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);
52 COVERAGE_DEFINE(xlate_actions_oversize);
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
56 /* Maximum depth of flow table recursion (due to resubmit actions) in a
57 * flow translation. */
58 #define MAX_RESUBMIT_RECURSION 64
60 /* Maximum number of resubmit actions in a flow translation, whether they are
61 * recursive or not. */
62 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
64 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
67 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
68 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
70 struct list xbundles; /* Owned xbundles. */
71 struct hmap xports; /* Indexed by ofp_port. */
73 char *name; /* Name used in log messages. */
74 struct dpif *dpif; /* Datapath interface. */
75 struct mac_learning *ml; /* Mac learning handle. */
76 struct mbridge *mbridge; /* Mirroring. */
77 struct dpif_sflow *sflow; /* SFlow handle, or null. */
78 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
79 struct netflow *netflow; /* Netflow handle, or null. */
80 struct stp *stp; /* STP or null if disabled. */
82 /* Special rules installed by ofproto-dpif. */
83 struct rule_dpif *miss_rule;
84 struct rule_dpif *no_packet_in_rule;
86 enum ofp_config_flags frag; /* Fragmentation handling. */
87 bool has_in_band; /* Bridge has in band control? */
88 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
92 struct hmap_node hmap_node; /* In global 'xbundles' map. */
93 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
95 struct list list_node; /* In parent 'xbridges' list. */
96 struct xbridge *xbridge; /* Parent xbridge. */
98 struct list xports; /* Contains "struct xport"s. */
100 char *name; /* Name used in log messages. */
101 struct bond *bond; /* Nonnull iff more than one port. */
102 struct lacp *lacp; /* LACP handle or null. */
104 enum port_vlan_mode vlan_mode; /* VLAN mode. */
105 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
106 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
107 * NULL if all VLANs are trunked. */
108 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
109 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
113 struct hmap_node hmap_node; /* Node in global 'xports' map. */
114 struct ofport_dpif *ofport; /* Key in global 'xports map. */
116 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
117 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
119 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
121 struct list bundle_node; /* In parent xbundle (if it exists). */
122 struct xbundle *xbundle; /* Parent xbundle or null. */
124 struct netdev *netdev; /* 'ofport''s netdev. */
126 struct xbridge *xbridge; /* Parent bridge. */
127 struct xport *peer; /* Patch port peer or null. */
129 enum ofputil_port_config config; /* OpenFlow port configuration. */
130 enum ofputil_port_state state; /* OpenFlow port state. */
131 int stp_port_no; /* STP port number or -1 if not in use. */
133 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
135 bool may_enable; /* May be enabled in bonds. */
136 bool is_tunnel; /* Is a tunnel port. */
138 struct cfm *cfm; /* CFM handle or null. */
139 struct bfd *bfd; /* BFD handle or null. */
143 struct xlate_in *xin;
144 struct xlate_out *xout;
146 const struct xbridge *xbridge;
148 /* Flow at the last commit. */
149 struct flow base_flow;
151 /* Tunnel IP destination address as received. This is stored separately
152 * as the base_flow.tunnel is cleared on init to reflect the datapath
153 * behavior. Used to make sure not to send tunneled output to ourselves,
154 * which might lead to an infinite loop. This could happen easily
155 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
156 * actually set the tun_dst field. */
157 ovs_be32 orig_tunnel_ip_dst;
159 /* Stack for the push and pop actions. Each stack element is of type
160 * "union mf_subvalue". */
161 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
164 /* The rule that we are currently translating, or NULL. */
165 struct rule_dpif *rule;
167 int mpls_depth_delta; /* Delta of the mpls stack depth since
168 * actions were last committed.
169 * Must be between -1 and 1 inclusive. */
170 ovs_be32 pre_push_mpls_lse; /* Used to record the top-most MPLS LSE
171 * prior to an mpls_push so that it may be
172 * used for a subsequent mpls_pop. */
174 /* Resubmit statistics, via xlate_table_action(). */
175 int recurse; /* Current resubmit nesting depth. */
176 int resubmits; /* Total number of resubmits. */
178 uint32_t orig_skb_priority; /* Priority when packet arrived. */
179 uint8_t table_id; /* OpenFlow table ID where flow was found. */
180 uint32_t sflow_n_outputs; /* Number of output ports. */
181 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
182 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
183 bool exit; /* No further actions should be processed. */
185 /* OpenFlow 1.1+ action set.
187 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
188 * When translation is otherwise complete, ofpacts_execute_action_set()
189 * converts it to a set of "struct ofpact"s that can be translated into
190 * datapath actions. */
191 struct ofpbuf action_set; /* Action set. */
192 uint64_t action_set_stub[1024 / 8];
195 /* A controller may use OFPP_NONE as the ingress port to indicate that
196 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
197 * when an input bundle is needed for validation (e.g., mirroring or
198 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
199 * any 'port' structs, so care must be taken when dealing with it.
200 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
201 static struct xbundle ofpp_none_bundle;
203 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
204 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
205 * traffic egressing the 'ofport' with that priority should be marked with. */
206 struct skb_priority_to_dscp {
207 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
208 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
210 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
213 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
214 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
215 static struct hmap xports = HMAP_INITIALIZER(&xports);
217 static bool may_receive(const struct xport *, struct xlate_ctx *);
218 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
220 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
221 OVS_REQ_RDLOCK(xlate_rwlock);
222 static void xlate_normal(struct xlate_ctx *);
223 static void xlate_report(struct xlate_ctx *, const char *);
224 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
225 uint8_t table_id, bool may_packet_in);
226 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
227 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
228 static void output_normal(struct xlate_ctx *, const struct xbundle *,
230 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
232 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
233 static struct xbundle *xbundle_lookup(const struct ofbundle *);
234 static struct xport *xport_lookup(const struct ofport_dpif *);
235 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
236 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
237 uint32_t skb_priority);
238 static void clear_skb_priorities(struct xport *);
239 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
243 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
244 struct dpif *dpif, struct rule_dpif *miss_rule,
245 struct rule_dpif *no_packet_in_rule,
246 const struct mac_learning *ml, struct stp *stp,
247 const struct mbridge *mbridge,
248 const struct dpif_sflow *sflow,
249 const struct dpif_ipfix *ipfix,
250 const struct netflow *netflow, enum ofp_config_flags frag,
251 bool forward_bpdu, bool has_in_band)
253 struct xbridge *xbridge = xbridge_lookup(ofproto);
256 xbridge = xzalloc(sizeof *xbridge);
257 xbridge->ofproto = ofproto;
259 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
260 hmap_init(&xbridge->xports);
261 list_init(&xbridge->xbundles);
264 if (xbridge->ml != ml) {
265 mac_learning_unref(xbridge->ml);
266 xbridge->ml = mac_learning_ref(ml);
269 if (xbridge->mbridge != mbridge) {
270 mbridge_unref(xbridge->mbridge);
271 xbridge->mbridge = mbridge_ref(mbridge);
274 if (xbridge->sflow != sflow) {
275 dpif_sflow_unref(xbridge->sflow);
276 xbridge->sflow = dpif_sflow_ref(sflow);
279 if (xbridge->ipfix != ipfix) {
280 dpif_ipfix_unref(xbridge->ipfix);
281 xbridge->ipfix = dpif_ipfix_ref(ipfix);
284 if (xbridge->stp != stp) {
285 stp_unref(xbridge->stp);
286 xbridge->stp = stp_ref(stp);
289 if (xbridge->netflow != netflow) {
290 netflow_unref(xbridge->netflow);
291 xbridge->netflow = netflow_ref(netflow);
295 xbridge->name = xstrdup(name);
297 xbridge->dpif = dpif;
298 xbridge->forward_bpdu = forward_bpdu;
299 xbridge->has_in_band = has_in_band;
300 xbridge->frag = frag;
301 xbridge->miss_rule = miss_rule;
302 xbridge->no_packet_in_rule = no_packet_in_rule;
306 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
308 struct xbridge *xbridge = xbridge_lookup(ofproto);
309 struct xbundle *xbundle, *next_xbundle;
310 struct xport *xport, *next_xport;
316 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
317 xlate_ofport_remove(xport->ofport);
320 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
321 xlate_bundle_remove(xbundle->ofbundle);
324 hmap_remove(&xbridges, &xbridge->hmap_node);
325 mac_learning_unref(xbridge->ml);
326 mbridge_unref(xbridge->mbridge);
327 dpif_sflow_unref(xbridge->sflow);
328 dpif_ipfix_unref(xbridge->ipfix);
329 stp_unref(xbridge->stp);
330 hmap_destroy(&xbridge->xports);
336 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
337 const char *name, enum port_vlan_mode vlan_mode, int vlan,
338 unsigned long *trunks, bool use_priority_tags,
339 const struct bond *bond, const struct lacp *lacp,
342 struct xbundle *xbundle = xbundle_lookup(ofbundle);
345 xbundle = xzalloc(sizeof *xbundle);
346 xbundle->ofbundle = ofbundle;
347 xbundle->xbridge = xbridge_lookup(ofproto);
349 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
350 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
351 list_init(&xbundle->xports);
354 ovs_assert(xbundle->xbridge);
357 xbundle->name = xstrdup(name);
359 xbundle->vlan_mode = vlan_mode;
360 xbundle->vlan = vlan;
361 xbundle->trunks = trunks;
362 xbundle->use_priority_tags = use_priority_tags;
363 xbundle->floodable = floodable;
365 if (xbundle->bond != bond) {
366 bond_unref(xbundle->bond);
367 xbundle->bond = bond_ref(bond);
370 if (xbundle->lacp != lacp) {
371 lacp_unref(xbundle->lacp);
372 xbundle->lacp = lacp_ref(lacp);
377 xlate_bundle_remove(struct ofbundle *ofbundle)
379 struct xbundle *xbundle = xbundle_lookup(ofbundle);
380 struct xport *xport, *next;
386 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
387 list_remove(&xport->bundle_node);
388 xport->xbundle = NULL;
391 hmap_remove(&xbundles, &xbundle->hmap_node);
392 list_remove(&xbundle->list_node);
393 bond_unref(xbundle->bond);
394 lacp_unref(xbundle->lacp);
400 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
401 struct ofport_dpif *ofport, ofp_port_t ofp_port,
402 odp_port_t odp_port, const struct netdev *netdev,
403 const struct cfm *cfm, const struct bfd *bfd,
404 struct ofport_dpif *peer, int stp_port_no,
405 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
406 enum ofputil_port_config config,
407 enum ofputil_port_state state, bool is_tunnel,
410 struct xport *xport = xport_lookup(ofport);
414 xport = xzalloc(sizeof *xport);
415 xport->ofport = ofport;
416 xport->xbridge = xbridge_lookup(ofproto);
417 xport->ofp_port = ofp_port;
419 hmap_init(&xport->skb_priorities);
420 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
421 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
422 hash_ofp_port(xport->ofp_port));
425 ovs_assert(xport->ofp_port == ofp_port);
427 xport->config = config;
428 xport->state = state;
429 xport->stp_port_no = stp_port_no;
430 xport->is_tunnel = is_tunnel;
431 xport->may_enable = may_enable;
432 xport->odp_port = odp_port;
434 if (xport->netdev != netdev) {
435 netdev_close(xport->netdev);
436 xport->netdev = netdev_ref(netdev);
439 if (xport->cfm != cfm) {
440 cfm_unref(xport->cfm);
441 xport->cfm = cfm_ref(cfm);
444 if (xport->bfd != bfd) {
445 bfd_unref(xport->bfd);
446 xport->bfd = bfd_ref(bfd);
450 xport->peer->peer = NULL;
452 xport->peer = xport_lookup(peer);
454 xport->peer->peer = xport;
457 if (xport->xbundle) {
458 list_remove(&xport->bundle_node);
460 xport->xbundle = xbundle_lookup(ofbundle);
461 if (xport->xbundle) {
462 list_insert(&xport->xbundle->xports, &xport->bundle_node);
465 clear_skb_priorities(xport);
466 for (i = 0; i < n_qdscp; i++) {
467 struct skb_priority_to_dscp *pdscp;
468 uint32_t skb_priority;
470 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
475 pdscp = xmalloc(sizeof *pdscp);
476 pdscp->skb_priority = skb_priority;
477 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
478 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
479 hash_int(pdscp->skb_priority, 0));
484 xlate_ofport_remove(struct ofport_dpif *ofport)
486 struct xport *xport = xport_lookup(ofport);
493 xport->peer->peer = NULL;
497 if (xport->xbundle) {
498 list_remove(&xport->bundle_node);
501 clear_skb_priorities(xport);
502 hmap_destroy(&xport->skb_priorities);
504 hmap_remove(&xports, &xport->hmap_node);
505 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
507 netdev_close(xport->netdev);
508 cfm_unref(xport->cfm);
509 bfd_unref(xport->bfd);
513 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
514 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
515 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
516 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
517 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
518 * 'packet' ingressed.
520 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
521 * 'flow''s in_port to OFPP_NONE.
523 * This function does post-processing on data returned from
524 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
525 * of the upcall processing logic. In particular, if the extracted in_port is
526 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
527 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
528 * a VLAN header onto 'packet' (if it is nonnull).
530 * Similarly, this function also includes some logic to help with tunnels. It
531 * may modify 'flow' as necessary to make the tunneling implementation
532 * transparent to the upcall processing logic.
534 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
535 * or some other positive errno if there are other problems. */
537 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
538 const struct nlattr *key, size_t key_len,
539 struct flow *flow, enum odp_key_fitness *fitnessp,
540 struct ofproto_dpif **ofproto, odp_port_t *odp_in_port)
542 enum odp_key_fitness fitness;
543 const struct xport *xport;
546 ovs_rwlock_rdlock(&xlate_rwlock);
547 fitness = odp_flow_key_to_flow(key, key_len, flow);
548 if (fitness == ODP_FIT_ERROR) {
554 *odp_in_port = flow->in_port.odp_port;
557 xport = xport_lookup(tnl_port_should_receive(flow)
558 ? tnl_port_receive(flow)
559 : odp_port_to_ofport(backer, flow->in_port.odp_port));
561 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
566 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
568 /* Make the packet resemble the flow, so that it gets sent to
569 * an OpenFlow controller properly, so that it looks correct
570 * for sFlow, and so that flow_extract() will get the correct
571 * vlan_tci if it is called on 'packet'.
573 * The allocated space inside 'packet' probably also contains
574 * 'key', that is, both 'packet' and 'key' are probably part of
575 * a struct dpif_upcall (see the large comment on that
576 * structure definition), so pushing data on 'packet' is in
577 * general not a good idea since it could overwrite 'key' or
578 * free it as a side effect. However, it's OK in this special
579 * case because we know that 'packet' is inside a Netlink
580 * attribute: pushing 4 bytes will just overwrite the 4-byte
581 * "struct nlattr", which is fine since we don't need that
583 eth_push_vlan(packet, flow->vlan_tci);
585 /* We can't reproduce 'key' from 'flow'. */
586 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
591 *ofproto = xport->xbridge->ofproto;
598 ovs_rwlock_unlock(&xlate_rwlock);
602 static struct xbridge *
603 xbridge_lookup(const struct ofproto_dpif *ofproto)
605 struct xbridge *xbridge;
611 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
613 if (xbridge->ofproto == ofproto) {
620 static struct xbundle *
621 xbundle_lookup(const struct ofbundle *ofbundle)
623 struct xbundle *xbundle;
629 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
631 if (xbundle->ofbundle == ofbundle) {
638 static struct xport *
639 xport_lookup(const struct ofport_dpif *ofport)
647 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
649 if (xport->ofport == ofport) {
656 static struct stp_port *
657 xport_get_stp_port(const struct xport *xport)
659 return xport->xbridge->stp && xport->stp_port_no != -1
660 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
664 static enum stp_state
665 xport_stp_learn_state(const struct xport *xport)
667 struct stp_port *sp = xport_get_stp_port(xport);
668 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
672 xport_stp_forward_state(const struct xport *xport)
674 struct stp_port *sp = xport_get_stp_port(xport);
675 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
678 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
679 * were used to make the determination.*/
681 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
683 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
684 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
688 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
690 struct stp_port *sp = xport_get_stp_port(xport);
691 struct ofpbuf payload = *packet;
692 struct eth_header *eth = payload.data;
694 /* Sink packets on ports that have STP disabled when the bridge has
696 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
700 /* Trim off padding on payload. */
701 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
702 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
705 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
706 stp_received_bpdu(sp, payload.data, payload.size);
710 static struct xport *
711 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
715 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
717 if (xport->ofp_port == ofp_port) {
725 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
727 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
728 return xport ? xport->odp_port : ODPP_NONE;
732 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
736 xport = get_ofp_port(ctx->xbridge, ofp_port);
737 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
738 xport->state & OFPUTIL_PS_LINK_DOWN) {
745 static const struct ofputil_bucket *
746 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
750 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
752 struct group_dpif *group;
755 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
760 hit = group_first_live_bucket(ctx, group, depth) != NULL;
762 group_dpif_release(group);
766 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
769 bucket_is_alive(const struct xlate_ctx *ctx,
770 const struct ofputil_bucket *bucket, int depth)
772 if (depth >= MAX_LIVENESS_RECURSION) {
773 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
775 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
776 MAX_LIVENESS_RECURSION);
780 return !ofputil_bucket_has_liveness(bucket) ||
781 (bucket->watch_port != OFPP_ANY &&
782 odp_port_is_alive(ctx, bucket->watch_port)) ||
783 (bucket->watch_group != OFPG_ANY &&
784 group_is_alive(ctx, bucket->watch_group, depth + 1));
787 static const struct ofputil_bucket *
788 group_first_live_bucket(const struct xlate_ctx *ctx,
789 const struct group_dpif *group, int depth)
791 struct ofputil_bucket *bucket;
792 const struct list *buckets;
794 group_dpif_get_buckets(group, &buckets);
795 LIST_FOR_EACH (bucket, list_node, buckets) {
796 if (bucket_is_alive(ctx, bucket, depth)) {
804 static const struct ofputil_bucket *
805 group_best_live_bucket(const struct xlate_ctx *ctx,
806 const struct group_dpif *group,
809 const struct ofputil_bucket *best_bucket = NULL;
810 uint32_t best_score = 0;
813 const struct ofputil_bucket *bucket;
814 const struct list *buckets;
816 group_dpif_get_buckets(group, &buckets);
817 LIST_FOR_EACH (bucket, list_node, buckets) {
818 if (bucket_is_alive(ctx, bucket, 0)) {
819 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
820 if (score >= best_score) {
821 best_bucket = bucket;
832 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
834 return (bundle->vlan_mode != PORT_VLAN_ACCESS
835 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
839 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
841 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
845 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
847 return xbundle != &ofpp_none_bundle
848 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
853 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
855 return xbundle != &ofpp_none_bundle
856 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
861 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
863 return xbundle != &ofpp_none_bundle
864 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
868 static struct xbundle *
869 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
870 bool warn, struct xport **in_xportp)
874 /* Find the port and bundle for the received packet. */
875 xport = get_ofp_port(xbridge, in_port);
879 if (xport && xport->xbundle) {
880 return xport->xbundle;
883 /* Special-case OFPP_NONE, which a controller may use as the ingress
884 * port for traffic that it is sourcing. */
885 if (in_port == OFPP_NONE) {
886 ofpp_none_bundle.name = "OFPP_NONE";
887 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
888 return &ofpp_none_bundle;
891 /* Odd. A few possible reasons here:
893 * - We deleted a port but there are still a few packets queued up
896 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
897 * we don't know about.
899 * - The ofproto client didn't configure the port as part of a bundle.
900 * This is particularly likely to happen if a packet was received on the
901 * port after it was created, but before the client had a chance to
902 * configure its bundle.
905 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
907 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
908 "port %"PRIu16, xbridge->name, in_port);
914 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
916 const struct xbridge *xbridge = ctx->xbridge;
917 mirror_mask_t mirrors;
918 struct xbundle *in_xbundle;
922 mirrors = ctx->xout->mirrors;
923 ctx->xout->mirrors = 0;
925 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
926 ctx->xin->packet != NULL, NULL);
930 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
932 /* Drop frames on bundles reserved for mirroring. */
933 if (xbundle_mirror_out(xbridge, in_xbundle)) {
934 if (ctx->xin->packet != NULL) {
935 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
936 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
937 "%s, which is reserved exclusively for mirroring",
938 ctx->xbridge->name, in_xbundle->name);
940 ofpbuf_clear(&ctx->xout->odp_actions);
945 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
946 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
949 vlan = input_vid_to_vlan(in_xbundle, vid);
955 /* Restore the original packet before adding the mirror actions. */
956 ctx->xin->flow = *orig_flow;
959 mirror_mask_t dup_mirrors;
960 struct ofbundle *out;
961 unsigned long *vlans;
966 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
967 &vlans, &dup_mirrors, &out, &out_vlan);
968 ovs_assert(has_mirror);
971 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
973 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
976 if (!vlan_mirrored) {
977 mirrors = zero_rightmost_1bit(mirrors);
981 mirrors &= ~dup_mirrors;
982 ctx->xout->mirrors |= dup_mirrors;
984 struct xbundle *out_xbundle = xbundle_lookup(out);
986 output_normal(ctx, out_xbundle, vlan);
988 } else if (vlan != out_vlan
989 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
990 struct xbundle *xbundle;
992 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
993 if (xbundle_includes_vlan(xbundle, out_vlan)
994 && !xbundle_mirror_out(xbridge, xbundle)) {
995 output_normal(ctx, xbundle, out_vlan);
1002 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1003 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1004 * the bundle on which the packet was received, returns the VLAN to which the
1007 * Both 'vid' and the return value are in the range 0...4095. */
1009 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1011 switch (in_xbundle->vlan_mode) {
1012 case PORT_VLAN_ACCESS:
1013 return in_xbundle->vlan;
1016 case PORT_VLAN_TRUNK:
1019 case PORT_VLAN_NATIVE_UNTAGGED:
1020 case PORT_VLAN_NATIVE_TAGGED:
1021 return vid ? vid : in_xbundle->vlan;
1028 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1029 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1032 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1033 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1036 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1038 /* Allow any VID on the OFPP_NONE port. */
1039 if (in_xbundle == &ofpp_none_bundle) {
1043 switch (in_xbundle->vlan_mode) {
1044 case PORT_VLAN_ACCESS:
1047 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1048 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1049 "packet received on port %s configured as VLAN "
1050 "%"PRIu16" access port", vid, in_xbundle->name,
1057 case PORT_VLAN_NATIVE_UNTAGGED:
1058 case PORT_VLAN_NATIVE_TAGGED:
1060 /* Port must always carry its native VLAN. */
1064 case PORT_VLAN_TRUNK:
1065 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1067 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1068 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1069 "received on port %s not configured for trunking "
1070 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1082 /* Given 'vlan', the VLAN that a packet belongs to, and
1083 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1084 * that should be included in the 802.1Q header. (If the return value is 0,
1085 * then the 802.1Q header should only be included in the packet if there is a
1088 * Both 'vlan' and the return value are in the range 0...4095. */
1090 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1092 switch (out_xbundle->vlan_mode) {
1093 case PORT_VLAN_ACCESS:
1096 case PORT_VLAN_TRUNK:
1097 case PORT_VLAN_NATIVE_TAGGED:
1100 case PORT_VLAN_NATIVE_UNTAGGED:
1101 return vlan == out_xbundle->vlan ? 0 : vlan;
1109 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1112 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1114 ovs_be16 tci, old_tci;
1115 struct xport *xport;
1117 vid = output_vlan_to_vid(out_xbundle, vlan);
1118 if (list_is_empty(&out_xbundle->xports)) {
1119 /* Partially configured bundle with no slaves. Drop the packet. */
1121 } else if (!out_xbundle->bond) {
1122 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1125 struct ofport_dpif *ofport;
1127 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1128 &ctx->xout->wc, vid);
1129 xport = xport_lookup(ofport);
1132 /* No slaves enabled, so drop packet. */
1137 old_tci = *flow_tci;
1139 if (tci || out_xbundle->use_priority_tags) {
1140 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1142 tci |= htons(VLAN_CFI);
1147 compose_output_action(ctx, xport->ofp_port);
1148 *flow_tci = old_tci;
1151 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1152 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1153 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1155 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1157 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1161 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1162 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1166 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1167 if (flow->nw_proto == ARP_OP_REPLY) {
1169 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1170 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1171 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1173 return flow->nw_src == flow->nw_dst;
1179 /* Checks whether a MAC learning update is necessary for MAC learning table
1180 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1183 * Most packets processed through the MAC learning table do not actually
1184 * change it in any way. This function requires only a read lock on the MAC
1185 * learning table, so it is much cheaper in this common case.
1187 * Keep the code here synchronized with that in update_learning_table__()
1190 is_mac_learning_update_needed(const struct mac_learning *ml,
1191 const struct flow *flow,
1192 struct flow_wildcards *wc,
1193 int vlan, struct xbundle *in_xbundle)
1194 OVS_REQ_RDLOCK(ml->rwlock)
1196 struct mac_entry *mac;
1198 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1202 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1203 if (!mac || mac_entry_age(ml, mac)) {
1207 if (is_gratuitous_arp(flow, wc)) {
1208 /* We don't want to learn from gratuitous ARP packets that are
1209 * reflected back over bond slaves so we lock the learning table. */
1210 if (!in_xbundle->bond) {
1212 } else if (mac_entry_is_grat_arp_locked(mac)) {
1217 return mac->port.p != in_xbundle->ofbundle;
1221 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1222 * received on 'in_xbundle' in 'vlan'.
1224 * This code repeats all the checks in is_mac_learning_update_needed() because
1225 * the lock was released between there and here and thus the MAC learning state
1226 * could have changed.
1228 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1231 update_learning_table__(const struct xbridge *xbridge,
1232 const struct flow *flow, struct flow_wildcards *wc,
1233 int vlan, struct xbundle *in_xbundle)
1234 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1236 struct mac_entry *mac;
1238 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1242 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1243 if (is_gratuitous_arp(flow, wc)) {
1244 /* We don't want to learn from gratuitous ARP packets that are
1245 * reflected back over bond slaves so we lock the learning table. */
1246 if (!in_xbundle->bond) {
1247 mac_entry_set_grat_arp_lock(mac);
1248 } else if (mac_entry_is_grat_arp_locked(mac)) {
1253 if (mac->port.p != in_xbundle->ofbundle) {
1254 /* The log messages here could actually be useful in debugging,
1255 * so keep the rate limit relatively high. */
1256 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1258 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1259 "on port %s in VLAN %d",
1260 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1261 in_xbundle->name, vlan);
1263 mac->port.p = in_xbundle->ofbundle;
1264 mac_learning_changed(xbridge->ml);
1269 update_learning_table(const struct xbridge *xbridge,
1270 const struct flow *flow, struct flow_wildcards *wc,
1271 int vlan, struct xbundle *in_xbundle)
1275 /* Don't learn the OFPP_NONE port. */
1276 if (in_xbundle == &ofpp_none_bundle) {
1280 /* First try the common case: no change to MAC learning table. */
1281 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1282 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1284 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1287 /* Slow path: MAC learning table might need an update. */
1288 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1289 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1290 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1294 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1295 * dropped. Returns true if they may be forwarded, false if they should be
1298 * 'in_port' must be the xport that corresponds to flow->in_port.
1299 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1301 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1302 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1303 * checked by input_vid_is_valid().
1305 * May also add tags to '*tags', although the current implementation only does
1306 * so in one special case.
1309 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1312 struct xbundle *in_xbundle = in_port->xbundle;
1313 const struct xbridge *xbridge = ctx->xbridge;
1314 struct flow *flow = &ctx->xin->flow;
1316 /* Drop frames for reserved multicast addresses
1317 * only if forward_bpdu option is absent. */
1318 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1319 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1323 if (in_xbundle->bond) {
1324 struct mac_entry *mac;
1326 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1332 xlate_report(ctx, "bonding refused admissibility, dropping");
1335 case BV_DROP_IF_MOVED:
1336 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1337 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1338 if (mac && mac->port.p != in_xbundle->ofbundle &&
1339 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1340 || mac_entry_is_grat_arp_locked(mac))) {
1341 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1342 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1346 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1355 xlate_normal(struct xlate_ctx *ctx)
1357 struct flow_wildcards *wc = &ctx->xout->wc;
1358 struct flow *flow = &ctx->xin->flow;
1359 struct xbundle *in_xbundle;
1360 struct xport *in_port;
1361 struct mac_entry *mac;
1366 ctx->xout->has_normal = true;
1368 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1369 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1370 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1372 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1373 ctx->xin->packet != NULL, &in_port);
1375 xlate_report(ctx, "no input bundle, dropping");
1379 /* Drop malformed frames. */
1380 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1381 !(flow->vlan_tci & htons(VLAN_CFI))) {
1382 if (ctx->xin->packet != NULL) {
1383 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1384 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1385 "VLAN tag received on port %s",
1386 ctx->xbridge->name, in_xbundle->name);
1388 xlate_report(ctx, "partial VLAN tag, dropping");
1392 /* Drop frames on bundles reserved for mirroring. */
1393 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1394 if (ctx->xin->packet != NULL) {
1395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1396 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1397 "%s, which is reserved exclusively for mirroring",
1398 ctx->xbridge->name, in_xbundle->name);
1400 xlate_report(ctx, "input port is mirror output port, dropping");
1405 vid = vlan_tci_to_vid(flow->vlan_tci);
1406 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1407 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1410 vlan = input_vid_to_vlan(in_xbundle, vid);
1412 /* Check other admissibility requirements. */
1413 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1417 /* Learn source MAC. */
1418 if (ctx->xin->may_learn) {
1419 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1422 /* Determine output bundle. */
1423 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1424 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1425 mac_port = mac ? mac->port.p : NULL;
1426 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1429 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1430 if (mac_xbundle && mac_xbundle != in_xbundle) {
1431 xlate_report(ctx, "forwarding to learned port");
1432 output_normal(ctx, mac_xbundle, vlan);
1433 } else if (!mac_xbundle) {
1434 xlate_report(ctx, "learned port is unknown, dropping");
1436 xlate_report(ctx, "learned port is input port, dropping");
1439 struct xbundle *xbundle;
1441 xlate_report(ctx, "no learned MAC for destination, flooding");
1442 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1443 if (xbundle != in_xbundle
1444 && xbundle_includes_vlan(xbundle, vlan)
1445 && xbundle->floodable
1446 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1447 output_normal(ctx, xbundle, vlan);
1450 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1454 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1455 * the number of packets out of UINT32_MAX to sample. The given
1456 * cookie is passed back in the callback for each sampled packet.
1459 compose_sample_action(const struct xbridge *xbridge,
1460 struct ofpbuf *odp_actions,
1461 const struct flow *flow,
1462 const uint32_t probability,
1463 const union user_action_cookie *cookie,
1464 const size_t cookie_size)
1466 size_t sample_offset, actions_offset;
1467 odp_port_t odp_port;
1471 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1473 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1475 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1477 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1478 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1479 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1481 nl_msg_end_nested(odp_actions, actions_offset);
1482 nl_msg_end_nested(odp_actions, sample_offset);
1483 return cookie_offset;
1487 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1488 odp_port_t odp_port, unsigned int n_outputs,
1489 union user_action_cookie *cookie)
1493 cookie->type = USER_ACTION_COOKIE_SFLOW;
1494 cookie->sflow.vlan_tci = vlan_tci;
1496 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1497 * port information") for the interpretation of cookie->output. */
1498 switch (n_outputs) {
1500 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1501 cookie->sflow.output = 0x40000000 | 256;
1505 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1507 cookie->sflow.output = ifindex;
1512 /* 0x80000000 means "multiple output ports. */
1513 cookie->sflow.output = 0x80000000 | n_outputs;
1518 /* Compose SAMPLE action for sFlow bridge sampling. */
1520 compose_sflow_action(const struct xbridge *xbridge,
1521 struct ofpbuf *odp_actions,
1522 const struct flow *flow,
1523 odp_port_t odp_port)
1525 uint32_t probability;
1526 union user_action_cookie cookie;
1528 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1532 probability = dpif_sflow_get_probability(xbridge->sflow);
1533 compose_sflow_cookie(xbridge, htons(0), odp_port,
1534 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1536 return compose_sample_action(xbridge, odp_actions, flow, probability,
1537 &cookie, sizeof cookie.sflow);
1541 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1542 uint32_t obs_domain_id, uint32_t obs_point_id,
1543 union user_action_cookie *cookie)
1545 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1546 cookie->flow_sample.probability = probability;
1547 cookie->flow_sample.collector_set_id = collector_set_id;
1548 cookie->flow_sample.obs_domain_id = obs_domain_id;
1549 cookie->flow_sample.obs_point_id = obs_point_id;
1553 compose_ipfix_cookie(union user_action_cookie *cookie)
1555 cookie->type = USER_ACTION_COOKIE_IPFIX;
1558 /* Compose SAMPLE action for IPFIX bridge sampling. */
1560 compose_ipfix_action(const struct xbridge *xbridge,
1561 struct ofpbuf *odp_actions,
1562 const struct flow *flow)
1564 uint32_t probability;
1565 union user_action_cookie cookie;
1567 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1571 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1572 compose_ipfix_cookie(&cookie);
1574 compose_sample_action(xbridge, odp_actions, flow, probability,
1575 &cookie, sizeof cookie.ipfix);
1578 /* SAMPLE action for sFlow must be first action in any given list of
1579 * actions. At this point we do not have all information required to
1580 * build it. So try to build sample action as complete as possible. */
1582 add_sflow_action(struct xlate_ctx *ctx)
1584 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1585 &ctx->xout->odp_actions,
1586 &ctx->xin->flow, ODPP_NONE);
1587 ctx->sflow_odp_port = 0;
1588 ctx->sflow_n_outputs = 0;
1591 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1592 * of actions, eventually after the SAMPLE action for sFlow. */
1594 add_ipfix_action(struct xlate_ctx *ctx)
1596 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1600 /* Fix SAMPLE action according to data collected while composing ODP actions.
1601 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1602 * USERSPACE action's user-cookie which is required for sflow. */
1604 fix_sflow_action(struct xlate_ctx *ctx)
1606 const struct flow *base = &ctx->base_flow;
1607 union user_action_cookie *cookie;
1609 if (!ctx->user_cookie_offset) {
1613 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1614 sizeof cookie->sflow);
1615 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1617 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1618 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1621 static enum slow_path_reason
1622 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1623 const struct xport *xport, const struct ofpbuf *packet)
1625 struct flow_wildcards *wc = &ctx->xout->wc;
1626 const struct xbridge *xbridge = ctx->xbridge;
1630 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1632 cfm_process_heartbeat(xport->cfm, packet);
1635 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1637 bfd_process_packet(xport->bfd, flow, packet);
1640 } else if (xport->xbundle && xport->xbundle->lacp
1641 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1643 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1646 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1648 stp_process_packet(xport, packet);
1657 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1660 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1661 struct flow_wildcards *wc = &ctx->xout->wc;
1662 struct flow *flow = &ctx->xin->flow;
1663 ovs_be16 flow_vlan_tci;
1664 uint32_t flow_pkt_mark;
1665 uint8_t flow_nw_tos;
1666 odp_port_t out_port, odp_port;
1669 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1670 * before traversing a patch port. */
1671 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 23);
1674 xlate_report(ctx, "Nonexistent output port");
1676 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1677 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1679 } else if (check_stp && !xport_stp_forward_state(xport)) {
1680 xlate_report(ctx, "STP not in forwarding state, skipping output");
1684 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1685 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1690 const struct xport *peer = xport->peer;
1691 struct flow old_flow = ctx->xin->flow;
1692 enum slow_path_reason special;
1694 ctx->xbridge = peer->xbridge;
1695 flow->in_port.ofp_port = peer->ofp_port;
1696 flow->metadata = htonll(0);
1697 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1698 memset(flow->regs, 0, sizeof flow->regs);
1700 special = process_special(ctx, &ctx->xin->flow, peer,
1703 ctx->xout->slow |= special;
1704 } else if (may_receive(peer, ctx)) {
1705 if (xport_stp_forward_state(peer)) {
1706 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1708 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1709 * learning action look at the packet, then drop it. */
1710 struct flow old_base_flow = ctx->base_flow;
1711 size_t old_size = ctx->xout->odp_actions.size;
1712 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1713 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1714 ctx->xout->mirrors = old_mirrors;
1715 ctx->base_flow = old_base_flow;
1716 ctx->xout->odp_actions.size = old_size;
1720 ctx->xin->flow = old_flow;
1721 ctx->xbridge = xport->xbridge;
1723 if (ctx->xin->resubmit_stats) {
1724 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1725 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1727 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1734 flow_vlan_tci = flow->vlan_tci;
1735 flow_pkt_mark = flow->pkt_mark;
1736 flow_nw_tos = flow->nw_tos;
1738 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1739 wc->masks.nw_tos |= IP_ECN_MASK;
1740 flow->nw_tos &= ~IP_DSCP_MASK;
1741 flow->nw_tos |= dscp;
1744 if (xport->is_tunnel) {
1745 /* Save tunnel metadata so that changes made due to
1746 * the Logical (tunnel) Port are not visible for any further
1747 * matches, while explicit set actions on tunnel metadata are.
1749 struct flow_tnl flow_tnl = flow->tunnel;
1750 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1751 if (odp_port == ODPP_NONE) {
1752 xlate_report(ctx, "Tunneling decided against output");
1753 goto out; /* restore flow_nw_tos */
1755 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1756 xlate_report(ctx, "Not tunneling to our own address");
1757 goto out; /* restore flow_nw_tos */
1759 if (ctx->xin->resubmit_stats) {
1760 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1762 out_port = odp_port;
1763 commit_odp_tunnel_action(flow, &ctx->base_flow,
1764 &ctx->xout->odp_actions);
1765 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1767 ofp_port_t vlandev_port;
1769 odp_port = xport->odp_port;
1770 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1771 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1773 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1775 if (vlandev_port == ofp_port) {
1776 out_port = odp_port;
1778 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1779 flow->vlan_tci = htons(0);
1783 if (out_port != ODPP_NONE) {
1784 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1785 &ctx->xout->odp_actions,
1787 &ctx->mpls_depth_delta);
1788 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1791 ctx->sflow_odp_port = odp_port;
1792 ctx->sflow_n_outputs++;
1793 ctx->xout->nf_output_iface = ofp_port;
1798 flow->vlan_tci = flow_vlan_tci;
1799 flow->pkt_mark = flow_pkt_mark;
1800 flow->nw_tos = flow_nw_tos;
1804 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1806 compose_output_action__(ctx, ofp_port, true);
1810 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1812 struct rule_dpif *old_rule = ctx->rule;
1813 struct rule_actions *actions;
1815 if (ctx->xin->resubmit_stats) {
1816 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1822 actions = rule_dpif_get_actions(rule);
1823 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1824 rule_actions_unref(actions);
1825 ctx->rule = old_rule;
1830 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
1832 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1834 if (ctx->recurse >= MAX_RESUBMIT_RECURSION) {
1835 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
1836 MAX_RESUBMIT_RECURSION);
1837 } else if (ctx->resubmits >= MAX_RESUBMITS) {
1838 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
1839 } else if (ctx->xout->odp_actions.size > UINT16_MAX) {
1840 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
1841 } else if (ctx->stack.size >= 65536) {
1842 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
1851 xlate_table_action(struct xlate_ctx *ctx,
1852 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1854 if (xlate_resubmit_resource_check(ctx)) {
1855 struct rule_dpif *rule;
1856 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1857 uint8_t old_table_id = ctx->table_id;
1859 ctx->table_id = table_id;
1861 /* Look up a flow with 'in_port' as the input port. Then restore the
1862 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1863 * have surprising behavior). */
1864 ctx->xin->flow.in_port.ofp_port = in_port;
1865 rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1866 &ctx->xin->flow, &ctx->xout->wc,
1868 ctx->xin->flow.in_port.ofp_port = old_in_port;
1870 if (ctx->xin->resubmit_hook) {
1871 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1874 if (!rule && may_packet_in) {
1875 struct xport *xport;
1878 * check if table configuration flags
1879 * OFPTC11_TABLE_MISS_CONTROLLER, default.
1880 * OFPTC11_TABLE_MISS_CONTINUE,
1881 * OFPTC11_TABLE_MISS_DROP
1882 * When OF1.0, OFPTC11_TABLE_MISS_CONTINUE is used. What to do? */
1883 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1884 choose_miss_rule(xport ? xport->config : 0,
1885 ctx->xbridge->miss_rule,
1886 ctx->xbridge->no_packet_in_rule, &rule);
1889 xlate_recursively(ctx, rule);
1890 rule_dpif_unref(rule);
1893 ctx->table_id = old_table_id;
1901 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
1903 uint64_t action_list_stub[1024 / 8];
1904 struct ofpbuf action_list, action_set;
1906 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
1907 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
1909 ofpacts_execute_action_set(&action_list, &action_set);
1911 do_xlate_actions(action_list.data, action_list.size, ctx);
1914 ofpbuf_uninit(&action_set);
1915 ofpbuf_uninit(&action_list);
1919 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
1921 const struct ofputil_bucket *bucket;
1922 const struct list *buckets;
1923 struct flow old_flow = ctx->xin->flow;
1925 group_dpif_get_buckets(group, &buckets);
1927 LIST_FOR_EACH (bucket, list_node, buckets) {
1928 xlate_group_bucket(ctx, bucket);
1929 /* Roll back flow to previous state.
1930 * This is equivalent to cloning the packet for each bucket.
1932 * As a side effect any subsequently applied actions will
1933 * also effectively be applied to a clone of the packet taken
1934 * just before applying the all or indirect group. */
1935 ctx->xin->flow = old_flow;
1940 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
1942 const struct ofputil_bucket *bucket;
1944 bucket = group_first_live_bucket(ctx, group, 0);
1946 xlate_group_bucket(ctx, bucket);
1951 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
1953 struct flow_wildcards *wc = &ctx->xout->wc;
1954 const struct ofputil_bucket *bucket;
1957 basis = hash_bytes(ctx->xin->flow.dl_dst, sizeof ctx->xin->flow.dl_dst, 0);
1958 bucket = group_best_live_bucket(ctx, group, basis);
1960 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1961 xlate_group_bucket(ctx, bucket);
1966 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
1968 switch (group_dpif_get_type(group)) {
1970 case OFPGT11_INDIRECT:
1971 xlate_all_group(ctx, group);
1973 case OFPGT11_SELECT:
1974 xlate_select_group(ctx, group);
1977 xlate_ff_group(ctx, group);
1982 group_dpif_release(group);
1986 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
1988 if (xlate_resubmit_resource_check(ctx)) {
1989 struct group_dpif *group;
1992 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
1994 xlate_group_action__(ctx, group);
2004 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2005 const struct ofpact_resubmit *resubmit)
2010 in_port = resubmit->in_port;
2011 if (in_port == OFPP_IN_PORT) {
2012 in_port = ctx->xin->flow.in_port.ofp_port;
2015 table_id = resubmit->table_id;
2016 if (table_id == 255) {
2017 table_id = ctx->table_id;
2020 xlate_table_action(ctx, in_port, table_id, false);
2024 flood_packets(struct xlate_ctx *ctx, bool all)
2026 const struct xport *xport;
2028 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2029 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2034 compose_output_action__(ctx, xport->ofp_port, false);
2035 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2036 compose_output_action(ctx, xport->ofp_port);
2040 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2044 execute_controller_action(struct xlate_ctx *ctx, int len,
2045 enum ofp_packet_in_reason reason,
2046 uint16_t controller_id)
2048 struct ofproto_packet_in *pin;
2049 struct ofpbuf *packet;
2052 ctx->xout->slow |= SLOW_CONTROLLER;
2053 if (!ctx->xin->packet) {
2057 packet = ofpbuf_clone(ctx->xin->packet);
2059 key.skb_priority = 0;
2061 memset(&key.tunnel, 0, sizeof key.tunnel);
2063 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2064 &ctx->xout->odp_actions,
2066 &ctx->mpls_depth_delta);
2068 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
2069 ctx->xout->odp_actions.size, NULL, NULL);
2071 pin = xmalloc(sizeof *pin);
2072 pin->up.packet_len = packet->size;
2073 pin->up.packet = ofpbuf_steal_data(packet);
2074 pin->up.reason = reason;
2075 pin->up.table_id = ctx->table_id;
2076 pin->up.cookie = (ctx->rule
2077 ? rule_dpif_get_flow_cookie(ctx->rule)
2080 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2082 pin->controller_id = controller_id;
2083 pin->send_len = len;
2084 pin->generated_by_table_miss = (ctx->rule
2085 && rule_dpif_is_table_miss(ctx->rule));
2086 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2087 ofpbuf_delete(packet);
2091 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2093 struct flow_wildcards *wc = &ctx->xout->wc;
2094 struct flow *flow = &ctx->xin->flow;
2096 ovs_assert(eth_type_mpls(eth_type));
2098 /* If mpls_depth_delta is negative then an MPLS POP action has been
2099 * composed and the resulting MPLS label stack is unknown. This means
2100 * an MPLS PUSH action can't be composed as it needs to know either the
2101 * top-most MPLS LSE to use as a template for the new MPLS LSE, or that
2102 * there is no MPLS label stack present. Thus, stop processing.
2104 * If mpls_depth_delta is positive then an MPLS PUSH action has been
2105 * composed and no further MPLS PUSH action may be performed without
2106 * losing MPLS LSE and ether type information held in xtx->xin->flow.
2107 * Thus, stop processing.
2109 * If the MPLS LSE of the flow and base_flow differ then the MPLS LSE
2110 * has been updated. Performing a MPLS PUSH action may be would result in
2111 * losing MPLS LSE and ether type information held in xtx->xin->flow.
2112 * Thus, stop processing.
2114 * It is planned that in the future this case will be handled
2115 * by recirculation */
2116 if (ctx->mpls_depth_delta ||
2117 ctx->xin->flow.mpls_lse != ctx->base_flow.mpls_lse) {
2121 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2123 ctx->pre_push_mpls_lse = ctx->xin->flow.mpls_lse;
2125 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2126 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
2131 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2132 label = htonl(0x2); /* IPV6 Explicit Null. */
2134 label = htonl(0x0); /* IPV4 Explicit Null. */
2136 wc->masks.nw_tos |= IP_DSCP_MASK;
2137 wc->masks.nw_ttl = 0xff;
2138 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2139 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
2140 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
2142 flow->dl_type = eth_type;
2143 ctx->mpls_depth_delta++;
2149 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2151 struct flow_wildcards *wc = &ctx->xout->wc;
2153 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2157 /* If mpls_depth_delta is negative then an MPLS POP action has been
2158 * composed. Performing another MPLS POP action
2159 * would result in losing ether type that results from
2160 * the already composed MPLS POP. Thus, stop processing.
2162 * It is planned that in the future this case will be handled
2163 * by recirculation */
2164 if (ctx->mpls_depth_delta < 0) {
2168 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2170 /* If mpls_depth_delta is positive then an MPLS PUSH action has been
2171 * executed and the previous MPLS LSE saved in ctx->pre_push_mpls_lse. The
2172 * flow's MPLS LSE should be restored to that value to allow any
2173 * subsequent actions that update of the LSE to be executed correctly.
2175 if (ctx->mpls_depth_delta > 0) {
2176 ctx->xin->flow.mpls_lse = ctx->pre_push_mpls_lse;
2179 ctx->xin->flow.dl_type = eth_type;
2180 ctx->mpls_depth_delta--;
2186 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2188 struct flow *flow = &ctx->xin->flow;
2190 if (!is_ip_any(flow)) {
2194 ctx->xout->wc.masks.nw_ttl = 0xff;
2195 if (flow->nw_ttl > 1) {
2201 for (i = 0; i < ids->n_controllers; i++) {
2202 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2206 /* Stop processing for current table. */
2212 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2214 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2218 /* If mpls_depth_delta is negative then an MPLS POP action has been
2219 * executed and the resulting MPLS label stack is unknown. This means
2220 * a SET MPLS LABEL action can't be executed as it needs to manipulate
2221 * the top-most MPLS LSE. Thus, stop processing.
2223 * It is planned that in the future this case will be handled
2226 if (ctx->mpls_depth_delta < 0) {
2230 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_LABEL_MASK);
2231 set_mpls_lse_label(&ctx->xin->flow.mpls_lse, label);
2236 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2238 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2242 /* If mpls_depth_delta is negative then an MPLS POP action has been
2243 * executed and the resulting MPLS label stack is unknown. This means
2244 * a SET MPLS TC action can't be executed as it needs to manipulate
2245 * the top-most MPLS LSE. Thus, stop processing.
2247 * It is planned that in the future this case will be handled
2250 if (ctx->mpls_depth_delta < 0) {
2254 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TC_MASK);
2255 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse, tc);
2260 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2262 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2266 /* If mpls_depth_delta is negative then an MPLS POP action has been
2267 * executed and the resulting MPLS label stack is unknown. This means
2268 * a SET MPLS TTL push action can't be executed as it needs to manipulate
2269 * the top-most MPLS LSE. Thus, stop processing.
2271 * It is planned that in the future this case will be handled
2274 if (ctx->mpls_depth_delta < 0) {
2278 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
2279 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
2284 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2286 struct flow *flow = &ctx->xin->flow;
2287 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
2288 struct flow_wildcards *wc = &ctx->xout->wc;
2290 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2292 if (!eth_type_mpls(flow->dl_type)) {
2298 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
2301 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2303 /* Stop processing for current table. */
2309 xlate_output_action(struct xlate_ctx *ctx,
2310 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2312 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2314 ctx->xout->nf_output_iface = NF_OUT_DROP;
2318 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2321 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2328 flood_packets(ctx, false);
2331 flood_packets(ctx, true);
2333 case OFPP_CONTROLLER:
2334 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2340 if (port != ctx->xin->flow.in_port.ofp_port) {
2341 compose_output_action(ctx, port);
2343 xlate_report(ctx, "skipping output to input port");
2348 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2349 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2350 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2351 ctx->xout->nf_output_iface = prev_nf_output_iface;
2352 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2353 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2354 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2359 xlate_output_reg_action(struct xlate_ctx *ctx,
2360 const struct ofpact_output_reg *or)
2362 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2363 if (port <= UINT16_MAX) {
2364 union mf_subvalue value;
2366 memset(&value, 0xff, sizeof value);
2367 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2368 xlate_output_action(ctx, u16_to_ofp(port),
2369 or->max_len, false);
2374 xlate_enqueue_action(struct xlate_ctx *ctx,
2375 const struct ofpact_enqueue *enqueue)
2377 ofp_port_t ofp_port = enqueue->port;
2378 uint32_t queue_id = enqueue->queue;
2379 uint32_t flow_priority, priority;
2382 /* Translate queue to priority. */
2383 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2385 /* Fall back to ordinary output action. */
2386 xlate_output_action(ctx, enqueue->port, 0, false);
2390 /* Check output port. */
2391 if (ofp_port == OFPP_IN_PORT) {
2392 ofp_port = ctx->xin->flow.in_port.ofp_port;
2393 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2397 /* Add datapath actions. */
2398 flow_priority = ctx->xin->flow.skb_priority;
2399 ctx->xin->flow.skb_priority = priority;
2400 compose_output_action(ctx, ofp_port);
2401 ctx->xin->flow.skb_priority = flow_priority;
2403 /* Update NetFlow output port. */
2404 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2405 ctx->xout->nf_output_iface = ofp_port;
2406 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2407 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2412 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2414 uint32_t skb_priority;
2416 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2417 ctx->xin->flow.skb_priority = skb_priority;
2419 /* Couldn't translate queue to a priority. Nothing to do. A warning
2420 * has already been logged. */
2425 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2427 const struct xbridge *xbridge = xbridge_;
2438 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2441 port = get_ofp_port(xbridge, ofp_port);
2442 return port ? port->may_enable : false;
2447 xlate_bundle_action(struct xlate_ctx *ctx,
2448 const struct ofpact_bundle *bundle)
2452 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2454 CONST_CAST(struct xbridge *, ctx->xbridge));
2455 if (bundle->dst.field) {
2456 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2459 xlate_output_action(ctx, port, 0, false);
2464 xlate_learn_action(struct xlate_ctx *ctx,
2465 const struct ofpact_learn *learn)
2467 uint64_t ofpacts_stub[1024 / 8];
2468 struct ofputil_flow_mod fm;
2469 struct ofpbuf ofpacts;
2471 ctx->xout->has_learn = true;
2473 learn_mask(learn, &ctx->xout->wc);
2475 if (!ctx->xin->may_learn) {
2479 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2480 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2481 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2482 ofpbuf_uninit(&ofpacts);
2486 xlate_fin_timeout(struct xlate_ctx *ctx,
2487 const struct ofpact_fin_timeout *oft)
2489 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2490 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2491 oft->fin_hard_timeout);
2496 xlate_sample_action(struct xlate_ctx *ctx,
2497 const struct ofpact_sample *os)
2499 union user_action_cookie cookie;
2500 /* Scale the probability from 16-bit to 32-bit while representing
2501 * the same percentage. */
2502 uint32_t probability = (os->probability << 16) | os->probability;
2504 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2505 &ctx->xout->odp_actions,
2507 &ctx->mpls_depth_delta);
2509 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2510 os->obs_domain_id, os->obs_point_id, &cookie);
2511 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2512 probability, &cookie, sizeof cookie.flow_sample);
2516 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2518 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2519 ? OFPUTIL_PC_NO_RECV_STP
2520 : OFPUTIL_PC_NO_RECV)) {
2524 /* Only drop packets here if both forwarding and learning are
2525 * disabled. If just learning is enabled, we need to have
2526 * OFPP_NORMAL and the learning action have a look at the packet
2527 * before we can drop it. */
2528 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2536 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2538 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2539 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2540 ofpact_pad(&ctx->action_set);
2544 xlate_action_set(struct xlate_ctx *ctx)
2546 uint64_t action_list_stub[1024 / 64];
2547 struct ofpbuf action_list;
2549 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2550 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2551 do_xlate_actions(action_list.data, action_list.size, ctx);
2552 ofpbuf_uninit(&action_list);
2556 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2557 struct xlate_ctx *ctx)
2559 struct flow_wildcards *wc = &ctx->xout->wc;
2560 struct flow *flow = &ctx->xin->flow;
2561 const struct ofpact *a;
2563 /* dl_type already in the mask, not set below. */
2565 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2566 struct ofpact_controller *controller;
2567 const struct ofpact_metadata *metadata;
2568 const struct ofpact_set_field *set_field;
2569 const struct mf_field *mf;
2577 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2578 ofpact_get_OUTPUT(a)->max_len, true);
2582 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2587 case OFPACT_CONTROLLER:
2588 controller = ofpact_get_CONTROLLER(a);
2589 execute_controller_action(ctx, controller->max_len,
2591 controller->controller_id);
2594 case OFPACT_ENQUEUE:
2595 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2598 case OFPACT_SET_VLAN_VID:
2599 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2600 if (flow->vlan_tci & htons(VLAN_CFI) ||
2601 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2602 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2603 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2608 case OFPACT_SET_VLAN_PCP:
2609 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2610 if (flow->vlan_tci & htons(VLAN_CFI) ||
2611 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2612 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2613 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2614 << VLAN_PCP_SHIFT) | VLAN_CFI);
2618 case OFPACT_STRIP_VLAN:
2619 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2620 flow->vlan_tci = htons(0);
2623 case OFPACT_PUSH_VLAN:
2624 /* XXX 802.1AD(QinQ) */
2625 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2626 flow->vlan_tci = htons(VLAN_CFI);
2629 case OFPACT_SET_ETH_SRC:
2630 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2631 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2634 case OFPACT_SET_ETH_DST:
2635 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2636 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2639 case OFPACT_SET_IPV4_SRC:
2640 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2641 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2642 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2646 case OFPACT_SET_IPV4_DST:
2647 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2648 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2649 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2653 case OFPACT_SET_IP_DSCP:
2654 if (is_ip_any(flow)) {
2655 wc->masks.nw_tos |= IP_DSCP_MASK;
2656 flow->nw_tos &= ~IP_DSCP_MASK;
2657 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2661 case OFPACT_SET_IP_ECN:
2662 if (is_ip_any(flow)) {
2663 wc->masks.nw_tos |= IP_ECN_MASK;
2664 flow->nw_tos &= ~IP_ECN_MASK;
2665 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2669 case OFPACT_SET_IP_TTL:
2670 if (is_ip_any(flow)) {
2671 wc->masks.nw_ttl = 0xff;
2672 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2676 case OFPACT_SET_L4_SRC_PORT:
2677 if (is_ip_any(flow)) {
2678 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2679 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2680 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2684 case OFPACT_SET_L4_DST_PORT:
2685 if (is_ip_any(flow)) {
2686 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2687 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2688 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2692 case OFPACT_RESUBMIT:
2693 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2696 case OFPACT_SET_TUNNEL:
2697 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2700 case OFPACT_SET_QUEUE:
2701 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2704 case OFPACT_POP_QUEUE:
2705 flow->skb_priority = ctx->orig_skb_priority;
2708 case OFPACT_REG_MOVE:
2709 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2712 case OFPACT_REG_LOAD:
2713 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2716 case OFPACT_SET_FIELD:
2717 set_field = ofpact_get_SET_FIELD(a);
2718 mf = set_field->field;
2719 mf_mask_field_and_prereqs(mf, &wc->masks);
2721 /* Set field action only ever overwrites packet's outermost
2722 * applicable header fields. Do nothing if no header exists. */
2723 if ((mf->id != MFF_VLAN_VID || flow->vlan_tci & htons(VLAN_CFI))
2724 && ((mf->id != MFF_MPLS_LABEL && mf->id != MFF_MPLS_TC)
2725 || flow->mpls_lse)) {
2726 mf_set_flow_value(mf, &set_field->value, flow);
2730 case OFPACT_STACK_PUSH:
2731 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2735 case OFPACT_STACK_POP:
2736 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2740 case OFPACT_PUSH_MPLS:
2741 if (compose_mpls_push_action(ctx,
2742 ofpact_get_PUSH_MPLS(a)->ethertype)) {
2747 case OFPACT_POP_MPLS:
2748 if (compose_mpls_pop_action(ctx,
2749 ofpact_get_POP_MPLS(a)->ethertype)) {
2754 case OFPACT_SET_MPLS_LABEL:
2755 if (compose_set_mpls_label_action(ctx,
2756 ofpact_get_SET_MPLS_LABEL(a)->label)) {
2761 case OFPACT_SET_MPLS_TC:
2762 if (compose_set_mpls_tc_action(ctx,
2763 ofpact_get_SET_MPLS_TC(a)->tc)) {
2768 case OFPACT_SET_MPLS_TTL:
2769 if (compose_set_mpls_ttl_action(ctx,
2770 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2775 case OFPACT_DEC_MPLS_TTL:
2776 if (compose_dec_mpls_ttl_action(ctx)) {
2781 case OFPACT_DEC_TTL:
2782 wc->masks.nw_ttl = 0xff;
2783 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2789 /* Nothing to do. */
2792 case OFPACT_MULTIPATH:
2793 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2797 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2800 case OFPACT_OUTPUT_REG:
2801 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2805 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2812 case OFPACT_FIN_TIMEOUT:
2813 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2814 ctx->xout->has_fin_timeout = true;
2815 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2818 case OFPACT_CLEAR_ACTIONS:
2819 ofpbuf_clear(&ctx->action_set);
2822 case OFPACT_WRITE_ACTIONS:
2823 xlate_write_actions(ctx, a);
2826 case OFPACT_WRITE_METADATA:
2827 metadata = ofpact_get_WRITE_METADATA(a);
2828 flow->metadata &= ~metadata->mask;
2829 flow->metadata |= metadata->metadata & metadata->mask;
2833 /* Not implemented yet. */
2836 case OFPACT_GOTO_TABLE: {
2837 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2839 ovs_assert(ctx->table_id < ogt->table_id);
2840 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2841 ogt->table_id, true);
2846 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2853 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2854 const struct flow *flow, struct rule_dpif *rule,
2855 uint16_t tcp_flags, const struct ofpbuf *packet)
2857 xin->ofproto = ofproto;
2859 xin->packet = packet;
2860 xin->may_learn = packet != NULL;
2862 xin->ofpacts = NULL;
2863 xin->ofpacts_len = 0;
2864 xin->tcp_flags = tcp_flags;
2865 xin->resubmit_hook = NULL;
2866 xin->report_hook = NULL;
2867 xin->resubmit_stats = NULL;
2871 xlate_out_uninit(struct xlate_out *xout)
2874 ofpbuf_uninit(&xout->odp_actions);
2878 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2879 * into datapath actions, using 'ctx', and discards the datapath actions. */
2881 xlate_actions_for_side_effects(struct xlate_in *xin)
2883 struct xlate_out xout;
2885 xlate_actions(xin, &xout);
2886 xlate_out_uninit(&xout);
2890 xlate_report(struct xlate_ctx *ctx, const char *s)
2892 if (ctx->xin->report_hook) {
2893 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2898 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2901 dst->slow = src->slow;
2902 dst->has_learn = src->has_learn;
2903 dst->has_normal = src->has_normal;
2904 dst->has_fin_timeout = src->has_fin_timeout;
2905 dst->nf_output_iface = src->nf_output_iface;
2906 dst->mirrors = src->mirrors;
2908 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2909 sizeof dst->odp_actions_stub);
2910 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2911 src->odp_actions.size);
2914 /* Returns a reference to the sflow handled associated with ofproto, or NULL if
2915 * there is none. The caller is responsible for decrementing the results ref
2916 * count with dpif_sflow_unref(). */
2918 xlate_get_sflow(const struct ofproto_dpif *ofproto)
2920 struct dpif_sflow *sflow = NULL;
2921 struct xbridge *xbridge;
2923 ovs_rwlock_rdlock(&xlate_rwlock);
2924 xbridge = xbridge_lookup(ofproto);
2926 sflow = dpif_sflow_ref(xbridge->sflow);
2928 ovs_rwlock_unlock(&xlate_rwlock);
2933 /* Returns a reference to the ipfix handled associated with ofproto, or NULL if
2934 * there is none. The caller is responsible for decrementing the results ref
2935 * count with dpif_ipfix_unref(). */
2937 xlate_get_ipfix(const struct ofproto_dpif *ofproto)
2939 struct dpif_ipfix *ipfix = NULL;
2940 struct xbridge *xbridge;
2942 ovs_rwlock_rdlock(&xlate_rwlock);
2943 xbridge = xbridge_lookup(ofproto);
2945 ipfix = dpif_ipfix_ref(xbridge->ipfix);
2947 ovs_rwlock_unlock(&xlate_rwlock);
2952 static struct skb_priority_to_dscp *
2953 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2955 struct skb_priority_to_dscp *pdscp;
2958 hash = hash_int(skb_priority, 0);
2959 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2960 if (pdscp->skb_priority == skb_priority) {
2968 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2971 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2972 *dscp = pdscp ? pdscp->dscp : 0;
2973 return pdscp != NULL;
2977 clear_skb_priorities(struct xport *xport)
2979 struct skb_priority_to_dscp *pdscp, *next;
2981 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2982 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2988 actions_output_to_local_port(const struct xlate_ctx *ctx)
2990 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2991 const struct nlattr *a;
2994 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2995 ctx->xout->odp_actions.size) {
2996 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2997 && nl_attr_get_odp_port(a) == local_odp_port) {
3004 /* Thread safe call to xlate_actions__(). */
3006 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
3008 ovs_rwlock_rdlock(&xlate_rwlock);
3009 xlate_actions__(xin, xout);
3010 ovs_rwlock_unlock(&xlate_rwlock);
3013 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
3014 * into datapath actions in 'odp_actions', using 'ctx'.
3016 * The caller must take responsibility for eventually freeing 'xout', with
3017 * xlate_out_uninit(). */
3019 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
3020 OVS_REQ_RDLOCK(xlate_rwlock)
3022 struct flow_wildcards *wc = &xout->wc;
3023 struct flow *flow = &xin->flow;
3024 struct rule_dpif *rule = NULL;
3026 struct rule_actions *actions = NULL;
3027 enum slow_path_reason special;
3028 const struct ofpact *ofpacts;
3029 struct xport *in_port;
3030 struct flow orig_flow;
3031 struct xlate_ctx ctx;
3035 COVERAGE_INC(xlate_actions);
3037 /* Flow initialization rules:
3038 * - 'base_flow' must match the kernel's view of the packet at the
3039 * time that action processing starts. 'flow' represents any
3040 * transformations we wish to make through actions.
3041 * - By default 'base_flow' and 'flow' are the same since the input
3042 * packet matches the output before any actions are applied.
3043 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3044 * of the received packet as seen by the kernel. If we later output
3045 * to another device without any modifications this will cause us to
3046 * insert a new tag since the original one was stripped off by the
3048 * - Tunnel metadata as received is retained in 'flow'. This allows
3049 * tunnel metadata matching also in later tables.
3050 * Since a kernel action for setting the tunnel metadata will only be
3051 * generated with actual tunnel output, changing the tunnel metadata
3052 * values in 'flow' (such as tun_id) will only have effect with a later
3053 * tunnel output action.
3054 * - Tunnel 'base_flow' is completely cleared since that is what the
3055 * kernel does. If we wish to maintain the original values an action
3056 * needs to be generated. */
3061 ctx.xout->has_learn = false;
3062 ctx.xout->has_normal = false;
3063 ctx.xout->has_fin_timeout = false;
3064 ctx.xout->nf_output_iface = NF_OUT_DROP;
3065 ctx.xout->mirrors = 0;
3066 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3067 sizeof ctx.xout->odp_actions_stub);
3068 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3070 ctx.xbridge = xbridge_lookup(xin->ofproto);
3075 ctx.rule = xin->rule;
3077 ctx.base_flow = *flow;
3078 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3079 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3081 flow_wildcards_init_catchall(wc);
3082 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3083 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3084 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3085 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3087 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3088 if (ctx.xbridge->netflow) {
3089 netflow_mask_wc(flow, wc);
3094 ctx.orig_skb_priority = flow->skb_priority;
3097 ctx.mpls_depth_delta = 0;
3099 if (!xin->ofpacts && !ctx.rule) {
3100 rule_dpif_lookup(ctx.xbridge->ofproto, flow, wc, &rule);
3101 if (ctx.xin->resubmit_stats) {
3102 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3106 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3109 ofpacts = xin->ofpacts;
3110 ofpacts_len = xin->ofpacts_len;
3111 } else if (ctx.rule) {
3112 actions = rule_dpif_get_actions(ctx.rule);
3113 ofpacts = actions->ofpacts;
3114 ofpacts_len = actions->ofpacts_len;
3119 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3120 ofpbuf_use_stub(&ctx.action_set,
3121 ctx.action_set_stub, sizeof ctx.action_set_stub);
3123 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3124 /* Do this conditionally because the copy is expensive enough that it
3125 * shows up in profiles. */
3129 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3130 switch (ctx.xbridge->frag) {
3131 case OFPC_FRAG_NORMAL:
3132 /* We must pretend that transport ports are unavailable. */
3133 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3134 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3137 case OFPC_FRAG_DROP:
3140 case OFPC_FRAG_REASM:
3143 case OFPC_FRAG_NX_MATCH:
3144 /* Nothing to do. */
3147 case OFPC_INVALID_TTL_TO_CONTROLLER:
3152 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3153 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3155 ctx.xout->slow |= special;
3157 size_t sample_actions_len;
3159 if (flow->in_port.ofp_port
3160 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3161 flow->in_port.ofp_port,
3163 ctx.base_flow.vlan_tci = 0;
3166 add_sflow_action(&ctx);
3167 add_ipfix_action(&ctx);
3168 sample_actions_len = ctx.xout->odp_actions.size;
3170 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3171 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3173 /* We've let OFPP_NORMAL and the learning action look at the
3174 * packet, so drop it now if forwarding is disabled. */
3175 if (in_port && !xport_stp_forward_state(in_port)) {
3176 ctx.xout->odp_actions.size = sample_actions_len;
3180 if (ctx.action_set.size) {
3181 xlate_action_set(&ctx);
3184 if (ctx.xbridge->has_in_band
3185 && in_band_must_output_to_local_port(flow)
3186 && !actions_output_to_local_port(&ctx)) {
3187 compose_output_action(&ctx, OFPP_LOCAL);
3190 fix_sflow_action(&ctx);
3192 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3193 add_mirror_actions(&ctx, &orig_flow);
3197 if (nl_attr_oversized(ctx.xout->odp_actions.size)) {
3198 /* These datapath actions are too big for a Netlink attribute, so we
3199 * can't hand them to the kernel directly. dpif_execute() can execute
3200 * them one by one with help, so just mark the result as SLOW_ACTION to
3201 * prevent the flow from being installed. */
3202 COVERAGE_INC(xlate_actions_oversize);
3203 ctx.xout->slow |= SLOW_ACTION;
3206 ofpbuf_uninit(&ctx.stack);
3207 ofpbuf_uninit(&ctx.action_set);
3209 /* Clear the metadata and register wildcard masks, because we won't
3210 * use non-header fields as part of the cache. */
3211 flow_wildcards_clear_non_packet_fields(wc);
3214 rule_actions_unref(actions);
3215 rule_dpif_unref(rule);
3218 /* Sends 'packet' out 'ofport'.
3219 * May modify 'packet'.
3220 * Returns 0 if successful, otherwise a positive errno value. */
3222 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3224 struct xport *xport;
3225 struct ofpact_output output;
3227 union flow_in_port in_port_;
3230 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3231 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3232 in_port_.ofp_port = OFPP_NONE;
3233 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
3235 ovs_rwlock_rdlock(&xlate_rwlock);
3236 xport = xport_lookup(ofport);
3238 ovs_rwlock_unlock(&xlate_rwlock);
3241 output.port = xport->ofp_port;
3243 error = ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3244 &output.ofpact, sizeof output,
3246 ovs_rwlock_unlock(&xlate_rwlock);