1 /* Copyright (c) 2009-2017, 2019-2020 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"
20 #include <sys/types.h>
21 #include <netinet/in.h>
22 #include <arpa/inet.h>
24 #include <sys/socket.h>
30 #include "byte-order.h"
35 #include "dp-packet.h"
40 #include "mac-learning.h"
41 #include "mcast-snooping.h"
42 #include "multipath.h"
43 #include "netdev-vport.h"
46 #include "odp-execute.h"
47 #include "ofproto/ofproto-dpif-ipfix.h"
48 #include "ofproto/ofproto-dpif-mirror.h"
49 #include "ofproto/ofproto-dpif-monitor.h"
50 #include "ofproto/ofproto-dpif-sflow.h"
51 #include "ofproto/ofproto-dpif-trace.h"
52 #include "ofproto/ofproto-dpif-xlate-cache.h"
53 #include "ofproto/ofproto-dpif.h"
54 #include "ofproto/ofproto-provider.h"
55 #include "openvswitch/dynamic-string.h"
56 #include "openvswitch/meta-flow.h"
57 #include "openvswitch/list.h"
58 #include "openvswitch/ofp-actions.h"
59 #include "openvswitch/ofp-ed-props.h"
60 #include "openvswitch/vlog.h"
62 #include "ovs-router.h"
64 #include "tnl-neigh-cache.h"
65 #include "tnl-ports.h"
70 COVERAGE_DEFINE(xlate_actions
);
71 COVERAGE_DEFINE(xlate_actions_oversize
);
72 COVERAGE_DEFINE(xlate_actions_too_many_output
);
74 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate
);
76 /* Maximum depth of flow table recursion (due to resubmit actions) in a
79 * The goal of limiting the depth of resubmits is to ensure that flow
80 * translation eventually terminates. Only resubmits to the same table or an
81 * earlier table count against the maximum depth. This is because resubmits to
82 * strictly monotonically increasing table IDs will eventually terminate, since
83 * any OpenFlow switch has a finite number of tables. OpenFlow tables are most
84 * commonly traversed in numerically increasing order, so this limit has little
85 * effect on conventionally designed OpenFlow pipelines.
87 * Outputs to patch ports and to groups also count against the depth limit. */
90 /* Maximum number of resubmit actions in a flow translation, whether they are
91 * recursive or not. */
92 #define MAX_RESUBMITS (MAX_DEPTH * MAX_DEPTH)
94 /* The structure holds an array of IP addresses assigned to a bridge and the
95 * number of elements in the array. These data are mutable and are evaluated
96 * when ARP or Neighbor Advertisement packets received on a native tunnel
97 * port are xlated. So 'ref_cnt' and RCU are used for synchronization. */
99 struct in6_addr
*addr
; /* Array of IP addresses of xbridge. */
100 int n_addr
; /* Number of IP addresses. */
101 struct ovs_refcount ref_cnt
;
105 struct hmap_node hmap_node
; /* Node in global 'xbridges' map. */
106 struct ofproto_dpif
*ofproto
; /* Key in global 'xbridges' map. */
108 struct ovs_list xbundles
; /* Owned xbundles. */
109 struct hmap xports
; /* Indexed by ofp_port. */
111 char *name
; /* Name used in log messages. */
112 struct dpif
*dpif
; /* Datapath interface. */
113 struct mac_learning
*ml
; /* Mac learning handle. */
114 struct mcast_snooping
*ms
; /* Multicast Snooping handle. */
115 struct mbridge
*mbridge
; /* Mirroring. */
116 struct dpif_sflow
*sflow
; /* SFlow handle, or null. */
117 struct dpif_ipfix
*ipfix
; /* Ipfix handle, or null. */
118 struct netflow
*netflow
; /* Netflow handle, or null. */
119 struct stp
*stp
; /* STP or null if disabled. */
120 struct rstp
*rstp
; /* RSTP or null if disabled. */
122 bool has_in_band
; /* Bridge has in band control? */
123 bool forward_bpdu
; /* Bridge forwards STP BPDUs? */
125 /* Datapath feature support. */
126 struct dpif_backer_support support
;
128 struct xbridge_addr
*addr
;
132 struct hmap_node hmap_node
; /* In global 'xbundles' map. */
133 struct ofbundle
*ofbundle
; /* Key in global 'xbundles' map. */
135 struct ovs_list list_node
; /* In parent 'xbridges' list. */
136 struct xbridge
*xbridge
; /* Parent xbridge. */
138 struct ovs_list xports
; /* Contains "struct xport"s. */
140 char *name
; /* Name used in log messages. */
141 struct bond
*bond
; /* Nonnull iff more than one port. */
142 struct lacp
*lacp
; /* LACP handle or null. */
144 enum port_vlan_mode vlan_mode
; /* VLAN mode. */
145 uint16_t qinq_ethtype
; /* Ethertype of dot1q-tunnel interface
146 * either 0x8100 or 0x88a8. */
147 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
148 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
149 * NULL if all VLANs are trunked. */
150 unsigned long *cvlans
; /* Bitmap of allowed customer vlans,
151 * NULL if all VLANs are allowed */
152 enum port_priority_tags_mode use_priority_tags
;
153 /* Use 802.1p tag for frames in VLAN 0? */
154 bool floodable
; /* No port has OFPUTIL_PC_NO_FLOOD set? */
155 bool protected; /* Protected port mode */
159 struct hmap_node hmap_node
; /* Node in global 'xports' map. */
160 struct ofport_dpif
*ofport
; /* Key in global 'xports map. */
162 struct hmap_node ofp_node
; /* Node in parent xbridge 'xports' map. */
163 ofp_port_t ofp_port
; /* Key in parent xbridge 'xports' map. */
165 struct hmap_node uuid_node
; /* Node in global 'xports_uuid' map. */
166 struct uuid uuid
; /* Key in global 'xports_uuid' map. */
168 odp_port_t odp_port
; /* Datapath port number or ODPP_NONE. */
170 struct ovs_list bundle_node
; /* In parent xbundle (if it exists). */
171 struct xbundle
*xbundle
; /* Parent xbundle or null. */
173 struct netdev
*netdev
; /* 'ofport''s netdev. */
175 struct xbridge
*xbridge
; /* Parent bridge. */
176 struct xport
*peer
; /* Patch port peer or null. */
178 enum ofputil_port_config config
; /* OpenFlow port configuration. */
179 enum ofputil_port_state state
; /* OpenFlow port state. */
180 int stp_port_no
; /* STP port number or -1 if not in use. */
181 struct rstp_port
*rstp_port
; /* RSTP port or null. */
183 struct hmap skb_priorities
; /* Map of 'skb_priority_to_dscp's. */
185 bool may_enable
; /* May be enabled in bonds. */
186 bool is_tunnel
; /* Is a tunnel port. */
187 enum netdev_pt_mode pt_mode
; /* packet_type handling. */
189 struct cfm
*cfm
; /* CFM handle or null. */
190 struct bfd
*bfd
; /* BFD handle or null. */
191 struct lldp
*lldp
; /* LLDP handle or null. */
195 struct xlate_in
*xin
;
196 struct xlate_out
*xout
;
198 struct xlate_cfg
*xcfg
;
199 const struct xbridge
*xbridge
;
201 /* Flow at the last commit. */
202 struct flow base_flow
;
204 /* Tunnel IP destination address as received. This is stored separately
205 * as the base_flow.tunnel is cleared on init to reflect the datapath
206 * behavior. Used to make sure not to send tunneled output to ourselves,
207 * which might lead to an infinite loop. This could happen easily
208 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
209 * actually set the tun_dst field. */
210 struct in6_addr orig_tunnel_ipv6_dst
;
212 /* Stack for the push and pop actions. See comment above nx_stack_push()
213 * in nx-match.c for info on how the stack is stored. */
216 /* The rule that we are currently translating, or NULL. */
217 struct rule_dpif
*rule
;
219 /* Flow translation populates this with wildcards relevant in translation.
220 * When 'xin->wc' is nonnull, this is the same pointer. When 'xin->wc' is
221 * null, this is a pointer to a temporary buffer. */
222 struct flow_wildcards
*wc
;
224 /* Output buffer for datapath actions. When 'xin->odp_actions' is nonnull,
225 * this is the same pointer. When 'xin->odp_actions' is null, this points
226 * to a scratch ofpbuf. This allows code to add actions to
227 * 'ctx->odp_actions' without worrying about whether the caller really
229 struct ofpbuf
*odp_actions
;
231 /* Statistics maintained by xlate_table_action().
233 * These statistics limit the amount of work that a single flow
234 * translation can perform. The goal of the first of these, 'depth', is
235 * primarily to prevent translation from performing an infinite amount of
236 * work. It counts the current depth of nested "resubmit"s (and a few
237 * other activities); when a resubmit returns, it decreases. Resubmits to
238 * tables in strictly monotonically increasing order don't contribute to
239 * 'depth' because they cannot cause a flow translation to take an infinite
240 * amount of time (because the number of tables is finite). Translation
241 * aborts when 'depth' exceeds MAX_DEPTH.
243 * 'resubmits', on the other hand, prevents flow translation from
244 * performing an extraordinarily large while still finite amount of work.
245 * It counts the total number of resubmits (and a few other activities)
246 * that have been executed. Returning from a resubmit does not affect this
247 * counter. Thus, this limits the amount of work that a particular
248 * translation can perform. Translation aborts when 'resubmits' exceeds
249 * MAX_RESUBMITS (which is much larger than MAX_DEPTH).
251 int depth
; /* Current resubmit nesting depth. */
252 int resubmits
; /* Total number of resubmits. */
253 bool in_action_set
; /* Currently translating action_set, if true. */
254 bool in_packet_out
; /* Currently translating a packet_out msg, if
256 bool pending_encap
; /* True when waiting to commit a pending
258 bool pending_decap
; /* True when waiting to commit a pending
260 struct ofpbuf
*encap_data
; /* May contain a pointer to an ofpbuf with
261 * context for the datapath encap action.*/
263 uint8_t table_id
; /* OpenFlow table ID where flow was found. */
264 ovs_be64 rule_cookie
; /* Cookie of the rule being translated. */
265 uint32_t orig_skb_priority
; /* Priority when packet arrived. */
266 uint32_t sflow_n_outputs
; /* Number of output ports. */
267 odp_port_t sflow_odp_port
; /* Output port for composing sFlow action. */
268 ofp_port_t nf_output_iface
; /* Output interface index for NetFlow. */
269 bool exit
; /* No further actions should be processed. */
270 mirror_mask_t mirrors
; /* Bitmap of associated mirrors. */
271 int mirror_snaplen
; /* Max size of a mirror packet in byte. */
273 /* Freezing Translation
274 * ====================
276 * At some point during translation, the code may recognize the need to halt
277 * and checkpoint the translation in a way that it can be restarted again
278 * later. We call the checkpointing process "freezing" and the restarting
281 * The use cases for freezing are:
283 * - "Recirculation", where the translation process discovers that it
284 * doesn't have enough information to complete translation without
285 * actually executing the actions that have already been translated,
286 * which provides the additionally needed information. In these
287 * situations, translation freezes translation and assigns the frozen
288 * data a unique "recirculation ID", which it associates with the data
289 * in a table in userspace (see ofproto-dpif-rid.h). It also adds a
290 * OVS_ACTION_ATTR_RECIRC action specifying that ID to the datapath
291 * actions. When a packet hits that action, the datapath looks its
292 * flow up again using the ID. If there's a miss, it comes back to
293 * userspace, which find the recirculation table entry for the ID,
294 * thaws the associated frozen data, and continues translation from
295 * that point given the additional information that is now known.
297 * The archetypal example is MPLS. As MPLS is implemented in
298 * OpenFlow, the protocol that follows the last MPLS label becomes
299 * known only when that label is popped by an OpenFlow action. That
300 * means that Open vSwitch can't extract the headers beyond the MPLS
301 * labels until the pop action is executed. Thus, at that point
302 * translation uses the recirculation process to extract the headers
303 * beyond the MPLS labels.
305 * (OVS also uses OVS_ACTION_ATTR_RECIRC to implement hashing for
306 * output to bonds. OVS pre-populates all the datapath flows for bond
307 * output in the datapath, though, which means that the elaborate
308 * process of coming back to userspace for a second round of
309 * translation isn't needed, and so bonds don't follow the above
312 * - "Continuation". A continuation is a way for an OpenFlow controller
313 * to interpose on a packet's traversal of the OpenFlow tables. When
314 * the translation process encounters a "controller" action with the
315 * "pause" flag, it freezes translation, serializes the frozen data,
316 * and sends it to an OpenFlow controller. The controller then
317 * examines and possibly modifies the frozen data and eventually sends
318 * it back to the switch, which thaws it and continues translation.
320 * The main problem of freezing translation is preserving state, so that
321 * when the translation is thawed later it resumes from where it left off,
322 * without disruption. In particular, actions must be preserved as follows:
324 * - If we're freezing because an action needed more information, the
325 * action that prompted it.
327 * - Any actions remaining to be translated within the current flow.
329 * - If translation was frozen within a NXAST_RESUBMIT, then any actions
330 * following the resubmit action. Resubmit actions can be nested, so
331 * this has to go all the way up the control stack.
333 * - The OpenFlow 1.1+ action set.
335 * State that actions and flow table lookups can depend on, such as the
336 * following, must also be preserved:
338 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
340 * - The stack used by NXAST_STACK_PUSH and NXAST_STACK_POP actions.
342 * - The table ID and cookie of the flow being translated at each level
343 * of the control stack, because these can become visible through
344 * OFPAT_CONTROLLER actions (and other ways).
346 * Translation allows for the control of this state preservation via these
347 * members. When a need to freeze translation is identified, the
348 * translation process:
350 * 1. Sets 'freezing' to true.
352 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
353 * translation process.
355 * 3. Adds an OFPACT_UNROLL_XLATE action to 'frozen_actions', and points
356 * frozen_actions.header to the action to make it easy to find it later.
357 * This action holds the current table ID and cookie so that they can be
358 * restored during a post-recirculation upcall translation.
360 * 4. Adds the action that prompted recirculation and any actions following
361 * it within the same flow to 'frozen_actions', so that they can be
362 * executed during a post-recirculation upcall translation.
366 * 6. The action that prompted recirculation might be nested in a stack of
367 * nested "resubmit"s that have actions remaining. Each of these notices
368 * that we're exiting and freezing and responds by adding more
369 * OFPACT_UNROLL_XLATE actions to 'frozen_actions', as necessary,
370 * followed by any actions that were yet unprocessed.
372 * If we're freezing because of recirculation, the caller generates a
373 * recirculation ID and associates all the state produced by this process
374 * with it. For post-recirculation upcall translation, the caller passes it
375 * back in for the new translation to execute. The process yielded a set of
376 * ofpacts that can be translated directly, so it is not much of a special
377 * case at that point.
380 bool recirc_update_dp_hash
; /* Generated recirculation will be preceded
381 * by datapath HASH action to get an updated
382 * dp_hash after recirculation. */
383 uint32_t dp_hash_alg
;
384 uint32_t dp_hash_basis
;
385 struct ofpbuf frozen_actions
;
386 const struct ofpact_controller
*pause
;
388 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
389 * This is a trigger for recirculation in cases where translating an action
390 * or looking up a flow requires access to the fields of the packet after
391 * the MPLS label stack that was originally present. */
394 /* True if conntrack has been performed on this packet during processing
395 * on the current bridge. This is used to determine whether conntrack
396 * state from the datapath should be honored after thawing. */
399 /* Pointer to an embedded NAT action in a conntrack action, or NULL. */
400 struct ofpact_nat
*ct_nat_action
;
402 /* OpenFlow 1.1+ action set.
404 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
405 * When translation is otherwise complete, ofpacts_execute_action_set()
406 * converts it to a set of "struct ofpact"s that can be translated into
407 * datapath actions. */
408 bool action_set_has_group
; /* Action set contains OFPACT_GROUP? */
409 struct ofpbuf action_set
; /* Action set. */
411 enum xlate_error error
; /* Translation failed. */
414 /* Structure to track VLAN manipulation */
415 struct xvlan_single
{
422 struct xvlan_single v
[FLOW_MAX_VLAN_HEADERS
];
425 const char *xlate_strerror(enum xlate_error error
)
430 case XLATE_BRIDGE_NOT_FOUND
:
431 return "Bridge not found";
432 case XLATE_RECURSION_TOO_DEEP
:
433 return "Recursion too deep";
434 case XLATE_TOO_MANY_RESUBMITS
:
435 return "Too many resubmits";
436 case XLATE_STACK_TOO_DEEP
:
437 return "Stack too deep";
438 case XLATE_NO_RECIRCULATION_CONTEXT
:
439 return "No recirculation context";
440 case XLATE_RECIRCULATION_CONFLICT
:
441 return "Recirculation conflict";
442 case XLATE_TOO_MANY_MPLS_LABELS
:
443 return "Too many MPLS labels";
444 case XLATE_INVALID_TUNNEL_METADATA
:
445 return "Invalid tunnel metadata";
446 case XLATE_UNSUPPORTED_PACKET_TYPE
:
447 return "Unsupported packet type";
448 case XLATE_CONGESTION_DROP
:
449 return "Congestion Drop";
450 case XLATE_FORWARDING_DISABLED
:
451 return "Forwarding is disabled";
455 return "Unknown error";
458 static void xlate_action_set(struct xlate_ctx
*ctx
);
459 static void xlate_commit_actions(struct xlate_ctx
*ctx
);
462 patch_port_output(struct xlate_ctx
*ctx
, const struct xport
*in_dev
,
463 struct xport
*out_dev
);
466 ctx_trigger_freeze(struct xlate_ctx
*ctx
)
469 ctx
->freezing
= true;
473 ctx_trigger_recirculate_with_hash(struct xlate_ctx
*ctx
, uint32_t type
,
477 ctx
->freezing
= true;
478 ctx
->recirc_update_dp_hash
= true;
479 ctx
->dp_hash_alg
= type
;
480 ctx
->dp_hash_basis
= basis
;
484 ctx_first_frozen_action(const struct xlate_ctx
*ctx
)
486 return !ctx
->frozen_actions
.size
;
490 ctx_cancel_freeze(struct xlate_ctx
*ctx
)
493 ctx
->freezing
= false;
494 ctx
->recirc_update_dp_hash
= false;
495 ofpbuf_clear(&ctx
->frozen_actions
);
496 ctx
->frozen_actions
.header
= NULL
;
501 static void finish_freezing(struct xlate_ctx
*ctx
);
503 /* A controller may use OFPP_NONE as the ingress port to indicate that
504 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
505 * when an input bundle is needed for validation (e.g., mirroring or
506 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
507 * any 'port' structs, so care must be taken when dealing with it. */
508 static struct xbundle ofpp_none_bundle
= {
510 .vlan_mode
= PORT_VLAN_TRUNK
513 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
514 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
515 * traffic egressing the 'ofport' with that priority should be marked with. */
516 struct skb_priority_to_dscp
{
517 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'skb_priorities'. */
518 uint32_t skb_priority
; /* Priority of this queue (see struct flow). */
520 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
523 /* Xlate config contains hash maps of all bridges, bundles and ports.
524 * Xcfgp contains the pointer to the current xlate configuration.
525 * When the main thread needs to change the configuration, it copies xcfgp to
526 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
527 * does not block handler and revalidator threads. */
529 struct hmap xbridges
;
530 struct hmap xbundles
;
532 struct hmap xports_uuid
;
534 static OVSRCU_TYPE(struct xlate_cfg
*) xcfgp
= OVSRCU_INITIALIZER(NULL
);
535 static struct xlate_cfg
*new_xcfg
= NULL
;
537 typedef void xlate_actions_handler(const struct ofpact
*, size_t ofpacts_len
,
538 struct xlate_ctx
*, bool, bool);
539 static bool may_receive(const struct xport
*, struct xlate_ctx
*);
540 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
541 struct xlate_ctx
*, bool, bool);
542 static void clone_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
543 struct xlate_ctx
*, bool, bool);
544 static void xlate_normal(struct xlate_ctx
*);
545 static void xlate_normal_flood(struct xlate_ctx
*ct
,
546 struct xbundle
*in_xbundle
, struct xvlan
*);
547 static void xlate_table_action(struct xlate_ctx
*, ofp_port_t in_port
,
548 uint8_t table_id
, bool may_packet_in
,
549 bool honor_table_miss
, bool with_ct_orig
,
550 bool is_last_action
, xlate_actions_handler
*);
552 static bool input_vid_is_valid(const struct xlate_ctx
*,
553 uint16_t vid
, struct xbundle
*);
554 static void xvlan_copy(struct xvlan
*dst
, const struct xvlan
*src
);
555 static void xvlan_pop(struct xvlan
*src
);
556 static void xvlan_push_uninit(struct xvlan
*src
);
557 static void xvlan_extract(const struct flow
*, struct xvlan
*);
558 static void xvlan_put(struct flow
*, const struct xvlan
*,
559 enum port_priority_tags_mode
);
560 static void xvlan_input_translate(const struct xbundle
*,
561 const struct xvlan
*in
,
562 struct xvlan
*xvlan
);
563 static void xvlan_output_translate(const struct xbundle
*,
564 const struct xvlan
*xvlan
,
566 static void output_normal(struct xlate_ctx
*, const struct xbundle
*,
567 const struct xvlan
*);
569 /* Optional bond recirculation parameter to compose_output_action(). */
570 struct xlate_bond_recirc
{
571 uint32_t recirc_id
; /* !0 Use recirculation instead of output. */
572 uint8_t hash_alg
; /* !0 Compute hash for recirc before. */
573 uint32_t hash_basis
; /* Compute hash for recirc before. */
576 static void compose_output_action(struct xlate_ctx
*, ofp_port_t ofp_port
,
577 const struct xlate_bond_recirc
*xr
,
578 bool is_last_action
, bool truncate
);
580 static struct xbridge
*xbridge_lookup(struct xlate_cfg
*,
581 const struct ofproto_dpif
*);
582 static struct xbridge
*xbridge_lookup_by_uuid(struct xlate_cfg
*,
583 const struct uuid
*);
584 static struct xbundle
*xbundle_lookup(struct xlate_cfg
*,
585 const struct ofbundle
*);
586 static struct xport
*xport_lookup(struct xlate_cfg
*,
587 const struct ofport_dpif
*);
588 static struct xport
*xport_lookup_by_uuid(struct xlate_cfg
*,
589 const struct uuid
*);
590 static struct xport
*get_ofp_port(const struct xbridge
*, ofp_port_t ofp_port
);
591 static struct skb_priority_to_dscp
*get_skb_priority(const struct xport
*,
592 uint32_t skb_priority
);
593 static void clear_skb_priorities(struct xport
*);
594 static size_t count_skb_priorities(const struct xport
*);
595 static bool dscp_from_skb_priority(const struct xport
*, uint32_t skb_priority
,
598 static void xlate_xbridge_init(struct xlate_cfg
*, struct xbridge
*);
599 static void xlate_xbundle_init(struct xlate_cfg
*, struct xbundle
*);
600 static void xlate_xport_init(struct xlate_cfg
*, struct xport
*);
601 static void xlate_xbridge_set(struct xbridge
*, struct dpif
*,
602 const struct mac_learning
*, struct stp
*,
603 struct rstp
*, const struct mcast_snooping
*,
604 const struct mbridge
*,
605 const struct dpif_sflow
*,
606 const struct dpif_ipfix
*,
607 const struct netflow
*,
608 bool forward_bpdu
, bool has_in_band
,
609 const struct dpif_backer_support
*,
610 const struct xbridge_addr
*);
611 static void xlate_xbundle_set(struct xbundle
*xbundle
,
612 enum port_vlan_mode vlan_mode
,
613 uint16_t qinq_ethtype
, int vlan
,
614 unsigned long *trunks
, unsigned long *cvlans
,
615 enum port_priority_tags_mode
,
616 const struct bond
*bond
, const struct lacp
*lacp
,
617 bool floodable
, bool protected);
618 static void xlate_xport_set(struct xport
*xport
, odp_port_t odp_port
,
619 const struct netdev
*netdev
, const struct cfm
*cfm
,
620 const struct bfd
*bfd
, const struct lldp
*lldp
,
621 int stp_port_no
, const struct rstp_port
*rstp_port
,
622 enum ofputil_port_config config
,
623 enum ofputil_port_state state
, bool is_tunnel
,
625 static void xlate_xbridge_remove(struct xlate_cfg
*, struct xbridge
*);
626 static void xlate_xbundle_remove(struct xlate_cfg
*, struct xbundle
*);
627 static void xlate_xport_remove(struct xlate_cfg
*, struct xport
*);
628 static void xlate_xbridge_copy(struct xbridge
*);
629 static void xlate_xbundle_copy(struct xbridge
*, struct xbundle
*);
630 static void xlate_xport_copy(struct xbridge
*, struct xbundle
*,
632 static void xlate_xcfg_free(struct xlate_cfg
*);
634 /* Tracing helpers. */
636 /* If tracing is enabled in 'ctx', creates a new trace node and appends it to
637 * the list of nodes maintained in ctx->xin. The new node has type 'type' and
638 * its text is created from 'format' by treating it as a printf format string.
639 * Returns the list of nodes embedded within the new trace node; ordinarily,
640 * the calleer can ignore this, but it is useful if the caller needs to nest
641 * more trace nodes within the new node.
643 * If tracing is not enabled, does nothing and returns NULL. */
644 static struct ovs_list
* OVS_PRINTF_FORMAT(3, 4)
645 xlate_report(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
646 const char *format
, ...)
648 struct ovs_list
*subtrace
= NULL
;
649 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
651 va_start(args
, format
);
652 char *text
= xvasprintf(format
, args
);
653 subtrace
= &oftrace_report(ctx
->xin
->trace
, type
, text
)->subs
;
660 /* This is like xlate_report() for errors that are serious enough that we
661 * should log them even if we are not tracing. */
662 static void OVS_PRINTF_FORMAT(2, 3)
663 xlate_report_error(const struct xlate_ctx
*ctx
, const char *format
, ...)
665 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
666 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
667 && (!ctx
->xin
->packet
|| VLOG_DROP_WARN(&rl
))) {
671 struct ds s
= DS_EMPTY_INITIALIZER
;
673 va_start(args
, format
);
674 ds_put_format_valist(&s
, format
, args
);
677 if (ctx
->xin
->trace
) {
678 oftrace_report(ctx
->xin
->trace
, OFT_ERROR
, ds_cstr(&s
));
680 ds_put_format(&s
, " on bridge %s while processing ",
682 flow_format(&s
, &ctx
->base_flow
, NULL
);
683 VLOG_WARN("%s", ds_cstr(&s
));
688 /* This is like xlate_report() for messages that should be logged
689 at the info level (even when not tracing). */
690 static void OVS_PRINTF_FORMAT(2, 3)
691 xlate_report_info(const struct xlate_ctx
*ctx
, const char *format
, ...)
693 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
694 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
695 && (!ctx
->xin
->packet
|| VLOG_DROP_INFO(&rl
))) {
699 struct ds s
= DS_EMPTY_INITIALIZER
;
701 va_start(args
, format
);
702 ds_put_format_valist(&s
, format
, args
);
705 if (ctx
->xin
->trace
) {
706 oftrace_report(ctx
->xin
->trace
, OFT_WARN
, ds_cstr(&s
));
708 ds_put_format(&s
, " on bridge %s while processing ",
710 flow_format(&s
, &ctx
->base_flow
, NULL
);
711 VLOG_INFO("%s", ds_cstr(&s
));
716 /* This is like xlate_report() for messages that should be logged at debug
717 * level (even if we are not tracing) because they can be valuable for
719 static void OVS_PRINTF_FORMAT(3, 4)
720 xlate_report_debug(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
721 const char *format
, ...)
723 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
724 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
725 && (!ctx
->xin
->packet
|| VLOG_DROP_DBG(&rl
))) {
729 struct ds s
= DS_EMPTY_INITIALIZER
;
731 va_start(args
, format
);
732 ds_put_format_valist(&s
, format
, args
);
735 if (ctx
->xin
->trace
) {
736 oftrace_report(ctx
->xin
->trace
, type
, ds_cstr(&s
));
738 VLOG_DBG("bridge %s: %s", ctx
->xbridge
->name
, ds_cstr(&s
));
743 /* If tracing is enabled in 'ctx', appends a node of the given 'type' to the
744 * trace, whose text is 'title' followed by a formatted version of the
745 * 'ofpacts_len' OpenFlow actions in 'ofpacts'.
747 * If tracing is not enabled, does nothing. */
749 xlate_report_actions(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
751 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
753 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
754 struct ds s
= DS_EMPTY_INITIALIZER
;
755 ds_put_format(&s
, "%s: ", title
);
756 struct ofpact_format_params fp
= { .s
= &s
};
757 ofpacts_format(ofpacts
, ofpacts_len
, &fp
);
758 oftrace_report(ctx
->xin
->trace
, type
, ds_cstr(&s
));
763 /* If tracing is enabled in 'ctx', appends a node of type OFT_DETAIL to the
764 * trace, whose the message is a formatted version of the OpenFlow action set.
765 * 'verb' should be "was" or "is", depending on whether the action set reported
766 * is the new action set or the old one.
768 * If tracing is not enabled, does nothing. */
770 xlate_report_action_set(const struct xlate_ctx
*ctx
, const char *verb
)
772 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
773 struct ofpbuf action_list
;
774 ofpbuf_init(&action_list
, 0);
775 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
776 if (action_list
.size
) {
777 struct ds s
= DS_EMPTY_INITIALIZER
;
778 struct ofpact_format_params fp
= { .s
= &s
};
779 ofpacts_format(action_list
.data
, action_list
.size
, &fp
);
780 xlate_report(ctx
, OFT_DETAIL
, "action set %s: %s",
784 xlate_report(ctx
, OFT_DETAIL
, "action set %s empty", verb
);
786 ofpbuf_uninit(&action_list
);
791 /* If tracing is enabled in 'ctx', appends a node representing 'rule' (in
792 * OpenFlow table 'table_id') to the trace and makes this node the parent for
793 * future trace nodes. The caller should save ctx->xin->trace before calling
794 * this function, then after tracing all of the activities under the table,
795 * restore its previous value.
797 * If tracing is not enabled, does nothing. */
799 xlate_report_table(const struct xlate_ctx
*ctx
, struct rule_dpif
*rule
,
802 if (OVS_LIKELY(!ctx
->xin
->trace
)) {
806 struct ds s
= DS_EMPTY_INITIALIZER
;
807 ds_put_format(&s
, "%2d. ", table_id
);
808 if (rule
== ctx
->xin
->ofproto
->miss_rule
) {
809 ds_put_cstr(&s
, "No match, and a \"packet-in\" is called for.");
810 } else if (rule
== ctx
->xin
->ofproto
->no_packet_in_rule
) {
811 ds_put_cstr(&s
, "No match.");
812 } else if (rule
== ctx
->xin
->ofproto
->drop_frags_rule
) {
813 ds_put_cstr(&s
, "Packets are IP fragments and "
814 "the fragment handling mode is \"drop\".");
816 minimatch_format(&rule
->up
.cr
.match
,
817 ofproto_get_tun_tab(&ctx
->xin
->ofproto
->up
),
818 NULL
, &s
, OFP_DEFAULT_PRIORITY
);
819 if (ds_last(&s
) != ' ') {
820 ds_put_cstr(&s
, ", ");
822 ds_put_format(&s
, "priority %d", rule
->up
.cr
.priority
);
823 if (rule
->up
.flow_cookie
) {
824 ds_put_format(&s
, ", cookie %#"PRIx64
,
825 ntohll(rule
->up
.flow_cookie
));
828 ctx
->xin
->trace
= &oftrace_report(ctx
->xin
->trace
, OFT_TABLE
,
833 /* If tracing is enabled in 'ctx', adds an OFT_DETAIL trace node to 'ctx'
834 * reporting the value of subfield 'sf'.
836 * If tracing is not enabled, does nothing. */
838 xlate_report_subfield(const struct xlate_ctx
*ctx
,
839 const struct mf_subfield
*sf
)
841 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
842 struct ds s
= DS_EMPTY_INITIALIZER
;
843 mf_format_subfield(sf
, &s
);
844 ds_put_cstr(&s
, " is now ");
846 if (sf
->ofs
== 0 && sf
->n_bits
>= sf
->field
->n_bits
) {
847 union mf_value value
;
848 mf_get_value(sf
->field
, &ctx
->xin
->flow
, &value
);
849 mf_format(sf
->field
, &value
, NULL
, NULL
, &s
);
851 union mf_subvalue cst
;
852 mf_read_subfield(sf
, &ctx
->xin
->flow
, &cst
);
853 ds_put_hex(&s
, &cst
, sizeof cst
);
856 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
863 xlate_xbridge_init(struct xlate_cfg
*xcfg
, struct xbridge
*xbridge
)
865 ovs_list_init(&xbridge
->xbundles
);
866 hmap_init(&xbridge
->xports
);
867 hmap_insert(&xcfg
->xbridges
, &xbridge
->hmap_node
,
868 hash_pointer(xbridge
->ofproto
, 0));
872 xlate_xbundle_init(struct xlate_cfg
*xcfg
, struct xbundle
*xbundle
)
874 ovs_list_init(&xbundle
->xports
);
875 ovs_list_insert(&xbundle
->xbridge
->xbundles
, &xbundle
->list_node
);
876 hmap_insert(&xcfg
->xbundles
, &xbundle
->hmap_node
,
877 hash_pointer(xbundle
->ofbundle
, 0));
881 xlate_xport_init(struct xlate_cfg
*xcfg
, struct xport
*xport
)
883 hmap_init(&xport
->skb_priorities
);
884 hmap_insert(&xcfg
->xports
, &xport
->hmap_node
,
885 hash_pointer(xport
->ofport
, 0));
886 hmap_insert(&xport
->xbridge
->xports
, &xport
->ofp_node
,
887 hash_ofp_port(xport
->ofp_port
));
888 hmap_insert(&xcfg
->xports_uuid
, &xport
->uuid_node
,
889 uuid_hash(&xport
->uuid
));
892 static struct xbridge_addr
*
893 xbridge_addr_create(struct xbridge
*xbridge
)
895 struct xbridge_addr
*xbridge_addr
= xbridge
->addr
;
896 struct in6_addr
*addr
= NULL
, *mask
= NULL
;
900 err
= netdev_open(xbridge
->name
, NULL
, &dev
);
902 err
= netdev_get_addr_list(dev
, &addr
, &mask
, &n_addr
);
904 if (!xbridge
->addr
||
905 n_addr
!= xbridge
->addr
->n_addr
||
906 (xbridge
->addr
->addr
&& memcmp(addr
, xbridge
->addr
->addr
,
907 sizeof(*addr
) * n_addr
))) {
908 xbridge_addr
= xzalloc(sizeof *xbridge_addr
);
909 xbridge_addr
->addr
= addr
;
910 xbridge_addr
->n_addr
= n_addr
;
911 ovs_refcount_init(&xbridge_addr
->ref_cnt
);
923 static struct xbridge_addr
*
924 xbridge_addr_ref(const struct xbridge_addr
*addr_
)
926 struct xbridge_addr
*addr
= CONST_CAST(struct xbridge_addr
*, addr_
);
928 ovs_refcount_ref(&addr
->ref_cnt
);
934 xbridge_addr_unref(struct xbridge_addr
*addr
)
936 if (addr
&& ovs_refcount_unref_relaxed(&addr
->ref_cnt
) == 1) {
943 xlate_xbridge_set(struct xbridge
*xbridge
,
945 const struct mac_learning
*ml
, struct stp
*stp
,
946 struct rstp
*rstp
, const struct mcast_snooping
*ms
,
947 const struct mbridge
*mbridge
,
948 const struct dpif_sflow
*sflow
,
949 const struct dpif_ipfix
*ipfix
,
950 const struct netflow
*netflow
,
951 bool forward_bpdu
, bool has_in_band
,
952 const struct dpif_backer_support
*support
,
953 const struct xbridge_addr
*addr
)
955 if (xbridge
->ml
!= ml
) {
956 mac_learning_unref(xbridge
->ml
);
957 xbridge
->ml
= mac_learning_ref(ml
);
960 if (xbridge
->ms
!= ms
) {
961 mcast_snooping_unref(xbridge
->ms
);
962 xbridge
->ms
= mcast_snooping_ref(ms
);
965 if (xbridge
->mbridge
!= mbridge
) {
966 mbridge_unref(xbridge
->mbridge
);
967 xbridge
->mbridge
= mbridge_ref(mbridge
);
970 if (xbridge
->sflow
!= sflow
) {
971 dpif_sflow_unref(xbridge
->sflow
);
972 xbridge
->sflow
= dpif_sflow_ref(sflow
);
975 if (xbridge
->ipfix
!= ipfix
) {
976 dpif_ipfix_unref(xbridge
->ipfix
);
977 xbridge
->ipfix
= dpif_ipfix_ref(ipfix
);
980 if (xbridge
->stp
!= stp
) {
981 stp_unref(xbridge
->stp
);
982 xbridge
->stp
= stp_ref(stp
);
985 if (xbridge
->rstp
!= rstp
) {
986 rstp_unref(xbridge
->rstp
);
987 xbridge
->rstp
= rstp_ref(rstp
);
990 if (xbridge
->netflow
!= netflow
) {
991 netflow_unref(xbridge
->netflow
);
992 xbridge
->netflow
= netflow_ref(netflow
);
995 if (xbridge
->addr
!= addr
) {
996 xbridge_addr_unref(xbridge
->addr
);
997 xbridge
->addr
= xbridge_addr_ref(addr
);
1000 xbridge
->dpif
= dpif
;
1001 xbridge
->forward_bpdu
= forward_bpdu
;
1002 xbridge
->has_in_band
= has_in_band
;
1003 xbridge
->support
= *support
;
1007 xlate_xbundle_set(struct xbundle
*xbundle
,
1008 enum port_vlan_mode vlan_mode
, uint16_t qinq_ethtype
,
1009 int vlan
, unsigned long *trunks
, unsigned long *cvlans
,
1010 enum port_priority_tags_mode use_priority_tags
,
1011 const struct bond
*bond
, const struct lacp
*lacp
,
1012 bool floodable
, bool protected)
1014 ovs_assert(xbundle
->xbridge
);
1016 xbundle
->vlan_mode
= vlan_mode
;
1017 xbundle
->qinq_ethtype
= qinq_ethtype
;
1018 xbundle
->vlan
= vlan
;
1019 xbundle
->trunks
= trunks
;
1020 xbundle
->cvlans
= cvlans
;
1021 xbundle
->use_priority_tags
= use_priority_tags
;
1022 xbundle
->floodable
= floodable
;
1023 xbundle
->protected = protected;
1025 if (xbundle
->bond
!= bond
) {
1026 bond_unref(xbundle
->bond
);
1027 xbundle
->bond
= bond_ref(bond
);
1030 if (xbundle
->lacp
!= lacp
) {
1031 lacp_unref(xbundle
->lacp
);
1032 xbundle
->lacp
= lacp_ref(lacp
);
1037 xlate_xport_set(struct xport
*xport
, odp_port_t odp_port
,
1038 const struct netdev
*netdev
, const struct cfm
*cfm
,
1039 const struct bfd
*bfd
, const struct lldp
*lldp
, int stp_port_no
,
1040 const struct rstp_port
* rstp_port
,
1041 enum ofputil_port_config config
, enum ofputil_port_state state
,
1042 bool is_tunnel
, bool may_enable
)
1044 xport
->config
= config
;
1045 xport
->state
= state
;
1046 xport
->stp_port_no
= stp_port_no
;
1047 xport
->is_tunnel
= is_tunnel
;
1048 xport
->pt_mode
= netdev_get_pt_mode(netdev
);
1049 xport
->may_enable
= may_enable
;
1050 xport
->odp_port
= odp_port
;
1052 if (xport
->rstp_port
!= rstp_port
) {
1053 rstp_port_unref(xport
->rstp_port
);
1054 xport
->rstp_port
= rstp_port_ref(rstp_port
);
1057 if (xport
->cfm
!= cfm
) {
1058 cfm_unref(xport
->cfm
);
1059 xport
->cfm
= cfm_ref(cfm
);
1062 if (xport
->bfd
!= bfd
) {
1063 bfd_unref(xport
->bfd
);
1064 xport
->bfd
= bfd_ref(bfd
);
1067 if (xport
->lldp
!= lldp
) {
1068 lldp_unref(xport
->lldp
);
1069 xport
->lldp
= lldp_ref(lldp
);
1072 if (xport
->netdev
!= netdev
) {
1073 netdev_close(xport
->netdev
);
1074 xport
->netdev
= netdev_ref(netdev
);
1079 xlate_xbridge_copy(struct xbridge
*xbridge
)
1081 struct xbundle
*xbundle
;
1082 struct xport
*xport
;
1083 struct xbridge
*new_xbridge
= xzalloc(sizeof *xbridge
);
1084 new_xbridge
->ofproto
= xbridge
->ofproto
;
1085 new_xbridge
->name
= xstrdup(xbridge
->name
);
1086 xlate_xbridge_init(new_xcfg
, new_xbridge
);
1088 xlate_xbridge_set(new_xbridge
,
1089 xbridge
->dpif
, xbridge
->ml
, xbridge
->stp
,
1090 xbridge
->rstp
, xbridge
->ms
, xbridge
->mbridge
,
1091 xbridge
->sflow
, xbridge
->ipfix
, xbridge
->netflow
,
1092 xbridge
->forward_bpdu
, xbridge
->has_in_band
,
1093 &xbridge
->support
, xbridge
->addr
);
1094 LIST_FOR_EACH (xbundle
, list_node
, &xbridge
->xbundles
) {
1095 xlate_xbundle_copy(new_xbridge
, xbundle
);
1098 /* Copy xports which are not part of a xbundle */
1099 HMAP_FOR_EACH (xport
, ofp_node
, &xbridge
->xports
) {
1100 if (!xport
->xbundle
) {
1101 xlate_xport_copy(new_xbridge
, NULL
, xport
);
1107 xlate_xbundle_copy(struct xbridge
*xbridge
, struct xbundle
*xbundle
)
1109 struct xport
*xport
;
1110 struct xbundle
*new_xbundle
= xzalloc(sizeof *xbundle
);
1111 new_xbundle
->ofbundle
= xbundle
->ofbundle
;
1112 new_xbundle
->xbridge
= xbridge
;
1113 new_xbundle
->name
= xstrdup(xbundle
->name
);
1114 xlate_xbundle_init(new_xcfg
, new_xbundle
);
1116 xlate_xbundle_set(new_xbundle
, xbundle
->vlan_mode
, xbundle
->qinq_ethtype
,
1117 xbundle
->vlan
, xbundle
->trunks
, xbundle
->cvlans
,
1118 xbundle
->use_priority_tags
, xbundle
->bond
, xbundle
->lacp
,
1119 xbundle
->floodable
, xbundle
->protected);
1120 LIST_FOR_EACH (xport
, bundle_node
, &xbundle
->xports
) {
1121 xlate_xport_copy(xbridge
, new_xbundle
, xport
);
1126 xlate_xport_copy(struct xbridge
*xbridge
, struct xbundle
*xbundle
,
1127 struct xport
*xport
)
1129 struct skb_priority_to_dscp
*pdscp
, *new_pdscp
;
1130 struct xport
*new_xport
= xzalloc(sizeof *xport
);
1131 new_xport
->ofport
= xport
->ofport
;
1132 new_xport
->ofp_port
= xport
->ofp_port
;
1133 new_xport
->xbridge
= xbridge
;
1134 new_xport
->uuid
= xport
->uuid
;
1135 xlate_xport_init(new_xcfg
, new_xport
);
1137 xlate_xport_set(new_xport
, xport
->odp_port
, xport
->netdev
, xport
->cfm
,
1138 xport
->bfd
, xport
->lldp
, xport
->stp_port_no
,
1139 xport
->rstp_port
, xport
->config
, xport
->state
,
1140 xport
->is_tunnel
, xport
->may_enable
);
1143 struct xport
*peer
= xport_lookup(new_xcfg
, xport
->peer
->ofport
);
1145 new_xport
->peer
= peer
;
1146 new_xport
->peer
->peer
= new_xport
;
1151 new_xport
->xbundle
= xbundle
;
1152 ovs_list_insert(&new_xport
->xbundle
->xports
, &new_xport
->bundle_node
);
1155 HMAP_FOR_EACH (pdscp
, hmap_node
, &xport
->skb_priorities
) {
1156 new_pdscp
= xmalloc(sizeof *pdscp
);
1157 new_pdscp
->skb_priority
= pdscp
->skb_priority
;
1158 new_pdscp
->dscp
= pdscp
->dscp
;
1159 hmap_insert(&new_xport
->skb_priorities
, &new_pdscp
->hmap_node
,
1160 hash_int(new_pdscp
->skb_priority
, 0));
1164 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
1165 * configuration in xcfgp.
1167 * This needs to be called after editing the xlate configuration.
1169 * Functions that edit the new xlate configuration are
1170 * xlate_<ofproto/bundle/ofport>_set and xlate_<ofproto/bundle/ofport>_remove.
1172 * A sample workflow:
1174 * xlate_txn_start();
1176 * edit_xlate_configuration();
1178 * xlate_txn_commit();
1180 * The ovsrcu_synchronize() call here also ensures that the upcall threads
1181 * retain no references to anything in the previous configuration.
1184 xlate_txn_commit(void)
1186 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1188 ovsrcu_set(&xcfgp
, new_xcfg
);
1189 ovsrcu_synchronize();
1190 xlate_xcfg_free(xcfg
);
1194 /* Copies the current xlate configuration in xcfgp to new_xcfg.
1196 * This needs to be called prior to editing the xlate configuration. */
1198 xlate_txn_start(void)
1200 struct xbridge
*xbridge
;
1201 struct xlate_cfg
*xcfg
;
1203 ovs_assert(!new_xcfg
);
1205 new_xcfg
= xmalloc(sizeof *new_xcfg
);
1206 hmap_init(&new_xcfg
->xbridges
);
1207 hmap_init(&new_xcfg
->xbundles
);
1208 hmap_init(&new_xcfg
->xports
);
1209 hmap_init(&new_xcfg
->xports_uuid
);
1211 xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1216 HMAP_FOR_EACH (xbridge
, hmap_node
, &xcfg
->xbridges
) {
1217 xlate_xbridge_copy(xbridge
);
1223 xlate_xcfg_free(struct xlate_cfg
*xcfg
)
1225 struct xbridge
*xbridge
, *next_xbridge
;
1231 HMAP_FOR_EACH_SAFE (xbridge
, next_xbridge
, hmap_node
, &xcfg
->xbridges
) {
1232 xlate_xbridge_remove(xcfg
, xbridge
);
1235 hmap_destroy(&xcfg
->xbridges
);
1236 hmap_destroy(&xcfg
->xbundles
);
1237 hmap_destroy(&xcfg
->xports
);
1238 hmap_destroy(&xcfg
->xports_uuid
);
1243 xlate_ofproto_set(struct ofproto_dpif
*ofproto
, const char *name
,
1245 const struct mac_learning
*ml
, struct stp
*stp
,
1246 struct rstp
*rstp
, const struct mcast_snooping
*ms
,
1247 const struct mbridge
*mbridge
,
1248 const struct dpif_sflow
*sflow
,
1249 const struct dpif_ipfix
*ipfix
,
1250 const struct netflow
*netflow
,
1251 bool forward_bpdu
, bool has_in_band
,
1252 const struct dpif_backer_support
*support
)
1254 struct xbridge
*xbridge
;
1255 struct xbridge_addr
*xbridge_addr
, *old_addr
;
1257 ovs_assert(new_xcfg
);
1259 xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1261 xbridge
= xzalloc(sizeof *xbridge
);
1262 xbridge
->ofproto
= ofproto
;
1264 xlate_xbridge_init(new_xcfg
, xbridge
);
1267 free(xbridge
->name
);
1268 xbridge
->name
= xstrdup(name
);
1270 xbridge_addr
= xbridge_addr_create(xbridge
);
1271 old_addr
= xbridge
->addr
;
1273 xlate_xbridge_set(xbridge
, dpif
, ml
, stp
, rstp
, ms
, mbridge
, sflow
, ipfix
,
1274 netflow
, forward_bpdu
, has_in_band
, support
,
1277 if (xbridge_addr
!= old_addr
) {
1278 xbridge_addr_unref(xbridge_addr
);
1283 xlate_xbridge_remove(struct xlate_cfg
*xcfg
, struct xbridge
*xbridge
)
1285 struct xbundle
*xbundle
, *next_xbundle
;
1286 struct xport
*xport
, *next_xport
;
1292 HMAP_FOR_EACH_SAFE (xport
, next_xport
, ofp_node
, &xbridge
->xports
) {
1293 xlate_xport_remove(xcfg
, xport
);
1296 LIST_FOR_EACH_SAFE (xbundle
, next_xbundle
, list_node
, &xbridge
->xbundles
) {
1297 xlate_xbundle_remove(xcfg
, xbundle
);
1300 hmap_remove(&xcfg
->xbridges
, &xbridge
->hmap_node
);
1301 mac_learning_unref(xbridge
->ml
);
1302 mcast_snooping_unref(xbridge
->ms
);
1303 mbridge_unref(xbridge
->mbridge
);
1304 dpif_sflow_unref(xbridge
->sflow
);
1305 dpif_ipfix_unref(xbridge
->ipfix
);
1306 netflow_unref(xbridge
->netflow
);
1307 stp_unref(xbridge
->stp
);
1308 rstp_unref(xbridge
->rstp
);
1309 xbridge_addr_unref(xbridge
->addr
);
1310 hmap_destroy(&xbridge
->xports
);
1311 free(xbridge
->name
);
1316 xlate_remove_ofproto(struct ofproto_dpif
*ofproto
)
1318 struct xbridge
*xbridge
;
1320 ovs_assert(new_xcfg
);
1322 xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1323 xlate_xbridge_remove(new_xcfg
, xbridge
);
1327 xlate_bundle_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
1328 const char *name
, enum port_vlan_mode vlan_mode
,
1329 uint16_t qinq_ethtype
, int vlan
,
1330 unsigned long *trunks
, unsigned long *cvlans
,
1331 enum port_priority_tags_mode use_priority_tags
,
1332 const struct bond
*bond
, const struct lacp
*lacp
,
1333 bool floodable
, bool protected)
1335 struct xbundle
*xbundle
;
1337 ovs_assert(new_xcfg
);
1339 xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1341 xbundle
= xzalloc(sizeof *xbundle
);
1342 xbundle
->ofbundle
= ofbundle
;
1343 xbundle
->xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1345 xlate_xbundle_init(new_xcfg
, xbundle
);
1348 free(xbundle
->name
);
1349 xbundle
->name
= xstrdup(name
);
1351 xlate_xbundle_set(xbundle
, vlan_mode
, qinq_ethtype
, vlan
, trunks
, cvlans
,
1352 use_priority_tags
, bond
, lacp
, floodable
, protected);
1356 xlate_xbundle_remove(struct xlate_cfg
*xcfg
, struct xbundle
*xbundle
)
1358 struct xport
*xport
;
1364 LIST_FOR_EACH_POP (xport
, bundle_node
, &xbundle
->xports
) {
1365 xport
->xbundle
= NULL
;
1368 hmap_remove(&xcfg
->xbundles
, &xbundle
->hmap_node
);
1369 ovs_list_remove(&xbundle
->list_node
);
1370 bond_unref(xbundle
->bond
);
1371 lacp_unref(xbundle
->lacp
);
1372 free(xbundle
->name
);
1377 xlate_bundle_remove(struct ofbundle
*ofbundle
)
1379 struct xbundle
*xbundle
;
1381 ovs_assert(new_xcfg
);
1383 xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1384 xlate_xbundle_remove(new_xcfg
, xbundle
);
1388 xlate_ofport_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
1389 struct ofport_dpif
*ofport
, ofp_port_t ofp_port
,
1390 odp_port_t odp_port
, const struct netdev
*netdev
,
1391 const struct cfm
*cfm
, const struct bfd
*bfd
,
1392 const struct lldp
*lldp
, struct ofport_dpif
*peer
,
1393 int stp_port_no
, const struct rstp_port
*rstp_port
,
1394 const struct ofproto_port_queue
*qdscp_list
, size_t n_qdscp
,
1395 enum ofputil_port_config config
,
1396 enum ofputil_port_state state
, bool is_tunnel
,
1400 struct xport
*xport
;
1402 ovs_assert(new_xcfg
);
1404 xport
= xport_lookup(new_xcfg
, ofport
);
1406 xport
= xzalloc(sizeof *xport
);
1407 xport
->ofport
= ofport
;
1408 xport
->xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1409 xport
->ofp_port
= ofp_port
;
1410 uuid_generate(&xport
->uuid
);
1412 xlate_xport_init(new_xcfg
, xport
);
1415 ovs_assert(xport
->ofp_port
== ofp_port
);
1417 xlate_xport_set(xport
, odp_port
, netdev
, cfm
, bfd
, lldp
,
1418 stp_port_no
, rstp_port
, config
, state
, is_tunnel
,
1422 xport
->peer
->peer
= NULL
;
1424 xport
->peer
= xport_lookup(new_xcfg
, peer
);
1426 xport
->peer
->peer
= xport
;
1429 if (xport
->xbundle
) {
1430 ovs_list_remove(&xport
->bundle_node
);
1432 xport
->xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1433 if (xport
->xbundle
) {
1434 ovs_list_insert(&xport
->xbundle
->xports
, &xport
->bundle_node
);
1437 clear_skb_priorities(xport
);
1438 for (i
= 0; i
< n_qdscp
; i
++) {
1439 struct skb_priority_to_dscp
*pdscp
;
1440 uint32_t skb_priority
;
1442 if (dpif_queue_to_priority(xport
->xbridge
->dpif
, qdscp_list
[i
].queue
,
1447 pdscp
= xmalloc(sizeof *pdscp
);
1448 pdscp
->skb_priority
= skb_priority
;
1449 pdscp
->dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
1450 hmap_insert(&xport
->skb_priorities
, &pdscp
->hmap_node
,
1451 hash_int(pdscp
->skb_priority
, 0));
1456 xlate_xport_remove(struct xlate_cfg
*xcfg
, struct xport
*xport
)
1463 xport
->peer
->peer
= NULL
;
1467 if (xport
->xbundle
) {
1468 ovs_list_remove(&xport
->bundle_node
);
1471 clear_skb_priorities(xport
);
1472 hmap_destroy(&xport
->skb_priorities
);
1474 hmap_remove(&xcfg
->xports
, &xport
->hmap_node
);
1475 hmap_remove(&xcfg
->xports_uuid
, &xport
->uuid_node
);
1476 hmap_remove(&xport
->xbridge
->xports
, &xport
->ofp_node
);
1478 netdev_close(xport
->netdev
);
1479 rstp_port_unref(xport
->rstp_port
);
1480 cfm_unref(xport
->cfm
);
1481 bfd_unref(xport
->bfd
);
1482 lldp_unref(xport
->lldp
);
1487 xlate_ofport_remove(struct ofport_dpif
*ofport
)
1489 struct xport
*xport
;
1491 ovs_assert(new_xcfg
);
1493 xport
= xport_lookup(new_xcfg
, ofport
);
1495 tnl_neigh_flush(netdev_get_name(xport
->netdev
));
1497 xlate_xport_remove(new_xcfg
, xport
);
1500 static struct ofproto_dpif
*
1501 xlate_lookup_ofproto_(const struct dpif_backer
*backer
,
1502 const struct flow
*flow
,
1503 ofp_port_t
*ofp_in_port
, const struct xport
**xportp
,
1506 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1507 const struct xport
*xport
;
1509 /* If packet is recirculated, xport can be retrieved from frozen state. */
1510 if (flow
->recirc_id
) {
1511 const struct recirc_id_node
*recirc_id_node
;
1513 recirc_id_node
= recirc_id_node_find(flow
->recirc_id
);
1515 if (OVS_UNLIKELY(!recirc_id_node
)) {
1517 *errorp
= xasprintf("no recirculation data for recirc_id "
1518 "%"PRIu32
, flow
->recirc_id
);
1523 ofp_port_t in_port
= recirc_id_node
->state
.metadata
.in_port
;
1524 if (in_port
!= OFPP_NONE
&& in_port
!= OFPP_CONTROLLER
) {
1525 struct uuid xport_uuid
= recirc_id_node
->state
.xport_uuid
;
1526 xport
= xport_lookup_by_uuid(xcfg
, &xport_uuid
);
1527 if (xport
&& xport
->xbridge
&& xport
->xbridge
->ofproto
) {
1531 /* OFPP_NONE and OFPP_CONTROLLER are not real ports. They indicate
1532 * that the packet originated from the controller via an OpenFlow
1533 * "packet-out". The right thing to do is to find just the
1534 * ofproto. There is no xport, which is OK.
1536 * OFPP_NONE can also indicate that a bond caused recirculation. */
1537 struct uuid uuid
= recirc_id_node
->state
.ofproto_uuid
;
1538 const struct xbridge
*bridge
= xbridge_lookup_by_uuid(xcfg
, &uuid
);
1539 if (bridge
&& bridge
->ofproto
) {
1545 *ofp_in_port
= in_port
;
1547 return bridge
->ofproto
;
1552 xport
= xport_lookup(xcfg
, tnl_port_should_receive(flow
)
1553 ? tnl_port_receive(flow
)
1554 : odp_port_to_ofport(backer
, flow
->in_port
.odp_port
));
1555 if (OVS_UNLIKELY(!xport
)) {
1557 *errorp
= (tnl_port_should_receive(flow
)
1558 ? xstrdup("no OpenFlow tunnel port for this packet")
1559 : xasprintf("no OpenFlow tunnel port for datapath "
1560 "port %"PRIu32
, flow
->in_port
.odp_port
));
1571 *ofp_in_port
= xport
->ofp_port
;
1573 return xport
->xbridge
->ofproto
;
1576 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1577 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1578 struct ofproto_dpif
*
1579 xlate_lookup_ofproto(const struct dpif_backer
*backer
, const struct flow
*flow
,
1580 ofp_port_t
*ofp_in_port
, char **errorp
)
1582 const struct xport
*xport
;
1584 return xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, errorp
);
1587 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1588 * optionally populates 'ofprotop' with the ofproto_dpif, 'ofp_in_port' with the
1589 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1590 * handles for those protocols if they're enabled. Caller may use the returned
1591 * pointers until quiescing, for longer term use additional references must
1594 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1597 xlate_lookup(const struct dpif_backer
*backer
, const struct flow
*flow
,
1598 struct ofproto_dpif
**ofprotop
, struct dpif_ipfix
**ipfix
,
1599 struct dpif_sflow
**sflow
, struct netflow
**netflow
,
1600 ofp_port_t
*ofp_in_port
)
1602 struct ofproto_dpif
*ofproto
;
1603 const struct xport
*xport
;
1605 ofproto
= xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, NULL
);
1612 *ofprotop
= ofproto
;
1616 *ipfix
= xport
? xport
->xbridge
->ipfix
: NULL
;
1620 *sflow
= xport
? xport
->xbridge
->sflow
: NULL
;
1624 *netflow
= xport
? xport
->xbridge
->netflow
: NULL
;
1630 static struct xbridge
*
1631 xbridge_lookup(struct xlate_cfg
*xcfg
, const struct ofproto_dpif
*ofproto
)
1633 struct hmap
*xbridges
;
1634 struct xbridge
*xbridge
;
1636 if (!ofproto
|| !xcfg
) {
1640 xbridges
= &xcfg
->xbridges
;
1642 HMAP_FOR_EACH_IN_BUCKET (xbridge
, hmap_node
, hash_pointer(ofproto
, 0),
1644 if (xbridge
->ofproto
== ofproto
) {
1651 static struct xbridge
*
1652 xbridge_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1654 struct xbridge
*xbridge
;
1656 HMAP_FOR_EACH (xbridge
, hmap_node
, &xcfg
->xbridges
) {
1657 if (uuid_equals(&xbridge
->ofproto
->uuid
, uuid
)) {
1664 static struct xbundle
*
1665 xbundle_lookup(struct xlate_cfg
*xcfg
, const struct ofbundle
*ofbundle
)
1667 struct hmap
*xbundles
;
1668 struct xbundle
*xbundle
;
1670 if (!ofbundle
|| !xcfg
) {
1674 xbundles
= &xcfg
->xbundles
;
1676 HMAP_FOR_EACH_IN_BUCKET (xbundle
, hmap_node
, hash_pointer(ofbundle
, 0),
1678 if (xbundle
->ofbundle
== ofbundle
) {
1685 static struct xport
*
1686 xport_lookup(struct xlate_cfg
*xcfg
, const struct ofport_dpif
*ofport
)
1688 struct hmap
*xports
;
1689 struct xport
*xport
;
1691 if (!ofport
|| !xcfg
) {
1695 xports
= &xcfg
->xports
;
1697 HMAP_FOR_EACH_IN_BUCKET (xport
, hmap_node
, hash_pointer(ofport
, 0),
1699 if (xport
->ofport
== ofport
) {
1706 static struct xport
*
1707 xport_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1709 struct hmap
*xports
;
1710 struct xport
*xport
;
1712 if (uuid_is_zero(uuid
) || !xcfg
) {
1716 xports
= &xcfg
->xports_uuid
;
1718 HMAP_FOR_EACH_IN_BUCKET (xport
, uuid_node
, uuid_hash(uuid
), xports
) {
1719 if (uuid_equals(&xport
->uuid
, uuid
)) {
1726 static struct stp_port
*
1727 xport_get_stp_port(const struct xport
*xport
)
1729 return xport
->xbridge
->stp
&& xport
->stp_port_no
!= -1
1730 ? stp_get_port(xport
->xbridge
->stp
, xport
->stp_port_no
)
1735 xport_stp_learn_state(const struct xport
*xport
)
1737 struct stp_port
*sp
= xport_get_stp_port(xport
);
1739 ? stp_learn_in_state(stp_port_get_state(sp
))
1744 xport_stp_forward_state(const struct xport
*xport
)
1746 struct stp_port
*sp
= xport_get_stp_port(xport
);
1748 ? stp_forward_in_state(stp_port_get_state(sp
))
1753 xport_stp_should_forward_bpdu(const struct xport
*xport
)
1755 struct stp_port
*sp
= xport_get_stp_port(xport
);
1756 return stp_should_forward_bpdu(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
1759 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1760 * were used to make the determination.*/
1762 stp_should_process_flow(const struct flow
*flow
, struct flow_wildcards
*wc
)
1764 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1765 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1766 return is_stp(flow
);
1770 stp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1772 struct stp_port
*sp
= xport_get_stp_port(xport
);
1773 struct dp_packet payload
= *packet
;
1774 struct eth_header
*eth
= dp_packet_data(&payload
);
1776 /* Sink packets on ports that have STP disabled when the bridge has
1778 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
1782 /* Trim off padding on payload. */
1783 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1784 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1787 if (dp_packet_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
1788 stp_received_bpdu(sp
, dp_packet_data(&payload
), dp_packet_size(&payload
));
1792 static enum rstp_state
1793 xport_get_rstp_port_state(const struct xport
*xport
)
1795 return xport
->rstp_port
1796 ? rstp_port_get_state(xport
->rstp_port
)
1801 xport_rstp_learn_state(const struct xport
*xport
)
1803 return xport
->xbridge
->rstp
&& xport
->rstp_port
1804 ? rstp_learn_in_state(xport_get_rstp_port_state(xport
))
1809 xport_rstp_forward_state(const struct xport
*xport
)
1811 return xport
->xbridge
->rstp
&& xport
->rstp_port
1812 ? rstp_forward_in_state(xport_get_rstp_port_state(xport
))
1817 xport_rstp_should_manage_bpdu(const struct xport
*xport
)
1819 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport
));
1823 rstp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1825 struct dp_packet payload
= *packet
;
1826 struct eth_header
*eth
= dp_packet_data(&payload
);
1828 /* Sink packets on ports that have no RSTP. */
1829 if (!xport
->rstp_port
) {
1833 /* Trim off padding on payload. */
1834 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1835 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1838 int len
= ETH_HEADER_LEN
+ LLC_HEADER_LEN
;
1839 if (eth
->eth_type
== htons(ETH_TYPE_VLAN
)) {
1840 len
+= VLAN_HEADER_LEN
;
1842 if (dp_packet_try_pull(&payload
, len
)) {
1843 rstp_port_received_bpdu(xport
->rstp_port
, dp_packet_data(&payload
),
1844 dp_packet_size(&payload
));
1848 static struct xport
*
1849 get_ofp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1851 struct xport
*xport
;
1853 HMAP_FOR_EACH_IN_BUCKET (xport
, ofp_node
, hash_ofp_port(ofp_port
),
1855 if (xport
->ofp_port
== ofp_port
) {
1863 ofp_port_to_odp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1865 const struct xport
*xport
= get_ofp_port(xbridge
, ofp_port
);
1866 return xport
? xport
->odp_port
: ODPP_NONE
;
1870 odp_port_is_alive(const struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
1872 struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
1873 return xport
&& xport
->may_enable
;
1876 static struct ofputil_bucket
*
1877 group_first_live_bucket(const struct xlate_ctx
*, const struct group_dpif
*,
1881 group_is_alive(const struct xlate_ctx
*ctx
, uint32_t group_id
, int depth
)
1883 struct group_dpif
*group
;
1885 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
1886 ctx
->xin
->tables_version
, false);
1888 return group_first_live_bucket(ctx
, group
, depth
) != NULL
;
1894 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1897 bucket_is_alive(const struct xlate_ctx
*ctx
,
1898 struct ofputil_bucket
*bucket
, int depth
)
1900 if (depth
>= MAX_LIVENESS_RECURSION
) {
1901 xlate_report_error(ctx
, "bucket chaining exceeded %d links",
1902 MAX_LIVENESS_RECURSION
);
1906 return (!ofputil_bucket_has_liveness(bucket
)
1907 || (bucket
->watch_port
!= OFPP_ANY
1908 && bucket
->watch_port
!= OFPP_CONTROLLER
1909 && odp_port_is_alive(ctx
, bucket
->watch_port
))
1910 || (bucket
->watch_group
!= OFPG_ANY
1911 && group_is_alive(ctx
, bucket
->watch_group
, depth
+ 1))
1912 || (bucket
->watch_port
== OFPP_CONTROLLER
1913 && ofproto_is_alive(&ctx
->xbridge
->ofproto
->up
)));
1917 xlate_report_bucket_not_live(const struct xlate_ctx
*ctx
,
1918 const struct ofputil_bucket
*bucket
)
1920 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
1921 struct ds s
= DS_EMPTY_INITIALIZER
;
1922 if (bucket
->watch_port
!= OFPP_ANY
) {
1923 ds_put_cstr(&s
, "port ");
1924 ofputil_format_port(bucket
->watch_port
, NULL
, &s
);
1926 if (bucket
->watch_group
!= OFPG_ANY
) {
1928 ds_put_cstr(&s
, " and ");
1930 ds_put_format(&s
, "port %"PRIu32
, bucket
->watch_group
);
1933 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": not live due to %s",
1934 bucket
->bucket_id
, ds_cstr(&s
));
1940 static struct ofputil_bucket
*
1941 group_first_live_bucket(const struct xlate_ctx
*ctx
,
1942 const struct group_dpif
*group
, int depth
)
1944 struct ofputil_bucket
*bucket
;
1945 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1946 if (bucket_is_alive(ctx
, bucket
, depth
)) {
1949 xlate_report_bucket_not_live(ctx
, bucket
);
1955 static struct ofputil_bucket
*
1956 group_best_live_bucket(const struct xlate_ctx
*ctx
,
1957 const struct group_dpif
*group
,
1960 struct ofputil_bucket
*best_bucket
= NULL
;
1961 uint32_t best_score
= 0;
1963 struct ofputil_bucket
*bucket
;
1964 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1965 if (bucket_is_alive(ctx
, bucket
, 0)) {
1967 (hash_int(bucket
->bucket_id
, basis
) & 0xffff) * bucket
->weight
;
1968 if (score
>= best_score
) {
1969 best_bucket
= bucket
;
1972 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": score %"PRIu32
,
1973 bucket
->bucket_id
, score
);
1975 xlate_report_bucket_not_live(ctx
, bucket
);
1983 xbundle_trunks_vlan(const struct xbundle
*bundle
, uint16_t vlan
)
1985 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
1986 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
1990 xbundle_allows_cvlan(const struct xbundle
*bundle
, uint16_t vlan
)
1992 return (!bundle
->cvlans
|| bitmap_is_set(bundle
->cvlans
, vlan
));
1996 xbundle_includes_vlan(const struct xbundle
*xbundle
, const struct xvlan
*xvlan
)
1998 switch (xbundle
->vlan_mode
) {
1999 case PORT_VLAN_ACCESS
:
2000 return xvlan
->v
[0].vid
== xbundle
->vlan
&& xvlan
->v
[1].vid
== 0;
2002 case PORT_VLAN_TRUNK
:
2003 case PORT_VLAN_NATIVE_UNTAGGED
:
2004 case PORT_VLAN_NATIVE_TAGGED
:
2005 return xbundle_trunks_vlan(xbundle
, xvlan
->v
[0].vid
);
2007 case PORT_VLAN_DOT1Q_TUNNEL
:
2008 return xvlan
->v
[0].vid
== xbundle
->vlan
&&
2009 xbundle_allows_cvlan(xbundle
, xvlan
->v
[1].vid
);
2016 static mirror_mask_t
2017 xbundle_mirror_out(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2019 return xbundle
!= &ofpp_none_bundle
2020 ? mirror_bundle_out(xbridge
->mbridge
, xbundle
->ofbundle
)
2024 static mirror_mask_t
2025 xbundle_mirror_src(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2027 return xbundle
!= &ofpp_none_bundle
2028 ? mirror_bundle_src(xbridge
->mbridge
, xbundle
->ofbundle
)
2032 static mirror_mask_t
2033 xbundle_mirror_dst(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2035 return xbundle
!= &ofpp_none_bundle
2036 ? mirror_bundle_dst(xbridge
->mbridge
, xbundle
->ofbundle
)
2040 static struct xbundle
*
2041 lookup_input_bundle__(const struct xbridge
*xbridge
,
2042 ofp_port_t in_port
, struct xport
**in_xportp
)
2044 struct xport
*xport
;
2046 /* Find the port and bundle for the received packet. */
2047 xport
= get_ofp_port(xbridge
, in_port
);
2051 if (xport
&& xport
->xbundle
) {
2052 return xport
->xbundle
;
2055 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
2056 * which a controller may use as the ingress port for traffic that
2057 * it is sourcing. */
2058 if (in_port
== OFPP_CONTROLLER
|| in_port
== OFPP_NONE
) {
2059 return &ofpp_none_bundle
;
2064 static struct xbundle
*
2065 lookup_input_bundle(const struct xlate_ctx
*ctx
,
2066 ofp_port_t in_port
, struct xport
**in_xportp
)
2068 struct xbundle
*xbundle
= lookup_input_bundle__(ctx
->xbridge
,
2069 in_port
, in_xportp
);
2071 /* Odd. A few possible reasons here:
2073 * - We deleted a port but there are still a few packets queued up
2076 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
2077 * we don't know about.
2079 * - The ofproto client didn't configure the port as part of a bundle.
2080 * This is particularly likely to happen if a packet was received on
2081 * the port after it was created, but before the client had a chance
2082 * to configure its bundle.
2084 xlate_report_error(ctx
, "received packet on unknown port %"PRIu32
,
2090 /* Mirrors the packet represented by 'ctx' to appropriate mirror destinations,
2091 * given the packet is ingressing or egressing on 'xbundle', which has ingress
2092 * or egress (as appropriate) mirrors 'mirrors'. */
2094 mirror_packet(struct xlate_ctx
*ctx
, struct xbundle
*xbundle
,
2095 mirror_mask_t mirrors
)
2097 struct xvlan in_xvlan
;
2100 /* Figure out what VLAN the packet is in (because mirrors can select
2101 * packets on basis of VLAN). */
2102 xvlan_extract(&ctx
->xin
->flow
, &in_xvlan
);
2103 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, xbundle
)) {
2106 xvlan_input_translate(xbundle
, &in_xvlan
, &xvlan
);
2108 const struct xbridge
*xbridge
= ctx
->xbridge
;
2110 /* Don't mirror to destinations that we've already mirrored to. */
2111 mirrors
&= ~ctx
->mirrors
;
2116 /* 'mirrors' is a bit-mask of candidates for mirroring. Iterate through
2117 * the candidates, adding the ones that really should be mirrored to
2118 * 'used_mirrors', as long as some candidates remain. */
2119 mirror_mask_t used_mirrors
= 0;
2121 const unsigned long *vlans
;
2122 mirror_mask_t dup_mirrors
;
2123 struct ofbundle
*out
;
2127 /* Get the details of the mirror represented by the rightmost 1-bit. */
2128 ovs_assert(mirror_get(xbridge
->mbridge
, raw_ctz(mirrors
),
2129 &vlans
, &dup_mirrors
,
2130 &out
, &snaplen
, &out_vlan
));
2133 /* If this mirror selects on the basis of VLAN, and it does not select
2134 * 'vlan', then discard this mirror and go on to the next one. */
2136 ctx
->wc
->masks
.vlans
[0].tci
|= htons(VLAN_CFI
| VLAN_VID_MASK
);
2138 if (vlans
&& !bitmap_is_set(vlans
, xvlan
.v
[0].vid
)) {
2139 mirrors
= zero_rightmost_1bit(mirrors
);
2143 /* We sent a packet to this mirror. */
2144 used_mirrors
|= rightmost_1bit(mirrors
);
2146 /* Record the mirror, and the mirrors that output to the same
2147 * destination, so that we don't mirror to them again. This must be
2148 * done now to ensure that output_normal(), below, doesn't recursively
2149 * output to the same mirrors. */
2150 ctx
->mirrors
|= dup_mirrors
;
2151 ctx
->mirror_snaplen
= snaplen
;
2153 /* Send the packet to the mirror. */
2155 struct xbundle
*out_xbundle
= xbundle_lookup(ctx
->xcfg
, out
);
2157 output_normal(ctx
, out_xbundle
, &xvlan
);
2159 } else if (xvlan
.v
[0].vid
!= out_vlan
2160 && !eth_addr_is_reserved(ctx
->xin
->flow
.dl_dst
)) {
2162 uint16_t old_vid
= xvlan
.v
[0].vid
;
2164 xvlan
.v
[0].vid
= out_vlan
;
2165 LIST_FOR_EACH (xb
, list_node
, &xbridge
->xbundles
) {
2166 if (xbundle_includes_vlan(xb
, &xvlan
)
2167 && !xbundle_mirror_out(xbridge
, xb
)) {
2168 output_normal(ctx
, xb
, &xvlan
);
2171 xvlan
.v
[0].vid
= old_vid
;
2174 /* output_normal() could have recursively output (to different
2175 * mirrors), so make sure that we don't send duplicates. */
2176 mirrors
&= ~ctx
->mirrors
;
2177 ctx
->mirror_snaplen
= 0;
2181 if (ctx
->xin
->resubmit_stats
) {
2182 mirror_update_stats(xbridge
->mbridge
, used_mirrors
,
2183 ctx
->xin
->resubmit_stats
->n_packets
,
2184 ctx
->xin
->resubmit_stats
->n_bytes
);
2186 if (ctx
->xin
->xcache
) {
2187 struct xc_entry
*entry
;
2189 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_MIRROR
);
2190 entry
->mirror
.mbridge
= mbridge_ref(xbridge
->mbridge
);
2191 entry
->mirror
.mirrors
= used_mirrors
;
2197 mirror_ingress_packet(struct xlate_ctx
*ctx
)
2199 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
)) {
2200 struct xbundle
*xbundle
= lookup_input_bundle(
2201 ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
);
2203 mirror_packet(ctx
, xbundle
,
2204 xbundle_mirror_src(ctx
->xbridge
, xbundle
));
2209 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
2210 * If so, returns true. Otherwise, returns false.
2212 * 'vid' should be the VID obtained from the 802.1Q header that was received as
2213 * part of a packet (specify 0 if there was no 802.1Q header), in the range
2216 input_vid_is_valid(const struct xlate_ctx
*ctx
,
2217 uint16_t vid
, struct xbundle
*in_xbundle
)
2219 /* Allow any VID on the OFPP_NONE port. */
2220 if (in_xbundle
== &ofpp_none_bundle
) {
2224 switch (in_xbundle
->vlan_mode
) {
2225 case PORT_VLAN_ACCESS
:
2227 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" tagged "
2228 "packet received on port %s configured as VLAN "
2229 "%d access port", vid
, in_xbundle
->name
,
2235 case PORT_VLAN_NATIVE_UNTAGGED
:
2236 case PORT_VLAN_NATIVE_TAGGED
:
2238 /* Port must always carry its native VLAN. */
2242 case PORT_VLAN_TRUNK
:
2243 if (!xbundle_trunks_vlan(in_xbundle
, vid
)) {
2244 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet "
2245 "received on port %s not configured for "
2246 "trunking VLAN %"PRIu16
,
2247 vid
, in_xbundle
->name
, vid
);
2252 case PORT_VLAN_DOT1Q_TUNNEL
:
2253 if (!xbundle_allows_cvlan(in_xbundle
, vid
)) {
2254 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet received "
2255 "on dot1q-tunnel port %s that excludes this "
2256 "VLAN", vid
, in_xbundle
->name
);
2268 xvlan_copy(struct xvlan
*dst
, const struct xvlan
*src
)
2274 xvlan_pop(struct xvlan
*src
)
2276 memmove(&src
->v
[0], &src
->v
[1], sizeof(src
->v
) - sizeof(src
->v
[0]));
2277 memset(&src
->v
[FLOW_MAX_VLAN_HEADERS
- 1], 0,
2278 sizeof(src
->v
[FLOW_MAX_VLAN_HEADERS
- 1]));
2282 xvlan_push_uninit(struct xvlan
*src
)
2284 memmove(&src
->v
[1], &src
->v
[0], sizeof(src
->v
) - sizeof(src
->v
[0]));
2285 memset(&src
->v
[0], 0, sizeof(src
->v
[0]));
2288 /* Extract VLAN information (headers) from flow */
2290 xvlan_extract(const struct flow
*flow
, struct xvlan
*xvlan
)
2293 memset(xvlan
, 0, sizeof(*xvlan
));
2294 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2295 if (!eth_type_vlan(flow
->vlans
[i
].tpid
) ||
2296 !(flow
->vlans
[i
].tci
& htons(VLAN_CFI
))) {
2299 xvlan
->v
[i
].tpid
= ntohs(flow
->vlans
[i
].tpid
);
2300 xvlan
->v
[i
].vid
= vlan_tci_to_vid(flow
->vlans
[i
].tci
);
2301 xvlan
->v
[i
].pcp
= ntohs(flow
->vlans
[i
].tci
) & VLAN_PCP_MASK
;
2305 /* Put VLAN information (headers) to flow */
2307 xvlan_put(struct flow
*flow
, const struct xvlan
*xvlan
,
2308 enum port_priority_tags_mode use_priority_tags
)
2312 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2313 tci
= htons(xvlan
->v
[i
].vid
| (xvlan
->v
[i
].pcp
& VLAN_PCP_MASK
));
2314 if (tci
|| ((use_priority_tags
== PORT_PRIORITY_TAGS_ALWAYS
) &&
2315 xvlan
->v
[i
].tpid
)) {
2316 tci
|= htons(VLAN_CFI
);
2317 flow
->vlans
[i
].tpid
= xvlan
->v
[i
].tpid
?
2318 htons(xvlan
->v
[i
].tpid
) :
2319 htons(ETH_TYPE_VLAN_8021Q
);
2321 flow
->vlans
[i
].tci
= tci
;
2325 /* Given 'in_xvlan', extracted from the input 802.1Q headers received as part
2326 * of a packet, and 'in_xbundle', the bundle on which the packet was received,
2327 * returns the VLANs of the packet during bridge internal processing. */
2329 xvlan_input_translate(const struct xbundle
*in_xbundle
,
2330 const struct xvlan
*in_xvlan
, struct xvlan
*xvlan
)
2333 switch (in_xbundle
->vlan_mode
) {
2334 case PORT_VLAN_ACCESS
:
2335 memset(xvlan
, 0, sizeof(*xvlan
));
2336 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2337 ETH_TYPE_VLAN_8021Q
;
2338 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2339 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2342 case PORT_VLAN_TRUNK
:
2343 xvlan_copy(xvlan
, in_xvlan
);
2346 case PORT_VLAN_NATIVE_UNTAGGED
:
2347 case PORT_VLAN_NATIVE_TAGGED
:
2348 xvlan_copy(xvlan
, in_xvlan
);
2349 if (!in_xvlan
->v
[0].vid
) {
2350 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2351 ETH_TYPE_VLAN_8021Q
;
2352 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2353 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2357 case PORT_VLAN_DOT1Q_TUNNEL
:
2358 xvlan_copy(xvlan
, in_xvlan
);
2359 xvlan_push_uninit(xvlan
);
2360 xvlan
->v
[0].tpid
= in_xbundle
->qinq_ethtype
;
2361 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2362 xvlan
->v
[0].pcp
= 0;
2370 /* Given 'xvlan', the VLANs of a packet during internal processing, and
2371 * 'out_xbundle', a bundle on which the packet is to be output, returns the
2372 * VLANs that should be included in output packet. */
2374 xvlan_output_translate(const struct xbundle
*out_xbundle
,
2375 const struct xvlan
*xvlan
, struct xvlan
*out_xvlan
)
2377 switch (out_xbundle
->vlan_mode
) {
2378 case PORT_VLAN_ACCESS
:
2379 memset(out_xvlan
, 0, sizeof(*out_xvlan
));
2382 case PORT_VLAN_TRUNK
:
2383 case PORT_VLAN_NATIVE_TAGGED
:
2384 xvlan_copy(out_xvlan
, xvlan
);
2387 case PORT_VLAN_NATIVE_UNTAGGED
:
2388 xvlan_copy(out_xvlan
, xvlan
);
2389 if (xvlan
->v
[0].vid
== out_xbundle
->vlan
) {
2390 xvlan_pop(out_xvlan
);
2394 case PORT_VLAN_DOT1Q_TUNNEL
:
2395 xvlan_copy(out_xvlan
, xvlan
);
2396 xvlan_pop(out_xvlan
);
2404 /* If output xbundle is dot1q-tunnel, set mask bits of cvlan */
2406 check_and_set_cvlan_mask(struct flow_wildcards
*wc
,
2407 const struct xbundle
*xbundle
)
2409 if (xbundle
->vlan_mode
== PORT_VLAN_DOT1Q_TUNNEL
&& xbundle
->cvlans
) {
2410 wc
->masks
.vlans
[1].tci
= htons(0xffff);
2415 output_normal(struct xlate_ctx
*ctx
, const struct xbundle
*out_xbundle
,
2416 const struct xvlan
*xvlan
)
2419 union flow_vlan_hdr old_vlans
[FLOW_MAX_VLAN_HEADERS
];
2420 struct xport
*xport
;
2421 struct xlate_bond_recirc xr
;
2422 bool use_recirc
= false;
2423 struct xvlan out_xvlan
;
2425 check_and_set_cvlan_mask(ctx
->wc
, out_xbundle
);
2427 xvlan_output_translate(out_xbundle
, xvlan
, &out_xvlan
);
2428 if (out_xbundle
->use_priority_tags
) {
2429 out_xvlan
.v
[0].pcp
= ntohs(ctx
->xin
->flow
.vlans
[0].tci
) &
2432 vid
= out_xvlan
.v
[0].vid
;
2433 if (ovs_list_is_empty(&out_xbundle
->xports
)) {
2434 /* Partially configured bundle with no slaves. Drop the packet. */
2436 } else if (!out_xbundle
->bond
) {
2437 xport
= CONTAINER_OF(ovs_list_front(&out_xbundle
->xports
), struct xport
,
2440 struct flow_wildcards
*wc
= ctx
->wc
;
2441 struct ofport_dpif
*ofport
;
2443 if (ctx
->xbridge
->support
.odp
.recirc
) {
2444 /* In case recirculation is not actually in use, 'xr.recirc_id'
2445 * will be set to '0', since a valid 'recirc_id' can
2447 bond_update_post_recirc_rules(out_xbundle
->bond
,
2451 /* Use recirculation instead of output. */
2453 xr
.hash_alg
= OVS_HASH_ALG_L4
;
2454 /* Recirculation does not require unmasking hash fields. */
2459 ofport
= bond_choose_output_slave(out_xbundle
->bond
,
2460 &ctx
->xin
->flow
, wc
, vid
);
2461 xport
= xport_lookup(ctx
->xcfg
, ofport
);
2464 /* No slaves enabled, so drop packet. */
2468 /* If use_recirc is set, the main thread will handle stats
2469 * accounting for this bond. */
2471 if (ctx
->xin
->resubmit_stats
) {
2472 bond_account(out_xbundle
->bond
, &ctx
->xin
->flow
, vid
,
2473 ctx
->xin
->resubmit_stats
->n_bytes
);
2475 if (ctx
->xin
->xcache
) {
2476 struct xc_entry
*entry
;
2479 flow
= &ctx
->xin
->flow
;
2480 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_BOND
);
2481 entry
->bond
.bond
= bond_ref(out_xbundle
->bond
);
2482 entry
->bond
.flow
= xmemdup(flow
, sizeof *flow
);
2483 entry
->bond
.vid
= vid
;
2488 memcpy(&old_vlans
, &ctx
->xin
->flow
.vlans
, sizeof(old_vlans
));
2489 xvlan_put(&ctx
->xin
->flow
, &out_xvlan
, out_xbundle
->use_priority_tags
);
2491 compose_output_action(ctx
, xport
->ofp_port
, use_recirc
? &xr
: NULL
,
2493 memcpy(&ctx
->xin
->flow
.vlans
, &old_vlans
, sizeof(old_vlans
));
2496 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2497 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2498 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2500 is_gratuitous_arp(const struct flow
*flow
, struct flow_wildcards
*wc
)
2502 if (flow
->dl_type
!= htons(ETH_TYPE_ARP
)) {
2506 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2507 if (!eth_addr_is_broadcast(flow
->dl_dst
)) {
2511 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2512 if (flow
->nw_proto
== ARP_OP_REPLY
) {
2514 } else if (flow
->nw_proto
== ARP_OP_REQUEST
) {
2515 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
2516 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2518 return flow
->nw_src
== flow
->nw_dst
;
2524 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
2525 * dropped. Returns true if they may be forwarded, false if they should be
2528 * 'in_port' must be the xport that corresponds to flow->in_port.
2529 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
2531 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
2532 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
2533 * checked by input_vid_is_valid().
2535 * May also add tags to '*tags', although the current implementation only does
2536 * so in one special case.
2539 is_admissible(struct xlate_ctx
*ctx
, struct xport
*in_port
,
2542 struct xbundle
*in_xbundle
= in_port
->xbundle
;
2543 const struct xbridge
*xbridge
= ctx
->xbridge
;
2544 struct flow
*flow
= &ctx
->xin
->flow
;
2546 /* Drop frames for reserved multicast addresses
2547 * only if forward_bpdu option is absent. */
2548 if (!xbridge
->forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
2549 xlate_report(ctx
, OFT_DETAIL
,
2550 "packet has reserved destination MAC, dropping");
2554 if (in_xbundle
->bond
) {
2555 struct mac_entry
*mac
;
2557 switch (bond_check_admissibility(in_xbundle
->bond
, in_port
->ofport
,
2563 xlate_report(ctx
, OFT_DETAIL
,
2564 "bonding refused admissibility, dropping");
2567 case BV_DROP_IF_MOVED
:
2568 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
2569 mac
= mac_learning_lookup(xbridge
->ml
, flow
->dl_src
, vlan
);
2571 && mac_entry_get_port(xbridge
->ml
, mac
) != in_xbundle
->ofbundle
2572 && (!is_gratuitous_arp(flow
, ctx
->wc
)
2573 || mac_entry_is_grat_arp_locked(mac
))) {
2574 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2575 xlate_report(ctx
, OFT_DETAIL
,
2576 "SLB bond thinks this packet looped back, "
2580 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2589 update_learning_table__(const struct xbridge
*xbridge
,
2590 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2591 int vlan
, bool is_grat_arp
)
2593 return (in_xbundle
== &ofpp_none_bundle
2594 || !mac_learning_update(xbridge
->ml
, dl_src
, vlan
,
2596 in_xbundle
->bond
!= NULL
,
2597 in_xbundle
->ofbundle
));
2601 update_learning_table(const struct xlate_ctx
*ctx
,
2602 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2603 int vlan
, bool is_grat_arp
)
2605 if (!update_learning_table__(ctx
->xbridge
, in_xbundle
, dl_src
, vlan
,
2607 xlate_report_debug(ctx
, OFT_DETAIL
, "learned that "ETH_ADDR_FMT
" is "
2608 "on port %s in VLAN %d",
2609 ETH_ADDR_ARGS(dl_src
), in_xbundle
->name
, vlan
);
2613 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2614 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2616 update_mcast_snooping_table4__(const struct xlate_ctx
*ctx
,
2617 const struct flow
*flow
,
2618 struct mcast_snooping
*ms
, int vlan
,
2619 struct xbundle
*in_xbundle
,
2620 const struct dp_packet
*packet
)
2621 OVS_REQ_WRLOCK(ms
->rwlock
)
2623 const struct igmp_header
*igmp
;
2626 ovs_be32 ip4
= flow
->igmp_group_ip4
;
2628 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2629 igmp
= dp_packet_at(packet
, offset
, IGMP_HEADER_LEN
);
2630 if (!igmp
|| csum(igmp
, dp_packet_l4_size(packet
)) != 0) {
2631 xlate_report_debug(ctx
, OFT_DETAIL
,
2632 "multicast snooping received bad IGMP "
2633 "checksum on port %s in VLAN %d",
2634 in_xbundle
->name
, vlan
);
2638 switch (ntohs(flow
->tp_src
)) {
2639 case IGMP_HOST_MEMBERSHIP_REPORT
:
2640 case IGMPV2_HOST_MEMBERSHIP_REPORT
:
2641 if (mcast_snooping_add_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2642 xlate_report_debug(ctx
, OFT_DETAIL
,
2643 "multicast snooping learned that "
2644 IP_FMT
" is on port %s in VLAN %d",
2645 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2648 case IGMP_HOST_LEAVE_MESSAGE
:
2649 if (mcast_snooping_leave_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2650 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping leaving "
2651 IP_FMT
" is on port %s in VLAN %d",
2652 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2655 case IGMP_HOST_MEMBERSHIP_QUERY
:
2656 if (flow
->nw_src
&& mcast_snooping_add_mrouter(ms
, vlan
,
2657 in_xbundle
->ofbundle
)) {
2658 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query "
2659 "from "IP_FMT
" is on port %s in VLAN %d",
2660 IP_ARGS(flow
->nw_src
), in_xbundle
->name
, vlan
);
2663 case IGMPV3_HOST_MEMBERSHIP_REPORT
:
2664 count
= mcast_snooping_add_report(ms
, packet
, vlan
,
2665 in_xbundle
->ofbundle
);
2667 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2668 "%d addresses on port %s in VLAN %d",
2669 count
, in_xbundle
->name
, vlan
);
2676 update_mcast_snooping_table6__(const struct xlate_ctx
*ctx
,
2677 const struct flow
*flow
,
2678 struct mcast_snooping
*ms
, int vlan
,
2679 struct xbundle
*in_xbundle
,
2680 const struct dp_packet
*packet
)
2681 OVS_REQ_WRLOCK(ms
->rwlock
)
2683 const struct mld_header
*mld
;
2687 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2688 mld
= dp_packet_at(packet
, offset
, MLD_HEADER_LEN
);
2691 packet_csum_upperlayer6(dp_packet_l3(packet
),
2692 mld
, IPPROTO_ICMPV6
,
2693 dp_packet_l4_size(packet
)) != 0) {
2694 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping received "
2695 "bad MLD checksum on port %s in VLAN %d",
2696 in_xbundle
->name
, vlan
);
2700 switch (ntohs(flow
->tp_src
)) {
2702 if (!ipv6_addr_equals(&flow
->ipv6_src
, &in6addr_any
)
2703 && mcast_snooping_add_mrouter(ms
, vlan
, in_xbundle
->ofbundle
)) {
2704 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query on "
2705 "port %s in VLAN %d", in_xbundle
->name
, vlan
);
2711 count
= mcast_snooping_add_mld(ms
, packet
, vlan
, in_xbundle
->ofbundle
);
2713 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2714 "%d addresses on port %s in VLAN %d",
2715 count
, in_xbundle
->name
, vlan
);
2721 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2722 * was received on 'in_xbundle' in 'vlan'. */
2724 update_mcast_snooping_table(const struct xlate_ctx
*ctx
,
2725 const struct flow
*flow
, int vlan
,
2726 struct xbundle
*in_xbundle
,
2727 const struct dp_packet
*packet
)
2729 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
2730 struct xbundle
*mcast_xbundle
;
2731 struct mcast_port_bundle
*fport
;
2733 /* Don't learn the OFPP_NONE port. */
2734 if (in_xbundle
== &ofpp_none_bundle
) {
2738 /* Don't learn from flood ports */
2739 mcast_xbundle
= NULL
;
2740 ovs_rwlock_wrlock(&ms
->rwlock
);
2741 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2742 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2743 if (mcast_xbundle
== in_xbundle
) {
2748 if (!mcast_xbundle
|| mcast_xbundle
!= in_xbundle
) {
2749 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2750 update_mcast_snooping_table4__(ctx
, flow
, ms
, vlan
,
2751 in_xbundle
, packet
);
2753 update_mcast_snooping_table6__(ctx
, flow
, ms
, vlan
,
2754 in_xbundle
, packet
);
2757 ovs_rwlock_unlock(&ms
->rwlock
);
2760 /* A list of multicast output ports.
2762 * We accumulate output ports and then do all the outputs afterward. It would
2763 * be more natural to do the outputs one at a time as we discover the need for
2764 * each one, but this can cause a deadlock because we need to take the
2765 * mcast_snooping's rwlock for reading to iterate through the port lists and
2766 * doing an output, if it goes to a patch port, can eventually come back to the
2767 * same mcast_snooping and attempt to take the write lock (see
2768 * https://github.com/openvswitch/ovs-issues/issues/153). */
2769 struct mcast_output
{
2770 /* Discrete ports. */
2771 struct xbundle
**xbundles
;
2772 size_t n
, allocated
;
2774 /* If set, flood to all ports. */
2777 #define MCAST_OUTPUT_INIT { NULL, 0, 0, false }
2779 /* Add 'mcast_bundle' to 'out'. */
2781 mcast_output_add(struct mcast_output
*out
, struct xbundle
*mcast_xbundle
)
2783 if (out
->n
>= out
->allocated
) {
2784 out
->xbundles
= x2nrealloc(out
->xbundles
, &out
->allocated
,
2785 sizeof *out
->xbundles
);
2787 out
->xbundles
[out
->n
++] = mcast_xbundle
;
2790 /* Outputs the packet in 'ctx' to all of the output ports in 'out', given input
2791 * bundle 'in_xbundle' and the current 'xvlan'. */
2793 mcast_output_finish(struct xlate_ctx
*ctx
, struct mcast_output
*out
,
2794 struct xbundle
*in_xbundle
, struct xvlan
*xvlan
)
2797 xlate_normal_flood(ctx
, in_xbundle
, xvlan
);
2799 for (size_t i
= 0; i
< out
->n
; i
++) {
2800 output_normal(ctx
, out
->xbundles
[i
], xvlan
);
2804 free(out
->xbundles
);
2807 /* send the packet to ports having the multicast group learned */
2809 xlate_normal_mcast_send_group(struct xlate_ctx
*ctx
,
2810 struct mcast_snooping
*ms OVS_UNUSED
,
2811 struct mcast_group
*grp
,
2812 struct xbundle
*in_xbundle
,
2813 struct mcast_output
*out
)
2814 OVS_REQ_RDLOCK(ms
->rwlock
)
2816 struct mcast_group_bundle
*b
;
2817 struct xbundle
*mcast_xbundle
;
2819 LIST_FOR_EACH(b
, bundle_node
, &grp
->bundle_lru
) {
2820 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, b
->port
);
2821 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2822 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast group port");
2823 mcast_output_add(out
, mcast_xbundle
);
2824 } else if (!mcast_xbundle
) {
2825 xlate_report(ctx
, OFT_WARN
,
2826 "mcast group port is unknown, dropping");
2828 xlate_report(ctx
, OFT_DETAIL
,
2829 "mcast group port is input port, dropping");
2834 /* send the packet to ports connected to multicast routers */
2836 xlate_normal_mcast_send_mrouters(struct xlate_ctx
*ctx
,
2837 struct mcast_snooping
*ms
,
2838 struct xbundle
*in_xbundle
,
2839 const struct xvlan
*xvlan
,
2840 struct mcast_output
*out
)
2841 OVS_REQ_RDLOCK(ms
->rwlock
)
2843 struct mcast_mrouter_bundle
*mrouter
;
2844 struct xbundle
*mcast_xbundle
;
2846 LIST_FOR_EACH(mrouter
, mrouter_node
, &ms
->mrouter_lru
) {
2847 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, mrouter
->port
);
2848 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
2849 && mrouter
->vlan
== xvlan
->v
[0].vid
) {
2850 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast router port");
2851 mcast_output_add(out
, mcast_xbundle
);
2852 } else if (!mcast_xbundle
) {
2853 xlate_report(ctx
, OFT_WARN
,
2854 "mcast router port is unknown, dropping");
2855 } else if (mrouter
->vlan
!= xvlan
->v
[0].vid
) {
2856 xlate_report(ctx
, OFT_DETAIL
,
2857 "mcast router is on another vlan, dropping");
2859 xlate_report(ctx
, OFT_DETAIL
,
2860 "mcast router port is input port, dropping");
2865 /* send the packet to ports flagged to be flooded */
2867 xlate_normal_mcast_send_fports(struct xlate_ctx
*ctx
,
2868 struct mcast_snooping
*ms
,
2869 struct xbundle
*in_xbundle
,
2870 struct mcast_output
*out
)
2871 OVS_REQ_RDLOCK(ms
->rwlock
)
2873 struct mcast_port_bundle
*fport
;
2874 struct xbundle
*mcast_xbundle
;
2876 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2877 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2878 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2879 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast flood port");
2880 mcast_output_add(out
, mcast_xbundle
);
2881 } else if (!mcast_xbundle
) {
2882 xlate_report(ctx
, OFT_WARN
,
2883 "mcast flood port is unknown, dropping");
2885 xlate_report(ctx
, OFT_DETAIL
,
2886 "mcast flood port is input port, dropping");
2891 /* forward the Reports to configured ports */
2893 xlate_normal_mcast_send_rports(struct xlate_ctx
*ctx
,
2894 struct mcast_snooping
*ms
,
2895 struct xbundle
*in_xbundle
,
2896 struct mcast_output
*out
)
2897 OVS_REQ_RDLOCK(ms
->rwlock
)
2899 struct mcast_port_bundle
*rport
;
2900 struct xbundle
*mcast_xbundle
;
2902 LIST_FOR_EACH(rport
, node
, &ms
->rport_list
) {
2903 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, rport
->port
);
2905 && mcast_xbundle
!= in_xbundle
2906 && mcast_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
2907 xlate_report(ctx
, OFT_DETAIL
,
2908 "forwarding report to mcast flagged port");
2909 mcast_output_add(out
, mcast_xbundle
);
2910 } else if (!mcast_xbundle
) {
2911 xlate_report(ctx
, OFT_WARN
,
2912 "mcast port is unknown, dropping the report");
2914 xlate_report(ctx
, OFT_DETAIL
,
2915 "mcast port is input port, dropping the Report");
2921 xlate_normal_flood(struct xlate_ctx
*ctx
, struct xbundle
*in_xbundle
,
2922 struct xvlan
*xvlan
)
2924 struct xbundle
*xbundle
;
2926 LIST_FOR_EACH (xbundle
, list_node
, &ctx
->xbridge
->xbundles
) {
2927 if (xbundle
!= in_xbundle
2928 && xbundle
->ofbundle
!= in_xbundle
->ofbundle
2929 && xbundle_includes_vlan(xbundle
, xvlan
)
2930 && xbundle
->floodable
2931 && !xbundle_mirror_out(ctx
->xbridge
, xbundle
)) {
2932 output_normal(ctx
, xbundle
, xvlan
);
2935 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2939 is_ip_local_multicast(const struct flow
*flow
, struct flow_wildcards
*wc
)
2941 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2942 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2943 return ip_is_local_multicast(flow
->nw_dst
);
2944 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2945 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
2946 return ipv6_is_all_hosts(&flow
->ipv6_dst
);
2953 xlate_normal(struct xlate_ctx
*ctx
)
2955 struct flow_wildcards
*wc
= ctx
->wc
;
2956 struct flow
*flow
= &ctx
->xin
->flow
;
2957 struct xbundle
*in_xbundle
;
2958 struct xport
*in_port
;
2959 struct mac_entry
*mac
;
2961 struct xvlan in_xvlan
;
2965 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
2966 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2967 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
2969 in_xbundle
= lookup_input_bundle(ctx
, flow
->in_port
.ofp_port
, &in_port
);
2971 xlate_report(ctx
, OFT_WARN
, "no input bundle, dropping");
2975 /* Drop malformed frames. */
2976 if (eth_type_vlan(flow
->dl_type
) &&
2977 !(flow
->vlans
[0].tci
& htons(VLAN_CFI
))) {
2978 if (ctx
->xin
->packet
!= NULL
) {
2979 xlate_report_error(ctx
, "dropping packet with partial "
2980 "VLAN tag received on port %s",
2983 xlate_report(ctx
, OFT_WARN
, "partial VLAN tag, dropping");
2987 /* Drop frames on bundles reserved for mirroring. */
2988 if (xbundle_mirror_out(ctx
->xbridge
, in_xbundle
)) {
2989 if (ctx
->xin
->packet
!= NULL
) {
2990 xlate_report_error(ctx
, "dropping packet received on port %s, "
2991 "which is reserved exclusively for mirroring",
2994 xlate_report(ctx
, OFT_WARN
,
2995 "input port is mirror output port, dropping");
3000 xvlan_extract(flow
, &in_xvlan
);
3001 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, in_xbundle
)) {
3002 xlate_report(ctx
, OFT_WARN
,
3003 "disallowed VLAN VID for this input port, dropping");
3006 xvlan_input_translate(in_xbundle
, &in_xvlan
, &xvlan
);
3007 vlan
= xvlan
.v
[0].vid
;
3009 /* Check other admissibility requirements. */
3010 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
3014 /* Learn source MAC. */
3015 bool is_grat_arp
= is_gratuitous_arp(flow
, wc
);
3016 if (ctx
->xin
->allow_side_effects
3017 && flow
->packet_type
== htonl(PT_ETH
)
3018 && in_port
->pt_mode
!= NETDEV_PT_LEGACY_L3
3020 update_learning_table(ctx
, in_xbundle
, flow
->dl_src
, vlan
,
3023 if (ctx
->xin
->xcache
&& in_xbundle
!= &ofpp_none_bundle
) {
3024 struct xc_entry
*entry
;
3026 /* Save just enough info to update mac learning table later. */
3027 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NORMAL
);
3028 entry
->normal
.ofproto
= ctx
->xbridge
->ofproto
;
3029 entry
->normal
.in_port
= flow
->in_port
.ofp_port
;
3030 entry
->normal
.dl_src
= flow
->dl_src
;
3031 entry
->normal
.vlan
= vlan
;
3032 entry
->normal
.is_gratuitous_arp
= is_grat_arp
;
3035 /* Determine output bundle. */
3036 if (mcast_snooping_enabled(ctx
->xbridge
->ms
)
3037 && !eth_addr_is_broadcast(flow
->dl_dst
)
3038 && eth_addr_is_multicast(flow
->dl_dst
)
3039 && is_ip_any(flow
)) {
3040 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
3041 struct mcast_group
*grp
= NULL
;
3043 if (is_igmp(flow
, wc
)) {
3045 * IGMP packets need to take the slow path, in order to be
3046 * processed for mdb updates. That will prevent expires
3047 * firing off even after hosts have sent reports.
3049 ctx
->xout
->slow
|= SLOW_ACTION
;
3051 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
3052 if (mcast_snooping_is_membership(flow
->tp_src
) ||
3053 mcast_snooping_is_query(flow
->tp_src
)) {
3054 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3055 update_mcast_snooping_table(ctx
, flow
, vlan
,
3056 in_xbundle
, ctx
->xin
->packet
);
3060 if (mcast_snooping_is_membership(flow
->tp_src
)) {
3061 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3063 ovs_rwlock_rdlock(&ms
->rwlock
);
3064 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3066 /* RFC4541: section 2.1.1, item 1: A snooping switch should
3067 * forward IGMP Membership Reports only to those ports where
3068 * multicast routers are attached. Alternatively stated: a
3069 * snooping switch should not forward IGMP Membership Reports
3070 * to ports on which only hosts are attached.
3071 * An administrative control may be provided to override this
3072 * restriction, allowing the report messages to be flooded to
3074 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3075 ovs_rwlock_unlock(&ms
->rwlock
);
3077 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3079 xlate_report(ctx
, OFT_DETAIL
, "multicast traffic, flooding");
3080 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3083 } else if (is_mld(flow
, wc
)) {
3084 ctx
->xout
->slow
|= SLOW_ACTION
;
3085 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3086 update_mcast_snooping_table(ctx
, flow
, vlan
,
3087 in_xbundle
, ctx
->xin
->packet
);
3089 if (is_mld_report(flow
, wc
)) {
3090 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3092 ovs_rwlock_rdlock(&ms
->rwlock
);
3093 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3095 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3096 ovs_rwlock_unlock(&ms
->rwlock
);
3098 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3100 xlate_report(ctx
, OFT_DETAIL
, "MLD query, flooding");
3101 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3104 if (is_ip_local_multicast(flow
, wc
)) {
3105 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
3106 * address in the 224.0.0.x range which are not IGMP must
3107 * be forwarded on all ports */
3108 xlate_report(ctx
, OFT_DETAIL
,
3109 "RFC4541: section 2.1.2, item 2, flooding");
3110 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3115 /* forwarding to group base ports */
3116 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3118 ovs_rwlock_rdlock(&ms
->rwlock
);
3119 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3120 grp
= mcast_snooping_lookup4(ms
, flow
->nw_dst
, vlan
);
3121 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3122 grp
= mcast_snooping_lookup(ms
, &flow
->ipv6_dst
, vlan
);
3125 xlate_normal_mcast_send_group(ctx
, ms
, grp
, in_xbundle
, &out
);
3126 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3127 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3130 if (mcast_snooping_flood_unreg(ms
)) {
3131 xlate_report(ctx
, OFT_DETAIL
,
3132 "unregistered multicast, flooding");
3135 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3137 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3140 ovs_rwlock_unlock(&ms
->rwlock
);
3142 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3144 ovs_rwlock_rdlock(&ctx
->xbridge
->ml
->rwlock
);
3145 mac
= mac_learning_lookup(ctx
->xbridge
->ml
, flow
->dl_dst
, vlan
);
3146 mac_port
= mac
? mac_entry_get_port(ctx
->xbridge
->ml
, mac
) : NULL
;
3147 ovs_rwlock_unlock(&ctx
->xbridge
->ml
->rwlock
);
3150 struct xbundle
*mac_xbundle
= xbundle_lookup(ctx
->xcfg
, mac_port
);
3152 if (mac_xbundle
&& xbundle_mirror_out(ctx
->xbridge
, mac_xbundle
)) {
3153 xlate_report(ctx
, OFT_WARN
,
3154 "learned port is a mirror port, dropping");
3159 && mac_xbundle
!= in_xbundle
3160 && mac_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
3161 xlate_report(ctx
, OFT_DETAIL
, "forwarding to learned port");
3162 output_normal(ctx
, mac_xbundle
, &xvlan
);
3163 } else if (!mac_xbundle
) {
3164 xlate_report(ctx
, OFT_WARN
,
3165 "learned port is unknown, dropping");
3167 xlate_report(ctx
, OFT_DETAIL
,
3168 "learned port is input port, dropping");
3171 xlate_report(ctx
, OFT_DETAIL
,
3172 "no learned MAC for destination, flooding");
3173 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3178 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
3179 * 'probability' is the number of packets out of UINT32_MAX to sample. The
3180 * 'cookie' is passed back in the callback for each sampled packet.
3181 * 'tunnel_out_port', if not ODPP_NONE, is added as the
3182 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions',
3183 * an OVS_USERSPACE_ATTR_ACTIONS attribute is added. If
3184 * 'emit_set_tunnel', sample(sampling_port=1) would translate into
3185 * datapath sample action set(tunnel(...)), sample(...) and it is used
3186 * for sampling egress tunnel information.
3189 compose_sample_action(struct xlate_ctx
*ctx
,
3190 const uint32_t probability
,
3191 const struct user_action_cookie
*cookie
,
3192 const odp_port_t tunnel_out_port
,
3193 bool include_actions
)
3195 if (probability
== 0) {
3196 /* No need to generate sampling or the inner action. */
3200 /* If the slow path meter is configured by the controller,
3201 * insert a meter action before the user space action. */
3202 struct ofproto
*ofproto
= &ctx
->xin
->ofproto
->up
;
3203 uint32_t meter_id
= ofproto
->slowpath_meter_id
;
3205 /* When meter action is not required, avoid generate sample action
3206 * for 100% sampling rate. */
3207 bool is_sample
= probability
< UINT32_MAX
|| meter_id
!= UINT32_MAX
;
3208 size_t sample_offset
= 0, actions_offset
= 0;
3210 sample_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3211 OVS_ACTION_ATTR_SAMPLE
);
3212 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
3214 actions_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3215 OVS_SAMPLE_ATTR_ACTIONS
);
3218 if (meter_id
!= UINT32_MAX
) {
3219 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
3222 odp_port_t odp_port
= ofp_port_to_odp_port(
3223 ctx
->xbridge
, ctx
->xin
->flow
.in_port
.ofp_port
);
3224 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
3225 size_t cookie_offset
= odp_put_userspace_action(pid
, cookie
,
3232 nl_msg_end_nested(ctx
->odp_actions
, actions_offset
);
3233 nl_msg_end_nested(ctx
->odp_actions
, sample_offset
);
3236 return cookie_offset
;
3239 /* If sFLow is not enabled, returns 0 without doing anything.
3241 * If sFlow is enabled, appends a template "sample" action to the ODP actions
3242 * in 'ctx'. This action is a template because some of the information needed
3243 * to fill it out is not available until flow translation is complete. In this
3244 * case, this functions returns an offset, which is always nonzero, to pass
3245 * later to fix_sflow_action() to fill in the rest of the template. */
3247 compose_sflow_action(struct xlate_ctx
*ctx
)
3249 struct dpif_sflow
*sflow
= ctx
->xbridge
->sflow
;
3250 if (!sflow
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3254 struct user_action_cookie cookie
;
3256 memset(&cookie
, 0, sizeof cookie
);
3257 cookie
.type
= USER_ACTION_COOKIE_SFLOW
;
3258 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
3259 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
3261 return compose_sample_action(ctx
, dpif_sflow_get_probability(sflow
),
3262 &cookie
, ODPP_NONE
, true);
3265 /* If flow IPFIX is enabled, make sure IPFIX flow sample action
3266 * at egress point of tunnel port is just in front of corresponding
3267 * output action. If bridge IPFIX is enabled, this appends an IPFIX
3268 * sample action to 'ctx->odp_actions'. */
3270 compose_ipfix_action(struct xlate_ctx
*ctx
, odp_port_t output_odp_port
)
3272 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
3273 odp_port_t tunnel_out_port
= ODPP_NONE
;
3275 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3279 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
3281 if (output_odp_port
== ODPP_NONE
&&
3282 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix
)) {
3286 /* For output case, output_odp_port is valid. */
3287 if (output_odp_port
!= ODPP_NONE
) {
3288 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix
)) {
3291 /* If tunnel sampling is enabled, put an additional option attribute:
3292 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
3294 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix
) &&
3295 dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
) ) {
3296 tunnel_out_port
= output_odp_port
;
3300 struct user_action_cookie cookie
;
3302 memset(&cookie
, 0, sizeof cookie
);
3303 cookie
.type
= USER_ACTION_COOKIE_IPFIX
;
3304 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
3305 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
3306 cookie
.ipfix
.output_odp_port
= output_odp_port
;
3308 compose_sample_action(ctx
,
3309 dpif_ipfix_get_bridge_exporter_probability(ipfix
),
3310 &cookie
, tunnel_out_port
, false);
3313 /* Fix "sample" action according to data collected while composing ODP actions,
3314 * as described in compose_sflow_action().
3316 * 'user_cookie_offset' must be the offset returned by
3317 * compose_sflow_action(). */
3319 fix_sflow_action(struct xlate_ctx
*ctx
, unsigned int user_cookie_offset
)
3321 const struct flow
*base
= &ctx
->base_flow
;
3322 struct user_action_cookie
*cookie
;
3324 cookie
= ofpbuf_at(ctx
->odp_actions
, user_cookie_offset
, sizeof *cookie
);
3325 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
3327 cookie
->sflow
.vlan_tci
= base
->vlans
[0].tci
;
3329 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
3330 * port information") for the interpretation of cookie->output. */
3331 switch (ctx
->sflow_n_outputs
) {
3333 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
3334 cookie
->sflow
.output
= 0x40000000 | 256;
3338 cookie
->sflow
.output
= dpif_sflow_odp_port_to_ifindex(
3339 ctx
->xbridge
->sflow
, ctx
->sflow_odp_port
);
3340 if (cookie
->sflow
.output
) {
3345 /* 0x80000000 means "multiple output ports. */
3346 cookie
->sflow
.output
= 0x80000000 | ctx
->sflow_n_outputs
;
3352 process_special(struct xlate_ctx
*ctx
, const struct xport
*xport
)
3354 const struct flow
*flow
= &ctx
->xin
->flow
;
3355 struct flow_wildcards
*wc
= ctx
->wc
;
3356 const struct xbridge
*xbridge
= ctx
->xbridge
;
3357 const struct dp_packet
*packet
= ctx
->xin
->packet
;
3358 enum slow_path_reason slow
;
3359 bool lacp_may_enable
;
3363 } else if (xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
)) {
3365 cfm_process_heartbeat(xport
->cfm
, packet
);
3368 } else if (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
, wc
)) {
3370 bfd_process_packet(xport
->bfd
, flow
, packet
);
3371 /* If POLL received, immediately sends FINAL back. */
3372 if (bfd_should_send_packet(xport
->bfd
)) {
3373 ofproto_dpif_monitor_port_send_soon(xport
->ofport
);
3377 } else if (xport
->xbundle
&& xport
->xbundle
->lacp
3378 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
3380 lacp_may_enable
= lacp_process_packet(xport
->xbundle
->lacp
,
3381 xport
->ofport
, packet
);
3382 /* Update LACP status in bond-slave to avoid packet-drops until
3383 * LACP state machine is run by the main thread. */
3384 if (xport
->xbundle
->bond
&& lacp_may_enable
) {
3385 bond_slave_set_may_enable(xport
->xbundle
->bond
, xport
->ofport
,
3390 } else if ((xbridge
->stp
|| xbridge
->rstp
) &&
3391 stp_should_process_flow(flow
, wc
)) {
3394 ? stp_process_packet(xport
, packet
)
3395 : rstp_process_packet(xport
, packet
);
3398 } else if (xport
->lldp
&& lldp_should_process_flow(xport
->lldp
, flow
)) {
3400 lldp_process_packet(xport
->lldp
, packet
);
3408 ctx
->xout
->slow
|= slow
;
3416 tnl_route_lookup_flow(const struct xlate_ctx
*ctx
,
3417 const struct flow
*oflow
,
3418 struct in6_addr
*ip
, struct in6_addr
*src
,
3419 struct xport
**out_port
)
3421 char out_dev
[IFNAMSIZ
];
3422 struct xbridge
*xbridge
;
3424 struct in6_addr dst
;
3426 dst
= flow_tnl_dst(&oflow
->tunnel
);
3427 if (!ovs_router_lookup(oflow
->pkt_mark
, &dst
, out_dev
, src
, &gw
)) {
3431 if (ipv6_addr_is_set(&gw
) &&
3432 (!IN6_IS_ADDR_V4MAPPED(&gw
) || in6_addr_get_mapped_ipv4(&gw
))) {
3438 HMAP_FOR_EACH (xbridge
, hmap_node
, &ctx
->xcfg
->xbridges
) {
3439 if (!strncmp(xbridge
->name
, out_dev
, IFNAMSIZ
)) {
3442 HMAP_FOR_EACH (port
, ofp_node
, &xbridge
->xports
) {
3443 if (!strncmp(netdev_get_name(port
->netdev
), out_dev
, IFNAMSIZ
)) {
3451 /* If tunnel IP isn't configured on bridges, then we search all ports. */
3452 HMAP_FOR_EACH (xbridge
, hmap_node
, &ctx
->xcfg
->xbridges
) {
3455 HMAP_FOR_EACH (port
, ofp_node
, &xbridge
->xports
) {
3456 if (!strncmp(netdev_get_name(port
->netdev
),
3457 out_dev
, IFNAMSIZ
)) {
3467 compose_table_xlate(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3468 struct dp_packet
*packet
)
3470 struct xbridge
*xbridge
= out_dev
->xbridge
;
3471 ovs_version_t version
= ofproto_dpif_get_tables_version(xbridge
->ofproto
);
3472 struct ofpact_output output
;
3475 ofpact_init(&output
.ofpact
, OFPACT_OUTPUT
, sizeof output
);
3476 flow_extract(packet
, &flow
);
3477 flow
.in_port
.ofp_port
= out_dev
->ofp_port
;
3478 output
.port
= OFPP_TABLE
;
3481 return ofproto_dpif_execute_actions__(xbridge
->ofproto
, version
, &flow
,
3482 NULL
, &output
.ofpact
, sizeof output
,
3483 ctx
->depth
, ctx
->resubmits
, packet
);
3487 tnl_send_nd_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3488 const struct eth_addr eth_src
,
3489 struct in6_addr
* ipv6_src
, struct in6_addr
* ipv6_dst
)
3491 struct dp_packet packet
;
3493 dp_packet_init(&packet
, 0);
3494 compose_nd_ns(&packet
, eth_src
, ipv6_src
, ipv6_dst
);
3495 compose_table_xlate(ctx
, out_dev
, &packet
);
3496 dp_packet_uninit(&packet
);
3500 tnl_send_arp_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3501 const struct eth_addr eth_src
,
3502 ovs_be32 ip_src
, ovs_be32 ip_dst
)
3504 struct dp_packet packet
;
3506 dp_packet_init(&packet
, 0);
3507 compose_arp(&packet
, ARP_OP_REQUEST
,
3508 eth_src
, eth_addr_zero
, true, ip_src
, ip_dst
);
3510 compose_table_xlate(ctx
, out_dev
, &packet
);
3511 dp_packet_uninit(&packet
);
3515 propagate_tunnel_data_to_flow__(struct flow
*dst_flow
,
3516 const struct flow
*src_flow
,
3517 struct eth_addr dmac
, struct eth_addr smac
,
3518 struct in6_addr s_ip6
, ovs_be32 s_ip
,
3519 bool is_tnl_ipv6
, uint8_t nw_proto
)
3521 dst_flow
->dl_dst
= dmac
;
3522 dst_flow
->dl_src
= smac
;
3524 dst_flow
->packet_type
= htonl(PT_ETH
);
3525 dst_flow
->nw_dst
= src_flow
->tunnel
.ip_dst
;
3526 dst_flow
->nw_src
= src_flow
->tunnel
.ip_src
;
3527 dst_flow
->ipv6_dst
= src_flow
->tunnel
.ipv6_dst
;
3528 dst_flow
->ipv6_src
= src_flow
->tunnel
.ipv6_src
;
3530 dst_flow
->nw_frag
= 0; /* Tunnel packets are unfragmented. */
3531 dst_flow
->nw_tos
= src_flow
->tunnel
.ip_tos
;
3532 dst_flow
->nw_ttl
= src_flow
->tunnel
.ip_ttl
;
3533 dst_flow
->tp_dst
= src_flow
->tunnel
.tp_dst
;
3534 dst_flow
->tp_src
= src_flow
->tunnel
.tp_src
;
3537 dst_flow
->dl_type
= htons(ETH_TYPE_IPV6
);
3538 if (ipv6_mask_is_any(&dst_flow
->ipv6_src
)
3539 && !ipv6_mask_is_any(&s_ip6
)) {
3540 dst_flow
->ipv6_src
= s_ip6
;
3543 dst_flow
->dl_type
= htons(ETH_TYPE_IP
);
3544 if (dst_flow
->nw_src
== 0 && s_ip
) {
3545 dst_flow
->nw_src
= s_ip
;
3548 dst_flow
->nw_proto
= nw_proto
;
3552 * Populate the 'flow' and 'base_flow' L3 fields to do the post tunnel push
3556 propagate_tunnel_data_to_flow(struct xlate_ctx
*ctx
, struct eth_addr dmac
,
3557 struct eth_addr smac
, struct in6_addr s_ip6
,
3558 ovs_be32 s_ip
, bool is_tnl_ipv6
,
3559 enum ovs_vport_type tnl_type
)
3561 struct flow
*base_flow
, *flow
;
3562 flow
= &ctx
->xin
->flow
;
3563 base_flow
= &ctx
->base_flow
;
3564 uint8_t nw_proto
= 0;
3567 case OVS_VPORT_TYPE_GRE
:
3568 case OVS_VPORT_TYPE_ERSPAN
:
3569 case OVS_VPORT_TYPE_IP6ERSPAN
:
3570 case OVS_VPORT_TYPE_IP6GRE
:
3571 nw_proto
= IPPROTO_GRE
;
3573 case OVS_VPORT_TYPE_VXLAN
:
3574 case OVS_VPORT_TYPE_GENEVE
:
3575 case OVS_VPORT_TYPE_GTPU
:
3576 nw_proto
= IPPROTO_UDP
;
3578 case OVS_VPORT_TYPE_LISP
:
3579 case OVS_VPORT_TYPE_STT
:
3580 case OVS_VPORT_TYPE_UNSPEC
:
3581 case OVS_VPORT_TYPE_NETDEV
:
3582 case OVS_VPORT_TYPE_INTERNAL
:
3583 case __OVS_VPORT_TYPE_MAX
:
3588 * Update base_flow first followed by flow as the dst_flow gets modified
3591 propagate_tunnel_data_to_flow__(base_flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3592 is_tnl_ipv6
, nw_proto
);
3593 propagate_tunnel_data_to_flow__(flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3594 is_tnl_ipv6
, nw_proto
);
3598 native_tunnel_output(struct xlate_ctx
*ctx
, const struct xport
*xport
,
3599 const struct flow
*flow
, odp_port_t tunnel_odp_port
,
3602 struct netdev_tnl_build_header_params tnl_params
;
3603 struct ovs_action_push_tnl tnl_push_data
;
3604 struct xport
*out_dev
= NULL
;
3605 ovs_be32 s_ip
= 0, d_ip
= 0;
3606 struct in6_addr s_ip6
= in6addr_any
;
3607 struct in6_addr d_ip6
= in6addr_any
;
3608 struct eth_addr smac
;
3609 struct eth_addr dmac
;
3611 char buf_sip6
[INET6_ADDRSTRLEN
];
3612 char buf_dip6
[INET6_ADDRSTRLEN
];
3614 /* Store sFlow data. */
3615 uint32_t sflow_n_outputs
= ctx
->sflow_n_outputs
;
3617 /* Structures to backup Ethernet and IP of base_flow. */
3618 struct flow old_base_flow
;
3619 struct flow old_flow
;
3621 /* Backup flow & base_flow data. */
3622 memcpy(&old_base_flow
, &ctx
->base_flow
, sizeof old_base_flow
);
3623 memcpy(&old_flow
, &ctx
->xin
->flow
, sizeof old_flow
);
3625 if (flow
->tunnel
.ip_src
) {
3626 in6_addr_set_mapped_ipv4(&s_ip6
, flow
->tunnel
.ip_src
);
3629 err
= tnl_route_lookup_flow(ctx
, flow
, &d_ip6
, &s_ip6
, &out_dev
);
3631 xlate_report(ctx
, OFT_WARN
, "native tunnel routing failed");
3635 xlate_report(ctx
, OFT_DETAIL
, "tunneling to %s via %s",
3636 ipv6_string_mapped(buf_dip6
, &d_ip6
),
3637 netdev_get_name(out_dev
->netdev
));
3639 /* Use mac addr of bridge port of the peer. */
3640 err
= netdev_get_etheraddr(out_dev
->netdev
, &smac
);
3642 xlate_report(ctx
, OFT_WARN
,
3643 "tunnel output device lacks Ethernet address");
3647 d_ip
= in6_addr_get_mapped_ipv4(&d_ip6
);
3649 s_ip
= in6_addr_get_mapped_ipv4(&s_ip6
);
3652 err
= tnl_neigh_lookup(out_dev
->xbridge
->name
, &d_ip6
, &dmac
);
3654 xlate_report(ctx
, OFT_DETAIL
,
3655 "neighbor cache miss for %s on bridge %s, "
3656 "sending %s request",
3657 buf_dip6
, out_dev
->xbridge
->name
, d_ip
? "ARP" : "ND");
3659 tnl_send_arp_request(ctx
, out_dev
, smac
, s_ip
, d_ip
);
3661 tnl_send_nd_request(ctx
, out_dev
, smac
, &s_ip6
, &d_ip6
);
3666 if (ctx
->xin
->xcache
) {
3667 struct xc_entry
*entry
;
3669 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TNL_NEIGH
);
3670 ovs_strlcpy(entry
->tnl_neigh_cache
.br_name
, out_dev
->xbridge
->name
,
3671 sizeof entry
->tnl_neigh_cache
.br_name
);
3672 entry
->tnl_neigh_cache
.d_ipv6
= d_ip6
;
3675 xlate_report(ctx
, OFT_DETAIL
, "tunneling from "ETH_ADDR_FMT
" %s"
3676 " to "ETH_ADDR_FMT
" %s",
3677 ETH_ADDR_ARGS(smac
), ipv6_string_mapped(buf_sip6
, &s_ip6
),
3678 ETH_ADDR_ARGS(dmac
), buf_dip6
);
3680 netdev_init_tnl_build_header_params(&tnl_params
, flow
, &s_ip6
, dmac
, smac
);
3681 err
= tnl_port_build_header(xport
->ofport
, &tnl_push_data
, &tnl_params
);
3685 tnl_push_data
.tnl_port
= tunnel_odp_port
;
3686 tnl_push_data
.out_port
= out_dev
->odp_port
;
3688 /* After tunnel header has been added, MAC and IP data of flow and
3689 * base_flow need to be set properly, since there is not recirculation
3690 * any more when sending packet to tunnel. */
3692 propagate_tunnel_data_to_flow(ctx
, dmac
, smac
, s_ip6
,
3693 s_ip
, tnl_params
.is_ipv6
,
3694 tnl_push_data
.tnl_type
);
3696 size_t clone_ofs
= 0;
3697 size_t push_action_size
;
3699 clone_ofs
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
3700 odp_put_tnl_push_action(ctx
->odp_actions
, &tnl_push_data
);
3701 push_action_size
= ctx
->odp_actions
->size
;
3704 const struct dpif_flow_stats
*backup_resubmit_stats
;
3705 struct xlate_cache
*backup_xcache
;
3706 struct flow_wildcards
*backup_wc
, wc
;
3707 bool backup_side_effects
;
3708 const struct dp_packet
*backup_packet
;
3710 memset(&wc
, 0 , sizeof wc
);
3711 backup_wc
= ctx
->wc
;
3713 ctx
->xin
->wc
= NULL
;
3714 backup_resubmit_stats
= ctx
->xin
->resubmit_stats
;
3715 backup_xcache
= ctx
->xin
->xcache
;
3716 backup_side_effects
= ctx
->xin
->allow_side_effects
;
3717 backup_packet
= ctx
->xin
->packet
;
3719 ctx
->xin
->resubmit_stats
= NULL
;
3720 ctx
->xin
->xcache
= xlate_cache_new(); /* Use new temporary cache. */
3721 ctx
->xin
->allow_side_effects
= false;
3722 ctx
->xin
->packet
= NULL
;
3724 /* Push the cache entry for the tunnel first. */
3725 struct xc_entry
*entry
;
3726 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TUNNEL_HEADER
);
3727 entry
->tunnel_hdr
.hdr_size
= tnl_push_data
.header_len
;
3728 entry
->tunnel_hdr
.operation
= ADD
;
3730 patch_port_output(ctx
, xport
, out_dev
);
3732 /* Similar to the stats update in revalidation, the x_cache entries
3733 * are populated by the previous translation are used to update the
3736 if (backup_resubmit_stats
) {
3737 struct dpif_flow_stats stats
= *backup_resubmit_stats
;
3738 xlate_push_stats(ctx
->xin
->xcache
, &stats
, false);
3740 xlate_cache_steal_entries(backup_xcache
, ctx
->xin
->xcache
);
3742 if (ctx
->odp_actions
->size
> push_action_size
) {
3743 nl_msg_end_non_empty_nested(ctx
->odp_actions
, clone_ofs
);
3745 nl_msg_cancel_nested(ctx
->odp_actions
, clone_ofs
);
3748 /* Restore context status. */
3749 ctx
->xin
->resubmit_stats
= backup_resubmit_stats
;
3750 xlate_cache_delete(ctx
->xin
->xcache
);
3751 ctx
->xin
->xcache
= backup_xcache
;
3752 ctx
->xin
->allow_side_effects
= backup_side_effects
;
3753 ctx
->xin
->packet
= backup_packet
;
3754 ctx
->wc
= backup_wc
;
3756 /* In order to maintain accurate stats, use recirc for
3757 * natvie tunneling. */
3758 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, 0);
3759 nl_msg_end_nested(ctx
->odp_actions
, clone_ofs
);
3762 /* Restore the flows after the translation. */
3763 memcpy(&ctx
->xin
->flow
, &old_flow
, sizeof ctx
->xin
->flow
);
3764 memcpy(&ctx
->base_flow
, &old_base_flow
, sizeof ctx
->base_flow
);
3766 /* Restore sFlow data. */
3767 ctx
->sflow_n_outputs
= sflow_n_outputs
;
3773 xlate_commit_actions(struct xlate_ctx
*ctx
)
3775 bool use_masked
= ctx
->xbridge
->support
.masked_set_action
;
3777 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
3778 ctx
->odp_actions
, ctx
->wc
,
3779 use_masked
, ctx
->pending_encap
,
3780 ctx
->pending_decap
, ctx
->encap_data
);
3781 ctx
->pending_encap
= false;
3782 ctx
->pending_decap
= false;
3783 ofpbuf_delete(ctx
->encap_data
);
3784 ctx
->encap_data
= NULL
;
3788 clear_conntrack(struct xlate_ctx
*ctx
)
3790 ctx
->conntracked
= false;
3791 flow_clear_conntrack(&ctx
->xin
->flow
);
3795 xlate_flow_is_protected(const struct xlate_ctx
*ctx
, const struct flow
*flow
, const struct xport
*xport_out
)
3797 const struct xport
*xport_in
;
3803 xport_in
= get_ofp_port(ctx
->xbridge
, flow
->in_port
.ofp_port
);
3805 return (xport_in
&& xport_in
->xbundle
&& xport_out
->xbundle
&&
3806 xport_in
->xbundle
->protected && xport_out
->xbundle
->protected);
3809 /* Function handles when a packet is sent from one bridge to another bridge.
3811 * The bridges are internally connected, either with patch ports or with
3814 * The output action to another bridge causes translation to continue within
3815 * the next bridge. This process can be recursive; the next bridge can
3816 * output yet to another bridge.
3818 * The translated actions from the second bridge onwards are enclosed within
3819 * the clone action, so that any modification to the packet will not be visible
3820 * to the remaining actions of the originating bridge.
3823 patch_port_output(struct xlate_ctx
*ctx
, const struct xport
*in_dev
,
3824 struct xport
*out_dev
)
3826 struct flow
*flow
= &ctx
->xin
->flow
;
3827 struct flow old_flow
= ctx
->xin
->flow
;
3828 struct flow_tnl old_flow_tnl_wc
= ctx
->wc
->masks
.tunnel
;
3829 bool old_conntrack
= ctx
->conntracked
;
3830 bool old_was_mpls
= ctx
->was_mpls
;
3831 ovs_version_t old_version
= ctx
->xin
->tables_version
;
3832 struct ofpbuf old_stack
= ctx
->stack
;
3833 uint8_t new_stack
[1024];
3834 struct ofpbuf old_action_set
= ctx
->action_set
;
3835 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
3836 uint64_t actset_stub
[1024 / 8];
3838 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
3839 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
3840 flow
->in_port
.ofp_port
= out_dev
->ofp_port
;
3841 flow
->metadata
= htonll(0);
3842 memset(&flow
->tunnel
, 0, sizeof flow
->tunnel
);
3843 memset(&ctx
->wc
->masks
.tunnel
, 0, sizeof ctx
->wc
->masks
.tunnel
);
3844 flow
->tunnel
.metadata
.tab
=
3845 ofproto_get_tun_tab(&out_dev
->xbridge
->ofproto
->up
);
3846 ctx
->wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
3847 memset(flow
->regs
, 0, sizeof flow
->regs
);
3848 flow
->actset_output
= OFPP_UNSET
;
3849 clear_conntrack(ctx
);
3850 ctx
->xin
->trace
= xlate_report(ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
3851 out_dev
->xbridge
->name
);
3852 mirror_mask_t old_mirrors
= ctx
->mirrors
;
3853 bool independent_mirrors
= out_dev
->xbridge
!= ctx
->xbridge
;
3854 if (independent_mirrors
) {
3857 ctx
->xbridge
= out_dev
->xbridge
;
3859 /* The bridge is now known so obtain its table version. */
3860 ctx
->xin
->tables_version
3861 = ofproto_dpif_get_tables_version(ctx
->xbridge
->ofproto
);
3863 if (!process_special(ctx
, out_dev
) && may_receive(out_dev
, ctx
)) {
3864 if (xport_stp_forward_state(out_dev
) &&
3865 xport_rstp_forward_state(out_dev
)) {
3866 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3867 false, true, clone_xlate_actions
);
3868 if (!ctx
->freezing
) {
3869 xlate_action_set(ctx
);
3871 if (ctx
->freezing
) {
3872 finish_freezing(ctx
);
3875 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
3876 * the learning action look at the packet, then drop it. */
3877 struct flow old_base_flow
= ctx
->base_flow
;
3878 size_t old_size
= ctx
->odp_actions
->size
;
3879 mirror_mask_t old_mirrors2
= ctx
->mirrors
;
3881 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3882 false, true, clone_xlate_actions
);
3883 ctx
->mirrors
= old_mirrors2
;
3884 ctx
->base_flow
= old_base_flow
;
3885 ctx
->odp_actions
->size
= old_size
;
3887 /* Undo changes that may have been done for freezing. */
3888 ctx_cancel_freeze(ctx
);
3892 ctx
->xin
->trace
= old_trace
;
3893 if (independent_mirrors
) {
3894 ctx
->mirrors
= old_mirrors
;
3896 ctx
->xin
->flow
= old_flow
;
3897 ctx
->xbridge
= in_dev
->xbridge
;
3898 ofpbuf_uninit(&ctx
->action_set
);
3899 ctx
->action_set
= old_action_set
;
3900 ofpbuf_uninit(&ctx
->stack
);
3901 ctx
->stack
= old_stack
;
3903 /* Restore calling bridge's lookup version. */
3904 ctx
->xin
->tables_version
= old_version
;
3906 /* Restore to calling bridge tunneling information */
3907 ctx
->wc
->masks
.tunnel
= old_flow_tnl_wc
;
3909 /* The out bridge popping MPLS should have no effect on the original
3911 ctx
->was_mpls
= old_was_mpls
;
3913 /* The out bridge's conntrack execution should have no effect on the
3914 * original bridge. */
3915 ctx
->conntracked
= old_conntrack
;
3917 /* The fact that the out bridge exits (for any reason) does not mean
3918 * that the original bridge should exit. Specifically, if the out
3919 * bridge freezes translation, the original bridge must continue
3920 * processing with the original, not the frozen packet! */
3923 /* Out bridge errors do not propagate back. */
3924 ctx
->error
= XLATE_OK
;
3926 if (ctx
->xin
->resubmit_stats
) {
3927 netdev_vport_inc_tx(in_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3928 netdev_vport_inc_rx(out_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3930 bfd_account_rx(out_dev
->bfd
, ctx
->xin
->resubmit_stats
);
3933 if (ctx
->xin
->xcache
) {
3934 struct xc_entry
*entry
;
3936 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
3937 entry
->dev
.tx
= netdev_ref(in_dev
->netdev
);
3938 entry
->dev
.rx
= netdev_ref(out_dev
->netdev
);
3939 entry
->dev
.bfd
= bfd_ref(out_dev
->bfd
);
3944 check_output_prerequisites(struct xlate_ctx
*ctx
,
3945 const struct xport
*xport
,
3949 struct flow_wildcards
*wc
= ctx
->wc
;
3952 xlate_report(ctx
, OFT_WARN
, "Nonexistent output port");
3954 } else if (xport
->config
& OFPUTIL_PC_NO_FWD
) {
3955 xlate_report(ctx
, OFT_DETAIL
, "OFPPC_NO_FWD set, skipping output");
3957 } else if (ctx
->mirror_snaplen
!= 0 && xport
->odp_port
== ODPP_NONE
) {
3958 xlate_report(ctx
, OFT_WARN
,
3959 "Mirror truncate to ODPP_NONE, skipping output");
3961 } else if (xlate_flow_is_protected(ctx
, flow
, xport
)) {
3962 xlate_report(ctx
, OFT_WARN
,
3963 "Flow is between protected ports, skipping output.");
3965 } else if (check_stp
) {
3966 if (is_stp(&ctx
->base_flow
)) {
3967 if (!xport_stp_should_forward_bpdu(xport
) &&
3968 !xport_rstp_should_manage_bpdu(xport
)) {
3969 if (ctx
->xbridge
->stp
!= NULL
) {
3970 xlate_report(ctx
, OFT_WARN
,
3971 "STP not in listening state, "
3972 "skipping bpdu output");
3973 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3974 xlate_report(ctx
, OFT_WARN
,
3975 "RSTP not managing BPDU in this state, "
3976 "skipping bpdu output");
3980 } else if ((xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
))
3981 || (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
,
3983 /* Pass; STP should not block link health detection. */
3984 } else if (!xport_stp_forward_state(xport
) ||
3985 !xport_rstp_forward_state(xport
)) {
3986 if (ctx
->xbridge
->stp
!= NULL
) {
3987 xlate_report(ctx
, OFT_WARN
,
3988 "STP not in forwarding state, skipping output");
3989 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3990 xlate_report(ctx
, OFT_WARN
,
3991 "RSTP not in forwarding state, skipping output");
3997 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L2
&&
3998 flow
->packet_type
!= htonl(PT_ETH
)) {
3999 xlate_report(ctx
, OFT_WARN
, "Trying to send non-Ethernet packet "
4000 "through legacy L2 port. Dropping packet.");
4007 /* Function verifies if destination address of received Neighbor Advertisement
4008 * message stored in 'flow' is correct. It should be either FF02::1:FFXX:XXXX
4009 * where XX:XXXX stands for the last 24 bits of 'ipv6_addr' or it should match
4012 is_nd_dst_correct(const struct flow
*flow
, const struct in6_addr
*ipv6_addr
)
4014 const uint8_t *flow_ipv6_addr
= (uint8_t *) &flow
->ipv6_dst
;
4015 const uint8_t *addr
= (uint8_t *) ipv6_addr
;
4017 return (IN6_IS_ADDR_MC_LINKLOCAL(&flow
->ipv6_dst
) &&
4018 flow_ipv6_addr
[11] == 0x01 &&
4019 flow_ipv6_addr
[12] == 0xff &&
4020 flow_ipv6_addr
[13] == addr
[13] &&
4021 flow_ipv6_addr
[14] == addr
[14] &&
4022 flow_ipv6_addr
[15] == addr
[15]) ||
4023 IN6_ARE_ADDR_EQUAL(&flow
->ipv6_dst
, ipv6_addr
);
4027 is_neighbor_reply_matched(const struct flow
*flow
, struct in6_addr
*ip_addr
)
4029 return ((IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
4030 flow
->dl_type
== htons(ETH_TYPE_ARP
) &&
4031 in6_addr_get_mapped_ipv4(ip_addr
) == flow
->nw_dst
) ||
4032 (!IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
4033 is_nd_dst_correct(flow
, ip_addr
)));
4036 /* Function verifies if the ARP reply or Neighbor Advertisement represented by
4037 * 'flow' addresses the 'xbridge' of 'ctx'. Returns true if the ARP TA or
4038 * neighbor discovery destination is in the list of configured IP addresses of
4039 * the bridge. Otherwise, it returns false. */
4041 is_neighbor_reply_correct(const struct xlate_ctx
*ctx
, const struct flow
*flow
)
4045 struct xbridge_addr
*xbridge_addr
= xbridge_addr_ref(ctx
->xbridge
->addr
);
4047 /* Verify if 'nw_dst' of ARP or 'ipv6_dst' of ICMPV6 is in the list. */
4048 for (i
= 0; xbridge_addr
&& i
< xbridge_addr
->n_addr
; i
++) {
4049 struct in6_addr
*ip_addr
= &xbridge_addr
->addr
[i
];
4050 if (is_neighbor_reply_matched(flow
, ip_addr
)) {
4051 /* Found a match. */
4057 xbridge_addr_unref(xbridge_addr
);
4059 /* If not found in bridge's IPs, search in its ports. */
4061 struct in6_addr
*ip_addr
, *mask
;
4065 HMAP_FOR_EACH (port
, ofp_node
, &ctx
->xbridge
->xports
) {
4066 error
= netdev_get_addr_list(port
->netdev
, &ip_addr
,
4069 ret
= is_neighbor_reply_matched(flow
, ip_addr
);
4073 /* Found a match. */
4083 terminate_native_tunnel(struct xlate_ctx
*ctx
, struct flow
*flow
,
4084 struct flow_wildcards
*wc
, odp_port_t
*tnl_port
)
4086 *tnl_port
= ODPP_NONE
;
4088 /* XXX: Write better Filter for tunnel port. We can use in_port
4089 * in tunnel-port flow to avoid these checks completely. */
4090 if (ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4091 *tnl_port
= tnl_port_map_lookup(flow
, wc
);
4093 /* If no tunnel port was found and it's about an ARP or ICMPv6 packet,
4094 * do tunnel neighbor snooping. */
4095 if (*tnl_port
== ODPP_NONE
&&
4096 (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
4097 flow
->nw_proto
== IPPROTO_ICMPV6
) &&
4098 is_neighbor_reply_correct(ctx
, flow
)) {
4099 tnl_neigh_snoop(flow
, wc
, ctx
->xbridge
->name
);
4103 return *tnl_port
!= ODPP_NONE
;
4107 compose_output_action__(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4108 const struct xlate_bond_recirc
*xr
, bool check_stp
,
4109 bool is_last_action OVS_UNUSED
, bool truncate
)
4111 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
4112 struct flow_wildcards
*wc
= ctx
->wc
;
4113 struct flow
*flow
= &ctx
->xin
->flow
;
4114 struct flow_tnl flow_tnl
;
4115 union flow_vlan_hdr flow_vlans
[FLOW_MAX_VLAN_HEADERS
];
4116 uint8_t flow_nw_tos
;
4117 odp_port_t out_port
, odp_port
, odp_tnl_port
;
4118 bool is_native_tunnel
= false;
4120 struct eth_addr flow_dl_dst
= flow
->dl_dst
;
4121 struct eth_addr flow_dl_src
= flow
->dl_src
;
4122 ovs_be32 flow_packet_type
= flow
->packet_type
;
4123 ovs_be16 flow_dl_type
= flow
->dl_type
;
4125 /* If 'struct flow' gets additional metadata, we'll need to zero it out
4126 * before traversing a patch port. */
4127 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 42);
4128 memset(&flow_tnl
, 0, sizeof flow_tnl
);
4130 if (!check_output_prerequisites(ctx
, xport
, flow
, check_stp
)) {
4134 if (flow
->packet_type
== htonl(PT_ETH
)) {
4135 /* Strip Ethernet header for legacy L3 port. */
4136 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L3
) {
4137 flow
->packet_type
= PACKET_TYPE_BE(OFPHTN_ETHERTYPE
,
4138 ntohs(flow
->dl_type
));
4144 xlate_report_error(ctx
, "Cannot truncate output to patch port");
4146 patch_port_output(ctx
, xport
, xport
->peer
);
4150 memcpy(flow_vlans
, flow
->vlans
, sizeof flow_vlans
);
4151 flow_nw_tos
= flow
->nw_tos
;
4153 if (count_skb_priorities(xport
)) {
4154 memset(&wc
->masks
.skb_priority
, 0xff, sizeof wc
->masks
.skb_priority
);
4155 if (dscp_from_skb_priority(xport
, flow
->skb_priority
, &dscp
)) {
4156 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
4157 flow
->nw_tos
&= ~IP_DSCP_MASK
;
4158 flow
->nw_tos
|= dscp
;
4162 if (xport
->is_tunnel
) {
4163 struct in6_addr dst
;
4164 /* Save tunnel metadata so that changes made due to
4165 * the Logical (tunnel) Port are not visible for any further
4166 * matches, while explicit set actions on tunnel metadata are.
4168 flow_tnl
= flow
->tunnel
;
4169 odp_port
= tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
4170 if (odp_port
== ODPP_NONE
) {
4171 xlate_report(ctx
, OFT_WARN
, "Tunneling decided against output");
4172 goto out
; /* restore flow_nw_tos */
4174 dst
= flow_tnl_dst(&flow
->tunnel
);
4175 if (ipv6_addr_equals(&dst
, &ctx
->orig_tunnel_ipv6_dst
)) {
4176 xlate_report(ctx
, OFT_WARN
, "Not tunneling to our own address");
4177 goto out
; /* restore flow_nw_tos */
4179 if (ctx
->xin
->resubmit_stats
) {
4180 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
4182 if (ctx
->xin
->xcache
) {
4183 struct xc_entry
*entry
;
4185 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
4186 entry
->dev
.tx
= netdev_ref(xport
->netdev
);
4188 out_port
= odp_port
;
4189 if (ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4190 xlate_report(ctx
, OFT_DETAIL
, "output to native tunnel");
4191 is_native_tunnel
= true;
4193 const char *tnl_type
;
4195 xlate_report(ctx
, OFT_DETAIL
, "output to kernel tunnel");
4196 tnl_type
= tnl_port_get_type(xport
->ofport
);
4197 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
4198 ctx
->odp_actions
, tnl_type
);
4199 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4202 odp_port
= xport
->odp_port
;
4203 out_port
= odp_port
;
4206 if (out_port
!= ODPP_NONE
) {
4207 /* Commit accumulated flow updates before output. */
4208 xlate_commit_actions(ctx
);
4210 if (xr
&& bond_use_lb_output_action(xport
->xbundle
->bond
)) {
4212 * If bond mode is balance-tcp and optimize balance tcp is enabled
4213 * then use the hash directly for slave selection and avoid
4216 * Currently support for netdev datapath only.
4218 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_LB_OUTPUT
,
4221 /* Recirculate the packet. */
4222 struct ovs_action_hash
*act_hash
;
4225 enum ovs_hash_alg hash_alg
= xr
->hash_alg
;
4226 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4227 /* Algorithm supported by all datapaths. */
4228 hash_alg
= OVS_HASH_ALG_L4
;
4230 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4231 OVS_ACTION_ATTR_HASH
,
4233 act_hash
->hash_alg
= hash_alg
;
4234 act_hash
->hash_basis
= xr
->hash_basis
;
4236 /* Recirc action. */
4237 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
,
4239 } else if (is_native_tunnel
) {
4240 /* Output to native tunnel port. */
4241 native_tunnel_output(ctx
, xport
, flow
, odp_port
, truncate
);
4242 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4244 } else if (terminate_native_tunnel(ctx
, flow
, wc
,
4246 /* Intercept packet to be received on native tunnel port. */
4247 nl_msg_put_odp_port(ctx
->odp_actions
, OVS_ACTION_ATTR_TUNNEL_POP
,
4251 /* Tunnel push-pop action is not compatible with
4253 compose_ipfix_action(ctx
, out_port
);
4255 /* Handle truncation of the mirrored packet. */
4256 if (ctx
->mirror_snaplen
> 0 &&
4257 ctx
->mirror_snaplen
< UINT16_MAX
) {
4258 struct ovs_action_trunc
*trunc
;
4260 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4261 OVS_ACTION_ATTR_TRUNC
,
4263 trunc
->max_len
= ctx
->mirror_snaplen
;
4264 if (!ctx
->xbridge
->support
.trunc
) {
4265 ctx
->xout
->slow
|= SLOW_ACTION
;
4269 nl_msg_put_odp_port(ctx
->odp_actions
,
4270 OVS_ACTION_ATTR_OUTPUT
,
4274 ctx
->sflow_odp_port
= odp_port
;
4275 ctx
->sflow_n_outputs
++;
4276 ctx
->nf_output_iface
= ofp_port
;
4279 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
) && xport
->xbundle
) {
4280 mirror_packet(ctx
, xport
->xbundle
,
4281 xbundle_mirror_dst(xport
->xbundle
->xbridge
,
4287 memcpy(flow
->vlans
, flow_vlans
, sizeof flow
->vlans
);
4288 flow
->nw_tos
= flow_nw_tos
;
4289 flow
->dl_dst
= flow_dl_dst
;
4290 flow
->dl_src
= flow_dl_src
;
4291 flow
->packet_type
= flow_packet_type
;
4292 flow
->dl_type
= flow_dl_type
;
4296 compose_output_action(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4297 const struct xlate_bond_recirc
*xr
,
4298 bool is_last_action
, bool truncate
)
4300 compose_output_action__(ctx
, ofp_port
, xr
, true,
4301 is_last_action
, truncate
);
4305 xlate_recursively(struct xlate_ctx
*ctx
, struct rule_dpif
*rule
,
4306 bool deepens
, bool is_last_action
,
4307 xlate_actions_handler
*actions_xlator
)
4309 struct rule_dpif
*old_rule
= ctx
->rule
;
4310 ovs_be64 old_cookie
= ctx
->rule_cookie
;
4311 const struct rule_actions
*actions
;
4313 if (ctx
->xin
->resubmit_stats
) {
4314 rule_dpif_credit_stats(rule
, ctx
->xin
->resubmit_stats
, false);
4319 ctx
->depth
+= deepens
;
4321 ctx
->rule_cookie
= rule
->up
.flow_cookie
;
4322 actions
= rule_get_actions(&rule
->up
);
4323 actions_xlator(actions
->ofpacts
, actions
->ofpacts_len
, ctx
,
4324 is_last_action
, false);
4325 ctx
->rule_cookie
= old_cookie
;
4326 ctx
->rule
= old_rule
;
4327 ctx
->depth
-= deepens
;
4331 xlate_resubmit_resource_check(struct xlate_ctx
*ctx
)
4333 if (ctx
->depth
>= MAX_DEPTH
) {
4334 xlate_report_error(ctx
, "over max translation depth %d", MAX_DEPTH
);
4335 ctx
->error
= XLATE_RECURSION_TOO_DEEP
;
4336 } else if (ctx
->resubmits
>= MAX_RESUBMITS
) {
4337 xlate_report_error(ctx
, "over %d resubmit actions", MAX_RESUBMITS
);
4338 ctx
->error
= XLATE_TOO_MANY_RESUBMITS
;
4339 } else if (ctx
->odp_actions
->size
> UINT16_MAX
) {
4340 xlate_report_error(ctx
, "resubmits yielded over 64 kB of actions");
4341 /* NOT an error, as we'll be slow-pathing the flow in this case? */
4342 ctx
->exit
= true; /* XXX: translation still terminated! */
4343 } else if (ctx
->stack
.size
>= 65536) {
4344 xlate_report_error(ctx
, "resubmits yielded over 64 kB of stack");
4345 ctx
->error
= XLATE_STACK_TOO_DEEP
;
4354 tuple_swap_flow(struct flow
*flow
, bool ipv4
)
4356 uint8_t nw_proto
= flow
->nw_proto
;
4357 flow
->nw_proto
= flow
->ct_nw_proto
;
4358 flow
->ct_nw_proto
= nw_proto
;
4361 ovs_be32 nw_src
= flow
->nw_src
;
4362 flow
->nw_src
= flow
->ct_nw_src
;
4363 flow
->ct_nw_src
= nw_src
;
4365 ovs_be32 nw_dst
= flow
->nw_dst
;
4366 flow
->nw_dst
= flow
->ct_nw_dst
;
4367 flow
->ct_nw_dst
= nw_dst
;
4369 struct in6_addr ipv6_src
= flow
->ipv6_src
;
4370 flow
->ipv6_src
= flow
->ct_ipv6_src
;
4371 flow
->ct_ipv6_src
= ipv6_src
;
4373 struct in6_addr ipv6_dst
= flow
->ipv6_dst
;
4374 flow
->ipv6_dst
= flow
->ct_ipv6_dst
;
4375 flow
->ct_ipv6_dst
= ipv6_dst
;
4378 ovs_be16 tp_src
= flow
->tp_src
;
4379 flow
->tp_src
= flow
->ct_tp_src
;
4380 flow
->ct_tp_src
= tp_src
;
4382 ovs_be16 tp_dst
= flow
->tp_dst
;
4383 flow
->tp_dst
= flow
->ct_tp_dst
;
4384 flow
->ct_tp_dst
= tp_dst
;
4388 tuple_swap(struct flow
*flow
, struct flow_wildcards
*wc
)
4390 bool ipv4
= (flow
->dl_type
== htons(ETH_TYPE_IP
));
4392 tuple_swap_flow(flow
, ipv4
);
4393 tuple_swap_flow(&wc
->masks
, ipv4
);
4397 xlate_table_action(struct xlate_ctx
*ctx
, ofp_port_t in_port
, uint8_t table_id
,
4398 bool may_packet_in
, bool honor_table_miss
,
4399 bool with_ct_orig
, bool is_last_action
,
4400 xlate_actions_handler
*xlator
)
4402 /* Check if we need to recirculate before matching in a table. */
4403 if (ctx
->was_mpls
) {
4404 ctx_trigger_freeze(ctx
);
4407 if (xlate_resubmit_resource_check(ctx
)) {
4408 uint8_t old_table_id
= ctx
->table_id
;
4409 struct rule_dpif
*rule
;
4411 ctx
->table_id
= table_id
;
4413 /* Swap packet fields with CT 5-tuple if requested. */
4415 /* Do not swap if there is no CT tuple, or if key is not IP. */
4416 if (ctx
->xin
->flow
.ct_nw_proto
== 0 ||
4417 !is_ip_any(&ctx
->xin
->flow
)) {
4418 xlate_report_error(ctx
,
4419 "resubmit(ct) with non-tracked or non-IP packet!");
4420 ctx
->table_id
= old_table_id
;
4423 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4425 rule
= rule_dpif_lookup_from_table(ctx
->xbridge
->ofproto
,
4426 ctx
->xin
->tables_version
,
4427 &ctx
->xin
->flow
, ctx
->wc
,
4428 ctx
->xin
->resubmit_stats
,
4429 &ctx
->table_id
, in_port
,
4430 may_packet_in
, honor_table_miss
,
4434 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4438 /* Fill in the cache entry here instead of xlate_recursively
4439 * to make the reference counting more explicit. We take a
4440 * reference in the lookups above if we are going to cache the
4442 if (ctx
->xin
->xcache
) {
4443 struct xc_entry
*entry
;
4445 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_RULE
);
4447 ofproto_rule_ref(&rule
->up
);
4450 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4451 xlate_report_table(ctx
, rule
, table_id
);
4452 xlate_recursively(ctx
, rule
, table_id
<= old_table_id
,
4453 is_last_action
, xlator
);
4454 ctx
->xin
->trace
= old_trace
;
4457 ctx
->table_id
= old_table_id
;
4462 /* Consumes the group reference, which is only taken if xcache exists. */
4464 xlate_group_stats(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4465 struct ofputil_bucket
*bucket
)
4467 if (ctx
->xin
->resubmit_stats
) {
4468 group_dpif_credit_stats(group
, bucket
, ctx
->xin
->resubmit_stats
);
4470 if (ctx
->xin
->xcache
) {
4471 struct xc_entry
*entry
;
4473 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_GROUP
);
4474 entry
->group
.group
= group
;
4475 entry
->group
.bucket
= bucket
;
4480 xlate_group_bucket(struct xlate_ctx
*ctx
, struct ofputil_bucket
*bucket
,
4481 bool is_last_action
)
4483 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4484 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
4485 char *s
= xasprintf("bucket %"PRIu32
, bucket
->bucket_id
);
4486 ctx
->xin
->trace
= &oftrace_report(ctx
->xin
->trace
, OFT_BUCKET
,
4491 uint64_t action_list_stub
[1024 / 8];
4492 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
4493 struct ofpbuf action_set
= ofpbuf_const_initializer(bucket
->ofpacts
,
4494 bucket
->ofpacts_len
);
4495 struct flow old_flow
= ctx
->xin
->flow
;
4496 bool old_was_mpls
= ctx
->was_mpls
;
4498 ofpacts_execute_action_set(&action_list
, &action_set
);
4500 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, is_last_action
,
4504 ofpbuf_uninit(&action_list
);
4506 /* Check if need to freeze. */
4507 if (ctx
->freezing
) {
4508 finish_freezing(ctx
);
4511 /* Roll back flow to previous state.
4512 * This is equivalent to cloning the packet for each bucket.
4514 * As a side effect any subsequently applied actions will
4515 * also effectively be applied to a clone of the packet taken
4516 * just before applying the all or indirect group.
4518 * Note that group buckets are action sets, hence they cannot modify the
4519 * main action set. Also any stack actions are ignored when executing an
4520 * action set, so group buckets cannot directly change the stack either.
4521 * However, we do allow resubmit actions in group buckets, which could
4522 * recursively execute actions that do modify the action set or change the
4523 * stack. The controller must be careful about what it does to the
4524 * action_set and stack in the tables resubmitted to from group buckets. */
4525 ctx
->xin
->flow
= old_flow
;
4527 /* The group bucket popping MPLS should have no effect after bucket
4529 ctx
->was_mpls
= old_was_mpls
;
4531 /* The fact that the group bucket exits (for any reason) does not mean that
4532 * the translation after the group action should exit. Specifically, if
4533 * the group bucket freezes translation, the actions after the group action
4534 * must continue processing with the original, not the frozen packet! */
4537 /* Context error in a bucket should not impact processing of other buckets
4538 * or actions. This is similar to cloning a packet for group buckets.
4539 * There is no need to restore the error back to old value due to the fact
4540 * that we actually processed group action which can happen only when there
4541 * is no previous context error.
4543 * Exception to above is errors which are system limits to protect
4544 * translation from running too long or occupy too much space. These errors
4545 * should not be masked. XLATE_RECURSION_TOO_DEEP, XLATE_TOO_MANY_RESUBMITS
4546 * and XLATE_STACK_TOO_DEEP fall in this category. */
4547 if (ctx
->error
== XLATE_TOO_MANY_MPLS_LABELS
||
4548 ctx
->error
== XLATE_UNSUPPORTED_PACKET_TYPE
) {
4549 /* reset the error and continue processing other buckets */
4550 ctx
->error
= XLATE_OK
;
4553 ctx
->xin
->trace
= old_trace
;
4556 static struct ofputil_bucket
*
4557 pick_ff_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4559 return group_first_live_bucket(ctx
, group
, 0);
4562 static struct ofputil_bucket
*
4563 pick_default_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4565 flow_mask_hash_fields(&ctx
->xin
->flow
, ctx
->wc
,
4566 NX_HASH_FIELDS_SYMMETRIC_L4
);
4567 return group_best_live_bucket(ctx
, group
,
4568 flow_hash_symmetric_l4(&ctx
->xin
->flow
, 0));
4571 static struct ofputil_bucket
*
4572 pick_hash_fields_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4574 const struct field_array
*fields
= &group
->up
.props
.fields
;
4575 const uint8_t *mask_values
= fields
->values
;
4576 uint32_t basis
= hash_uint64(group
->up
.props
.selection_method_param
);
4579 BITMAP_FOR_EACH_1 (i
, MFF_N_IDS
, fields
->used
.bm
) {
4580 const struct mf_field
*mf
= mf_from_id(i
);
4582 /* Skip fields for which prerequisites are not met. */
4583 if (!mf_are_prereqs_ok(mf
, &ctx
->xin
->flow
, ctx
->wc
)) {
4584 /* Skip the mask bytes for this field. */
4585 mask_values
+= mf
->n_bytes
;
4589 union mf_value value
;
4590 union mf_value mask
;
4592 mf_get_value(mf
, &ctx
->xin
->flow
, &value
);
4593 /* Mask the value. */
4594 for (int j
= 0; j
< mf
->n_bytes
; j
++) {
4595 mask
.b
[j
] = *mask_values
++;
4596 value
.b
[j
] &= mask
.b
[j
];
4598 basis
= hash_bytes(&value
, mf
->n_bytes
, basis
);
4600 /* For tunnels, hash in whether the field is present. */
4601 if (mf_is_tun_metadata(mf
)) {
4602 basis
= hash_boolean(mf_is_set(mf
, &ctx
->xin
->flow
), basis
);
4605 mf_mask_field_masked(mf
, &mask
, ctx
->wc
);
4608 return group_best_live_bucket(ctx
, group
, basis
);
4611 static struct ofputil_bucket
*
4612 pick_dp_hash_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4614 uint32_t dp_hash
= ctx
->xin
->flow
.dp_hash
;
4616 /* dp_hash value 0 is special since it means that the dp_hash has not been
4617 * computed, as all computed dp_hash values are non-zero. Therefore
4618 * compare to zero can be used to decide if the dp_hash value is valid
4619 * without masking the dp_hash field. */
4621 enum ovs_hash_alg hash_alg
= group
->hash_alg
;
4622 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4623 /* Algorithm supported by all datapaths. */
4624 hash_alg
= OVS_HASH_ALG_L4
;
4626 ctx_trigger_recirculate_with_hash(ctx
, hash_alg
, group
->hash_basis
);
4629 uint32_t hash_mask
= group
->hash_mask
;
4630 ctx
->wc
->masks
.dp_hash
|= hash_mask
;
4632 /* Starting from the original masked dp_hash value iterate over the
4633 * hash mapping table to find the first live bucket. As the buckets
4634 * are quasi-randomly spread over the hash values, this maintains
4635 * a distribution according to bucket weights even when some buckets
4637 for (int i
= 0; i
<= hash_mask
; i
++) {
4638 struct ofputil_bucket
*b
=
4639 group
->hash_map
[(dp_hash
+ i
) & hash_mask
];
4640 if (bucket_is_alive(ctx
, b
, 0)) {
4649 static struct ofputil_bucket
*
4650 pick_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4652 /* Select groups may access flow keys beyond L2 in order to
4653 * select a bucket. Recirculate as appropriate to make this possible.
4655 if (ctx
->was_mpls
) {
4656 ctx_trigger_freeze(ctx
);
4660 switch (group
->selection_method
) {
4661 case SEL_METHOD_DEFAULT
:
4662 return pick_default_select_group(ctx
, group
);
4664 case SEL_METHOD_HASH
:
4665 return pick_hash_fields_select_group(ctx
, group
);
4667 case SEL_METHOD_DP_HASH
:
4668 return pick_dp_hash_select_group(ctx
, group
);
4671 /* Parsing of groups ensures this never happens */
4679 xlate_group_action__(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4680 bool is_last_action
)
4682 if (group
->up
.type
== OFPGT11_ALL
|| group
->up
.type
== OFPGT11_INDIRECT
) {
4683 struct ovs_list
*last_bucket
= group
->up
.buckets
.prev
;
4684 struct ofputil_bucket
*bucket
;
4685 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
4686 bool is_last_bucket
= &bucket
->list_node
== last_bucket
;
4687 xlate_group_bucket(ctx
, bucket
, is_last_action
&& is_last_bucket
);
4689 xlate_group_stats(ctx
, group
, NULL
);
4691 struct ofputil_bucket
*bucket
;
4692 if (group
->up
.type
== OFPGT11_SELECT
) {
4693 bucket
= pick_select_group(ctx
, group
);
4694 } else if (group
->up
.type
== OFPGT11_FF
) {
4695 bucket
= pick_ff_group(ctx
, group
);
4701 xlate_report(ctx
, OFT_DETAIL
, "using bucket %"PRIu32
,
4703 xlate_group_bucket(ctx
, bucket
, is_last_action
);
4704 xlate_group_stats(ctx
, group
, bucket
);
4706 xlate_report(ctx
, OFT_DETAIL
, "no live bucket");
4707 if (ctx
->xin
->xcache
) {
4708 ofproto_group_unref(&group
->up
);
4715 xlate_group_action(struct xlate_ctx
*ctx
, uint32_t group_id
,
4716 bool is_last_action
)
4718 if (xlate_resubmit_resource_check(ctx
)) {
4719 struct group_dpif
*group
;
4721 /* Take ref only if xcache exists. */
4722 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
4723 ctx
->xin
->tables_version
, ctx
->xin
->xcache
);
4725 /* XXX: Should set ctx->error ? */
4726 xlate_report(ctx
, OFT_WARN
, "output to nonexistent group %"PRIu32
,
4730 xlate_group_action__(ctx
, group
, is_last_action
);
4737 xlate_ofpact_resubmit(struct xlate_ctx
*ctx
,
4738 const struct ofpact_resubmit
*resubmit
,
4739 bool is_last_action
)
4743 bool may_packet_in
= false;
4744 bool honor_table_miss
= false;
4746 if (ctx
->rule
&& rule_dpif_is_internal(ctx
->rule
)) {
4747 /* Still allow missed packets to be sent to the controller
4748 * if resubmitting from an internal table. */
4749 may_packet_in
= true;
4750 honor_table_miss
= true;
4753 in_port
= resubmit
->in_port
;
4754 if (in_port
== OFPP_IN_PORT
) {
4755 in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
4758 table_id
= resubmit
->table_id
;
4759 if (table_id
== 255) {
4760 table_id
= ctx
->table_id
;
4763 xlate_table_action(ctx
, in_port
, table_id
, may_packet_in
,
4764 honor_table_miss
, resubmit
->with_ct_orig
,
4765 is_last_action
, do_xlate_actions
);
4769 flood_packet_to_port(struct xlate_ctx
*ctx
, const struct xport
*xport
,
4770 bool all
, bool is_last_action
)
4777 compose_output_action__(ctx
, xport
->ofp_port
, NULL
, false,
4778 is_last_action
, false);
4780 compose_output_action(ctx
, xport
->ofp_port
, NULL
, is_last_action
,
4786 flood_packets(struct xlate_ctx
*ctx
, bool all
, bool is_last_action
)
4788 const struct xport
*xport
, *last
= NULL
;
4790 /* Use 'last' the keep track of the last output port. */
4791 HMAP_FOR_EACH (xport
, ofp_node
, &ctx
->xbridge
->xports
) {
4792 if (xport
->ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
4796 if (all
|| !(xport
->config
& OFPUTIL_PC_NO_FLOOD
)) {
4797 /* 'last' is not the last port, send a packet out, and
4799 flood_packet_to_port(ctx
, last
, all
, false);
4804 /* Send the packet to the 'last' port. */
4805 flood_packet_to_port(ctx
, last
, all
, is_last_action
);
4806 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
4810 put_controller_user_action(struct xlate_ctx
*ctx
,
4811 bool dont_send
, bool continuation
,
4812 uint32_t recirc_id
, int len
,
4813 enum ofp_packet_in_reason reason
,
4814 uint16_t controller_id
)
4816 struct user_action_cookie cookie
;
4818 memset(&cookie
, 0, sizeof cookie
);
4819 cookie
.type
= USER_ACTION_COOKIE_CONTROLLER
;
4820 cookie
.ofp_in_port
= OFPP_NONE
,
4821 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
4822 cookie
.controller
.dont_send
= dont_send
;
4823 cookie
.controller
.continuation
= continuation
;
4824 cookie
.controller
.reason
= reason
;
4825 cookie
.controller
.recirc_id
= recirc_id
;
4826 put_32aligned_be64(&cookie
.controller
.rule_cookie
, ctx
->rule_cookie
);
4827 cookie
.controller
.controller_id
= controller_id
;
4828 cookie
.controller
.max_len
= len
;
4830 odp_port_t odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
4831 ctx
->xin
->flow
.in_port
.ofp_port
);
4832 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
4833 odp_put_userspace_action(pid
, &cookie
, sizeof cookie
, ODPP_NONE
,
4834 false, ctx
->odp_actions
);
4838 xlate_controller_action(struct xlate_ctx
*ctx
, int len
,
4839 enum ofp_packet_in_reason reason
,
4840 uint16_t controller_id
,
4841 uint32_t provider_meter_id
,
4842 const uint8_t *userdata
, size_t userdata_len
)
4844 xlate_commit_actions(ctx
);
4846 /* A packet sent by an action in a table-miss rule is considered an
4847 * explicit table miss. OpenFlow before 1.3 doesn't have that concept so
4848 * it will get translated back to OFPR_ACTION for those versions. */
4849 if (reason
== OFPR_ACTION
4850 && ctx
->rule
&& rule_is_table_miss(&ctx
->rule
->up
)) {
4851 reason
= OFPR_EXPLICIT_MISS
;
4854 struct frozen_state state
= {
4855 .table_id
= ctx
->table_id
,
4856 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4857 .stack
= ctx
->stack
.data
,
4858 .stack_size
= ctx
->stack
.size
,
4859 .mirrors
= ctx
->mirrors
,
4860 .conntracked
= ctx
->conntracked
,
4861 .was_mpls
= ctx
->was_mpls
,
4865 .action_set_len
= 0,
4866 .userdata
= CONST_CAST(uint8_t *, userdata
),
4867 .userdata_len
= userdata_len
,
4869 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4871 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4873 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4874 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4877 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4879 /* If the controller action didn't request a meter (indicated by a
4880 * 'meter_id' argument other than NX_CTLR_NO_METER), see if one was
4881 * configured through the "controller" virtual meter.
4883 * Internally, ovs-vswitchd uses UINT32_MAX to indicate no meter is
4886 if (provider_meter_id
== UINT32_MAX
) {
4887 meter_id
= ctx
->xbridge
->ofproto
->up
.controller_meter_id
;
4889 meter_id
= provider_meter_id
;
4894 if (meter_id
!= UINT32_MAX
) {
4895 /* If controller meter is configured, generate clone(meter, userspace)
4897 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
4898 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
4900 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
4901 OVS_SAMPLE_ATTR_ACTIONS
);
4902 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
4905 /* Generate the datapath flows even if we don't send the packet-in
4906 * so that debugging more closely represents normal state. */
4907 bool dont_send
= false;
4908 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4911 put_controller_user_action(ctx
, dont_send
, false, recirc_id
, len
,
4912 reason
, controller_id
);
4914 if (meter_id
!= UINT32_MAX
) {
4915 nl_msg_end_nested(ctx
->odp_actions
, ac_offset
);
4916 nl_msg_end_nested(ctx
->odp_actions
, offset
);
4920 /* Creates a frozen state, and allocates a unique recirc id for the given
4921 * state. Returns a non-zero recirc id if it is allocated successfully.
4922 * Returns 0 otherwise.
4925 finish_freezing__(struct xlate_ctx
*ctx
, uint8_t table
)
4927 ovs_assert(ctx
->freezing
);
4929 struct frozen_state state
= {
4931 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4932 .stack
= ctx
->stack
.data
,
4933 .stack_size
= ctx
->stack
.size
,
4934 .mirrors
= ctx
->mirrors
,
4935 .conntracked
= ctx
->conntracked
,
4936 .was_mpls
= ctx
->was_mpls
,
4937 .xport_uuid
= ctx
->xin
->xport_uuid
,
4938 .ofpacts
= ctx
->frozen_actions
.data
,
4939 .ofpacts_len
= ctx
->frozen_actions
.size
,
4940 .action_set
= ctx
->action_set
.data
,
4941 .action_set_len
= ctx
->action_set
.size
,
4942 .userdata
= ctx
->pause
? CONST_CAST(uint8_t *,ctx
->pause
->userdata
)
4944 .userdata_len
= ctx
->pause
? ctx
->pause
->userdata_len
: 0,
4946 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4948 /* Allocate a unique recirc id for the given metadata state in the
4949 * flow. An existing id, with a new reference to the corresponding
4950 * recirculation context, will be returned if possible.
4951 * The life-cycle of this recirc id is managed by associating it
4952 * with the udpif key ('ukey') created for each new datapath flow. */
4953 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4955 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4956 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4959 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4962 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4966 put_controller_user_action(ctx
, false, true, recirc_id
,
4967 ctx
->pause
->max_len
,
4969 ctx
->pause
->controller_id
);
4971 if (ctx
->recirc_update_dp_hash
) {
4972 struct ovs_action_hash
*act_hash
;
4975 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4976 OVS_ACTION_ATTR_HASH
,
4978 act_hash
->hash_alg
= ctx
->dp_hash_alg
;
4979 act_hash
->hash_basis
= ctx
->dp_hash_basis
;
4981 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, recirc_id
);
4984 /* Undo changes done by freezing. */
4985 ctx_cancel_freeze(ctx
);
4989 /* Called only when we're freezing. */
4991 finish_freezing(struct xlate_ctx
*ctx
)
4993 xlate_commit_actions(ctx
);
4994 finish_freezing__(ctx
, 0);
4997 /* Fork the pipeline here. The current packet will continue processing the
4998 * current action list. A clone of the current packet will recirculate, skip
4999 * the remainder of the current action list and asynchronously resume pipeline
5000 * processing in 'table' with the current metadata and action set. */
5002 compose_recirculate_and_fork(struct xlate_ctx
*ctx
, uint8_t table
,
5003 const uint16_t zone
)
5006 ctx
->freezing
= true;
5007 recirc_id
= finish_freezing__(ctx
, table
);
5009 if (OVS_UNLIKELY(ctx
->xin
->trace
) && recirc_id
) {
5010 if (oftrace_add_recirc_node(ctx
->xin
->recirc_queue
,
5011 OFT_RECIRC_CONNTRACK
, &ctx
->xin
->flow
,
5012 ctx
->ct_nat_action
, ctx
->xin
->packet
,
5014 xlate_report(ctx
, OFT_DETAIL
, "A clone of the packet is forked to "
5015 "recirculate. The forked pipeline will be resumed at "
5016 "table %u.", table
);
5018 xlate_report(ctx
, OFT_DETAIL
, "Failed to trace the conntrack "
5019 "forked pipeline with recirc_id = %d.", recirc_id
);
5025 compose_mpls_push_action(struct xlate_ctx
*ctx
, struct ofpact_push_mpls
*mpls
)
5027 struct flow
*flow
= &ctx
->xin
->flow
;
5030 ovs_assert(eth_type_mpls(mpls
->ethertype
));
5032 n
= flow_count_mpls_labels(flow
, ctx
->wc
);
5034 xlate_commit_actions(ctx
);
5035 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
5036 if (ctx
->xin
->packet
!= NULL
) {
5037 xlate_report_error(ctx
, "dropping packet on which an MPLS push "
5038 "action can't be performed as it would have "
5039 "more MPLS LSEs than the %d supported.",
5040 FLOW_MAX_MPLS_LABELS
);
5042 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
5046 /* Update flow's MPLS stack, and clear L3/4 fields to mark them invalid. */
5047 flow_push_mpls(flow
, n
, mpls
->ethertype
, ctx
->wc
, true);
5051 compose_mpls_pop_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
5053 struct flow
*flow
= &ctx
->xin
->flow
;
5054 int n
= flow_count_mpls_labels(flow
, ctx
->wc
);
5056 if (flow_pop_mpls(flow
, n
, eth_type
, ctx
->wc
)) {
5057 if (!eth_type_mpls(eth_type
) && ctx
->xbridge
->support
.odp
.recirc
) {
5058 ctx
->was_mpls
= true;
5060 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
5061 if (ctx
->xin
->packet
!= NULL
) {
5062 xlate_report_error(ctx
, "dropping packet on which an "
5063 "MPLS pop action can't be performed as it has "
5064 "more MPLS LSEs than the %d supported.",
5065 FLOW_MAX_MPLS_LABELS
);
5067 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
5068 ofpbuf_clear(ctx
->odp_actions
);
5073 compose_dec_ttl(struct xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
5075 struct flow
*flow
= &ctx
->xin
->flow
;
5077 if (!is_ip_any(flow
)) {
5081 ctx
->wc
->masks
.nw_ttl
= 0xff;
5082 if (flow
->nw_ttl
> 1) {
5088 for (i
= 0; i
< ids
->n_controllers
; i
++) {
5089 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5090 ids
->cnt_ids
[i
], UINT32_MAX
, NULL
, 0);
5093 /* Stop processing for current table. */
5094 xlate_report(ctx
, OFT_WARN
, "IPv%d decrement TTL exception",
5095 flow
->dl_type
== htons(ETH_TYPE_IP
) ? 4 : 6);
5101 compose_set_mpls_label_action(struct xlate_ctx
*ctx
, ovs_be32 label
)
5103 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5104 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_LABEL_MASK
);
5105 set_mpls_lse_label(&ctx
->xin
->flow
.mpls_lse
[0], label
);
5110 compose_set_mpls_tc_action(struct xlate_ctx
*ctx
, uint8_t tc
)
5112 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5113 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TC_MASK
);
5114 set_mpls_lse_tc(&ctx
->xin
->flow
.mpls_lse
[0], tc
);
5119 compose_dec_nsh_ttl_action(struct xlate_ctx
*ctx
)
5121 struct flow
*flow
= &ctx
->xin
->flow
;
5123 if ((flow
->packet_type
== htonl(PT_NSH
)) ||
5124 (flow
->dl_type
== htons(ETH_TYPE_NSH
))) {
5125 ctx
->wc
->masks
.nsh
.ttl
= 0xff;
5126 if (flow
->nsh
.ttl
> 1) {
5130 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5131 0, UINT32_MAX
, NULL
, 0);
5135 /* Stop processing for current table. */
5136 xlate_report(ctx
, OFT_WARN
, "NSH decrement TTL exception");
5141 compose_set_mpls_ttl_action(struct xlate_ctx
*ctx
, uint8_t ttl
)
5143 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5144 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5145 set_mpls_lse_ttl(&ctx
->xin
->flow
.mpls_lse
[0], ttl
);
5150 compose_dec_mpls_ttl_action(struct xlate_ctx
*ctx
)
5152 struct flow
*flow
= &ctx
->xin
->flow
;
5154 if (eth_type_mpls(flow
->dl_type
)) {
5155 uint8_t ttl
= mpls_lse_to_ttl(flow
->mpls_lse
[0]);
5157 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5160 set_mpls_lse_ttl(&flow
->mpls_lse
[0], ttl
);
5163 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0,
5164 UINT32_MAX
, NULL
, 0);
5168 /* Stop processing for current table. */
5169 xlate_report(ctx
, OFT_WARN
, "MPLS decrement TTL exception");
5174 xlate_delete_field(struct xlate_ctx
*ctx
,
5176 const struct ofpact_delete_field
*odf
)
5178 struct ds s
= DS_EMPTY_INITIALIZER
;
5180 /* Currently, only tun_metadata is allowed for delete_field action. */
5181 tun_metadata_delete(&flow
->tunnel
, odf
->field
);
5183 ds_put_format(&s
, "delete %s", odf
->field
->name
);
5184 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
5188 /* Emits an action that outputs to 'port', within 'ctx'.
5190 * 'controller_len' affects only packets sent to an OpenFlow controller. It
5191 * is the maximum number of bytes of the packet to send. UINT16_MAX means to
5192 * send the whole packet (and 0 means to omit the packet entirely).
5194 * 'may_packet_in' determines whether the packet may be sent to an OpenFlow
5195 * controller. If it is false, then the packet is never sent to the OpenFlow
5198 * 'is_last_action' should be true if this output is the last OpenFlow action
5199 * to be processed, which enables certain optimizations.
5201 * 'truncate' should be true if the packet to be output is being truncated,
5202 * which suppresses certain optimizations. */
5204 xlate_output_action(struct xlate_ctx
*ctx
, ofp_port_t port
,
5205 uint16_t controller_len
, bool may_packet_in
,
5206 bool is_last_action
, bool truncate
,
5207 bool group_bucket_action
)
5209 ofp_port_t prev_nf_output_iface
= ctx
->nf_output_iface
;
5211 ctx
->nf_output_iface
= NF_OUT_DROP
;
5215 compose_output_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
,
5216 is_last_action
, truncate
);
5219 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
5220 0, may_packet_in
, true, false, false,
5227 flood_packets(ctx
, false, is_last_action
);
5230 flood_packets(ctx
, true, is_last_action
);
5232 case OFPP_CONTROLLER
:
5233 xlate_controller_action(ctx
, controller_len
,
5234 (ctx
->in_packet_out
? OFPR_PACKET_OUT
5235 : group_bucket_action
? OFPR_GROUP
5236 : ctx
->in_action_set
? OFPR_ACTION_SET
5238 0, UINT32_MAX
, NULL
, 0);
5244 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5245 compose_output_action(ctx
, port
, NULL
, is_last_action
, truncate
);
5247 xlate_report_info(ctx
, "skipping output to input port");
5252 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
5253 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
5254 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5255 ctx
->nf_output_iface
= prev_nf_output_iface
;
5256 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
5257 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5258 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5263 xlate_output_reg_action(struct xlate_ctx
*ctx
,
5264 const struct ofpact_output_reg
*or,
5265 bool is_last_action
,
5266 bool group_bucket_action
)
5268 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->xin
->flow
);
5269 if (port
<= UINT16_MAX
) {
5270 xlate_report(ctx
, OFT_DETAIL
, "output port is %"PRIu64
, port
);
5272 union mf_subvalue value
;
5274 memset(&value
, 0xff, sizeof value
);
5275 mf_write_subfield_flow(&or->src
, &value
, &ctx
->wc
->masks
);
5276 xlate_output_action(ctx
, u16_to_ofp(port
), or->max_len
,
5277 false, is_last_action
, false,
5278 group_bucket_action
);
5280 xlate_report(ctx
, OFT_WARN
, "output port %"PRIu64
" is out of range",
5286 xlate_output_trunc_action(struct xlate_ctx
*ctx
,
5287 ofp_port_t port
, uint32_t max_len
,
5288 bool is_last_action
,
5289 bool group_bucket_action
)
5291 bool support_trunc
= ctx
->xbridge
->support
.trunc
;
5292 struct ovs_action_trunc
*trunc
;
5293 char name
[OFP_MAX_PORT_NAME_LEN
];
5300 case OFPP_CONTROLLER
:
5302 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5303 xlate_report(ctx
, OFT_WARN
,
5304 "output_trunc does not support port: %s", name
);
5309 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5310 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, port
);
5312 if (xport
== NULL
|| xport
->odp_port
== ODPP_NONE
) {
5313 /* Since truncate happens at its following output action, if
5314 * the output port is a patch port, the behavior is somehow
5315 * unpredictable. For simplicity, disallow this case. */
5316 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5317 xlate_report_error(ctx
, "output_trunc does not support "
5318 "patch port %s", name
);
5322 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
5323 OVS_ACTION_ATTR_TRUNC
,
5325 trunc
->max_len
= max_len
;
5326 xlate_output_action(ctx
, port
, 0, false, is_last_action
, true,
5327 group_bucket_action
);
5328 if (!support_trunc
) {
5329 ctx
->xout
->slow
|= SLOW_ACTION
;
5332 xlate_report_info(ctx
, "skipping output to input port");
5339 xlate_enqueue_action(struct xlate_ctx
*ctx
,
5340 const struct ofpact_enqueue
*enqueue
,
5341 bool is_last_action
,
5342 bool group_bucket_action
)
5344 ofp_port_t ofp_port
= enqueue
->port
;
5345 uint32_t queue_id
= enqueue
->queue
;
5346 uint32_t flow_priority
, priority
;
5349 /* Translate queue to priority. */
5350 error
= dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &priority
);
5352 /* Fall back to ordinary output action. */
5353 xlate_output_action(ctx
, enqueue
->port
, 0, false,
5354 is_last_action
, false,
5355 group_bucket_action
);
5359 /* Check output port. */
5360 if (ofp_port
== OFPP_IN_PORT
) {
5361 ofp_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
5362 } else if (ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
5366 /* Add datapath actions. */
5367 flow_priority
= ctx
->xin
->flow
.skb_priority
;
5368 ctx
->xin
->flow
.skb_priority
= priority
;
5369 compose_output_action(ctx
, ofp_port
, NULL
, is_last_action
, false);
5370 ctx
->xin
->flow
.skb_priority
= flow_priority
;
5372 /* Update NetFlow output port. */
5373 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5374 ctx
->nf_output_iface
= ofp_port
;
5375 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5376 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5381 xlate_set_queue_action(struct xlate_ctx
*ctx
, uint32_t queue_id
)
5383 uint32_t skb_priority
;
5385 if (!dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &skb_priority
)) {
5386 ctx
->xin
->flow
.skb_priority
= skb_priority
;
5388 /* Couldn't translate queue to a priority. Nothing to do. A warning
5389 * has already been logged. */
5394 slave_enabled_cb(ofp_port_t ofp_port
, void *xbridge_
)
5396 const struct xbridge
*xbridge
= xbridge_
;
5407 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
5410 port
= get_ofp_port(xbridge
, ofp_port
);
5411 return port
? port
->may_enable
: false;
5416 xlate_bundle_action(struct xlate_ctx
*ctx
,
5417 const struct ofpact_bundle
*bundle
,
5418 bool is_last_action
,
5419 bool group_bucket_action
)
5423 port
= bundle_execute(bundle
, &ctx
->xin
->flow
, ctx
->wc
, slave_enabled_cb
,
5424 CONST_CAST(struct xbridge
*, ctx
->xbridge
));
5425 if (bundle
->dst
.field
) {
5426 nxm_reg_load(&bundle
->dst
, ofp_to_u16(port
), &ctx
->xin
->flow
, ctx
->wc
);
5427 xlate_report_subfield(ctx
, &bundle
->dst
);
5429 xlate_output_action(ctx
, port
, 0, false, is_last_action
, false,
5430 group_bucket_action
);
5435 xlate_learn_action(struct xlate_ctx
*ctx
, const struct ofpact_learn
*learn
)
5437 learn_mask(learn
, ctx
->wc
);
5439 if (ctx
->xin
->xcache
|| ctx
->xin
->allow_side_effects
) {
5440 uint64_t ofpacts_stub
[1024 / 8];
5441 struct ofputil_flow_mod fm
;
5442 struct ofproto_flow_mod ofm__
, *ofm
;
5443 struct ofpbuf ofpacts
;
5446 if (ctx
->xin
->xcache
) {
5447 ofm
= xmalloc(sizeof *ofm
);
5452 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
5453 learn_execute(learn
, &ctx
->xin
->flow
, &fm
, &ofpacts
);
5454 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
5455 struct ds s
= DS_EMPTY_INITIALIZER
;
5456 ds_put_format(&s
, "table=%"PRIu8
" ", fm
.table_id
);
5457 minimatch_format(&fm
.match
,
5458 ofproto_get_tun_tab(&ctx
->xin
->ofproto
->up
),
5459 NULL
, &s
, OFP_DEFAULT_PRIORITY
);
5461 ds_put_format(&s
, " priority=%d", fm
.priority
);
5462 if (fm
.new_cookie
) {
5463 ds_put_format(&s
, " cookie=%#"PRIx64
, ntohll(fm
.new_cookie
));
5465 if (fm
.idle_timeout
!= OFP_FLOW_PERMANENT
) {
5466 ds_put_format(&s
, " idle=%"PRIu16
, fm
.idle_timeout
);
5468 if (fm
.hard_timeout
!= OFP_FLOW_PERMANENT
) {
5469 ds_put_format(&s
, " hard=%"PRIu16
, fm
.hard_timeout
);
5471 if (fm
.flags
& NX_LEARN_F_SEND_FLOW_REM
) {
5472 ds_put_cstr(&s
, " send_flow_rem");
5474 ds_put_cstr(&s
, " actions=");
5475 struct ofpact_format_params fp
= { .s
= &s
};
5476 ofpacts_format(fm
.ofpacts
, fm
.ofpacts_len
, &fp
);
5477 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
5480 error
= ofproto_dpif_flow_mod_init_for_learn(ctx
->xbridge
->ofproto
,
5482 ofpbuf_uninit(&ofpacts
);
5485 bool success
= true;
5486 if (ctx
->xin
->allow_side_effects
) {
5487 error
= ofproto_flow_mod_learn(ofm
, ctx
->xin
->xcache
!= NULL
,
5488 learn
->limit
, &success
);
5489 } else if (learn
->limit
) {
5491 || ofm
->temp_rule
->state
!= RULE_INSERTED
) {
5492 /* The learned rule expired and there are no packets, so
5493 * we cannot learn again. Since the translated actions
5494 * depend on the result of learning, we tell the caller
5495 * that there's no point in caching this result. */
5496 ctx
->xout
->avoid_caching
= true;
5500 if (learn
->flags
& NX_LEARN_F_WRITE_RESULT
) {
5501 nxm_reg_load(&learn
->result_dst
, success
? 1 : 0,
5502 &ctx
->xin
->flow
, ctx
->wc
);
5503 xlate_report_subfield(ctx
, &learn
->result_dst
);
5506 if (success
&& ctx
->xin
->xcache
) {
5507 struct xc_entry
*entry
;
5509 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_LEARN
);
5510 entry
->learn
.ofm
= ofm
;
5511 entry
->learn
.limit
= learn
->limit
;
5514 ofproto_flow_mod_uninit(ofm
);
5517 if (OVS_UNLIKELY(ctx
->xin
->trace
&& !success
)) {
5518 xlate_report(ctx
, OFT_DETAIL
, "Limit exceeded, learn failed");
5522 if (ofm
!= &ofm__
) {
5527 xlate_report_error(ctx
, "LEARN action execution failed (%s).",
5528 ofperr_to_string(error
));
5531 minimatch_destroy(&fm
.match
);
5533 xlate_report(ctx
, OFT_WARN
,
5534 "suppressing side effects, so learn action ignored");
5539 xlate_fin_timeout__(struct rule_dpif
*rule
, uint16_t tcp_flags
,
5540 uint16_t idle_timeout
, uint16_t hard_timeout
)
5542 if (tcp_flags
& (TCP_FIN
| TCP_RST
)) {
5543 ofproto_rule_reduce_timeouts(&rule
->up
, idle_timeout
, hard_timeout
);
5548 xlate_fin_timeout(struct xlate_ctx
*ctx
,
5549 const struct ofpact_fin_timeout
*oft
)
5552 if (ctx
->xin
->allow_side_effects
) {
5553 xlate_fin_timeout__(ctx
->rule
, ctx
->xin
->tcp_flags
,
5554 oft
->fin_idle_timeout
, oft
->fin_hard_timeout
);
5556 if (ctx
->xin
->xcache
) {
5557 struct xc_entry
*entry
;
5559 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_FIN_TIMEOUT
);
5560 /* XC_RULE already holds a reference on the rule, none is taken
5562 entry
->fin
.rule
= ctx
->rule
;
5563 entry
->fin
.idle
= oft
->fin_idle_timeout
;
5564 entry
->fin
.hard
= oft
->fin_hard_timeout
;
5570 xlate_sample_action(struct xlate_ctx
*ctx
,
5571 const struct ofpact_sample
*os
)
5573 odp_port_t output_odp_port
= ODPP_NONE
;
5574 odp_port_t tunnel_out_port
= ODPP_NONE
;
5575 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
5576 bool emit_set_tunnel
= false;
5578 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
5582 /* Scale the probability from 16-bit to 32-bit while representing
5583 * the same percentage. */
5584 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
5586 /* If ofp_port in flow sample action is equel to ofp_port,
5587 * this sample action is a input port action. */
5588 if (os
->sampling_port
!= OFPP_NONE
&&
5589 os
->sampling_port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5590 output_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
5592 if (output_odp_port
== ODPP_NONE
) {
5593 xlate_report_error(ctx
, "can't use unknown port %d in flow sample "
5594 "action", os
->sampling_port
);
5598 if (dpif_ipfix_get_flow_exporter_tunnel_sampling(ipfix
,
5599 os
->collector_set_id
)
5600 && dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
)) {
5601 tunnel_out_port
= output_odp_port
;
5602 emit_set_tunnel
= true;
5606 xlate_commit_actions(ctx
);
5607 /* If 'emit_set_tunnel', sample(sampling_port=1) would translate
5608 * into datapath sample action set(tunnel(...)), sample(...) and
5609 * it is used for sampling egress tunnel information. */
5610 if (emit_set_tunnel
) {
5611 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
,
5614 if (xport
&& xport
->is_tunnel
) {
5615 struct flow
*flow
= &ctx
->xin
->flow
;
5616 tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
5617 if (!ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
5618 struct flow_tnl flow_tnl
= flow
->tunnel
;
5619 const char *tnl_type
;
5621 tnl_type
= tnl_port_get_type(xport
->ofport
);
5622 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
5623 ctx
->odp_actions
, tnl_type
);
5624 flow
->tunnel
= flow_tnl
;
5627 xlate_report_error(ctx
,
5628 "sampling_port:%d should be a tunnel port.",
5633 struct user_action_cookie cookie
;
5635 memset(&cookie
, 0, sizeof cookie
);
5636 cookie
.type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
5637 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
5638 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
5639 cookie
.flow_sample
.probability
= os
->probability
;
5640 cookie
.flow_sample
.collector_set_id
= os
->collector_set_id
;
5641 cookie
.flow_sample
.obs_domain_id
= os
->obs_domain_id
;
5642 cookie
.flow_sample
.obs_point_id
= os
->obs_point_id
;
5643 cookie
.flow_sample
.output_odp_port
= output_odp_port
;
5644 cookie
.flow_sample
.direction
= os
->direction
;
5646 compose_sample_action(ctx
, probability
, &cookie
, tunnel_out_port
, false);
5649 /* Determine if an datapath action translated from the openflow action
5650 * can be reversed by another datapath action.
5652 * Openflow actions that do not emit datapath actions are trivially
5653 * reversible. Reversiblity of other actions depends on nature of
5654 * action and their translation. */
5656 reversible_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5658 const struct ofpact
*a
;
5660 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5663 case OFPACT_CLEAR_ACTIONS
:
5665 case OFPACT_CONJUNCTION
:
5666 case OFPACT_CONTROLLER
:
5667 case OFPACT_CT_CLEAR
:
5668 case OFPACT_DEBUG_RECIRC
:
5669 case OFPACT_DEBUG_SLOW
:
5670 case OFPACT_DEC_MPLS_TTL
:
5671 case OFPACT_DEC_TTL
:
5672 case OFPACT_ENQUEUE
:
5674 case OFPACT_FIN_TIMEOUT
:
5675 case OFPACT_GOTO_TABLE
:
5678 case OFPACT_MULTIPATH
:
5681 case OFPACT_OUTPUT_REG
:
5682 case OFPACT_POP_MPLS
:
5683 case OFPACT_POP_QUEUE
:
5684 case OFPACT_PUSH_MPLS
:
5685 case OFPACT_PUSH_VLAN
:
5686 case OFPACT_REG_MOVE
:
5687 case OFPACT_RESUBMIT
:
5689 case OFPACT_SET_ETH_DST
:
5690 case OFPACT_SET_ETH_SRC
:
5691 case OFPACT_SET_FIELD
:
5692 case OFPACT_SET_IP_DSCP
:
5693 case OFPACT_SET_IP_ECN
:
5694 case OFPACT_SET_IP_TTL
:
5695 case OFPACT_SET_IPV4_DST
:
5696 case OFPACT_SET_IPV4_SRC
:
5697 case OFPACT_SET_L4_DST_PORT
:
5698 case OFPACT_SET_L4_SRC_PORT
:
5699 case OFPACT_SET_MPLS_LABEL
:
5700 case OFPACT_SET_MPLS_TC
:
5701 case OFPACT_SET_MPLS_TTL
:
5702 case OFPACT_SET_QUEUE
:
5703 case OFPACT_SET_TUNNEL
:
5704 case OFPACT_SET_VLAN_PCP
:
5705 case OFPACT_SET_VLAN_VID
:
5706 case OFPACT_STACK_POP
:
5707 case OFPACT_STACK_PUSH
:
5708 case OFPACT_STRIP_VLAN
:
5709 case OFPACT_UNROLL_XLATE
:
5710 case OFPACT_WRITE_ACTIONS
:
5711 case OFPACT_WRITE_METADATA
:
5712 case OFPACT_CHECK_PKT_LARGER
:
5713 case OFPACT_DELETE_FIELD
:
5719 case OFPACT_OUTPUT_TRUNC
:
5722 case OFPACT_DEC_NSH_TTL
:
5730 clone_xlate_actions(const struct ofpact
*actions
, size_t actions_len
,
5731 struct xlate_ctx
*ctx
, bool is_last_action
,
5732 bool group_bucket_action OVS_UNUSED
)
5734 struct ofpbuf old_stack
= ctx
->stack
;
5735 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
5736 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
5737 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
5739 struct ofpbuf old_action_set
= ctx
->action_set
;
5740 uint64_t actset_stub
[1024 / 8];
5741 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
5742 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
5744 size_t offset
, ac_offset
;
5745 struct flow old_flow
= ctx
->xin
->flow
;
5747 if (reversible_actions(actions
, actions_len
) || is_last_action
) {
5748 old_flow
= ctx
->xin
->flow
;
5749 do_xlate_actions(actions
, actions_len
, ctx
, is_last_action
, false);
5750 if (!ctx
->freezing
) {
5751 xlate_action_set(ctx
);
5753 if (ctx
->freezing
) {
5754 finish_freezing(ctx
);
5759 /* Commit datapath actions before emitting the clone action to
5760 * avoid emitting those actions twice. Once inside
5761 * the clone, another time for the action after clone. */
5762 xlate_commit_actions(ctx
);
5763 struct flow old_base
= ctx
->base_flow
;
5764 bool old_was_mpls
= ctx
->was_mpls
;
5765 bool old_conntracked
= ctx
->conntracked
;
5767 /* The actions are not reversible, a datapath clone action is
5768 * required to encode the translation. Select the clone action
5769 * based on datapath capabilities. */
5770 if (ctx
->xbridge
->support
.clone
) { /* Use clone action */
5771 /* Use clone action as datapath clone. */
5772 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
5773 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5774 if (!ctx
->freezing
) {
5775 xlate_action_set(ctx
);
5777 if (ctx
->freezing
) {
5778 finish_freezing(ctx
);
5780 nl_msg_end_non_empty_nested(ctx
->odp_actions
, offset
);
5784 if (ctx
->xbridge
->support
.sample_nesting
> 3) {
5785 /* Use sample action as datapath clone. */
5786 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
5787 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
5788 OVS_SAMPLE_ATTR_ACTIONS
);
5789 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5790 if (!ctx
->freezing
) {
5791 xlate_action_set(ctx
);
5793 if (ctx
->freezing
) {
5794 finish_freezing(ctx
);
5796 if (nl_msg_end_non_empty_nested(ctx
->odp_actions
, ac_offset
)) {
5797 nl_msg_cancel_nested(ctx
->odp_actions
, offset
);
5799 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
5800 UINT32_MAX
); /* 100% probability. */
5801 nl_msg_end_nested(ctx
->odp_actions
, offset
);
5806 /* Datapath does not support clone, skip xlate 'oc' and
5807 * report an error */
5808 xlate_report_error(ctx
, "Failed to compose clone action");
5811 /* The clone's conntrack execution should have no effect on the original
5813 ctx
->conntracked
= old_conntracked
;
5815 /* Popping MPLS from the clone should have no effect on the original
5817 ctx
->was_mpls
= old_was_mpls
;
5819 /* Restore the 'base_flow' for the next action. */
5820 ctx
->base_flow
= old_base
;
5823 ofpbuf_uninit(&ctx
->action_set
);
5824 ctx
->action_set
= old_action_set
;
5825 ofpbuf_uninit(&ctx
->stack
);
5826 ctx
->stack
= old_stack
;
5827 ctx
->xin
->flow
= old_flow
;
5831 compose_clone(struct xlate_ctx
*ctx
, const struct ofpact_nest
*oc
,
5832 bool is_last_action
)
5834 size_t oc_actions_len
= ofpact_nest_get_action_len(oc
);
5836 clone_xlate_actions(oc
->actions
, oc_actions_len
, ctx
, is_last_action
,
5841 xlate_meter_action(struct xlate_ctx
*ctx
, const struct ofpact_meter
*meter
)
5843 if (meter
->provider_meter_id
!= UINT32_MAX
) {
5844 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
,
5845 meter
->provider_meter_id
);
5850 may_receive(const struct xport
*xport
, struct xlate_ctx
*ctx
)
5852 if (xport
->config
& (is_stp(&ctx
->xin
->flow
)
5853 ? OFPUTIL_PC_NO_RECV_STP
5854 : OFPUTIL_PC_NO_RECV
)) {
5858 /* Only drop packets here if both forwarding and learning are
5859 * disabled. If just learning is enabled, we need to have
5860 * OFPP_NORMAL and the learning action have a look at the packet
5861 * before we can drop it. */
5862 if ((!xport_stp_forward_state(xport
) && !xport_stp_learn_state(xport
)) ||
5863 (!xport_rstp_forward_state(xport
) && !xport_rstp_learn_state(xport
))) {
5871 xlate_write_actions__(struct xlate_ctx
*ctx
,
5872 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5874 /* Maintain actset_output depending on the contents of the action set:
5876 * - OFPP_UNSET, if there is no "output" action.
5878 * - The output port, if there is an "output" action and no "group"
5881 * - OFPP_UNSET, if there is a "group" action.
5883 if (!ctx
->action_set_has_group
) {
5884 const struct ofpact
*a
;
5885 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5886 if (a
->type
== OFPACT_OUTPUT
) {
5887 ctx
->xin
->flow
.actset_output
= ofpact_get_OUTPUT(a
)->port
;
5888 } else if (a
->type
== OFPACT_GROUP
) {
5889 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
5890 ctx
->action_set_has_group
= true;
5896 ofpbuf_put(&ctx
->action_set
, ofpacts
, ofpacts_len
);
5900 xlate_write_actions(struct xlate_ctx
*ctx
, const struct ofpact_nest
*a
)
5902 xlate_write_actions__(ctx
, a
->actions
, ofpact_nest_get_action_len(a
));
5906 xlate_action_set(struct xlate_ctx
*ctx
)
5908 uint64_t action_list_stub
[1024 / 8];
5909 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
5910 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
5911 /* Clear the action set, as it is not needed any more. */
5912 ofpbuf_clear(&ctx
->action_set
);
5913 if (action_list
.size
) {
5914 ctx
->in_action_set
= true;
5916 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
5917 ctx
->xin
->trace
= xlate_report(ctx
, OFT_TABLE
,
5918 "--. Executing action set:");
5919 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, true, false);
5920 ctx
->xin
->trace
= old_trace
;
5922 ctx
->in_action_set
= false;
5924 ofpbuf_uninit(&action_list
);
5928 freeze_put_unroll_xlate(struct xlate_ctx
*ctx
)
5930 struct ofpact_unroll_xlate
*unroll
= ctx
->frozen_actions
.header
;
5932 /* Restore the table_id and rule cookie for a potential PACKET
5935 (ctx
->table_id
!= unroll
->rule_table_id
5936 || ctx
->rule_cookie
!= unroll
->rule_cookie
)) {
5937 unroll
= ofpact_put_UNROLL_XLATE(&ctx
->frozen_actions
);
5938 unroll
->rule_table_id
= ctx
->table_id
;
5939 unroll
->rule_cookie
= ctx
->rule_cookie
;
5940 ctx
->frozen_actions
.header
= unroll
;
5945 /* Copy actions 'a' through 'end' to ctx->frozen_actions, which will be
5946 * executed after thawing. Inserts an UNROLL_XLATE action, if none is already
5947 * present, before any action that may depend on the current table ID or flow
5950 freeze_unroll_actions(const struct ofpact
*a
, const struct ofpact
*end
,
5951 struct xlate_ctx
*ctx
)
5953 for (; a
< end
; a
= ofpact_next(a
)) {
5955 case OFPACT_OUTPUT_REG
:
5956 case OFPACT_OUTPUT_TRUNC
:
5959 case OFPACT_CONTROLLER
:
5960 case OFPACT_DEC_MPLS_TTL
:
5961 case OFPACT_DEC_NSH_TTL
:
5962 case OFPACT_DEC_TTL
:
5963 /* These actions may generate asynchronous messages, which include
5964 * table ID and flow cookie information. */
5965 freeze_put_unroll_xlate(ctx
);
5968 case OFPACT_RESUBMIT
:
5969 if (ofpact_get_RESUBMIT(a
)->table_id
== 0xff) {
5970 /* This resubmit action is relative to the current table, so we
5971 * need to track what table that is.*/
5972 freeze_put_unroll_xlate(ctx
);
5976 case OFPACT_SET_TUNNEL
:
5977 case OFPACT_REG_MOVE
:
5978 case OFPACT_SET_FIELD
:
5979 case OFPACT_STACK_PUSH
:
5980 case OFPACT_STACK_POP
:
5982 case OFPACT_WRITE_METADATA
:
5983 case OFPACT_GOTO_TABLE
:
5984 case OFPACT_ENQUEUE
:
5985 case OFPACT_SET_VLAN_VID
:
5986 case OFPACT_SET_VLAN_PCP
:
5987 case OFPACT_STRIP_VLAN
:
5988 case OFPACT_PUSH_VLAN
:
5989 case OFPACT_SET_ETH_SRC
:
5990 case OFPACT_SET_ETH_DST
:
5991 case OFPACT_SET_IPV4_SRC
:
5992 case OFPACT_SET_IPV4_DST
:
5993 case OFPACT_SET_IP_DSCP
:
5994 case OFPACT_SET_IP_ECN
:
5995 case OFPACT_SET_IP_TTL
:
5996 case OFPACT_SET_L4_SRC_PORT
:
5997 case OFPACT_SET_L4_DST_PORT
:
5998 case OFPACT_SET_QUEUE
:
5999 case OFPACT_POP_QUEUE
:
6000 case OFPACT_PUSH_MPLS
:
6001 case OFPACT_POP_MPLS
:
6002 case OFPACT_SET_MPLS_LABEL
:
6003 case OFPACT_SET_MPLS_TC
:
6004 case OFPACT_SET_MPLS_TTL
:
6005 case OFPACT_MULTIPATH
:
6008 case OFPACT_UNROLL_XLATE
:
6009 case OFPACT_FIN_TIMEOUT
:
6010 case OFPACT_CLEAR_ACTIONS
:
6011 case OFPACT_WRITE_ACTIONS
:
6017 case OFPACT_DEBUG_RECIRC
:
6018 case OFPACT_DEBUG_SLOW
:
6020 case OFPACT_CT_CLEAR
:
6022 case OFPACT_CHECK_PKT_LARGER
:
6023 case OFPACT_DELETE_FIELD
:
6024 /* These may not generate PACKET INs. */
6028 case OFPACT_CONJUNCTION
:
6029 /* These need not be copied for restoration. */
6032 /* Copy the action over. */
6033 ofpbuf_put(&ctx
->frozen_actions
, a
, OFPACT_ALIGN(a
->len
));
6038 put_ct_mark(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
6039 struct flow_wildcards
*wc
)
6041 if (wc
->masks
.ct_mark
) {
6047 odp_ct_mark
= nl_msg_put_unspec_uninit(odp_actions
, OVS_CT_ATTR_MARK
,
6048 sizeof(*odp_ct_mark
));
6049 odp_ct_mark
->key
= flow
->ct_mark
& wc
->masks
.ct_mark
;
6050 odp_ct_mark
->mask
= wc
->masks
.ct_mark
;
6055 put_ct_label(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
6056 struct flow_wildcards
*wc
)
6058 if (!ovs_u128_is_zero(wc
->masks
.ct_label
)) {
6064 odp_ct_label
.key
= ovs_u128_and(flow
->ct_label
, wc
->masks
.ct_label
);
6065 odp_ct_label
.mask
= wc
->masks
.ct_label
;
6066 nl_msg_put_unspec(odp_actions
, OVS_CT_ATTR_LABELS
,
6067 &odp_ct_label
, sizeof odp_ct_label
);
6072 put_drop_action(struct ofpbuf
*odp_actions
, enum xlate_error error
)
6074 nl_msg_put_u32(odp_actions
, OVS_ACTION_ATTR_DROP
, error
);
6078 put_ct_helper(struct xlate_ctx
*ctx
,
6079 struct ofpbuf
*odp_actions
, struct ofpact_conntrack
*ofc
)
6084 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "ftp");
6087 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "tftp");
6090 xlate_report_error(ctx
, "cannot serialize ct_helper %d", ofc
->alg
);
6097 put_ct_timeout(struct ofpbuf
*odp_actions
, const struct dpif_backer
*backer
,
6098 const struct flow
*flow
, struct flow_wildcards
*wc
,
6102 char *tp_name
= NULL
;
6104 if (ofproto_dpif_ct_zone_timeout_policy_get_name(backer
, zone_id
,
6105 ntohs(flow
->dl_type
), flow
->nw_proto
, &tp_name
, &unwildcard
)) {
6106 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_TIMEOUT
, tp_name
);
6109 /* The underlying datapath requires separate timeout
6110 * policies for different Ethertypes and IP protocols. We
6111 * don't need to unwildcard 'wc->masks.dl_type' since that
6112 * field is always unwildcarded in megaflows. */
6113 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6120 put_ct_nat(struct xlate_ctx
*ctx
)
6122 struct ofpact_nat
*ofn
= ctx
->ct_nat_action
;
6129 nat_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_CT_ATTR_NAT
);
6130 if (ofn
->flags
& NX_NAT_F_SRC
|| ofn
->flags
& NX_NAT_F_DST
) {
6131 nl_msg_put_flag(ctx
->odp_actions
, ofn
->flags
& NX_NAT_F_SRC
6132 ? OVS_NAT_ATTR_SRC
: OVS_NAT_ATTR_DST
);
6133 if (ofn
->flags
& NX_NAT_F_PERSISTENT
) {
6134 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PERSISTENT
);
6136 if (ofn
->flags
& NX_NAT_F_PROTO_HASH
) {
6137 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_HASH
);
6138 } else if (ofn
->flags
& NX_NAT_F_PROTO_RANDOM
) {
6139 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_RANDOM
);
6141 if (ofn
->range_af
== AF_INET
) {
6142 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6143 ofn
->range
.addr
.ipv4
.min
);
6144 if (ofn
->range
.addr
.ipv4
.max
&&
6145 (ntohl(ofn
->range
.addr
.ipv4
.max
)
6146 > ntohl(ofn
->range
.addr
.ipv4
.min
))) {
6147 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6148 ofn
->range
.addr
.ipv4
.max
);
6150 } else if (ofn
->range_af
== AF_INET6
) {
6151 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6152 &ofn
->range
.addr
.ipv6
.min
,
6153 sizeof ofn
->range
.addr
.ipv6
.min
);
6154 if (!ipv6_mask_is_any(&ofn
->range
.addr
.ipv6
.max
) &&
6155 memcmp(&ofn
->range
.addr
.ipv6
.max
, &ofn
->range
.addr
.ipv6
.min
,
6156 sizeof ofn
->range
.addr
.ipv6
.max
) > 0) {
6157 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6158 &ofn
->range
.addr
.ipv6
.max
,
6159 sizeof ofn
->range
.addr
.ipv6
.max
);
6162 if (ofn
->range_af
!= AF_UNSPEC
&& ofn
->range
.proto
.min
) {
6163 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MIN
,
6164 ofn
->range
.proto
.min
);
6165 if (ofn
->range
.proto
.max
&&
6166 ofn
->range
.proto
.max
> ofn
->range
.proto
.min
) {
6167 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MAX
,
6168 ofn
->range
.proto
.max
);
6172 nl_msg_end_nested(ctx
->odp_actions
, nat_offset
);
6176 compose_conntrack_action(struct xlate_ctx
*ctx
, struct ofpact_conntrack
*ofc
,
6177 bool is_last_action
)
6179 ovs_u128 old_ct_label_mask
= ctx
->wc
->masks
.ct_label
;
6180 uint32_t old_ct_mark_mask
= ctx
->wc
->masks
.ct_mark
;
6184 /* Ensure that any prior actions are applied before composing the new
6185 * conntrack action. */
6186 xlate_commit_actions(ctx
);
6188 /* Process nested actions first, to populate the key. */
6189 ctx
->ct_nat_action
= NULL
;
6190 ctx
->wc
->masks
.ct_mark
= 0;
6191 ctx
->wc
->masks
.ct_label
= OVS_U128_ZERO
;
6192 do_xlate_actions(ofc
->actions
, ofpact_ct_get_action_len(ofc
), ctx
,
6193 is_last_action
, false);
6195 if (ofc
->zone_src
.field
) {
6196 zone
= mf_get_subfield(&ofc
->zone_src
, &ctx
->xin
->flow
);
6198 zone
= ofc
->zone_imm
;
6201 ct_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CT
);
6202 if (ofc
->flags
& NX_CT_F_COMMIT
) {
6203 nl_msg_put_flag(ctx
->odp_actions
, ofc
->flags
& NX_CT_F_FORCE
?
6204 OVS_CT_ATTR_FORCE_COMMIT
: OVS_CT_ATTR_COMMIT
);
6205 if (ctx
->xbridge
->support
.ct_eventmask
) {
6206 nl_msg_put_u32(ctx
->odp_actions
, OVS_CT_ATTR_EVENTMASK
,
6207 OVS_CT_EVENTMASK_DEFAULT
);
6209 if (ctx
->xbridge
->support
.ct_timeout
) {
6210 put_ct_timeout(ctx
->odp_actions
, ctx
->xbridge
->ofproto
->backer
,
6211 &ctx
->xin
->flow
, ctx
->wc
, zone
);
6214 nl_msg_put_u16(ctx
->odp_actions
, OVS_CT_ATTR_ZONE
, zone
);
6215 put_ct_mark(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6216 put_ct_label(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6217 put_ct_helper(ctx
, ctx
->odp_actions
, ofc
);
6219 nl_msg_end_nested(ctx
->odp_actions
, ct_offset
);
6221 ctx
->wc
->masks
.ct_mark
= old_ct_mark_mask
;
6222 ctx
->wc
->masks
.ct_label
= old_ct_label_mask
;
6224 if (ofc
->recirc_table
!= NX_CT_RECIRC_NONE
) {
6225 ctx
->conntracked
= true;
6226 compose_recirculate_and_fork(ctx
, ofc
->recirc_table
, zone
);
6229 ctx
->ct_nat_action
= NULL
;
6231 /* The ct_* fields are only available in the scope of the 'recirc_table'
6233 flow_clear_conntrack(&ctx
->xin
->flow
);
6234 xlate_report(ctx
, OFT_DETAIL
, "Sets the packet to an untracked state, "
6235 "and clears all the conntrack fields.");
6236 ctx
->conntracked
= false;
6240 compose_ct_clear_action(struct xlate_ctx
*ctx
)
6242 clear_conntrack(ctx
);
6243 /* This action originally existed without dpif support. So to preserve
6244 * compatibility, only append it if the dpif supports it. */
6245 if (ctx
->xbridge
->support
.ct_clear
) {
6246 nl_msg_put_flag(ctx
->odp_actions
, OVS_ACTION_ATTR_CT_CLEAR
);
6250 /* check_pkt_larger action checks the packet length and stores the
6251 * result in the register bit. We translate this action to the
6252 * datapath action - 'check_pkt_len' whose format
6253 * is: 'check_pkt_len(pkt_len, ge(actions), le(actions))'.
6255 * We first set the destination register bit to 1 and call
6256 * 'do_xlate_actions' for the case - packet len greater than
6257 * the specified packet length.
6259 * We then set the destination register bit to 0 and call
6260 * 'do_xlate_actions' for the case - packet length is lesser or
6261 * equal to the specified packet length.
6263 * It is possible for freezing to happen for both the cases.
6266 xlate_check_pkt_larger(struct xlate_ctx
*ctx
,
6267 struct ofpact_check_pkt_larger
*check_pkt_larger
,
6268 const struct ofpact
*remaining_acts
,
6269 size_t remaining_acts_len
)
6271 union mf_subvalue value
;
6272 memset(&value
, 0, sizeof value
);
6273 if (!ctx
->xbridge
->support
.check_pkt_len
) {
6274 uint8_t is_pkt_larger
= 0;
6275 if (ctx
->xin
->packet
) {
6277 dp_packet_size(ctx
->xin
->packet
) > check_pkt_larger
->pkt_len
;
6279 value
.u8_val
= is_pkt_larger
;
6280 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
,
6282 /* If datapath doesn't support check_pkt_len action, then set the
6283 * SLOW_ACTION flag. If we don't set SLOW_ACTION, we
6284 * will push a flow to the datapath based on the packet length
6285 * in ctx->xin->packet. For subsequent patches which match the
6286 * same flow, datapath will apply the actions without considering
6287 * the packet length. This results in wrong actions being applied.
6289 ctx
->xout
->slow
|= SLOW_ACTION
;
6293 struct ofpbuf old_stack
= ctx
->stack
;
6294 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
6295 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
6296 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
6298 struct ofpbuf old_action_set
= ctx
->action_set
;
6299 uint64_t actset_stub
[1024 / 8];
6300 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
6301 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
6303 struct flow old_flow
= ctx
->xin
->flow
;
6304 xlate_commit_actions(ctx
);
6305 struct flow old_base
= ctx
->base_flow
;
6306 bool old_was_mpls
= ctx
->was_mpls
;
6307 bool old_conntracked
= ctx
->conntracked
;
6309 size_t offset
= nl_msg_start_nested(ctx
->odp_actions
,
6310 OVS_ACTION_ATTR_CHECK_PKT_LEN
);
6311 nl_msg_put_u16(ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN
,
6312 check_pkt_larger
->pkt_len
);
6313 size_t offset_attr
= nl_msg_start_nested(
6314 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER
);
6316 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6317 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6318 if (!ctx
->freezing
) {
6319 xlate_action_set(ctx
);
6321 if (ctx
->freezing
) {
6322 finish_freezing(ctx
);
6324 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6326 ctx
->base_flow
= old_base
;
6327 ctx
->was_mpls
= old_was_mpls
;
6328 ctx
->conntracked
= old_conntracked
;
6329 ctx
->xin
->flow
= old_flow
;
6331 /* If the flow translation for the IF_GREATER case requires freezing,
6332 * then ctx->exit would be true. Reset to false so that we can
6333 * do flow translation for 'IF_LESS_EQUAL' case. finish_freezing()
6334 * would have taken care of Undoing the changes done for freeze. */
6337 offset_attr
= nl_msg_start_nested(
6338 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL
);
6340 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6341 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6342 if (!ctx
->freezing
) {
6343 xlate_action_set(ctx
);
6345 if (ctx
->freezing
) {
6346 finish_freezing(ctx
);
6348 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6349 nl_msg_end_nested(ctx
->odp_actions
, offset
);
6351 ofpbuf_uninit(&ctx
->action_set
);
6352 ctx
->action_set
= old_action_set
;
6353 ofpbuf_uninit(&ctx
->stack
);
6354 ctx
->stack
= old_stack
;
6355 ctx
->base_flow
= old_base
;
6356 ctx
->was_mpls
= old_was_mpls
;
6357 ctx
->conntracked
= old_conntracked
;
6358 ctx
->xin
->flow
= old_flow
;
6363 rewrite_flow_encap_ethernet(struct xlate_ctx
*ctx
,
6365 struct flow_wildcards
*wc
)
6367 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6368 if (pt_ns(flow
->packet_type
) == OFPHTN_ETHERTYPE
) {
6369 /* Only adjust the packet_type and zero the dummy Ethernet addresses. */
6370 ovs_be16 ethertype
= pt_ns_type_be(flow
->packet_type
);
6371 flow
->packet_type
= htonl(PT_ETH
);
6372 flow
->dl_src
= eth_addr_zero
;
6373 flow
->dl_dst
= eth_addr_zero
;
6374 flow
->dl_type
= ethertype
;
6376 /* Error handling: drop packet. */
6377 xlate_report_debug(ctx
, OFT_ACTION
,
6378 "Dropping packet as encap(ethernet) is not "
6379 "supported for packet type ethernet.");
6380 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6384 /* For an MD2 NSH header returns a pointer to an ofpbuf with the encoded
6385 * MD2 TLVs provided as encap properties to the encap operation. This
6386 * will be stored as encap_data in the ctx and copied into the push_nsh
6387 * action at the next commit. */
6388 static struct ofpbuf
*
6389 rewrite_flow_push_nsh(struct xlate_ctx
*ctx
,
6390 const struct ofpact_encap
*encap
,
6392 struct flow_wildcards
*wc
)
6394 ovs_be32 packet_type
= flow
->packet_type
;
6395 const char *ptr
= (char *) encap
->props
;
6396 struct ofpbuf
*buf
= ofpbuf_new(NSH_CTX_HDRS_MAX_LEN
);
6397 uint8_t md_type
= NSH_M_TYPE1
;
6401 /* Scan the optional NSH encap TLV properties, if any. */
6402 for (i
= 0; i
< encap
->n_props
; i
++) {
6403 struct ofpact_ed_prop
*prop_ptr
=
6404 ALIGNED_CAST(struct ofpact_ed_prop
*, ptr
);
6405 if (prop_ptr
->prop_class
== OFPPPC_NSH
) {
6406 switch (prop_ptr
->type
) {
6407 case OFPPPT_PROP_NSH_MDTYPE
: {
6408 struct ofpact_ed_prop_nsh_md_type
*prop_md_type
=
6409 ALIGNED_CAST(struct ofpact_ed_prop_nsh_md_type
*,
6411 md_type
= prop_md_type
->md_type
;
6414 case OFPPPT_PROP_NSH_TLV
: {
6415 struct ofpact_ed_prop_nsh_tlv
*tlv_prop
=
6416 ALIGNED_CAST(struct ofpact_ed_prop_nsh_tlv
*,
6418 struct nsh_md2_tlv
*md2_ctx
=
6419 ofpbuf_put_uninit(buf
, sizeof(*md2_ctx
));
6420 md2_ctx
->md_class
= tlv_prop
->tlv_class
;
6421 md2_ctx
->type
= tlv_prop
->tlv_type
;
6422 md2_ctx
->length
= tlv_prop
->tlv_len
;
6423 size_t len
= ROUND_UP(md2_ctx
->length
, 4);
6424 size_t padding
= len
- md2_ctx
->length
;
6425 ofpbuf_put(buf
, tlv_prop
->data
, md2_ctx
->length
);
6426 ofpbuf_put_zeros(buf
, padding
);
6430 /* No other NSH encap properties defined yet. */
6434 ptr
+= ROUND_UP(prop_ptr
->len
, 8);
6436 if (buf
->size
== 0 || buf
->size
> NSH_CTX_HDRS_MAX_LEN
) {
6441 /* Determine the Next Protocol field for NSH header. */
6442 switch (ntohl(packet_type
)) {
6444 np
= NSH_P_ETHERNET
;
6456 /* Error handling: drop packet. */
6457 xlate_report_debug(ctx
, OFT_ACTION
,
6458 "Dropping packet as encap(nsh) is not "
6459 "supported for packet type (%d,0x%x)",
6460 pt_ns(packet_type
), pt_ns_type(packet_type
));
6461 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6464 /* Note that we have matched on packet_type! */
6465 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6467 /* Reset all current flow packet headers. */
6468 memset(&flow
->dl_dst
, 0,
6469 sizeof(struct flow
) - offsetof(struct flow
, dl_dst
));
6471 /* Populate the flow with the new NSH header. */
6472 flow
->packet_type
= htonl(PT_NSH
);
6473 flow
->dl_type
= htons(ETH_TYPE_NSH
);
6474 flow
->nsh
.flags
= 0;
6477 flow
->nsh
.path_hdr
= htonl(255);
6479 if (md_type
== NSH_M_TYPE1
) {
6480 flow
->nsh
.mdtype
= NSH_M_TYPE1
;
6481 memset(flow
->nsh
.context
, 0, sizeof flow
->nsh
.context
);
6483 /* Drop any MD2 context TLVs. */
6487 } else if (md_type
== NSH_M_TYPE2
) {
6488 flow
->nsh
.mdtype
= NSH_M_TYPE2
;
6490 flow
->nsh
.mdtype
&= NSH_MDTYPE_MASK
;
6496 xlate_generic_encap_action(struct xlate_ctx
*ctx
,
6497 const struct ofpact_encap
*encap
)
6499 struct flow
*flow
= &ctx
->xin
->flow
;
6500 struct flow_wildcards
*wc
= ctx
->wc
;
6501 struct ofpbuf
*encap_data
= NULL
;
6503 /* Ensure that any pending actions on the inner packet are applied before
6504 * rewriting the flow */
6505 xlate_commit_actions(ctx
);
6507 /* Rewrite the flow to reflect the effect of pushing the new encap header. */
6508 switch (ntohl(encap
->new_pkt_type
)) {
6510 rewrite_flow_encap_ethernet(ctx
, flow
, wc
);
6513 encap_data
= rewrite_flow_push_nsh(ctx
, encap
, flow
, wc
);
6516 /* New packet type was checked during decoding. */
6521 /* The actual encap datapath action will be generated at next commit. */
6522 ctx
->pending_encap
= true;
6523 ctx
->encap_data
= encap_data
;
6527 /* Returns true if packet must be recirculated after decapsulation. */
6529 xlate_generic_decap_action(struct xlate_ctx
*ctx
,
6530 const struct ofpact_decap
*decap OVS_UNUSED
)
6532 struct flow
*flow
= &ctx
->xin
->flow
;
6534 /* Ensure that any pending actions on the current packet are applied
6535 * before generating the decap action. */
6536 xlate_commit_actions(ctx
);
6538 /* We assume for now that the new_pkt_type is PT_USE_NEXT_PROTO. */
6539 switch (ntohl(flow
->packet_type
)) {
6541 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
)) {
6542 /* Error handling: drop packet. */
6543 xlate_report_debug(ctx
, OFT_ACTION
, "Dropping packet, cannot "
6544 "decap Ethernet if VLAN is present.");
6545 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6547 /* Just change the packet_type.
6548 * Delay generating pop_eth to the next commit. */
6549 flow
->packet_type
= htonl(PACKET_TYPE(OFPHTN_ETHERTYPE
,
6550 ntohs(flow
->dl_type
)));
6551 ctx
->wc
->masks
.dl_type
= OVS_BE16_MAX
;
6555 /* The pop_nsh action is generated at the commit executed as
6556 * part of freezing the ctx for recirculation. Here we just set
6557 * the new packet type based on the NSH next protocol field. */
6558 switch (flow
->nsh
.np
) {
6559 case NSH_P_ETHERNET
:
6560 flow
->packet_type
= htonl(PT_ETH
);
6563 flow
->packet_type
= htonl(PT_IPV4
);
6566 flow
->packet_type
= htonl(PT_IPV6
);
6569 flow
->packet_type
= htonl(PT_NSH
);
6572 /* Error handling: drop packet. */
6573 xlate_report_debug(ctx
, OFT_ACTION
,
6574 "Dropping packet as NSH next protocol %d "
6575 "is not supported", flow
->nsh
.np
);
6576 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6580 ctx
->wc
->masks
.nsh
.np
= UINT8_MAX
;
6581 ctx
->pending_decap
= true;
6582 /* Trigger recirculation. */
6585 /* Error handling: drop packet. */
6588 "Dropping packet as the decap() does not support "
6589 "packet type (%d,0x%x)",
6590 pt_ns(flow
->packet_type
), pt_ns_type(flow
->packet_type
));
6591 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6597 recirc_for_mpls(const struct ofpact
*a
, struct xlate_ctx
*ctx
)
6599 /* No need to recirculate if already exiting. */
6604 /* Do not consider recirculating unless the packet was previously MPLS. */
6605 if (!ctx
->was_mpls
) {
6609 /* Special case these actions, only recirculating if necessary.
6610 * This avoids the overhead of recirculation in common use-cases.
6614 /* Output actions do not require recirculation. */
6616 case OFPACT_OUTPUT_TRUNC
:
6617 case OFPACT_ENQUEUE
:
6618 case OFPACT_OUTPUT_REG
:
6619 /* Set actions that don't touch L3+ fields do not require recirculation. */
6620 case OFPACT_SET_VLAN_VID
:
6621 case OFPACT_SET_VLAN_PCP
:
6622 case OFPACT_SET_ETH_SRC
:
6623 case OFPACT_SET_ETH_DST
:
6624 case OFPACT_SET_TUNNEL
:
6625 case OFPACT_SET_QUEUE
:
6626 /* If actions of a group require recirculation that can be detected
6627 * when translating them. */
6631 /* Set field that don't touch L3+ fields don't require recirculation. */
6632 case OFPACT_SET_FIELD
:
6633 if (mf_is_l3_or_higher(ofpact_get_SET_FIELD(a
)->field
)) {
6638 /* For simplicity, recirculate in all other cases. */
6639 case OFPACT_CONTROLLER
:
6641 case OFPACT_STRIP_VLAN
:
6642 case OFPACT_PUSH_VLAN
:
6643 case OFPACT_SET_IPV4_SRC
:
6644 case OFPACT_SET_IPV4_DST
:
6645 case OFPACT_SET_IP_DSCP
:
6646 case OFPACT_SET_IP_ECN
:
6647 case OFPACT_SET_IP_TTL
:
6648 case OFPACT_SET_L4_SRC_PORT
:
6649 case OFPACT_SET_L4_DST_PORT
:
6650 case OFPACT_REG_MOVE
:
6651 case OFPACT_STACK_PUSH
:
6652 case OFPACT_STACK_POP
:
6653 case OFPACT_DEC_TTL
:
6654 case OFPACT_SET_MPLS_LABEL
:
6655 case OFPACT_SET_MPLS_TC
:
6656 case OFPACT_SET_MPLS_TTL
:
6657 case OFPACT_DEC_MPLS_TTL
:
6658 case OFPACT_PUSH_MPLS
:
6659 case OFPACT_POP_MPLS
:
6660 case OFPACT_POP_QUEUE
:
6661 case OFPACT_FIN_TIMEOUT
:
6662 case OFPACT_RESUBMIT
:
6664 case OFPACT_CONJUNCTION
:
6665 case OFPACT_MULTIPATH
:
6672 case OFPACT_DEC_NSH_TTL
:
6673 case OFPACT_UNROLL_XLATE
:
6675 case OFPACT_CT_CLEAR
:
6677 case OFPACT_DEBUG_RECIRC
:
6678 case OFPACT_DEBUG_SLOW
:
6680 case OFPACT_CLEAR_ACTIONS
:
6681 case OFPACT_WRITE_ACTIONS
:
6682 case OFPACT_WRITE_METADATA
:
6683 case OFPACT_GOTO_TABLE
:
6684 case OFPACT_CHECK_PKT_LARGER
:
6685 case OFPACT_DELETE_FIELD
:
6691 ctx_trigger_freeze(ctx
);
6695 xlate_ofpact_reg_move(struct xlate_ctx
*ctx
, const struct ofpact_reg_move
*a
)
6697 mf_subfield_copy(&a
->src
, &a
->dst
, &ctx
->xin
->flow
, ctx
->wc
);
6698 xlate_report_subfield(ctx
, &a
->dst
);
6702 xlate_ofpact_stack_pop(struct xlate_ctx
*ctx
, const struct ofpact_stack
*a
)
6704 if (nxm_execute_stack_pop(a
, &ctx
->xin
->flow
, ctx
->wc
, &ctx
->stack
)) {
6705 xlate_report_subfield(ctx
, &a
->subfield
);
6707 xlate_report_error(ctx
, "stack underflow");
6711 /* Restore translation context data that was stored earlier. */
6713 xlate_ofpact_unroll_xlate(struct xlate_ctx
*ctx
,
6714 const struct ofpact_unroll_xlate
*a
)
6716 ctx
->table_id
= a
->rule_table_id
;
6717 ctx
->rule_cookie
= a
->rule_cookie
;
6718 xlate_report(ctx
, OFT_THAW
, "restored state: table=%"PRIu8
", "
6719 "cookie=%#"PRIx64
, a
->rule_table_id
, a
->rule_cookie
);
6723 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
6724 struct xlate_ctx
*ctx
, bool is_last_action
,
6725 bool group_bucket_action
)
6727 struct flow_wildcards
*wc
= ctx
->wc
;
6728 struct flow
*flow
= &ctx
->xin
->flow
;
6729 const struct ofpact
*a
;
6731 /* dl_type already in the mask, not set below. */
6734 xlate_report(ctx
, OFT_ACTION
, "drop");
6738 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
6739 struct ofpact_controller
*controller
;
6740 const struct ofpact_metadata
*metadata
;
6741 const struct ofpact_set_field
*set_field
;
6742 const struct mf_field
*mf
;
6743 bool last
= is_last_action
&& ofpact_last(a
, ofpacts
, ofpacts_len
)
6744 && ctx
->action_set
.size
;
6750 recirc_for_mpls(a
, ctx
);
6753 /* Check if need to store the remaining actions for later
6755 if (ctx
->freezing
) {
6756 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
),
6762 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
6763 struct ds s
= DS_EMPTY_INITIALIZER
;
6764 struct ofpact_format_params fp
= { .s
= &s
};
6765 ofpacts_format(a
, OFPACT_ALIGN(a
->len
), &fp
);
6766 xlate_report(ctx
, OFT_ACTION
, "%s", ds_cstr(&s
));
6772 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
6773 ofpact_get_OUTPUT(a
)->max_len
, true, last
,
6774 false, group_bucket_action
);
6778 if (xlate_group_action(ctx
, ofpact_get_GROUP(a
)->group_id
, last
)) {
6779 /* Group could not be found. */
6781 /* XXX: Terminates action list translation, but does not
6782 * terminate the pipeline. */
6787 case OFPACT_CONTROLLER
:
6788 controller
= ofpact_get_CONTROLLER(a
);
6789 if (controller
->pause
) {
6790 ctx
->pause
= controller
;
6791 ctx_trigger_freeze(ctx
);
6794 xlate_controller_action(ctx
, controller
->max_len
,
6796 controller
->controller_id
,
6797 controller
->provider_meter_id
,
6798 controller
->userdata
,
6799 controller
->userdata_len
);
6803 case OFPACT_ENQUEUE
:
6804 memset(&wc
->masks
.skb_priority
, 0xff,
6805 sizeof wc
->masks
.skb_priority
);
6806 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
), last
,
6807 group_bucket_action
);
6810 case OFPACT_SET_VLAN_VID
:
6811 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
6812 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6813 ofpact_get_SET_VLAN_VID(a
)->push_vlan_if_needed
) {
6814 if (!flow
->vlans
[0].tpid
) {
6815 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6817 flow
->vlans
[0].tci
&= ~htons(VLAN_VID_MASK
);
6818 flow
->vlans
[0].tci
|=
6819 (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
) |
6824 case OFPACT_SET_VLAN_PCP
:
6825 wc
->masks
.vlans
[0].tci
|= htons(VLAN_PCP_MASK
| VLAN_CFI
);
6826 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6827 ofpact_get_SET_VLAN_PCP(a
)->push_vlan_if_needed
) {
6828 if (!flow
->vlans
[0].tpid
) {
6829 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6831 flow
->vlans
[0].tci
&= ~htons(VLAN_PCP_MASK
);
6832 flow
->vlans
[0].tci
|=
6833 htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
6834 << VLAN_PCP_SHIFT
) | VLAN_CFI
);
6838 case OFPACT_STRIP_VLAN
:
6839 flow_pop_vlan(flow
, wc
);
6842 case OFPACT_PUSH_VLAN
:
6843 flow_push_vlan_uninit(flow
, wc
);
6844 flow
->vlans
[0].tpid
= ofpact_get_PUSH_VLAN(a
)->ethertype
;
6845 flow
->vlans
[0].tci
= htons(VLAN_CFI
);
6848 case OFPACT_SET_ETH_SRC
:
6849 WC_MASK_FIELD(wc
, dl_src
);
6850 flow
->dl_src
= ofpact_get_SET_ETH_SRC(a
)->mac
;
6853 case OFPACT_SET_ETH_DST
:
6854 WC_MASK_FIELD(wc
, dl_dst
);
6855 flow
->dl_dst
= ofpact_get_SET_ETH_DST(a
)->mac
;
6858 case OFPACT_SET_IPV4_SRC
:
6859 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6860 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
6861 flow
->nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
6865 case OFPACT_SET_IPV4_DST
:
6866 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6867 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
6868 flow
->nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
6872 case OFPACT_SET_IP_DSCP
:
6873 if (is_ip_any(flow
)) {
6874 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
6875 flow
->nw_tos
&= ~IP_DSCP_MASK
;
6876 flow
->nw_tos
|= ofpact_get_SET_IP_DSCP(a
)->dscp
;
6880 case OFPACT_SET_IP_ECN
:
6881 if (is_ip_any(flow
)) {
6882 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
6883 flow
->nw_tos
&= ~IP_ECN_MASK
;
6884 flow
->nw_tos
|= ofpact_get_SET_IP_ECN(a
)->ecn
;
6888 case OFPACT_SET_IP_TTL
:
6889 if (is_ip_any(flow
)) {
6890 wc
->masks
.nw_ttl
= 0xff;
6891 flow
->nw_ttl
= ofpact_get_SET_IP_TTL(a
)->ttl
;
6895 case OFPACT_SET_L4_SRC_PORT
:
6896 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6897 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6898 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
6899 flow
->tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
6903 case OFPACT_SET_L4_DST_PORT
:
6904 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6905 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6906 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
6907 flow
->tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
6911 case OFPACT_RESUBMIT
:
6912 /* Freezing complicates resubmit. Some action in the flow
6913 * entry found by resubmit might trigger freezing. If that
6914 * happens, then we do not want to execute the resubmit again after
6915 * during thawing, so we want to skip back to the head of the loop
6916 * to avoid that, only adding any actions that follow the resubmit
6917 * to the frozen actions.
6919 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
), last
);
6922 case OFPACT_SET_TUNNEL
:
6923 flow
->tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
6926 case OFPACT_SET_QUEUE
:
6927 memset(&wc
->masks
.skb_priority
, 0xff,
6928 sizeof wc
->masks
.skb_priority
);
6929 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
6932 case OFPACT_POP_QUEUE
:
6933 memset(&wc
->masks
.skb_priority
, 0xff,
6934 sizeof wc
->masks
.skb_priority
);
6935 if (flow
->skb_priority
!= ctx
->orig_skb_priority
) {
6936 flow
->skb_priority
= ctx
->orig_skb_priority
;
6937 xlate_report(ctx
, OFT_DETAIL
, "queue = %#"PRIx32
,
6938 flow
->skb_priority
);
6942 case OFPACT_REG_MOVE
:
6943 xlate_ofpact_reg_move(ctx
, ofpact_get_REG_MOVE(a
));
6946 case OFPACT_SET_FIELD
:
6947 set_field
= ofpact_get_SET_FIELD(a
);
6948 mf
= set_field
->field
;
6950 /* Set the field only if the packet actually has it. */
6951 if (mf_are_prereqs_ok(mf
, flow
, wc
)) {
6952 mf_mask_field_masked(mf
, ofpact_set_field_mask(set_field
), wc
);
6953 mf_set_flow_value_masked(mf
, set_field
->value
,
6954 ofpact_set_field_mask(set_field
),
6957 xlate_report(ctx
, OFT_WARN
,
6958 "unmet prerequisites for %s, set_field ignored",
6964 case OFPACT_STACK_PUSH
:
6965 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), flow
, wc
,
6969 case OFPACT_STACK_POP
:
6970 xlate_ofpact_stack_pop(ctx
, ofpact_get_STACK_POP(a
));
6973 case OFPACT_PUSH_MPLS
:
6974 compose_mpls_push_action(ctx
, ofpact_get_PUSH_MPLS(a
));
6977 case OFPACT_POP_MPLS
:
6978 compose_mpls_pop_action(ctx
, ofpact_get_POP_MPLS(a
)->ethertype
);
6981 case OFPACT_SET_MPLS_LABEL
:
6982 compose_set_mpls_label_action(
6983 ctx
, ofpact_get_SET_MPLS_LABEL(a
)->label
);
6986 case OFPACT_SET_MPLS_TC
:
6987 compose_set_mpls_tc_action(ctx
, ofpact_get_SET_MPLS_TC(a
)->tc
);
6990 case OFPACT_SET_MPLS_TTL
:
6991 compose_set_mpls_ttl_action(ctx
, ofpact_get_SET_MPLS_TTL(a
)->ttl
);
6994 case OFPACT_DEC_MPLS_TTL
:
6995 if (compose_dec_mpls_ttl_action(ctx
)) {
7000 case OFPACT_DEC_NSH_TTL
:
7001 if (compose_dec_nsh_ttl_action(ctx
)) {
7006 case OFPACT_DEC_TTL
:
7007 wc
->masks
.nw_ttl
= 0xff;
7008 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
7014 /* Nothing to do. */
7017 case OFPACT_MULTIPATH
:
7018 multipath_execute(ofpact_get_MULTIPATH(a
), flow
, wc
);
7019 xlate_report_subfield(ctx
, &ofpact_get_MULTIPATH(a
)->dst
);
7023 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
), last
,
7024 group_bucket_action
);
7027 case OFPACT_OUTPUT_REG
:
7028 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
), last
,
7029 group_bucket_action
);
7032 case OFPACT_OUTPUT_TRUNC
:
7033 xlate_output_trunc_action(ctx
, ofpact_get_OUTPUT_TRUNC(a
)->port
,
7034 ofpact_get_OUTPUT_TRUNC(a
)->max_len
, last
,
7035 group_bucket_action
);
7039 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
7042 case OFPACT_CONJUNCTION
:
7043 /* A flow with a "conjunction" action represents part of a special
7044 * kind of "set membership match". Such a flow should not actually
7045 * get executed, but it could via, say, a "packet-out", even though
7046 * that wouldn't be useful. Log it to help debugging. */
7047 xlate_report_error(ctx
, "executing no-op conjunction action");
7054 case OFPACT_UNROLL_XLATE
:
7055 xlate_ofpact_unroll_xlate(ctx
, ofpact_get_UNROLL_XLATE(a
));
7058 case OFPACT_FIN_TIMEOUT
:
7059 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
7060 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
7063 case OFPACT_DELETE_FIELD
:
7064 xlate_delete_field(ctx
, flow
, ofpact_get_DELETE_FIELD(a
));
7067 case OFPACT_CLEAR_ACTIONS
:
7068 xlate_report_action_set(ctx
, "was");
7069 ofpbuf_clear(&ctx
->action_set
);
7070 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
7071 ctx
->action_set_has_group
= false;
7074 case OFPACT_WRITE_ACTIONS
:
7075 xlate_write_actions(ctx
, ofpact_get_WRITE_ACTIONS(a
));
7076 xlate_report_action_set(ctx
, "is");
7079 case OFPACT_WRITE_METADATA
:
7080 metadata
= ofpact_get_WRITE_METADATA(a
);
7081 flow
->metadata
&= ~metadata
->mask
;
7082 flow
->metadata
|= metadata
->metadata
& metadata
->mask
;
7086 xlate_meter_action(ctx
, ofpact_get_METER(a
));
7089 case OFPACT_GOTO_TABLE
: {
7090 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
7092 ovs_assert(ctx
->table_id
< ogt
->table_id
);
7094 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
7095 ogt
->table_id
, true, true, false, last
,
7101 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
7105 compose_clone(ctx
, ofpact_get_CLONE(a
), last
);
7109 xlate_generic_encap_action(ctx
, ofpact_get_ENCAP(a
));
7112 case OFPACT_DECAP
: {
7113 bool recirc_needed
=
7114 xlate_generic_decap_action(ctx
, ofpact_get_DECAP(a
));
7115 if (!ctx
->error
&& recirc_needed
) {
7116 /* Recirculate for parsing of inner packet. */
7117 ctx_trigger_freeze(ctx
);
7118 /* Then continue with next action. */
7125 compose_conntrack_action(ctx
, ofpact_get_CT(a
), last
);
7128 case OFPACT_CT_CLEAR
:
7129 compose_ct_clear_action(ctx
);
7133 /* This will be processed by compose_conntrack_action(). */
7134 ctx
->ct_nat_action
= ofpact_get_NAT(a
);
7137 case OFPACT_DEBUG_RECIRC
:
7138 ctx_trigger_freeze(ctx
);
7142 case OFPACT_DEBUG_SLOW
:
7143 ctx
->xout
->slow
|= SLOW_ACTION
;
7146 case OFPACT_CHECK_PKT_LARGER
: {
7148 /* If this is last action, then there is no need to
7149 * translate the action. */
7152 const struct ofpact
*remaining_acts
= ofpact_next(a
);
7153 size_t remaining_acts_len
= ofpact_remaining_len(remaining_acts
,
7156 xlate_check_pkt_larger(ctx
, ofpact_get_CHECK_PKT_LARGER(a
),
7157 remaining_acts
, remaining_acts_len
);
7162 /* Check if need to store this and the remaining actions for later
7164 if (!ctx
->error
&& ctx
->exit
&& ctx_first_frozen_action(ctx
)) {
7165 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
), ctx
);
7172 xlate_in_init(struct xlate_in
*xin
, struct ofproto_dpif
*ofproto
,
7173 ovs_version_t version
, const struct flow
*flow
,
7174 ofp_port_t in_port
, struct rule_dpif
*rule
, uint16_t tcp_flags
,
7175 const struct dp_packet
*packet
, struct flow_wildcards
*wc
,
7176 struct ofpbuf
*odp_actions
)
7178 xin
->ofproto
= ofproto
;
7179 xin
->tables_version
= version
;
7181 xin
->upcall_flow
= flow
;
7182 xin
->flow
.in_port
.ofp_port
= in_port
;
7183 xin
->flow
.actset_output
= OFPP_UNSET
;
7184 xin
->packet
= packet
;
7185 xin
->allow_side_effects
= packet
!= NULL
;
7188 xin
->ofpacts
= NULL
;
7189 xin
->ofpacts_len
= 0;
7190 xin
->tcp_flags
= tcp_flags
;
7192 xin
->resubmit_stats
= NULL
;
7196 xin
->odp_actions
= odp_actions
;
7197 xin
->in_packet_out
= false;
7198 xin
->recirc_queue
= NULL
;
7199 xin
->xport_uuid
= UUID_ZERO
;
7201 /* Do recirc lookup. */
7202 xin
->frozen_state
= NULL
;
7203 if (flow
->recirc_id
) {
7204 const struct recirc_id_node
*node
7205 = recirc_id_node_find(flow
->recirc_id
);
7207 xin
->frozen_state
= &node
->state
;
7213 xlate_out_uninit(struct xlate_out
*xout
)
7216 recirc_refs_unref(&xout
->recircs
);
7220 static struct skb_priority_to_dscp
*
7221 get_skb_priority(const struct xport
*xport
, uint32_t skb_priority
)
7223 struct skb_priority_to_dscp
*pdscp
;
7226 hash
= hash_int(skb_priority
, 0);
7227 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &xport
->skb_priorities
) {
7228 if (pdscp
->skb_priority
== skb_priority
) {
7236 dscp_from_skb_priority(const struct xport
*xport
, uint32_t skb_priority
,
7239 struct skb_priority_to_dscp
*pdscp
= get_skb_priority(xport
, skb_priority
);
7240 *dscp
= pdscp
? pdscp
->dscp
: 0;
7241 return pdscp
!= NULL
;
7245 count_skb_priorities(const struct xport
*xport
)
7247 return hmap_count(&xport
->skb_priorities
);
7251 clear_skb_priorities(struct xport
*xport
)
7253 struct skb_priority_to_dscp
*pdscp
;
7255 HMAP_FOR_EACH_POP (pdscp
, hmap_node
, &xport
->skb_priorities
) {
7261 actions_output_to_local_port(const struct xlate_ctx
*ctx
)
7263 odp_port_t local_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
, OFPP_LOCAL
);
7264 const struct nlattr
*a
;
7267 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, ctx
->odp_actions
->data
,
7268 ctx
->odp_actions
->size
) {
7269 if (nl_attr_type(a
) == OVS_ACTION_ATTR_OUTPUT
7270 && nl_attr_get_odp_port(a
) == local_odp_port
) {
7277 #if defined(__linux__)
7278 /* Returns the maximum number of packets that the Linux kernel is willing to
7279 * queue up internally to certain kinds of software-implemented ports, or the
7280 * default (and rarely modified) value if it cannot be determined. */
7282 netdev_max_backlog(void)
7284 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
7285 static int max_backlog
= 1000; /* The normal default value. */
7287 if (ovsthread_once_start(&once
)) {
7288 static const char filename
[] = "/proc/sys/net/core/netdev_max_backlog";
7292 stream
= fopen(filename
, "r");
7294 VLOG_INFO("%s: open failed (%s)", filename
, ovs_strerror(errno
));
7296 if (fscanf(stream
, "%d", &n
) != 1) {
7297 VLOG_WARN("%s: read error", filename
);
7298 } else if (n
<= 100) {
7299 VLOG_WARN("%s: unexpectedly small value %d", filename
, n
);
7305 ovsthread_once_done(&once
);
7307 VLOG_DBG("%s: using %d max_backlog", filename
, max_backlog
);
7313 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
7316 count_output_actions(const struct ofpbuf
*odp_actions
)
7318 const struct nlattr
*a
;
7322 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, odp_actions
->data
, odp_actions
->size
) {
7323 if ((a
->nla_type
== OVS_ACTION_ATTR_OUTPUT
) ||
7324 (a
->nla_type
== OVS_ACTION_ATTR_LB_OUTPUT
)) {
7330 #endif /* defined(__linux__) */
7332 /* Returns true if 'odp_actions' contains more output actions than the datapath
7333 * can reliably handle in one go. On Linux, this is the value of the
7334 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
7335 * packets that the kernel is willing to queue up for processing while the
7336 * datapath is processing a set of actions. */
7338 too_many_output_actions(const struct ofpbuf
*odp_actions OVS_UNUSED
)
7341 return (odp_actions
->size
/ NL_A_U32_SIZE
> netdev_max_backlog()
7342 && count_output_actions(odp_actions
) > netdev_max_backlog());
7344 /* OSes other than Linux might have similar limits, but we don't know how
7345 * to determine them.*/
7351 xlate_wc_init(struct xlate_ctx
*ctx
)
7353 flow_wildcards_init_catchall(ctx
->wc
);
7355 /* Some fields we consider to always be examined. */
7356 WC_MASK_FIELD(ctx
->wc
, packet_type
);
7357 WC_MASK_FIELD(ctx
->wc
, in_port
);
7358 WC_MASK_FIELD(ctx
->wc
, dl_type
);
7359 if (is_ip_any(&ctx
->xin
->flow
)) {
7360 WC_MASK_FIELD_MASK(ctx
->wc
, nw_frag
, FLOW_NW_FRAG_MASK
);
7363 if (ctx
->xbridge
->support
.odp
.recirc
) {
7364 /* Always exactly match recirc_id when datapath supports
7366 WC_MASK_FIELD(ctx
->wc
, recirc_id
);
7369 if (ctx
->xbridge
->netflow
) {
7370 netflow_mask_wc(&ctx
->xin
->flow
, ctx
->wc
);
7373 tnl_wc_init(&ctx
->xin
->flow
, ctx
->wc
);
7377 xlate_wc_finish(struct xlate_ctx
*ctx
)
7381 /* Clear the metadata and register wildcard masks, because we won't
7382 * use non-header fields as part of the cache. */
7383 flow_wildcards_clear_non_packet_fields(ctx
->wc
);
7385 /* Wildcard Ethernet address fields if the original packet type was not
7388 * (The Ethertype field is used even when the original packet type is not
7390 if (ctx
->xin
->upcall_flow
->packet_type
!= htonl(PT_ETH
)) {
7391 ctx
->wc
->masks
.dl_dst
= eth_addr_zero
;
7392 ctx
->wc
->masks
.dl_src
= eth_addr_zero
;
7395 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
7396 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
7397 * represent these fields. The datapath interface, on the other hand,
7398 * represents them with just 8 bits each. This means that if the high
7399 * 8 bits of the masks for these fields somehow become set, then they
7400 * will get chopped off by a round trip through the datapath, and
7401 * revalidation will spot that as an inconsistency and delete the flow.
7402 * Avoid the problem here by making sure that only the low 8 bits of
7403 * either field can be unwildcarded for ICMP.
7405 if (is_icmpv4(&ctx
->xin
->flow
, NULL
) || is_icmpv6(&ctx
->xin
->flow
, NULL
)) {
7406 ctx
->wc
->masks
.tp_src
&= htons(UINT8_MAX
);
7407 ctx
->wc
->masks
.tp_dst
&= htons(UINT8_MAX
);
7409 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
7410 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
7411 if (ctx
->wc
->masks
.vlans
[i
].tci
) {
7412 ctx
->wc
->masks
.vlans
[i
].tci
|= htons(VLAN_CFI
);
7416 /* The classifier might return masks that match on tp_src and tp_dst even
7417 * for later fragments. This happens because there might be flows that
7418 * match on tp_src or tp_dst without matching on the frag bits, because
7419 * it is not a prerequisite for OpenFlow. Since it is a prerequisite for
7420 * datapath flows and since tp_src and tp_dst are always going to be 0,
7421 * wildcard the fields here. */
7422 if (ctx
->xin
->flow
.nw_frag
& FLOW_NW_FRAG_LATER
) {
7423 ctx
->wc
->masks
.tp_src
= 0;
7424 ctx
->wc
->masks
.tp_dst
= 0;
7427 /* Clear flow wildcard bits for fields which are not present
7428 * in the original packet header. These wildcards may get set
7429 * due to push/set_field actions. This results into frequent
7430 * invalidation of datapath flows by revalidator thread. */
7432 /* Clear mpls label wc bits if original packet is non-mpls. */
7433 if (!eth_type_mpls(ctx
->xin
->upcall_flow
->dl_type
)) {
7434 for (i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
7435 ctx
->wc
->masks
.mpls_lse
[i
] = 0;
7438 /* Clear vlan header wc bits if original packet does not have
7440 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
7441 if (!eth_type_vlan(ctx
->xin
->upcall_flow
->vlans
[i
].tpid
)) {
7442 ctx
->wc
->masks
.vlans
[i
].tpid
= 0;
7443 ctx
->wc
->masks
.vlans
[i
].tci
= 0;
7448 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
7450 * The caller must take responsibility for eventually freeing 'xout', with
7451 * xlate_out_uninit().
7452 * Returns 'XLATE_OK' if translation was successful. In case of an error an
7453 * empty set of actions will be returned in 'xin->odp_actions' (if non-NULL),
7454 * so that most callers may ignore the return value and transparently install a
7455 * drop flow when the translation fails. */
7457 xlate_actions(struct xlate_in
*xin
, struct xlate_out
*xout
)
7459 *xout
= (struct xlate_out
) {
7461 .recircs
= RECIRC_REFS_EMPTY_INITIALIZER
,
7464 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7465 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, xin
->ofproto
);
7467 return XLATE_BRIDGE_NOT_FOUND
;
7470 struct flow
*flow
= &xin
->flow
;
7472 uint8_t stack_stub
[1024];
7473 uint64_t action_set_stub
[1024 / 8];
7474 uint64_t frozen_actions_stub
[1024 / 8];
7475 uint64_t actions_stub
[256 / 8];
7476 struct ofpbuf scratch_actions
= OFPBUF_STUB_INITIALIZER(actions_stub
);
7477 struct xlate_ctx ctx
= {
7481 .orig_tunnel_ipv6_dst
= flow_tnl_dst(&flow
->tunnel
),
7484 .stack
= OFPBUF_STUB_INITIALIZER(stack_stub
),
7488 : &(struct flow_wildcards
) { .masks
= { .dl_type
= 0 } }),
7489 .odp_actions
= xin
->odp_actions
? xin
->odp_actions
: &scratch_actions
,
7491 .depth
= xin
->depth
,
7492 .resubmits
= xin
->resubmits
,
7493 .in_action_set
= false,
7494 .in_packet_out
= xin
->in_packet_out
,
7495 .pending_encap
= false,
7496 .pending_decap
= false,
7500 .rule_cookie
= OVS_BE64_MAX
,
7501 .orig_skb_priority
= flow
->skb_priority
,
7502 .sflow_n_outputs
= 0,
7503 .sflow_odp_port
= 0,
7504 .nf_output_iface
= NF_OUT_DROP
,
7510 .recirc_update_dp_hash
= false,
7511 .frozen_actions
= OFPBUF_STUB_INITIALIZER(frozen_actions_stub
),
7515 .conntracked
= false,
7517 .ct_nat_action
= NULL
,
7519 .action_set_has_group
= false,
7520 .action_set
= OFPBUF_STUB_INITIALIZER(action_set_stub
),
7523 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
7524 * the packet as the datapath will treat it for output actions. Our
7525 * datapath doesn't retain tunneling information without us re-setting
7526 * it, so clear the tunnel data.
7529 memset(&ctx
.base_flow
.tunnel
, 0, sizeof ctx
.base_flow
.tunnel
);
7531 ofpbuf_reserve(ctx
.odp_actions
, NL_A_U32_SIZE
);
7532 xlate_wc_init(&ctx
);
7534 COVERAGE_INC(xlate_actions
);
7536 xin
->trace
= xlate_report(&ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
7538 if (xin
->frozen_state
) {
7539 const struct frozen_state
*state
= xin
->frozen_state
;
7541 struct ovs_list
*old_trace
= xin
->trace
;
7542 xin
->trace
= xlate_report(&ctx
, OFT_THAW
, "thaw");
7544 if (xin
->ofpacts_len
> 0 || ctx
.rule
) {
7545 xlate_report_error(&ctx
, "Recirculation conflict (%s)!",
7546 xin
->ofpacts_len
? "actions" : "rule");
7547 ctx
.error
= XLATE_RECIRCULATION_CONFLICT
;
7551 /* Set the bridge for post-recirculation processing if needed. */
7552 if (!uuid_equals(&ctx
.xbridge
->ofproto
->uuid
, &state
->ofproto_uuid
)) {
7553 const struct xbridge
*new_bridge
7554 = xbridge_lookup_by_uuid(xcfg
, &state
->ofproto_uuid
);
7556 if (OVS_UNLIKELY(!new_bridge
)) {
7557 /* Drop the packet if the bridge cannot be found. */
7558 xlate_report_error(&ctx
, "Frozen bridge no longer exists.");
7559 ctx
.error
= XLATE_BRIDGE_NOT_FOUND
;
7560 xin
->trace
= old_trace
;
7563 ctx
.xbridge
= new_bridge
;
7564 /* The bridge is now known so obtain its table version. */
7565 ctx
.xin
->tables_version
7566 = ofproto_dpif_get_tables_version(ctx
.xbridge
->ofproto
);
7569 /* Set the thawed table id. Note: A table lookup is done only if there
7570 * are no frozen actions. */
7571 ctx
.table_id
= state
->table_id
;
7572 xlate_report(&ctx
, OFT_THAW
,
7573 "Resuming from table %"PRIu8
, ctx
.table_id
);
7575 ctx
.conntracked
= state
->conntracked
;
7576 if (!state
->conntracked
) {
7577 clear_conntrack(&ctx
);
7580 /* Restore pipeline metadata. May change flow's in_port and other
7581 * metadata to the values that existed when freezing was triggered. */
7582 frozen_metadata_to_flow(&ctx
.xbridge
->ofproto
->up
,
7583 &state
->metadata
, flow
);
7585 /* Restore stack, if any. */
7587 ofpbuf_put(&ctx
.stack
, state
->stack
, state
->stack_size
);
7590 /* Restore mirror state. */
7591 ctx
.mirrors
= state
->mirrors
;
7593 /* Restore action set, if any. */
7594 if (state
->action_set_len
) {
7595 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring action set",
7596 state
->action_set
, state
->action_set_len
);
7598 flow
->actset_output
= OFPP_UNSET
;
7599 xlate_write_actions__(&ctx
, state
->action_set
,
7600 state
->action_set_len
);
7603 /* Restore frozen actions. If there are no actions, processing will
7604 * start with a lookup in the table set above. */
7605 xin
->ofpacts
= state
->ofpacts
;
7606 xin
->ofpacts_len
= state
->ofpacts_len
;
7607 if (state
->ofpacts_len
) {
7608 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring actions",
7609 xin
->ofpacts
, xin
->ofpacts_len
);
7612 xin
->trace
= old_trace
;
7613 } else if (OVS_UNLIKELY(flow
->recirc_id
)) {
7614 xlate_report_error(&ctx
,
7615 "Recirculation context not found for ID %"PRIx32
,
7617 ctx
.error
= XLATE_NO_RECIRCULATION_CONTEXT
;
7621 /* Tunnel metadata in udpif format must be normalized before translation. */
7622 if (flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7623 const struct tun_table
*tun_tab
= ofproto_get_tun_tab(
7624 &ctx
.xbridge
->ofproto
->up
);
7627 err
= tun_metadata_from_geneve_udpif(tun_tab
, &xin
->upcall_flow
->tunnel
,
7628 &xin
->upcall_flow
->tunnel
,
7631 xlate_report_error(&ctx
, "Invalid Geneve tunnel metadata");
7632 ctx
.error
= XLATE_INVALID_TUNNEL_METADATA
;
7635 } else if (!flow
->tunnel
.metadata
.tab
) {
7636 /* If the original flow did not come in on a tunnel, then it won't have
7637 * FLOW_TNL_F_UDPIF set. However, we still need to have a metadata
7638 * table in case we generate tunnel actions. */
7639 flow
->tunnel
.metadata
.tab
= ofproto_get_tun_tab(
7640 &ctx
.xbridge
->ofproto
->up
);
7642 ctx
.wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
7644 /* Get the proximate input port of the packet. (If xin->frozen_state,
7645 * flow->in_port is the ultimate input port of the packet.) */
7646 struct xport
*in_port
= get_ofp_port(xbridge
,
7647 ctx
.base_flow
.in_port
.ofp_port
);
7648 if (in_port
&& !in_port
->peer
) {
7649 ctx
.xin
->xport_uuid
= in_port
->uuid
;
7652 if (flow
->packet_type
!= htonl(PT_ETH
) && in_port
&&
7653 in_port
->pt_mode
== NETDEV_PT_LEGACY_L3
&& ctx
.table_id
== 0) {
7654 /* Add dummy Ethernet header to non-L2 packet if it's coming from a
7655 * L3 port. So all packets will be L2 packets for lookup.
7656 * The dl_type has already been set from the packet_type. */
7657 flow
->packet_type
= htonl(PT_ETH
);
7658 flow
->dl_src
= eth_addr_zero
;
7659 flow
->dl_dst
= eth_addr_zero
;
7660 ctx
.pending_encap
= true;
7663 if (!xin
->ofpacts
&& !ctx
.rule
) {
7664 ctx
.rule
= rule_dpif_lookup_from_table(
7665 ctx
.xbridge
->ofproto
, ctx
.xin
->tables_version
, flow
, ctx
.wc
,
7666 ctx
.xin
->resubmit_stats
, &ctx
.table_id
,
7667 flow
->in_port
.ofp_port
, true, true, ctx
.xin
->xcache
);
7668 if (ctx
.xin
->resubmit_stats
) {
7669 rule_dpif_credit_stats(ctx
.rule
, ctx
.xin
->resubmit_stats
, false);
7671 if (ctx
.xin
->xcache
) {
7672 struct xc_entry
*entry
;
7674 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_RULE
);
7675 entry
->rule
= ctx
.rule
;
7676 ofproto_rule_ref(&ctx
.rule
->up
);
7679 xlate_report_table(&ctx
, ctx
.rule
, ctx
.table_id
);
7682 /* Tunnel stats only for not-thawed packets. */
7683 if (!xin
->frozen_state
&& in_port
&& in_port
->is_tunnel
) {
7684 if (ctx
.xin
->resubmit_stats
) {
7685 netdev_vport_inc_rx(in_port
->netdev
, ctx
.xin
->resubmit_stats
);
7687 bfd_account_rx(in_port
->bfd
, ctx
.xin
->resubmit_stats
);
7690 if (ctx
.xin
->xcache
) {
7691 struct xc_entry
*entry
;
7693 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETDEV
);
7694 entry
->dev
.rx
= netdev_ref(in_port
->netdev
);
7695 entry
->dev
.bfd
= bfd_ref(in_port
->bfd
);
7699 if (!xin
->frozen_state
&& process_special(&ctx
, in_port
)) {
7700 /* process_special() did all the processing for this packet.
7702 * We do not perform special processing on thawed packets, since that
7703 * was done before they were frozen and should not be redone. */
7704 mirror_ingress_packet(&ctx
);
7705 } else if (in_port
&& in_port
->xbundle
7706 && xbundle_mirror_out(xbridge
, in_port
->xbundle
)) {
7707 xlate_report_error(&ctx
, "dropping packet received on port "
7708 "%s, which is reserved exclusively for mirroring",
7709 in_port
->xbundle
->name
);
7711 /* Sampling is done on initial reception; don't redo after thawing. */
7712 unsigned int user_cookie_offset
= 0;
7713 if (!xin
->frozen_state
) {
7714 user_cookie_offset
= compose_sflow_action(&ctx
);
7715 compose_ipfix_action(&ctx
, ODPP_NONE
);
7717 size_t sample_actions_len
= ctx
.odp_actions
->size
;
7718 bool ecn_drop
= !tnl_process_ecn(flow
);
7721 && (!in_port
|| may_receive(in_port
, &ctx
))) {
7722 const struct ofpact
*ofpacts
;
7726 ofpacts
= xin
->ofpacts
;
7727 ofpacts_len
= xin
->ofpacts_len
;
7728 } else if (ctx
.rule
) {
7729 const struct rule_actions
*actions
7730 = rule_get_actions(&ctx
.rule
->up
);
7731 ofpacts
= actions
->ofpacts
;
7732 ofpacts_len
= actions
->ofpacts_len
;
7733 ctx
.rule_cookie
= ctx
.rule
->up
.flow_cookie
;
7738 mirror_ingress_packet(&ctx
);
7739 do_xlate_actions(ofpacts
, ofpacts_len
, &ctx
, true, false);
7744 /* We've let OFPP_NORMAL and the learning action look at the
7745 * packet, so cancel all actions and freezing if forwarding is
7747 if (in_port
&& (!xport_stp_forward_state(in_port
) ||
7748 !xport_rstp_forward_state(in_port
))) {
7749 ctx
.odp_actions
->size
= sample_actions_len
;
7750 ctx_cancel_freeze(&ctx
);
7751 ofpbuf_clear(&ctx
.action_set
);
7752 ctx
.error
= XLATE_FORWARDING_DISABLED
;
7755 if (!ctx
.freezing
) {
7756 xlate_action_set(&ctx
);
7759 finish_freezing(&ctx
);
7761 } else if (ecn_drop
) {
7762 ctx
.error
= XLATE_CONGESTION_DROP
;
7765 /* Output only fully processed packets. */
7767 && xbridge
->has_in_band
7768 && in_band_must_output_to_local_port(flow
)
7769 && !actions_output_to_local_port(&ctx
)) {
7770 WC_MASK_FIELD(ctx
.wc
, nw_proto
);
7771 WC_MASK_FIELD(ctx
.wc
, tp_src
);
7772 WC_MASK_FIELD(ctx
.wc
, tp_dst
);
7773 WC_MASK_FIELD(ctx
.wc
, dl_type
);
7774 xlate_report(&ctx
, OFT_DETAIL
, "outputting DHCP packet "
7775 "to local port for in-band control");
7776 compose_output_action(&ctx
, OFPP_LOCAL
, NULL
, false, false);
7779 if (user_cookie_offset
) {
7780 fix_sflow_action(&ctx
, user_cookie_offset
);
7784 if (nl_attr_oversized(ctx
.odp_actions
->size
)) {
7785 /* These datapath actions are too big for a Netlink attribute, so we
7786 * can't hand them to the kernel directly. dpif_execute() can execute
7787 * them one by one with help, so just mark the result as SLOW_ACTION to
7788 * prevent the flow from being installed. */
7789 COVERAGE_INC(xlate_actions_oversize
);
7790 ctx
.xout
->slow
|= SLOW_ACTION
;
7791 } else if (too_many_output_actions(ctx
.odp_actions
)) {
7792 COVERAGE_INC(xlate_actions_too_many_output
);
7793 ctx
.xout
->slow
|= SLOW_ACTION
;
7796 /* Update NetFlow for non-frozen traffic. */
7797 if (xbridge
->netflow
&& !xin
->frozen_state
) {
7798 if (ctx
.xin
->resubmit_stats
) {
7799 netflow_flow_update(xbridge
->netflow
, flow
,
7800 ctx
.nf_output_iface
,
7801 ctx
.xin
->resubmit_stats
);
7803 if (ctx
.xin
->xcache
) {
7804 struct xc_entry
*entry
;
7806 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETFLOW
);
7807 entry
->nf
.netflow
= netflow_ref(xbridge
->netflow
);
7808 entry
->nf
.flow
= xmemdup(flow
, sizeof *flow
);
7809 entry
->nf
.iface
= ctx
.nf_output_iface
;
7813 /* Translate tunnel metadata masks to udpif format if necessary. */
7814 if (xin
->upcall_flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7815 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7816 const struct flow_tnl
*upcall_tnl
= &xin
->upcall_flow
->tunnel
;
7817 struct geneve_opt opts
[TLV_TOT_OPT_SIZE
/
7818 sizeof(struct geneve_opt
)];
7820 tun_metadata_to_geneve_udpif_mask(&flow
->tunnel
,
7821 &ctx
.wc
->masks
.tunnel
,
7822 upcall_tnl
->metadata
.opts
.gnv
,
7823 upcall_tnl
->metadata
.present
.len
,
7825 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7826 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7827 memcpy(&ctx
.wc
->masks
.tunnel
.metadata
.opts
.gnv
, opts
,
7828 upcall_tnl
->metadata
.present
.len
);
7830 ctx
.wc
->masks
.tunnel
.metadata
.present
.len
= 0xff;
7831 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7832 ctx
.wc
->masks
.tunnel
.flags
|= FLOW_TNL_F_UDPIF
;
7833 } else if (!xin
->upcall_flow
->tunnel
.metadata
.tab
) {
7834 /* If we didn't have options in UDPIF format and didn't have an existing
7835 * metadata table, then it means that there were no options at all when
7836 * we started processing and any wildcards we picked up were from
7837 * action generation. Without options on the incoming packet, wildcards
7838 * aren't meaningful. To avoid them possibly getting misinterpreted,
7839 * just clear everything. */
7840 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7841 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7842 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7844 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7848 xlate_wc_finish(&ctx
);
7851 /* Reset the table to what it was when we came in. If we only fetched
7852 * it locally, then it has no meaning outside of flow translation. */
7853 flow
->tunnel
.metadata
.tab
= xin
->upcall_flow
->tunnel
.metadata
.tab
;
7855 ofpbuf_uninit(&ctx
.stack
);
7856 ofpbuf_uninit(&ctx
.action_set
);
7857 ofpbuf_uninit(&ctx
.frozen_actions
);
7858 ofpbuf_uninit(&scratch_actions
);
7859 ofpbuf_delete(ctx
.encap_data
);
7861 /* Make sure we return a "drop flow" in case of an error. */
7864 if (xin
->odp_actions
) {
7865 ofpbuf_clear(xin
->odp_actions
);
7869 /* Install drop action if datapath supports explicit drop action. */
7870 if (xin
->odp_actions
&& !xin
->odp_actions
->size
&&
7871 ovs_explicit_drop_action_supported(ctx
.xbridge
->ofproto
)) {
7872 put_drop_action(xin
->odp_actions
, ctx
.error
);
7875 /* Since congestion drop and forwarding drop are not exactly
7876 * translation error, we are resetting the translation error.
7878 if (ctx
.error
== XLATE_CONGESTION_DROP
||
7879 ctx
.error
== XLATE_FORWARDING_DISABLED
) {
7880 ctx
.error
= XLATE_OK
;
7887 xlate_resume(struct ofproto_dpif
*ofproto
,
7888 const struct ofputil_packet_in_private
*pin
,
7889 struct ofpbuf
*odp_actions
,
7890 enum slow_path_reason
*slow
,
7892 struct xlate_cache
*xcache
)
7894 struct dp_packet packet
;
7895 dp_packet_use_const(&packet
, pin
->base
.packet
,
7896 pin
->base
.packet_len
);
7898 pkt_metadata_from_flow(&packet
.md
, &pin
->base
.flow_metadata
.flow
);
7899 flow_extract(&packet
, flow
);
7901 struct xlate_in xin
;
7902 xlate_in_init(&xin
, ofproto
, ofproto_dpif_get_tables_version(ofproto
),
7903 flow
, 0, NULL
, ntohs(flow
->tcp_flags
),
7904 &packet
, NULL
, odp_actions
);
7905 xin
.xcache
= xcache
;
7907 struct ofpact_note noop
;
7908 ofpact_init_NOTE(&noop
);
7911 bool any_actions
= pin
->actions_len
> 0;
7912 struct frozen_state state
= {
7913 .table_id
= 0, /* Not the table where NXAST_PAUSE was executed. */
7914 .ofproto_uuid
= pin
->bridge
,
7915 .stack
= pin
->stack
,
7916 .stack_size
= pin
->stack_size
,
7917 .mirrors
= pin
->mirrors
,
7918 .conntracked
= pin
->conntracked
,
7919 .xport_uuid
= UUID_ZERO
,
7921 /* When there are no actions, xlate_actions() will search the flow
7922 * table. We don't want it to do that (we want it to resume), so
7923 * supply a no-op action if there aren't any.
7925 * (We can't necessarily avoid translating actions entirely if there
7926 * aren't any actions, because there might be some finishing-up to do
7927 * at the end of the pipeline, and we don't check for those
7929 .ofpacts
= any_actions
? pin
->actions
: &noop
.ofpact
,
7930 .ofpacts_len
= any_actions
? pin
->actions_len
: sizeof noop
,
7932 .action_set
= pin
->action_set
,
7933 .action_set_len
= pin
->action_set_len
,
7935 frozen_metadata_from_flow(&state
.metadata
,
7936 &pin
->base
.flow_metadata
.flow
);
7937 xin
.frozen_state
= &state
;
7939 struct xlate_out xout
;
7940 enum xlate_error error
= xlate_actions(&xin
, &xout
);
7942 xlate_out_uninit(&xout
);
7944 /* xlate_actions() can generate a number of errors, but only
7945 * XLATE_BRIDGE_NOT_FOUND really stands out to me as one that we should be
7946 * sure to report over OpenFlow. The others could come up in packet-outs
7947 * or regular flow translation and I don't think that it's going to be too
7948 * useful to report them to the controller. */
7949 return error
== XLATE_BRIDGE_NOT_FOUND
? OFPERR_NXR_STALE
: 0;
7952 /* Sends 'packet' out 'ofport'. If 'port' is a tunnel and that tunnel type
7953 * supports a notion of an OAM flag, sets it if 'oam' is true.
7954 * May modify 'packet'.
7955 * Returns 0 if successful, otherwise a positive errno value. */
7957 xlate_send_packet(const struct ofport_dpif
*ofport
, bool oam
,
7958 struct dp_packet
*packet
)
7960 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7961 struct xport
*xport
;
7962 uint64_t ofpacts_stub
[1024 / 8];
7963 struct ofpbuf ofpacts
;
7966 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
7967 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
7968 flow_extract(packet
, &flow
);
7969 flow
.in_port
.ofp_port
= OFPP_NONE
;
7971 xport
= xport_lookup(xcfg
, ofport
);
7977 const ovs_be16 flag
= htons(NX_TUN_FLAG_OAM
);
7978 ofpact_put_set_field(&ofpacts
, mf_from_id(MFF_TUN_FLAGS
),
7982 ofpact_put_OUTPUT(&ofpacts
)->port
= xport
->ofp_port
;
7984 /* Actions here are not referring to anything versionable (flow tables or
7985 * groups) so we don't need to worry about the version here. */
7986 return ofproto_dpif_execute_actions(xport
->xbridge
->ofproto
,
7987 OVS_VERSION_MAX
, &flow
, NULL
,
7988 ofpacts
.data
, ofpacts
.size
, packet
);
7992 xlate_mac_learning_update(const struct ofproto_dpif
*ofproto
,
7993 ofp_port_t in_port
, struct eth_addr dl_src
,
7994 int vlan
, bool is_grat_arp
)
7996 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7997 struct xbridge
*xbridge
;
7998 struct xbundle
*xbundle
;
8000 xbridge
= xbridge_lookup(xcfg
, ofproto
);
8005 xbundle
= lookup_input_bundle__(xbridge
, in_port
, NULL
);
8010 update_learning_table__(xbridge
, xbundle
, dl_src
, vlan
, is_grat_arp
);
8014 xlate_set_support(const struct ofproto_dpif
*ofproto
,
8015 const struct dpif_backer_support
*support
)
8017 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
8018 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, ofproto
);
8021 xbridge
->support
= *support
;