1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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-monitor.h"
46 #include "ofproto/ofproto-dpif-sflow.h"
47 #include "ofproto/ofproto-dpif.h"
48 #include "ofproto/ofproto-provider.h"
52 COVERAGE_DEFINE(xlate_actions);
53 COVERAGE_DEFINE(xlate_actions_oversize);
54 COVERAGE_DEFINE(xlate_actions_too_many_output);
55 COVERAGE_DEFINE(xlate_actions_mpls_overflow);
57 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
59 /* Maximum depth of flow table recursion (due to resubmit actions) in a
60 * flow translation. */
61 #define MAX_RESUBMIT_RECURSION 64
62 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
65 /* Maximum number of resubmit actions in a flow translation, whether they are
66 * recursive or not. */
67 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
69 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
72 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
73 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
75 struct list xbundles; /* Owned xbundles. */
76 struct hmap xports; /* Indexed by ofp_port. */
78 char *name; /* Name used in log messages. */
79 struct dpif *dpif; /* Datapath interface. */
80 struct mac_learning *ml; /* Mac learning handle. */
81 struct mbridge *mbridge; /* Mirroring. */
82 struct dpif_sflow *sflow; /* SFlow handle, or null. */
83 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
84 struct netflow *netflow; /* Netflow handle, or null. */
85 struct stp *stp; /* STP or null if disabled. */
87 /* Special rules installed by ofproto-dpif. */
88 struct rule_dpif *miss_rule;
89 struct rule_dpif *no_packet_in_rule;
91 enum ofp_config_flags frag; /* Fragmentation handling. */
92 bool has_in_band; /* Bridge has in band control? */
93 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
95 /* True if the datapath supports recirculation. */
98 /* True if the datapath supports variable-length
99 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
100 * False if the datapath supports only 8-byte (or shorter) userdata. */
101 bool variable_length_userdata;
103 /* Number of MPLS label stack entries that the datapath supports
105 size_t max_mpls_depth;
109 struct hmap_node hmap_node; /* In global 'xbundles' map. */
110 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
112 struct list list_node; /* In parent 'xbridges' list. */
113 struct xbridge *xbridge; /* Parent xbridge. */
115 struct list xports; /* Contains "struct xport"s. */
117 char *name; /* Name used in log messages. */
118 struct bond *bond; /* Nonnull iff more than one port. */
119 struct lacp *lacp; /* LACP handle or null. */
121 enum port_vlan_mode vlan_mode; /* VLAN mode. */
122 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
123 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
124 * NULL if all VLANs are trunked. */
125 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
126 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
130 struct hmap_node hmap_node; /* Node in global 'xports' map. */
131 struct ofport_dpif *ofport; /* Key in global 'xports map. */
133 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
134 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
136 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
138 struct list bundle_node; /* In parent xbundle (if it exists). */
139 struct xbundle *xbundle; /* Parent xbundle or null. */
141 struct netdev *netdev; /* 'ofport''s netdev. */
143 struct xbridge *xbridge; /* Parent bridge. */
144 struct xport *peer; /* Patch port peer or null. */
146 enum ofputil_port_config config; /* OpenFlow port configuration. */
147 enum ofputil_port_state state; /* OpenFlow port state. */
148 int stp_port_no; /* STP port number or -1 if not in use. */
150 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
152 bool may_enable; /* May be enabled in bonds. */
153 bool is_tunnel; /* Is a tunnel port. */
155 struct cfm *cfm; /* CFM handle or null. */
156 struct bfd *bfd; /* BFD handle or null. */
160 struct xlate_in *xin;
161 struct xlate_out *xout;
163 const struct xbridge *xbridge;
165 /* Flow at the last commit. */
166 struct flow base_flow;
168 /* Tunnel IP destination address as received. This is stored separately
169 * as the base_flow.tunnel is cleared on init to reflect the datapath
170 * behavior. Used to make sure not to send tunneled output to ourselves,
171 * which might lead to an infinite loop. This could happen easily
172 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
173 * actually set the tun_dst field. */
174 ovs_be32 orig_tunnel_ip_dst;
176 /* Stack for the push and pop actions. Each stack element is of type
177 * "union mf_subvalue". */
178 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
181 /* The rule that we are currently translating, or NULL. */
182 struct rule_dpif *rule;
184 /* Resubmit statistics, via xlate_table_action(). */
185 int recurse; /* Current resubmit nesting depth. */
186 int resubmits; /* Total number of resubmits. */
187 bool in_group; /* Currently translating ofgroup, if true. */
189 uint32_t orig_skb_priority; /* Priority when packet arrived. */
190 uint8_t table_id; /* OpenFlow table ID where flow was found. */
191 uint32_t sflow_n_outputs; /* Number of output ports. */
192 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
193 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
194 bool exit; /* No further actions should be processed. */
196 bool use_recirc; /* Should generate recirc? */
197 struct xlate_recirc recirc; /* Information used for generating
198 * recirculation actions */
200 /* OpenFlow 1.1+ action set.
202 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
203 * When translation is otherwise complete, ofpacts_execute_action_set()
204 * converts it to a set of "struct ofpact"s that can be translated into
205 * datapath actions. */
206 struct ofpbuf action_set; /* Action set. */
207 uint64_t action_set_stub[1024 / 8];
210 /* A controller may use OFPP_NONE as the ingress port to indicate that
211 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
212 * when an input bundle is needed for validation (e.g., mirroring or
213 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
214 * any 'port' structs, so care must be taken when dealing with it. */
215 static struct xbundle ofpp_none_bundle = {
217 .vlan_mode = PORT_VLAN_TRUNK
220 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
221 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
222 * traffic egressing the 'ofport' with that priority should be marked with. */
223 struct skb_priority_to_dscp {
224 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
225 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
227 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
241 /* xlate_cache entries hold enough information to perform the side effects of
242 * xlate_actions() for a rule, without needing to perform rule translation
243 * from scratch. The primary usage of these is to submit statistics to objects
244 * that a flow relates to, although they may be used for other effects as well
245 * (for instance, refreshing hard timeouts for learned flows). */
249 struct rule_dpif *rule;
256 struct netflow *netflow;
261 struct mbridge *mbridge;
262 mirror_mask_t mirrors;
270 struct ofproto_dpif *ofproto;
271 struct ofputil_flow_mod *fm;
272 struct ofpbuf *ofpacts;
275 struct ofproto_dpif *ofproto;
280 struct rule_dpif *rule;
287 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
288 entries = xcache->entries; \
289 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
291 entry = ofpbuf_try_pull(&entries, sizeof *entry))
294 struct ofpbuf entries;
297 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
298 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
299 static struct hmap xports = HMAP_INITIALIZER(&xports);
301 static bool may_receive(const struct xport *, struct xlate_ctx *);
302 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
304 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
305 OVS_REQ_RDLOCK(xlate_rwlock);
306 static void xlate_normal(struct xlate_ctx *);
307 static void xlate_report(struct xlate_ctx *, const char *);
308 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
309 uint8_t table_id, bool may_packet_in,
310 bool honor_table_miss);
311 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
312 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
313 static void output_normal(struct xlate_ctx *, const struct xbundle *,
315 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
317 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
318 static struct xbundle *xbundle_lookup(const struct ofbundle *);
319 static struct xport *xport_lookup(const struct ofport_dpif *);
320 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
321 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
322 uint32_t skb_priority);
323 static void clear_skb_priorities(struct xport *);
324 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
327 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
331 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
332 struct dpif *dpif, struct rule_dpif *miss_rule,
333 struct rule_dpif *no_packet_in_rule,
334 const struct mac_learning *ml, struct stp *stp,
335 const struct mbridge *mbridge,
336 const struct dpif_sflow *sflow,
337 const struct dpif_ipfix *ipfix,
338 const struct netflow *netflow, enum ofp_config_flags frag,
339 bool forward_bpdu, bool has_in_band,
341 bool variable_length_userdata,
342 size_t max_mpls_depth)
344 struct xbridge *xbridge = xbridge_lookup(ofproto);
347 xbridge = xzalloc(sizeof *xbridge);
348 xbridge->ofproto = ofproto;
350 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
351 hmap_init(&xbridge->xports);
352 list_init(&xbridge->xbundles);
355 if (xbridge->ml != ml) {
356 mac_learning_unref(xbridge->ml);
357 xbridge->ml = mac_learning_ref(ml);
360 if (xbridge->mbridge != mbridge) {
361 mbridge_unref(xbridge->mbridge);
362 xbridge->mbridge = mbridge_ref(mbridge);
365 if (xbridge->sflow != sflow) {
366 dpif_sflow_unref(xbridge->sflow);
367 xbridge->sflow = dpif_sflow_ref(sflow);
370 if (xbridge->ipfix != ipfix) {
371 dpif_ipfix_unref(xbridge->ipfix);
372 xbridge->ipfix = dpif_ipfix_ref(ipfix);
375 if (xbridge->stp != stp) {
376 stp_unref(xbridge->stp);
377 xbridge->stp = stp_ref(stp);
380 if (xbridge->netflow != netflow) {
381 netflow_unref(xbridge->netflow);
382 xbridge->netflow = netflow_ref(netflow);
386 xbridge->name = xstrdup(name);
388 xbridge->dpif = dpif;
389 xbridge->forward_bpdu = forward_bpdu;
390 xbridge->has_in_band = has_in_band;
391 xbridge->frag = frag;
392 xbridge->miss_rule = miss_rule;
393 xbridge->no_packet_in_rule = no_packet_in_rule;
394 xbridge->enable_recirc = enable_recirc;
395 xbridge->variable_length_userdata = variable_length_userdata;
396 xbridge->max_mpls_depth = max_mpls_depth;
400 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
402 struct xbridge *xbridge = xbridge_lookup(ofproto);
403 struct xbundle *xbundle, *next_xbundle;
404 struct xport *xport, *next_xport;
410 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
411 xlate_ofport_remove(xport->ofport);
414 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
415 xlate_bundle_remove(xbundle->ofbundle);
418 hmap_remove(&xbridges, &xbridge->hmap_node);
419 mac_learning_unref(xbridge->ml);
420 mbridge_unref(xbridge->mbridge);
421 dpif_sflow_unref(xbridge->sflow);
422 dpif_ipfix_unref(xbridge->ipfix);
423 stp_unref(xbridge->stp);
424 hmap_destroy(&xbridge->xports);
430 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
431 const char *name, enum port_vlan_mode vlan_mode, int vlan,
432 unsigned long *trunks, bool use_priority_tags,
433 const struct bond *bond, const struct lacp *lacp,
436 struct xbundle *xbundle = xbundle_lookup(ofbundle);
439 xbundle = xzalloc(sizeof *xbundle);
440 xbundle->ofbundle = ofbundle;
441 xbundle->xbridge = xbridge_lookup(ofproto);
443 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
444 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
445 list_init(&xbundle->xports);
448 ovs_assert(xbundle->xbridge);
451 xbundle->name = xstrdup(name);
453 xbundle->vlan_mode = vlan_mode;
454 xbundle->vlan = vlan;
455 xbundle->trunks = trunks;
456 xbundle->use_priority_tags = use_priority_tags;
457 xbundle->floodable = floodable;
459 if (xbundle->bond != bond) {
460 bond_unref(xbundle->bond);
461 xbundle->bond = bond_ref(bond);
464 if (xbundle->lacp != lacp) {
465 lacp_unref(xbundle->lacp);
466 xbundle->lacp = lacp_ref(lacp);
471 xlate_bundle_remove(struct ofbundle *ofbundle)
473 struct xbundle *xbundle = xbundle_lookup(ofbundle);
474 struct xport *xport, *next;
480 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
481 list_remove(&xport->bundle_node);
482 xport->xbundle = NULL;
485 hmap_remove(&xbundles, &xbundle->hmap_node);
486 list_remove(&xbundle->list_node);
487 bond_unref(xbundle->bond);
488 lacp_unref(xbundle->lacp);
494 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
495 struct ofport_dpif *ofport, ofp_port_t ofp_port,
496 odp_port_t odp_port, const struct netdev *netdev,
497 const struct cfm *cfm, const struct bfd *bfd,
498 struct ofport_dpif *peer, int stp_port_no,
499 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
500 enum ofputil_port_config config,
501 enum ofputil_port_state state, bool is_tunnel,
504 struct xport *xport = xport_lookup(ofport);
508 xport = xzalloc(sizeof *xport);
509 xport->ofport = ofport;
510 xport->xbridge = xbridge_lookup(ofproto);
511 xport->ofp_port = ofp_port;
513 hmap_init(&xport->skb_priorities);
514 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
515 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
516 hash_ofp_port(xport->ofp_port));
519 ovs_assert(xport->ofp_port == ofp_port);
521 xport->config = config;
522 xport->state = state;
523 xport->stp_port_no = stp_port_no;
524 xport->is_tunnel = is_tunnel;
525 xport->may_enable = may_enable;
526 xport->odp_port = odp_port;
528 if (xport->netdev != netdev) {
529 netdev_close(xport->netdev);
530 xport->netdev = netdev_ref(netdev);
533 if (xport->cfm != cfm) {
534 cfm_unref(xport->cfm);
535 xport->cfm = cfm_ref(cfm);
538 if (xport->bfd != bfd) {
539 bfd_unref(xport->bfd);
540 xport->bfd = bfd_ref(bfd);
544 xport->peer->peer = NULL;
546 xport->peer = xport_lookup(peer);
548 xport->peer->peer = xport;
551 if (xport->xbundle) {
552 list_remove(&xport->bundle_node);
554 xport->xbundle = xbundle_lookup(ofbundle);
555 if (xport->xbundle) {
556 list_insert(&xport->xbundle->xports, &xport->bundle_node);
559 clear_skb_priorities(xport);
560 for (i = 0; i < n_qdscp; i++) {
561 struct skb_priority_to_dscp *pdscp;
562 uint32_t skb_priority;
564 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
569 pdscp = xmalloc(sizeof *pdscp);
570 pdscp->skb_priority = skb_priority;
571 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
572 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
573 hash_int(pdscp->skb_priority, 0));
578 xlate_ofport_remove(struct ofport_dpif *ofport)
580 struct xport *xport = xport_lookup(ofport);
587 xport->peer->peer = NULL;
591 if (xport->xbundle) {
592 list_remove(&xport->bundle_node);
595 clear_skb_priorities(xport);
596 hmap_destroy(&xport->skb_priorities);
598 hmap_remove(&xports, &xport->hmap_node);
599 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
601 netdev_close(xport->netdev);
602 cfm_unref(xport->cfm);
603 bfd_unref(xport->bfd);
607 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
608 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
609 * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
610 * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
611 * 'netflow' with the appropriate handles for those protocols if they're
612 * enabled. Caller is responsible for unrefing them.
614 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
615 * 'flow''s in_port to OFPP_NONE.
617 * This function does post-processing on data returned from
618 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
619 * of the upcall processing logic. In particular, if the extracted in_port is
620 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
621 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
622 * a VLAN header onto 'packet' (if it is nonnull).
624 * Similarly, this function also includes some logic to help with tunnels. It
625 * may modify 'flow' as necessary to make the tunneling implementation
626 * transparent to the upcall processing logic.
628 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
629 * or some other positive errno if there are other problems. */
631 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
632 const struct nlattr *key, size_t key_len, struct flow *flow,
633 struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
634 struct dpif_sflow **sflow, struct netflow **netflow,
635 odp_port_t *odp_in_port)
637 const struct xport *xport;
640 ovs_rwlock_rdlock(&xlate_rwlock);
641 if (odp_flow_key_to_flow(key, key_len, flow) == ODP_FIT_ERROR) {
647 *odp_in_port = flow->in_port.odp_port;
650 xport = xport_lookup(tnl_port_should_receive(flow)
651 ? tnl_port_receive(flow)
652 : odp_port_to_ofport(backer, flow->in_port.odp_port));
654 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
659 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
661 /* Make the packet resemble the flow, so that it gets sent to
662 * an OpenFlow controller properly, so that it looks correct
663 * for sFlow, and so that flow_extract() will get the correct
664 * vlan_tci if it is called on 'packet'. */
665 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
671 *ofproto = xport->xbridge->ofproto;
675 *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
679 *sflow = dpif_sflow_ref(xport->xbridge->sflow);
683 *netflow = netflow_ref(xport->xbridge->netflow);
687 ovs_rwlock_unlock(&xlate_rwlock);
691 static struct xbridge *
692 xbridge_lookup(const struct ofproto_dpif *ofproto)
694 struct xbridge *xbridge;
700 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
702 if (xbridge->ofproto == ofproto) {
709 static struct xbundle *
710 xbundle_lookup(const struct ofbundle *ofbundle)
712 struct xbundle *xbundle;
718 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
720 if (xbundle->ofbundle == ofbundle) {
727 static struct xport *
728 xport_lookup(const struct ofport_dpif *ofport)
736 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
738 if (xport->ofport == ofport) {
745 static struct stp_port *
746 xport_get_stp_port(const struct xport *xport)
748 return xport->xbridge->stp && xport->stp_port_no != -1
749 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
754 xport_stp_learn_state(const struct xport *xport)
756 struct stp_port *sp = xport_get_stp_port(xport);
757 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
761 xport_stp_forward_state(const struct xport *xport)
763 struct stp_port *sp = xport_get_stp_port(xport);
764 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
768 xport_stp_should_forward_bpdu(const struct xport *xport)
770 struct stp_port *sp = xport_get_stp_port(xport);
771 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
774 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
775 * were used to make the determination.*/
777 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
779 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
780 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
785 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
787 struct stp_port *sp = xport_get_stp_port(xport);
788 struct ofpbuf payload = *packet;
789 struct eth_header *eth = ofpbuf_data(&payload);
791 /* Sink packets on ports that have STP disabled when the bridge has
793 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
797 /* Trim off padding on payload. */
798 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
799 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
802 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
803 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
807 static struct xport *
808 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
812 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
814 if (xport->ofp_port == ofp_port) {
822 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
824 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
825 return xport ? xport->odp_port : ODPP_NONE;
829 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
833 xport = get_ofp_port(ctx->xbridge, ofp_port);
834 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
835 xport->state & OFPUTIL_PS_LINK_DOWN) {
842 static const struct ofputil_bucket *
843 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
847 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
849 struct group_dpif *group;
852 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
857 hit = group_first_live_bucket(ctx, group, depth) != NULL;
859 group_dpif_release(group);
863 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
866 bucket_is_alive(const struct xlate_ctx *ctx,
867 const struct ofputil_bucket *bucket, int depth)
869 if (depth >= MAX_LIVENESS_RECURSION) {
870 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
872 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
873 MAX_LIVENESS_RECURSION);
877 return !ofputil_bucket_has_liveness(bucket) ||
878 (bucket->watch_port != OFPP_ANY &&
879 odp_port_is_alive(ctx, bucket->watch_port)) ||
880 (bucket->watch_group != OFPG_ANY &&
881 group_is_alive(ctx, bucket->watch_group, depth + 1));
884 static const struct ofputil_bucket *
885 group_first_live_bucket(const struct xlate_ctx *ctx,
886 const struct group_dpif *group, int depth)
888 struct ofputil_bucket *bucket;
889 const struct list *buckets;
891 group_dpif_get_buckets(group, &buckets);
892 LIST_FOR_EACH (bucket, list_node, buckets) {
893 if (bucket_is_alive(ctx, bucket, depth)) {
901 static const struct ofputil_bucket *
902 group_best_live_bucket(const struct xlate_ctx *ctx,
903 const struct group_dpif *group,
906 const struct ofputil_bucket *best_bucket = NULL;
907 uint32_t best_score = 0;
910 const struct ofputil_bucket *bucket;
911 const struct list *buckets;
913 group_dpif_get_buckets(group, &buckets);
914 LIST_FOR_EACH (bucket, list_node, buckets) {
915 if (bucket_is_alive(ctx, bucket, 0)) {
916 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
917 if (score >= best_score) {
918 best_bucket = bucket;
929 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
931 return (bundle->vlan_mode != PORT_VLAN_ACCESS
932 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
936 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
938 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
942 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
944 return xbundle != &ofpp_none_bundle
945 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
950 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
952 return xbundle != &ofpp_none_bundle
953 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
958 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
960 return xbundle != &ofpp_none_bundle
961 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
965 static struct xbundle *
966 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
967 bool warn, struct xport **in_xportp)
971 /* Find the port and bundle for the received packet. */
972 xport = get_ofp_port(xbridge, in_port);
976 if (xport && xport->xbundle) {
977 return xport->xbundle;
980 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
981 * which a controller may use as the ingress port for traffic that
983 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
984 return &ofpp_none_bundle;
987 /* Odd. A few possible reasons here:
989 * - We deleted a port but there are still a few packets queued up
992 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
993 * we don't know about.
995 * - The ofproto client didn't configure the port as part of a bundle.
996 * This is particularly likely to happen if a packet was received on the
997 * port after it was created, but before the client had a chance to
998 * configure its bundle.
1001 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1003 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1004 "port %"PRIu16, xbridge->name, in_port);
1010 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1012 const struct xbridge *xbridge = ctx->xbridge;
1013 mirror_mask_t mirrors;
1014 struct xbundle *in_xbundle;
1018 mirrors = ctx->xout->mirrors;
1019 ctx->xout->mirrors = 0;
1021 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1022 ctx->xin->packet != NULL, NULL);
1026 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1028 /* Drop frames on bundles reserved for mirroring. */
1029 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1030 if (ctx->xin->packet != NULL) {
1031 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1032 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1033 "%s, which is reserved exclusively for mirroring",
1034 ctx->xbridge->name, in_xbundle->name);
1036 ofpbuf_clear(&ctx->xout->odp_actions);
1041 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1042 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1045 vlan = input_vid_to_vlan(in_xbundle, vid);
1051 /* Restore the original packet before adding the mirror actions. */
1052 ctx->xin->flow = *orig_flow;
1055 mirror_mask_t dup_mirrors;
1056 struct ofbundle *out;
1057 unsigned long *vlans;
1062 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1063 &vlans, &dup_mirrors, &out, &out_vlan);
1064 ovs_assert(has_mirror);
1067 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1069 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1072 if (!vlan_mirrored) {
1073 mirrors = zero_rightmost_1bit(mirrors);
1077 mirrors &= ~dup_mirrors;
1078 ctx->xout->mirrors |= dup_mirrors;
1080 struct xbundle *out_xbundle = xbundle_lookup(out);
1082 output_normal(ctx, out_xbundle, vlan);
1084 } else if (vlan != out_vlan
1085 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1086 struct xbundle *xbundle;
1088 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1089 if (xbundle_includes_vlan(xbundle, out_vlan)
1090 && !xbundle_mirror_out(xbridge, xbundle)) {
1091 output_normal(ctx, xbundle, out_vlan);
1098 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1099 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1100 * the bundle on which the packet was received, returns the VLAN to which the
1103 * Both 'vid' and the return value are in the range 0...4095. */
1105 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1107 switch (in_xbundle->vlan_mode) {
1108 case PORT_VLAN_ACCESS:
1109 return in_xbundle->vlan;
1112 case PORT_VLAN_TRUNK:
1115 case PORT_VLAN_NATIVE_UNTAGGED:
1116 case PORT_VLAN_NATIVE_TAGGED:
1117 return vid ? vid : in_xbundle->vlan;
1124 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1125 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1128 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1129 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1132 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1134 /* Allow any VID on the OFPP_NONE port. */
1135 if (in_xbundle == &ofpp_none_bundle) {
1139 switch (in_xbundle->vlan_mode) {
1140 case PORT_VLAN_ACCESS:
1143 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1144 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1145 "packet received on port %s configured as VLAN "
1146 "%"PRIu16" access port", vid, in_xbundle->name,
1153 case PORT_VLAN_NATIVE_UNTAGGED:
1154 case PORT_VLAN_NATIVE_TAGGED:
1156 /* Port must always carry its native VLAN. */
1160 case PORT_VLAN_TRUNK:
1161 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1163 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1164 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1165 "received on port %s not configured for trunking "
1166 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1178 /* Given 'vlan', the VLAN that a packet belongs to, and
1179 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1180 * that should be included in the 802.1Q header. (If the return value is 0,
1181 * then the 802.1Q header should only be included in the packet if there is a
1184 * Both 'vlan' and the return value are in the range 0...4095. */
1186 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1188 switch (out_xbundle->vlan_mode) {
1189 case PORT_VLAN_ACCESS:
1192 case PORT_VLAN_TRUNK:
1193 case PORT_VLAN_NATIVE_TAGGED:
1196 case PORT_VLAN_NATIVE_UNTAGGED:
1197 return vlan == out_xbundle->vlan ? 0 : vlan;
1205 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1208 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1210 ovs_be16 tci, old_tci;
1211 struct xport *xport;
1213 vid = output_vlan_to_vid(out_xbundle, vlan);
1214 if (list_is_empty(&out_xbundle->xports)) {
1215 /* Partially configured bundle with no slaves. Drop the packet. */
1217 } else if (!out_xbundle->bond) {
1218 ctx->use_recirc = false;
1219 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1222 struct ofport_dpif *ofport;
1223 struct xlate_recirc *xr = &ctx->recirc;
1224 struct flow_wildcards *wc = &ctx->xout->wc;
1226 if (ctx->xbridge->enable_recirc) {
1227 ctx->use_recirc = bond_may_recirc(
1228 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1230 if (ctx->use_recirc) {
1231 /* Only TCP mode uses recirculation. */
1232 xr->hash_alg = OVS_HASH_ALG_L4;
1233 bond_update_post_recirc_rules(out_xbundle->bond, false);
1235 /* Recirculation does not require unmasking hash fields. */
1240 ofport = bond_choose_output_slave(out_xbundle->bond,
1241 &ctx->xin->flow, wc, vid);
1242 xport = xport_lookup(ofport);
1245 /* No slaves enabled, so drop packet. */
1249 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1250 * accounting for this bond. */
1251 if (!ctx->use_recirc) {
1252 if (ctx->xin->resubmit_stats) {
1253 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1254 ctx->xin->resubmit_stats->n_bytes);
1256 if (ctx->xin->xcache) {
1257 struct xc_entry *entry;
1260 flow = &ctx->xin->flow;
1261 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1262 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1263 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1264 entry->u.bond.vid = vid;
1269 old_tci = *flow_tci;
1271 if (tci || out_xbundle->use_priority_tags) {
1272 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1274 tci |= htons(VLAN_CFI);
1279 compose_output_action(ctx, xport->ofp_port);
1280 *flow_tci = old_tci;
1283 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1284 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1285 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1287 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1289 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1293 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1294 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1298 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1299 if (flow->nw_proto == ARP_OP_REPLY) {
1301 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1302 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1303 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1305 return flow->nw_src == flow->nw_dst;
1311 /* Checks whether a MAC learning update is necessary for MAC learning table
1312 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1315 * Most packets processed through the MAC learning table do not actually
1316 * change it in any way. This function requires only a read lock on the MAC
1317 * learning table, so it is much cheaper in this common case.
1319 * Keep the code here synchronized with that in update_learning_table__()
1322 is_mac_learning_update_needed(const struct mac_learning *ml,
1323 const struct flow *flow,
1324 struct flow_wildcards *wc,
1325 int vlan, struct xbundle *in_xbundle)
1326 OVS_REQ_RDLOCK(ml->rwlock)
1328 struct mac_entry *mac;
1330 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1334 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1335 if (!mac || mac_entry_age(ml, mac)) {
1339 if (is_gratuitous_arp(flow, wc)) {
1340 /* We don't want to learn from gratuitous ARP packets that are
1341 * reflected back over bond slaves so we lock the learning table. */
1342 if (!in_xbundle->bond) {
1344 } else if (mac_entry_is_grat_arp_locked(mac)) {
1349 return mac->port.p != in_xbundle->ofbundle;
1353 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1354 * received on 'in_xbundle' in 'vlan'.
1356 * This code repeats all the checks in is_mac_learning_update_needed() because
1357 * the lock was released between there and here and thus the MAC learning state
1358 * could have changed.
1360 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1363 update_learning_table__(const struct xbridge *xbridge,
1364 const struct flow *flow, struct flow_wildcards *wc,
1365 int vlan, struct xbundle *in_xbundle)
1366 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1368 struct mac_entry *mac;
1370 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1374 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1375 if (is_gratuitous_arp(flow, wc)) {
1376 /* We don't want to learn from gratuitous ARP packets that are
1377 * reflected back over bond slaves so we lock the learning table. */
1378 if (!in_xbundle->bond) {
1379 mac_entry_set_grat_arp_lock(mac);
1380 } else if (mac_entry_is_grat_arp_locked(mac)) {
1385 if (mac->port.p != in_xbundle->ofbundle) {
1386 /* The log messages here could actually be useful in debugging,
1387 * so keep the rate limit relatively high. */
1388 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1390 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1391 "on port %s in VLAN %d",
1392 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1393 in_xbundle->name, vlan);
1395 mac->port.p = in_xbundle->ofbundle;
1396 mac_learning_changed(xbridge->ml);
1401 update_learning_table(const struct xbridge *xbridge,
1402 const struct flow *flow, struct flow_wildcards *wc,
1403 int vlan, struct xbundle *in_xbundle)
1407 /* Don't learn the OFPP_NONE port. */
1408 if (in_xbundle == &ofpp_none_bundle) {
1412 /* First try the common case: no change to MAC learning table. */
1413 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1414 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1416 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1419 /* Slow path: MAC learning table might need an update. */
1420 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1421 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1422 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1426 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1427 * dropped. Returns true if they may be forwarded, false if they should be
1430 * 'in_port' must be the xport that corresponds to flow->in_port.
1431 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1433 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1434 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1435 * checked by input_vid_is_valid().
1437 * May also add tags to '*tags', although the current implementation only does
1438 * so in one special case.
1441 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1444 struct xbundle *in_xbundle = in_port->xbundle;
1445 const struct xbridge *xbridge = ctx->xbridge;
1446 struct flow *flow = &ctx->xin->flow;
1448 /* Drop frames for reserved multicast addresses
1449 * only if forward_bpdu option is absent. */
1450 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1451 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1455 if (in_xbundle->bond) {
1456 struct mac_entry *mac;
1458 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1464 xlate_report(ctx, "bonding refused admissibility, dropping");
1467 case BV_DROP_IF_MOVED:
1468 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1469 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1470 if (mac && mac->port.p != in_xbundle->ofbundle &&
1471 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1472 || mac_entry_is_grat_arp_locked(mac))) {
1473 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1474 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1478 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1487 xlate_normal(struct xlate_ctx *ctx)
1489 struct flow_wildcards *wc = &ctx->xout->wc;
1490 struct flow *flow = &ctx->xin->flow;
1491 struct xbundle *in_xbundle;
1492 struct xport *in_port;
1493 struct mac_entry *mac;
1498 ctx->xout->has_normal = true;
1500 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1501 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1502 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1504 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1505 ctx->xin->packet != NULL, &in_port);
1507 xlate_report(ctx, "no input bundle, dropping");
1511 /* Drop malformed frames. */
1512 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1513 !(flow->vlan_tci & htons(VLAN_CFI))) {
1514 if (ctx->xin->packet != NULL) {
1515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1516 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1517 "VLAN tag received on port %s",
1518 ctx->xbridge->name, in_xbundle->name);
1520 xlate_report(ctx, "partial VLAN tag, dropping");
1524 /* Drop frames on bundles reserved for mirroring. */
1525 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1526 if (ctx->xin->packet != NULL) {
1527 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1528 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1529 "%s, which is reserved exclusively for mirroring",
1530 ctx->xbridge->name, in_xbundle->name);
1532 xlate_report(ctx, "input port is mirror output port, dropping");
1537 vid = vlan_tci_to_vid(flow->vlan_tci);
1538 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1539 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1542 vlan = input_vid_to_vlan(in_xbundle, vid);
1544 /* Check other admissibility requirements. */
1545 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1549 /* Learn source MAC. */
1550 if (ctx->xin->may_learn) {
1551 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1553 if (ctx->xin->xcache) {
1554 struct xc_entry *entry;
1556 /* Save enough info to update mac learning table later. */
1557 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
1558 entry->u.normal.ofproto = ctx->xbridge->ofproto;
1559 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
1560 entry->u.normal.vlan = vlan;
1563 /* Determine output bundle. */
1564 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1565 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1566 mac_port = mac ? mac->port.p : NULL;
1567 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1570 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1571 if (mac_xbundle && mac_xbundle != in_xbundle) {
1572 xlate_report(ctx, "forwarding to learned port");
1573 output_normal(ctx, mac_xbundle, vlan);
1574 } else if (!mac_xbundle) {
1575 xlate_report(ctx, "learned port is unknown, dropping");
1577 xlate_report(ctx, "learned port is input port, dropping");
1580 struct xbundle *xbundle;
1582 xlate_report(ctx, "no learned MAC for destination, flooding");
1583 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1584 if (xbundle != in_xbundle
1585 && xbundle_includes_vlan(xbundle, vlan)
1586 && xbundle->floodable
1587 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1588 output_normal(ctx, xbundle, vlan);
1591 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1595 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1596 * the number of packets out of UINT32_MAX to sample. The given
1597 * cookie is passed back in the callback for each sampled packet.
1600 compose_sample_action(const struct xbridge *xbridge,
1601 struct ofpbuf *odp_actions,
1602 const struct flow *flow,
1603 const uint32_t probability,
1604 const union user_action_cookie *cookie,
1605 const size_t cookie_size)
1607 size_t sample_offset, actions_offset;
1608 odp_port_t odp_port;
1612 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1614 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1616 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1618 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1619 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
1620 flow_hash_5tuple(flow, 0));
1621 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
1624 nl_msg_end_nested(odp_actions, actions_offset);
1625 nl_msg_end_nested(odp_actions, sample_offset);
1626 return cookie_offset;
1630 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1631 odp_port_t odp_port, unsigned int n_outputs,
1632 union user_action_cookie *cookie)
1636 cookie->type = USER_ACTION_COOKIE_SFLOW;
1637 cookie->sflow.vlan_tci = vlan_tci;
1639 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1640 * port information") for the interpretation of cookie->output. */
1641 switch (n_outputs) {
1643 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1644 cookie->sflow.output = 0x40000000 | 256;
1648 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1650 cookie->sflow.output = ifindex;
1655 /* 0x80000000 means "multiple output ports. */
1656 cookie->sflow.output = 0x80000000 | n_outputs;
1661 /* Compose SAMPLE action for sFlow bridge sampling. */
1663 compose_sflow_action(const struct xbridge *xbridge,
1664 struct ofpbuf *odp_actions,
1665 const struct flow *flow,
1666 odp_port_t odp_port)
1668 uint32_t probability;
1669 union user_action_cookie cookie;
1671 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1675 probability = dpif_sflow_get_probability(xbridge->sflow);
1676 compose_sflow_cookie(xbridge, htons(0), odp_port,
1677 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1679 return compose_sample_action(xbridge, odp_actions, flow, probability,
1680 &cookie, sizeof cookie.sflow);
1684 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1685 uint32_t obs_domain_id, uint32_t obs_point_id,
1686 union user_action_cookie *cookie)
1688 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1689 cookie->flow_sample.probability = probability;
1690 cookie->flow_sample.collector_set_id = collector_set_id;
1691 cookie->flow_sample.obs_domain_id = obs_domain_id;
1692 cookie->flow_sample.obs_point_id = obs_point_id;
1696 compose_ipfix_cookie(union user_action_cookie *cookie)
1698 cookie->type = USER_ACTION_COOKIE_IPFIX;
1701 /* Compose SAMPLE action for IPFIX bridge sampling. */
1703 compose_ipfix_action(const struct xbridge *xbridge,
1704 struct ofpbuf *odp_actions,
1705 const struct flow *flow)
1707 uint32_t probability;
1708 union user_action_cookie cookie;
1710 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1714 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1715 compose_ipfix_cookie(&cookie);
1717 compose_sample_action(xbridge, odp_actions, flow, probability,
1718 &cookie, sizeof cookie.ipfix);
1721 /* SAMPLE action for sFlow must be first action in any given list of
1722 * actions. At this point we do not have all information required to
1723 * build it. So try to build sample action as complete as possible. */
1725 add_sflow_action(struct xlate_ctx *ctx)
1727 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1728 &ctx->xout->odp_actions,
1729 &ctx->xin->flow, ODPP_NONE);
1730 ctx->sflow_odp_port = 0;
1731 ctx->sflow_n_outputs = 0;
1734 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1735 * of actions, eventually after the SAMPLE action for sFlow. */
1737 add_ipfix_action(struct xlate_ctx *ctx)
1739 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1743 /* Fix SAMPLE action according to data collected while composing ODP actions.
1744 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1745 * USERSPACE action's user-cookie which is required for sflow. */
1747 fix_sflow_action(struct xlate_ctx *ctx)
1749 const struct flow *base = &ctx->base_flow;
1750 union user_action_cookie *cookie;
1752 if (!ctx->user_cookie_offset) {
1756 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1757 sizeof cookie->sflow);
1758 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1760 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1761 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1764 static enum slow_path_reason
1765 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1766 const struct xport *xport, const struct ofpbuf *packet)
1768 struct flow_wildcards *wc = &ctx->xout->wc;
1769 const struct xbridge *xbridge = ctx->xbridge;
1773 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1775 cfm_process_heartbeat(xport->cfm, packet);
1778 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1780 bfd_process_packet(xport->bfd, flow, packet);
1781 /* If POLL received, immediately sends FINAL back. */
1782 if (bfd_should_send_packet(xport->bfd)) {
1783 ofproto_dpif_monitor_port_send_soon(xport->ofport);
1787 } else if (xport->xbundle && xport->xbundle->lacp
1788 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1790 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1793 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1795 stp_process_packet(xport, packet);
1804 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1807 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1808 struct flow_wildcards *wc = &ctx->xout->wc;
1809 struct flow *flow = &ctx->xin->flow;
1810 ovs_be16 flow_vlan_tci;
1811 uint32_t flow_pkt_mark;
1812 uint8_t flow_nw_tos;
1813 odp_port_t out_port, odp_port;
1816 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1817 * before traversing a patch port. */
1818 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 26);
1821 xlate_report(ctx, "Nonexistent output port");
1823 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1824 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1826 } else if (check_stp) {
1827 if (is_stp(&ctx->base_flow)) {
1828 if (!xport_stp_should_forward_bpdu(xport)) {
1829 xlate_report(ctx, "STP not in listening state, "
1830 "skipping bpdu output");
1833 } else if (!xport_stp_forward_state(xport)) {
1834 xlate_report(ctx, "STP not in forwarding state, "
1840 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1841 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1846 const struct xport *peer = xport->peer;
1847 struct flow old_flow = ctx->xin->flow;
1848 enum slow_path_reason special;
1850 ctx->xbridge = peer->xbridge;
1851 flow->in_port.ofp_port = peer->ofp_port;
1852 flow->metadata = htonll(0);
1853 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1854 memset(flow->regs, 0, sizeof flow->regs);
1856 special = process_special(ctx, &ctx->xin->flow, peer,
1859 ctx->xout->slow |= special;
1860 } else if (may_receive(peer, ctx)) {
1861 if (xport_stp_forward_state(peer)) {
1862 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1864 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1865 * learning action look at the packet, then drop it. */
1866 struct flow old_base_flow = ctx->base_flow;
1867 size_t old_size = ofpbuf_size(&ctx->xout->odp_actions);
1868 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1869 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1870 ctx->xout->mirrors = old_mirrors;
1871 ctx->base_flow = old_base_flow;
1872 ofpbuf_set_size(&ctx->xout->odp_actions, old_size);
1876 ctx->xin->flow = old_flow;
1877 ctx->xbridge = xport->xbridge;
1879 if (ctx->xin->resubmit_stats) {
1880 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1881 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1883 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1886 if (ctx->xin->xcache) {
1887 struct xc_entry *entry;
1889 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1890 entry->u.dev.tx = netdev_ref(xport->netdev);
1891 entry->u.dev.rx = netdev_ref(peer->netdev);
1892 entry->u.dev.bfd = bfd_ref(peer->bfd);
1898 flow_vlan_tci = flow->vlan_tci;
1899 flow_pkt_mark = flow->pkt_mark;
1900 flow_nw_tos = flow->nw_tos;
1902 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1903 wc->masks.nw_tos |= IP_DSCP_MASK;
1904 flow->nw_tos &= ~IP_DSCP_MASK;
1905 flow->nw_tos |= dscp;
1908 if (xport->is_tunnel) {
1909 /* Save tunnel metadata so that changes made due to
1910 * the Logical (tunnel) Port are not visible for any further
1911 * matches, while explicit set actions on tunnel metadata are.
1913 struct flow_tnl flow_tnl = flow->tunnel;
1914 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1915 if (odp_port == ODPP_NONE) {
1916 xlate_report(ctx, "Tunneling decided against output");
1917 goto out; /* restore flow_nw_tos */
1919 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1920 xlate_report(ctx, "Not tunneling to our own address");
1921 goto out; /* restore flow_nw_tos */
1923 if (ctx->xin->resubmit_stats) {
1924 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1926 if (ctx->xin->xcache) {
1927 struct xc_entry *entry;
1929 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1930 entry->u.dev.tx = netdev_ref(xport->netdev);
1932 out_port = odp_port;
1933 commit_odp_tunnel_action(flow, &ctx->base_flow,
1934 &ctx->xout->odp_actions);
1935 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1937 odp_port = xport->odp_port;
1938 out_port = odp_port;
1939 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1940 ofp_port_t vlandev_port;
1942 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1943 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1944 ofp_port, flow->vlan_tci);
1945 if (vlandev_port != ofp_port) {
1946 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1947 flow->vlan_tci = htons(0);
1952 if (out_port != ODPP_NONE) {
1953 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1954 &ctx->xout->odp_actions,
1957 if (ctx->use_recirc) {
1958 struct ovs_action_hash *act_hash;
1959 struct xlate_recirc *xr = &ctx->recirc;
1962 act_hash = nl_msg_put_unspec_uninit(&ctx->xout->odp_actions,
1963 OVS_ACTION_ATTR_HASH,
1965 act_hash->hash_alg = xr->hash_alg;
1966 act_hash->hash_basis = xr->hash_basis;
1968 /* Recirc action. */
1969 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
1972 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1976 ctx->sflow_odp_port = odp_port;
1977 ctx->sflow_n_outputs++;
1978 ctx->xout->nf_output_iface = ofp_port;
1983 flow->vlan_tci = flow_vlan_tci;
1984 flow->pkt_mark = flow_pkt_mark;
1985 flow->nw_tos = flow_nw_tos;
1989 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1991 compose_output_action__(ctx, ofp_port, true);
1995 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1997 struct rule_dpif *old_rule = ctx->rule;
1998 const struct rule_actions *actions;
2000 if (ctx->xin->resubmit_stats) {
2001 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2007 actions = rule_dpif_get_actions(rule);
2008 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2009 ctx->rule = old_rule;
2014 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2016 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2018 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2019 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2020 MAX_RESUBMIT_RECURSION);
2021 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2022 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2023 } else if (ofpbuf_size(&ctx->xout->odp_actions) > UINT16_MAX) {
2024 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2025 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2026 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2035 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2036 bool may_packet_in, bool honor_table_miss)
2038 if (xlate_resubmit_resource_check(ctx)) {
2039 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
2040 bool skip_wildcards = ctx->xin->skip_wildcards;
2041 uint8_t old_table_id = ctx->table_id;
2042 struct rule_dpif *rule;
2043 enum rule_dpif_lookup_verdict verdict;
2044 enum ofputil_port_config config = 0;
2046 ctx->table_id = table_id;
2048 /* Look up a flow with 'in_port' as the input port. Then restore the
2049 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2050 * have surprising behavior). */
2051 ctx->xin->flow.in_port.ofp_port = in_port;
2052 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2055 ? &ctx->xout->wc : NULL,
2057 &ctx->table_id, &rule,
2058 ctx->xin->xcache != NULL);
2059 ctx->xin->flow.in_port.ofp_port = old_in_port;
2061 if (ctx->xin->resubmit_hook) {
2062 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
2066 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
2068 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
2069 if (may_packet_in) {
2070 struct xport *xport;
2072 xport = get_ofp_port(ctx->xbridge,
2073 ctx->xin->flow.in_port.ofp_port);
2074 config = xport ? xport->config : 0;
2077 /* Fall through to drop */
2078 case RULE_DPIF_LOOKUP_VERDICT_DROP:
2079 config = OFPUTIL_PC_NO_PACKET_IN;
2081 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
2082 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
2083 config = OFPUTIL_PC_NO_PACKET_IN;
2090 choose_miss_rule(config, ctx->xbridge->miss_rule,
2091 ctx->xbridge->no_packet_in_rule, &rule,
2092 ctx->xin->xcache != NULL);
2096 /* Fill in the cache entry here instead of xlate_recursively
2097 * to make the reference counting more explicit. We take a
2098 * reference in the lookups above if we are going to cache the
2100 if (ctx->xin->xcache) {
2101 struct xc_entry *entry;
2103 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2104 entry->u.rule = rule;
2106 xlate_recursively(ctx, rule);
2109 ctx->table_id = old_table_id;
2117 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
2119 uint64_t action_list_stub[1024 / 8];
2120 struct ofpbuf action_list, action_set;
2122 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2123 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2125 ofpacts_execute_action_set(&action_list, &action_set);
2127 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2130 ofpbuf_uninit(&action_set);
2131 ofpbuf_uninit(&action_list);
2135 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2137 const struct ofputil_bucket *bucket;
2138 const struct list *buckets;
2139 struct flow old_flow = ctx->xin->flow;
2141 group_dpif_get_buckets(group, &buckets);
2143 LIST_FOR_EACH (bucket, list_node, buckets) {
2144 xlate_group_bucket(ctx, bucket);
2145 /* Roll back flow to previous state.
2146 * This is equivalent to cloning the packet for each bucket.
2148 * As a side effect any subsequently applied actions will
2149 * also effectively be applied to a clone of the packet taken
2150 * just before applying the all or indirect group. */
2151 ctx->xin->flow = old_flow;
2156 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2158 const struct ofputil_bucket *bucket;
2160 bucket = group_first_live_bucket(ctx, group, 0);
2162 xlate_group_bucket(ctx, bucket);
2167 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2169 struct flow_wildcards *wc = &ctx->xout->wc;
2170 const struct ofputil_bucket *bucket;
2173 basis = hash_mac(ctx->xin->flow.dl_dst, 0, 0);
2174 bucket = group_best_live_bucket(ctx, group, basis);
2176 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2177 xlate_group_bucket(ctx, bucket);
2182 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2184 ctx->in_group = true;
2186 switch (group_dpif_get_type(group)) {
2188 case OFPGT11_INDIRECT:
2189 xlate_all_group(ctx, group);
2191 case OFPGT11_SELECT:
2192 xlate_select_group(ctx, group);
2195 xlate_ff_group(ctx, group);
2200 group_dpif_release(group);
2202 ctx->in_group = false;
2206 xlate_group_resource_check(struct xlate_ctx *ctx)
2208 if (!xlate_resubmit_resource_check(ctx)) {
2210 } else if (ctx->in_group) {
2211 /* Prevent nested translation of OpenFlow groups.
2213 * OpenFlow allows this restriction. We enforce this restriction only
2214 * because, with the current architecture, we would otherwise have to
2215 * take a possibly recursive read lock on the ofgroup rwlock, which is
2216 * unsafe given that POSIX allows taking a read lock to block if there
2217 * is a thread blocked on taking the write lock. Other solutions
2218 * without this restriction are also possible, but seem unwarranted
2219 * given the current limited use of groups. */
2220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2222 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2230 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2232 if (xlate_group_resource_check(ctx)) {
2233 struct group_dpif *group;
2236 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2238 xlate_group_action__(ctx, group);
2248 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2249 const struct ofpact_resubmit *resubmit)
2253 bool may_packet_in = false;
2254 bool honor_table_miss = false;
2256 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2257 /* Still allow missed packets to be sent to the controller
2258 * if resubmitting from an internal table. */
2259 may_packet_in = true;
2260 honor_table_miss = true;
2263 in_port = resubmit->in_port;
2264 if (in_port == OFPP_IN_PORT) {
2265 in_port = ctx->xin->flow.in_port.ofp_port;
2268 table_id = resubmit->table_id;
2269 if (table_id == 255) {
2270 table_id = ctx->table_id;
2273 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2278 flood_packets(struct xlate_ctx *ctx, bool all)
2280 const struct xport *xport;
2282 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2283 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2288 compose_output_action__(ctx, xport->ofp_port, false);
2289 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2290 compose_output_action(ctx, xport->ofp_port);
2294 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2298 execute_controller_action(struct xlate_ctx *ctx, int len,
2299 enum ofp_packet_in_reason reason,
2300 uint16_t controller_id)
2302 struct ofproto_packet_in *pin;
2303 struct ofpbuf *packet;
2304 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2306 ctx->xout->slow |= SLOW_CONTROLLER;
2307 if (!ctx->xin->packet) {
2311 packet = ofpbuf_clone(ctx->xin->packet);
2313 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2314 &ctx->xout->odp_actions,
2317 odp_execute_actions(NULL, packet, false, &md,
2318 ofpbuf_data(&ctx->xout->odp_actions),
2319 ofpbuf_size(&ctx->xout->odp_actions), NULL);
2321 pin = xmalloc(sizeof *pin);
2322 pin->up.packet_len = ofpbuf_size(packet);
2323 pin->up.packet = ofpbuf_steal_data(packet);
2324 pin->up.reason = reason;
2325 pin->up.table_id = ctx->table_id;
2326 pin->up.cookie = (ctx->rule
2327 ? rule_dpif_get_flow_cookie(ctx->rule)
2330 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2332 pin->controller_id = controller_id;
2333 pin->send_len = len;
2334 /* If a rule is a table-miss rule then this is
2335 * a table-miss handled by a table-miss rule.
2337 * Else, if rule is internal and has a controller action,
2338 * the later being implied by the rule being processed here,
2339 * then this is a table-miss handled without a table-miss rule.
2341 * Otherwise this is not a table-miss. */
2342 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
2344 if (rule_dpif_is_table_miss(ctx->rule)) {
2345 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
2346 } else if (rule_dpif_is_internal(ctx->rule)) {
2347 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
2350 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2351 ofpbuf_delete(packet);
2355 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2357 struct flow_wildcards *wc = &ctx->xout->wc;
2358 struct flow *flow = &ctx->xin->flow;
2361 ovs_assert(eth_type_mpls(mpls->ethertype));
2363 n = flow_count_mpls_labels(flow, wc);
2365 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2366 &ctx->xout->odp_actions,
2368 } else if (n >= FLOW_MAX_MPLS_LABELS) {
2369 if (ctx->xin->packet != NULL) {
2370 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2371 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2372 "MPLS push action can't be performed as it would "
2373 "have more MPLS LSEs than the %d supported.",
2374 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2378 } else if (n >= ctx->xbridge->max_mpls_depth) {
2379 COVERAGE_INC(xlate_actions_mpls_overflow);
2380 ctx->xout->slow |= SLOW_ACTION;
2383 flow_push_mpls(flow, n, mpls->ethertype, wc);
2387 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2389 struct flow_wildcards *wc = &ctx->xout->wc;
2390 struct flow *flow = &ctx->xin->flow;
2391 int n = flow_count_mpls_labels(flow, wc);
2393 if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2394 if (ctx->xin->packet != NULL) {
2395 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2396 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2397 "MPLS pop action can't be performed as it has "
2398 "more MPLS LSEs than the %d supported.",
2399 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2402 ofpbuf_clear(&ctx->xout->odp_actions);
2407 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2409 struct flow *flow = &ctx->xin->flow;
2411 if (!is_ip_any(flow)) {
2415 ctx->xout->wc.masks.nw_ttl = 0xff;
2416 if (flow->nw_ttl > 1) {
2422 for (i = 0; i < ids->n_controllers; i++) {
2423 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2427 /* Stop processing for current table. */
2433 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2435 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2436 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2437 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2442 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2444 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2445 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2446 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2451 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2453 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2454 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2455 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2460 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2462 struct flow *flow = &ctx->xin->flow;
2463 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2464 struct flow_wildcards *wc = &ctx->xout->wc;
2466 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2467 if (eth_type_mpls(flow->dl_type)) {
2470 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2473 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2475 /* Stop processing for current table. */
2484 xlate_output_action(struct xlate_ctx *ctx,
2485 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2487 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2489 ctx->xout->nf_output_iface = NF_OUT_DROP;
2493 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2496 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2497 0, may_packet_in, true);
2503 flood_packets(ctx, false);
2506 flood_packets(ctx, true);
2508 case OFPP_CONTROLLER:
2509 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2515 if (port != ctx->xin->flow.in_port.ofp_port) {
2516 compose_output_action(ctx, port);
2518 xlate_report(ctx, "skipping output to input port");
2523 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2524 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2525 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2526 ctx->xout->nf_output_iface = prev_nf_output_iface;
2527 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2528 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2529 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2534 xlate_output_reg_action(struct xlate_ctx *ctx,
2535 const struct ofpact_output_reg *or)
2537 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2538 if (port <= UINT16_MAX) {
2539 union mf_subvalue value;
2541 memset(&value, 0xff, sizeof value);
2542 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2543 xlate_output_action(ctx, u16_to_ofp(port),
2544 or->max_len, false);
2549 xlate_enqueue_action(struct xlate_ctx *ctx,
2550 const struct ofpact_enqueue *enqueue)
2552 ofp_port_t ofp_port = enqueue->port;
2553 uint32_t queue_id = enqueue->queue;
2554 uint32_t flow_priority, priority;
2557 /* Translate queue to priority. */
2558 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2560 /* Fall back to ordinary output action. */
2561 xlate_output_action(ctx, enqueue->port, 0, false);
2565 /* Check output port. */
2566 if (ofp_port == OFPP_IN_PORT) {
2567 ofp_port = ctx->xin->flow.in_port.ofp_port;
2568 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2572 /* Add datapath actions. */
2573 flow_priority = ctx->xin->flow.skb_priority;
2574 ctx->xin->flow.skb_priority = priority;
2575 compose_output_action(ctx, ofp_port);
2576 ctx->xin->flow.skb_priority = flow_priority;
2578 /* Update NetFlow output port. */
2579 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2580 ctx->xout->nf_output_iface = ofp_port;
2581 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2582 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2587 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2589 uint32_t skb_priority;
2591 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2592 ctx->xin->flow.skb_priority = skb_priority;
2594 /* Couldn't translate queue to a priority. Nothing to do. A warning
2595 * has already been logged. */
2600 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2602 const struct xbridge *xbridge = xbridge_;
2613 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2616 port = get_ofp_port(xbridge, ofp_port);
2617 return port ? port->may_enable : false;
2622 xlate_bundle_action(struct xlate_ctx *ctx,
2623 const struct ofpact_bundle *bundle)
2627 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2629 CONST_CAST(struct xbridge *, ctx->xbridge));
2630 if (bundle->dst.field) {
2631 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2634 xlate_output_action(ctx, port, 0, false);
2639 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
2640 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
2642 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
2643 if (ctx->xin->may_learn) {
2644 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
2649 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
2651 ctx->xout->has_learn = true;
2652 learn_mask(learn, &ctx->xout->wc);
2654 if (ctx->xin->xcache) {
2655 struct xc_entry *entry;
2657 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
2658 entry->u.learn.ofproto = ctx->xbridge->ofproto;
2659 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
2660 entry->u.learn.ofpacts = ofpbuf_new(64);
2661 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
2662 entry->u.learn.ofpacts);
2663 } else if (ctx->xin->may_learn) {
2664 uint64_t ofpacts_stub[1024 / 8];
2665 struct ofputil_flow_mod fm;
2666 struct ofpbuf ofpacts;
2668 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2669 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
2670 ofpbuf_uninit(&ofpacts);
2675 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
2676 uint16_t idle_timeout, uint16_t hard_timeout)
2678 if (tcp_flags & (TCP_FIN | TCP_RST)) {
2679 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
2684 xlate_fin_timeout(struct xlate_ctx *ctx,
2685 const struct ofpact_fin_timeout *oft)
2688 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
2689 oft->fin_idle_timeout, oft->fin_hard_timeout);
2690 if (ctx->xin->xcache) {
2691 struct xc_entry *entry;
2693 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
2694 /* XC_RULE already holds a reference on the rule, none is taken
2696 entry->u.fin.rule = ctx->rule;
2697 entry->u.fin.idle = oft->fin_idle_timeout;
2698 entry->u.fin.hard = oft->fin_hard_timeout;
2704 xlate_sample_action(struct xlate_ctx *ctx,
2705 const struct ofpact_sample *os)
2707 union user_action_cookie cookie;
2708 /* Scale the probability from 16-bit to 32-bit while representing
2709 * the same percentage. */
2710 uint32_t probability = (os->probability << 16) | os->probability;
2712 if (!ctx->xbridge->variable_length_userdata) {
2713 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2715 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2716 "lacks support (needs Linux 3.10+ or kernel module from "
2721 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2722 &ctx->xout->odp_actions,
2725 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2726 os->obs_domain_id, os->obs_point_id, &cookie);
2727 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2728 probability, &cookie, sizeof cookie.flow_sample);
2732 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2734 if (xport->config & (is_stp(&ctx->xin->flow)
2735 ? OFPUTIL_PC_NO_RECV_STP
2736 : OFPUTIL_PC_NO_RECV)) {
2740 /* Only drop packets here if both forwarding and learning are
2741 * disabled. If just learning is enabled, we need to have
2742 * OFPP_NORMAL and the learning action have a look at the packet
2743 * before we can drop it. */
2744 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2752 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2754 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2755 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2756 ofpact_pad(&ctx->action_set);
2760 xlate_action_set(struct xlate_ctx *ctx)
2762 uint64_t action_list_stub[1024 / 64];
2763 struct ofpbuf action_list;
2765 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2766 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2767 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2768 ofpbuf_uninit(&action_list);
2772 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2773 struct xlate_ctx *ctx)
2775 struct flow_wildcards *wc = &ctx->xout->wc;
2776 struct flow *flow = &ctx->xin->flow;
2777 const struct ofpact *a;
2779 /* dl_type already in the mask, not set below. */
2781 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2782 struct ofpact_controller *controller;
2783 const struct ofpact_metadata *metadata;
2784 const struct ofpact_set_field *set_field;
2785 const struct mf_field *mf;
2793 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2794 ofpact_get_OUTPUT(a)->max_len, true);
2798 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2803 case OFPACT_CONTROLLER:
2804 controller = ofpact_get_CONTROLLER(a);
2805 execute_controller_action(ctx, controller->max_len,
2807 controller->controller_id);
2810 case OFPACT_ENQUEUE:
2811 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2814 case OFPACT_SET_VLAN_VID:
2815 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2816 if (flow->vlan_tci & htons(VLAN_CFI) ||
2817 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2818 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2819 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2824 case OFPACT_SET_VLAN_PCP:
2825 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2826 if (flow->vlan_tci & htons(VLAN_CFI) ||
2827 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2828 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2829 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2830 << VLAN_PCP_SHIFT) | VLAN_CFI);
2834 case OFPACT_STRIP_VLAN:
2835 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2836 flow->vlan_tci = htons(0);
2839 case OFPACT_PUSH_VLAN:
2840 /* XXX 802.1AD(QinQ) */
2841 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2842 flow->vlan_tci = htons(VLAN_CFI);
2845 case OFPACT_SET_ETH_SRC:
2846 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2847 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2850 case OFPACT_SET_ETH_DST:
2851 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2852 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2855 case OFPACT_SET_IPV4_SRC:
2856 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2857 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2858 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2862 case OFPACT_SET_IPV4_DST:
2863 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2864 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2865 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2869 case OFPACT_SET_IP_DSCP:
2870 if (is_ip_any(flow)) {
2871 wc->masks.nw_tos |= IP_DSCP_MASK;
2872 flow->nw_tos &= ~IP_DSCP_MASK;
2873 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2877 case OFPACT_SET_IP_ECN:
2878 if (is_ip_any(flow)) {
2879 wc->masks.nw_tos |= IP_ECN_MASK;
2880 flow->nw_tos &= ~IP_ECN_MASK;
2881 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2885 case OFPACT_SET_IP_TTL:
2886 if (is_ip_any(flow)) {
2887 wc->masks.nw_ttl = 0xff;
2888 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2892 case OFPACT_SET_L4_SRC_PORT:
2893 if (is_ip_any(flow)) {
2894 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2895 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2896 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2900 case OFPACT_SET_L4_DST_PORT:
2901 if (is_ip_any(flow)) {
2902 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2903 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2904 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2908 case OFPACT_RESUBMIT:
2909 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2912 case OFPACT_SET_TUNNEL:
2913 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2916 case OFPACT_SET_QUEUE:
2917 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2920 case OFPACT_POP_QUEUE:
2921 flow->skb_priority = ctx->orig_skb_priority;
2924 case OFPACT_REG_MOVE:
2925 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2928 case OFPACT_REG_LOAD:
2929 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2932 case OFPACT_SET_FIELD:
2933 set_field = ofpact_get_SET_FIELD(a);
2934 mf = set_field->field;
2936 /* Set field action only ever overwrites packet's outermost
2937 * applicable header fields. Do nothing if no header exists. */
2938 if (mf->id == MFF_VLAN_VID) {
2939 wc->masks.vlan_tci |= htons(VLAN_CFI);
2940 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2943 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
2944 /* 'dl_type' is already unwildcarded. */
2945 && !eth_type_mpls(flow->dl_type)) {
2949 mf_mask_field_and_prereqs(mf, &wc->masks);
2950 mf_set_flow_value(mf, &set_field->value, flow);
2953 case OFPACT_STACK_PUSH:
2954 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2958 case OFPACT_STACK_POP:
2959 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2963 case OFPACT_PUSH_MPLS:
2964 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2967 case OFPACT_POP_MPLS:
2968 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2971 case OFPACT_SET_MPLS_LABEL:
2972 compose_set_mpls_label_action(
2973 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2976 case OFPACT_SET_MPLS_TC:
2977 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2980 case OFPACT_SET_MPLS_TTL:
2981 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2984 case OFPACT_DEC_MPLS_TTL:
2985 if (compose_dec_mpls_ttl_action(ctx)) {
2990 case OFPACT_DEC_TTL:
2991 wc->masks.nw_ttl = 0xff;
2992 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2998 /* Nothing to do. */
3001 case OFPACT_MULTIPATH:
3002 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
3006 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
3009 case OFPACT_OUTPUT_REG:
3010 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
3014 xlate_learn_action(ctx, ofpact_get_LEARN(a));
3021 case OFPACT_FIN_TIMEOUT:
3022 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3023 ctx->xout->has_fin_timeout = true;
3024 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
3027 case OFPACT_CLEAR_ACTIONS:
3028 ofpbuf_clear(&ctx->action_set);
3031 case OFPACT_WRITE_ACTIONS:
3032 xlate_write_actions(ctx, a);
3035 case OFPACT_WRITE_METADATA:
3036 metadata = ofpact_get_WRITE_METADATA(a);
3037 flow->metadata &= ~metadata->mask;
3038 flow->metadata |= metadata->metadata & metadata->mask;
3042 /* Not implemented yet. */
3045 case OFPACT_GOTO_TABLE: {
3046 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
3048 ovs_assert(ctx->table_id < ogt->table_id);
3049 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3050 ogt->table_id, true, true);
3055 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
3062 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
3063 const struct flow *flow, struct rule_dpif *rule,
3064 uint16_t tcp_flags, const struct ofpbuf *packet)
3066 xin->ofproto = ofproto;
3068 xin->packet = packet;
3069 xin->may_learn = packet != NULL;
3072 xin->ofpacts = NULL;
3073 xin->ofpacts_len = 0;
3074 xin->tcp_flags = tcp_flags;
3075 xin->resubmit_hook = NULL;
3076 xin->report_hook = NULL;
3077 xin->resubmit_stats = NULL;
3078 xin->skip_wildcards = false;
3082 xlate_out_uninit(struct xlate_out *xout)
3085 ofpbuf_uninit(&xout->odp_actions);
3089 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
3090 * into datapath actions, using 'ctx', and discards the datapath actions. */
3092 xlate_actions_for_side_effects(struct xlate_in *xin)
3094 struct xlate_out xout;
3096 xlate_actions(xin, &xout);
3097 xlate_out_uninit(&xout);
3101 xlate_report(struct xlate_ctx *ctx, const char *s)
3103 if (ctx->xin->report_hook) {
3104 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
3109 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
3112 dst->slow = src->slow;
3113 dst->has_learn = src->has_learn;
3114 dst->has_normal = src->has_normal;
3115 dst->has_fin_timeout = src->has_fin_timeout;
3116 dst->nf_output_iface = src->nf_output_iface;
3117 dst->mirrors = src->mirrors;
3119 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
3120 sizeof dst->odp_actions_stub);
3121 ofpbuf_put(&dst->odp_actions, ofpbuf_data(&src->odp_actions),
3122 ofpbuf_size(&src->odp_actions));
3125 static struct skb_priority_to_dscp *
3126 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
3128 struct skb_priority_to_dscp *pdscp;
3131 hash = hash_int(skb_priority, 0);
3132 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
3133 if (pdscp->skb_priority == skb_priority) {
3141 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
3144 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
3145 *dscp = pdscp ? pdscp->dscp : 0;
3146 return pdscp != NULL;
3150 clear_skb_priorities(struct xport *xport)
3152 struct skb_priority_to_dscp *pdscp, *next;
3154 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
3155 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
3161 actions_output_to_local_port(const struct xlate_ctx *ctx)
3163 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
3164 const struct nlattr *a;
3167 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(&ctx->xout->odp_actions),
3168 ofpbuf_size(&ctx->xout->odp_actions)) {
3169 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
3170 && nl_attr_get_odp_port(a) == local_odp_port) {
3177 /* Thread safe call to xlate_actions__(). */
3179 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
3180 OVS_EXCLUDED(xlate_rwlock)
3182 ovs_rwlock_rdlock(&xlate_rwlock);
3183 xlate_actions__(xin, xout);
3184 ovs_rwlock_unlock(&xlate_rwlock);
3187 /* Returns the maximum number of packets that the Linux kernel is willing to
3188 * queue up internally to certain kinds of software-implemented ports, or the
3189 * default (and rarely modified) value if it cannot be determined. */
3191 netdev_max_backlog(void)
3193 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
3194 static int max_backlog = 1000; /* The normal default value. */
3196 if (ovsthread_once_start(&once)) {
3197 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
3201 stream = fopen(filename, "r");
3203 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
3205 if (fscanf(stream, "%d", &n) != 1) {
3206 VLOG_WARN("%s: read error", filename);
3207 } else if (n <= 100) {
3208 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
3214 ovsthread_once_done(&once);
3216 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
3222 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
3225 count_output_actions(const struct ofpbuf *odp_actions)
3227 const struct nlattr *a;
3231 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(odp_actions),
3232 ofpbuf_size(odp_actions)) {
3233 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
3240 /* Returns true if 'odp_actions' contains more output actions than the datapath
3241 * can reliably handle in one go. On Linux, this is the value of the
3242 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
3243 * packets that the kernel is willing to queue up for processing while the
3244 * datapath is processing a set of actions. */
3246 too_many_output_actions(const struct ofpbuf *odp_actions)
3249 return (ofpbuf_size(odp_actions) / NL_A_U32_SIZE > netdev_max_backlog()
3250 && count_output_actions(odp_actions) > netdev_max_backlog());
3252 /* OSes other than Linux might have similar limits, but we don't know how
3253 * to determine them.*/
3258 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
3259 * into datapath actions in 'odp_actions', using 'ctx'.
3261 * The caller must take responsibility for eventually freeing 'xout', with
3262 * xlate_out_uninit(). */
3264 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
3265 OVS_REQ_RDLOCK(xlate_rwlock)
3267 struct flow_wildcards *wc = &xout->wc;
3268 struct flow *flow = &xin->flow;
3269 struct rule_dpif *rule = NULL;
3271 const struct rule_actions *actions = NULL;
3272 enum slow_path_reason special;
3273 const struct ofpact *ofpacts;
3274 struct xport *in_port;
3275 struct flow orig_flow;
3276 struct xlate_ctx ctx;
3281 COVERAGE_INC(xlate_actions);
3283 /* Flow initialization rules:
3284 * - 'base_flow' must match the kernel's view of the packet at the
3285 * time that action processing starts. 'flow' represents any
3286 * transformations we wish to make through actions.
3287 * - By default 'base_flow' and 'flow' are the same since the input
3288 * packet matches the output before any actions are applied.
3289 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3290 * of the received packet as seen by the kernel. If we later output
3291 * to another device without any modifications this will cause us to
3292 * insert a new tag since the original one was stripped off by the
3294 * - Tunnel metadata as received is retained in 'flow'. This allows
3295 * tunnel metadata matching also in later tables.
3296 * Since a kernel action for setting the tunnel metadata will only be
3297 * generated with actual tunnel output, changing the tunnel metadata
3298 * values in 'flow' (such as tun_id) will only have effect with a later
3299 * tunnel output action.
3300 * - Tunnel 'base_flow' is completely cleared since that is what the
3301 * kernel does. If we wish to maintain the original values an action
3302 * needs to be generated. */
3307 ctx.xout->has_learn = false;
3308 ctx.xout->has_normal = false;
3309 ctx.xout->has_fin_timeout = false;
3310 ctx.xout->nf_output_iface = NF_OUT_DROP;
3311 ctx.xout->mirrors = 0;
3312 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3313 sizeof ctx.xout->odp_actions_stub);
3314 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3316 ctx.xbridge = xbridge_lookup(xin->ofproto);
3321 ctx.rule = xin->rule;
3323 ctx.base_flow = *flow;
3324 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3325 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3327 flow_wildcards_init_catchall(wc);
3328 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3329 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3330 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3331 if (is_ip_any(flow)) {
3332 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3334 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
3336 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3337 if (ctx.xbridge->netflow) {
3338 netflow_mask_wc(flow, wc);
3343 ctx.in_group = false;
3344 ctx.orig_skb_priority = flow->skb_priority;
3347 ctx.use_recirc = false;
3349 if (!xin->ofpacts && !ctx.rule) {
3350 ctx.table_id = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3351 !xin->skip_wildcards ? wc : NULL,
3352 &rule, ctx.xin->xcache != NULL);
3353 if (ctx.xin->resubmit_stats) {
3354 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3356 if (ctx.xin->xcache) {
3357 struct xc_entry *entry;
3359 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
3360 entry->u.rule = rule;
3364 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3367 ofpacts = xin->ofpacts;
3368 ofpacts_len = xin->ofpacts_len;
3369 } else if (ctx.rule) {
3370 actions = rule_dpif_get_actions(ctx.rule);
3371 ofpacts = actions->ofpacts;
3372 ofpacts_len = actions->ofpacts_len;
3377 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3378 ofpbuf_use_stub(&ctx.action_set,
3379 ctx.action_set_stub, sizeof ctx.action_set_stub);
3381 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3382 /* Do this conditionally because the copy is expensive enough that it
3383 * shows up in profiles. */
3387 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3388 switch (ctx.xbridge->frag) {
3389 case OFPC_FRAG_NORMAL:
3390 /* We must pretend that transport ports are unavailable. */
3391 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3392 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3395 case OFPC_FRAG_DROP:
3398 case OFPC_FRAG_REASM:
3401 case OFPC_FRAG_NX_MATCH:
3402 /* Nothing to do. */
3405 case OFPC_INVALID_TTL_TO_CONTROLLER:
3410 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3411 if (in_port && in_port->is_tunnel) {
3412 if (ctx.xin->resubmit_stats) {
3413 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3415 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3418 if (ctx.xin->xcache) {
3419 struct xc_entry *entry;
3421 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
3422 entry->u.dev.rx = netdev_ref(in_port->netdev);
3423 entry->u.dev.bfd = bfd_ref(in_port->bfd);
3427 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3429 ctx.xout->slow |= special;
3431 size_t sample_actions_len;
3433 if (flow->in_port.ofp_port
3434 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3435 flow->in_port.ofp_port,
3437 ctx.base_flow.vlan_tci = 0;
3440 add_sflow_action(&ctx);
3441 add_ipfix_action(&ctx);
3442 sample_actions_len = ofpbuf_size(&ctx.xout->odp_actions);
3444 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3445 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3447 /* We've let OFPP_NORMAL and the learning action look at the
3448 * packet, so drop it now if forwarding is disabled. */
3449 if (in_port && !xport_stp_forward_state(in_port)) {
3450 ofpbuf_set_size(&ctx.xout->odp_actions, sample_actions_len);
3454 if (ofpbuf_size(&ctx.action_set)) {
3455 xlate_action_set(&ctx);
3458 if (ctx.xbridge->has_in_band
3459 && in_band_must_output_to_local_port(flow)
3460 && !actions_output_to_local_port(&ctx)) {
3461 compose_output_action(&ctx, OFPP_LOCAL);
3464 fix_sflow_action(&ctx);
3466 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3467 add_mirror_actions(&ctx, &orig_flow);
3471 if (nl_attr_oversized(ofpbuf_size(&ctx.xout->odp_actions))) {
3472 /* These datapath actions are too big for a Netlink attribute, so we
3473 * can't hand them to the kernel directly. dpif_execute() can execute
3474 * them one by one with help, so just mark the result as SLOW_ACTION to
3475 * prevent the flow from being installed. */
3476 COVERAGE_INC(xlate_actions_oversize);
3477 ctx.xout->slow |= SLOW_ACTION;
3478 } else if (too_many_output_actions(&ctx.xout->odp_actions)) {
3479 COVERAGE_INC(xlate_actions_too_many_output);
3480 ctx.xout->slow |= SLOW_ACTION;
3483 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3484 if (ctx.xin->resubmit_stats) {
3485 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3486 ctx.xin->resubmit_stats->n_packets,
3487 ctx.xin->resubmit_stats->n_bytes);
3489 if (ctx.xin->xcache) {
3490 struct xc_entry *entry;
3492 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
3493 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
3494 entry->u.mirror.mirrors = xout->mirrors;
3498 if (ctx.xbridge->netflow) {
3499 /* Only update netflow if we don't have controller flow. We don't
3500 * report NetFlow expiration messages for such facets because they
3501 * are just part of the control logic for the network, not real
3503 if (ofpacts_len == 0
3504 || ofpacts->type != OFPACT_CONTROLLER
3505 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3506 if (ctx.xin->resubmit_stats) {
3507 netflow_flow_update(ctx.xbridge->netflow, flow,
3508 xout->nf_output_iface,
3509 ctx.xin->resubmit_stats);
3511 if (ctx.xin->xcache) {
3512 struct xc_entry *entry;
3514 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
3515 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
3516 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
3517 entry->u.nf.iface = xout->nf_output_iface;
3522 ofpbuf_uninit(&ctx.stack);
3523 ofpbuf_uninit(&ctx.action_set);
3525 /* Clear the metadata and register wildcard masks, because we won't
3526 * use non-header fields as part of the cache. */
3527 flow_wildcards_clear_non_packet_fields(wc);
3529 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3530 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3531 * these fields. The datapath interface, on the other hand, represents
3532 * them with just 8 bits each. This means that if the high 8 bits of the
3533 * masks for these fields somehow become set, then they will get chopped
3534 * off by a round trip through the datapath, and revalidation will spot
3535 * that as an inconsistency and delete the flow. Avoid the problem here by
3536 * making sure that only the low 8 bits of either field can be unwildcarded
3540 wc->masks.tp_src &= htons(UINT8_MAX);
3541 wc->masks.tp_dst &= htons(UINT8_MAX);
3545 /* Sends 'packet' out 'ofport'.
3546 * May modify 'packet'.
3547 * Returns 0 if successful, otherwise a positive errno value. */
3549 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3551 struct xport *xport;
3552 struct ofpact_output output;
3555 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3556 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3557 flow_extract(packet, NULL, &flow);
3558 flow.in_port.ofp_port = OFPP_NONE;
3560 ovs_rwlock_rdlock(&xlate_rwlock);
3561 xport = xport_lookup(ofport);
3563 ovs_rwlock_unlock(&xlate_rwlock);
3566 output.port = xport->ofp_port;
3568 ovs_rwlock_unlock(&xlate_rwlock);
3570 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3571 &output.ofpact, sizeof output,
3575 struct xlate_cache *
3576 xlate_cache_new(void)
3578 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
3580 ofpbuf_init(&xcache->entries, 512);
3584 static struct xc_entry *
3585 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
3587 struct xc_entry *entry;
3589 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
3596 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
3598 if (entry->u.dev.tx) {
3599 netdev_vport_inc_tx(entry->u.dev.tx, stats);
3601 if (entry->u.dev.rx) {
3602 netdev_vport_inc_rx(entry->u.dev.rx, stats);
3604 if (entry->u.dev.bfd) {
3605 bfd_account_rx(entry->u.dev.bfd, stats);
3610 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
3612 struct xbridge *xbridge;
3613 struct xbundle *xbundle;
3614 struct flow_wildcards wc;
3616 xbridge = xbridge_lookup(ofproto);
3621 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
3627 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
3630 /* Push stats and perform side effects of flow translation. */
3632 xlate_push_stats(struct xlate_cache *xcache, bool may_learn,
3633 const struct dpif_flow_stats *stats)
3635 struct xc_entry *entry;
3636 struct ofpbuf entries = xcache->entries;
3638 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3639 switch (entry->type) {
3641 rule_dpif_credit_stats(entry->u.rule, stats);
3644 bond_account(entry->u.bond.bond, entry->u.bond.flow,
3645 entry->u.bond.vid, stats->n_bytes);
3648 xlate_cache_netdev(entry, stats);
3651 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
3652 entry->u.nf.iface, stats);
3655 mirror_update_stats(entry->u.mirror.mbridge,
3656 entry->u.mirror.mirrors,
3657 stats->n_packets, stats->n_bytes);
3661 ofproto_dpif_flow_mod(entry->u.learn.ofproto,
3666 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
3667 entry->u.normal.vlan);
3669 case XC_FIN_TIMEOUT:
3670 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
3671 entry->u.fin.idle, entry->u.fin.hard);
3680 xlate_dev_unref(struct xc_entry *entry)
3682 if (entry->u.dev.tx) {
3683 netdev_close(entry->u.dev.tx);
3685 if (entry->u.dev.rx) {
3686 netdev_close(entry->u.dev.rx);
3688 if (entry->u.dev.bfd) {
3689 bfd_unref(entry->u.dev.bfd);
3694 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
3696 netflow_flow_clear(netflow, flow);
3697 netflow_unref(netflow);
3702 xlate_cache_clear(struct xlate_cache *xcache)
3704 struct xc_entry *entry;
3705 struct ofpbuf entries;
3711 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3712 switch (entry->type) {
3714 rule_dpif_unref(entry->u.rule);
3717 free(entry->u.bond.flow);
3718 bond_unref(entry->u.bond.bond);
3721 xlate_dev_unref(entry);
3724 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
3727 mbridge_unref(entry->u.mirror.mbridge);
3730 free(entry->u.learn.fm);
3731 ofpbuf_delete(entry->u.learn.ofpacts);
3734 free(entry->u.normal.flow);
3736 case XC_FIN_TIMEOUT:
3737 /* 'u.fin.rule' is always already held as a XC_RULE, which
3738 * has already released it's reference above. */
3745 ofpbuf_clear(&xcache->entries);
3749 xlate_cache_delete(struct xlate_cache *xcache)
3751 xlate_cache_clear(xcache);
3752 ofpbuf_uninit(&xcache->entries);