641 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1643 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1645 if (vlandev_port == ofp_port) {
1646 out_port = odp_port;
1648 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1649 flow->vlan_tci = htons(0);
1653 if (out_port != ODPP_NONE) {
1654 commit_odp_actions(flow, &ctx->base_flow,
1655 &ctx->xout->odp_actions, &ctx->xout->wc,
1656 &ctx->mpls_depth_delta);
1657 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1660 ctx->sflow_odp_port = odp_port;
1661 ctx->sflow_n_outputs++;
1662 ctx->xout->nf_output_iface = ofp_port;
1667 flow->vlan_tci = flow_vlan_tci;
1668 flow->pkt_mark = flow_pkt_mark;
1669 flow->nw_tos = flow_nw_tos;
1673 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1675 compose_output_action__(ctx, ofp_port, true);
1679 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1681 struct rule_dpif *old_rule = ctx->rule;
1682 struct rule_actions *actions;
1684 if (ctx->xin->resubmit_stats) {
1685 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1690 actions = rule_dpif_get_actions(rule);
1691 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1692 rule_actions_unref(actions);
1693 ctx->rule = old_rule;
1698 xlate_table_action(struct xlate_ctx *ctx,
1699 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1701 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1702 struct rule_dpif *rule;
1703 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1704 uint8_t old_table_id = ctx->table_id;
1706 ctx->table_id = table_id;
1708 /* Look up a flow with 'in_port' as the input port. Then restore the
1709 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1710 * have surprising behavior). */
1711 ctx->xin->flow.in_port.ofp_port = in_port;
1712 rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1713 &ctx->xin->flow, &ctx->xout->wc,
1715 ctx->xin->flow.in_port.ofp_port = old_in_port;
1717 if (ctx->xin->resubmit_hook) {
1718 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1721 if (!rule && may_packet_in) {
1722 struct xport *xport;
1725 * check if table configuration flags
1726 * OFPTC_TABLE_MISS_CONTROLLER, default.
1727 * OFPTC_TABLE_MISS_CONTINUE,
1728 * OFPTC_TABLE_MISS_DROP
1729 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1730 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1731 choose_miss_rule(xport ? xport->config : 0,
1732 ctx->xbridge->miss_rule,
1733 ctx->xbridge->no_packet_in_rule, &rule);
1736 xlate_recursively(ctx, rule);
1737 rule_dpif_unref(rule);
1740 ctx->table_id = old_table_id;
1742 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1744 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1745 MAX_RESUBMIT_RECURSION);
1750 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1751 const struct ofpact_resubmit *resubmit)
1756 in_port = resubmit->in_port;
1757 if (in_port == OFPP_IN_PORT) {
1758 in_port = ctx->xin->flow.in_port.ofp_port;
1761 table_id = resubmit->table_id;
1762 if (table_id == 255) {
1763 table_id = ctx->table_id;
1766 xlate_table_action(ctx, in_port, table_id, false);
1770 flood_packets(struct xlate_ctx *ctx, bool all)
1772 const struct xport *xport;
1774 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1775 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1780 compose_output_action__(ctx, xport->ofp_port, false);
1781 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1782 compose_output_action(ctx, xport->ofp_port);
1786 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1790 execute_controller_action(struct xlate_ctx *ctx, int len,
1791 enum ofp_packet_in_reason reason,
1792 uint16_t controller_id)
1794 struct ofputil_packet_in *pin;
1795 struct ofpbuf *packet;
1798 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1799 ctx->xout->slow = SLOW_CONTROLLER;
1800 if (!ctx->xin->packet) {
1804 packet = ofpbuf_clone(ctx->xin->packet);
1806 key.skb_priority = 0;
1808 memset(&key.tunnel, 0, sizeof key.tunnel);
1810 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1811 &ctx->xout->odp_actions, &ctx->xout->wc,
1812 &ctx->mpls_depth_delta);
1814 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1815 ctx->xout->odp_actions.size, NULL, NULL);
1817 pin = xmalloc(sizeof *pin);
1818 pin->packet_len = packet->size;
1819 pin->packet = ofpbuf_steal_data(packet);
1820 pin->reason = reason;
1821 pin->controller_id = controller_id;
1822 pin->table_id = ctx->table_id;
1823 pin->cookie = ctx->rule ? rule_dpif_get_flow_cookie(ctx->rule) : 0;
1825 pin->send_len = len;
1826 flow_get_metadata(&ctx->xin->flow, &pin->fmd);
1828 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
1829 ofpbuf_delete(packet);
1833 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1835 struct flow_wildcards *wc = &ctx->xout->wc;
1836 struct flow *flow = &ctx->xin->flow;
1838 ovs_assert(eth_type_mpls(eth_type));
1840 /* If mpls_depth_delta is negative then an MPLS POP action has been
1841 * composed and the resulting MPLS label stack is unknown. This means
1842 * an MPLS PUSH action can't be composed as it needs to know either the
1843 * top-most MPLS LSE to use as a template for the new MPLS LSE, or that
1844 * there is no MPLS label stack present. Thus, stop processing.
1846 * If mpls_depth_delta is positive then an MPLS PUSH action has been
1847 * composed and no further MPLS PUSH action may be performed without
1848 * losing MPLS LSE and ether type information held in xtx->xin->flow.
1849 * Thus, stop processing.
1851 * If the MPLS LSE of the flow and base_flow differ then the MPLS LSE
1852 * has been updated. Performing a MPLS PUSH action may be would result in
1853 * losing MPLS LSE and ether type information held in xtx->xin->flow.
1854 * Thus, stop processing.
1856 * It is planned that in the future this case will be handled
1857 * by recirculation */
1858 if (ctx->mpls_depth_delta ||
1859 ctx->xin->flow.mpls_lse != ctx->base_flow.mpls_lse) {
1863 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1865 ctx->pre_push_mpls_lse = ctx->xin->flow.mpls_lse;
1867 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
1868 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1873 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1874 label = htonl(0x2); /* IPV6 Explicit Null. */
1876 label = htonl(0x0); /* IPV4 Explicit Null. */
1878 wc->masks.nw_tos |= IP_DSCP_MASK;
1879 wc->masks.nw_ttl = 0xff;
1880 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1881 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1882 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1884 flow->dl_type = eth_type;
1885 ctx->mpls_depth_delta++;
1891 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1893 struct flow_wildcards *wc = &ctx->xout->wc;
1895 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1899 /* If mpls_depth_delta is negative then an MPLS POP action has been
1900 * composed. Performing another MPLS POP action
1901 * would result in losing ether type that results from
1902 * the already composed MPLS POP. Thus, stop processing.
1904 * It is planned that in the future this case will be handled
1905 * by recirculation */
1906 if (ctx->mpls_depth_delta < 0) {
1910 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1912 /* If mpls_depth_delta is positive then an MPLS PUSH action has been
1913 * executed and the previous MPLS LSE saved in ctx->pre_push_mpls_lse. The
1914 * flow's MPLS LSE should be restored to that value to allow any
1915 * subsequent actions that update of the LSE to be executed correctly.
1917 if (ctx->mpls_depth_delta > 0) {
1918 ctx->xin->flow.mpls_lse = ctx->pre_push_mpls_lse;
1921 ctx->xin->flow.dl_type = eth_type;
1922 ctx->mpls_depth_delta--;
1928 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1930 struct flow *flow = &ctx->xin->flow;
1932 if (!is_ip_any(flow)) {
1936 ctx->xout->wc.masks.nw_ttl = 0xff;
1937 if (flow->nw_ttl > 1) {
1943 for (i = 0; i < ids->n_controllers; i++) {
1944 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1948 /* Stop processing for current table. */
1954 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1956 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1960 /* If mpls_depth_delta is negative then an MPLS POP action has been
1961 * executed and the resulting MPLS label stack is unknown. This means
1962 * a SET MPLS TTL push action can't be executed as it needs to manipulate
1963 * the top-most MPLS LSE. Thus, stop processing.
1965 * It is planned that in the future this case will be handled
1968 if (ctx->mpls_depth_delta < 0) {
1972 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
1973 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1978 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1980 struct flow *flow = &ctx->xin->flow;
1981 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1982 struct flow_wildcards *wc = &ctx->xout->wc;
1984 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1986 if (!eth_type_mpls(flow->dl_type)) {
1992 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1995 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1997 /* Stop processing for current table. */
2003 xlate_output_action(struct xlate_ctx *ctx,
2004 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2006 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2008 ctx->xout->nf_output_iface = NF_OUT_DROP;
2012 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2015 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2022 flood_packets(ctx, false);
2025 flood_packets(ctx, true);
2027 case OFPP_CONTROLLER:
2028 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2034 if (port != ctx->xin->flow.in_port.ofp_port) {
2035 compose_output_action(ctx, port);
2037 xlate_report(ctx, "skipping output to input port");
2042 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2043 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2044 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2045 ctx->xout->nf_output_iface = prev_nf_output_iface;
2046 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2047 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2048 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2053 xlate_output_reg_action(struct xlate_ctx *ctx,
2054 const struct ofpact_output_reg *or)
2056 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2057 if (port <= UINT16_MAX) {
2058 union mf_subvalue value;
2060 memset(&value, 0xff, sizeof value);
2061 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2062 xlate_output_action(ctx, u16_to_ofp(port),
2063 or->max_len, false);
2068 xlate_enqueue_action(struct xlate_ctx *ctx,
2069 const struct ofpact_enqueue *enqueue)
2071 ofp_port_t ofp_port = enqueue->port;
2072 uint32_t queue_id = enqueue->queue;
2073 uint32_t flow_priority, priority;
2076 /* Translate queue to priority. */
2077 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2079 /* Fall back to ordinary output action. */
2080 xlate_output_action(ctx, enqueue->port, 0, false);
2084 /* Check output port. */
2085 if (ofp_port == OFPP_IN_PORT) {
2086 ofp_port = ctx->xin->flow.in_port.ofp_port;
2087 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2091 /* Add datapath actions. */
2092 flow_priority = ctx->xin->flow.skb_priority;
2093 ctx->xin->flow.skb_priority = priority;
2094 compose_output_action(ctx, ofp_port);
2095 ctx->xin->flow.skb_priority = flow_priority;
2097 /* Update NetFlow output port. */
2098 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2099 ctx->xout->nf_output_iface = ofp_port;
2100 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2101 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2106 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2108 uint32_t skb_priority;
2110 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2111 ctx->xin->flow.skb_priority = skb_priority;
2113 /* Couldn't translate queue to a priority. Nothing to do. A warning
2114 * has already been logged. */
2119 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2121 const struct xbridge *xbridge = xbridge_;
2132 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2135 port = get_ofp_port(xbridge, ofp_port);
2136 return port ? port->may_enable : false;
2141 xlate_bundle_action(struct xlate_ctx *ctx,
2142 const struct ofpact_bundle *bundle)
2146 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2148 CONST_CAST(struct xbridge *, ctx->xbridge));
2149 if (bundle->dst.field) {
2150 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2153 xlate_output_action(ctx, port, 0, false);
2158 xlate_learn_action(struct xlate_ctx *ctx,
2159 const struct ofpact_learn *learn)
2161 uint64_t ofpacts_stub[1024 / 8];
2162 struct ofputil_flow_mod fm;
2163 struct ofpbuf ofpacts;
2165 ctx->xout->has_learn = true;
2167 learn_mask(learn, &ctx->xout->wc);
2169 if (!ctx->xin->may_learn) {
2173 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2174 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2175 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2176 ofpbuf_uninit(&ofpacts);
2180 xlate_fin_timeout(struct xlate_ctx *ctx,
2181 const struct ofpact_fin_timeout *oft)
2183 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2184 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2185 oft->fin_hard_timeout);
2190 xlate_sample_action(struct xlate_ctx *ctx,
2191 const struct ofpact_sample *os)
2193 union user_action_cookie cookie;
2194 /* Scale the probability from 16-bit to 32-bit while representing
2195 * the same percentage. */
2196 uint32_t probability = (os->probability << 16) | os->probability;
2198 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2199 &ctx->xout->odp_actions, &ctx->xout->wc,
2200 &ctx->mpls_depth_delta);
2202 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2203 os->obs_domain_id, os->obs_point_id, &cookie);
2204 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2205 probability, &cookie, sizeof cookie.flow_sample);
2209 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2211 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2212 ? OFPUTIL_PC_NO_RECV_STP
2213 : OFPUTIL_PC_NO_RECV)) {
2217 /* Only drop packets here if both forwarding and learning are
2218 * disabled. If just learning is enabled, we need to have
2219 * OFPP_NORMAL and the learning action have a look at the packet
2220 * before we can drop it. */
2221 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2229 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2230 struct xlate_ctx *ctx)
2232 struct flow_wildcards *wc = &ctx->xout->wc;
2233 struct flow *flow = &ctx->xin->flow;
2234 const struct ofpact *a;
2236 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2237 struct ofpact_controller *controller;
2238 const struct ofpact_metadata *metadata;
2246 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2247 ofpact_get_OUTPUT(a)->max_len, true);
2251 /* XXX not yet implemented */
2254 case OFPACT_CONTROLLER:
2255 controller = ofpact_get_CONTROLLER(a);
2256 execute_controller_action(ctx, controller->max_len,
2258 controller->controller_id);
2261 case OFPACT_ENQUEUE:
2262 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2265 case OFPACT_SET_VLAN_VID:
2266 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2267 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2268 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2272 case OFPACT_SET_VLAN_PCP:
2273 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2274 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2276 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2280 case OFPACT_STRIP_VLAN:
2281 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2282 flow->vlan_tci = htons(0);
2285 case OFPACT_PUSH_VLAN:
2286 /* XXX 802.1AD(QinQ) */
2287 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2288 flow->vlan_tci = htons(VLAN_CFI);
2291 case OFPACT_SET_ETH_SRC:
2292 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2293 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2296 case OFPACT_SET_ETH_DST:
2297 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2298 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2301 case OFPACT_SET_IPV4_SRC:
2302 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2303 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2304 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2308 case OFPACT_SET_IPV4_DST:
2309 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2310 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2311 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2315 case OFPACT_SET_IPV4_DSCP:
2316 wc->masks.nw_tos |= IP_DSCP_MASK;
2317 /* OpenFlow 1.0 only supports IPv4. */
2318 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2319 flow->nw_tos &= ~IP_DSCP_MASK;
2320 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2324 case OFPACT_SET_L4_SRC_PORT:
2325 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2326 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2327 if (is_ip_any(flow)) {
2328 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2332 case OFPACT_SET_L4_DST_PORT:
2333 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2334 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2335 if (is_ip_any(flow)) {
2336 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2340 case OFPACT_RESUBMIT:
2341 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2344 case OFPACT_SET_TUNNEL:
2345 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2348 case OFPACT_SET_QUEUE:
2349 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2352 case OFPACT_POP_QUEUE:
2353 flow->skb_priority = ctx->orig_skb_priority;
2356 case OFPACT_REG_MOVE:
2357 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2360 case OFPACT_REG_LOAD:
2361 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2364 case OFPACT_STACK_PUSH:
2365 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2369 case OFPACT_STACK_POP:
2370 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2374 case OFPACT_PUSH_MPLS:
2375 if (compose_mpls_push_action(ctx,
2376 ofpact_get_PUSH_MPLS(a)->ethertype)) {
2381 case OFPACT_POP_MPLS:
2382 if (compose_mpls_pop_action(ctx,
2383 ofpact_get_POP_MPLS(a)->ethertype)) {
2388 case OFPACT_SET_MPLS_TTL:
2389 if (compose_set_mpls_ttl_action(ctx,
2390 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2395 case OFPACT_DEC_MPLS_TTL:
2396 if (compose_dec_mpls_ttl_action(ctx)) {
2401 case OFPACT_DEC_TTL:
2402 wc->masks.nw_ttl = 0xff;
2403 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2409 /* Nothing to do. */
2412 case OFPACT_MULTIPATH:
2413 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2417 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2420 case OFPACT_OUTPUT_REG:
2421 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2425 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2432 case OFPACT_FIN_TIMEOUT:
2433 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2434 ctx->xout->has_fin_timeout = true;
2435 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2438 case OFPACT_CLEAR_ACTIONS:
2440 * Nothing to do because writa-actions is not supported for now.
2441 * When writa-actions is supported, clear-actions also must
2442 * be supported at the same time.
2446 case OFPACT_WRITE_METADATA:
2447 metadata = ofpact_get_WRITE_METADATA(a);
2448 flow->metadata &= ~metadata->mask;
2449 flow->metadata |= metadata->metadata & metadata->mask;
2453 /* Not implemented yet. */
2456 case OFPACT_GOTO_TABLE: {
2457 /* It is assumed that goto-table is the last action. */
2458 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2460 ovs_assert(ctx->table_id < ogt->table_id);
2461 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2462 ogt->table_id, true);
2467 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2474 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2475 const struct flow *flow, struct rule_dpif *rule,
2476 uint8_t tcp_flags, const struct ofpbuf *packet)
2478 xin->ofproto = ofproto;
2480 xin->packet = packet;
2481 xin->may_learn = packet != NULL;
2483 xin->ofpacts = NULL;
2484 xin->ofpacts_len = 0;
2485 xin->tcp_flags = tcp_flags;
2486 xin->resubmit_hook = NULL;
2487 xin->report_hook = NULL;
2488 xin->resubmit_stats = NULL;
2492 xlate_out_uninit(struct xlate_out *xout)
2495 ofpbuf_uninit(&xout->odp_actions);
2499 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2500 * into datapath actions, using 'ctx', and discards the datapath actions. */
2502 xlate_actions_for_side_effects(struct xlate_in *xin)
2504 struct xlate_out xout;
2506 xlate_actions(xin, &xout);
2507 xlate_out_uninit(&xout);
2511 xlate_report(struct xlate_ctx *ctx, const char *s)
2513 if (ctx->xin->report_hook) {
2514 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2519 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2522 dst->slow = src->slow;
2523 dst->has_learn = src->has_learn;
2524 dst->has_normal = src->has_normal;
2525 dst->has_fin_timeout = src->has_fin_timeout;
2526 dst->nf_output_iface = src->nf_output_iface;
2527 dst->mirrors = src->mirrors;
2529 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2530 sizeof dst->odp_actions_stub);
2531 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2532 src->odp_actions.size);
2535 /* Returns a reference to the sflow handled associated with ofproto, or NULL if
2536 * there is none. The caller is responsible for decrementing the results ref
2537 * count with dpif_sflow_unref(). */
2539 xlate_get_sflow(const struct ofproto_dpif *ofproto)
2541 struct dpif_sflow *sflow = NULL;
2542 struct xbridge *xbridge;
2544 ovs_rwlock_rdlock(&xlate_rwlock);
2545 xbridge = xbridge_lookup(ofproto);
2547 sflow = dpif_sflow_ref(xbridge->sflow);
2549 ovs_rwlock_unlock(&xlate_rwlock);
2554 /* Returns a reference to the ipfix handled associated with ofproto, or NULL if
2555 * there is none. The caller is responsible for decrementing the results ref
2556 * count with dpif_ipfix_unref(). */
2558 xlate_get_ipfix(const struct ofproto_dpif *ofproto)
2560 struct dpif_ipfix *ipfix = NULL;
2561 struct xbridge *xbridge;
2563 ovs_rwlock_rdlock(&xlate_rwlock);
2564 xbridge = xbridge_lookup(ofproto);
2566 ipfix = dpif_ipfix_ref(xbridge->ipfix);
2568 ovs_rwlock_unlock(&xlate_rwlock);
2573 static struct skb_priority_to_dscp *
2574 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2576 struct skb_priority_to_dscp *pdscp;
2579 hash = hash_int(skb_priority, 0);
2580 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2581 if (pdscp->skb_priority == skb_priority) {
2589 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2592 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2593 *dscp = pdscp ? pdscp->dscp : 0;
2594 return pdscp != NULL;
2598 clear_skb_priorities(struct xport *xport)
2600 struct skb_priority_to_dscp *pdscp, *next;
2602 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2603 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2609 actions_output_to_local_port(const struct xlate_ctx *ctx)
2611 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2612 const struct nlattr *a;
2615 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2616 ctx->xout->odp_actions.size) {
2617 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2618 && nl_attr_get_odp_port(a) == local_odp_port) {
2625 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2626 * into datapath actions in 'odp_actions', using 'ctx'.
2628 * The caller must take responsibility for eventually freeing 'xout', with
2629 * xlate_out_uninit(). */
2631 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2633 struct flow_wildcards *wc = &xout->wc;
2634 struct flow *flow = &xin->flow;
2636 struct rule_actions *actions = NULL;
2637 enum slow_path_reason special;
2638 const struct ofpact *ofpacts;
2639 struct xport *in_port;
2640 struct flow orig_flow;
2641 struct xlate_ctx ctx;
2645 COVERAGE_INC(xlate_actions);
2647 ovs_rwlock_rdlock(&xlate_rwlock);
2649 /* Flow initialization rules:
2650 * - 'base_flow' must match the kernel's view of the packet at the
2651 * time that action processing starts. 'flow' represents any
2652 * transformations we wish to make through actions.
2653 * - By default 'base_flow' and 'flow' are the same since the input
2654 * packet matches the output before any actions are applied.
2655 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2656 * of the received packet as seen by the kernel. If we later output
2657 * to another device without any modifications this will cause us to
2658 * insert a new tag since the original one was stripped off by the
2660 * - Tunnel metadata as received is retained in 'flow'. This allows
2661 * tunnel metadata matching also in later tables.
2662 * Since a kernel action for setting the tunnel metadata will only be
2663 * generated with actual tunnel output, changing the tunnel metadata
2664 * values in 'flow' (such as tun_id) will only have effect with a later
2665 * tunnel output action.
2666 * - Tunnel 'base_flow' is completely cleared since that is what the
2667 * kernel does. If we wish to maintain the original values an action
2668 * needs to be generated. */
2673 ctx.xout->has_learn = false;
2674 ctx.xout->has_normal = false;
2675 ctx.xout->has_fin_timeout = false;
2676 ctx.xout->nf_output_iface = NF_OUT_DROP;
2677 ctx.xout->mirrors = 0;
2678 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2679 sizeof ctx.xout->odp_actions_stub);
2680 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2682 ctx.xbridge = xbridge_lookup(xin->ofproto);
2687 ctx.rule = xin->rule;
2689 ctx.base_flow = *flow;
2690 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2691 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2693 flow_wildcards_init_catchall(wc);
2694 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2695 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2696 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2697 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2699 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
2700 if (ctx.xbridge->has_netflow) {
2701 netflow_mask_wc(flow, wc);
2705 ctx.orig_skb_priority = flow->skb_priority;
2708 ctx.mpls_depth_delta = 0;
2711 ofpacts = xin->ofpacts;
2712 ofpacts_len = xin->ofpacts_len;
2713 } else if (xin->rule) {
2714 actions = rule_dpif_get_actions(xin->rule);
2715 ofpacts = actions->ofpacts;
2716 ofpacts_len = actions->ofpacts_len;
2721 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2723 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2724 /* Do this conditionally because the copy is expensive enough that it
2725 * shows up in profiles. */
2729 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2730 switch (ctx.xbridge->frag) {
2731 case OFPC_FRAG_NORMAL:
2732 /* We must pretend that transport ports are unavailable. */
2733 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2734 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2737 case OFPC_FRAG_DROP:
2740 case OFPC_FRAG_REASM:
2743 case OFPC_FRAG_NX_MATCH:
2744 /* Nothing to do. */
2747 case OFPC_INVALID_TTL_TO_CONTROLLER:
2752 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2753 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2755 ctx.xout->slow = special;
2757 size_t sample_actions_len;
2759 if (flow->in_port.ofp_port
2760 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2761 flow->in_port.ofp_port,
2763 ctx.base_flow.vlan_tci = 0;
2766 add_sflow_action(&ctx);
2767 add_ipfix_action(&ctx);
2768 sample_actions_len = ctx.xout->odp_actions.size;
2770 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
2771 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2773 /* We've let OFPP_NORMAL and the learning action look at the
2774 * packet, so drop it now if forwarding is disabled. */
2775 if (in_port && !xport_stp_forward_state(in_port)) {
2776 ctx.xout->odp_actions.size = sample_actions_len;
2780 if (ctx.xbridge->has_in_band
2781 && in_band_must_output_to_local_port(flow)
2782 && !actions_output_to_local_port(&ctx)) {
2783 compose_output_action(&ctx, OFPP_LOCAL);
2786 fix_sflow_action(&ctx);
2788 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2789 add_mirror_actions(&ctx, &orig_flow);
2793 ofpbuf_uninit(&ctx.stack);
2795 /* Clear the metadata and register wildcard masks, because we won't
2796 * use non-header fields as part of the cache. */
2797 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2798 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);
2801 ovs_rwlock_unlock(&xlate_rwlock);
2803 rule_actions_unref(actions);