1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
17 #include "ofproto/ofproto-dpif-xlate.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
41 #include "odp-execute.h"
42 #include "ofp-actions.h"
43 #include "ofproto/ofproto-dpif-ipfix.h"
44 #include "ofproto/ofproto-dpif-mirror.h"
45 #include "ofproto/ofproto-dpif-sflow.h"
46 #include "ofproto/ofproto-dpif.h"
47 #include "ofproto/ofproto-provider.h"
51 COVERAGE_DEFINE(xlate_actions
);
52 COVERAGE_DEFINE(xlate_actions_oversize
);
54 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate
);
56 /* Maximum depth of flow table recursion (due to resubmit actions) in a
57 * flow translation. */
58 #define MAX_RESUBMIT_RECURSION 64
60 /* Maximum number of resubmit actions in a flow translation, whether they are
61 * recursive or not. */
62 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
64 struct ovs_rwlock xlate_rwlock
= OVS_RWLOCK_INITIALIZER
;
67 struct hmap_node hmap_node
; /* Node in global 'xbridges' map. */
68 struct ofproto_dpif
*ofproto
; /* Key in global 'xbridges' map. */
70 struct list xbundles
; /* Owned xbundles. */
71 struct hmap xports
; /* Indexed by ofp_port. */
73 char *name
; /* Name used in log messages. */
74 struct dpif
*dpif
; /* Datapath interface. */
75 struct mac_learning
*ml
; /* Mac learning handle. */
76 struct mbridge
*mbridge
; /* Mirroring. */
77 struct dpif_sflow
*sflow
; /* SFlow handle, or null. */
78 struct dpif_ipfix
*ipfix
; /* Ipfix handle, or null. */
79 struct stp
*stp
; /* STP or null if disabled. */
81 /* Special rules installed by ofproto-dpif. */
82 struct rule_dpif
*miss_rule
;
83 struct rule_dpif
*no_packet_in_rule
;
85 enum ofp_config_flags frag
; /* Fragmentation handling. */
86 bool has_netflow
; /* Bridge runs netflow? */
87 bool has_in_band
; /* Bridge has in band control? */
88 bool forward_bpdu
; /* Bridge forwards STP BPDUs? */
92 struct hmap_node hmap_node
; /* In global 'xbundles' map. */
93 struct ofbundle
*ofbundle
; /* Key in global 'xbundles' map. */
95 struct list list_node
; /* In parent 'xbridges' list. */
96 struct xbridge
*xbridge
; /* Parent xbridge. */
98 struct list xports
; /* Contains "struct xport"s. */
100 char *name
; /* Name used in log messages. */
101 struct bond
*bond
; /* Nonnull iff more than one port. */
102 struct lacp
*lacp
; /* LACP handle or null. */
104 enum port_vlan_mode vlan_mode
; /* VLAN mode. */
105 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
106 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
107 * NULL if all VLANs are trunked. */
108 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
109 bool floodable
; /* No port has OFPUTIL_PC_NO_FLOOD set? */
113 struct hmap_node hmap_node
; /* Node in global 'xports' map. */
114 struct ofport_dpif
*ofport
; /* Key in global 'xports map. */
116 struct hmap_node ofp_node
; /* Node in parent xbridge 'xports' map. */
117 ofp_port_t ofp_port
; /* Key in parent xbridge 'xports' map. */
119 odp_port_t odp_port
; /* Datapath port number or ODPP_NONE. */
121 struct list bundle_node
; /* In parent xbundle (if it exists). */
122 struct xbundle
*xbundle
; /* Parent xbundle or null. */
124 struct netdev
*netdev
; /* 'ofport''s netdev. */
126 struct xbridge
*xbridge
; /* Parent bridge. */
127 struct xport
*peer
; /* Patch port peer or null. */
129 enum ofputil_port_config config
; /* OpenFlow port configuration. */
130 int stp_port_no
; /* STP port number or -1 if not in use. */
132 struct hmap skb_priorities
; /* Map of 'skb_priority_to_dscp's. */
134 bool may_enable
; /* May be enabled in bonds. */
135 bool is_tunnel
; /* Is a tunnel port. */
137 struct cfm
*cfm
; /* CFM handle or null. */
138 struct bfd
*bfd
; /* BFD handle or null. */
142 struct xlate_in
*xin
;
143 struct xlate_out
*xout
;
145 const struct xbridge
*xbridge
;
147 /* Flow at the last commit. */
148 struct flow base_flow
;
150 /* Tunnel IP destination address as received. This is stored separately
151 * as the base_flow.tunnel is cleared on init to reflect the datapath
152 * behavior. Used to make sure not to send tunneled output to ourselves,
153 * which might lead to an infinite loop. This could happen easily
154 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
155 * actually set the tun_dst field. */
156 ovs_be32 orig_tunnel_ip_dst
;
158 /* Stack for the push and pop actions. Each stack element is of type
159 * "union mf_subvalue". */
160 union mf_subvalue init_stack
[1024 / sizeof(union mf_subvalue
)];
163 /* The rule that we are currently translating, or NULL. */
164 struct rule_dpif
*rule
;
166 int mpls_depth_delta
; /* Delta of the mpls stack depth since
167 * actions were last committed.
168 * Must be between -1 and 1 inclusive. */
169 ovs_be32 pre_push_mpls_lse
; /* Used to record the top-most MPLS LSE
170 * prior to an mpls_push so that it may be
171 * used for a subsequent mpls_pop. */
173 /* Resubmit statistics, via xlate_table_action(). */
174 int recurse
; /* Current resubmit nesting depth. */
175 int resubmits
; /* Total number of resubmits. */
177 uint32_t orig_skb_priority
; /* Priority when packet arrived. */
178 uint8_t table_id
; /* OpenFlow table ID where flow was found. */
179 uint32_t sflow_n_outputs
; /* Number of output ports. */
180 odp_port_t sflow_odp_port
; /* Output port for composing sFlow action. */
181 uint16_t user_cookie_offset
;/* Used for user_action_cookie fixup. */
182 bool exit
; /* No further actions should be processed. */
184 /* OpenFlow 1.1+ action set.
186 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
187 * When translation is otherwise complete, ofpacts_execute_action_set()
188 * converts it to a set of "struct ofpact"s that can be translated into
189 * datapath actions. */
190 struct ofpbuf action_set
; /* Action set. */
191 uint64_t action_set_stub
[1024 / 8];
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it.
199 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
200 static struct xbundle ofpp_none_bundle
;
202 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
203 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
204 * traffic egressing the 'ofport' with that priority should be marked with. */
205 struct skb_priority_to_dscp
{
206 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'skb_priorities'. */
207 uint32_t skb_priority
; /* Priority of this queue (see struct flow). */
209 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
212 static struct hmap xbridges
= HMAP_INITIALIZER(&xbridges
);
213 static struct hmap xbundles
= HMAP_INITIALIZER(&xbundles
);
214 static struct hmap xports
= HMAP_INITIALIZER(&xports
);
216 static bool may_receive(const struct xport
*, struct xlate_ctx
*);
217 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
219 static void xlate_actions__(struct xlate_in
*, struct xlate_out
*)
220 OVS_REQ_RDLOCK(xlate_rwlock
);
221 static void xlate_normal(struct xlate_ctx
*);
222 static void xlate_report(struct xlate_ctx
*, const char *);
223 static void xlate_table_action(struct xlate_ctx
*, ofp_port_t in_port
,
224 uint8_t table_id
, bool may_packet_in
);
225 static bool input_vid_is_valid(uint16_t vid
, struct xbundle
*, bool warn
);
226 static uint16_t input_vid_to_vlan(const struct xbundle
*, uint16_t vid
);
227 static void output_normal(struct xlate_ctx
*, const struct xbundle
*,
229 static void compose_output_action(struct xlate_ctx
*, ofp_port_t ofp_port
);
231 static struct xbridge
*xbridge_lookup(const struct ofproto_dpif
*);
232 static struct xbundle
*xbundle_lookup(const struct ofbundle
*);
233 static struct xport
*xport_lookup(const struct ofport_dpif
*);
234 static struct xport
*get_ofp_port(const struct xbridge
*, ofp_port_t ofp_port
);
235 static struct skb_priority_to_dscp
*get_skb_priority(const struct xport
*,
236 uint32_t skb_priority
);
237 static void clear_skb_priorities(struct xport
*);
238 static bool dscp_from_skb_priority(const struct xport
*, uint32_t skb_priority
,
242 xlate_ofproto_set(struct ofproto_dpif
*ofproto
, const char *name
,
243 struct dpif
*dpif
, struct rule_dpif
*miss_rule
,
244 struct rule_dpif
*no_packet_in_rule
,
245 const struct mac_learning
*ml
, struct stp
*stp
,
246 const struct mbridge
*mbridge
,
247 const struct dpif_sflow
*sflow
,
248 const struct dpif_ipfix
*ipfix
, enum ofp_config_flags frag
,
249 bool forward_bpdu
, bool has_in_band
, bool has_netflow
)
251 struct xbridge
*xbridge
= xbridge_lookup(ofproto
);
254 xbridge
= xzalloc(sizeof *xbridge
);
255 xbridge
->ofproto
= ofproto
;
257 hmap_insert(&xbridges
, &xbridge
->hmap_node
, hash_pointer(ofproto
, 0));
258 hmap_init(&xbridge
->xports
);
259 list_init(&xbridge
->xbundles
);
262 if (xbridge
->ml
!= ml
) {
263 mac_learning_unref(xbridge
->ml
);
264 xbridge
->ml
= mac_learning_ref(ml
);
267 if (xbridge
->mbridge
!= mbridge
) {
268 mbridge_unref(xbridge
->mbridge
);
269 xbridge
->mbridge
= mbridge_ref(mbridge
);
272 if (xbridge
->sflow
!= sflow
) {
273 dpif_sflow_unref(xbridge
->sflow
);
274 xbridge
->sflow
= dpif_sflow_ref(sflow
);
277 if (xbridge
->ipfix
!= ipfix
) {
278 dpif_ipfix_unref(xbridge
->ipfix
);
279 xbridge
->ipfix
= dpif_ipfix_ref(ipfix
);
282 if (xbridge
->stp
!= stp
) {
283 stp_unref(xbridge
->stp
);
284 xbridge
->stp
= stp_ref(stp
);
288 xbridge
->name
= xstrdup(name
);
290 xbridge
->dpif
= dpif
;
291 xbridge
->forward_bpdu
= forward_bpdu
;
292 xbridge
->has_in_band
= has_in_band
;
293 xbridge
->has_netflow
= has_netflow
;
294 xbridge
->frag
= frag
;
295 xbridge
->miss_rule
= miss_rule
;
296 xbridge
->no_packet_in_rule
= no_packet_in_rule
;
300 xlate_remove_ofproto(struct ofproto_dpif
*ofproto
)
302 struct xbridge
*xbridge
= xbridge_lookup(ofproto
);
303 struct xbundle
*xbundle
, *next_xbundle
;
304 struct xport
*xport
, *next_xport
;
310 HMAP_FOR_EACH_SAFE (xport
, next_xport
, ofp_node
, &xbridge
->xports
) {
311 xlate_ofport_remove(xport
->ofport
);
314 LIST_FOR_EACH_SAFE (xbundle
, next_xbundle
, list_node
, &xbridge
->xbundles
) {
315 xlate_bundle_remove(xbundle
->ofbundle
);
318 hmap_remove(&xbridges
, &xbridge
->hmap_node
);
319 mac_learning_unref(xbridge
->ml
);
320 mbridge_unref(xbridge
->mbridge
);
321 dpif_sflow_unref(xbridge
->sflow
);
322 dpif_ipfix_unref(xbridge
->ipfix
);
323 stp_unref(xbridge
->stp
);
324 hmap_destroy(&xbridge
->xports
);
330 xlate_bundle_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
331 const char *name
, enum port_vlan_mode vlan_mode
, int vlan
,
332 unsigned long *trunks
, bool use_priority_tags
,
333 const struct bond
*bond
, const struct lacp
*lacp
,
336 struct xbundle
*xbundle
= xbundle_lookup(ofbundle
);
339 xbundle
= xzalloc(sizeof *xbundle
);
340 xbundle
->ofbundle
= ofbundle
;
341 xbundle
->xbridge
= xbridge_lookup(ofproto
);
343 hmap_insert(&xbundles
, &xbundle
->hmap_node
, hash_pointer(ofbundle
, 0));
344 list_insert(&xbundle
->xbridge
->xbundles
, &xbundle
->list_node
);
345 list_init(&xbundle
->xports
);
348 ovs_assert(xbundle
->xbridge
);
351 xbundle
->name
= xstrdup(name
);
353 xbundle
->vlan_mode
= vlan_mode
;
354 xbundle
->vlan
= vlan
;
355 xbundle
->trunks
= trunks
;
356 xbundle
->use_priority_tags
= use_priority_tags
;
357 xbundle
->floodable
= floodable
;
359 if (xbundle
->bond
!= bond
) {
360 bond_unref(xbundle
->bond
);
361 xbundle
->bond
= bond_ref(bond
);
364 if (xbundle
->lacp
!= lacp
) {
365 lacp_unref(xbundle
->lacp
);
366 xbundle
->lacp
= lacp_ref(lacp
);
371 xlate_bundle_remove(struct ofbundle
*ofbundle
)
373 struct xbundle
*xbundle
= xbundle_lookup(ofbundle
);
374 struct xport
*xport
, *next
;
380 LIST_FOR_EACH_SAFE (xport
, next
, bundle_node
, &xbundle
->xports
) {
381 list_remove(&xport
->bundle_node
);
382 xport
->xbundle
= NULL
;
385 hmap_remove(&xbundles
, &xbundle
->hmap_node
);
386 list_remove(&xbundle
->list_node
);
387 bond_unref(xbundle
->bond
);
388 lacp_unref(xbundle
->lacp
);
394 xlate_ofport_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
395 struct ofport_dpif
*ofport
, ofp_port_t ofp_port
,
396 odp_port_t odp_port
, const struct netdev
*netdev
,
397 const struct cfm
*cfm
, const struct bfd
*bfd
,
398 struct ofport_dpif
*peer
, int stp_port_no
,
399 const struct ofproto_port_queue
*qdscp_list
, size_t n_qdscp
,
400 enum ofputil_port_config config
, bool is_tunnel
,
403 struct xport
*xport
= xport_lookup(ofport
);
407 xport
= xzalloc(sizeof *xport
);
408 xport
->ofport
= ofport
;
409 xport
->xbridge
= xbridge_lookup(ofproto
);
410 xport
->ofp_port
= ofp_port
;
412 hmap_init(&xport
->skb_priorities
);
413 hmap_insert(&xports
, &xport
->hmap_node
, hash_pointer(ofport
, 0));
414 hmap_insert(&xport
->xbridge
->xports
, &xport
->ofp_node
,
415 hash_ofp_port(xport
->ofp_port
));
418 ovs_assert(xport
->ofp_port
== ofp_port
);
420 xport
->config
= config
;
421 xport
->stp_port_no
= stp_port_no
;
422 xport
->is_tunnel
= is_tunnel
;
423 xport
->may_enable
= may_enable
;
424 xport
->odp_port
= odp_port
;
426 if (xport
->netdev
!= netdev
) {
427 netdev_close(xport
->netdev
);
428 xport
->netdev
= netdev_ref(netdev
);
431 if (xport
->cfm
!= cfm
) {
432 cfm_unref(xport
->cfm
);
433 xport
->cfm
= cfm_ref(cfm
);
436 if (xport
->bfd
!= bfd
) {
437 bfd_unref(xport
->bfd
);
438 xport
->bfd
= bfd_ref(bfd
);
442 xport
->peer
->peer
= NULL
;
444 xport
->peer
= xport_lookup(peer
);
446 xport
->peer
->peer
= xport
;
449 if (xport
->xbundle
) {
450 list_remove(&xport
->bundle_node
);
452 xport
->xbundle
= xbundle_lookup(ofbundle
);
453 if (xport
->xbundle
) {
454 list_insert(&xport
->xbundle
->xports
, &xport
->bundle_node
);
457 clear_skb_priorities(xport
);
458 for (i
= 0; i
< n_qdscp
; i
++) {
459 struct skb_priority_to_dscp
*pdscp
;
460 uint32_t skb_priority
;
462 if (dpif_queue_to_priority(xport
->xbridge
->dpif
, qdscp_list
[i
].queue
,
467 pdscp
= xmalloc(sizeof *pdscp
);
468 pdscp
->skb_priority
= skb_priority
;
469 pdscp
->dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
470 hmap_insert(&xport
->skb_priorities
, &pdscp
->hmap_node
,
471 hash_int(pdscp
->skb_priority
, 0));
476 xlate_ofport_remove(struct ofport_dpif
*ofport
)
478 struct xport
*xport
= xport_lookup(ofport
);
485 xport
->peer
->peer
= NULL
;
489 if (xport
->xbundle
) {
490 list_remove(&xport
->bundle_node
);
493 clear_skb_priorities(xport
);
494 hmap_destroy(&xport
->skb_priorities
);
496 hmap_remove(&xports
, &xport
->hmap_node
);
497 hmap_remove(&xport
->xbridge
->xports
, &xport
->ofp_node
);
499 netdev_close(xport
->netdev
);
500 cfm_unref(xport
->cfm
);
501 bfd_unref(xport
->bfd
);
505 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
506 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
507 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
508 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
509 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
510 * 'packet' ingressed.
512 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
513 * 'flow''s in_port to OFPP_NONE.
515 * This function does post-processing on data returned from
516 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
517 * of the upcall processing logic. In particular, if the extracted in_port is
518 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
519 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
520 * a VLAN header onto 'packet' (if it is nonnull).
522 * Similarly, this function also includes some logic to help with tunnels. It
523 * may modify 'flow' as necessary to make the tunneling implementation
524 * transparent to the upcall processing logic.
526 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
527 * or some other positive errno if there are other problems. */
529 xlate_receive(const struct dpif_backer
*backer
, struct ofpbuf
*packet
,
530 const struct nlattr
*key
, size_t key_len
,
531 struct flow
*flow
, enum odp_key_fitness
*fitnessp
,
532 struct ofproto_dpif
**ofproto
, odp_port_t
*odp_in_port
)
534 enum odp_key_fitness fitness
;
535 const struct xport
*xport
;
538 ovs_rwlock_rdlock(&xlate_rwlock
);
539 fitness
= odp_flow_key_to_flow(key
, key_len
, flow
);
540 if (fitness
== ODP_FIT_ERROR
) {
546 *odp_in_port
= flow
->in_port
.odp_port
;
549 xport
= xport_lookup(tnl_port_should_receive(flow
)
550 ? tnl_port_receive(flow
)
551 : odp_port_to_ofport(backer
, flow
->in_port
.odp_port
));
553 flow
->in_port
.ofp_port
= xport
? xport
->ofp_port
: OFPP_NONE
;
558 if (vsp_adjust_flow(xport
->xbridge
->ofproto
, flow
)) {
560 /* Make the packet resemble the flow, so that it gets sent to
561 * an OpenFlow controller properly, so that it looks correct
562 * for sFlow, and so that flow_extract() will get the correct
563 * vlan_tci if it is called on 'packet'.
565 * The allocated space inside 'packet' probably also contains
566 * 'key', that is, both 'packet' and 'key' are probably part of
567 * a struct dpif_upcall (see the large comment on that
568 * structure definition), so pushing data on 'packet' is in
569 * general not a good idea since it could overwrite 'key' or
570 * free it as a side effect. However, it's OK in this special
571 * case because we know that 'packet' is inside a Netlink
572 * attribute: pushing 4 bytes will just overwrite the 4-byte
573 * "struct nlattr", which is fine since we don't need that
575 eth_push_vlan(packet
, flow
->vlan_tci
);
577 /* We can't reproduce 'key' from 'flow'. */
578 fitness
= fitness
== ODP_FIT_PERFECT
? ODP_FIT_TOO_MUCH
: fitness
;
583 *ofproto
= xport
->xbridge
->ofproto
;
590 ovs_rwlock_unlock(&xlate_rwlock
);
594 static struct xbridge
*
595 xbridge_lookup(const struct ofproto_dpif
*ofproto
)
597 struct xbridge
*xbridge
;
603 HMAP_FOR_EACH_IN_BUCKET (xbridge
, hmap_node
, hash_pointer(ofproto
, 0),
605 if (xbridge
->ofproto
== ofproto
) {
612 static struct xbundle
*
613 xbundle_lookup(const struct ofbundle
*ofbundle
)
615 struct xbundle
*xbundle
;
621 HMAP_FOR_EACH_IN_BUCKET (xbundle
, hmap_node
, hash_pointer(ofbundle
, 0),
623 if (xbundle
->ofbundle
== ofbundle
) {
630 static struct xport
*
631 xport_lookup(const struct ofport_dpif
*ofport
)
639 HMAP_FOR_EACH_IN_BUCKET (xport
, hmap_node
, hash_pointer(ofport
, 0),
641 if (xport
->ofport
== ofport
) {
648 static struct stp_port
*
649 xport_get_stp_port(const struct xport
*xport
)
651 return xport
->xbridge
->stp
&& xport
->stp_port_no
!= -1
652 ? stp_get_port(xport
->xbridge
->stp
, xport
->stp_port_no
)
656 static enum stp_state
657 xport_stp_learn_state(const struct xport
*xport
)
659 struct stp_port
*sp
= xport_get_stp_port(xport
);
660 return stp_learn_in_state(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
664 xport_stp_forward_state(const struct xport
*xport
)
666 struct stp_port
*sp
= xport_get_stp_port(xport
);
667 return stp_forward_in_state(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
670 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
671 * were used to make the determination.*/
673 stp_should_process_flow(const struct flow
*flow
, struct flow_wildcards
*wc
)
675 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
676 return eth_addr_equals(flow
->dl_dst
, eth_addr_stp
);
680 stp_process_packet(const struct xport
*xport
, const struct ofpbuf
*packet
)
682 struct stp_port
*sp
= xport_get_stp_port(xport
);
683 struct ofpbuf payload
= *packet
;
684 struct eth_header
*eth
= payload
.data
;
686 /* Sink packets on ports that have STP disabled when the bridge has
688 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
692 /* Trim off padding on payload. */
693 if (payload
.size
> ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
694 payload
.size
= ntohs(eth
->eth_type
) + ETH_HEADER_LEN
;
697 if (ofpbuf_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
698 stp_received_bpdu(sp
, payload
.data
, payload
.size
);
702 static struct xport
*
703 get_ofp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
707 HMAP_FOR_EACH_IN_BUCKET (xport
, ofp_node
, hash_ofp_port(ofp_port
),
709 if (xport
->ofp_port
== ofp_port
) {
717 ofp_port_to_odp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
719 const struct xport
*xport
= get_ofp_port(xbridge
, ofp_port
);
720 return xport
? xport
->odp_port
: ODPP_NONE
;
724 xbundle_trunks_vlan(const struct xbundle
*bundle
, uint16_t vlan
)
726 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
727 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
731 xbundle_includes_vlan(const struct xbundle
*xbundle
, uint16_t vlan
)
733 return vlan
== xbundle
->vlan
|| xbundle_trunks_vlan(xbundle
, vlan
);
737 xbundle_mirror_out(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
739 return xbundle
!= &ofpp_none_bundle
740 ? mirror_bundle_out(xbridge
->mbridge
, xbundle
->ofbundle
)
745 xbundle_mirror_src(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
747 return xbundle
!= &ofpp_none_bundle
748 ? mirror_bundle_src(xbridge
->mbridge
, xbundle
->ofbundle
)
753 xbundle_mirror_dst(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
755 return xbundle
!= &ofpp_none_bundle
756 ? mirror_bundle_dst(xbridge
->mbridge
, xbundle
->ofbundle
)
760 static struct xbundle
*
761 lookup_input_bundle(const struct xbridge
*xbridge
, ofp_port_t in_port
,
762 bool warn
, struct xport
**in_xportp
)
766 /* Find the port and bundle for the received packet. */
767 xport
= get_ofp_port(xbridge
, in_port
);
771 if (xport
&& xport
->xbundle
) {
772 return xport
->xbundle
;
775 /* Special-case OFPP_NONE, which a controller may use as the ingress
776 * port for traffic that it is sourcing. */
777 if (in_port
== OFPP_NONE
) {
778 ofpp_none_bundle
.name
= "OFPP_NONE";
779 ofpp_none_bundle
.vlan_mode
= PORT_VLAN_TRUNK
;
780 return &ofpp_none_bundle
;
783 /* Odd. A few possible reasons here:
785 * - We deleted a port but there are still a few packets queued up
788 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
789 * we don't know about.
791 * - The ofproto client didn't configure the port as part of a bundle.
792 * This is particularly likely to happen if a packet was received on the
793 * port after it was created, but before the client had a chance to
794 * configure its bundle.
797 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
799 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
800 "port %"PRIu16
, xbridge
->name
, in_port
);
806 add_mirror_actions(struct xlate_ctx
*ctx
, const struct flow
*orig_flow
)
808 const struct xbridge
*xbridge
= ctx
->xbridge
;
809 mirror_mask_t mirrors
;
810 struct xbundle
*in_xbundle
;
814 mirrors
= ctx
->xout
->mirrors
;
815 ctx
->xout
->mirrors
= 0;
817 in_xbundle
= lookup_input_bundle(xbridge
, orig_flow
->in_port
.ofp_port
,
818 ctx
->xin
->packet
!= NULL
, NULL
);
822 mirrors
|= xbundle_mirror_src(xbridge
, in_xbundle
);
824 /* Drop frames on bundles reserved for mirroring. */
825 if (xbundle_mirror_out(xbridge
, in_xbundle
)) {
826 if (ctx
->xin
->packet
!= NULL
) {
827 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
828 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
829 "%s, which is reserved exclusively for mirroring",
830 ctx
->xbridge
->name
, in_xbundle
->name
);
832 ofpbuf_clear(&ctx
->xout
->odp_actions
);
837 vid
= vlan_tci_to_vid(orig_flow
->vlan_tci
);
838 if (!input_vid_is_valid(vid
, in_xbundle
, ctx
->xin
->packet
!= NULL
)) {
841 vlan
= input_vid_to_vlan(in_xbundle
, vid
);
847 /* Restore the original packet before adding the mirror actions. */
848 ctx
->xin
->flow
= *orig_flow
;
851 mirror_mask_t dup_mirrors
;
852 struct ofbundle
*out
;
853 unsigned long *vlans
;
858 has_mirror
= mirror_get(xbridge
->mbridge
, mirror_mask_ffs(mirrors
) - 1,
859 &vlans
, &dup_mirrors
, &out
, &out_vlan
);
860 ovs_assert(has_mirror
);
863 ctx
->xout
->wc
.masks
.vlan_tci
|= htons(VLAN_CFI
| VLAN_VID_MASK
);
865 vlan_mirrored
= !vlans
|| bitmap_is_set(vlans
, vlan
);
868 if (!vlan_mirrored
) {
869 mirrors
= zero_rightmost_1bit(mirrors
);
873 mirrors
&= ~dup_mirrors
;
874 ctx
->xout
->mirrors
|= dup_mirrors
;
876 struct xbundle
*out_xbundle
= xbundle_lookup(out
);
878 output_normal(ctx
, out_xbundle
, vlan
);
880 } else if (vlan
!= out_vlan
881 && !eth_addr_is_reserved(orig_flow
->dl_dst
)) {
882 struct xbundle
*xbundle
;
884 LIST_FOR_EACH (xbundle
, list_node
, &xbridge
->xbundles
) {
885 if (xbundle_includes_vlan(xbundle
, out_vlan
)
886 && !xbundle_mirror_out(xbridge
, xbundle
)) {
887 output_normal(ctx
, xbundle
, out_vlan
);
894 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
895 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
896 * the bundle on which the packet was received, returns the VLAN to which the
899 * Both 'vid' and the return value are in the range 0...4095. */
901 input_vid_to_vlan(const struct xbundle
*in_xbundle
, uint16_t vid
)
903 switch (in_xbundle
->vlan_mode
) {
904 case PORT_VLAN_ACCESS
:
905 return in_xbundle
->vlan
;
908 case PORT_VLAN_TRUNK
:
911 case PORT_VLAN_NATIVE_UNTAGGED
:
912 case PORT_VLAN_NATIVE_TAGGED
:
913 return vid
? vid
: in_xbundle
->vlan
;
920 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
921 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
924 * 'vid' should be the VID obtained from the 802.1Q header that was received as
925 * part of a packet (specify 0 if there was no 802.1Q header), in the range
928 input_vid_is_valid(uint16_t vid
, struct xbundle
*in_xbundle
, bool warn
)
930 /* Allow any VID on the OFPP_NONE port. */
931 if (in_xbundle
== &ofpp_none_bundle
) {
935 switch (in_xbundle
->vlan_mode
) {
936 case PORT_VLAN_ACCESS
:
939 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
940 VLOG_WARN_RL(&rl
, "dropping VLAN %"PRIu16
" tagged "
941 "packet received on port %s configured as VLAN "
942 "%"PRIu16
" access port", vid
, in_xbundle
->name
,
949 case PORT_VLAN_NATIVE_UNTAGGED
:
950 case PORT_VLAN_NATIVE_TAGGED
:
952 /* Port must always carry its native VLAN. */
956 case PORT_VLAN_TRUNK
:
957 if (!xbundle_includes_vlan(in_xbundle
, vid
)) {
959 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
960 VLOG_WARN_RL(&rl
, "dropping VLAN %"PRIu16
" packet "
961 "received on port %s not configured for trunking "
962 "VLAN %"PRIu16
, vid
, in_xbundle
->name
, vid
);
974 /* Given 'vlan', the VLAN that a packet belongs to, and
975 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
976 * that should be included in the 802.1Q header. (If the return value is 0,
977 * then the 802.1Q header should only be included in the packet if there is a
980 * Both 'vlan' and the return value are in the range 0...4095. */
982 output_vlan_to_vid(const struct xbundle
*out_xbundle
, uint16_t vlan
)
984 switch (out_xbundle
->vlan_mode
) {
985 case PORT_VLAN_ACCESS
:
988 case PORT_VLAN_TRUNK
:
989 case PORT_VLAN_NATIVE_TAGGED
:
992 case PORT_VLAN_NATIVE_UNTAGGED
:
993 return vlan
== out_xbundle
->vlan
? 0 : vlan
;
1001 output_normal(struct xlate_ctx
*ctx
, const struct xbundle
*out_xbundle
,
1004 ovs_be16
*flow_tci
= &ctx
->xin
->flow
.vlan_tci
;
1006 ovs_be16 tci
, old_tci
;
1007 struct xport
*xport
;
1009 vid
= output_vlan_to_vid(out_xbundle
, vlan
);
1010 if (list_is_empty(&out_xbundle
->xports
)) {
1011 /* Partially configured bundle with no slaves. Drop the packet. */
1013 } else if (!out_xbundle
->bond
) {
1014 xport
= CONTAINER_OF(list_front(&out_xbundle
->xports
), struct xport
,
1017 struct ofport_dpif
*ofport
;
1019 ofport
= bond_choose_output_slave(out_xbundle
->bond
, &ctx
->xin
->flow
,
1020 &ctx
->xout
->wc
, vid
);
1021 xport
= xport_lookup(ofport
);
1024 /* No slaves enabled, so drop packet. */
1029 old_tci
= *flow_tci
;
1031 if (tci
|| out_xbundle
->use_priority_tags
) {
1032 tci
|= *flow_tci
& htons(VLAN_PCP_MASK
);
1034 tci
|= htons(VLAN_CFI
);
1039 compose_output_action(ctx
, xport
->ofp_port
);
1040 *flow_tci
= old_tci
;
1043 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1044 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1045 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1047 is_gratuitous_arp(const struct flow
*flow
, struct flow_wildcards
*wc
)
1049 if (flow
->dl_type
!= htons(ETH_TYPE_ARP
)) {
1053 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1054 if (!eth_addr_is_broadcast(flow
->dl_dst
)) {
1058 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
1059 if (flow
->nw_proto
== ARP_OP_REPLY
) {
1061 } else if (flow
->nw_proto
== ARP_OP_REQUEST
) {
1062 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
1063 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
1065 return flow
->nw_src
== flow
->nw_dst
;
1071 /* Checks whether a MAC learning update is necessary for MAC learning table
1072 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1075 * Most packets processed through the MAC learning table do not actually
1076 * change it in any way. This function requires only a read lock on the MAC
1077 * learning table, so it is much cheaper in this common case.
1079 * Keep the code here synchronized with that in update_learning_table__()
1082 is_mac_learning_update_needed(const struct mac_learning
*ml
,
1083 const struct flow
*flow
,
1084 struct flow_wildcards
*wc
,
1085 int vlan
, struct xbundle
*in_xbundle
)
1086 OVS_REQ_RDLOCK(ml
->rwlock
)
1088 struct mac_entry
*mac
;
1090 if (!mac_learning_may_learn(ml
, flow
->dl_src
, vlan
)) {
1094 mac
= mac_learning_lookup(ml
, flow
->dl_src
, vlan
);
1095 if (!mac
|| mac_entry_age(ml
, mac
)) {
1099 if (is_gratuitous_arp(flow
, wc
)) {
1100 /* We don't want to learn from gratuitous ARP packets that are
1101 * reflected back over bond slaves so we lock the learning table. */
1102 if (!in_xbundle
->bond
) {
1104 } else if (mac_entry_is_grat_arp_locked(mac
)) {
1109 return mac
->port
.p
!= in_xbundle
->ofbundle
;
1113 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1114 * received on 'in_xbundle' in 'vlan'.
1116 * This code repeats all the checks in is_mac_learning_update_needed() because
1117 * the lock was released between there and here and thus the MAC learning state
1118 * could have changed.
1120 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1123 update_learning_table__(const struct xbridge
*xbridge
,
1124 const struct flow
*flow
, struct flow_wildcards
*wc
,
1125 int vlan
, struct xbundle
*in_xbundle
)
1126 OVS_REQ_WRLOCK(xbridge
->ml
->rwlock
)
1128 struct mac_entry
*mac
;
1130 if (!mac_learning_may_learn(xbridge
->ml
, flow
->dl_src
, vlan
)) {
1134 mac
= mac_learning_insert(xbridge
->ml
, flow
->dl_src
, vlan
);
1135 if (is_gratuitous_arp(flow
, wc
)) {
1136 /* We don't want to learn from gratuitous ARP packets that are
1137 * reflected back over bond slaves so we lock the learning table. */
1138 if (!in_xbundle
->bond
) {
1139 mac_entry_set_grat_arp_lock(mac
);
1140 } else if (mac_entry_is_grat_arp_locked(mac
)) {
1145 if (mac
->port
.p
!= in_xbundle
->ofbundle
) {
1146 /* The log messages here could actually be useful in debugging,
1147 * so keep the rate limit relatively high. */
1148 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
1150 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
1151 "on port %s in VLAN %d",
1152 xbridge
->name
, ETH_ADDR_ARGS(flow
->dl_src
),
1153 in_xbundle
->name
, vlan
);
1155 mac
->port
.p
= in_xbundle
->ofbundle
;
1156 mac_learning_changed(xbridge
->ml
);
1161 update_learning_table(const struct xbridge
*xbridge
,
1162 const struct flow
*flow
, struct flow_wildcards
*wc
,
1163 int vlan
, struct xbundle
*in_xbundle
)
1167 /* Don't learn the OFPP_NONE port. */
1168 if (in_xbundle
== &ofpp_none_bundle
) {
1172 /* First try the common case: no change to MAC learning table. */
1173 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
1174 need_update
= is_mac_learning_update_needed(xbridge
->ml
, flow
, wc
, vlan
,
1176 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1179 /* Slow path: MAC learning table might need an update. */
1180 ovs_rwlock_wrlock(&xbridge
->ml
->rwlock
);
1181 update_learning_table__(xbridge
, flow
, wc
, vlan
, in_xbundle
);
1182 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1186 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1187 * dropped. Returns true if they may be forwarded, false if they should be
1190 * 'in_port' must be the xport that corresponds to flow->in_port.
1191 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1193 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1194 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1195 * checked by input_vid_is_valid().
1197 * May also add tags to '*tags', although the current implementation only does
1198 * so in one special case.
1201 is_admissible(struct xlate_ctx
*ctx
, struct xport
*in_port
,
1204 struct xbundle
*in_xbundle
= in_port
->xbundle
;
1205 const struct xbridge
*xbridge
= ctx
->xbridge
;
1206 struct flow
*flow
= &ctx
->xin
->flow
;
1208 /* Drop frames for reserved multicast addresses
1209 * only if forward_bpdu option is absent. */
1210 if (!xbridge
->forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
1211 xlate_report(ctx
, "packet has reserved destination MAC, dropping");
1215 if (in_xbundle
->bond
) {
1216 struct mac_entry
*mac
;
1218 switch (bond_check_admissibility(in_xbundle
->bond
, in_port
->ofport
,
1224 xlate_report(ctx
, "bonding refused admissibility, dropping");
1227 case BV_DROP_IF_MOVED
:
1228 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
1229 mac
= mac_learning_lookup(xbridge
->ml
, flow
->dl_src
, vlan
);
1230 if (mac
&& mac
->port
.p
!= in_xbundle
->ofbundle
&&
1231 (!is_gratuitous_arp(flow
, &ctx
->xout
->wc
)
1232 || mac_entry_is_grat_arp_locked(mac
))) {
1233 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1234 xlate_report(ctx
, "SLB bond thinks this packet looped back, "
1238 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1247 xlate_normal(struct xlate_ctx
*ctx
)
1249 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1250 struct flow
*flow
= &ctx
->xin
->flow
;
1251 struct xbundle
*in_xbundle
;
1252 struct xport
*in_port
;
1253 struct mac_entry
*mac
;
1258 ctx
->xout
->has_normal
= true;
1260 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1261 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1262 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1264 in_xbundle
= lookup_input_bundle(ctx
->xbridge
, flow
->in_port
.ofp_port
,
1265 ctx
->xin
->packet
!= NULL
, &in_port
);
1267 xlate_report(ctx
, "no input bundle, dropping");
1271 /* Drop malformed frames. */
1272 if (flow
->dl_type
== htons(ETH_TYPE_VLAN
) &&
1273 !(flow
->vlan_tci
& htons(VLAN_CFI
))) {
1274 if (ctx
->xin
->packet
!= NULL
) {
1275 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1276 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet with partial "
1277 "VLAN tag received on port %s",
1278 ctx
->xbridge
->name
, in_xbundle
->name
);
1280 xlate_report(ctx
, "partial VLAN tag, dropping");
1284 /* Drop frames on bundles reserved for mirroring. */
1285 if (xbundle_mirror_out(ctx
->xbridge
, in_xbundle
)) {
1286 if (ctx
->xin
->packet
!= NULL
) {
1287 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1288 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
1289 "%s, which is reserved exclusively for mirroring",
1290 ctx
->xbridge
->name
, in_xbundle
->name
);
1292 xlate_report(ctx
, "input port is mirror output port, dropping");
1297 vid
= vlan_tci_to_vid(flow
->vlan_tci
);
1298 if (!input_vid_is_valid(vid
, in_xbundle
, ctx
->xin
->packet
!= NULL
)) {
1299 xlate_report(ctx
, "disallowed VLAN VID for this input port, dropping");
1302 vlan
= input_vid_to_vlan(in_xbundle
, vid
);
1304 /* Check other admissibility requirements. */
1305 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
1309 /* Learn source MAC. */
1310 if (ctx
->xin
->may_learn
) {
1311 update_learning_table(ctx
->xbridge
, flow
, wc
, vlan
, in_xbundle
);
1314 /* Determine output bundle. */
1315 ovs_rwlock_rdlock(&ctx
->xbridge
->ml
->rwlock
);
1316 mac
= mac_learning_lookup(ctx
->xbridge
->ml
, flow
->dl_dst
, vlan
);
1317 mac_port
= mac
? mac
->port
.p
: NULL
;
1318 ovs_rwlock_unlock(&ctx
->xbridge
->ml
->rwlock
);
1321 struct xbundle
*mac_xbundle
= xbundle_lookup(mac_port
);
1322 if (mac_xbundle
&& mac_xbundle
!= in_xbundle
) {
1323 xlate_report(ctx
, "forwarding to learned port");
1324 output_normal(ctx
, mac_xbundle
, vlan
);
1325 } else if (!mac_xbundle
) {
1326 xlate_report(ctx
, "learned port is unknown, dropping");
1328 xlate_report(ctx
, "learned port is input port, dropping");
1331 struct xbundle
*xbundle
;
1333 xlate_report(ctx
, "no learned MAC for destination, flooding");
1334 LIST_FOR_EACH (xbundle
, list_node
, &ctx
->xbridge
->xbundles
) {
1335 if (xbundle
!= in_xbundle
1336 && xbundle_includes_vlan(xbundle
, vlan
)
1337 && xbundle
->floodable
1338 && !xbundle_mirror_out(ctx
->xbridge
, xbundle
)) {
1339 output_normal(ctx
, xbundle
, vlan
);
1342 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
1346 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1347 * the number of packets out of UINT32_MAX to sample. The given
1348 * cookie is passed back in the callback for each sampled packet.
1351 compose_sample_action(const struct xbridge
*xbridge
,
1352 struct ofpbuf
*odp_actions
,
1353 const struct flow
*flow
,
1354 const uint32_t probability
,
1355 const union user_action_cookie
*cookie
,
1356 const size_t cookie_size
)
1358 size_t sample_offset
, actions_offset
;
1359 odp_port_t odp_port
;
1363 sample_offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
1365 nl_msg_put_u32(odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
, probability
);
1367 actions_offset
= nl_msg_start_nested(odp_actions
, OVS_SAMPLE_ATTR_ACTIONS
);
1369 odp_port
= ofp_port_to_odp_port(xbridge
, flow
->in_port
.ofp_port
);
1370 pid
= dpif_port_get_pid(xbridge
->dpif
, odp_port
);
1371 cookie_offset
= odp_put_userspace_action(pid
, cookie
, cookie_size
, odp_actions
);
1373 nl_msg_end_nested(odp_actions
, actions_offset
);
1374 nl_msg_end_nested(odp_actions
, sample_offset
);
1375 return cookie_offset
;
1379 compose_sflow_cookie(const struct xbridge
*xbridge
, ovs_be16 vlan_tci
,
1380 odp_port_t odp_port
, unsigned int n_outputs
,
1381 union user_action_cookie
*cookie
)
1385 cookie
->type
= USER_ACTION_COOKIE_SFLOW
;
1386 cookie
->sflow
.vlan_tci
= vlan_tci
;
1388 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1389 * port information") for the interpretation of cookie->output. */
1390 switch (n_outputs
) {
1392 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1393 cookie
->sflow
.output
= 0x40000000 | 256;
1397 ifindex
= dpif_sflow_odp_port_to_ifindex(xbridge
->sflow
, odp_port
);
1399 cookie
->sflow
.output
= ifindex
;
1404 /* 0x80000000 means "multiple output ports. */
1405 cookie
->sflow
.output
= 0x80000000 | n_outputs
;
1410 /* Compose SAMPLE action for sFlow bridge sampling. */
1412 compose_sflow_action(const struct xbridge
*xbridge
,
1413 struct ofpbuf
*odp_actions
,
1414 const struct flow
*flow
,
1415 odp_port_t odp_port
)
1417 uint32_t probability
;
1418 union user_action_cookie cookie
;
1420 if (!xbridge
->sflow
|| flow
->in_port
.ofp_port
== OFPP_NONE
) {
1424 probability
= dpif_sflow_get_probability(xbridge
->sflow
);
1425 compose_sflow_cookie(xbridge
, htons(0), odp_port
,
1426 odp_port
== ODPP_NONE
? 0 : 1, &cookie
);
1428 return compose_sample_action(xbridge
, odp_actions
, flow
, probability
,
1429 &cookie
, sizeof cookie
.sflow
);
1433 compose_flow_sample_cookie(uint16_t probability
, uint32_t collector_set_id
,
1434 uint32_t obs_domain_id
, uint32_t obs_point_id
,
1435 union user_action_cookie
*cookie
)
1437 cookie
->type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
1438 cookie
->flow_sample
.probability
= probability
;
1439 cookie
->flow_sample
.collector_set_id
= collector_set_id
;
1440 cookie
->flow_sample
.obs_domain_id
= obs_domain_id
;
1441 cookie
->flow_sample
.obs_point_id
= obs_point_id
;
1445 compose_ipfix_cookie(union user_action_cookie
*cookie
)
1447 cookie
->type
= USER_ACTION_COOKIE_IPFIX
;
1450 /* Compose SAMPLE action for IPFIX bridge sampling. */
1452 compose_ipfix_action(const struct xbridge
*xbridge
,
1453 struct ofpbuf
*odp_actions
,
1454 const struct flow
*flow
)
1456 uint32_t probability
;
1457 union user_action_cookie cookie
;
1459 if (!xbridge
->ipfix
|| flow
->in_port
.ofp_port
== OFPP_NONE
) {
1463 probability
= dpif_ipfix_get_bridge_exporter_probability(xbridge
->ipfix
);
1464 compose_ipfix_cookie(&cookie
);
1466 compose_sample_action(xbridge
, odp_actions
, flow
, probability
,
1467 &cookie
, sizeof cookie
.ipfix
);
1470 /* SAMPLE action for sFlow must be first action in any given list of
1471 * actions. At this point we do not have all information required to
1472 * build it. So try to build sample action as complete as possible. */
1474 add_sflow_action(struct xlate_ctx
*ctx
)
1476 ctx
->user_cookie_offset
= compose_sflow_action(ctx
->xbridge
,
1477 &ctx
->xout
->odp_actions
,
1478 &ctx
->xin
->flow
, ODPP_NONE
);
1479 ctx
->sflow_odp_port
= 0;
1480 ctx
->sflow_n_outputs
= 0;
1483 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1484 * of actions, eventually after the SAMPLE action for sFlow. */
1486 add_ipfix_action(struct xlate_ctx
*ctx
)
1488 compose_ipfix_action(ctx
->xbridge
, &ctx
->xout
->odp_actions
,
1492 /* Fix SAMPLE action according to data collected while composing ODP actions.
1493 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1494 * USERSPACE action's user-cookie which is required for sflow. */
1496 fix_sflow_action(struct xlate_ctx
*ctx
)
1498 const struct flow
*base
= &ctx
->base_flow
;
1499 union user_action_cookie
*cookie
;
1501 if (!ctx
->user_cookie_offset
) {
1505 cookie
= ofpbuf_at(&ctx
->xout
->odp_actions
, ctx
->user_cookie_offset
,
1506 sizeof cookie
->sflow
);
1507 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
1509 compose_sflow_cookie(ctx
->xbridge
, base
->vlan_tci
,
1510 ctx
->sflow_odp_port
, ctx
->sflow_n_outputs
, cookie
);
1513 static enum slow_path_reason
1514 process_special(struct xlate_ctx
*ctx
, const struct flow
*flow
,
1515 const struct xport
*xport
, const struct ofpbuf
*packet
)
1517 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1518 const struct xbridge
*xbridge
= ctx
->xbridge
;
1522 } else if (xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
)) {
1524 cfm_process_heartbeat(xport
->cfm
, packet
);
1527 } else if (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
, wc
)) {
1529 bfd_process_packet(xport
->bfd
, flow
, packet
);
1532 } else if (xport
->xbundle
&& xport
->xbundle
->lacp
1533 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
1535 lacp_process_packet(xport
->xbundle
->lacp
, xport
->ofport
, packet
);
1538 } else if (xbridge
->stp
&& stp_should_process_flow(flow
, wc
)) {
1540 stp_process_packet(xport
, packet
);
1549 compose_output_action__(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
1552 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
1553 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1554 struct flow
*flow
= &ctx
->xin
->flow
;
1555 ovs_be16 flow_vlan_tci
;
1556 uint32_t flow_pkt_mark
;
1557 uint8_t flow_nw_tos
;
1558 odp_port_t out_port
, odp_port
;
1561 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1562 * before traversing a patch port. */
1563 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 22);
1566 xlate_report(ctx
, "Nonexistent output port");
1568 } else if (xport
->config
& OFPUTIL_PC_NO_FWD
) {
1569 xlate_report(ctx
, "OFPPC_NO_FWD set, skipping output");
1571 } else if (check_stp
&& !xport_stp_forward_state(xport
)) {
1572 xlate_report(ctx
, "STP not in forwarding state, skipping output");
1576 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
) && xport
->xbundle
) {
1577 ctx
->xout
->mirrors
|= xbundle_mirror_dst(xport
->xbundle
->xbridge
,
1582 const struct xport
*peer
= xport
->peer
;
1583 struct flow old_flow
= ctx
->xin
->flow
;
1584 enum slow_path_reason special
;
1586 ctx
->xbridge
= peer
->xbridge
;
1587 flow
->in_port
.ofp_port
= peer
->ofp_port
;
1588 flow
->metadata
= htonll(0);
1589 memset(&flow
->tunnel
, 0, sizeof flow
->tunnel
);
1590 memset(flow
->regs
, 0, sizeof flow
->regs
);
1592 special
= process_special(ctx
, &ctx
->xin
->flow
, peer
,
1595 ctx
->xout
->slow
|= special
;
1596 } else if (may_receive(peer
, ctx
)) {
1597 if (xport_stp_forward_state(peer
)) {
1598 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true);
1600 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1601 * learning action look at the packet, then drop it. */
1602 struct flow old_base_flow
= ctx
->base_flow
;
1603 size_t old_size
= ctx
->xout
->odp_actions
.size
;
1604 mirror_mask_t old_mirrors
= ctx
->xout
->mirrors
;
1605 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true);
1606 ctx
->xout
->mirrors
= old_mirrors
;
1607 ctx
->base_flow
= old_base_flow
;
1608 ctx
->xout
->odp_actions
.size
= old_size
;
1612 ctx
->xin
->flow
= old_flow
;
1613 ctx
->xbridge
= xport
->xbridge
;
1615 if (ctx
->xin
->resubmit_stats
) {
1616 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
1617 netdev_vport_inc_rx(peer
->netdev
, ctx
->xin
->resubmit_stats
);
1623 flow_vlan_tci
= flow
->vlan_tci
;
1624 flow_pkt_mark
= flow
->pkt_mark
;
1625 flow_nw_tos
= flow
->nw_tos
;
1627 if (dscp_from_skb_priority(xport
, flow
->skb_priority
, &dscp
)) {
1628 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
1629 flow
->nw_tos
&= ~IP_DSCP_MASK
;
1630 flow
->nw_tos
|= dscp
;
1633 if (xport
->is_tunnel
) {
1634 /* Save tunnel metadata so that changes made due to
1635 * the Logical (tunnel) Port are not visible for any further
1636 * matches, while explicit set actions on tunnel metadata are.
1638 struct flow_tnl flow_tnl
= flow
->tunnel
;
1639 odp_port
= tnl_port_send(xport
->ofport
, flow
, &ctx
->xout
->wc
);
1640 if (odp_port
== ODPP_NONE
) {
1641 xlate_report(ctx
, "Tunneling decided against output");
1642 goto out
; /* restore flow_nw_tos */
1644 if (flow
->tunnel
.ip_dst
== ctx
->orig_tunnel_ip_dst
) {
1645 xlate_report(ctx
, "Not tunneling to our own address");
1646 goto out
; /* restore flow_nw_tos */
1648 if (ctx
->xin
->resubmit_stats
) {
1649 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
1651 out_port
= odp_port
;
1652 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
1653 &ctx
->xout
->odp_actions
);
1654 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
1656 ofp_port_t vlandev_port
;
1658 odp_port
= xport
->odp_port
;
1659 if (ofproto_has_vlan_splinters(ctx
->xbridge
->ofproto
)) {
1660 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1662 vlandev_port
= vsp_realdev_to_vlandev(ctx
->xbridge
->ofproto
, ofp_port
,
1664 if (vlandev_port
== ofp_port
) {
1665 out_port
= odp_port
;
1667 out_port
= ofp_port_to_odp_port(ctx
->xbridge
, vlandev_port
);
1668 flow
->vlan_tci
= htons(0);
1672 if (out_port
!= ODPP_NONE
) {
1673 ctx
->xout
->slow
|= commit_odp_actions(flow
, &ctx
->base_flow
,
1674 &ctx
->xout
->odp_actions
,
1676 &ctx
->mpls_depth_delta
);
1677 nl_msg_put_odp_port(&ctx
->xout
->odp_actions
, OVS_ACTION_ATTR_OUTPUT
,
1680 ctx
->sflow_odp_port
= odp_port
;
1681 ctx
->sflow_n_outputs
++;
1682 ctx
->xout
->nf_output_iface
= ofp_port
;
1687 flow
->vlan_tci
= flow_vlan_tci
;
1688 flow
->pkt_mark
= flow_pkt_mark
;
1689 flow
->nw_tos
= flow_nw_tos
;
1693 compose_output_action(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
1695 compose_output_action__(ctx
, ofp_port
, true);
1699 xlate_recursively(struct xlate_ctx
*ctx
, struct rule_dpif
*rule
)
1701 struct rule_dpif
*old_rule
= ctx
->rule
;
1702 struct rule_actions
*actions
;
1704 if (ctx
->xin
->resubmit_stats
) {
1705 rule_dpif_credit_stats(rule
, ctx
->xin
->resubmit_stats
);
1711 actions
= rule_dpif_get_actions(rule
);
1712 do_xlate_actions(actions
->ofpacts
, actions
->ofpacts_len
, ctx
);
1713 rule_actions_unref(actions
);
1714 ctx
->rule
= old_rule
;
1719 xlate_resubmit_resource_check(struct xlate_ctx
*ctx
)
1721 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
1723 if (ctx
->recurse
>= MAX_RESUBMIT_RECURSION
) {
1724 VLOG_ERR_RL(&rl
, "resubmit actions recursed over %d times",
1725 MAX_RESUBMIT_RECURSION
);
1726 } else if (ctx
->resubmits
>= MAX_RESUBMITS
) {
1727 VLOG_ERR_RL(&rl
, "over %d resubmit actions", MAX_RESUBMITS
);
1728 } else if (ctx
->xout
->odp_actions
.size
> UINT16_MAX
) {
1729 VLOG_ERR_RL(&rl
, "resubmits yielded over 64 kB of actions");
1730 } else if (ctx
->stack
.size
>= 65536) {
1731 VLOG_ERR_RL(&rl
, "resubmits yielded over 64 kB of stack");
1740 xlate_table_action(struct xlate_ctx
*ctx
,
1741 ofp_port_t in_port
, uint8_t table_id
, bool may_packet_in
)
1743 if (xlate_resubmit_resource_check(ctx
)) {
1744 struct rule_dpif
*rule
;
1745 ofp_port_t old_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
1746 uint8_t old_table_id
= ctx
->table_id
;
1748 ctx
->table_id
= table_id
;
1750 /* Look up a flow with 'in_port' as the input port. Then restore the
1751 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1752 * have surprising behavior). */
1753 ctx
->xin
->flow
.in_port
.ofp_port
= in_port
;
1754 rule_dpif_lookup_in_table(ctx
->xbridge
->ofproto
,
1755 &ctx
->xin
->flow
, &ctx
->xout
->wc
,
1757 ctx
->xin
->flow
.in_port
.ofp_port
= old_in_port
;
1759 if (ctx
->xin
->resubmit_hook
) {
1760 ctx
->xin
->resubmit_hook(ctx
->xin
, rule
, ctx
->recurse
);
1763 if (!rule
&& may_packet_in
) {
1764 struct xport
*xport
;
1767 * check if table configuration flags
1768 * OFPTC_TABLE_MISS_CONTROLLER, default.
1769 * OFPTC_TABLE_MISS_CONTINUE,
1770 * OFPTC_TABLE_MISS_DROP
1771 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do? */
1772 xport
= get_ofp_port(ctx
->xbridge
, ctx
->xin
->flow
.in_port
.ofp_port
);
1773 choose_miss_rule(xport
? xport
->config
: 0,
1774 ctx
->xbridge
->miss_rule
,
1775 ctx
->xbridge
->no_packet_in_rule
, &rule
);
1778 xlate_recursively(ctx
, rule
);
1779 rule_dpif_unref(rule
);
1782 ctx
->table_id
= old_table_id
;
1790 xlate_group_bucket(struct xlate_ctx
*ctx
, const struct ofputil_bucket
*bucket
)
1792 uint64_t action_list_stub
[1024 / 8];
1793 struct ofpbuf action_list
, action_set
;
1795 ofpbuf_use_const(&action_set
, bucket
->ofpacts
, bucket
->ofpacts_len
);
1796 ofpbuf_use_stub(&action_list
, action_list_stub
, sizeof action_list_stub
);
1798 ofpacts_execute_action_set(&action_list
, &action_set
);
1800 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
);
1803 ofpbuf_uninit(&action_set
);
1804 ofpbuf_uninit(&action_list
);
1808 xlate_all_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
1810 const struct ofputil_bucket
*bucket
;
1811 const struct list
*buckets
;
1812 struct flow old_flow
= ctx
->xin
->flow
;
1814 group_dpif_get_buckets(group
, &buckets
);
1816 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
1817 xlate_group_bucket(ctx
, bucket
);
1818 /* Roll back flow to previous state.
1819 * This is equivalent to cloning the packet for each bucket.
1821 * As a side effect any subsequently applied actions will
1822 * also effectively be applied to a clone of the packet taken
1823 * just before applying the all or indirect group. */
1824 ctx
->xin
->flow
= old_flow
;
1829 xlate_group_action__(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
1831 switch (group_dpif_get_type(group
)) {
1833 case OFPGT11_INDIRECT
:
1834 xlate_all_group(ctx
, group
);
1836 case OFPGT11_SELECT
:
1838 /* XXX not yet implemented */
1843 group_dpif_release(group
);
1847 xlate_group_action(struct xlate_ctx
*ctx
, uint32_t group_id
)
1849 if (xlate_resubmit_resource_check(ctx
)) {
1850 struct group_dpif
*group
;
1853 got_group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
, &group
);
1855 xlate_group_action__(ctx
, group
);
1865 xlate_ofpact_resubmit(struct xlate_ctx
*ctx
,
1866 const struct ofpact_resubmit
*resubmit
)
1871 in_port
= resubmit
->in_port
;
1872 if (in_port
== OFPP_IN_PORT
) {
1873 in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
1876 table_id
= resubmit
->table_id
;
1877 if (table_id
== 255) {
1878 table_id
= ctx
->table_id
;
1881 xlate_table_action(ctx
, in_port
, table_id
, false);
1885 flood_packets(struct xlate_ctx
*ctx
, bool all
)
1887 const struct xport
*xport
;
1889 HMAP_FOR_EACH (xport
, ofp_node
, &ctx
->xbridge
->xports
) {
1890 if (xport
->ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
1895 compose_output_action__(ctx
, xport
->ofp_port
, false);
1896 } else if (!(xport
->config
& OFPUTIL_PC_NO_FLOOD
)) {
1897 compose_output_action(ctx
, xport
->ofp_port
);
1901 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
1905 execute_controller_action(struct xlate_ctx
*ctx
, int len
,
1906 enum ofp_packet_in_reason reason
,
1907 uint16_t controller_id
)
1909 struct ofproto_packet_in
*pin
;
1910 struct ofpbuf
*packet
;
1913 ctx
->xout
->slow
|= SLOW_CONTROLLER
;
1914 if (!ctx
->xin
->packet
) {
1918 packet
= ofpbuf_clone(ctx
->xin
->packet
);
1920 key
.skb_priority
= 0;
1922 memset(&key
.tunnel
, 0, sizeof key
.tunnel
);
1924 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
1925 &ctx
->xout
->odp_actions
,
1927 &ctx
->mpls_depth_delta
);
1929 odp_execute_actions(NULL
, packet
, &key
, ctx
->xout
->odp_actions
.data
,
1930 ctx
->xout
->odp_actions
.size
, NULL
, NULL
);
1932 pin
= xmalloc(sizeof *pin
);
1933 pin
->up
.packet_len
= packet
->size
;
1934 pin
->up
.packet
= ofpbuf_steal_data(packet
);
1935 pin
->up
.reason
= reason
;
1936 pin
->up
.table_id
= ctx
->table_id
;
1937 pin
->up
.cookie
= (ctx
->rule
1938 ? rule_dpif_get_flow_cookie(ctx
->rule
)
1941 flow_get_metadata(&ctx
->xin
->flow
, &pin
->up
.fmd
);
1943 pin
->controller_id
= controller_id
;
1944 pin
->send_len
= len
;
1945 pin
->generated_by_table_miss
= (ctx
->rule
1946 && rule_dpif_is_table_miss(ctx
->rule
));
1947 ofproto_dpif_send_packet_in(ctx
->xbridge
->ofproto
, pin
);
1948 ofpbuf_delete(packet
);
1952 compose_mpls_push_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
1954 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1955 struct flow
*flow
= &ctx
->xin
->flow
;
1957 ovs_assert(eth_type_mpls(eth_type
));
1959 /* If mpls_depth_delta is negative then an MPLS POP action has been
1960 * composed and the resulting MPLS label stack is unknown. This means
1961 * an MPLS PUSH action can't be composed as it needs to know either the
1962 * top-most MPLS LSE to use as a template for the new MPLS LSE, or that
1963 * there is no MPLS label stack present. Thus, stop processing.
1965 * If mpls_depth_delta is positive then an MPLS PUSH action has been
1966 * composed and no further MPLS PUSH action may be performed without
1967 * losing MPLS LSE and ether type information held in xtx->xin->flow.
1968 * Thus, stop processing.
1970 * If the MPLS LSE of the flow and base_flow differ then the MPLS LSE
1971 * has been updated. Performing a MPLS PUSH action may be would result in
1972 * losing MPLS LSE and ether type information held in xtx->xin->flow.
1973 * Thus, stop processing.
1975 * It is planned that in the future this case will be handled
1976 * by recirculation */
1977 if (ctx
->mpls_depth_delta
||
1978 ctx
->xin
->flow
.mpls_lse
!= ctx
->base_flow
.mpls_lse
) {
1982 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof wc
->masks
.mpls_lse
);
1984 ctx
->pre_push_mpls_lse
= ctx
->xin
->flow
.mpls_lse
;
1986 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
1987 flow
->mpls_lse
&= ~htonl(MPLS_BOS_MASK
);
1992 if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
1993 label
= htonl(0x2); /* IPV6 Explicit Null. */
1995 label
= htonl(0x0); /* IPV4 Explicit Null. */
1997 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
1998 wc
->masks
.nw_ttl
= 0xff;
1999 tc
= (flow
->nw_tos
& IP_DSCP_MASK
) >> 2;
2000 ttl
= flow
->nw_ttl
? flow
->nw_ttl
: 0x40;
2001 flow
->mpls_lse
= set_mpls_lse_values(ttl
, tc
, 1, label
);
2003 flow
->dl_type
= eth_type
;
2004 ctx
->mpls_depth_delta
++;
2010 compose_mpls_pop_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
2012 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2014 if (!eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2018 /* If mpls_depth_delta is negative then an MPLS POP action has been
2019 * composed. Performing another MPLS POP action
2020 * would result in losing ether type that results from
2021 * the already composed MPLS POP. Thus, stop processing.
2023 * It is planned that in the future this case will be handled
2024 * by recirculation */
2025 if (ctx
->mpls_depth_delta
< 0) {
2029 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof wc
->masks
.mpls_lse
);
2031 /* If mpls_depth_delta is positive then an MPLS PUSH action has been
2032 * executed and the previous MPLS LSE saved in ctx->pre_push_mpls_lse. The
2033 * flow's MPLS LSE should be restored to that value to allow any
2034 * subsequent actions that update of the LSE to be executed correctly.
2036 if (ctx
->mpls_depth_delta
> 0) {
2037 ctx
->xin
->flow
.mpls_lse
= ctx
->pre_push_mpls_lse
;
2040 ctx
->xin
->flow
.dl_type
= eth_type
;
2041 ctx
->mpls_depth_delta
--;
2047 compose_dec_ttl(struct xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
2049 struct flow
*flow
= &ctx
->xin
->flow
;
2051 if (!is_ip_any(flow
)) {
2055 ctx
->xout
->wc
.masks
.nw_ttl
= 0xff;
2056 if (flow
->nw_ttl
> 1) {
2062 for (i
= 0; i
< ids
->n_controllers
; i
++) {
2063 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
2067 /* Stop processing for current table. */
2073 compose_set_mpls_label_action(struct xlate_ctx
*ctx
, ovs_be32 label
)
2075 if (!eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2079 /* If mpls_depth_delta is negative then an MPLS POP action has been
2080 * executed and the resulting MPLS label stack is unknown. This means
2081 * a SET MPLS LABEL action can't be executed as it needs to manipulate
2082 * the top-most MPLS LSE. Thus, stop processing.
2084 * It is planned that in the future this case will be handled
2087 if (ctx
->mpls_depth_delta
< 0) {
2091 ctx
->xout
->wc
.masks
.mpls_lse
|= htonl(MPLS_LABEL_MASK
);
2092 set_mpls_lse_label(&ctx
->xin
->flow
.mpls_lse
, label
);
2097 compose_set_mpls_tc_action(struct xlate_ctx
*ctx
, uint8_t tc
)
2099 if (!eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2103 /* If mpls_depth_delta is negative then an MPLS POP action has been
2104 * executed and the resulting MPLS label stack is unknown. This means
2105 * a SET MPLS TC action can't be executed as it needs to manipulate
2106 * the top-most MPLS LSE. Thus, stop processing.
2108 * It is planned that in the future this case will be handled
2111 if (ctx
->mpls_depth_delta
< 0) {
2115 ctx
->xout
->wc
.masks
.mpls_lse
|= htonl(MPLS_TC_MASK
);
2116 set_mpls_lse_tc(&ctx
->xin
->flow
.mpls_lse
, tc
);
2121 compose_set_mpls_ttl_action(struct xlate_ctx
*ctx
, uint8_t ttl
)
2123 if (!eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2127 /* If mpls_depth_delta is negative then an MPLS POP action has been
2128 * executed and the resulting MPLS label stack is unknown. This means
2129 * a SET MPLS TTL push action can't be executed as it needs to manipulate
2130 * the top-most MPLS LSE. Thus, stop processing.
2132 * It is planned that in the future this case will be handled
2135 if (ctx
->mpls_depth_delta
< 0) {
2139 ctx
->xout
->wc
.masks
.mpls_lse
|= htonl(MPLS_TTL_MASK
);
2140 set_mpls_lse_ttl(&ctx
->xin
->flow
.mpls_lse
, ttl
);
2145 compose_dec_mpls_ttl_action(struct xlate_ctx
*ctx
)
2147 struct flow
*flow
= &ctx
->xin
->flow
;
2148 uint8_t ttl
= mpls_lse_to_ttl(flow
->mpls_lse
);
2149 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2151 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof wc
->masks
.mpls_lse
);
2153 if (!eth_type_mpls(flow
->dl_type
)) {
2159 set_mpls_lse_ttl(&flow
->mpls_lse
, ttl
);
2162 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0);
2164 /* Stop processing for current table. */
2170 xlate_output_action(struct xlate_ctx
*ctx
,
2171 ofp_port_t port
, uint16_t max_len
, bool may_packet_in
)
2173 ofp_port_t prev_nf_output_iface
= ctx
->xout
->nf_output_iface
;
2175 ctx
->xout
->nf_output_iface
= NF_OUT_DROP
;
2179 compose_output_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
);
2182 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
2189 flood_packets(ctx
, false);
2192 flood_packets(ctx
, true);
2194 case OFPP_CONTROLLER
:
2195 execute_controller_action(ctx
, max_len
, OFPR_ACTION
, 0);
2201 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
2202 compose_output_action(ctx
, port
);
2204 xlate_report(ctx
, "skipping output to input port");
2209 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2210 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
2211 } else if (ctx
->xout
->nf_output_iface
== NF_OUT_DROP
) {
2212 ctx
->xout
->nf_output_iface
= prev_nf_output_iface
;
2213 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2214 ctx
->xout
->nf_output_iface
!= NF_OUT_FLOOD
) {
2215 ctx
->xout
->nf_output_iface
= NF_OUT_MULTI
;
2220 xlate_output_reg_action(struct xlate_ctx
*ctx
,
2221 const struct ofpact_output_reg
*or)
2223 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->xin
->flow
);
2224 if (port
<= UINT16_MAX
) {
2225 union mf_subvalue value
;
2227 memset(&value
, 0xff, sizeof value
);
2228 mf_write_subfield_flow(&or->src
, &value
, &ctx
->xout
->wc
.masks
);
2229 xlate_output_action(ctx
, u16_to_ofp(port
),
2230 or->max_len
, false);
2235 xlate_enqueue_action(struct xlate_ctx
*ctx
,
2236 const struct ofpact_enqueue
*enqueue
)
2238 ofp_port_t ofp_port
= enqueue
->port
;
2239 uint32_t queue_id
= enqueue
->queue
;
2240 uint32_t flow_priority
, priority
;
2243 /* Translate queue to priority. */
2244 error
= dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &priority
);
2246 /* Fall back to ordinary output action. */
2247 xlate_output_action(ctx
, enqueue
->port
, 0, false);
2251 /* Check output port. */
2252 if (ofp_port
== OFPP_IN_PORT
) {
2253 ofp_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
2254 } else if (ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
2258 /* Add datapath actions. */
2259 flow_priority
= ctx
->xin
->flow
.skb_priority
;
2260 ctx
->xin
->flow
.skb_priority
= priority
;
2261 compose_output_action(ctx
, ofp_port
);
2262 ctx
->xin
->flow
.skb_priority
= flow_priority
;
2264 /* Update NetFlow output port. */
2265 if (ctx
->xout
->nf_output_iface
== NF_OUT_DROP
) {
2266 ctx
->xout
->nf_output_iface
= ofp_port
;
2267 } else if (ctx
->xout
->nf_output_iface
!= NF_OUT_FLOOD
) {
2268 ctx
->xout
->nf_output_iface
= NF_OUT_MULTI
;
2273 xlate_set_queue_action(struct xlate_ctx
*ctx
, uint32_t queue_id
)
2275 uint32_t skb_priority
;
2277 if (!dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &skb_priority
)) {
2278 ctx
->xin
->flow
.skb_priority
= skb_priority
;
2280 /* Couldn't translate queue to a priority. Nothing to do. A warning
2281 * has already been logged. */
2286 slave_enabled_cb(ofp_port_t ofp_port
, void *xbridge_
)
2288 const struct xbridge
*xbridge
= xbridge_
;
2299 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
2302 port
= get_ofp_port(xbridge
, ofp_port
);
2303 return port
? port
->may_enable
: false;
2308 xlate_bundle_action(struct xlate_ctx
*ctx
,
2309 const struct ofpact_bundle
*bundle
)
2313 port
= bundle_execute(bundle
, &ctx
->xin
->flow
, &ctx
->xout
->wc
,
2315 CONST_CAST(struct xbridge
*, ctx
->xbridge
));
2316 if (bundle
->dst
.field
) {
2317 nxm_reg_load(&bundle
->dst
, ofp_to_u16(port
), &ctx
->xin
->flow
,
2320 xlate_output_action(ctx
, port
, 0, false);
2325 xlate_learn_action(struct xlate_ctx
*ctx
,
2326 const struct ofpact_learn
*learn
)
2328 uint64_t ofpacts_stub
[1024 / 8];
2329 struct ofputil_flow_mod fm
;
2330 struct ofpbuf ofpacts
;
2332 ctx
->xout
->has_learn
= true;
2334 learn_mask(learn
, &ctx
->xout
->wc
);
2336 if (!ctx
->xin
->may_learn
) {
2340 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2341 learn_execute(learn
, &ctx
->xin
->flow
, &fm
, &ofpacts
);
2342 ofproto_dpif_flow_mod(ctx
->xbridge
->ofproto
, &fm
);
2343 ofpbuf_uninit(&ofpacts
);
2347 xlate_fin_timeout(struct xlate_ctx
*ctx
,
2348 const struct ofpact_fin_timeout
*oft
)
2350 if (ctx
->xin
->tcp_flags
& (TCP_FIN
| TCP_RST
) && ctx
->rule
) {
2351 rule_dpif_reduce_timeouts(ctx
->rule
, oft
->fin_idle_timeout
,
2352 oft
->fin_hard_timeout
);
2357 xlate_sample_action(struct xlate_ctx
*ctx
,
2358 const struct ofpact_sample
*os
)
2360 union user_action_cookie cookie
;
2361 /* Scale the probability from 16-bit to 32-bit while representing
2362 * the same percentage. */
2363 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
2365 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
2366 &ctx
->xout
->odp_actions
,
2368 &ctx
->mpls_depth_delta
);
2370 compose_flow_sample_cookie(os
->probability
, os
->collector_set_id
,
2371 os
->obs_domain_id
, os
->obs_point_id
, &cookie
);
2372 compose_sample_action(ctx
->xbridge
, &ctx
->xout
->odp_actions
, &ctx
->xin
->flow
,
2373 probability
, &cookie
, sizeof cookie
.flow_sample
);
2377 may_receive(const struct xport
*xport
, struct xlate_ctx
*ctx
)
2379 if (xport
->config
& (eth_addr_equals(ctx
->xin
->flow
.dl_dst
, eth_addr_stp
)
2380 ? OFPUTIL_PC_NO_RECV_STP
2381 : OFPUTIL_PC_NO_RECV
)) {
2385 /* Only drop packets here if both forwarding and learning are
2386 * disabled. If just learning is enabled, we need to have
2387 * OFPP_NORMAL and the learning action have a look at the packet
2388 * before we can drop it. */
2389 if (!xport_stp_forward_state(xport
) && !xport_stp_learn_state(xport
)) {
2397 xlate_write_actions(struct xlate_ctx
*ctx
, const struct ofpact
*a
)
2399 struct ofpact_nest
*on
= ofpact_get_WRITE_ACTIONS(a
);
2400 ofpbuf_put(&ctx
->action_set
, on
->actions
, ofpact_nest_get_action_len(on
));
2401 ofpact_pad(&ctx
->action_set
);
2405 xlate_action_set(struct xlate_ctx
*ctx
)
2407 uint64_t action_list_stub
[1024 / 64];
2408 struct ofpbuf action_list
;
2410 ofpbuf_use_stub(&action_list
, action_list_stub
, sizeof action_list_stub
);
2411 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
2412 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
);
2413 ofpbuf_uninit(&action_list
);
2417 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
2418 struct xlate_ctx
*ctx
)
2420 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2421 struct flow
*flow
= &ctx
->xin
->flow
;
2422 const struct ofpact
*a
;
2424 /* dl_type already in the mask, not set below. */
2426 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
2427 struct ofpact_controller
*controller
;
2428 const struct ofpact_metadata
*metadata
;
2429 const struct ofpact_set_field
*set_field
;
2430 const struct mf_field
*mf
;
2438 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
2439 ofpact_get_OUTPUT(a
)->max_len
, true);
2443 if (xlate_group_action(ctx
, ofpact_get_GROUP(a
)->group_id
)) {
2448 case OFPACT_CONTROLLER
:
2449 controller
= ofpact_get_CONTROLLER(a
);
2450 execute_controller_action(ctx
, controller
->max_len
,
2452 controller
->controller_id
);
2455 case OFPACT_ENQUEUE
:
2456 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
));
2459 case OFPACT_SET_VLAN_VID
:
2460 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
2461 if (flow
->vlan_tci
& htons(VLAN_CFI
) ||
2462 ofpact_get_SET_VLAN_VID(a
)->push_vlan_if_needed
) {
2463 flow
->vlan_tci
&= ~htons(VLAN_VID_MASK
);
2464 flow
->vlan_tci
|= (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
)
2469 case OFPACT_SET_VLAN_PCP
:
2470 wc
->masks
.vlan_tci
|= htons(VLAN_PCP_MASK
| VLAN_CFI
);
2471 if (flow
->vlan_tci
& htons(VLAN_CFI
) ||
2472 ofpact_get_SET_VLAN_PCP(a
)->push_vlan_if_needed
) {
2473 flow
->vlan_tci
&= ~htons(VLAN_PCP_MASK
);
2474 flow
->vlan_tci
|= htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
2475 << VLAN_PCP_SHIFT
) | VLAN_CFI
);
2479 case OFPACT_STRIP_VLAN
:
2480 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
2481 flow
->vlan_tci
= htons(0);
2484 case OFPACT_PUSH_VLAN
:
2485 /* XXX 802.1AD(QinQ) */
2486 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
2487 flow
->vlan_tci
= htons(VLAN_CFI
);
2490 case OFPACT_SET_ETH_SRC
:
2491 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
2492 memcpy(flow
->dl_src
, ofpact_get_SET_ETH_SRC(a
)->mac
, ETH_ADDR_LEN
);
2495 case OFPACT_SET_ETH_DST
:
2496 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2497 memcpy(flow
->dl_dst
, ofpact_get_SET_ETH_DST(a
)->mac
, ETH_ADDR_LEN
);
2500 case OFPACT_SET_IPV4_SRC
:
2501 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2502 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
2503 flow
->nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
2507 case OFPACT_SET_IPV4_DST
:
2508 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2509 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2510 flow
->nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
2514 case OFPACT_SET_IP_DSCP
:
2515 if (is_ip_any(flow
)) {
2516 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
2517 flow
->nw_tos
&= ~IP_DSCP_MASK
;
2518 flow
->nw_tos
|= ofpact_get_SET_IP_DSCP(a
)->dscp
;
2522 case OFPACT_SET_IP_ECN
:
2523 if (is_ip_any(flow
)) {
2524 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
2525 flow
->nw_tos
&= ~IP_ECN_MASK
;
2526 flow
->nw_tos
|= ofpact_get_SET_IP_ECN(a
)->ecn
;
2530 case OFPACT_SET_IP_TTL
:
2531 if (is_ip_any(flow
)) {
2532 wc
->masks
.nw_ttl
= 0xff;
2533 flow
->nw_ttl
= ofpact_get_SET_IP_TTL(a
)->ttl
;
2537 case OFPACT_SET_L4_SRC_PORT
:
2538 if (is_ip_any(flow
)) {
2539 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2540 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
2541 flow
->tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
2545 case OFPACT_SET_L4_DST_PORT
:
2546 if (is_ip_any(flow
)) {
2547 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2548 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
2549 flow
->tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
2553 case OFPACT_RESUBMIT
:
2554 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
));
2557 case OFPACT_SET_TUNNEL
:
2558 flow
->tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
2561 case OFPACT_SET_QUEUE
:
2562 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
2565 case OFPACT_POP_QUEUE
:
2566 flow
->skb_priority
= ctx
->orig_skb_priority
;
2569 case OFPACT_REG_MOVE
:
2570 nxm_execute_reg_move(ofpact_get_REG_MOVE(a
), flow
, wc
);
2573 case OFPACT_REG_LOAD
:
2574 nxm_execute_reg_load(ofpact_get_REG_LOAD(a
), flow
, wc
);
2577 case OFPACT_SET_FIELD
:
2578 set_field
= ofpact_get_SET_FIELD(a
);
2579 mf
= set_field
->field
;
2580 mf_mask_field_and_prereqs(mf
, &wc
->masks
);
2582 /* Set field action only ever overwrites packet's outermost
2583 * applicable header fields. Do nothing if no header exists. */
2584 if ((mf
->id
!= MFF_VLAN_VID
|| flow
->vlan_tci
& htons(VLAN_CFI
))
2585 && ((mf
->id
!= MFF_MPLS_LABEL
&& mf
->id
!= MFF_MPLS_TC
)
2586 || flow
->mpls_lse
)) {
2587 mf_set_flow_value(mf
, &set_field
->value
, flow
);
2591 case OFPACT_STACK_PUSH
:
2592 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), flow
, wc
,
2596 case OFPACT_STACK_POP
:
2597 nxm_execute_stack_pop(ofpact_get_STACK_POP(a
), flow
, wc
,
2601 case OFPACT_PUSH_MPLS
:
2602 if (compose_mpls_push_action(ctx
,
2603 ofpact_get_PUSH_MPLS(a
)->ethertype
)) {
2608 case OFPACT_POP_MPLS
:
2609 if (compose_mpls_pop_action(ctx
,
2610 ofpact_get_POP_MPLS(a
)->ethertype
)) {
2615 case OFPACT_SET_MPLS_LABEL
:
2616 if (compose_set_mpls_label_action(ctx
,
2617 ofpact_get_SET_MPLS_LABEL(a
)->label
)) {
2622 case OFPACT_SET_MPLS_TC
:
2623 if (compose_set_mpls_tc_action(ctx
,
2624 ofpact_get_SET_MPLS_TC(a
)->tc
)) {
2629 case OFPACT_SET_MPLS_TTL
:
2630 if (compose_set_mpls_ttl_action(ctx
,
2631 ofpact_get_SET_MPLS_TTL(a
)->ttl
)) {
2636 case OFPACT_DEC_MPLS_TTL
:
2637 if (compose_dec_mpls_ttl_action(ctx
)) {
2642 case OFPACT_DEC_TTL
:
2643 wc
->masks
.nw_ttl
= 0xff;
2644 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
2650 /* Nothing to do. */
2653 case OFPACT_MULTIPATH
:
2654 multipath_execute(ofpact_get_MULTIPATH(a
), flow
, wc
);
2658 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
));
2661 case OFPACT_OUTPUT_REG
:
2662 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
));
2666 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
2673 case OFPACT_FIN_TIMEOUT
:
2674 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2675 ctx
->xout
->has_fin_timeout
= true;
2676 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
2679 case OFPACT_CLEAR_ACTIONS
:
2680 ofpbuf_clear(&ctx
->action_set
);
2683 case OFPACT_WRITE_ACTIONS
:
2684 xlate_write_actions(ctx
, a
);
2687 case OFPACT_WRITE_METADATA
:
2688 metadata
= ofpact_get_WRITE_METADATA(a
);
2689 flow
->metadata
&= ~metadata
->mask
;
2690 flow
->metadata
|= metadata
->metadata
& metadata
->mask
;
2694 /* Not implemented yet. */
2697 case OFPACT_GOTO_TABLE
: {
2698 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
2700 ovs_assert(ctx
->table_id
< ogt
->table_id
);
2701 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
2702 ogt
->table_id
, true);
2707 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
2714 xlate_in_init(struct xlate_in
*xin
, struct ofproto_dpif
*ofproto
,
2715 const struct flow
*flow
, struct rule_dpif
*rule
,
2716 uint16_t tcp_flags
, const struct ofpbuf
*packet
)
2718 xin
->ofproto
= ofproto
;
2720 xin
->packet
= packet
;
2721 xin
->may_learn
= packet
!= NULL
;
2723 xin
->ofpacts
= NULL
;
2724 xin
->ofpacts_len
= 0;
2725 xin
->tcp_flags
= tcp_flags
;
2726 xin
->resubmit_hook
= NULL
;
2727 xin
->report_hook
= NULL
;
2728 xin
->resubmit_stats
= NULL
;
2732 xlate_out_uninit(struct xlate_out
*xout
)
2735 ofpbuf_uninit(&xout
->odp_actions
);
2739 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2740 * into datapath actions, using 'ctx', and discards the datapath actions. */
2742 xlate_actions_for_side_effects(struct xlate_in
*xin
)
2744 struct xlate_out xout
;
2746 xlate_actions(xin
, &xout
);
2747 xlate_out_uninit(&xout
);
2751 xlate_report(struct xlate_ctx
*ctx
, const char *s
)
2753 if (ctx
->xin
->report_hook
) {
2754 ctx
->xin
->report_hook(ctx
->xin
, s
, ctx
->recurse
);
2759 xlate_out_copy(struct xlate_out
*dst
, const struct xlate_out
*src
)
2762 dst
->slow
= src
->slow
;
2763 dst
->has_learn
= src
->has_learn
;
2764 dst
->has_normal
= src
->has_normal
;
2765 dst
->has_fin_timeout
= src
->has_fin_timeout
;
2766 dst
->nf_output_iface
= src
->nf_output_iface
;
2767 dst
->mirrors
= src
->mirrors
;
2769 ofpbuf_use_stub(&dst
->odp_actions
, dst
->odp_actions_stub
,
2770 sizeof dst
->odp_actions_stub
);
2771 ofpbuf_put(&dst
->odp_actions
, src
->odp_actions
.data
,
2772 src
->odp_actions
.size
);
2775 /* Returns a reference to the sflow handled associated with ofproto, or NULL if
2776 * there is none. The caller is responsible for decrementing the results ref
2777 * count with dpif_sflow_unref(). */
2779 xlate_get_sflow(const struct ofproto_dpif
*ofproto
)
2781 struct dpif_sflow
*sflow
= NULL
;
2782 struct xbridge
*xbridge
;
2784 ovs_rwlock_rdlock(&xlate_rwlock
);
2785 xbridge
= xbridge_lookup(ofproto
);
2787 sflow
= dpif_sflow_ref(xbridge
->sflow
);
2789 ovs_rwlock_unlock(&xlate_rwlock
);
2794 /* Returns a reference to the ipfix handled associated with ofproto, or NULL if
2795 * there is none. The caller is responsible for decrementing the results ref
2796 * count with dpif_ipfix_unref(). */
2798 xlate_get_ipfix(const struct ofproto_dpif
*ofproto
)
2800 struct dpif_ipfix
*ipfix
= NULL
;
2801 struct xbridge
*xbridge
;
2803 ovs_rwlock_rdlock(&xlate_rwlock
);
2804 xbridge
= xbridge_lookup(ofproto
);
2806 ipfix
= dpif_ipfix_ref(xbridge
->ipfix
);
2808 ovs_rwlock_unlock(&xlate_rwlock
);
2813 static struct skb_priority_to_dscp
*
2814 get_skb_priority(const struct xport
*xport
, uint32_t skb_priority
)
2816 struct skb_priority_to_dscp
*pdscp
;
2819 hash
= hash_int(skb_priority
, 0);
2820 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &xport
->skb_priorities
) {
2821 if (pdscp
->skb_priority
== skb_priority
) {
2829 dscp_from_skb_priority(const struct xport
*xport
, uint32_t skb_priority
,
2832 struct skb_priority_to_dscp
*pdscp
= get_skb_priority(xport
, skb_priority
);
2833 *dscp
= pdscp
? pdscp
->dscp
: 0;
2834 return pdscp
!= NULL
;
2838 clear_skb_priorities(struct xport
*xport
)
2840 struct skb_priority_to_dscp
*pdscp
, *next
;
2842 HMAP_FOR_EACH_SAFE (pdscp
, next
, hmap_node
, &xport
->skb_priorities
) {
2843 hmap_remove(&xport
->skb_priorities
, &pdscp
->hmap_node
);
2849 actions_output_to_local_port(const struct xlate_ctx
*ctx
)
2851 odp_port_t local_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
, OFPP_LOCAL
);
2852 const struct nlattr
*a
;
2855 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, ctx
->xout
->odp_actions
.data
,
2856 ctx
->xout
->odp_actions
.size
) {
2857 if (nl_attr_type(a
) == OVS_ACTION_ATTR_OUTPUT
2858 && nl_attr_get_odp_port(a
) == local_odp_port
) {
2865 /* Thread safe call to xlate_actions__(). */
2867 xlate_actions(struct xlate_in
*xin
, struct xlate_out
*xout
)
2869 ovs_rwlock_rdlock(&xlate_rwlock
);
2870 xlate_actions__(xin
, xout
);
2871 ovs_rwlock_unlock(&xlate_rwlock
);
2874 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2875 * into datapath actions in 'odp_actions', using 'ctx'.
2877 * The caller must take responsibility for eventually freeing 'xout', with
2878 * xlate_out_uninit(). */
2880 xlate_actions__(struct xlate_in
*xin
, struct xlate_out
*xout
)
2881 OVS_REQ_RDLOCK(xlate_rwlock
)
2883 struct flow_wildcards
*wc
= &xout
->wc
;
2884 struct flow
*flow
= &xin
->flow
;
2885 struct rule_dpif
*rule
= NULL
;
2887 struct rule_actions
*actions
= NULL
;
2888 enum slow_path_reason special
;
2889 const struct ofpact
*ofpacts
;
2890 struct xport
*in_port
;
2891 struct flow orig_flow
;
2892 struct xlate_ctx ctx
;
2896 COVERAGE_INC(xlate_actions
);
2898 /* Flow initialization rules:
2899 * - 'base_flow' must match the kernel's view of the packet at the
2900 * time that action processing starts. 'flow' represents any
2901 * transformations we wish to make through actions.
2902 * - By default 'base_flow' and 'flow' are the same since the input
2903 * packet matches the output before any actions are applied.
2904 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2905 * of the received packet as seen by the kernel. If we later output
2906 * to another device without any modifications this will cause us to
2907 * insert a new tag since the original one was stripped off by the
2909 * - Tunnel metadata as received is retained in 'flow'. This allows
2910 * tunnel metadata matching also in later tables.
2911 * Since a kernel action for setting the tunnel metadata will only be
2912 * generated with actual tunnel output, changing the tunnel metadata
2913 * values in 'flow' (such as tun_id) will only have effect with a later
2914 * tunnel output action.
2915 * - Tunnel 'base_flow' is completely cleared since that is what the
2916 * kernel does. If we wish to maintain the original values an action
2917 * needs to be generated. */
2922 ctx
.xout
->has_learn
= false;
2923 ctx
.xout
->has_normal
= false;
2924 ctx
.xout
->has_fin_timeout
= false;
2925 ctx
.xout
->nf_output_iface
= NF_OUT_DROP
;
2926 ctx
.xout
->mirrors
= 0;
2927 ofpbuf_use_stub(&ctx
.xout
->odp_actions
, ctx
.xout
->odp_actions_stub
,
2928 sizeof ctx
.xout
->odp_actions_stub
);
2929 ofpbuf_reserve(&ctx
.xout
->odp_actions
, NL_A_U32_SIZE
);
2931 ctx
.xbridge
= xbridge_lookup(xin
->ofproto
);
2936 ctx
.rule
= xin
->rule
;
2938 ctx
.base_flow
= *flow
;
2939 memset(&ctx
.base_flow
.tunnel
, 0, sizeof ctx
.base_flow
.tunnel
);
2940 ctx
.orig_tunnel_ip_dst
= flow
->tunnel
.ip_dst
;
2942 flow_wildcards_init_catchall(wc
);
2943 memset(&wc
->masks
.in_port
, 0xff, sizeof wc
->masks
.in_port
);
2944 memset(&wc
->masks
.skb_priority
, 0xff, sizeof wc
->masks
.skb_priority
);
2945 memset(&wc
->masks
.dl_type
, 0xff, sizeof wc
->masks
.dl_type
);
2946 wc
->masks
.nw_frag
|= FLOW_NW_FRAG_MASK
;
2948 tnl_may_send
= tnl_xlate_init(&ctx
.base_flow
, flow
, wc
);
2949 if (ctx
.xbridge
->has_netflow
) {
2950 netflow_mask_wc(flow
, wc
);
2955 ctx
.orig_skb_priority
= flow
->skb_priority
;
2958 ctx
.mpls_depth_delta
= 0;
2960 if (!xin
->ofpacts
&& !ctx
.rule
) {
2961 rule_dpif_lookup(ctx
.xbridge
->ofproto
, flow
, wc
, &rule
);
2962 if (ctx
.xin
->resubmit_stats
) {
2963 rule_dpif_credit_stats(rule
, ctx
.xin
->resubmit_stats
);
2967 xout
->fail_open
= ctx
.rule
&& rule_dpif_is_fail_open(ctx
.rule
);
2970 ofpacts
= xin
->ofpacts
;
2971 ofpacts_len
= xin
->ofpacts_len
;
2972 } else if (ctx
.rule
) {
2973 actions
= rule_dpif_get_actions(ctx
.rule
);
2974 ofpacts
= actions
->ofpacts
;
2975 ofpacts_len
= actions
->ofpacts_len
;
2980 ofpbuf_use_stub(&ctx
.stack
, ctx
.init_stack
, sizeof ctx
.init_stack
);
2981 ofpbuf_use_stub(&ctx
.action_set
,
2982 ctx
.action_set_stub
, sizeof ctx
.action_set_stub
);
2984 if (mbridge_has_mirrors(ctx
.xbridge
->mbridge
)) {
2985 /* Do this conditionally because the copy is expensive enough that it
2986 * shows up in profiles. */
2990 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
2991 switch (ctx
.xbridge
->frag
) {
2992 case OFPC_FRAG_NORMAL
:
2993 /* We must pretend that transport ports are unavailable. */
2994 flow
->tp_src
= ctx
.base_flow
.tp_src
= htons(0);
2995 flow
->tp_dst
= ctx
.base_flow
.tp_dst
= htons(0);
2998 case OFPC_FRAG_DROP
:
3001 case OFPC_FRAG_REASM
:
3004 case OFPC_FRAG_NX_MATCH
:
3005 /* Nothing to do. */
3008 case OFPC_INVALID_TTL_TO_CONTROLLER
:
3013 in_port
= get_ofp_port(ctx
.xbridge
, flow
->in_port
.ofp_port
);
3014 special
= process_special(&ctx
, flow
, in_port
, ctx
.xin
->packet
);
3016 ctx
.xout
->slow
|= special
;
3018 size_t sample_actions_len
;
3020 if (flow
->in_port
.ofp_port
3021 != vsp_realdev_to_vlandev(ctx
.xbridge
->ofproto
,
3022 flow
->in_port
.ofp_port
,
3024 ctx
.base_flow
.vlan_tci
= 0;
3027 add_sflow_action(&ctx
);
3028 add_ipfix_action(&ctx
);
3029 sample_actions_len
= ctx
.xout
->odp_actions
.size
;
3031 if (tnl_may_send
&& (!in_port
|| may_receive(in_port
, &ctx
))) {
3032 do_xlate_actions(ofpacts
, ofpacts_len
, &ctx
);
3034 /* We've let OFPP_NORMAL and the learning action look at the
3035 * packet, so drop it now if forwarding is disabled. */
3036 if (in_port
&& !xport_stp_forward_state(in_port
)) {
3037 ctx
.xout
->odp_actions
.size
= sample_actions_len
;
3041 if (ctx
.action_set
.size
) {
3042 xlate_action_set(&ctx
);
3045 if (ctx
.xbridge
->has_in_band
3046 && in_band_must_output_to_local_port(flow
)
3047 && !actions_output_to_local_port(&ctx
)) {
3048 compose_output_action(&ctx
, OFPP_LOCAL
);
3051 fix_sflow_action(&ctx
);
3053 if (mbridge_has_mirrors(ctx
.xbridge
->mbridge
)) {
3054 add_mirror_actions(&ctx
, &orig_flow
);
3058 if (nl_attr_oversized(ctx
.xout
->odp_actions
.size
)) {
3059 /* These datapath actions are too big for a Netlink attribute, so we
3060 * can't hand them to the kernel directly. dpif_execute() can execute
3061 * them one by one with help, so just mark the result as SLOW_ACTION to
3062 * prevent the flow from being installed. */
3063 COVERAGE_INC(xlate_actions_oversize
);
3064 ctx
.xout
->slow
|= SLOW_ACTION
;
3067 ofpbuf_uninit(&ctx
.stack
);
3068 ofpbuf_uninit(&ctx
.action_set
);
3070 /* Clear the metadata and register wildcard masks, because we won't
3071 * use non-header fields as part of the cache. */
3072 memset(&wc
->masks
.metadata
, 0, sizeof wc
->masks
.metadata
);
3073 memset(&wc
->masks
.regs
, 0, sizeof wc
->masks
.regs
);
3076 rule_actions_unref(actions
);
3077 rule_dpif_unref(rule
);
3080 /* Sends 'packet' out 'ofport'.
3081 * May modify 'packet'.
3082 * Returns 0 if successful, otherwise a positive errno value. */
3084 xlate_send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
3086 struct xport
*xport
;
3087 struct ofpact_output output
;
3089 union flow_in_port in_port_
;
3092 ofpact_init(&output
.ofpact
, OFPACT_OUTPUT
, sizeof output
);
3093 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3094 in_port_
.ofp_port
= OFPP_NONE
;
3095 flow_extract(packet
, 0, 0, NULL
, &in_port_
, &flow
);
3097 ovs_rwlock_rdlock(&xlate_rwlock
);
3098 xport
= xport_lookup(ofport
);
3100 ovs_rwlock_unlock(&xlate_rwlock
);
3103 output
.port
= xport
->ofp_port
;
3105 error
= ofproto_dpif_execute_actions(xport
->xbridge
->ofproto
, &flow
, NULL
,
3106 &output
.ofpact
, sizeof output
,
3108 ovs_rwlock_unlock(&xlate_rwlock
);