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_mpls_overflow
);
56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate
);
58 /* Maximum depth of flow table recursion (due to resubmit actions) in a
59 * flow translation. */
60 #define MAX_RESUBMIT_RECURSION 64
62 /* Maximum number of resubmit actions in a flow translation, whether they are
63 * recursive or not. */
64 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
66 struct ovs_rwlock xlate_rwlock
= OVS_RWLOCK_INITIALIZER
;
69 struct hmap_node hmap_node
; /* Node in global 'xbridges' map. */
70 struct ofproto_dpif
*ofproto
; /* Key in global 'xbridges' map. */
72 struct list xbundles
; /* Owned xbundles. */
73 struct hmap xports
; /* Indexed by ofp_port. */
75 char *name
; /* Name used in log messages. */
76 struct dpif
*dpif
; /* Datapath interface. */
77 struct mac_learning
*ml
; /* Mac learning handle. */
78 struct mbridge
*mbridge
; /* Mirroring. */
79 struct dpif_sflow
*sflow
; /* SFlow handle, or null. */
80 struct dpif_ipfix
*ipfix
; /* Ipfix handle, or null. */
81 struct netflow
*netflow
; /* Netflow handle, or null. */
82 struct stp
*stp
; /* STP or null if disabled. */
84 /* Special rules installed by ofproto-dpif. */
85 struct rule_dpif
*miss_rule
;
86 struct rule_dpif
*no_packet_in_rule
;
88 enum ofp_config_flags frag
; /* Fragmentation handling. */
89 bool has_in_band
; /* Bridge has in band control? */
90 bool forward_bpdu
; /* Bridge forwards STP BPDUs? */
92 /* True if the datapath supports variable-length
93 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
94 * False if the datapath supports only 8-byte (or shorter) userdata. */
95 bool variable_length_userdata
;
97 /* Number of MPLS label stack entries that the datapath supports
99 size_t max_mpls_depth
;
103 struct hmap_node hmap_node
; /* In global 'xbundles' map. */
104 struct ofbundle
*ofbundle
; /* Key in global 'xbundles' map. */
106 struct list list_node
; /* In parent 'xbridges' list. */
107 struct xbridge
*xbridge
; /* Parent xbridge. */
109 struct list xports
; /* Contains "struct xport"s. */
111 char *name
; /* Name used in log messages. */
112 struct bond
*bond
; /* Nonnull iff more than one port. */
113 struct lacp
*lacp
; /* LACP handle or null. */
115 enum port_vlan_mode vlan_mode
; /* VLAN mode. */
116 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
117 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
118 * NULL if all VLANs are trunked. */
119 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
120 bool floodable
; /* No port has OFPUTIL_PC_NO_FLOOD set? */
124 struct hmap_node hmap_node
; /* Node in global 'xports' map. */
125 struct ofport_dpif
*ofport
; /* Key in global 'xports map. */
127 struct hmap_node ofp_node
; /* Node in parent xbridge 'xports' map. */
128 ofp_port_t ofp_port
; /* Key in parent xbridge 'xports' map. */
130 odp_port_t odp_port
; /* Datapath port number or ODPP_NONE. */
132 struct list bundle_node
; /* In parent xbundle (if it exists). */
133 struct xbundle
*xbundle
; /* Parent xbundle or null. */
135 struct netdev
*netdev
; /* 'ofport''s netdev. */
137 struct xbridge
*xbridge
; /* Parent bridge. */
138 struct xport
*peer
; /* Patch port peer or null. */
140 enum ofputil_port_config config
; /* OpenFlow port configuration. */
141 enum ofputil_port_state state
; /* OpenFlow port state. */
142 int stp_port_no
; /* STP port number or -1 if not in use. */
144 struct hmap skb_priorities
; /* Map of 'skb_priority_to_dscp's. */
146 bool may_enable
; /* May be enabled in bonds. */
147 bool is_tunnel
; /* Is a tunnel port. */
149 struct cfm
*cfm
; /* CFM handle or null. */
150 struct bfd
*bfd
; /* BFD handle or null. */
154 struct xlate_in
*xin
;
155 struct xlate_out
*xout
;
157 const struct xbridge
*xbridge
;
159 /* Flow at the last commit. */
160 struct flow base_flow
;
162 /* Tunnel IP destination address as received. This is stored separately
163 * as the base_flow.tunnel is cleared on init to reflect the datapath
164 * behavior. Used to make sure not to send tunneled output to ourselves,
165 * which might lead to an infinite loop. This could happen easily
166 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
167 * actually set the tun_dst field. */
168 ovs_be32 orig_tunnel_ip_dst
;
170 /* Stack for the push and pop actions. Each stack element is of type
171 * "union mf_subvalue". */
172 union mf_subvalue init_stack
[1024 / sizeof(union mf_subvalue
)];
175 /* The rule that we are currently translating, or NULL. */
176 struct rule_dpif
*rule
;
178 /* Resubmit statistics, via xlate_table_action(). */
179 int recurse
; /* Current resubmit nesting depth. */
180 int resubmits
; /* Total number of resubmits. */
181 bool in_group
; /* Currently translating ofgroup, if true. */
183 uint32_t orig_skb_priority
; /* Priority when packet arrived. */
184 uint8_t table_id
; /* OpenFlow table ID where flow was found. */
185 uint32_t sflow_n_outputs
; /* Number of output ports. */
186 odp_port_t sflow_odp_port
; /* Output port for composing sFlow action. */
187 uint16_t user_cookie_offset
;/* Used for user_action_cookie fixup. */
188 bool exit
; /* No further actions should be processed. */
190 /* OpenFlow 1.1+ action set.
192 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
193 * When translation is otherwise complete, ofpacts_execute_action_set()
194 * converts it to a set of "struct ofpact"s that can be translated into
195 * datapath actions. */
196 struct ofpbuf action_set
; /* Action set. */
197 uint64_t action_set_stub
[1024 / 8];
200 /* A controller may use OFPP_NONE as the ingress port to indicate that
201 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
202 * when an input bundle is needed for validation (e.g., mirroring or
203 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
204 * any 'port' structs, so care must be taken when dealing with it. */
205 static struct xbundle ofpp_none_bundle
= {
207 .vlan_mode
= PORT_VLAN_TRUNK
210 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
211 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
212 * traffic egressing the 'ofport' with that priority should be marked with. */
213 struct skb_priority_to_dscp
{
214 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'skb_priorities'. */
215 uint32_t skb_priority
; /* Priority of this queue (see struct flow). */
217 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
220 static struct hmap xbridges
= HMAP_INITIALIZER(&xbridges
);
221 static struct hmap xbundles
= HMAP_INITIALIZER(&xbundles
);
222 static struct hmap xports
= HMAP_INITIALIZER(&xports
);
224 static bool may_receive(const struct xport
*, struct xlate_ctx
*);
225 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
227 static void xlate_actions__(struct xlate_in
*, struct xlate_out
*)
228 OVS_REQ_RDLOCK(xlate_rwlock
);
229 static void xlate_normal(struct xlate_ctx
*);
230 static void xlate_report(struct xlate_ctx
*, const char *);
231 static void xlate_table_action(struct xlate_ctx
*, ofp_port_t in_port
,
232 uint8_t table_id
, bool may_packet_in
,
233 bool honor_table_miss
);
234 static bool input_vid_is_valid(uint16_t vid
, struct xbundle
*, bool warn
);
235 static uint16_t input_vid_to_vlan(const struct xbundle
*, uint16_t vid
);
236 static void output_normal(struct xlate_ctx
*, const struct xbundle
*,
238 static void compose_output_action(struct xlate_ctx
*, ofp_port_t ofp_port
);
240 static struct xbridge
*xbridge_lookup(const struct ofproto_dpif
*);
241 static struct xbundle
*xbundle_lookup(const struct ofbundle
*);
242 static struct xport
*xport_lookup(const struct ofport_dpif
*);
243 static struct xport
*get_ofp_port(const struct xbridge
*, ofp_port_t ofp_port
);
244 static struct skb_priority_to_dscp
*get_skb_priority(const struct xport
*,
245 uint32_t skb_priority
);
246 static void clear_skb_priorities(struct xport
*);
247 static bool dscp_from_skb_priority(const struct xport
*, uint32_t skb_priority
,
251 xlate_ofproto_set(struct ofproto_dpif
*ofproto
, const char *name
,
252 struct dpif
*dpif
, struct rule_dpif
*miss_rule
,
253 struct rule_dpif
*no_packet_in_rule
,
254 const struct mac_learning
*ml
, struct stp
*stp
,
255 const struct mbridge
*mbridge
,
256 const struct dpif_sflow
*sflow
,
257 const struct dpif_ipfix
*ipfix
,
258 const struct netflow
*netflow
, enum ofp_config_flags frag
,
259 bool forward_bpdu
, bool has_in_band
,
260 bool variable_length_userdata
,
261 size_t max_mpls_depth
)
263 struct xbridge
*xbridge
= xbridge_lookup(ofproto
);
266 xbridge
= xzalloc(sizeof *xbridge
);
267 xbridge
->ofproto
= ofproto
;
269 hmap_insert(&xbridges
, &xbridge
->hmap_node
, hash_pointer(ofproto
, 0));
270 hmap_init(&xbridge
->xports
);
271 list_init(&xbridge
->xbundles
);
274 if (xbridge
->ml
!= ml
) {
275 mac_learning_unref(xbridge
->ml
);
276 xbridge
->ml
= mac_learning_ref(ml
);
279 if (xbridge
->mbridge
!= mbridge
) {
280 mbridge_unref(xbridge
->mbridge
);
281 xbridge
->mbridge
= mbridge_ref(mbridge
);
284 if (xbridge
->sflow
!= sflow
) {
285 dpif_sflow_unref(xbridge
->sflow
);
286 xbridge
->sflow
= dpif_sflow_ref(sflow
);
289 if (xbridge
->ipfix
!= ipfix
) {
290 dpif_ipfix_unref(xbridge
->ipfix
);
291 xbridge
->ipfix
= dpif_ipfix_ref(ipfix
);
294 if (xbridge
->stp
!= stp
) {
295 stp_unref(xbridge
->stp
);
296 xbridge
->stp
= stp_ref(stp
);
299 if (xbridge
->netflow
!= netflow
) {
300 netflow_unref(xbridge
->netflow
);
301 xbridge
->netflow
= netflow_ref(netflow
);
305 xbridge
->name
= xstrdup(name
);
307 xbridge
->dpif
= dpif
;
308 xbridge
->forward_bpdu
= forward_bpdu
;
309 xbridge
->has_in_band
= has_in_band
;
310 xbridge
->frag
= frag
;
311 xbridge
->miss_rule
= miss_rule
;
312 xbridge
->no_packet_in_rule
= no_packet_in_rule
;
313 xbridge
->variable_length_userdata
= variable_length_userdata
;
314 xbridge
->max_mpls_depth
= max_mpls_depth
;
318 xlate_remove_ofproto(struct ofproto_dpif
*ofproto
)
320 struct xbridge
*xbridge
= xbridge_lookup(ofproto
);
321 struct xbundle
*xbundle
, *next_xbundle
;
322 struct xport
*xport
, *next_xport
;
328 HMAP_FOR_EACH_SAFE (xport
, next_xport
, ofp_node
, &xbridge
->xports
) {
329 xlate_ofport_remove(xport
->ofport
);
332 LIST_FOR_EACH_SAFE (xbundle
, next_xbundle
, list_node
, &xbridge
->xbundles
) {
333 xlate_bundle_remove(xbundle
->ofbundle
);
336 hmap_remove(&xbridges
, &xbridge
->hmap_node
);
337 mac_learning_unref(xbridge
->ml
);
338 mbridge_unref(xbridge
->mbridge
);
339 dpif_sflow_unref(xbridge
->sflow
);
340 dpif_ipfix_unref(xbridge
->ipfix
);
341 stp_unref(xbridge
->stp
);
342 hmap_destroy(&xbridge
->xports
);
348 xlate_bundle_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
349 const char *name
, enum port_vlan_mode vlan_mode
, int vlan
,
350 unsigned long *trunks
, bool use_priority_tags
,
351 const struct bond
*bond
, const struct lacp
*lacp
,
354 struct xbundle
*xbundle
= xbundle_lookup(ofbundle
);
357 xbundle
= xzalloc(sizeof *xbundle
);
358 xbundle
->ofbundle
= ofbundle
;
359 xbundle
->xbridge
= xbridge_lookup(ofproto
);
361 hmap_insert(&xbundles
, &xbundle
->hmap_node
, hash_pointer(ofbundle
, 0));
362 list_insert(&xbundle
->xbridge
->xbundles
, &xbundle
->list_node
);
363 list_init(&xbundle
->xports
);
366 ovs_assert(xbundle
->xbridge
);
369 xbundle
->name
= xstrdup(name
);
371 xbundle
->vlan_mode
= vlan_mode
;
372 xbundle
->vlan
= vlan
;
373 xbundle
->trunks
= trunks
;
374 xbundle
->use_priority_tags
= use_priority_tags
;
375 xbundle
->floodable
= floodable
;
377 if (xbundle
->bond
!= bond
) {
378 bond_unref(xbundle
->bond
);
379 xbundle
->bond
= bond_ref(bond
);
382 if (xbundle
->lacp
!= lacp
) {
383 lacp_unref(xbundle
->lacp
);
384 xbundle
->lacp
= lacp_ref(lacp
);
389 xlate_bundle_remove(struct ofbundle
*ofbundle
)
391 struct xbundle
*xbundle
= xbundle_lookup(ofbundle
);
392 struct xport
*xport
, *next
;
398 LIST_FOR_EACH_SAFE (xport
, next
, bundle_node
, &xbundle
->xports
) {
399 list_remove(&xport
->bundle_node
);
400 xport
->xbundle
= NULL
;
403 hmap_remove(&xbundles
, &xbundle
->hmap_node
);
404 list_remove(&xbundle
->list_node
);
405 bond_unref(xbundle
->bond
);
406 lacp_unref(xbundle
->lacp
);
412 xlate_ofport_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
413 struct ofport_dpif
*ofport
, ofp_port_t ofp_port
,
414 odp_port_t odp_port
, const struct netdev
*netdev
,
415 const struct cfm
*cfm
, const struct bfd
*bfd
,
416 struct ofport_dpif
*peer
, int stp_port_no
,
417 const struct ofproto_port_queue
*qdscp_list
, size_t n_qdscp
,
418 enum ofputil_port_config config
,
419 enum ofputil_port_state state
, bool is_tunnel
,
422 struct xport
*xport
= xport_lookup(ofport
);
426 xport
= xzalloc(sizeof *xport
);
427 xport
->ofport
= ofport
;
428 xport
->xbridge
= xbridge_lookup(ofproto
);
429 xport
->ofp_port
= ofp_port
;
431 hmap_init(&xport
->skb_priorities
);
432 hmap_insert(&xports
, &xport
->hmap_node
, hash_pointer(ofport
, 0));
433 hmap_insert(&xport
->xbridge
->xports
, &xport
->ofp_node
,
434 hash_ofp_port(xport
->ofp_port
));
437 ovs_assert(xport
->ofp_port
== ofp_port
);
439 xport
->config
= config
;
440 xport
->state
= state
;
441 xport
->stp_port_no
= stp_port_no
;
442 xport
->is_tunnel
= is_tunnel
;
443 xport
->may_enable
= may_enable
;
444 xport
->odp_port
= odp_port
;
446 if (xport
->netdev
!= netdev
) {
447 netdev_close(xport
->netdev
);
448 xport
->netdev
= netdev_ref(netdev
);
451 if (xport
->cfm
!= cfm
) {
452 cfm_unref(xport
->cfm
);
453 xport
->cfm
= cfm_ref(cfm
);
456 if (xport
->bfd
!= bfd
) {
457 bfd_unref(xport
->bfd
);
458 xport
->bfd
= bfd_ref(bfd
);
462 xport
->peer
->peer
= NULL
;
464 xport
->peer
= xport_lookup(peer
);
466 xport
->peer
->peer
= xport
;
469 if (xport
->xbundle
) {
470 list_remove(&xport
->bundle_node
);
472 xport
->xbundle
= xbundle_lookup(ofbundle
);
473 if (xport
->xbundle
) {
474 list_insert(&xport
->xbundle
->xports
, &xport
->bundle_node
);
477 clear_skb_priorities(xport
);
478 for (i
= 0; i
< n_qdscp
; i
++) {
479 struct skb_priority_to_dscp
*pdscp
;
480 uint32_t skb_priority
;
482 if (dpif_queue_to_priority(xport
->xbridge
->dpif
, qdscp_list
[i
].queue
,
487 pdscp
= xmalloc(sizeof *pdscp
);
488 pdscp
->skb_priority
= skb_priority
;
489 pdscp
->dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
490 hmap_insert(&xport
->skb_priorities
, &pdscp
->hmap_node
,
491 hash_int(pdscp
->skb_priority
, 0));
496 xlate_ofport_remove(struct ofport_dpif
*ofport
)
498 struct xport
*xport
= xport_lookup(ofport
);
505 xport
->peer
->peer
= NULL
;
509 if (xport
->xbundle
) {
510 list_remove(&xport
->bundle_node
);
513 clear_skb_priorities(xport
);
514 hmap_destroy(&xport
->skb_priorities
);
516 hmap_remove(&xports
, &xport
->hmap_node
);
517 hmap_remove(&xport
->xbridge
->xports
, &xport
->ofp_node
);
519 netdev_close(xport
->netdev
);
520 cfm_unref(xport
->cfm
);
521 bfd_unref(xport
->bfd
);
525 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
526 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
527 * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
528 * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
529 * 'netflow' with the appropriate handles for those protocols if they're
530 * enabled. Caller is responsible for unrefing them.
532 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
533 * 'flow''s in_port to OFPP_NONE.
535 * This function does post-processing on data returned from
536 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
537 * of the upcall processing logic. In particular, if the extracted in_port is
538 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
539 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
540 * a VLAN header onto 'packet' (if it is nonnull).
542 * Similarly, this function also includes some logic to help with tunnels. It
543 * may modify 'flow' as necessary to make the tunneling implementation
544 * transparent to the upcall processing logic.
546 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
547 * or some other positive errno if there are other problems. */
549 xlate_receive(const struct dpif_backer
*backer
, struct ofpbuf
*packet
,
550 const struct nlattr
*key
, size_t key_len
, struct flow
*flow
,
551 struct ofproto_dpif
**ofproto
, struct dpif_ipfix
**ipfix
,
552 struct dpif_sflow
**sflow
, struct netflow
**netflow
,
553 odp_port_t
*odp_in_port
)
555 const struct xport
*xport
;
558 ovs_rwlock_rdlock(&xlate_rwlock
);
559 if (odp_flow_key_to_flow(key
, key_len
, flow
) == ODP_FIT_ERROR
) {
565 *odp_in_port
= flow
->in_port
.odp_port
;
568 xport
= xport_lookup(tnl_port_should_receive(flow
)
569 ? tnl_port_receive(flow
)
570 : odp_port_to_ofport(backer
, flow
->in_port
.odp_port
));
572 flow
->in_port
.ofp_port
= xport
? xport
->ofp_port
: OFPP_NONE
;
577 if (vsp_adjust_flow(xport
->xbridge
->ofproto
, flow
)) {
579 /* Make the packet resemble the flow, so that it gets sent to
580 * an OpenFlow controller properly, so that it looks correct
581 * for sFlow, and so that flow_extract() will get the correct
582 * vlan_tci if it is called on 'packet'. */
583 eth_push_vlan(packet
, htons(ETH_TYPE_VLAN
), flow
->vlan_tci
);
589 *ofproto
= xport
->xbridge
->ofproto
;
593 *ipfix
= dpif_ipfix_ref(xport
->xbridge
->ipfix
);
597 *sflow
= dpif_sflow_ref(xport
->xbridge
->sflow
);
601 *netflow
= netflow_ref(xport
->xbridge
->netflow
);
605 ovs_rwlock_unlock(&xlate_rwlock
);
609 static struct xbridge
*
610 xbridge_lookup(const struct ofproto_dpif
*ofproto
)
612 struct xbridge
*xbridge
;
618 HMAP_FOR_EACH_IN_BUCKET (xbridge
, hmap_node
, hash_pointer(ofproto
, 0),
620 if (xbridge
->ofproto
== ofproto
) {
627 static struct xbundle
*
628 xbundle_lookup(const struct ofbundle
*ofbundle
)
630 struct xbundle
*xbundle
;
636 HMAP_FOR_EACH_IN_BUCKET (xbundle
, hmap_node
, hash_pointer(ofbundle
, 0),
638 if (xbundle
->ofbundle
== ofbundle
) {
645 static struct xport
*
646 xport_lookup(const struct ofport_dpif
*ofport
)
654 HMAP_FOR_EACH_IN_BUCKET (xport
, hmap_node
, hash_pointer(ofport
, 0),
656 if (xport
->ofport
== ofport
) {
663 static struct stp_port
*
664 xport_get_stp_port(const struct xport
*xport
)
666 return xport
->xbridge
->stp
&& xport
->stp_port_no
!= -1
667 ? stp_get_port(xport
->xbridge
->stp
, xport
->stp_port_no
)
672 xport_stp_learn_state(const struct xport
*xport
)
674 struct stp_port
*sp
= xport_get_stp_port(xport
);
675 return stp_learn_in_state(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
679 xport_stp_forward_state(const struct xport
*xport
)
681 struct stp_port
*sp
= xport_get_stp_port(xport
);
682 return stp_forward_in_state(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
686 xport_stp_listen_state(const struct xport
*xport
)
688 struct stp_port
*sp
= xport_get_stp_port(xport
);
689 return stp_listen_in_state(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
692 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
693 * were used to make the determination.*/
695 stp_should_process_flow(const struct flow
*flow
, struct flow_wildcards
*wc
)
697 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
698 return eth_addr_equals(flow
->dl_dst
, eth_addr_stp
);
702 stp_process_packet(const struct xport
*xport
, const struct ofpbuf
*packet
)
704 struct stp_port
*sp
= xport_get_stp_port(xport
);
705 struct ofpbuf payload
= *packet
;
706 struct eth_header
*eth
= payload
.data
;
708 /* Sink packets on ports that have STP disabled when the bridge has
710 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
714 /* Trim off padding on payload. */
715 if (payload
.size
> ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
716 payload
.size
= ntohs(eth
->eth_type
) + ETH_HEADER_LEN
;
719 if (ofpbuf_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
720 stp_received_bpdu(sp
, payload
.data
, payload
.size
);
724 static struct xport
*
725 get_ofp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
729 HMAP_FOR_EACH_IN_BUCKET (xport
, ofp_node
, hash_ofp_port(ofp_port
),
731 if (xport
->ofp_port
== ofp_port
) {
739 ofp_port_to_odp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
741 const struct xport
*xport
= get_ofp_port(xbridge
, ofp_port
);
742 return xport
? xport
->odp_port
: ODPP_NONE
;
746 odp_port_is_alive(const struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
750 xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
751 if (!xport
|| xport
->config
& OFPUTIL_PC_PORT_DOWN
||
752 xport
->state
& OFPUTIL_PS_LINK_DOWN
) {
759 static const struct ofputil_bucket
*
760 group_first_live_bucket(const struct xlate_ctx
*, const struct group_dpif
*,
764 group_is_alive(const struct xlate_ctx
*ctx
, uint32_t group_id
, int depth
)
766 struct group_dpif
*group
;
769 hit
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
, &group
);
774 hit
= group_first_live_bucket(ctx
, group
, depth
) != NULL
;
776 group_dpif_release(group
);
780 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
783 bucket_is_alive(const struct xlate_ctx
*ctx
,
784 const struct ofputil_bucket
*bucket
, int depth
)
786 if (depth
>= MAX_LIVENESS_RECURSION
) {
787 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
789 VLOG_WARN_RL(&rl
, "bucket chaining exceeded %d links",
790 MAX_LIVENESS_RECURSION
);
794 return !ofputil_bucket_has_liveness(bucket
) ||
795 (bucket
->watch_port
!= OFPP_ANY
&&
796 odp_port_is_alive(ctx
, bucket
->watch_port
)) ||
797 (bucket
->watch_group
!= OFPG_ANY
&&
798 group_is_alive(ctx
, bucket
->watch_group
, depth
+ 1));
801 static const struct ofputil_bucket
*
802 group_first_live_bucket(const struct xlate_ctx
*ctx
,
803 const struct group_dpif
*group
, int depth
)
805 struct ofputil_bucket
*bucket
;
806 const struct list
*buckets
;
808 group_dpif_get_buckets(group
, &buckets
);
809 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
810 if (bucket_is_alive(ctx
, bucket
, depth
)) {
818 static const struct ofputil_bucket
*
819 group_best_live_bucket(const struct xlate_ctx
*ctx
,
820 const struct group_dpif
*group
,
823 const struct ofputil_bucket
*best_bucket
= NULL
;
824 uint32_t best_score
= 0;
827 const struct ofputil_bucket
*bucket
;
828 const struct list
*buckets
;
830 group_dpif_get_buckets(group
, &buckets
);
831 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
832 if (bucket_is_alive(ctx
, bucket
, 0)) {
833 uint32_t score
= (hash_int(i
, basis
) & 0xffff) * bucket
->weight
;
834 if (score
>= best_score
) {
835 best_bucket
= bucket
;
846 xbundle_trunks_vlan(const struct xbundle
*bundle
, uint16_t vlan
)
848 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
849 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
853 xbundle_includes_vlan(const struct xbundle
*xbundle
, uint16_t vlan
)
855 return vlan
== xbundle
->vlan
|| xbundle_trunks_vlan(xbundle
, vlan
);
859 xbundle_mirror_out(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
861 return xbundle
!= &ofpp_none_bundle
862 ? mirror_bundle_out(xbridge
->mbridge
, xbundle
->ofbundle
)
867 xbundle_mirror_src(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
869 return xbundle
!= &ofpp_none_bundle
870 ? mirror_bundle_src(xbridge
->mbridge
, xbundle
->ofbundle
)
875 xbundle_mirror_dst(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
877 return xbundle
!= &ofpp_none_bundle
878 ? mirror_bundle_dst(xbridge
->mbridge
, xbundle
->ofbundle
)
882 static struct xbundle
*
883 lookup_input_bundle(const struct xbridge
*xbridge
, ofp_port_t in_port
,
884 bool warn
, struct xport
**in_xportp
)
888 /* Find the port and bundle for the received packet. */
889 xport
= get_ofp_port(xbridge
, in_port
);
893 if (xport
&& xport
->xbundle
) {
894 return xport
->xbundle
;
897 /* Special-case OFPP_NONE, which a controller may use as the ingress
898 * port for traffic that it is sourcing. */
899 if (in_port
== OFPP_NONE
) {
900 return &ofpp_none_bundle
;
903 /* Odd. A few possible reasons here:
905 * - We deleted a port but there are still a few packets queued up
908 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
909 * we don't know about.
911 * - The ofproto client didn't configure the port as part of a bundle.
912 * This is particularly likely to happen if a packet was received on the
913 * port after it was created, but before the client had a chance to
914 * configure its bundle.
917 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
919 VLOG_WARN_RL(&rl
, "bridge %s: received packet on unknown "
920 "port %"PRIu16
, xbridge
->name
, in_port
);
926 add_mirror_actions(struct xlate_ctx
*ctx
, const struct flow
*orig_flow
)
928 const struct xbridge
*xbridge
= ctx
->xbridge
;
929 mirror_mask_t mirrors
;
930 struct xbundle
*in_xbundle
;
934 mirrors
= ctx
->xout
->mirrors
;
935 ctx
->xout
->mirrors
= 0;
937 in_xbundle
= lookup_input_bundle(xbridge
, orig_flow
->in_port
.ofp_port
,
938 ctx
->xin
->packet
!= NULL
, NULL
);
942 mirrors
|= xbundle_mirror_src(xbridge
, in_xbundle
);
944 /* Drop frames on bundles reserved for mirroring. */
945 if (xbundle_mirror_out(xbridge
, in_xbundle
)) {
946 if (ctx
->xin
->packet
!= NULL
) {
947 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
948 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
949 "%s, which is reserved exclusively for mirroring",
950 ctx
->xbridge
->name
, in_xbundle
->name
);
952 ofpbuf_clear(&ctx
->xout
->odp_actions
);
957 vid
= vlan_tci_to_vid(orig_flow
->vlan_tci
);
958 if (!input_vid_is_valid(vid
, in_xbundle
, ctx
->xin
->packet
!= NULL
)) {
961 vlan
= input_vid_to_vlan(in_xbundle
, vid
);
967 /* Restore the original packet before adding the mirror actions. */
968 ctx
->xin
->flow
= *orig_flow
;
971 mirror_mask_t dup_mirrors
;
972 struct ofbundle
*out
;
973 unsigned long *vlans
;
978 has_mirror
= mirror_get(xbridge
->mbridge
, raw_ctz(mirrors
),
979 &vlans
, &dup_mirrors
, &out
, &out_vlan
);
980 ovs_assert(has_mirror
);
983 ctx
->xout
->wc
.masks
.vlan_tci
|= htons(VLAN_CFI
| VLAN_VID_MASK
);
985 vlan_mirrored
= !vlans
|| bitmap_is_set(vlans
, vlan
);
988 if (!vlan_mirrored
) {
989 mirrors
= zero_rightmost_1bit(mirrors
);
993 mirrors
&= ~dup_mirrors
;
994 ctx
->xout
->mirrors
|= dup_mirrors
;
996 struct xbundle
*out_xbundle
= xbundle_lookup(out
);
998 output_normal(ctx
, out_xbundle
, vlan
);
1000 } else if (vlan
!= out_vlan
1001 && !eth_addr_is_reserved(orig_flow
->dl_dst
)) {
1002 struct xbundle
*xbundle
;
1004 LIST_FOR_EACH (xbundle
, list_node
, &xbridge
->xbundles
) {
1005 if (xbundle_includes_vlan(xbundle
, out_vlan
)
1006 && !xbundle_mirror_out(xbridge
, xbundle
)) {
1007 output_normal(ctx
, xbundle
, out_vlan
);
1014 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1015 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1016 * the bundle on which the packet was received, returns the VLAN to which the
1019 * Both 'vid' and the return value are in the range 0...4095. */
1021 input_vid_to_vlan(const struct xbundle
*in_xbundle
, uint16_t vid
)
1023 switch (in_xbundle
->vlan_mode
) {
1024 case PORT_VLAN_ACCESS
:
1025 return in_xbundle
->vlan
;
1028 case PORT_VLAN_TRUNK
:
1031 case PORT_VLAN_NATIVE_UNTAGGED
:
1032 case PORT_VLAN_NATIVE_TAGGED
:
1033 return vid
? vid
: in_xbundle
->vlan
;
1040 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1041 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1044 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1045 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1048 input_vid_is_valid(uint16_t vid
, struct xbundle
*in_xbundle
, bool warn
)
1050 /* Allow any VID on the OFPP_NONE port. */
1051 if (in_xbundle
== &ofpp_none_bundle
) {
1055 switch (in_xbundle
->vlan_mode
) {
1056 case PORT_VLAN_ACCESS
:
1059 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1060 VLOG_WARN_RL(&rl
, "dropping VLAN %"PRIu16
" tagged "
1061 "packet received on port %s configured as VLAN "
1062 "%"PRIu16
" access port", vid
, in_xbundle
->name
,
1069 case PORT_VLAN_NATIVE_UNTAGGED
:
1070 case PORT_VLAN_NATIVE_TAGGED
:
1072 /* Port must always carry its native VLAN. */
1076 case PORT_VLAN_TRUNK
:
1077 if (!xbundle_includes_vlan(in_xbundle
, vid
)) {
1079 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1080 VLOG_WARN_RL(&rl
, "dropping VLAN %"PRIu16
" packet "
1081 "received on port %s not configured for trunking "
1082 "VLAN %"PRIu16
, vid
, in_xbundle
->name
, vid
);
1094 /* Given 'vlan', the VLAN that a packet belongs to, and
1095 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1096 * that should be included in the 802.1Q header. (If the return value is 0,
1097 * then the 802.1Q header should only be included in the packet if there is a
1100 * Both 'vlan' and the return value are in the range 0...4095. */
1102 output_vlan_to_vid(const struct xbundle
*out_xbundle
, uint16_t vlan
)
1104 switch (out_xbundle
->vlan_mode
) {
1105 case PORT_VLAN_ACCESS
:
1108 case PORT_VLAN_TRUNK
:
1109 case PORT_VLAN_NATIVE_TAGGED
:
1112 case PORT_VLAN_NATIVE_UNTAGGED
:
1113 return vlan
== out_xbundle
->vlan
? 0 : vlan
;
1121 output_normal(struct xlate_ctx
*ctx
, const struct xbundle
*out_xbundle
,
1124 ovs_be16
*flow_tci
= &ctx
->xin
->flow
.vlan_tci
;
1126 ovs_be16 tci
, old_tci
;
1127 struct xport
*xport
;
1129 vid
= output_vlan_to_vid(out_xbundle
, vlan
);
1130 if (list_is_empty(&out_xbundle
->xports
)) {
1131 /* Partially configured bundle with no slaves. Drop the packet. */
1133 } else if (!out_xbundle
->bond
) {
1134 xport
= CONTAINER_OF(list_front(&out_xbundle
->xports
), struct xport
,
1137 struct ofport_dpif
*ofport
;
1139 ofport
= bond_choose_output_slave(out_xbundle
->bond
, &ctx
->xin
->flow
,
1140 &ctx
->xout
->wc
, vid
);
1141 xport
= xport_lookup(ofport
);
1144 /* No slaves enabled, so drop packet. */
1148 if (ctx
->xin
->resubmit_stats
) {
1149 bond_account(out_xbundle
->bond
, &ctx
->xin
->flow
, vid
,
1150 ctx
->xin
->resubmit_stats
->n_bytes
);
1154 old_tci
= *flow_tci
;
1156 if (tci
|| out_xbundle
->use_priority_tags
) {
1157 tci
|= *flow_tci
& htons(VLAN_PCP_MASK
);
1159 tci
|= htons(VLAN_CFI
);
1164 compose_output_action(ctx
, xport
->ofp_port
);
1165 *flow_tci
= old_tci
;
1168 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1169 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1170 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1172 is_gratuitous_arp(const struct flow
*flow
, struct flow_wildcards
*wc
)
1174 if (flow
->dl_type
!= htons(ETH_TYPE_ARP
)) {
1178 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1179 if (!eth_addr_is_broadcast(flow
->dl_dst
)) {
1183 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
1184 if (flow
->nw_proto
== ARP_OP_REPLY
) {
1186 } else if (flow
->nw_proto
== ARP_OP_REQUEST
) {
1187 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
1188 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
1190 return flow
->nw_src
== flow
->nw_dst
;
1196 /* Checks whether a MAC learning update is necessary for MAC learning table
1197 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1200 * Most packets processed through the MAC learning table do not actually
1201 * change it in any way. This function requires only a read lock on the MAC
1202 * learning table, so it is much cheaper in this common case.
1204 * Keep the code here synchronized with that in update_learning_table__()
1207 is_mac_learning_update_needed(const struct mac_learning
*ml
,
1208 const struct flow
*flow
,
1209 struct flow_wildcards
*wc
,
1210 int vlan
, struct xbundle
*in_xbundle
)
1211 OVS_REQ_RDLOCK(ml
->rwlock
)
1213 struct mac_entry
*mac
;
1215 if (!mac_learning_may_learn(ml
, flow
->dl_src
, vlan
)) {
1219 mac
= mac_learning_lookup(ml
, flow
->dl_src
, vlan
);
1220 if (!mac
|| mac_entry_age(ml
, mac
)) {
1224 if (is_gratuitous_arp(flow
, wc
)) {
1225 /* We don't want to learn from gratuitous ARP packets that are
1226 * reflected back over bond slaves so we lock the learning table. */
1227 if (!in_xbundle
->bond
) {
1229 } else if (mac_entry_is_grat_arp_locked(mac
)) {
1234 return mac
->port
.p
!= in_xbundle
->ofbundle
;
1238 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1239 * received on 'in_xbundle' in 'vlan'.
1241 * This code repeats all the checks in is_mac_learning_update_needed() because
1242 * the lock was released between there and here and thus the MAC learning state
1243 * could have changed.
1245 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1248 update_learning_table__(const struct xbridge
*xbridge
,
1249 const struct flow
*flow
, struct flow_wildcards
*wc
,
1250 int vlan
, struct xbundle
*in_xbundle
)
1251 OVS_REQ_WRLOCK(xbridge
->ml
->rwlock
)
1253 struct mac_entry
*mac
;
1255 if (!mac_learning_may_learn(xbridge
->ml
, flow
->dl_src
, vlan
)) {
1259 mac
= mac_learning_insert(xbridge
->ml
, flow
->dl_src
, vlan
);
1260 if (is_gratuitous_arp(flow
, wc
)) {
1261 /* We don't want to learn from gratuitous ARP packets that are
1262 * reflected back over bond slaves so we lock the learning table. */
1263 if (!in_xbundle
->bond
) {
1264 mac_entry_set_grat_arp_lock(mac
);
1265 } else if (mac_entry_is_grat_arp_locked(mac
)) {
1270 if (mac
->port
.p
!= in_xbundle
->ofbundle
) {
1271 /* The log messages here could actually be useful in debugging,
1272 * so keep the rate limit relatively high. */
1273 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
1275 VLOG_DBG_RL(&rl
, "bridge %s: learned that "ETH_ADDR_FMT
" is "
1276 "on port %s in VLAN %d",
1277 xbridge
->name
, ETH_ADDR_ARGS(flow
->dl_src
),
1278 in_xbundle
->name
, vlan
);
1280 mac
->port
.p
= in_xbundle
->ofbundle
;
1281 mac_learning_changed(xbridge
->ml
);
1286 update_learning_table(const struct xbridge
*xbridge
,
1287 const struct flow
*flow
, struct flow_wildcards
*wc
,
1288 int vlan
, struct xbundle
*in_xbundle
)
1292 /* Don't learn the OFPP_NONE port. */
1293 if (in_xbundle
== &ofpp_none_bundle
) {
1297 /* First try the common case: no change to MAC learning table. */
1298 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
1299 need_update
= is_mac_learning_update_needed(xbridge
->ml
, flow
, wc
, vlan
,
1301 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1304 /* Slow path: MAC learning table might need an update. */
1305 ovs_rwlock_wrlock(&xbridge
->ml
->rwlock
);
1306 update_learning_table__(xbridge
, flow
, wc
, vlan
, in_xbundle
);
1307 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1311 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1312 * dropped. Returns true if they may be forwarded, false if they should be
1315 * 'in_port' must be the xport that corresponds to flow->in_port.
1316 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1318 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1319 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1320 * checked by input_vid_is_valid().
1322 * May also add tags to '*tags', although the current implementation only does
1323 * so in one special case.
1326 is_admissible(struct xlate_ctx
*ctx
, struct xport
*in_port
,
1329 struct xbundle
*in_xbundle
= in_port
->xbundle
;
1330 const struct xbridge
*xbridge
= ctx
->xbridge
;
1331 struct flow
*flow
= &ctx
->xin
->flow
;
1333 /* Drop frames for reserved multicast addresses
1334 * only if forward_bpdu option is absent. */
1335 if (!xbridge
->forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
1336 xlate_report(ctx
, "packet has reserved destination MAC, dropping");
1340 if (in_xbundle
->bond
) {
1341 struct mac_entry
*mac
;
1343 switch (bond_check_admissibility(in_xbundle
->bond
, in_port
->ofport
,
1349 xlate_report(ctx
, "bonding refused admissibility, dropping");
1352 case BV_DROP_IF_MOVED
:
1353 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
1354 mac
= mac_learning_lookup(xbridge
->ml
, flow
->dl_src
, vlan
);
1355 if (mac
&& mac
->port
.p
!= in_xbundle
->ofbundle
&&
1356 (!is_gratuitous_arp(flow
, &ctx
->xout
->wc
)
1357 || mac_entry_is_grat_arp_locked(mac
))) {
1358 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1359 xlate_report(ctx
, "SLB bond thinks this packet looped back, "
1363 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
1372 xlate_normal(struct xlate_ctx
*ctx
)
1374 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1375 struct flow
*flow
= &ctx
->xin
->flow
;
1376 struct xbundle
*in_xbundle
;
1377 struct xport
*in_port
;
1378 struct mac_entry
*mac
;
1383 ctx
->xout
->has_normal
= true;
1385 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
1386 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1387 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1389 in_xbundle
= lookup_input_bundle(ctx
->xbridge
, flow
->in_port
.ofp_port
,
1390 ctx
->xin
->packet
!= NULL
, &in_port
);
1392 xlate_report(ctx
, "no input bundle, dropping");
1396 /* Drop malformed frames. */
1397 if (flow
->dl_type
== htons(ETH_TYPE_VLAN
) &&
1398 !(flow
->vlan_tci
& htons(VLAN_CFI
))) {
1399 if (ctx
->xin
->packet
!= NULL
) {
1400 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1401 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet with partial "
1402 "VLAN tag received on port %s",
1403 ctx
->xbridge
->name
, in_xbundle
->name
);
1405 xlate_report(ctx
, "partial VLAN tag, dropping");
1409 /* Drop frames on bundles reserved for mirroring. */
1410 if (xbundle_mirror_out(ctx
->xbridge
, in_xbundle
)) {
1411 if (ctx
->xin
->packet
!= NULL
) {
1412 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
1413 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet received on port "
1414 "%s, which is reserved exclusively for mirroring",
1415 ctx
->xbridge
->name
, in_xbundle
->name
);
1417 xlate_report(ctx
, "input port is mirror output port, dropping");
1422 vid
= vlan_tci_to_vid(flow
->vlan_tci
);
1423 if (!input_vid_is_valid(vid
, in_xbundle
, ctx
->xin
->packet
!= NULL
)) {
1424 xlate_report(ctx
, "disallowed VLAN VID for this input port, dropping");
1427 vlan
= input_vid_to_vlan(in_xbundle
, vid
);
1429 /* Check other admissibility requirements. */
1430 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
1434 /* Learn source MAC. */
1435 if (ctx
->xin
->may_learn
) {
1436 update_learning_table(ctx
->xbridge
, flow
, wc
, vlan
, in_xbundle
);
1439 /* Determine output bundle. */
1440 ovs_rwlock_rdlock(&ctx
->xbridge
->ml
->rwlock
);
1441 mac
= mac_learning_lookup(ctx
->xbridge
->ml
, flow
->dl_dst
, vlan
);
1442 mac_port
= mac
? mac
->port
.p
: NULL
;
1443 ovs_rwlock_unlock(&ctx
->xbridge
->ml
->rwlock
);
1446 struct xbundle
*mac_xbundle
= xbundle_lookup(mac_port
);
1447 if (mac_xbundle
&& mac_xbundle
!= in_xbundle
) {
1448 xlate_report(ctx
, "forwarding to learned port");
1449 output_normal(ctx
, mac_xbundle
, vlan
);
1450 } else if (!mac_xbundle
) {
1451 xlate_report(ctx
, "learned port is unknown, dropping");
1453 xlate_report(ctx
, "learned port is input port, dropping");
1456 struct xbundle
*xbundle
;
1458 xlate_report(ctx
, "no learned MAC for destination, flooding");
1459 LIST_FOR_EACH (xbundle
, list_node
, &ctx
->xbridge
->xbundles
) {
1460 if (xbundle
!= in_xbundle
1461 && xbundle_includes_vlan(xbundle
, vlan
)
1462 && xbundle
->floodable
1463 && !xbundle_mirror_out(ctx
->xbridge
, xbundle
)) {
1464 output_normal(ctx
, xbundle
, vlan
);
1467 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
1471 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1472 * the number of packets out of UINT32_MAX to sample. The given
1473 * cookie is passed back in the callback for each sampled packet.
1476 compose_sample_action(const struct xbridge
*xbridge
,
1477 struct ofpbuf
*odp_actions
,
1478 const struct flow
*flow
,
1479 const uint32_t probability
,
1480 const union user_action_cookie
*cookie
,
1481 const size_t cookie_size
)
1483 size_t sample_offset
, actions_offset
;
1484 odp_port_t odp_port
;
1488 sample_offset
= nl_msg_start_nested(odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
1490 nl_msg_put_u32(odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
, probability
);
1492 actions_offset
= nl_msg_start_nested(odp_actions
, OVS_SAMPLE_ATTR_ACTIONS
);
1494 odp_port
= ofp_port_to_odp_port(xbridge
, flow
->in_port
.ofp_port
);
1495 pid
= dpif_port_get_pid(xbridge
->dpif
, odp_port
, 0);
1496 cookie_offset
= odp_put_userspace_action(pid
, cookie
, cookie_size
, odp_actions
);
1498 nl_msg_end_nested(odp_actions
, actions_offset
);
1499 nl_msg_end_nested(odp_actions
, sample_offset
);
1500 return cookie_offset
;
1504 compose_sflow_cookie(const struct xbridge
*xbridge
, ovs_be16 vlan_tci
,
1505 odp_port_t odp_port
, unsigned int n_outputs
,
1506 union user_action_cookie
*cookie
)
1510 cookie
->type
= USER_ACTION_COOKIE_SFLOW
;
1511 cookie
->sflow
.vlan_tci
= vlan_tci
;
1513 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1514 * port information") for the interpretation of cookie->output. */
1515 switch (n_outputs
) {
1517 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1518 cookie
->sflow
.output
= 0x40000000 | 256;
1522 ifindex
= dpif_sflow_odp_port_to_ifindex(xbridge
->sflow
, odp_port
);
1524 cookie
->sflow
.output
= ifindex
;
1529 /* 0x80000000 means "multiple output ports. */
1530 cookie
->sflow
.output
= 0x80000000 | n_outputs
;
1535 /* Compose SAMPLE action for sFlow bridge sampling. */
1537 compose_sflow_action(const struct xbridge
*xbridge
,
1538 struct ofpbuf
*odp_actions
,
1539 const struct flow
*flow
,
1540 odp_port_t odp_port
)
1542 uint32_t probability
;
1543 union user_action_cookie cookie
;
1545 if (!xbridge
->sflow
|| flow
->in_port
.ofp_port
== OFPP_NONE
) {
1549 probability
= dpif_sflow_get_probability(xbridge
->sflow
);
1550 compose_sflow_cookie(xbridge
, htons(0), odp_port
,
1551 odp_port
== ODPP_NONE
? 0 : 1, &cookie
);
1553 return compose_sample_action(xbridge
, odp_actions
, flow
, probability
,
1554 &cookie
, sizeof cookie
.sflow
);
1558 compose_flow_sample_cookie(uint16_t probability
, uint32_t collector_set_id
,
1559 uint32_t obs_domain_id
, uint32_t obs_point_id
,
1560 union user_action_cookie
*cookie
)
1562 cookie
->type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
1563 cookie
->flow_sample
.probability
= probability
;
1564 cookie
->flow_sample
.collector_set_id
= collector_set_id
;
1565 cookie
->flow_sample
.obs_domain_id
= obs_domain_id
;
1566 cookie
->flow_sample
.obs_point_id
= obs_point_id
;
1570 compose_ipfix_cookie(union user_action_cookie
*cookie
)
1572 cookie
->type
= USER_ACTION_COOKIE_IPFIX
;
1575 /* Compose SAMPLE action for IPFIX bridge sampling. */
1577 compose_ipfix_action(const struct xbridge
*xbridge
,
1578 struct ofpbuf
*odp_actions
,
1579 const struct flow
*flow
)
1581 uint32_t probability
;
1582 union user_action_cookie cookie
;
1584 if (!xbridge
->ipfix
|| flow
->in_port
.ofp_port
== OFPP_NONE
) {
1588 probability
= dpif_ipfix_get_bridge_exporter_probability(xbridge
->ipfix
);
1589 compose_ipfix_cookie(&cookie
);
1591 compose_sample_action(xbridge
, odp_actions
, flow
, probability
,
1592 &cookie
, sizeof cookie
.ipfix
);
1595 /* SAMPLE action for sFlow must be first action in any given list of
1596 * actions. At this point we do not have all information required to
1597 * build it. So try to build sample action as complete as possible. */
1599 add_sflow_action(struct xlate_ctx
*ctx
)
1601 ctx
->user_cookie_offset
= compose_sflow_action(ctx
->xbridge
,
1602 &ctx
->xout
->odp_actions
,
1603 &ctx
->xin
->flow
, ODPP_NONE
);
1604 ctx
->sflow_odp_port
= 0;
1605 ctx
->sflow_n_outputs
= 0;
1608 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1609 * of actions, eventually after the SAMPLE action for sFlow. */
1611 add_ipfix_action(struct xlate_ctx
*ctx
)
1613 compose_ipfix_action(ctx
->xbridge
, &ctx
->xout
->odp_actions
,
1617 /* Fix SAMPLE action according to data collected while composing ODP actions.
1618 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1619 * USERSPACE action's user-cookie which is required for sflow. */
1621 fix_sflow_action(struct xlate_ctx
*ctx
)
1623 const struct flow
*base
= &ctx
->base_flow
;
1624 union user_action_cookie
*cookie
;
1626 if (!ctx
->user_cookie_offset
) {
1630 cookie
= ofpbuf_at(&ctx
->xout
->odp_actions
, ctx
->user_cookie_offset
,
1631 sizeof cookie
->sflow
);
1632 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
1634 compose_sflow_cookie(ctx
->xbridge
, base
->vlan_tci
,
1635 ctx
->sflow_odp_port
, ctx
->sflow_n_outputs
, cookie
);
1638 static enum slow_path_reason
1639 process_special(struct xlate_ctx
*ctx
, const struct flow
*flow
,
1640 const struct xport
*xport
, const struct ofpbuf
*packet
)
1642 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1643 const struct xbridge
*xbridge
= ctx
->xbridge
;
1647 } else if (xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
)) {
1649 cfm_process_heartbeat(xport
->cfm
, packet
);
1652 } else if (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
, wc
)) {
1654 bfd_process_packet(xport
->bfd
, flow
, packet
);
1655 /* If POLL received, immediately sends FINAL back. */
1656 if (bfd_should_send_packet(xport
->bfd
)) {
1658 ofproto_dpif_monitor_port_send_soon(xport
->ofport
);
1660 ofproto_dpif_monitor_port_send_soon_safe(xport
->ofport
);
1665 } else if (xport
->xbundle
&& xport
->xbundle
->lacp
1666 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
1668 lacp_process_packet(xport
->xbundle
->lacp
, xport
->ofport
, packet
);
1671 } else if (xbridge
->stp
&& stp_should_process_flow(flow
, wc
)) {
1673 stp_process_packet(xport
, packet
);
1682 compose_output_action__(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
1685 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
1686 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
1687 struct flow
*flow
= &ctx
->xin
->flow
;
1688 ovs_be16 flow_vlan_tci
;
1689 uint32_t flow_pkt_mark
;
1690 uint8_t flow_nw_tos
;
1691 odp_port_t out_port
, odp_port
;
1694 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1695 * before traversing a patch port. */
1696 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 25);
1699 xlate_report(ctx
, "Nonexistent output port");
1701 } else if (xport
->config
& OFPUTIL_PC_NO_FWD
) {
1702 xlate_report(ctx
, "OFPPC_NO_FWD set, skipping output");
1704 } else if (check_stp
) {
1705 if (eth_addr_equals(ctx
->base_flow
.dl_dst
, eth_addr_stp
)) {
1706 if (!xport_stp_listen_state(xport
)) {
1707 xlate_report(ctx
, "STP not in listening state, "
1708 "skipping bpdu output");
1711 } else if (!xport_stp_forward_state(xport
)) {
1712 xlate_report(ctx
, "STP not in forwarding state, "
1718 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
) && xport
->xbundle
) {
1719 ctx
->xout
->mirrors
|= xbundle_mirror_dst(xport
->xbundle
->xbridge
,
1724 const struct xport
*peer
= xport
->peer
;
1725 struct flow old_flow
= ctx
->xin
->flow
;
1726 enum slow_path_reason special
;
1728 ctx
->xbridge
= peer
->xbridge
;
1729 flow
->in_port
.ofp_port
= peer
->ofp_port
;
1730 flow
->metadata
= htonll(0);
1731 memset(&flow
->tunnel
, 0, sizeof flow
->tunnel
);
1732 memset(flow
->regs
, 0, sizeof flow
->regs
);
1734 special
= process_special(ctx
, &ctx
->xin
->flow
, peer
,
1737 ctx
->xout
->slow
|= special
;
1738 } else if (may_receive(peer
, ctx
)) {
1739 if (xport_stp_forward_state(peer
)) {
1740 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true);
1742 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1743 * learning action look at the packet, then drop it. */
1744 struct flow old_base_flow
= ctx
->base_flow
;
1745 size_t old_size
= ctx
->xout
->odp_actions
.size
;
1746 mirror_mask_t old_mirrors
= ctx
->xout
->mirrors
;
1747 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true);
1748 ctx
->xout
->mirrors
= old_mirrors
;
1749 ctx
->base_flow
= old_base_flow
;
1750 ctx
->xout
->odp_actions
.size
= old_size
;
1754 ctx
->xin
->flow
= old_flow
;
1755 ctx
->xbridge
= xport
->xbridge
;
1757 if (ctx
->xin
->resubmit_stats
) {
1758 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
1759 netdev_vport_inc_rx(peer
->netdev
, ctx
->xin
->resubmit_stats
);
1761 bfd_account_rx(peer
->bfd
, ctx
->xin
->resubmit_stats
);
1768 flow_vlan_tci
= flow
->vlan_tci
;
1769 flow_pkt_mark
= flow
->pkt_mark
;
1770 flow_nw_tos
= flow
->nw_tos
;
1772 if (dscp_from_skb_priority(xport
, flow
->skb_priority
, &dscp
)) {
1773 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
1774 flow
->nw_tos
&= ~IP_DSCP_MASK
;
1775 flow
->nw_tos
|= dscp
;
1778 if (xport
->is_tunnel
) {
1779 /* Save tunnel metadata so that changes made due to
1780 * the Logical (tunnel) Port are not visible for any further
1781 * matches, while explicit set actions on tunnel metadata are.
1783 struct flow_tnl flow_tnl
= flow
->tunnel
;
1784 odp_port
= tnl_port_send(xport
->ofport
, flow
, &ctx
->xout
->wc
);
1785 if (odp_port
== ODPP_NONE
) {
1786 xlate_report(ctx
, "Tunneling decided against output");
1787 goto out
; /* restore flow_nw_tos */
1789 if (flow
->tunnel
.ip_dst
== ctx
->orig_tunnel_ip_dst
) {
1790 xlate_report(ctx
, "Not tunneling to our own address");
1791 goto out
; /* restore flow_nw_tos */
1793 if (ctx
->xin
->resubmit_stats
) {
1794 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
1796 out_port
= odp_port
;
1797 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
1798 &ctx
->xout
->odp_actions
);
1799 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
1801 odp_port
= xport
->odp_port
;
1802 out_port
= odp_port
;
1803 if (ofproto_has_vlan_splinters(ctx
->xbridge
->ofproto
)) {
1804 ofp_port_t vlandev_port
;
1806 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
1807 vlandev_port
= vsp_realdev_to_vlandev(ctx
->xbridge
->ofproto
,
1808 ofp_port
, flow
->vlan_tci
);
1809 if (vlandev_port
!= ofp_port
) {
1810 out_port
= ofp_port_to_odp_port(ctx
->xbridge
, vlandev_port
);
1811 flow
->vlan_tci
= htons(0);
1816 if (out_port
!= ODPP_NONE
) {
1817 ctx
->xout
->slow
|= commit_odp_actions(flow
, &ctx
->base_flow
,
1818 &ctx
->xout
->odp_actions
,
1820 nl_msg_put_odp_port(&ctx
->xout
->odp_actions
, OVS_ACTION_ATTR_OUTPUT
,
1823 ctx
->sflow_odp_port
= odp_port
;
1824 ctx
->sflow_n_outputs
++;
1825 ctx
->xout
->nf_output_iface
= ofp_port
;
1830 flow
->vlan_tci
= flow_vlan_tci
;
1831 flow
->pkt_mark
= flow_pkt_mark
;
1832 flow
->nw_tos
= flow_nw_tos
;
1836 compose_output_action(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
1838 compose_output_action__(ctx
, ofp_port
, true);
1842 xlate_recursively(struct xlate_ctx
*ctx
, struct rule_dpif
*rule
)
1844 struct rule_dpif
*old_rule
= ctx
->rule
;
1845 struct rule_actions
*actions
;
1847 if (ctx
->xin
->resubmit_stats
) {
1848 rule_dpif_credit_stats(rule
, ctx
->xin
->resubmit_stats
);
1854 actions
= rule_dpif_get_actions(rule
);
1855 do_xlate_actions(actions
->ofpacts
, actions
->ofpacts_len
, ctx
);
1856 ctx
->rule
= old_rule
;
1861 xlate_resubmit_resource_check(struct xlate_ctx
*ctx
)
1863 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
1865 if (ctx
->recurse
>= MAX_RESUBMIT_RECURSION
) {
1866 VLOG_ERR_RL(&rl
, "resubmit actions recursed over %d times",
1867 MAX_RESUBMIT_RECURSION
);
1868 } else if (ctx
->resubmits
>= MAX_RESUBMITS
) {
1869 VLOG_ERR_RL(&rl
, "over %d resubmit actions", MAX_RESUBMITS
);
1870 } else if (ctx
->xout
->odp_actions
.size
> UINT16_MAX
) {
1871 VLOG_ERR_RL(&rl
, "resubmits yielded over 64 kB of actions");
1872 } else if (ctx
->stack
.size
>= 65536) {
1873 VLOG_ERR_RL(&rl
, "resubmits yielded over 64 kB of stack");
1882 xlate_table_action(struct xlate_ctx
*ctx
, ofp_port_t in_port
, uint8_t table_id
,
1883 bool may_packet_in
, bool honor_table_miss
)
1885 if (xlate_resubmit_resource_check(ctx
)) {
1886 ofp_port_t old_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
1887 bool skip_wildcards
= ctx
->xin
->skip_wildcards
;
1888 uint8_t old_table_id
= ctx
->table_id
;
1889 struct rule_dpif
*rule
;
1890 enum rule_dpif_lookup_verdict verdict
;
1891 enum ofputil_port_config config
= 0;
1893 ctx
->table_id
= table_id
;
1895 /* Look up a flow with 'in_port' as the input port. Then restore the
1896 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1897 * have surprising behavior). */
1898 ctx
->xin
->flow
.in_port
.ofp_port
= in_port
;
1899 verdict
= rule_dpif_lookup_from_table(ctx
->xbridge
->ofproto
,
1902 ? &ctx
->xout
->wc
: NULL
,
1904 &ctx
->table_id
, &rule
);
1905 ctx
->xin
->flow
.in_port
.ofp_port
= old_in_port
;
1907 if (ctx
->xin
->resubmit_hook
) {
1908 ctx
->xin
->resubmit_hook(ctx
->xin
, rule
, ctx
->recurse
);
1912 case RULE_DPIF_LOOKUP_VERDICT_MATCH
:
1914 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
:
1915 if (may_packet_in
) {
1916 struct xport
*xport
;
1918 xport
= get_ofp_port(ctx
->xbridge
,
1919 ctx
->xin
->flow
.in_port
.ofp_port
);
1920 config
= xport
? xport
->config
: 0;
1923 /* Fall through to drop */
1924 case RULE_DPIF_LOOKUP_VERDICT_DROP
:
1925 config
= OFPUTIL_PC_NO_PACKET_IN
;
1931 choose_miss_rule(config
, ctx
->xbridge
->miss_rule
,
1932 ctx
->xbridge
->no_packet_in_rule
, &rule
);
1936 xlate_recursively(ctx
, rule
);
1937 rule_dpif_unref(rule
);
1940 ctx
->table_id
= old_table_id
;
1948 xlate_group_bucket(struct xlate_ctx
*ctx
, const struct ofputil_bucket
*bucket
)
1950 uint64_t action_list_stub
[1024 / 8];
1951 struct ofpbuf action_list
, action_set
;
1953 ofpbuf_use_const(&action_set
, bucket
->ofpacts
, bucket
->ofpacts_len
);
1954 ofpbuf_use_stub(&action_list
, action_list_stub
, sizeof action_list_stub
);
1956 ofpacts_execute_action_set(&action_list
, &action_set
);
1958 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
);
1961 ofpbuf_uninit(&action_set
);
1962 ofpbuf_uninit(&action_list
);
1966 xlate_all_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
1968 const struct ofputil_bucket
*bucket
;
1969 const struct list
*buckets
;
1970 struct flow old_flow
= ctx
->xin
->flow
;
1972 group_dpif_get_buckets(group
, &buckets
);
1974 LIST_FOR_EACH (bucket
, list_node
, buckets
) {
1975 xlate_group_bucket(ctx
, bucket
);
1976 /* Roll back flow to previous state.
1977 * This is equivalent to cloning the packet for each bucket.
1979 * As a side effect any subsequently applied actions will
1980 * also effectively be applied to a clone of the packet taken
1981 * just before applying the all or indirect group. */
1982 ctx
->xin
->flow
= old_flow
;
1987 xlate_ff_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
1989 const struct ofputil_bucket
*bucket
;
1991 bucket
= group_first_live_bucket(ctx
, group
, 0);
1993 xlate_group_bucket(ctx
, bucket
);
1998 xlate_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
2000 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2001 const struct ofputil_bucket
*bucket
;
2004 basis
= hash_bytes(ctx
->xin
->flow
.dl_dst
, sizeof ctx
->xin
->flow
.dl_dst
, 0);
2005 bucket
= group_best_live_bucket(ctx
, group
, basis
);
2007 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2008 xlate_group_bucket(ctx
, bucket
);
2013 xlate_group_action__(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
2015 ctx
->in_group
= true;
2017 switch (group_dpif_get_type(group
)) {
2019 case OFPGT11_INDIRECT
:
2020 xlate_all_group(ctx
, group
);
2022 case OFPGT11_SELECT
:
2023 xlate_select_group(ctx
, group
);
2026 xlate_ff_group(ctx
, group
);
2031 group_dpif_release(group
);
2033 ctx
->in_group
= false;
2037 xlate_group_resource_check(struct xlate_ctx
*ctx
)
2039 if (!xlate_resubmit_resource_check(ctx
)) {
2041 } else if (ctx
->in_group
) {
2042 /* Prevent nested translation of OpenFlow groups.
2044 * OpenFlow allows this restriction. We enforce this restriction only
2045 * because, with the current architecture, we would otherwise have to
2046 * take a possibly recursive read lock on the ofgroup rwlock, which is
2047 * unsafe given that POSIX allows taking a read lock to block if there
2048 * is a thread blocked on taking the write lock. Other solutions
2049 * without this restriction are also possible, but seem unwarranted
2050 * given the current limited use of groups. */
2051 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2053 VLOG_ERR_RL(&rl
, "cannot recursively translate OpenFlow group");
2061 xlate_group_action(struct xlate_ctx
*ctx
, uint32_t group_id
)
2063 if (xlate_group_resource_check(ctx
)) {
2064 struct group_dpif
*group
;
2067 got_group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
, &group
);
2069 xlate_group_action__(ctx
, group
);
2079 xlate_ofpact_resubmit(struct xlate_ctx
*ctx
,
2080 const struct ofpact_resubmit
*resubmit
)
2085 in_port
= resubmit
->in_port
;
2086 if (in_port
== OFPP_IN_PORT
) {
2087 in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
2090 table_id
= resubmit
->table_id
;
2091 if (table_id
== 255) {
2092 table_id
= ctx
->table_id
;
2095 xlate_table_action(ctx
, in_port
, table_id
, false, false);
2099 flood_packets(struct xlate_ctx
*ctx
, bool all
)
2101 const struct xport
*xport
;
2103 HMAP_FOR_EACH (xport
, ofp_node
, &ctx
->xbridge
->xports
) {
2104 if (xport
->ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
2109 compose_output_action__(ctx
, xport
->ofp_port
, false);
2110 } else if (!(xport
->config
& OFPUTIL_PC_NO_FLOOD
)) {
2111 compose_output_action(ctx
, xport
->ofp_port
);
2115 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
2119 execute_controller_action(struct xlate_ctx
*ctx
, int len
,
2120 enum ofp_packet_in_reason reason
,
2121 uint16_t controller_id
)
2123 struct ofproto_packet_in
*pin
;
2124 struct ofpbuf
*packet
;
2125 struct pkt_metadata md
= PKT_METADATA_INITIALIZER(0);
2127 ctx
->xout
->slow
|= SLOW_CONTROLLER
;
2128 if (!ctx
->xin
->packet
) {
2132 packet
= ofpbuf_clone(ctx
->xin
->packet
);
2134 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
2135 &ctx
->xout
->odp_actions
,
2138 odp_execute_actions(NULL
, packet
, false, &md
, ctx
->xout
->odp_actions
.data
,
2139 ctx
->xout
->odp_actions
.size
, NULL
);
2141 pin
= xmalloc(sizeof *pin
);
2142 pin
->up
.packet_len
= packet
->size
;
2143 pin
->up
.packet
= ofpbuf_steal_data(packet
);
2144 pin
->up
.reason
= reason
;
2145 pin
->up
.table_id
= ctx
->table_id
;
2146 pin
->up
.cookie
= (ctx
->rule
2147 ? rule_dpif_get_flow_cookie(ctx
->rule
)
2150 flow_get_metadata(&ctx
->xin
->flow
, &pin
->up
.fmd
);
2152 pin
->controller_id
= controller_id
;
2153 pin
->send_len
= len
;
2154 /* If a rule is a table-miss rule then this is
2155 * a table-miss handled by a table-miss rule.
2157 * Else, if rule is internal and has a controller action,
2158 * the later being implied by the rule being processed here,
2159 * then this is a table-miss handled without a table-miss rule.
2161 * Otherwise this is not a table-miss. */
2162 pin
->miss_type
= OFPROTO_PACKET_IN_NO_MISS
;
2164 if (rule_dpif_is_table_miss(ctx
->rule
)) {
2165 pin
->miss_type
= OFPROTO_PACKET_IN_MISS_FLOW
;
2166 } else if (rule_dpif_is_internal(ctx
->rule
)) {
2167 pin
->miss_type
= OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW
;
2170 ofproto_dpif_send_packet_in(ctx
->xbridge
->ofproto
, pin
);
2171 ofpbuf_delete(packet
);
2175 compose_mpls_push_action(struct xlate_ctx
*ctx
, struct ofpact_push_mpls
*mpls
)
2177 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2178 struct flow
*flow
= &ctx
->xin
->flow
;
2181 ovs_assert(eth_type_mpls(mpls
->ethertype
));
2183 n
= flow_count_mpls_labels(flow
, wc
);
2185 ctx
->xout
->slow
|= commit_odp_actions(flow
, &ctx
->base_flow
,
2186 &ctx
->xout
->odp_actions
,
2188 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
2189 if (ctx
->xin
->packet
!= NULL
) {
2190 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2191 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet on which an "
2192 "MPLS push action can't be performed as it would "
2193 "have more MPLS LSEs than the %d supported.",
2194 ctx
->xbridge
->name
, FLOW_MAX_MPLS_LABELS
);
2198 } else if (n
>= ctx
->xbridge
->max_mpls_depth
) {
2199 COVERAGE_INC(xlate_actions_mpls_overflow
);
2200 ctx
->xout
->slow
|= SLOW_ACTION
;
2203 flow_push_mpls(flow
, n
, mpls
->ethertype
, wc
);
2207 compose_mpls_pop_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
2209 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2210 struct flow
*flow
= &ctx
->xin
->flow
;
2211 int n
= flow_count_mpls_labels(flow
, wc
);
2213 if (!flow_pop_mpls(flow
, n
, eth_type
, wc
) && n
>= FLOW_MAX_MPLS_LABELS
) {
2214 if (ctx
->xin
->packet
!= NULL
) {
2215 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2216 VLOG_WARN_RL(&rl
, "bridge %s: dropping packet on which an "
2217 "MPLS pop action can't be performed as it has "
2218 "more MPLS LSEs than the %d supported.",
2219 ctx
->xbridge
->name
, FLOW_MAX_MPLS_LABELS
);
2222 ofpbuf_clear(&ctx
->xout
->odp_actions
);
2227 compose_dec_ttl(struct xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
2229 struct flow
*flow
= &ctx
->xin
->flow
;
2231 if (!is_ip_any(flow
)) {
2235 ctx
->xout
->wc
.masks
.nw_ttl
= 0xff;
2236 if (flow
->nw_ttl
> 1) {
2242 for (i
= 0; i
< ids
->n_controllers
; i
++) {
2243 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
2247 /* Stop processing for current table. */
2253 compose_set_mpls_label_action(struct xlate_ctx
*ctx
, ovs_be32 label
)
2255 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2256 ctx
->xout
->wc
.masks
.mpls_lse
[0] |= htonl(MPLS_LABEL_MASK
);
2257 set_mpls_lse_label(&ctx
->xin
->flow
.mpls_lse
[0], label
);
2262 compose_set_mpls_tc_action(struct xlate_ctx
*ctx
, uint8_t tc
)
2264 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2265 ctx
->xout
->wc
.masks
.mpls_lse
[0] |= htonl(MPLS_TC_MASK
);
2266 set_mpls_lse_tc(&ctx
->xin
->flow
.mpls_lse
[0], tc
);
2271 compose_set_mpls_ttl_action(struct xlate_ctx
*ctx
, uint8_t ttl
)
2273 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
2274 ctx
->xout
->wc
.masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
2275 set_mpls_lse_ttl(&ctx
->xin
->flow
.mpls_lse
[0], ttl
);
2280 compose_dec_mpls_ttl_action(struct xlate_ctx
*ctx
)
2282 struct flow
*flow
= &ctx
->xin
->flow
;
2283 uint8_t ttl
= mpls_lse_to_ttl(flow
->mpls_lse
[0]);
2284 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2286 memset(&wc
->masks
.mpls_lse
, 0xff, sizeof wc
->masks
.mpls_lse
);
2287 if (eth_type_mpls(flow
->dl_type
)) {
2290 set_mpls_lse_ttl(&flow
->mpls_lse
[0], ttl
);
2293 execute_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0);
2295 /* Stop processing for current table. */
2304 xlate_output_action(struct xlate_ctx
*ctx
,
2305 ofp_port_t port
, uint16_t max_len
, bool may_packet_in
)
2307 ofp_port_t prev_nf_output_iface
= ctx
->xout
->nf_output_iface
;
2309 ctx
->xout
->nf_output_iface
= NF_OUT_DROP
;
2313 compose_output_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
);
2316 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
2317 0, may_packet_in
, true);
2323 flood_packets(ctx
, false);
2326 flood_packets(ctx
, true);
2328 case OFPP_CONTROLLER
:
2329 execute_controller_action(ctx
, max_len
, OFPR_ACTION
, 0);
2335 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
2336 compose_output_action(ctx
, port
);
2338 xlate_report(ctx
, "skipping output to input port");
2343 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2344 ctx
->xout
->nf_output_iface
= NF_OUT_FLOOD
;
2345 } else if (ctx
->xout
->nf_output_iface
== NF_OUT_DROP
) {
2346 ctx
->xout
->nf_output_iface
= prev_nf_output_iface
;
2347 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2348 ctx
->xout
->nf_output_iface
!= NF_OUT_FLOOD
) {
2349 ctx
->xout
->nf_output_iface
= NF_OUT_MULTI
;
2354 xlate_output_reg_action(struct xlate_ctx
*ctx
,
2355 const struct ofpact_output_reg
*or)
2357 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->xin
->flow
);
2358 if (port
<= UINT16_MAX
) {
2359 union mf_subvalue value
;
2361 memset(&value
, 0xff, sizeof value
);
2362 mf_write_subfield_flow(&or->src
, &value
, &ctx
->xout
->wc
.masks
);
2363 xlate_output_action(ctx
, u16_to_ofp(port
),
2364 or->max_len
, false);
2369 xlate_enqueue_action(struct xlate_ctx
*ctx
,
2370 const struct ofpact_enqueue
*enqueue
)
2372 ofp_port_t ofp_port
= enqueue
->port
;
2373 uint32_t queue_id
= enqueue
->queue
;
2374 uint32_t flow_priority
, priority
;
2377 /* Translate queue to priority. */
2378 error
= dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &priority
);
2380 /* Fall back to ordinary output action. */
2381 xlate_output_action(ctx
, enqueue
->port
, 0, false);
2385 /* Check output port. */
2386 if (ofp_port
== OFPP_IN_PORT
) {
2387 ofp_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
2388 } else if (ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
2392 /* Add datapath actions. */
2393 flow_priority
= ctx
->xin
->flow
.skb_priority
;
2394 ctx
->xin
->flow
.skb_priority
= priority
;
2395 compose_output_action(ctx
, ofp_port
);
2396 ctx
->xin
->flow
.skb_priority
= flow_priority
;
2398 /* Update NetFlow output port. */
2399 if (ctx
->xout
->nf_output_iface
== NF_OUT_DROP
) {
2400 ctx
->xout
->nf_output_iface
= ofp_port
;
2401 } else if (ctx
->xout
->nf_output_iface
!= NF_OUT_FLOOD
) {
2402 ctx
->xout
->nf_output_iface
= NF_OUT_MULTI
;
2407 xlate_set_queue_action(struct xlate_ctx
*ctx
, uint32_t queue_id
)
2409 uint32_t skb_priority
;
2411 if (!dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &skb_priority
)) {
2412 ctx
->xin
->flow
.skb_priority
= skb_priority
;
2414 /* Couldn't translate queue to a priority. Nothing to do. A warning
2415 * has already been logged. */
2420 slave_enabled_cb(ofp_port_t ofp_port
, void *xbridge_
)
2422 const struct xbridge
*xbridge
= xbridge_
;
2433 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
2436 port
= get_ofp_port(xbridge
, ofp_port
);
2437 return port
? port
->may_enable
: false;
2442 xlate_bundle_action(struct xlate_ctx
*ctx
,
2443 const struct ofpact_bundle
*bundle
)
2447 port
= bundle_execute(bundle
, &ctx
->xin
->flow
, &ctx
->xout
->wc
,
2449 CONST_CAST(struct xbridge
*, ctx
->xbridge
));
2450 if (bundle
->dst
.field
) {
2451 nxm_reg_load(&bundle
->dst
, ofp_to_u16(port
), &ctx
->xin
->flow
,
2454 xlate_output_action(ctx
, port
, 0, false);
2459 xlate_learn_action(struct xlate_ctx
*ctx
,
2460 const struct ofpact_learn
*learn
)
2462 uint64_t ofpacts_stub
[1024 / 8];
2463 struct ofputil_flow_mod fm
;
2464 struct ofpbuf ofpacts
;
2466 ctx
->xout
->has_learn
= true;
2468 learn_mask(learn
, &ctx
->xout
->wc
);
2470 if (!ctx
->xin
->may_learn
) {
2474 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
2475 learn_execute(learn
, &ctx
->xin
->flow
, &fm
, &ofpacts
);
2476 ofproto_dpif_flow_mod(ctx
->xbridge
->ofproto
, &fm
);
2477 ofpbuf_uninit(&ofpacts
);
2481 xlate_fin_timeout(struct xlate_ctx
*ctx
,
2482 const struct ofpact_fin_timeout
*oft
)
2484 if (ctx
->xin
->tcp_flags
& (TCP_FIN
| TCP_RST
) && ctx
->rule
) {
2485 rule_dpif_reduce_timeouts(ctx
->rule
, oft
->fin_idle_timeout
,
2486 oft
->fin_hard_timeout
);
2491 xlate_sample_action(struct xlate_ctx
*ctx
,
2492 const struct ofpact_sample
*os
)
2494 union user_action_cookie cookie
;
2495 /* Scale the probability from 16-bit to 32-bit while representing
2496 * the same percentage. */
2497 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
2499 if (!ctx
->xbridge
->variable_length_userdata
) {
2500 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2502 VLOG_ERR_RL(&rl
, "ignoring NXAST_SAMPLE action because datapath "
2503 "lacks support (needs Linux 3.10+ or kernel module from "
2508 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
2509 &ctx
->xout
->odp_actions
,
2512 compose_flow_sample_cookie(os
->probability
, os
->collector_set_id
,
2513 os
->obs_domain_id
, os
->obs_point_id
, &cookie
);
2514 compose_sample_action(ctx
->xbridge
, &ctx
->xout
->odp_actions
, &ctx
->xin
->flow
,
2515 probability
, &cookie
, sizeof cookie
.flow_sample
);
2519 may_receive(const struct xport
*xport
, struct xlate_ctx
*ctx
)
2521 if (xport
->config
& (eth_addr_equals(ctx
->xin
->flow
.dl_dst
, eth_addr_stp
)
2522 ? OFPUTIL_PC_NO_RECV_STP
2523 : OFPUTIL_PC_NO_RECV
)) {
2527 /* Only drop packets here if both forwarding and learning are
2528 * disabled. If just learning is enabled, we need to have
2529 * OFPP_NORMAL and the learning action have a look at the packet
2530 * before we can drop it. */
2531 if (!xport_stp_forward_state(xport
) && !xport_stp_learn_state(xport
)) {
2539 xlate_write_actions(struct xlate_ctx
*ctx
, const struct ofpact
*a
)
2541 struct ofpact_nest
*on
= ofpact_get_WRITE_ACTIONS(a
);
2542 ofpbuf_put(&ctx
->action_set
, on
->actions
, ofpact_nest_get_action_len(on
));
2543 ofpact_pad(&ctx
->action_set
);
2547 xlate_action_set(struct xlate_ctx
*ctx
)
2549 uint64_t action_list_stub
[1024 / 64];
2550 struct ofpbuf action_list
;
2552 ofpbuf_use_stub(&action_list
, action_list_stub
, sizeof action_list_stub
);
2553 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
2554 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
);
2555 ofpbuf_uninit(&action_list
);
2559 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
2560 struct xlate_ctx
*ctx
)
2562 struct flow_wildcards
*wc
= &ctx
->xout
->wc
;
2563 struct flow
*flow
= &ctx
->xin
->flow
;
2564 const struct ofpact
*a
;
2566 /* dl_type already in the mask, not set below. */
2568 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
2569 struct ofpact_controller
*controller
;
2570 const struct ofpact_metadata
*metadata
;
2571 const struct ofpact_set_field
*set_field
;
2572 const struct mf_field
*mf
;
2580 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
2581 ofpact_get_OUTPUT(a
)->max_len
, true);
2585 if (xlate_group_action(ctx
, ofpact_get_GROUP(a
)->group_id
)) {
2590 case OFPACT_CONTROLLER
:
2591 controller
= ofpact_get_CONTROLLER(a
);
2592 execute_controller_action(ctx
, controller
->max_len
,
2594 controller
->controller_id
);
2597 case OFPACT_ENQUEUE
:
2598 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
));
2601 case OFPACT_SET_VLAN_VID
:
2602 wc
->masks
.vlan_tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
2603 if (flow
->vlan_tci
& htons(VLAN_CFI
) ||
2604 ofpact_get_SET_VLAN_VID(a
)->push_vlan_if_needed
) {
2605 flow
->vlan_tci
&= ~htons(VLAN_VID_MASK
);
2606 flow
->vlan_tci
|= (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
)
2611 case OFPACT_SET_VLAN_PCP
:
2612 wc
->masks
.vlan_tci
|= htons(VLAN_PCP_MASK
| VLAN_CFI
);
2613 if (flow
->vlan_tci
& htons(VLAN_CFI
) ||
2614 ofpact_get_SET_VLAN_PCP(a
)->push_vlan_if_needed
) {
2615 flow
->vlan_tci
&= ~htons(VLAN_PCP_MASK
);
2616 flow
->vlan_tci
|= htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
2617 << VLAN_PCP_SHIFT
) | VLAN_CFI
);
2621 case OFPACT_STRIP_VLAN
:
2622 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
2623 flow
->vlan_tci
= htons(0);
2626 case OFPACT_PUSH_VLAN
:
2627 /* XXX 802.1AD(QinQ) */
2628 memset(&wc
->masks
.vlan_tci
, 0xff, sizeof wc
->masks
.vlan_tci
);
2629 flow
->vlan_tci
= htons(VLAN_CFI
);
2632 case OFPACT_SET_ETH_SRC
:
2633 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
2634 memcpy(flow
->dl_src
, ofpact_get_SET_ETH_SRC(a
)->mac
, ETH_ADDR_LEN
);
2637 case OFPACT_SET_ETH_DST
:
2638 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2639 memcpy(flow
->dl_dst
, ofpact_get_SET_ETH_DST(a
)->mac
, ETH_ADDR_LEN
);
2642 case OFPACT_SET_IPV4_SRC
:
2643 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2644 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
2645 flow
->nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
2649 case OFPACT_SET_IPV4_DST
:
2650 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2651 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2652 flow
->nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
2656 case OFPACT_SET_IP_DSCP
:
2657 if (is_ip_any(flow
)) {
2658 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
2659 flow
->nw_tos
&= ~IP_DSCP_MASK
;
2660 flow
->nw_tos
|= ofpact_get_SET_IP_DSCP(a
)->dscp
;
2664 case OFPACT_SET_IP_ECN
:
2665 if (is_ip_any(flow
)) {
2666 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
2667 flow
->nw_tos
&= ~IP_ECN_MASK
;
2668 flow
->nw_tos
|= ofpact_get_SET_IP_ECN(a
)->ecn
;
2672 case OFPACT_SET_IP_TTL
:
2673 if (is_ip_any(flow
)) {
2674 wc
->masks
.nw_ttl
= 0xff;
2675 flow
->nw_ttl
= ofpact_get_SET_IP_TTL(a
)->ttl
;
2679 case OFPACT_SET_L4_SRC_PORT
:
2680 if (is_ip_any(flow
)) {
2681 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2682 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
2683 flow
->tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
2687 case OFPACT_SET_L4_DST_PORT
:
2688 if (is_ip_any(flow
)) {
2689 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2690 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
2691 flow
->tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
2695 case OFPACT_RESUBMIT
:
2696 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
));
2699 case OFPACT_SET_TUNNEL
:
2700 flow
->tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
2703 case OFPACT_SET_QUEUE
:
2704 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
2707 case OFPACT_POP_QUEUE
:
2708 flow
->skb_priority
= ctx
->orig_skb_priority
;
2711 case OFPACT_REG_MOVE
:
2712 nxm_execute_reg_move(ofpact_get_REG_MOVE(a
), flow
, wc
);
2715 case OFPACT_REG_LOAD
:
2716 nxm_execute_reg_load(ofpact_get_REG_LOAD(a
), flow
, wc
);
2719 case OFPACT_SET_FIELD
:
2720 set_field
= ofpact_get_SET_FIELD(a
);
2721 mf
= set_field
->field
;
2722 mf_mask_field_and_prereqs(mf
, &wc
->masks
);
2724 /* Set field action only ever overwrites packet's outermost
2725 * applicable header fields. Do nothing if no header exists. */
2726 if ((mf
->id
!= MFF_VLAN_VID
|| flow
->vlan_tci
& htons(VLAN_CFI
))
2727 && ((mf
->id
!= MFF_MPLS_LABEL
&& mf
->id
!= MFF_MPLS_TC
)
2728 || eth_type_mpls(flow
->dl_type
))) {
2729 mf_set_flow_value(mf
, &set_field
->value
, flow
);
2733 case OFPACT_STACK_PUSH
:
2734 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), flow
, wc
,
2738 case OFPACT_STACK_POP
:
2739 nxm_execute_stack_pop(ofpact_get_STACK_POP(a
), flow
, wc
,
2743 case OFPACT_PUSH_MPLS
:
2744 compose_mpls_push_action(ctx
, ofpact_get_PUSH_MPLS(a
));
2747 case OFPACT_POP_MPLS
:
2748 compose_mpls_pop_action(ctx
, ofpact_get_POP_MPLS(a
)->ethertype
);
2751 case OFPACT_SET_MPLS_LABEL
:
2752 compose_set_mpls_label_action(
2753 ctx
, ofpact_get_SET_MPLS_LABEL(a
)->label
);
2756 case OFPACT_SET_MPLS_TC
:
2757 compose_set_mpls_tc_action(ctx
, ofpact_get_SET_MPLS_TC(a
)->tc
);
2760 case OFPACT_SET_MPLS_TTL
:
2761 compose_set_mpls_ttl_action(ctx
, ofpact_get_SET_MPLS_TTL(a
)->ttl
);
2764 case OFPACT_DEC_MPLS_TTL
:
2765 if (compose_dec_mpls_ttl_action(ctx
)) {
2770 case OFPACT_DEC_TTL
:
2771 wc
->masks
.nw_ttl
= 0xff;
2772 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
2778 /* Nothing to do. */
2781 case OFPACT_MULTIPATH
:
2782 multipath_execute(ofpact_get_MULTIPATH(a
), flow
, wc
);
2786 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
));
2789 case OFPACT_OUTPUT_REG
:
2790 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
));
2794 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
2801 case OFPACT_FIN_TIMEOUT
:
2802 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2803 ctx
->xout
->has_fin_timeout
= true;
2804 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
2807 case OFPACT_CLEAR_ACTIONS
:
2808 ofpbuf_clear(&ctx
->action_set
);
2811 case OFPACT_WRITE_ACTIONS
:
2812 xlate_write_actions(ctx
, a
);
2815 case OFPACT_WRITE_METADATA
:
2816 metadata
= ofpact_get_WRITE_METADATA(a
);
2817 flow
->metadata
&= ~metadata
->mask
;
2818 flow
->metadata
|= metadata
->metadata
& metadata
->mask
;
2822 /* Not implemented yet. */
2825 case OFPACT_GOTO_TABLE
: {
2826 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
2828 ovs_assert(ctx
->table_id
< ogt
->table_id
);
2829 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
2830 ogt
->table_id
, true, true);
2835 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
2842 xlate_in_init(struct xlate_in
*xin
, struct ofproto_dpif
*ofproto
,
2843 const struct flow
*flow
, struct rule_dpif
*rule
,
2844 uint16_t tcp_flags
, const struct ofpbuf
*packet
)
2846 xin
->ofproto
= ofproto
;
2848 xin
->packet
= packet
;
2849 xin
->may_learn
= packet
!= NULL
;
2851 xin
->ofpacts
= NULL
;
2852 xin
->ofpacts_len
= 0;
2853 xin
->tcp_flags
= tcp_flags
;
2854 xin
->resubmit_hook
= NULL
;
2855 xin
->report_hook
= NULL
;
2856 xin
->resubmit_stats
= NULL
;
2857 xin
->skip_wildcards
= false;
2861 xlate_out_uninit(struct xlate_out
*xout
)
2864 ofpbuf_uninit(&xout
->odp_actions
);
2868 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2869 * into datapath actions, using 'ctx', and discards the datapath actions. */
2871 xlate_actions_for_side_effects(struct xlate_in
*xin
)
2873 struct xlate_out xout
;
2875 xlate_actions(xin
, &xout
);
2876 xlate_out_uninit(&xout
);
2880 xlate_report(struct xlate_ctx
*ctx
, const char *s
)
2882 if (ctx
->xin
->report_hook
) {
2883 ctx
->xin
->report_hook(ctx
->xin
, s
, ctx
->recurse
);
2888 xlate_out_copy(struct xlate_out
*dst
, const struct xlate_out
*src
)
2891 dst
->slow
= src
->slow
;
2892 dst
->has_learn
= src
->has_learn
;
2893 dst
->has_normal
= src
->has_normal
;
2894 dst
->has_fin_timeout
= src
->has_fin_timeout
;
2895 dst
->nf_output_iface
= src
->nf_output_iface
;
2896 dst
->mirrors
= src
->mirrors
;
2898 ofpbuf_use_stub(&dst
->odp_actions
, dst
->odp_actions_stub
,
2899 sizeof dst
->odp_actions_stub
);
2900 ofpbuf_put(&dst
->odp_actions
, src
->odp_actions
.data
,
2901 src
->odp_actions
.size
);
2904 static struct skb_priority_to_dscp
*
2905 get_skb_priority(const struct xport
*xport
, uint32_t skb_priority
)
2907 struct skb_priority_to_dscp
*pdscp
;
2910 hash
= hash_int(skb_priority
, 0);
2911 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &xport
->skb_priorities
) {
2912 if (pdscp
->skb_priority
== skb_priority
) {
2920 dscp_from_skb_priority(const struct xport
*xport
, uint32_t skb_priority
,
2923 struct skb_priority_to_dscp
*pdscp
= get_skb_priority(xport
, skb_priority
);
2924 *dscp
= pdscp
? pdscp
->dscp
: 0;
2925 return pdscp
!= NULL
;
2929 clear_skb_priorities(struct xport
*xport
)
2931 struct skb_priority_to_dscp
*pdscp
, *next
;
2933 HMAP_FOR_EACH_SAFE (pdscp
, next
, hmap_node
, &xport
->skb_priorities
) {
2934 hmap_remove(&xport
->skb_priorities
, &pdscp
->hmap_node
);
2940 actions_output_to_local_port(const struct xlate_ctx
*ctx
)
2942 odp_port_t local_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
, OFPP_LOCAL
);
2943 const struct nlattr
*a
;
2946 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, ctx
->xout
->odp_actions
.data
,
2947 ctx
->xout
->odp_actions
.size
) {
2948 if (nl_attr_type(a
) == OVS_ACTION_ATTR_OUTPUT
2949 && nl_attr_get_odp_port(a
) == local_odp_port
) {
2956 /* Thread safe call to xlate_actions__(). */
2958 xlate_actions(struct xlate_in
*xin
, struct xlate_out
*xout
)
2959 OVS_EXCLUDED(xlate_rwlock
)
2961 ovs_rwlock_rdlock(&xlate_rwlock
);
2962 xlate_actions__(xin
, xout
);
2963 ovs_rwlock_unlock(&xlate_rwlock
);
2966 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2967 * into datapath actions in 'odp_actions', using 'ctx'.
2969 * The caller must take responsibility for eventually freeing 'xout', with
2970 * xlate_out_uninit(). */
2972 xlate_actions__(struct xlate_in
*xin
, struct xlate_out
*xout
)
2973 OVS_REQ_RDLOCK(xlate_rwlock
)
2975 struct flow_wildcards
*wc
= &xout
->wc
;
2976 struct flow
*flow
= &xin
->flow
;
2977 struct rule_dpif
*rule
= NULL
;
2979 struct rule_actions
*actions
= NULL
;
2980 enum slow_path_reason special
;
2981 const struct ofpact
*ofpacts
;
2982 struct xport
*in_port
;
2983 struct flow orig_flow
;
2984 struct xlate_ctx ctx
;
2989 COVERAGE_INC(xlate_actions
);
2991 /* Flow initialization rules:
2992 * - 'base_flow' must match the kernel's view of the packet at the
2993 * time that action processing starts. 'flow' represents any
2994 * transformations we wish to make through actions.
2995 * - By default 'base_flow' and 'flow' are the same since the input
2996 * packet matches the output before any actions are applied.
2997 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2998 * of the received packet as seen by the kernel. If we later output
2999 * to another device without any modifications this will cause us to
3000 * insert a new tag since the original one was stripped off by the
3002 * - Tunnel metadata as received is retained in 'flow'. This allows
3003 * tunnel metadata matching also in later tables.
3004 * Since a kernel action for setting the tunnel metadata will only be
3005 * generated with actual tunnel output, changing the tunnel metadata
3006 * values in 'flow' (such as tun_id) will only have effect with a later
3007 * tunnel output action.
3008 * - Tunnel 'base_flow' is completely cleared since that is what the
3009 * kernel does. If we wish to maintain the original values an action
3010 * needs to be generated. */
3015 ctx
.xout
->has_learn
= false;
3016 ctx
.xout
->has_normal
= false;
3017 ctx
.xout
->has_fin_timeout
= false;
3018 ctx
.xout
->nf_output_iface
= NF_OUT_DROP
;
3019 ctx
.xout
->mirrors
= 0;
3020 ofpbuf_use_stub(&ctx
.xout
->odp_actions
, ctx
.xout
->odp_actions_stub
,
3021 sizeof ctx
.xout
->odp_actions_stub
);
3022 ofpbuf_reserve(&ctx
.xout
->odp_actions
, NL_A_U32_SIZE
);
3024 ctx
.xbridge
= xbridge_lookup(xin
->ofproto
);
3029 ctx
.rule
= xin
->rule
;
3031 ctx
.base_flow
= *flow
;
3032 memset(&ctx
.base_flow
.tunnel
, 0, sizeof ctx
.base_flow
.tunnel
);
3033 ctx
.orig_tunnel_ip_dst
= flow
->tunnel
.ip_dst
;
3035 flow_wildcards_init_catchall(wc
);
3036 memset(&wc
->masks
.in_port
, 0xff, sizeof wc
->masks
.in_port
);
3037 memset(&wc
->masks
.skb_priority
, 0xff, sizeof wc
->masks
.skb_priority
);
3038 memset(&wc
->masks
.dl_type
, 0xff, sizeof wc
->masks
.dl_type
);
3039 if (is_ip_any(flow
)) {
3040 wc
->masks
.nw_frag
|= FLOW_NW_FRAG_MASK
;
3042 is_icmp
= is_icmpv4(flow
) || is_icmpv6(flow
);
3044 tnl_may_send
= tnl_xlate_init(&ctx
.base_flow
, flow
, wc
);
3045 if (ctx
.xbridge
->netflow
) {
3046 netflow_mask_wc(flow
, wc
);
3051 ctx
.in_group
= false;
3052 ctx
.orig_skb_priority
= flow
->skb_priority
;
3056 if (!xin
->ofpacts
&& !ctx
.rule
) {
3057 ctx
.table_id
= rule_dpif_lookup(ctx
.xbridge
->ofproto
, flow
,
3058 !xin
->skip_wildcards
? wc
: NULL
,
3060 if (ctx
.xin
->resubmit_stats
) {
3061 rule_dpif_credit_stats(rule
, ctx
.xin
->resubmit_stats
);
3065 xout
->fail_open
= ctx
.rule
&& rule_dpif_is_fail_open(ctx
.rule
);
3068 ofpacts
= xin
->ofpacts
;
3069 ofpacts_len
= xin
->ofpacts_len
;
3070 } else if (ctx
.rule
) {
3071 actions
= rule_dpif_get_actions(ctx
.rule
);
3072 ofpacts
= actions
->ofpacts
;
3073 ofpacts_len
= actions
->ofpacts_len
;
3078 ofpbuf_use_stub(&ctx
.stack
, ctx
.init_stack
, sizeof ctx
.init_stack
);
3079 ofpbuf_use_stub(&ctx
.action_set
,
3080 ctx
.action_set_stub
, sizeof ctx
.action_set_stub
);
3082 if (mbridge_has_mirrors(ctx
.xbridge
->mbridge
)) {
3083 /* Do this conditionally because the copy is expensive enough that it
3084 * shows up in profiles. */
3088 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
3089 switch (ctx
.xbridge
->frag
) {
3090 case OFPC_FRAG_NORMAL
:
3091 /* We must pretend that transport ports are unavailable. */
3092 flow
->tp_src
= ctx
.base_flow
.tp_src
= htons(0);
3093 flow
->tp_dst
= ctx
.base_flow
.tp_dst
= htons(0);
3096 case OFPC_FRAG_DROP
:
3099 case OFPC_FRAG_REASM
:
3102 case OFPC_FRAG_NX_MATCH
:
3103 /* Nothing to do. */
3106 case OFPC_INVALID_TTL_TO_CONTROLLER
:
3111 in_port
= get_ofp_port(ctx
.xbridge
, flow
->in_port
.ofp_port
);
3112 if (in_port
&& in_port
->is_tunnel
&& ctx
.xin
->resubmit_stats
) {
3113 netdev_vport_inc_rx(in_port
->netdev
, ctx
.xin
->resubmit_stats
);
3115 bfd_account_rx(in_port
->bfd
, ctx
.xin
->resubmit_stats
);
3119 special
= process_special(&ctx
, flow
, in_port
, ctx
.xin
->packet
);
3121 ctx
.xout
->slow
|= special
;
3123 size_t sample_actions_len
;
3125 if (flow
->in_port
.ofp_port
3126 != vsp_realdev_to_vlandev(ctx
.xbridge
->ofproto
,
3127 flow
->in_port
.ofp_port
,
3129 ctx
.base_flow
.vlan_tci
= 0;
3132 add_sflow_action(&ctx
);
3133 add_ipfix_action(&ctx
);
3134 sample_actions_len
= ctx
.xout
->odp_actions
.size
;
3136 if (tnl_may_send
&& (!in_port
|| may_receive(in_port
, &ctx
))) {
3137 do_xlate_actions(ofpacts
, ofpacts_len
, &ctx
);
3139 /* We've let OFPP_NORMAL and the learning action look at the
3140 * packet, so drop it now if forwarding is disabled. */
3141 if (in_port
&& !xport_stp_forward_state(in_port
)) {
3142 ctx
.xout
->odp_actions
.size
= sample_actions_len
;
3146 if (ctx
.action_set
.size
) {
3147 xlate_action_set(&ctx
);
3150 if (ctx
.xbridge
->has_in_band
3151 && in_band_must_output_to_local_port(flow
)
3152 && !actions_output_to_local_port(&ctx
)) {
3153 compose_output_action(&ctx
, OFPP_LOCAL
);
3156 fix_sflow_action(&ctx
);
3158 if (mbridge_has_mirrors(ctx
.xbridge
->mbridge
)) {
3159 add_mirror_actions(&ctx
, &orig_flow
);
3163 if (nl_attr_oversized(ctx
.xout
->odp_actions
.size
)) {
3164 /* These datapath actions are too big for a Netlink attribute, so we
3165 * can't hand them to the kernel directly. dpif_execute() can execute
3166 * them one by one with help, so just mark the result as SLOW_ACTION to
3167 * prevent the flow from being installed. */
3168 COVERAGE_INC(xlate_actions_oversize
);
3169 ctx
.xout
->slow
|= SLOW_ACTION
;
3172 if (ctx
.xin
->resubmit_stats
) {
3173 mirror_update_stats(ctx
.xbridge
->mbridge
, xout
->mirrors
,
3174 ctx
.xin
->resubmit_stats
->n_packets
,
3175 ctx
.xin
->resubmit_stats
->n_bytes
);
3177 if (ctx
.xbridge
->netflow
) {
3178 const struct ofpact
*ofpacts
;
3181 ofpacts_len
= actions
->ofpacts_len
;
3182 ofpacts
= actions
->ofpacts
;
3183 if (ofpacts_len
== 0
3184 || ofpacts
->type
!= OFPACT_CONTROLLER
3185 || ofpact_next(ofpacts
) < ofpact_end(ofpacts
, ofpacts_len
)) {
3186 /* Only update netflow if we don't have controller flow. We don't
3187 * report NetFlow expiration messages for such facets because they
3188 * are just part of the control logic for the network, not real
3190 netflow_flow_update(ctx
.xbridge
->netflow
, flow
,
3191 xout
->nf_output_iface
,
3192 ctx
.xin
->resubmit_stats
);
3197 ofpbuf_uninit(&ctx
.stack
);
3198 ofpbuf_uninit(&ctx
.action_set
);
3200 /* Clear the metadata and register wildcard masks, because we won't
3201 * use non-header fields as part of the cache. */
3202 flow_wildcards_clear_non_packet_fields(wc
);
3204 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3205 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3206 * these fields. The datapath interface, on the other hand, represents
3207 * them with just 8 bits each. This means that if the high 8 bits of the
3208 * masks for these fields somehow become set, then they will get chopped
3209 * off by a round trip through the datapath, and revalidation will spot
3210 * that as an inconsistency and delete the flow. Avoid the problem here by
3211 * making sure that only the low 8 bits of either field can be unwildcarded
3215 wc
->masks
.tp_src
&= htons(UINT8_MAX
);
3216 wc
->masks
.tp_dst
&= htons(UINT8_MAX
);
3220 rule_dpif_unref(rule
);
3223 /* Sends 'packet' out 'ofport'.
3224 * May modify 'packet'.
3225 * Returns 0 if successful, otherwise a positive errno value. */
3227 xlate_send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
3229 struct xport
*xport
;
3230 struct ofpact_output output
;
3233 ofpact_init(&output
.ofpact
, OFPACT_OUTPUT
, sizeof output
);
3234 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3235 flow_extract(packet
, NULL
, &flow
);
3236 flow
.in_port
.ofp_port
= OFPP_NONE
;
3238 ovs_rwlock_rdlock(&xlate_rwlock
);
3239 xport
= xport_lookup(ofport
);
3241 ovs_rwlock_unlock(&xlate_rwlock
);
3244 output
.port
= xport
->ofp_port
;
3246 ovs_rwlock_unlock(&xlate_rwlock
);
3248 return ofproto_dpif_execute_actions(xport
->xbridge
->ofproto
, &flow
, NULL
,
3249 &output
.ofpact
, sizeof output
,