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, 'odp_in_port' with the datapath in_port, that
518 * 'packet' ingressed, and 'ipfix', 'sflow', and 'netflow' with the appropriate
519 * handles for those protocols if they're enabled. Caller is responsible for
522 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
523 * 'flow''s in_port to OFPP_NONE.
525 * This function does post-processing on data returned from
526 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
527 * of the upcall processing logic. In particular, if the extracted in_port is
528 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
529 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
530 * a VLAN header onto 'packet' (if it is nonnull).
532 * Similarly, this function also includes some logic to help with tunnels. It
533 * may modify 'flow' as necessary to make the tunneling implementation
534 * transparent to the upcall processing logic.
536 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
537 * or some other positive errno if there are other problems. */
539 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
540 const struct nlattr *key, size_t key_len,
541 struct flow *flow, enum odp_key_fitness *fitnessp,
542 struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
543 struct dpif_sflow **sflow, struct netflow **netflow,
544 odp_port_t *odp_in_port)
546 enum odp_key_fitness fitness;
547 const struct xport *xport;
550 ovs_rwlock_rdlock(&xlate_rwlock);
551 fitness = odp_flow_key_to_flow(key, key_len, flow);
552 if (fitness == ODP_FIT_ERROR) {
558 *odp_in_port = flow->in_port.odp_port;
561 xport = xport_lookup(tnl_port_should_receive(flow)
562 ? tnl_port_receive(flow)
563 : odp_port_to_ofport(backer, flow->in_port.odp_port));
565 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
570 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
572 /* Make the packet resemble the flow, so that it gets sent to
573 * an OpenFlow controller properly, so that it looks correct
574 * for sFlow, and so that flow_extract() will get the correct
575 * vlan_tci if it is called on 'packet'. */
576 eth_push_vlan(packet, flow->vlan_tci);
578 /* We can't reproduce 'key' from 'flow'. */
579 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
584 *ofproto = xport->xbridge->ofproto;
588 *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
592 *sflow = dpif_sflow_ref(xport->xbridge->sflow);
596 *netflow = netflow_ref(xport->xbridge->netflow);
603 ovs_rwlock_unlock(&xlate_rwlock);
607 static struct xbridge *
608 xbridge_lookup(const struct ofproto_dpif *ofproto)
610 struct xbridge *xbridge;
616 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
618 if (xbridge->ofproto == ofproto) {
625 static struct xbundle *
626 xbundle_lookup(const struct ofbundle *ofbundle)
628 struct xbundle *xbundle;
634 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
636 if (xbundle->ofbundle == ofbundle) {
643 static struct xport *
644 xport_lookup(const struct ofport_dpif *ofport)
652 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
654 if (xport->ofport == ofport) {
661 static struct stp_port *
662 xport_get_stp_port(const struct xport *xport)
664 return xport->xbridge->stp && xport->stp_port_no != -1
665 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
669 static enum stp_state
670 xport_stp_learn_state(const struct xport *xport)
672 struct stp_port *sp = xport_get_stp_port(xport);
673 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
677 xport_stp_forward_state(const struct xport *xport)
679 struct stp_port *sp = xport_get_stp_port(xport);
680 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
683 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
684 * were used to make the determination.*/
686 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
688 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
689 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
693 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
695 struct stp_port *sp = xport_get_stp_port(xport);
696 struct ofpbuf payload = *packet;
697 struct eth_header *eth = payload.data;
699 /* Sink packets on ports that have STP disabled when the bridge has
701 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
705 /* Trim off padding on payload. */
706 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
707 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
710 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
711 stp_received_bpdu(sp, payload.data, payload.size);
715 static struct xport *
716 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
720 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
722 if (xport->ofp_port == ofp_port) {
730 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
732 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
733 return xport ? xport->odp_port : ODPP_NONE;
737 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
741 xport = get_ofp_port(ctx->xbridge, ofp_port);
742 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
743 xport->state & OFPUTIL_PS_LINK_DOWN) {
750 static const struct ofputil_bucket *
751 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
755 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
757 struct group_dpif *group;
760 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
765 hit = group_first_live_bucket(ctx, group, depth) != NULL;
767 group_dpif_release(group);
771 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
774 bucket_is_alive(const struct xlate_ctx *ctx,
775 const struct ofputil_bucket *bucket, int depth)
777 if (depth >= MAX_LIVENESS_RECURSION) {
778 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
780 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
781 MAX_LIVENESS_RECURSION);
785 return !ofputil_bucket_has_liveness(bucket) ||
786 (bucket->watch_port != OFPP_ANY &&
787 odp_port_is_alive(ctx, bucket->watch_port)) ||
788 (bucket->watch_group != OFPG_ANY &&
789 group_is_alive(ctx, bucket->watch_group, depth + 1));
792 static const struct ofputil_bucket *
793 group_first_live_bucket(const struct xlate_ctx *ctx,
794 const struct group_dpif *group, int depth)
796 struct ofputil_bucket *bucket;
797 const struct list *buckets;
799 group_dpif_get_buckets(group, &buckets);
800 LIST_FOR_EACH (bucket, list_node, buckets) {
801 if (bucket_is_alive(ctx, bucket, depth)) {
809 static const struct ofputil_bucket *
810 group_best_live_bucket(const struct xlate_ctx *ctx,
811 const struct group_dpif *group,
814 const struct ofputil_bucket *best_bucket = NULL;
815 uint32_t best_score = 0;
818 const struct ofputil_bucket *bucket;
819 const struct list *buckets;
821 group_dpif_get_buckets(group, &buckets);
822 LIST_FOR_EACH (bucket, list_node, buckets) {
823 if (bucket_is_alive(ctx, bucket, 0)) {
824 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
825 if (score >= best_score) {
826 best_bucket = bucket;
837 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
839 return (bundle->vlan_mode != PORT_VLAN_ACCESS
840 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
844 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
846 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
850 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
852 return xbundle != &ofpp_none_bundle
853 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
858 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
860 return xbundle != &ofpp_none_bundle
861 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
866 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
868 return xbundle != &ofpp_none_bundle
869 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
873 static struct xbundle *
874 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
875 bool warn, struct xport **in_xportp)
879 /* Find the port and bundle for the received packet. */
880 xport = get_ofp_port(xbridge, in_port);
884 if (xport && xport->xbundle) {
885 return xport->xbundle;
888 /* Special-case OFPP_NONE, which a controller may use as the ingress
889 * port for traffic that it is sourcing. */
890 if (in_port == OFPP_NONE) {
891 ofpp_none_bundle.name = "OFPP_NONE";
892 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
893 return &ofpp_none_bundle;
896 /* Odd. A few possible reasons here:
898 * - We deleted a port but there are still a few packets queued up
901 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
902 * we don't know about.
904 * - The ofproto client didn't configure the port as part of a bundle.
905 * This is particularly likely to happen if a packet was received on the
906 * port after it was created, but before the client had a chance to
907 * configure its bundle.
910 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
912 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
913 "port %"PRIu16, xbridge->name, in_port);
919 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
921 const struct xbridge *xbridge = ctx->xbridge;
922 mirror_mask_t mirrors;
923 struct xbundle *in_xbundle;
927 mirrors = ctx->xout->mirrors;
928 ctx->xout->mirrors = 0;
930 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
931 ctx->xin->packet != NULL, NULL);
935 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
937 /* Drop frames on bundles reserved for mirroring. */
938 if (xbundle_mirror_out(xbridge, in_xbundle)) {
939 if (ctx->xin->packet != NULL) {
940 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
941 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
942 "%s, which is reserved exclusively for mirroring",
943 ctx->xbridge->name, in_xbundle->name);
945 ofpbuf_clear(&ctx->xout->odp_actions);
950 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
951 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
954 vlan = input_vid_to_vlan(in_xbundle, vid);
960 /* Restore the original packet before adding the mirror actions. */
961 ctx->xin->flow = *orig_flow;
964 mirror_mask_t dup_mirrors;
965 struct ofbundle *out;
966 unsigned long *vlans;
971 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
972 &vlans, &dup_mirrors, &out, &out_vlan);
973 ovs_assert(has_mirror);
976 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
978 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
981 if (!vlan_mirrored) {
982 mirrors = zero_rightmost_1bit(mirrors);
986 mirrors &= ~dup_mirrors;
987 ctx->xout->mirrors |= dup_mirrors;
989 struct xbundle *out_xbundle = xbundle_lookup(out);
991 output_normal(ctx, out_xbundle, vlan);
993 } else if (vlan != out_vlan
994 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
995 struct xbundle *xbundle;
997 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
998 if (xbundle_includes_vlan(xbundle, out_vlan)
999 && !xbundle_mirror_out(xbridge, xbundle)) {
1000 output_normal(ctx, xbundle, out_vlan);
1007 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1008 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1009 * the bundle on which the packet was received, returns the VLAN to which the
1012 * Both 'vid' and the return value are in the range 0...4095. */
1014 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1016 switch (in_xbundle->vlan_mode) {
1017 case PORT_VLAN_ACCESS:
1018 return in_xbundle->vlan;
1021 case PORT_VLAN_TRUNK:
1024 case PORT_VLAN_NATIVE_UNTAGGED:
1025 case PORT_VLAN_NATIVE_TAGGED:
1026 return vid ? vid : in_xbundle->vlan;
1033 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1034 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1037 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1038 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1041 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1043 /* Allow any VID on the OFPP_NONE port. */
1044 if (in_xbundle == &ofpp_none_bundle) {
1048 switch (in_xbundle->vlan_mode) {
1049 case PORT_VLAN_ACCESS:
1052 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1053 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1054 "packet received on port %s configured as VLAN "
1055 "%"PRIu16" access port", vid, in_xbundle->name,
1062 case PORT_VLAN_NATIVE_UNTAGGED:
1063 case PORT_VLAN_NATIVE_TAGGED:
1065 /* Port must always carry its native VLAN. */
1069 case PORT_VLAN_TRUNK:
1070 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1073 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1074 "received on port %s not configured for trunking "
1075 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1087 /* Given 'vlan', the VLAN that a packet belongs to, and
1088 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1089 * that should be included in the 802.1Q header. (If the return value is 0,
1090 * then the 802.1Q header should only be included in the packet if there is a
1093 * Both 'vlan' and the return value are in the range 0...4095. */
1095 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1097 switch (out_xbundle->vlan_mode) {
1098 case PORT_VLAN_ACCESS:
1101 case PORT_VLAN_TRUNK:
1102 case PORT_VLAN_NATIVE_TAGGED:
1105 case PORT_VLAN_NATIVE_UNTAGGED:
1106 return vlan == out_xbundle->vlan ? 0 : vlan;
1114 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1117 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1119 ovs_be16 tci, old_tci;
1120 struct xport *xport;
1122 vid = output_vlan_to_vid(out_xbundle, vlan);
1123 if (list_is_empty(&out_xbundle->xports)) {
1124 /* Partially configured bundle with no slaves. Drop the packet. */
1126 } else if (!out_xbundle->bond) {
1127 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1130 struct ofport_dpif *ofport;
1132 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1133 &ctx->xout->wc, vid);
1134 xport = xport_lookup(ofport);
1137 /* No slaves enabled, so drop packet. */
1141 if (ctx->xin->resubmit_stats) {
1142 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1143 ctx->xin->resubmit_stats->n_bytes);
1147 old_tci = *flow_tci;
1149 if (tci || out_xbundle->use_priority_tags) {
1150 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1152 tci |= htons(VLAN_CFI);
1157 compose_output_action(ctx, xport->ofp_port);
1158 *flow_tci = old_tci;
1161 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1162 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1163 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1165 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1167 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1171 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1172 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1176 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1177 if (flow->nw_proto == ARP_OP_REPLY) {
1179 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1180 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1181 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1183 return flow->nw_src == flow->nw_dst;
1189 /* Checks whether a MAC learning update is necessary for MAC learning table
1190 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1193 * Most packets processed through the MAC learning table do not actually
1194 * change it in any way. This function requires only a read lock on the MAC
1195 * learning table, so it is much cheaper in this common case.
1197 * Keep the code here synchronized with that in update_learning_table__()
1200 is_mac_learning_update_needed(const struct mac_learning *ml,
1201 const struct flow *flow,
1202 struct flow_wildcards *wc,
1203 int vlan, struct xbundle *in_xbundle)
1204 OVS_REQ_RDLOCK(ml->rwlock)
1206 struct mac_entry *mac;
1208 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1212 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1213 if (!mac || mac_entry_age(ml, mac)) {
1217 if (is_gratuitous_arp(flow, wc)) {
1218 /* We don't want to learn from gratuitous ARP packets that are
1219 * reflected back over bond slaves so we lock the learning table. */
1220 if (!in_xbundle->bond) {
1222 } else if (mac_entry_is_grat_arp_locked(mac)) {
1227 return mac->port.p != in_xbundle->ofbundle;
1231 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1232 * received on 'in_xbundle' in 'vlan'.
1234 * This code repeats all the checks in is_mac_learning_update_needed() because
1235 * the lock was released between there and here and thus the MAC learning state
1236 * could have changed.
1238 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1241 update_learning_table__(const struct xbridge *xbridge,
1242 const struct flow *flow, struct flow_wildcards *wc,
1243 int vlan, struct xbundle *in_xbundle)
1244 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1246 struct mac_entry *mac;
1248 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1252 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1253 if (is_gratuitous_arp(flow, wc)) {
1254 /* We don't want to learn from gratuitous ARP packets that are
1255 * reflected back over bond slaves so we lock the learning table. */
1256 if (!in_xbundle->bond) {
1257 mac_entry_set_grat_arp_lock(mac);
1258 } else if (mac_entry_is_grat_arp_locked(mac)) {
1263 if (mac->port.p != in_xbundle->ofbundle) {
1264 /* The log messages here could actually be useful in debugging,
1265 * so keep the rate limit relatively high. */
1266 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1268 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1269 "on port %s in VLAN %d",
1270 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1271 in_xbundle->name, vlan);
1273 mac->port.p = in_xbundle->ofbundle;
1274 mac_learning_changed(xbridge->ml);
1279 update_learning_table(const struct xbridge *xbridge,
1280 const struct flow *flow, struct flow_wildcards *wc,
1281 int vlan, struct xbundle *in_xbundle)
1285 /* Don't learn the OFPP_NONE port. */
1286 if (in_xbundle == &ofpp_none_bundle) {
1290 /* First try the common case: no change to MAC learning table. */
1291 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1292 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1294 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1297 /* Slow path: MAC learning table might need an update. */
1298 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1299 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1300 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1304 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1305 * dropped. Returns true if they may be forwarded, false if they should be
1308 * 'in_port' must be the xport that corresponds to flow->in_port.
1309 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1311 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1312 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1313 * checked by input_vid_is_valid().
1315 * May also add tags to '*tags', although the current implementation only does
1316 * so in one special case.
1319 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1322 struct xbundle *in_xbundle = in_port->xbundle;
1323 const struct xbridge *xbridge = ctx->xbridge;
1324 struct flow *flow = &ctx->xin->flow;
1326 /* Drop frames for reserved multicast addresses
1327 * only if forward_bpdu option is absent. */
1328 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1329 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1333 if (in_xbundle->bond) {
1334 struct mac_entry *mac;
1336 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1342 xlate_report(ctx, "bonding refused admissibility, dropping");
1345 case BV_DROP_IF_MOVED:
1346 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1347 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1348 if (mac && mac->port.p != in_xbundle->ofbundle &&
1349 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1350 || mac_entry_is_grat_arp_locked(mac))) {
1351 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1352 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1356 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1365 xlate_normal(struct xlate_ctx *ctx)
1367 struct flow_wildcards *wc = &ctx->xout->wc;
1368 struct flow *flow = &ctx->xin->flow;
1369 struct xbundle *in_xbundle;
1370 struct xport *in_port;
1371 struct mac_entry *mac;
1376 ctx->xout->has_normal = true;
1378 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1379 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1380 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1382 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1383 ctx->xin->packet != NULL, &in_port);
1385 xlate_report(ctx, "no input bundle, dropping");
1389 /* Drop malformed frames. */
1390 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1391 !(flow->vlan_tci & htons(VLAN_CFI))) {
1392 if (ctx->xin->packet != NULL) {
1393 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1394 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1395 "VLAN tag received on port %s",
1396 ctx->xbridge->name, in_xbundle->name);
1398 xlate_report(ctx, "partial VLAN tag, dropping");
1402 /* Drop frames on bundles reserved for mirroring. */
1403 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1404 if (ctx->xin->packet != NULL) {
1405 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1406 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1407 "%s, which is reserved exclusively for mirroring",
1408 ctx->xbridge->name, in_xbundle->name);
1410 xlate_report(ctx, "input port is mirror output port, dropping");
1415 vid = vlan_tci_to_vid(flow->vlan_tci);
1416 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1417 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1420 vlan = input_vid_to_vlan(in_xbundle, vid);
1422 /* Check other admissibility requirements. */
1423 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1427 /* Learn source MAC. */
1428 if (ctx->xin->may_learn) {
1429 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1432 /* Determine output bundle. */
1433 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1434 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1435 mac_port = mac ? mac->port.p : NULL;
1436 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1439 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1440 if (mac_xbundle && mac_xbundle != in_xbundle) {
1441 xlate_report(ctx, "forwarding to learned port");
1442 output_normal(ctx, mac_xbundle, vlan);
1443 } else if (!mac_xbundle) {
1444 xlate_report(ctx, "learned port is unknown, dropping");
1446 xlate_report(ctx, "learned port is input port, dropping");
1449 struct xbundle *xbundle;
1451 xlate_report(ctx, "no learned MAC for destination, flooding");
1452 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1453 if (xbundle != in_xbundle
1454 && xbundle_includes_vlan(xbundle, vlan)
1455 && xbundle->floodable
1456 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1457 output_normal(ctx, xbundle, vlan);
1460 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1464 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1465 * the number of packets out of UINT32_MAX to sample. The given
1466 * cookie is passed back in the callback for each sampled packet.
1469 compose_sample_action(const struct xbridge *xbridge,
1470 struct ofpbuf *odp_actions,
1471 const struct flow *flow,
1472 const uint32_t probability,
1473 const union user_action_cookie *cookie,
1474 const size_t cookie_size)
1476 size_t sample_offset, actions_offset;
1477 odp_port_t odp_port;
1481 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1483 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1485 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1487 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1488 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1489 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1491 nl_msg_end_nested(odp_actions, actions_offset);
1492 nl_msg_end_nested(odp_actions, sample_offset);
1493 return cookie_offset;
1497 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1498 odp_port_t odp_port, unsigned int n_outputs,
1499 union user_action_cookie *cookie)
1503 cookie->type = USER_ACTION_COOKIE_SFLOW;
1504 cookie->sflow.vlan_tci = vlan_tci;
1506 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1507 * port information") for the interpretation of cookie->output. */
1508 switch (n_outputs) {
1510 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1511 cookie->sflow.output = 0x40000000 | 256;
1515 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1517 cookie->sflow.output = ifindex;
1522 /* 0x80000000 means "multiple output ports. */
1523 cookie->sflow.output = 0x80000000 | n_outputs;
1528 /* Compose SAMPLE action for sFlow bridge sampling. */
1530 compose_sflow_action(const struct xbridge *xbridge,
1531 struct ofpbuf *odp_actions,
1532 const struct flow *flow,
1533 odp_port_t odp_port)
1535 uint32_t probability;
1536 union user_action_cookie cookie;
1538 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1542 probability = dpif_sflow_get_probability(xbridge->sflow);
1543 compose_sflow_cookie(xbridge, htons(0), odp_port,
1544 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1546 return compose_sample_action(xbridge, odp_actions, flow, probability,
1547 &cookie, sizeof cookie.sflow);
1551 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1552 uint32_t obs_domain_id, uint32_t obs_point_id,
1553 union user_action_cookie *cookie)
1555 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1556 cookie->flow_sample.probability = probability;
1557 cookie->flow_sample.collector_set_id = collector_set_id;
1558 cookie->flow_sample.obs_domain_id = obs_domain_id;
1559 cookie->flow_sample.obs_point_id = obs_point_id;
1563 compose_ipfix_cookie(union user_action_cookie *cookie)
1565 cookie->type = USER_ACTION_COOKIE_IPFIX;
1568 /* Compose SAMPLE action for IPFIX bridge sampling. */
1570 compose_ipfix_action(const struct xbridge *xbridge,
1571 struct ofpbuf *odp_actions,
1572 const struct flow *flow)
1574 uint32_t probability;
1575 union user_action_cookie cookie;
1577 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1581 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1582 compose_ipfix_cookie(&cookie);
1584 compose_sample_action(xbridge, odp_actions, flow, probability,
1585 &cookie, sizeof cookie.ipfix);
1588 /* SAMPLE action for sFlow must be first action in any given list of
1589 * actions. At this point we do not have all information required to
1590 * build it. So try to build sample action as complete as possible. */
1592 add_sflow_action(struct xlate_ctx *ctx)
1594 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1595 &ctx->xout->odp_actions,
1596 &ctx->xin->flow, ODPP_NONE);
1597 ctx->sflow_odp_port = 0;
1598 ctx->sflow_n_outputs = 0;
1601 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1602 * of actions, eventually after the SAMPLE action for sFlow. */
1604 add_ipfix_action(struct xlate_ctx *ctx)
1606 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1610 /* Fix SAMPLE action according to data collected while composing ODP actions.
1611 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1612 * USERSPACE action's user-cookie which is required for sflow. */
1614 fix_sflow_action(struct xlate_ctx *ctx)
1616 const struct flow *base = &ctx->base_flow;
1617 union user_action_cookie *cookie;
1619 if (!ctx->user_cookie_offset) {
1623 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1624 sizeof cookie->sflow);
1625 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1627 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1628 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1631 static enum slow_path_reason
1632 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1633 const struct xport *xport, const struct ofpbuf *packet)
1635 struct flow_wildcards *wc = &ctx->xout->wc;
1636 const struct xbridge *xbridge = ctx->xbridge;
1640 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1642 cfm_process_heartbeat(xport->cfm, packet);
1645 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1647 bfd_process_packet(xport->bfd, flow, packet);
1650 } else if (xport->xbundle && xport->xbundle->lacp
1651 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1653 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1656 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1658 stp_process_packet(xport, packet);
1667 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1670 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1671 struct flow_wildcards *wc = &ctx->xout->wc;
1672 struct flow *flow = &ctx->xin->flow;
1673 ovs_be16 flow_vlan_tci;
1674 uint32_t flow_pkt_mark;
1675 uint8_t flow_nw_tos;
1676 odp_port_t out_port, odp_port;
1679 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1680 * before traversing a patch port. */
1681 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 23);
1684 xlate_report(ctx, "Nonexistent output port");
1686 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1687 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1689 } else if (check_stp && !xport_stp_forward_state(xport)) {
1690 xlate_report(ctx, "STP not in forwarding state, skipping output");
1694 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1695 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1700 const struct xport *peer = xport->peer;
1701 struct flow old_flow = ctx->xin->flow;
1702 enum slow_path_reason special;
1704 ctx->xbridge = peer->xbridge;
1705 flow->in_port.ofp_port = peer->ofp_port;
1706 flow->metadata = htonll(0);
1707 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1708 memset(flow->regs, 0, sizeof flow->regs);
1710 special = process_special(ctx, &ctx->xin->flow, peer,
1713 ctx->xout->slow |= special;
1714 } else if (may_receive(peer, ctx)) {
1715 if (xport_stp_forward_state(peer)) {
1716 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1718 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1719 * learning action look at the packet, then drop it. */
1720 struct flow old_base_flow = ctx->base_flow;
1721 size_t old_size = ctx->xout->odp_actions.size;
1722 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1723 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1724 ctx->xout->mirrors = old_mirrors;
1725 ctx->base_flow = old_base_flow;
1726 ctx->xout->odp_actions.size = old_size;
1730 ctx->xin->flow = old_flow;
1731 ctx->xbridge = xport->xbridge;
1733 if (ctx->xin->resubmit_stats) {
1734 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1735 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1737 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1744 flow_vlan_tci = flow->vlan_tci;
1745 flow_pkt_mark = flow->pkt_mark;
1746 flow_nw_tos = flow->nw_tos;
1748 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1749 wc->masks.nw_tos |= IP_ECN_MASK;
1750 flow->nw_tos &= ~IP_DSCP_MASK;
1751 flow->nw_tos |= dscp;
1754 if (xport->is_tunnel) {
1755 /* Save tunnel metadata so that changes made due to
1756 * the Logical (tunnel) Port are not visible for any further
1757 * matches, while explicit set actions on tunnel metadata are.
1759 struct flow_tnl flow_tnl = flow->tunnel;
1760 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1761 if (odp_port == ODPP_NONE) {
1762 xlate_report(ctx, "Tunneling decided against output");
1763 goto out; /* restore flow_nw_tos */
1765 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1766 xlate_report(ctx, "Not tunneling to our own address");
1767 goto out; /* restore flow_nw_tos */
1769 if (ctx->xin->resubmit_stats) {
1770 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1772 out_port = odp_port;
1773 commit_odp_tunnel_action(flow, &ctx->base_flow,
1774 &ctx->xout->odp_actions);
1775 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1777 ofp_port_t vlandev_port;
1779 odp_port = xport->odp_port;
1780 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1781 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1783 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1785 if (vlandev_port == ofp_port) {
1786 out_port = odp_port;
1788 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1789 flow->vlan_tci = htons(0);
1793 if (out_port != ODPP_NONE) {
1794 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1795 &ctx->xout->odp_actions,
1797 &ctx->mpls_depth_delta);
1798 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1801 ctx->sflow_odp_port = odp_port;
1802 ctx->sflow_n_outputs++;
1803 ctx->xout->nf_output_iface = ofp_port;
1808 flow->vlan_tci = flow_vlan_tci;
1809 flow->pkt_mark = flow_pkt_mark;
1810 flow->nw_tos = flow_nw_tos;
1814 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1816 compose_output_action__(ctx, ofp_port, true);
1820 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1822 struct rule_dpif *old_rule = ctx->rule;
1823 struct rule_actions *actions;
1825 if (ctx->xin->resubmit_stats) {
1826 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1832 actions = rule_dpif_get_actions(rule);
1833 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1834 rule_actions_unref(actions);
1835 ctx->rule = old_rule;
1840 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
1842 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1844 if (ctx->recurse >= MAX_RESUBMIT_RECURSION) {
1845 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
1846 MAX_RESUBMIT_RECURSION);
1847 } else if (ctx->resubmits >= MAX_RESUBMITS) {
1848 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
1849 } else if (ctx->xout->odp_actions.size > UINT16_MAX) {
1850 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
1851 } else if (ctx->stack.size >= 65536) {
1852 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
1861 xlate_table_action(struct xlate_ctx *ctx,
1862 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1864 if (xlate_resubmit_resource_check(ctx)) {
1865 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1866 bool skip_wildcards = ctx->xin->skip_wildcards;
1867 uint8_t old_table_id = ctx->table_id;
1868 struct rule_dpif *rule;
1870 ctx->table_id = table_id;
1872 /* Look up a flow with 'in_port' as the input port. Then restore the
1873 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1874 * have surprising behavior). */
1875 ctx->xin->flow.in_port.ofp_port = in_port;
1876 rule_dpif_lookup_in_table(ctx->xbridge->ofproto, &ctx->xin->flow,
1877 !skip_wildcards ? &ctx->xout->wc : NULL,
1879 ctx->xin->flow.in_port.ofp_port = old_in_port;
1881 if (ctx->xin->resubmit_hook) {
1882 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1885 if (!rule && may_packet_in) {
1886 struct xport *xport;
1889 * check if table configuration flags
1890 * OFPTC11_TABLE_MISS_CONTROLLER, default.
1891 * OFPTC11_TABLE_MISS_CONTINUE,
1892 * OFPTC11_TABLE_MISS_DROP
1893 * When OF1.0, OFPTC11_TABLE_MISS_CONTINUE is used. What to do? */
1894 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1895 choose_miss_rule(xport ? xport->config : 0,
1896 ctx->xbridge->miss_rule,
1897 ctx->xbridge->no_packet_in_rule, &rule);
1900 xlate_recursively(ctx, rule);
1901 rule_dpif_unref(rule);
1904 ctx->table_id = old_table_id;
1912 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
1914 uint64_t action_list_stub[1024 / 8];
1915 struct ofpbuf action_list, action_set;
1917 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
1918 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
1920 ofpacts_execute_action_set(&action_list, &action_set);
1922 do_xlate_actions(action_list.data, action_list.size, ctx);
1925 ofpbuf_uninit(&action_set);
1926 ofpbuf_uninit(&action_list);
1930 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
1932 const struct ofputil_bucket *bucket;
1933 const struct list *buckets;
1934 struct flow old_flow = ctx->xin->flow;
1936 group_dpif_get_buckets(group, &buckets);
1938 LIST_FOR_EACH (bucket, list_node, buckets) {
1939 xlate_group_bucket(ctx, bucket);
1940 /* Roll back flow to previous state.
1941 * This is equivalent to cloning the packet for each bucket.
1943 * As a side effect any subsequently applied actions will
1944 * also effectively be applied to a clone of the packet taken
1945 * just before applying the all or indirect group. */
1946 ctx->xin->flow = old_flow;
1951 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
1953 const struct ofputil_bucket *bucket;
1955 bucket = group_first_live_bucket(ctx, group, 0);
1957 xlate_group_bucket(ctx, bucket);
1962 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
1964 struct flow_wildcards *wc = &ctx->xout->wc;
1965 const struct ofputil_bucket *bucket;
1968 basis = hash_bytes(ctx->xin->flow.dl_dst, sizeof ctx->xin->flow.dl_dst, 0);
1969 bucket = group_best_live_bucket(ctx, group, basis);
1971 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1972 xlate_group_bucket(ctx, bucket);
1977 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
1979 switch (group_dpif_get_type(group)) {
1981 case OFPGT11_INDIRECT:
1982 xlate_all_group(ctx, group);
1984 case OFPGT11_SELECT:
1985 xlate_select_group(ctx, group);
1988 xlate_ff_group(ctx, group);
1993 group_dpif_release(group);
1997 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
1999 if (xlate_resubmit_resource_check(ctx)) {
2000 struct group_dpif *group;
2003 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2005 xlate_group_action__(ctx, group);
2015 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2016 const struct ofpact_resubmit *resubmit)
2021 in_port = resubmit->in_port;
2022 if (in_port == OFPP_IN_PORT) {
2023 in_port = ctx->xin->flow.in_port.ofp_port;
2026 table_id = resubmit->table_id;
2027 if (table_id == 255) {
2028 table_id = ctx->table_id;
2031 xlate_table_action(ctx, in_port, table_id, false);
2035 flood_packets(struct xlate_ctx *ctx, bool all)
2037 const struct xport *xport;
2039 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2040 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2045 compose_output_action__(ctx, xport->ofp_port, false);
2046 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2047 compose_output_action(ctx, xport->ofp_port);
2051 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2055 execute_controller_action(struct xlate_ctx *ctx, int len,
2056 enum ofp_packet_in_reason reason,
2057 uint16_t controller_id)
2059 struct ofproto_packet_in *pin;
2060 struct ofpbuf *packet;
2063 ctx->xout->slow |= SLOW_CONTROLLER;
2064 if (!ctx->xin->packet) {
2068 packet = ofpbuf_clone(ctx->xin->packet);
2070 key.skb_priority = 0;
2072 memset(&key.tunnel, 0, sizeof key.tunnel);
2074 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2075 &ctx->xout->odp_actions,
2077 &ctx->mpls_depth_delta);
2079 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
2080 ctx->xout->odp_actions.size, NULL, NULL);
2082 pin = xmalloc(sizeof *pin);
2083 pin->up.packet_len = packet->size;
2084 pin->up.packet = ofpbuf_steal_data(packet);
2085 pin->up.reason = reason;
2086 pin->up.table_id = ctx->table_id;
2087 pin->up.cookie = (ctx->rule
2088 ? rule_dpif_get_flow_cookie(ctx->rule)
2091 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2093 pin->controller_id = controller_id;
2094 pin->send_len = len;
2095 pin->generated_by_table_miss = (ctx->rule
2096 && rule_dpif_is_table_miss(ctx->rule));
2097 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2098 ofpbuf_delete(packet);
2102 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2104 struct flow_wildcards *wc = &ctx->xout->wc;
2105 struct flow *flow = &ctx->xin->flow;
2107 ovs_assert(eth_type_mpls(eth_type));
2109 /* If mpls_depth_delta is negative then an MPLS POP action has been
2110 * composed and the resulting MPLS label stack is unknown. This means
2111 * an MPLS PUSH action can't be composed as it needs to know either the
2112 * top-most MPLS LSE to use as a template for the new MPLS LSE, or that
2113 * there is no MPLS label stack present. Thus, stop processing.
2115 * If mpls_depth_delta is positive then an MPLS PUSH action has been
2116 * composed and no further MPLS PUSH action may be performed without
2117 * losing MPLS LSE and ether type information held in xtx->xin->flow.
2118 * Thus, stop processing.
2120 * If the MPLS LSE of the flow and base_flow differ then the MPLS LSE
2121 * has been updated. Performing a MPLS PUSH action may be would result in
2122 * losing MPLS LSE and ether type information held in xtx->xin->flow.
2123 * Thus, stop processing.
2125 * It is planned that in the future this case will be handled
2126 * by recirculation */
2127 if (ctx->mpls_depth_delta ||
2128 ctx->xin->flow.mpls_lse != ctx->base_flow.mpls_lse) {
2132 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2134 ctx->pre_push_mpls_lse = ctx->xin->flow.mpls_lse;
2136 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2137 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
2142 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2143 label = htonl(0x2); /* IPV6 Explicit Null. */
2145 label = htonl(0x0); /* IPV4 Explicit Null. */
2147 wc->masks.nw_tos |= IP_DSCP_MASK;
2148 wc->masks.nw_ttl = 0xff;
2149 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
2150 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
2151 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
2153 flow->dl_type = eth_type;
2154 ctx->mpls_depth_delta++;
2160 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2162 struct flow_wildcards *wc = &ctx->xout->wc;
2164 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2168 /* If mpls_depth_delta is negative then an MPLS POP action has been
2169 * composed. Performing another MPLS POP action
2170 * would result in losing ether type that results from
2171 * the already composed MPLS POP. Thus, stop processing.
2173 * It is planned that in the future this case will be handled
2174 * by recirculation */
2175 if (ctx->mpls_depth_delta < 0) {
2179 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2181 /* If mpls_depth_delta is positive then an MPLS PUSH action has been
2182 * executed and the previous MPLS LSE saved in ctx->pre_push_mpls_lse. The
2183 * flow's MPLS LSE should be restored to that value to allow any
2184 * subsequent actions that update of the LSE to be executed correctly.
2186 if (ctx->mpls_depth_delta > 0) {
2187 ctx->xin->flow.mpls_lse = ctx->pre_push_mpls_lse;
2190 ctx->xin->flow.dl_type = eth_type;
2191 ctx->mpls_depth_delta--;
2197 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2199 struct flow *flow = &ctx->xin->flow;
2201 if (!is_ip_any(flow)) {
2205 ctx->xout->wc.masks.nw_ttl = 0xff;
2206 if (flow->nw_ttl > 1) {
2212 for (i = 0; i < ids->n_controllers; i++) {
2213 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2217 /* Stop processing for current table. */
2223 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2225 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2229 /* If mpls_depth_delta is negative then an MPLS POP action has been
2230 * executed and the resulting MPLS label stack is unknown. This means
2231 * a SET MPLS LABEL action can't be executed as it needs to manipulate
2232 * the top-most MPLS LSE. Thus, stop processing.
2234 * It is planned that in the future this case will be handled
2237 if (ctx->mpls_depth_delta < 0) {
2241 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_LABEL_MASK);
2242 set_mpls_lse_label(&ctx->xin->flow.mpls_lse, label);
2247 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2249 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2253 /* If mpls_depth_delta is negative then an MPLS POP action has been
2254 * executed and the resulting MPLS label stack is unknown. This means
2255 * a SET MPLS TC action can't be executed as it needs to manipulate
2256 * the top-most MPLS LSE. Thus, stop processing.
2258 * It is planned that in the future this case will be handled
2261 if (ctx->mpls_depth_delta < 0) {
2265 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TC_MASK);
2266 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse, tc);
2271 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2273 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
2277 /* If mpls_depth_delta is negative then an MPLS POP action has been
2278 * executed and the resulting MPLS label stack is unknown. This means
2279 * a SET MPLS TTL push action can't be executed as it needs to manipulate
2280 * the top-most MPLS LSE. Thus, stop processing.
2282 * It is planned that in the future this case will be handled
2285 if (ctx->mpls_depth_delta < 0) {
2289 ctx->xout->wc.masks.mpls_lse |= htonl(MPLS_TTL_MASK);
2290 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
2295 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2297 struct flow *flow = &ctx->xin->flow;
2298 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
2299 struct flow_wildcards *wc = &ctx->xout->wc;
2301 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2303 if (!eth_type_mpls(flow->dl_type)) {
2309 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
2312 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2314 /* Stop processing for current table. */
2320 xlate_output_action(struct xlate_ctx *ctx,
2321 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2323 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2325 ctx->xout->nf_output_iface = NF_OUT_DROP;
2329 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2332 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2339 flood_packets(ctx, false);
2342 flood_packets(ctx, true);
2344 case OFPP_CONTROLLER:
2345 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2351 if (port != ctx->xin->flow.in_port.ofp_port) {
2352 compose_output_action(ctx, port);
2354 xlate_report(ctx, "skipping output to input port");
2359 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2360 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2361 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2362 ctx->xout->nf_output_iface = prev_nf_output_iface;
2363 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2364 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2365 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2370 xlate_output_reg_action(struct xlate_ctx *ctx,
2371 const struct ofpact_output_reg *or)
2373 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2374 if (port <= UINT16_MAX) {
2375 union mf_subvalue value;
2377 memset(&value, 0xff, sizeof value);
2378 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2379 xlate_output_action(ctx, u16_to_ofp(port),
2380 or->max_len, false);
2385 xlate_enqueue_action(struct xlate_ctx *ctx,
2386 const struct ofpact_enqueue *enqueue)
2388 ofp_port_t ofp_port = enqueue->port;
2389 uint32_t queue_id = enqueue->queue;
2390 uint32_t flow_priority, priority;
2393 /* Translate queue to priority. */
2394 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2396 /* Fall back to ordinary output action. */
2397 xlate_output_action(ctx, enqueue->port, 0, false);
2401 /* Check output port. */
2402 if (ofp_port == OFPP_IN_PORT) {
2403 ofp_port = ctx->xin->flow.in_port.ofp_port;
2404 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2408 /* Add datapath actions. */
2409 flow_priority = ctx->xin->flow.skb_priority;
2410 ctx->xin->flow.skb_priority = priority;
2411 compose_output_action(ctx, ofp_port);
2412 ctx->xin->flow.skb_priority = flow_priority;
2414 /* Update NetFlow output port. */
2415 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2416 ctx->xout->nf_output_iface = ofp_port;
2417 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2418 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2423 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2425 uint32_t skb_priority;
2427 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2428 ctx->xin->flow.skb_priority = skb_priority;
2430 /* Couldn't translate queue to a priority. Nothing to do. A warning
2431 * has already been logged. */
2436 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2438 const struct xbridge *xbridge = xbridge_;
2449 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2452 port = get_ofp_port(xbridge, ofp_port);
2453 return port ? port->may_enable : false;
2458 xlate_bundle_action(struct xlate_ctx *ctx,
2459 const struct ofpact_bundle *bundle)
2463 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2465 CONST_CAST(struct xbridge *, ctx->xbridge));
2466 if (bundle->dst.field) {
2467 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2470 xlate_output_action(ctx, port, 0, false);
2475 xlate_learn_action(struct xlate_ctx *ctx,
2476 const struct ofpact_learn *learn)
2478 uint64_t ofpacts_stub[1024 / 8];
2479 struct ofputil_flow_mod fm;
2480 struct ofpbuf ofpacts;
2482 ctx->xout->has_learn = true;
2484 learn_mask(learn, &ctx->xout->wc);
2486 if (!ctx->xin->may_learn) {
2490 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2491 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2492 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2493 ofpbuf_uninit(&ofpacts);
2497 xlate_fin_timeout(struct xlate_ctx *ctx,
2498 const struct ofpact_fin_timeout *oft)
2500 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2501 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2502 oft->fin_hard_timeout);
2507 xlate_sample_action(struct xlate_ctx *ctx,
2508 const struct ofpact_sample *os)
2510 union user_action_cookie cookie;
2511 /* Scale the probability from 16-bit to 32-bit while representing
2512 * the same percentage. */
2513 uint32_t probability = (os->probability << 16) | os->probability;
2515 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2516 &ctx->xout->odp_actions,
2518 &ctx->mpls_depth_delta);
2520 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2521 os->obs_domain_id, os->obs_point_id, &cookie);
2522 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2523 probability, &cookie, sizeof cookie.flow_sample);
2527 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2529 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2530 ? OFPUTIL_PC_NO_RECV_STP
2531 : OFPUTIL_PC_NO_RECV)) {
2535 /* Only drop packets here if both forwarding and learning are
2536 * disabled. If just learning is enabled, we need to have
2537 * OFPP_NORMAL and the learning action have a look at the packet
2538 * before we can drop it. */
2539 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2547 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2549 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2550 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2551 ofpact_pad(&ctx->action_set);
2555 xlate_action_set(struct xlate_ctx *ctx)
2557 uint64_t action_list_stub[1024 / 64];
2558 struct ofpbuf action_list;
2560 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2561 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2562 do_xlate_actions(action_list.data, action_list.size, ctx);
2563 ofpbuf_uninit(&action_list);
2567 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2568 struct xlate_ctx *ctx)
2570 struct flow_wildcards *wc = &ctx->xout->wc;
2571 struct flow *flow = &ctx->xin->flow;
2572 const struct ofpact *a;
2574 /* dl_type already in the mask, not set below. */
2576 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2577 struct ofpact_controller *controller;
2578 const struct ofpact_metadata *metadata;
2579 const struct ofpact_set_field *set_field;
2580 const struct mf_field *mf;
2588 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2589 ofpact_get_OUTPUT(a)->max_len, true);
2593 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2598 case OFPACT_CONTROLLER:
2599 controller = ofpact_get_CONTROLLER(a);
2600 execute_controller_action(ctx, controller->max_len,
2602 controller->controller_id);
2605 case OFPACT_ENQUEUE:
2606 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2609 case OFPACT_SET_VLAN_VID:
2610 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2611 if (flow->vlan_tci & htons(VLAN_CFI) ||
2612 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2613 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2614 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2619 case OFPACT_SET_VLAN_PCP:
2620 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2621 if (flow->vlan_tci & htons(VLAN_CFI) ||
2622 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2623 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2624 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2625 << VLAN_PCP_SHIFT) | VLAN_CFI);
2629 case OFPACT_STRIP_VLAN:
2630 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2631 flow->vlan_tci = htons(0);
2634 case OFPACT_PUSH_VLAN:
2635 /* XXX 802.1AD(QinQ) */
2636 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2637 flow->vlan_tci = htons(VLAN_CFI);
2640 case OFPACT_SET_ETH_SRC:
2641 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2642 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2645 case OFPACT_SET_ETH_DST:
2646 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2647 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2650 case OFPACT_SET_IPV4_SRC:
2651 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2652 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2653 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2657 case OFPACT_SET_IPV4_DST:
2658 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2659 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2660 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2664 case OFPACT_SET_IP_DSCP:
2665 if (is_ip_any(flow)) {
2666 wc->masks.nw_tos |= IP_DSCP_MASK;
2667 flow->nw_tos &= ~IP_DSCP_MASK;
2668 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2672 case OFPACT_SET_IP_ECN:
2673 if (is_ip_any(flow)) {
2674 wc->masks.nw_tos |= IP_ECN_MASK;
2675 flow->nw_tos &= ~IP_ECN_MASK;
2676 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2680 case OFPACT_SET_IP_TTL:
2681 if (is_ip_any(flow)) {
2682 wc->masks.nw_ttl = 0xff;
2683 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2687 case OFPACT_SET_L4_SRC_PORT:
2688 if (is_ip_any(flow)) {
2689 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2690 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2691 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2695 case OFPACT_SET_L4_DST_PORT:
2696 if (is_ip_any(flow)) {
2697 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2698 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2699 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2703 case OFPACT_RESUBMIT:
2704 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2707 case OFPACT_SET_TUNNEL:
2708 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2711 case OFPACT_SET_QUEUE:
2712 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2715 case OFPACT_POP_QUEUE:
2716 flow->skb_priority = ctx->orig_skb_priority;
2719 case OFPACT_REG_MOVE:
2720 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2723 case OFPACT_REG_LOAD:
2724 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2727 case OFPACT_SET_FIELD:
2728 set_field = ofpact_get_SET_FIELD(a);
2729 mf = set_field->field;
2730 mf_mask_field_and_prereqs(mf, &wc->masks);
2732 /* Set field action only ever overwrites packet's outermost
2733 * applicable header fields. Do nothing if no header exists. */
2734 if ((mf->id != MFF_VLAN_VID || flow->vlan_tci & htons(VLAN_CFI))
2735 && ((mf->id != MFF_MPLS_LABEL && mf->id != MFF_MPLS_TC)
2736 || flow->mpls_lse)) {
2737 mf_set_flow_value(mf, &set_field->value, flow);
2741 case OFPACT_STACK_PUSH:
2742 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2746 case OFPACT_STACK_POP:
2747 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2751 case OFPACT_PUSH_MPLS:
2752 if (compose_mpls_push_action(ctx,
2753 ofpact_get_PUSH_MPLS(a)->ethertype)) {
2758 case OFPACT_POP_MPLS:
2759 if (compose_mpls_pop_action(ctx,
2760 ofpact_get_POP_MPLS(a)->ethertype)) {
2765 case OFPACT_SET_MPLS_LABEL:
2766 if (compose_set_mpls_label_action(ctx,
2767 ofpact_get_SET_MPLS_LABEL(a)->label)) {
2772 case OFPACT_SET_MPLS_TC:
2773 if (compose_set_mpls_tc_action(ctx,
2774 ofpact_get_SET_MPLS_TC(a)->tc)) {
2779 case OFPACT_SET_MPLS_TTL:
2780 if (compose_set_mpls_ttl_action(ctx,
2781 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2786 case OFPACT_DEC_MPLS_TTL:
2787 if (compose_dec_mpls_ttl_action(ctx)) {
2792 case OFPACT_DEC_TTL:
2793 wc->masks.nw_ttl = 0xff;
2794 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2800 /* Nothing to do. */
2803 case OFPACT_MULTIPATH:
2804 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2808 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2811 case OFPACT_OUTPUT_REG:
2812 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2816 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2823 case OFPACT_FIN_TIMEOUT:
2824 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2825 ctx->xout->has_fin_timeout = true;
2826 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2829 case OFPACT_CLEAR_ACTIONS:
2830 ofpbuf_clear(&ctx->action_set);
2833 case OFPACT_WRITE_ACTIONS:
2834 xlate_write_actions(ctx, a);
2837 case OFPACT_WRITE_METADATA:
2838 metadata = ofpact_get_WRITE_METADATA(a);
2839 flow->metadata &= ~metadata->mask;
2840 flow->metadata |= metadata->metadata & metadata->mask;
2844 /* Not implemented yet. */
2847 case OFPACT_GOTO_TABLE: {
2848 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2850 ovs_assert(ctx->table_id < ogt->table_id);
2851 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2852 ogt->table_id, true);
2857 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2864 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2865 const struct flow *flow, struct rule_dpif *rule,
2866 uint16_t tcp_flags, const struct ofpbuf *packet)
2868 xin->ofproto = ofproto;
2870 xin->packet = packet;
2871 xin->may_learn = packet != NULL;
2873 xin->ofpacts = NULL;
2874 xin->ofpacts_len = 0;
2875 xin->tcp_flags = tcp_flags;
2876 xin->resubmit_hook = NULL;
2877 xin->report_hook = NULL;
2878 xin->resubmit_stats = NULL;
2879 xin->skip_wildcards = false;
2883 xlate_out_uninit(struct xlate_out *xout)
2886 ofpbuf_uninit(&xout->odp_actions);
2890 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2891 * into datapath actions, using 'ctx', and discards the datapath actions. */
2893 xlate_actions_for_side_effects(struct xlate_in *xin)
2895 struct xlate_out xout;
2897 xlate_actions(xin, &xout);
2898 xlate_out_uninit(&xout);
2902 xlate_report(struct xlate_ctx *ctx, const char *s)
2904 if (ctx->xin->report_hook) {
2905 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2910 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2913 dst->slow = src->slow;
2914 dst->has_learn = src->has_learn;
2915 dst->has_normal = src->has_normal;
2916 dst->has_fin_timeout = src->has_fin_timeout;
2917 dst->nf_output_iface = src->nf_output_iface;
2918 dst->mirrors = src->mirrors;
2920 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2921 sizeof dst->odp_actions_stub);
2922 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2923 src->odp_actions.size);
2926 static struct skb_priority_to_dscp *
2927 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2929 struct skb_priority_to_dscp *pdscp;
2932 hash = hash_int(skb_priority, 0);
2933 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2934 if (pdscp->skb_priority == skb_priority) {
2942 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2945 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2946 *dscp = pdscp ? pdscp->dscp : 0;
2947 return pdscp != NULL;
2951 clear_skb_priorities(struct xport *xport)
2953 struct skb_priority_to_dscp *pdscp, *next;
2955 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2956 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2962 actions_output_to_local_port(const struct xlate_ctx *ctx)
2964 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2965 const struct nlattr *a;
2968 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2969 ctx->xout->odp_actions.size) {
2970 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2971 && nl_attr_get_odp_port(a) == local_odp_port) {
2978 /* Thread safe call to xlate_actions__(). */
2980 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2982 ovs_rwlock_rdlock(&xlate_rwlock);
2983 xlate_actions__(xin, xout);
2984 ovs_rwlock_unlock(&xlate_rwlock);
2987 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2988 * into datapath actions in 'odp_actions', using 'ctx'.
2990 * The caller must take responsibility for eventually freeing 'xout', with
2991 * xlate_out_uninit(). */
2993 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
2994 OVS_REQ_RDLOCK(xlate_rwlock)
2996 struct flow_wildcards *wc = &xout->wc;
2997 struct flow *flow = &xin->flow;
2998 struct rule_dpif *rule = NULL;
3000 struct rule_actions *actions = NULL;
3001 enum slow_path_reason special;
3002 const struct ofpact *ofpacts;
3003 struct xport *in_port;
3004 struct flow orig_flow;
3005 struct xlate_ctx ctx;
3009 COVERAGE_INC(xlate_actions);
3011 /* Flow initialization rules:
3012 * - 'base_flow' must match the kernel's view of the packet at the
3013 * time that action processing starts. 'flow' represents any
3014 * transformations we wish to make through actions.
3015 * - By default 'base_flow' and 'flow' are the same since the input
3016 * packet matches the output before any actions are applied.
3017 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3018 * of the received packet as seen by the kernel. If we later output
3019 * to another device without any modifications this will cause us to
3020 * insert a new tag since the original one was stripped off by the
3022 * - Tunnel metadata as received is retained in 'flow'. This allows
3023 * tunnel metadata matching also in later tables.
3024 * Since a kernel action for setting the tunnel metadata will only be
3025 * generated with actual tunnel output, changing the tunnel metadata
3026 * values in 'flow' (such as tun_id) will only have effect with a later
3027 * tunnel output action.
3028 * - Tunnel 'base_flow' is completely cleared since that is what the
3029 * kernel does. If we wish to maintain the original values an action
3030 * needs to be generated. */
3035 ctx.xout->has_learn = false;
3036 ctx.xout->has_normal = false;
3037 ctx.xout->has_fin_timeout = false;
3038 ctx.xout->nf_output_iface = NF_OUT_DROP;
3039 ctx.xout->mirrors = 0;
3040 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3041 sizeof ctx.xout->odp_actions_stub);
3042 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3044 ctx.xbridge = xbridge_lookup(xin->ofproto);
3049 ctx.rule = xin->rule;
3051 ctx.base_flow = *flow;
3052 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3053 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3055 flow_wildcards_init_catchall(wc);
3056 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3057 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3058 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3059 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3061 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3062 if (ctx.xbridge->netflow) {
3063 netflow_mask_wc(flow, wc);
3068 ctx.orig_skb_priority = flow->skb_priority;
3071 ctx.mpls_depth_delta = 0;
3073 if (!xin->ofpacts && !ctx.rule) {
3074 rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3075 !xin->skip_wildcards ? wc : NULL, &rule);
3076 if (ctx.xin->resubmit_stats) {
3077 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3081 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3084 ofpacts = xin->ofpacts;
3085 ofpacts_len = xin->ofpacts_len;
3086 } else if (ctx.rule) {
3087 actions = rule_dpif_get_actions(ctx.rule);
3088 ofpacts = actions->ofpacts;
3089 ofpacts_len = actions->ofpacts_len;
3094 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3095 ofpbuf_use_stub(&ctx.action_set,
3096 ctx.action_set_stub, sizeof ctx.action_set_stub);
3098 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3099 /* Do this conditionally because the copy is expensive enough that it
3100 * shows up in profiles. */
3104 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3105 switch (ctx.xbridge->frag) {
3106 case OFPC_FRAG_NORMAL:
3107 /* We must pretend that transport ports are unavailable. */
3108 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3109 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3112 case OFPC_FRAG_DROP:
3115 case OFPC_FRAG_REASM:
3118 case OFPC_FRAG_NX_MATCH:
3119 /* Nothing to do. */
3122 case OFPC_INVALID_TTL_TO_CONTROLLER:
3127 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3128 if (in_port && in_port->is_tunnel && ctx.xin->resubmit_stats) {
3129 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3131 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3135 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3137 ctx.xout->slow |= special;
3139 size_t sample_actions_len;
3141 if (flow->in_port.ofp_port
3142 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3143 flow->in_port.ofp_port,
3145 ctx.base_flow.vlan_tci = 0;
3148 add_sflow_action(&ctx);
3149 add_ipfix_action(&ctx);
3150 sample_actions_len = ctx.xout->odp_actions.size;
3152 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3153 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3155 /* We've let OFPP_NORMAL and the learning action look at the
3156 * packet, so drop it now if forwarding is disabled. */
3157 if (in_port && !xport_stp_forward_state(in_port)) {
3158 ctx.xout->odp_actions.size = sample_actions_len;
3162 if (ctx.action_set.size) {
3163 xlate_action_set(&ctx);
3166 if (ctx.xbridge->has_in_band
3167 && in_band_must_output_to_local_port(flow)
3168 && !actions_output_to_local_port(&ctx)) {
3169 compose_output_action(&ctx, OFPP_LOCAL);
3172 fix_sflow_action(&ctx);
3174 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3175 add_mirror_actions(&ctx, &orig_flow);
3179 if (nl_attr_oversized(ctx.xout->odp_actions.size)) {
3180 /* These datapath actions are too big for a Netlink attribute, so we
3181 * can't hand them to the kernel directly. dpif_execute() can execute
3182 * them one by one with help, so just mark the result as SLOW_ACTION to
3183 * prevent the flow from being installed. */
3184 COVERAGE_INC(xlate_actions_oversize);
3185 ctx.xout->slow |= SLOW_ACTION;
3188 if (ctx.xin->resubmit_stats) {
3189 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3190 ctx.xin->resubmit_stats->n_packets,
3191 ctx.xin->resubmit_stats->n_bytes);
3193 if (ctx.xbridge->netflow) {
3194 const struct ofpact *ofpacts;
3197 ofpacts_len = actions->ofpacts_len;
3198 ofpacts = actions->ofpacts;
3199 if (ofpacts_len == 0
3200 || ofpacts->type != OFPACT_CONTROLLER
3201 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3202 /* Only update netflow if we don't have controller flow. We don't
3203 * report NetFlow expiration messages for such facets because they
3204 * are just part of the control logic for the network, not real
3206 netflow_flow_update(ctx.xbridge->netflow, flow,
3207 xout->nf_output_iface,
3208 ctx.xin->resubmit_stats);
3213 ofpbuf_uninit(&ctx.stack);
3214 ofpbuf_uninit(&ctx.action_set);
3216 /* Clear the metadata and register wildcard masks, because we won't
3217 * use non-header fields as part of the cache. */
3218 flow_wildcards_clear_non_packet_fields(wc);
3221 rule_actions_unref(actions);
3222 rule_dpif_unref(rule);
3225 /* Sends 'packet' out 'ofport'.
3226 * May modify 'packet'.
3227 * Returns 0 if successful, otherwise a positive errno value. */
3229 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3231 struct xport *xport;
3232 struct ofpact_output output;
3234 union flow_in_port in_port_;
3237 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3238 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3239 in_port_.ofp_port = OFPP_NONE;
3240 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
3242 ovs_rwlock_rdlock(&xlate_rwlock);
3243 xport = xport_lookup(ofport);
3245 ovs_rwlock_unlock(&xlate_rwlock);
3248 output.port = xport->ofp_port;
3250 error = ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3251 &output.ofpact, sizeof output,
3253 ovs_rwlock_unlock(&xlate_rwlock);