1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2019 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 /* If recirculation was initiated due to bond (in_port = OFPP_NONE)
1524 * then frozen state is static and xport_uuid is not defined, so xport
1525 * cannot be restored from frozen state. */
1526 if (recirc_id_node
->state
.metadata
.in_port
!= OFPP_NONE
) {
1527 struct uuid xport_uuid
= recirc_id_node
->state
.xport_uuid
;
1528 xport
= xport_lookup_by_uuid(xcfg
, &xport_uuid
);
1529 if (xport
&& xport
->xbridge
&& xport
->xbridge
->ofproto
) {
1535 xport
= xport_lookup(xcfg
, tnl_port_should_receive(flow
)
1536 ? tnl_port_receive(flow
)
1537 : odp_port_to_ofport(backer
, flow
->in_port
.odp_port
));
1538 if (OVS_UNLIKELY(!xport
)) {
1540 *errorp
= (tnl_port_should_receive(flow
)
1541 ? xstrdup("no OpenFlow tunnel port for this packet")
1542 : xasprintf("no OpenFlow tunnel port for datapath "
1543 "port %"PRIu32
, flow
->in_port
.odp_port
));
1554 *ofp_in_port
= xport
->ofp_port
;
1556 return xport
->xbridge
->ofproto
;
1559 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1560 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1561 struct ofproto_dpif
*
1562 xlate_lookup_ofproto(const struct dpif_backer
*backer
, const struct flow
*flow
,
1563 ofp_port_t
*ofp_in_port
, char **errorp
)
1565 const struct xport
*xport
;
1567 return xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, errorp
);
1570 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1571 * optionally populates 'ofprotop' with the ofproto_dpif, 'ofp_in_port' with the
1572 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1573 * handles for those protocols if they're enabled. Caller may use the returned
1574 * pointers until quiescing, for longer term use additional references must
1577 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1580 xlate_lookup(const struct dpif_backer
*backer
, const struct flow
*flow
,
1581 struct ofproto_dpif
**ofprotop
, struct dpif_ipfix
**ipfix
,
1582 struct dpif_sflow
**sflow
, struct netflow
**netflow
,
1583 ofp_port_t
*ofp_in_port
)
1585 struct ofproto_dpif
*ofproto
;
1586 const struct xport
*xport
;
1588 ofproto
= xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, NULL
);
1595 *ofprotop
= ofproto
;
1599 *ipfix
= xport
? xport
->xbridge
->ipfix
: NULL
;
1603 *sflow
= xport
? xport
->xbridge
->sflow
: NULL
;
1607 *netflow
= xport
? xport
->xbridge
->netflow
: NULL
;
1613 static struct xbridge
*
1614 xbridge_lookup(struct xlate_cfg
*xcfg
, const struct ofproto_dpif
*ofproto
)
1616 struct hmap
*xbridges
;
1617 struct xbridge
*xbridge
;
1619 if (!ofproto
|| !xcfg
) {
1623 xbridges
= &xcfg
->xbridges
;
1625 HMAP_FOR_EACH_IN_BUCKET (xbridge
, hmap_node
, hash_pointer(ofproto
, 0),
1627 if (xbridge
->ofproto
== ofproto
) {
1634 static struct xbridge
*
1635 xbridge_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1637 struct xbridge
*xbridge
;
1639 HMAP_FOR_EACH (xbridge
, hmap_node
, &xcfg
->xbridges
) {
1640 if (uuid_equals(&xbridge
->ofproto
->uuid
, uuid
)) {
1647 static struct xbundle
*
1648 xbundle_lookup(struct xlate_cfg
*xcfg
, const struct ofbundle
*ofbundle
)
1650 struct hmap
*xbundles
;
1651 struct xbundle
*xbundle
;
1653 if (!ofbundle
|| !xcfg
) {
1657 xbundles
= &xcfg
->xbundles
;
1659 HMAP_FOR_EACH_IN_BUCKET (xbundle
, hmap_node
, hash_pointer(ofbundle
, 0),
1661 if (xbundle
->ofbundle
== ofbundle
) {
1668 static struct xport
*
1669 xport_lookup(struct xlate_cfg
*xcfg
, const struct ofport_dpif
*ofport
)
1671 struct hmap
*xports
;
1672 struct xport
*xport
;
1674 if (!ofport
|| !xcfg
) {
1678 xports
= &xcfg
->xports
;
1680 HMAP_FOR_EACH_IN_BUCKET (xport
, hmap_node
, hash_pointer(ofport
, 0),
1682 if (xport
->ofport
== ofport
) {
1689 static struct xport
*
1690 xport_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1692 struct hmap
*xports
;
1693 struct xport
*xport
;
1695 if (uuid_is_zero(uuid
) || !xcfg
) {
1699 xports
= &xcfg
->xports_uuid
;
1701 HMAP_FOR_EACH_IN_BUCKET (xport
, uuid_node
, uuid_hash(uuid
), xports
) {
1702 if (uuid_equals(&xport
->uuid
, uuid
)) {
1709 static struct stp_port
*
1710 xport_get_stp_port(const struct xport
*xport
)
1712 return xport
->xbridge
->stp
&& xport
->stp_port_no
!= -1
1713 ? stp_get_port(xport
->xbridge
->stp
, xport
->stp_port_no
)
1718 xport_stp_learn_state(const struct xport
*xport
)
1720 struct stp_port
*sp
= xport_get_stp_port(xport
);
1722 ? stp_learn_in_state(stp_port_get_state(sp
))
1727 xport_stp_forward_state(const struct xport
*xport
)
1729 struct stp_port
*sp
= xport_get_stp_port(xport
);
1731 ? stp_forward_in_state(stp_port_get_state(sp
))
1736 xport_stp_should_forward_bpdu(const struct xport
*xport
)
1738 struct stp_port
*sp
= xport_get_stp_port(xport
);
1739 return stp_should_forward_bpdu(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
1742 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1743 * were used to make the determination.*/
1745 stp_should_process_flow(const struct flow
*flow
, struct flow_wildcards
*wc
)
1747 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1748 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1749 return is_stp(flow
);
1753 stp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1755 struct stp_port
*sp
= xport_get_stp_port(xport
);
1756 struct dp_packet payload
= *packet
;
1757 struct eth_header
*eth
= dp_packet_data(&payload
);
1759 /* Sink packets on ports that have STP disabled when the bridge has
1761 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
1765 /* Trim off padding on payload. */
1766 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1767 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1770 if (dp_packet_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
1771 stp_received_bpdu(sp
, dp_packet_data(&payload
), dp_packet_size(&payload
));
1775 static enum rstp_state
1776 xport_get_rstp_port_state(const struct xport
*xport
)
1778 return xport
->rstp_port
1779 ? rstp_port_get_state(xport
->rstp_port
)
1784 xport_rstp_learn_state(const struct xport
*xport
)
1786 return xport
->xbridge
->rstp
&& xport
->rstp_port
1787 ? rstp_learn_in_state(xport_get_rstp_port_state(xport
))
1792 xport_rstp_forward_state(const struct xport
*xport
)
1794 return xport
->xbridge
->rstp
&& xport
->rstp_port
1795 ? rstp_forward_in_state(xport_get_rstp_port_state(xport
))
1800 xport_rstp_should_manage_bpdu(const struct xport
*xport
)
1802 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport
));
1806 rstp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1808 struct dp_packet payload
= *packet
;
1809 struct eth_header
*eth
= dp_packet_data(&payload
);
1811 /* Sink packets on ports that have no RSTP. */
1812 if (!xport
->rstp_port
) {
1816 /* Trim off padding on payload. */
1817 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1818 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1821 int len
= ETH_HEADER_LEN
+ LLC_HEADER_LEN
;
1822 if (eth
->eth_type
== htons(ETH_TYPE_VLAN
)) {
1823 len
+= VLAN_HEADER_LEN
;
1825 if (dp_packet_try_pull(&payload
, len
)) {
1826 rstp_port_received_bpdu(xport
->rstp_port
, dp_packet_data(&payload
),
1827 dp_packet_size(&payload
));
1831 static struct xport
*
1832 get_ofp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1834 struct xport
*xport
;
1836 HMAP_FOR_EACH_IN_BUCKET (xport
, ofp_node
, hash_ofp_port(ofp_port
),
1838 if (xport
->ofp_port
== ofp_port
) {
1846 ofp_port_to_odp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1848 const struct xport
*xport
= get_ofp_port(xbridge
, ofp_port
);
1849 return xport
? xport
->odp_port
: ODPP_NONE
;
1853 odp_port_is_alive(const struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
1855 struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
1856 return xport
&& xport
->may_enable
;
1859 static struct ofputil_bucket
*
1860 group_first_live_bucket(const struct xlate_ctx
*, const struct group_dpif
*,
1864 group_is_alive(const struct xlate_ctx
*ctx
, uint32_t group_id
, int depth
)
1866 struct group_dpif
*group
;
1868 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
1869 ctx
->xin
->tables_version
, false);
1871 return group_first_live_bucket(ctx
, group
, depth
) != NULL
;
1877 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1880 bucket_is_alive(const struct xlate_ctx
*ctx
,
1881 struct ofputil_bucket
*bucket
, int depth
)
1883 if (depth
>= MAX_LIVENESS_RECURSION
) {
1884 xlate_report_error(ctx
, "bucket chaining exceeded %d links",
1885 MAX_LIVENESS_RECURSION
);
1889 return (!ofputil_bucket_has_liveness(bucket
)
1890 || (bucket
->watch_port
!= OFPP_ANY
1891 && odp_port_is_alive(ctx
, bucket
->watch_port
))
1892 || (bucket
->watch_group
!= OFPG_ANY
1893 && group_is_alive(ctx
, bucket
->watch_group
, depth
+ 1)));
1897 xlate_report_bucket_not_live(const struct xlate_ctx
*ctx
,
1898 const struct ofputil_bucket
*bucket
)
1900 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
1901 struct ds s
= DS_EMPTY_INITIALIZER
;
1902 if (bucket
->watch_port
!= OFPP_ANY
) {
1903 ds_put_cstr(&s
, "port ");
1904 ofputil_format_port(bucket
->watch_port
, NULL
, &s
);
1906 if (bucket
->watch_group
!= OFPG_ANY
) {
1908 ds_put_cstr(&s
, " and ");
1910 ds_put_format(&s
, "port %"PRIu32
, bucket
->watch_group
);
1913 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": not live due to %s",
1914 bucket
->bucket_id
, ds_cstr(&s
));
1920 static struct ofputil_bucket
*
1921 group_first_live_bucket(const struct xlate_ctx
*ctx
,
1922 const struct group_dpif
*group
, int depth
)
1924 struct ofputil_bucket
*bucket
;
1925 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1926 if (bucket_is_alive(ctx
, bucket
, depth
)) {
1929 xlate_report_bucket_not_live(ctx
, bucket
);
1935 static struct ofputil_bucket
*
1936 group_best_live_bucket(const struct xlate_ctx
*ctx
,
1937 const struct group_dpif
*group
,
1940 struct ofputil_bucket
*best_bucket
= NULL
;
1941 uint32_t best_score
= 0;
1943 struct ofputil_bucket
*bucket
;
1944 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1945 if (bucket_is_alive(ctx
, bucket
, 0)) {
1947 (hash_int(bucket
->bucket_id
, basis
) & 0xffff) * bucket
->weight
;
1948 if (score
>= best_score
) {
1949 best_bucket
= bucket
;
1952 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": score %"PRIu32
,
1953 bucket
->bucket_id
, score
);
1955 xlate_report_bucket_not_live(ctx
, bucket
);
1963 xbundle_trunks_vlan(const struct xbundle
*bundle
, uint16_t vlan
)
1965 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
1966 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
1970 xbundle_allows_cvlan(const struct xbundle
*bundle
, uint16_t vlan
)
1972 return (!bundle
->cvlans
|| bitmap_is_set(bundle
->cvlans
, vlan
));
1976 xbundle_includes_vlan(const struct xbundle
*xbundle
, const struct xvlan
*xvlan
)
1978 switch (xbundle
->vlan_mode
) {
1979 case PORT_VLAN_ACCESS
:
1980 return xvlan
->v
[0].vid
== xbundle
->vlan
&& xvlan
->v
[1].vid
== 0;
1982 case PORT_VLAN_TRUNK
:
1983 case PORT_VLAN_NATIVE_UNTAGGED
:
1984 case PORT_VLAN_NATIVE_TAGGED
:
1985 return xbundle_trunks_vlan(xbundle
, xvlan
->v
[0].vid
);
1987 case PORT_VLAN_DOT1Q_TUNNEL
:
1988 return xvlan
->v
[0].vid
== xbundle
->vlan
&&
1989 xbundle_allows_cvlan(xbundle
, xvlan
->v
[1].vid
);
1996 static mirror_mask_t
1997 xbundle_mirror_out(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
1999 return xbundle
!= &ofpp_none_bundle
2000 ? mirror_bundle_out(xbridge
->mbridge
, xbundle
->ofbundle
)
2004 static mirror_mask_t
2005 xbundle_mirror_src(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2007 return xbundle
!= &ofpp_none_bundle
2008 ? mirror_bundle_src(xbridge
->mbridge
, xbundle
->ofbundle
)
2012 static mirror_mask_t
2013 xbundle_mirror_dst(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2015 return xbundle
!= &ofpp_none_bundle
2016 ? mirror_bundle_dst(xbridge
->mbridge
, xbundle
->ofbundle
)
2020 static struct xbundle
*
2021 lookup_input_bundle__(const struct xbridge
*xbridge
,
2022 ofp_port_t in_port
, struct xport
**in_xportp
)
2024 struct xport
*xport
;
2026 /* Find the port and bundle for the received packet. */
2027 xport
= get_ofp_port(xbridge
, in_port
);
2031 if (xport
&& xport
->xbundle
) {
2032 return xport
->xbundle
;
2035 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
2036 * which a controller may use as the ingress port for traffic that
2037 * it is sourcing. */
2038 if (in_port
== OFPP_CONTROLLER
|| in_port
== OFPP_NONE
) {
2039 return &ofpp_none_bundle
;
2044 static struct xbundle
*
2045 lookup_input_bundle(const struct xlate_ctx
*ctx
,
2046 ofp_port_t in_port
, struct xport
**in_xportp
)
2048 struct xbundle
*xbundle
= lookup_input_bundle__(ctx
->xbridge
,
2049 in_port
, in_xportp
);
2051 /* Odd. A few possible reasons here:
2053 * - We deleted a port but there are still a few packets queued up
2056 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
2057 * we don't know about.
2059 * - The ofproto client didn't configure the port as part of a bundle.
2060 * This is particularly likely to happen if a packet was received on
2061 * the port after it was created, but before the client had a chance
2062 * to configure its bundle.
2064 xlate_report_error(ctx
, "received packet on unknown port %"PRIu32
,
2070 /* Mirrors the packet represented by 'ctx' to appropriate mirror destinations,
2071 * given the packet is ingressing or egressing on 'xbundle', which has ingress
2072 * or egress (as appropriate) mirrors 'mirrors'. */
2074 mirror_packet(struct xlate_ctx
*ctx
, struct xbundle
*xbundle
,
2075 mirror_mask_t mirrors
)
2077 struct xvlan in_xvlan
;
2080 /* Figure out what VLAN the packet is in (because mirrors can select
2081 * packets on basis of VLAN). */
2082 xvlan_extract(&ctx
->xin
->flow
, &in_xvlan
);
2083 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, xbundle
)) {
2086 xvlan_input_translate(xbundle
, &in_xvlan
, &xvlan
);
2088 const struct xbridge
*xbridge
= ctx
->xbridge
;
2090 /* Don't mirror to destinations that we've already mirrored to. */
2091 mirrors
&= ~ctx
->mirrors
;
2096 /* 'mirrors' is a bit-mask of candidates for mirroring. Iterate through
2097 * the candidates, adding the ones that really should be mirrored to
2098 * 'used_mirrors', as long as some candidates remain. */
2099 mirror_mask_t used_mirrors
= 0;
2101 const unsigned long *vlans
;
2102 mirror_mask_t dup_mirrors
;
2103 struct ofbundle
*out
;
2107 /* Get the details of the mirror represented by the rightmost 1-bit. */
2108 ovs_assert(mirror_get(xbridge
->mbridge
, raw_ctz(mirrors
),
2109 &vlans
, &dup_mirrors
,
2110 &out
, &snaplen
, &out_vlan
));
2113 /* If this mirror selects on the basis of VLAN, and it does not select
2114 * 'vlan', then discard this mirror and go on to the next one. */
2116 ctx
->wc
->masks
.vlans
[0].tci
|= htons(VLAN_CFI
| VLAN_VID_MASK
);
2118 if (vlans
&& !bitmap_is_set(vlans
, xvlan
.v
[0].vid
)) {
2119 mirrors
= zero_rightmost_1bit(mirrors
);
2123 /* We sent a packet to this mirror. */
2124 used_mirrors
|= rightmost_1bit(mirrors
);
2126 /* Record the mirror, and the mirrors that output to the same
2127 * destination, so that we don't mirror to them again. This must be
2128 * done now to ensure that output_normal(), below, doesn't recursively
2129 * output to the same mirrors. */
2130 ctx
->mirrors
|= dup_mirrors
;
2131 ctx
->mirror_snaplen
= snaplen
;
2133 /* Send the packet to the mirror. */
2135 struct xbundle
*out_xbundle
= xbundle_lookup(ctx
->xcfg
, out
);
2137 output_normal(ctx
, out_xbundle
, &xvlan
);
2139 } else if (xvlan
.v
[0].vid
!= out_vlan
2140 && !eth_addr_is_reserved(ctx
->xin
->flow
.dl_dst
)) {
2142 uint16_t old_vid
= xvlan
.v
[0].vid
;
2144 xvlan
.v
[0].vid
= out_vlan
;
2145 LIST_FOR_EACH (xb
, list_node
, &xbridge
->xbundles
) {
2146 if (xbundle_includes_vlan(xb
, &xvlan
)
2147 && !xbundle_mirror_out(xbridge
, xb
)) {
2148 output_normal(ctx
, xb
, &xvlan
);
2151 xvlan
.v
[0].vid
= old_vid
;
2154 /* output_normal() could have recursively output (to different
2155 * mirrors), so make sure that we don't send duplicates. */
2156 mirrors
&= ~ctx
->mirrors
;
2157 ctx
->mirror_snaplen
= 0;
2161 if (ctx
->xin
->resubmit_stats
) {
2162 mirror_update_stats(xbridge
->mbridge
, used_mirrors
,
2163 ctx
->xin
->resubmit_stats
->n_packets
,
2164 ctx
->xin
->resubmit_stats
->n_bytes
);
2166 if (ctx
->xin
->xcache
) {
2167 struct xc_entry
*entry
;
2169 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_MIRROR
);
2170 entry
->mirror
.mbridge
= mbridge_ref(xbridge
->mbridge
);
2171 entry
->mirror
.mirrors
= used_mirrors
;
2177 mirror_ingress_packet(struct xlate_ctx
*ctx
)
2179 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
)) {
2180 struct xbundle
*xbundle
= lookup_input_bundle(
2181 ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
);
2183 mirror_packet(ctx
, xbundle
,
2184 xbundle_mirror_src(ctx
->xbridge
, xbundle
));
2189 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
2190 * If so, returns true. Otherwise, returns false.
2192 * 'vid' should be the VID obtained from the 802.1Q header that was received as
2193 * part of a packet (specify 0 if there was no 802.1Q header), in the range
2196 input_vid_is_valid(const struct xlate_ctx
*ctx
,
2197 uint16_t vid
, struct xbundle
*in_xbundle
)
2199 /* Allow any VID on the OFPP_NONE port. */
2200 if (in_xbundle
== &ofpp_none_bundle
) {
2204 switch (in_xbundle
->vlan_mode
) {
2205 case PORT_VLAN_ACCESS
:
2207 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" tagged "
2208 "packet received on port %s configured as VLAN "
2209 "%d access port", vid
, in_xbundle
->name
,
2215 case PORT_VLAN_NATIVE_UNTAGGED
:
2216 case PORT_VLAN_NATIVE_TAGGED
:
2218 /* Port must always carry its native VLAN. */
2222 case PORT_VLAN_TRUNK
:
2223 if (!xbundle_trunks_vlan(in_xbundle
, vid
)) {
2224 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet "
2225 "received on port %s not configured for "
2226 "trunking VLAN %"PRIu16
,
2227 vid
, in_xbundle
->name
, vid
);
2232 case PORT_VLAN_DOT1Q_TUNNEL
:
2233 if (!xbundle_allows_cvlan(in_xbundle
, vid
)) {
2234 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet received "
2235 "on dot1q-tunnel port %s that excludes this "
2236 "VLAN", vid
, in_xbundle
->name
);
2248 xvlan_copy(struct xvlan
*dst
, const struct xvlan
*src
)
2254 xvlan_pop(struct xvlan
*src
)
2256 memmove(&src
->v
[0], &src
->v
[1], sizeof(src
->v
) - sizeof(src
->v
[0]));
2257 memset(&src
->v
[FLOW_MAX_VLAN_HEADERS
- 1], 0,
2258 sizeof(src
->v
[FLOW_MAX_VLAN_HEADERS
- 1]));
2262 xvlan_push_uninit(struct xvlan
*src
)
2264 memmove(&src
->v
[1], &src
->v
[0], sizeof(src
->v
) - sizeof(src
->v
[0]));
2265 memset(&src
->v
[0], 0, sizeof(src
->v
[0]));
2268 /* Extract VLAN information (headers) from flow */
2270 xvlan_extract(const struct flow
*flow
, struct xvlan
*xvlan
)
2273 memset(xvlan
, 0, sizeof(*xvlan
));
2274 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2275 if (!eth_type_vlan(flow
->vlans
[i
].tpid
) ||
2276 !(flow
->vlans
[i
].tci
& htons(VLAN_CFI
))) {
2279 xvlan
->v
[i
].tpid
= ntohs(flow
->vlans
[i
].tpid
);
2280 xvlan
->v
[i
].vid
= vlan_tci_to_vid(flow
->vlans
[i
].tci
);
2281 xvlan
->v
[i
].pcp
= ntohs(flow
->vlans
[i
].tci
) & VLAN_PCP_MASK
;
2285 /* Put VLAN information (headers) to flow */
2287 xvlan_put(struct flow
*flow
, const struct xvlan
*xvlan
,
2288 enum port_priority_tags_mode use_priority_tags
)
2292 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2293 tci
= htons(xvlan
->v
[i
].vid
| (xvlan
->v
[i
].pcp
& VLAN_PCP_MASK
));
2294 if (tci
|| ((use_priority_tags
== PORT_PRIORITY_TAGS_ALWAYS
) &&
2295 xvlan
->v
[i
].tpid
)) {
2296 tci
|= htons(VLAN_CFI
);
2297 flow
->vlans
[i
].tpid
= xvlan
->v
[i
].tpid
?
2298 htons(xvlan
->v
[i
].tpid
) :
2299 htons(ETH_TYPE_VLAN_8021Q
);
2301 flow
->vlans
[i
].tci
= tci
;
2305 /* Given 'in_xvlan', extracted from the input 802.1Q headers received as part
2306 * of a packet, and 'in_xbundle', the bundle on which the packet was received,
2307 * returns the VLANs of the packet during bridge internal processing. */
2309 xvlan_input_translate(const struct xbundle
*in_xbundle
,
2310 const struct xvlan
*in_xvlan
, struct xvlan
*xvlan
)
2313 switch (in_xbundle
->vlan_mode
) {
2314 case PORT_VLAN_ACCESS
:
2315 memset(xvlan
, 0, sizeof(*xvlan
));
2316 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2317 ETH_TYPE_VLAN_8021Q
;
2318 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2319 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2322 case PORT_VLAN_TRUNK
:
2323 xvlan_copy(xvlan
, in_xvlan
);
2326 case PORT_VLAN_NATIVE_UNTAGGED
:
2327 case PORT_VLAN_NATIVE_TAGGED
:
2328 xvlan_copy(xvlan
, in_xvlan
);
2329 if (!in_xvlan
->v
[0].vid
) {
2330 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2331 ETH_TYPE_VLAN_8021Q
;
2332 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2333 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2337 case PORT_VLAN_DOT1Q_TUNNEL
:
2338 xvlan_copy(xvlan
, in_xvlan
);
2339 xvlan_push_uninit(xvlan
);
2340 xvlan
->v
[0].tpid
= in_xbundle
->qinq_ethtype
;
2341 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2342 xvlan
->v
[0].pcp
= 0;
2350 /* Given 'xvlan', the VLANs of a packet during internal processing, and
2351 * 'out_xbundle', a bundle on which the packet is to be output, returns the
2352 * VLANs that should be included in output packet. */
2354 xvlan_output_translate(const struct xbundle
*out_xbundle
,
2355 const struct xvlan
*xvlan
, struct xvlan
*out_xvlan
)
2357 switch (out_xbundle
->vlan_mode
) {
2358 case PORT_VLAN_ACCESS
:
2359 memset(out_xvlan
, 0, sizeof(*out_xvlan
));
2362 case PORT_VLAN_TRUNK
:
2363 case PORT_VLAN_NATIVE_TAGGED
:
2364 xvlan_copy(out_xvlan
, xvlan
);
2367 case PORT_VLAN_NATIVE_UNTAGGED
:
2368 xvlan_copy(out_xvlan
, xvlan
);
2369 if (xvlan
->v
[0].vid
== out_xbundle
->vlan
) {
2370 xvlan_pop(out_xvlan
);
2374 case PORT_VLAN_DOT1Q_TUNNEL
:
2375 xvlan_copy(out_xvlan
, xvlan
);
2376 xvlan_pop(out_xvlan
);
2384 /* If output xbundle is dot1q-tunnel, set mask bits of cvlan */
2386 check_and_set_cvlan_mask(struct flow_wildcards
*wc
,
2387 const struct xbundle
*xbundle
)
2389 if (xbundle
->vlan_mode
== PORT_VLAN_DOT1Q_TUNNEL
&& xbundle
->cvlans
) {
2390 wc
->masks
.vlans
[1].tci
= htons(0xffff);
2395 output_normal(struct xlate_ctx
*ctx
, const struct xbundle
*out_xbundle
,
2396 const struct xvlan
*xvlan
)
2399 union flow_vlan_hdr old_vlans
[FLOW_MAX_VLAN_HEADERS
];
2400 struct xport
*xport
;
2401 struct xlate_bond_recirc xr
;
2402 bool use_recirc
= false;
2403 struct xvlan out_xvlan
;
2405 check_and_set_cvlan_mask(ctx
->wc
, out_xbundle
);
2407 xvlan_output_translate(out_xbundle
, xvlan
, &out_xvlan
);
2408 if (out_xbundle
->use_priority_tags
) {
2409 out_xvlan
.v
[0].pcp
= ntohs(ctx
->xin
->flow
.vlans
[0].tci
) &
2412 vid
= out_xvlan
.v
[0].vid
;
2413 if (ovs_list_is_empty(&out_xbundle
->xports
)) {
2414 /* Partially configured bundle with no slaves. Drop the packet. */
2416 } else if (!out_xbundle
->bond
) {
2417 xport
= CONTAINER_OF(ovs_list_front(&out_xbundle
->xports
), struct xport
,
2420 struct flow_wildcards
*wc
= ctx
->wc
;
2421 struct ofport_dpif
*ofport
;
2423 if (ctx
->xbridge
->support
.odp
.recirc
) {
2424 /* In case recirculation is not actually in use, 'xr.recirc_id'
2425 * will be set to '0', since a valid 'recirc_id' can
2427 bond_update_post_recirc_rules(out_xbundle
->bond
,
2431 /* Use recirculation instead of output. */
2433 xr
.hash_alg
= OVS_HASH_ALG_L4
;
2434 /* Recirculation does not require unmasking hash fields. */
2439 ofport
= bond_choose_output_slave(out_xbundle
->bond
,
2440 &ctx
->xin
->flow
, wc
, vid
);
2441 xport
= xport_lookup(ctx
->xcfg
, ofport
);
2444 /* No slaves enabled, so drop packet. */
2448 /* If use_recirc is set, the main thread will handle stats
2449 * accounting for this bond. */
2451 if (ctx
->xin
->resubmit_stats
) {
2452 bond_account(out_xbundle
->bond
, &ctx
->xin
->flow
, vid
,
2453 ctx
->xin
->resubmit_stats
->n_bytes
);
2455 if (ctx
->xin
->xcache
) {
2456 struct xc_entry
*entry
;
2459 flow
= &ctx
->xin
->flow
;
2460 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_BOND
);
2461 entry
->bond
.bond
= bond_ref(out_xbundle
->bond
);
2462 entry
->bond
.flow
= xmemdup(flow
, sizeof *flow
);
2463 entry
->bond
.vid
= vid
;
2468 memcpy(&old_vlans
, &ctx
->xin
->flow
.vlans
, sizeof(old_vlans
));
2469 xvlan_put(&ctx
->xin
->flow
, &out_xvlan
, out_xbundle
->use_priority_tags
);
2471 compose_output_action(ctx
, xport
->ofp_port
, use_recirc
? &xr
: NULL
,
2473 memcpy(&ctx
->xin
->flow
.vlans
, &old_vlans
, sizeof(old_vlans
));
2476 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2477 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2478 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2480 is_gratuitous_arp(const struct flow
*flow
, struct flow_wildcards
*wc
)
2482 if (flow
->dl_type
!= htons(ETH_TYPE_ARP
)) {
2486 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2487 if (!eth_addr_is_broadcast(flow
->dl_dst
)) {
2491 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2492 if (flow
->nw_proto
== ARP_OP_REPLY
) {
2494 } else if (flow
->nw_proto
== ARP_OP_REQUEST
) {
2495 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
2496 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2498 return flow
->nw_src
== flow
->nw_dst
;
2504 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
2505 * dropped. Returns true if they may be forwarded, false if they should be
2508 * 'in_port' must be the xport that corresponds to flow->in_port.
2509 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
2511 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
2512 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
2513 * checked by input_vid_is_valid().
2515 * May also add tags to '*tags', although the current implementation only does
2516 * so in one special case.
2519 is_admissible(struct xlate_ctx
*ctx
, struct xport
*in_port
,
2522 struct xbundle
*in_xbundle
= in_port
->xbundle
;
2523 const struct xbridge
*xbridge
= ctx
->xbridge
;
2524 struct flow
*flow
= &ctx
->xin
->flow
;
2526 /* Drop frames for reserved multicast addresses
2527 * only if forward_bpdu option is absent. */
2528 if (!xbridge
->forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
2529 xlate_report(ctx
, OFT_DETAIL
,
2530 "packet has reserved destination MAC, dropping");
2534 if (in_xbundle
->bond
) {
2535 struct mac_entry
*mac
;
2537 switch (bond_check_admissibility(in_xbundle
->bond
, in_port
->ofport
,
2543 xlate_report(ctx
, OFT_DETAIL
,
2544 "bonding refused admissibility, dropping");
2547 case BV_DROP_IF_MOVED
:
2548 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
2549 mac
= mac_learning_lookup(xbridge
->ml
, flow
->dl_src
, vlan
);
2551 && mac_entry_get_port(xbridge
->ml
, mac
) != in_xbundle
->ofbundle
2552 && (!is_gratuitous_arp(flow
, ctx
->wc
)
2553 || mac_entry_is_grat_arp_locked(mac
))) {
2554 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2555 xlate_report(ctx
, OFT_DETAIL
,
2556 "SLB bond thinks this packet looped back, "
2560 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2569 update_learning_table__(const struct xbridge
*xbridge
,
2570 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2571 int vlan
, bool is_grat_arp
)
2573 return (in_xbundle
== &ofpp_none_bundle
2574 || !mac_learning_update(xbridge
->ml
, dl_src
, vlan
,
2576 in_xbundle
->bond
!= NULL
,
2577 in_xbundle
->ofbundle
));
2581 update_learning_table(const struct xlate_ctx
*ctx
,
2582 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2583 int vlan
, bool is_grat_arp
)
2585 if (!update_learning_table__(ctx
->xbridge
, in_xbundle
, dl_src
, vlan
,
2587 xlate_report_debug(ctx
, OFT_DETAIL
, "learned that "ETH_ADDR_FMT
" is "
2588 "on port %s in VLAN %d",
2589 ETH_ADDR_ARGS(dl_src
), in_xbundle
->name
, vlan
);
2593 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2594 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2596 update_mcast_snooping_table4__(const struct xlate_ctx
*ctx
,
2597 const struct flow
*flow
,
2598 struct mcast_snooping
*ms
, int vlan
,
2599 struct xbundle
*in_xbundle
,
2600 const struct dp_packet
*packet
)
2601 OVS_REQ_WRLOCK(ms
->rwlock
)
2603 const struct igmp_header
*igmp
;
2606 ovs_be32 ip4
= flow
->igmp_group_ip4
;
2608 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2609 igmp
= dp_packet_at(packet
, offset
, IGMP_HEADER_LEN
);
2610 if (!igmp
|| csum(igmp
, dp_packet_l4_size(packet
)) != 0) {
2611 xlate_report_debug(ctx
, OFT_DETAIL
,
2612 "multicast snooping received bad IGMP "
2613 "checksum on port %s in VLAN %d",
2614 in_xbundle
->name
, vlan
);
2618 switch (ntohs(flow
->tp_src
)) {
2619 case IGMP_HOST_MEMBERSHIP_REPORT
:
2620 case IGMPV2_HOST_MEMBERSHIP_REPORT
:
2621 if (mcast_snooping_add_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2622 xlate_report_debug(ctx
, OFT_DETAIL
,
2623 "multicast snooping learned that "
2624 IP_FMT
" is on port %s in VLAN %d",
2625 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2628 case IGMP_HOST_LEAVE_MESSAGE
:
2629 if (mcast_snooping_leave_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2630 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping leaving "
2631 IP_FMT
" is on port %s in VLAN %d",
2632 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2635 case IGMP_HOST_MEMBERSHIP_QUERY
:
2636 if (flow
->nw_src
&& mcast_snooping_add_mrouter(ms
, vlan
,
2637 in_xbundle
->ofbundle
)) {
2638 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query "
2639 "from "IP_FMT
" is on port %s in VLAN %d",
2640 IP_ARGS(flow
->nw_src
), in_xbundle
->name
, vlan
);
2643 case IGMPV3_HOST_MEMBERSHIP_REPORT
:
2644 count
= mcast_snooping_add_report(ms
, packet
, vlan
,
2645 in_xbundle
->ofbundle
);
2647 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2648 "%d addresses on port %s in VLAN %d",
2649 count
, in_xbundle
->name
, vlan
);
2656 update_mcast_snooping_table6__(const struct xlate_ctx
*ctx
,
2657 const struct flow
*flow
,
2658 struct mcast_snooping
*ms
, int vlan
,
2659 struct xbundle
*in_xbundle
,
2660 const struct dp_packet
*packet
)
2661 OVS_REQ_WRLOCK(ms
->rwlock
)
2663 const struct mld_header
*mld
;
2667 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2668 mld
= dp_packet_at(packet
, offset
, MLD_HEADER_LEN
);
2671 packet_csum_upperlayer6(dp_packet_l3(packet
),
2672 mld
, IPPROTO_ICMPV6
,
2673 dp_packet_l4_size(packet
)) != 0) {
2674 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping received "
2675 "bad MLD checksum on port %s in VLAN %d",
2676 in_xbundle
->name
, vlan
);
2680 switch (ntohs(flow
->tp_src
)) {
2682 if (!ipv6_addr_equals(&flow
->ipv6_src
, &in6addr_any
)
2683 && mcast_snooping_add_mrouter(ms
, vlan
, in_xbundle
->ofbundle
)) {
2684 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query on "
2685 "port %s in VLAN %d", in_xbundle
->name
, vlan
);
2691 count
= mcast_snooping_add_mld(ms
, packet
, vlan
, in_xbundle
->ofbundle
);
2693 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2694 "%d addresses on port %s in VLAN %d",
2695 count
, in_xbundle
->name
, vlan
);
2701 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2702 * was received on 'in_xbundle' in 'vlan'. */
2704 update_mcast_snooping_table(const struct xlate_ctx
*ctx
,
2705 const struct flow
*flow
, int vlan
,
2706 struct xbundle
*in_xbundle
,
2707 const struct dp_packet
*packet
)
2709 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
2710 struct xbundle
*mcast_xbundle
;
2711 struct mcast_port_bundle
*fport
;
2713 /* Don't learn the OFPP_NONE port. */
2714 if (in_xbundle
== &ofpp_none_bundle
) {
2718 /* Don't learn from flood ports */
2719 mcast_xbundle
= NULL
;
2720 ovs_rwlock_wrlock(&ms
->rwlock
);
2721 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2722 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2723 if (mcast_xbundle
== in_xbundle
) {
2728 if (!mcast_xbundle
|| mcast_xbundle
!= in_xbundle
) {
2729 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2730 update_mcast_snooping_table4__(ctx
, flow
, ms
, vlan
,
2731 in_xbundle
, packet
);
2733 update_mcast_snooping_table6__(ctx
, flow
, ms
, vlan
,
2734 in_xbundle
, packet
);
2737 ovs_rwlock_unlock(&ms
->rwlock
);
2740 /* A list of multicast output ports.
2742 * We accumulate output ports and then do all the outputs afterward. It would
2743 * be more natural to do the outputs one at a time as we discover the need for
2744 * each one, but this can cause a deadlock because we need to take the
2745 * mcast_snooping's rwlock for reading to iterate through the port lists and
2746 * doing an output, if it goes to a patch port, can eventually come back to the
2747 * same mcast_snooping and attempt to take the write lock (see
2748 * https://github.com/openvswitch/ovs-issues/issues/153). */
2749 struct mcast_output
{
2750 /* Discrete ports. */
2751 struct xbundle
**xbundles
;
2752 size_t n
, allocated
;
2754 /* If set, flood to all ports. */
2757 #define MCAST_OUTPUT_INIT { NULL, 0, 0, false }
2759 /* Add 'mcast_bundle' to 'out'. */
2761 mcast_output_add(struct mcast_output
*out
, struct xbundle
*mcast_xbundle
)
2763 if (out
->n
>= out
->allocated
) {
2764 out
->xbundles
= x2nrealloc(out
->xbundles
, &out
->allocated
,
2765 sizeof *out
->xbundles
);
2767 out
->xbundles
[out
->n
++] = mcast_xbundle
;
2770 /* Outputs the packet in 'ctx' to all of the output ports in 'out', given input
2771 * bundle 'in_xbundle' and the current 'xvlan'. */
2773 mcast_output_finish(struct xlate_ctx
*ctx
, struct mcast_output
*out
,
2774 struct xbundle
*in_xbundle
, struct xvlan
*xvlan
)
2777 xlate_normal_flood(ctx
, in_xbundle
, xvlan
);
2779 for (size_t i
= 0; i
< out
->n
; i
++) {
2780 output_normal(ctx
, out
->xbundles
[i
], xvlan
);
2784 free(out
->xbundles
);
2787 /* send the packet to ports having the multicast group learned */
2789 xlate_normal_mcast_send_group(struct xlate_ctx
*ctx
,
2790 struct mcast_snooping
*ms OVS_UNUSED
,
2791 struct mcast_group
*grp
,
2792 struct xbundle
*in_xbundle
,
2793 struct mcast_output
*out
)
2794 OVS_REQ_RDLOCK(ms
->rwlock
)
2796 struct mcast_group_bundle
*b
;
2797 struct xbundle
*mcast_xbundle
;
2799 LIST_FOR_EACH(b
, bundle_node
, &grp
->bundle_lru
) {
2800 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, b
->port
);
2801 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2802 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast group port");
2803 mcast_output_add(out
, mcast_xbundle
);
2804 } else if (!mcast_xbundle
) {
2805 xlate_report(ctx
, OFT_WARN
,
2806 "mcast group port is unknown, dropping");
2808 xlate_report(ctx
, OFT_DETAIL
,
2809 "mcast group port is input port, dropping");
2814 /* send the packet to ports connected to multicast routers */
2816 xlate_normal_mcast_send_mrouters(struct xlate_ctx
*ctx
,
2817 struct mcast_snooping
*ms
,
2818 struct xbundle
*in_xbundle
,
2819 const struct xvlan
*xvlan
,
2820 struct mcast_output
*out
)
2821 OVS_REQ_RDLOCK(ms
->rwlock
)
2823 struct mcast_mrouter_bundle
*mrouter
;
2824 struct xbundle
*mcast_xbundle
;
2826 LIST_FOR_EACH(mrouter
, mrouter_node
, &ms
->mrouter_lru
) {
2827 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, mrouter
->port
);
2828 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
2829 && mrouter
->vlan
== xvlan
->v
[0].vid
) {
2830 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast router port");
2831 mcast_output_add(out
, mcast_xbundle
);
2832 } else if (!mcast_xbundle
) {
2833 xlate_report(ctx
, OFT_WARN
,
2834 "mcast router port is unknown, dropping");
2835 } else if (mrouter
->vlan
!= xvlan
->v
[0].vid
) {
2836 xlate_report(ctx
, OFT_DETAIL
,
2837 "mcast router is on another vlan, dropping");
2839 xlate_report(ctx
, OFT_DETAIL
,
2840 "mcast router port is input port, dropping");
2845 /* send the packet to ports flagged to be flooded */
2847 xlate_normal_mcast_send_fports(struct xlate_ctx
*ctx
,
2848 struct mcast_snooping
*ms
,
2849 struct xbundle
*in_xbundle
,
2850 struct mcast_output
*out
)
2851 OVS_REQ_RDLOCK(ms
->rwlock
)
2853 struct mcast_port_bundle
*fport
;
2854 struct xbundle
*mcast_xbundle
;
2856 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2857 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2858 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2859 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast flood port");
2860 mcast_output_add(out
, mcast_xbundle
);
2861 } else if (!mcast_xbundle
) {
2862 xlate_report(ctx
, OFT_WARN
,
2863 "mcast flood port is unknown, dropping");
2865 xlate_report(ctx
, OFT_DETAIL
,
2866 "mcast flood port is input port, dropping");
2871 /* forward the Reports to configured ports */
2873 xlate_normal_mcast_send_rports(struct xlate_ctx
*ctx
,
2874 struct mcast_snooping
*ms
,
2875 struct xbundle
*in_xbundle
,
2876 struct mcast_output
*out
)
2877 OVS_REQ_RDLOCK(ms
->rwlock
)
2879 struct mcast_port_bundle
*rport
;
2880 struct xbundle
*mcast_xbundle
;
2882 LIST_FOR_EACH(rport
, node
, &ms
->rport_list
) {
2883 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, rport
->port
);
2885 && mcast_xbundle
!= in_xbundle
2886 && mcast_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
2887 xlate_report(ctx
, OFT_DETAIL
,
2888 "forwarding report to mcast flagged port");
2889 mcast_output_add(out
, mcast_xbundle
);
2890 } else if (!mcast_xbundle
) {
2891 xlate_report(ctx
, OFT_WARN
,
2892 "mcast port is unknown, dropping the report");
2894 xlate_report(ctx
, OFT_DETAIL
,
2895 "mcast port is input port, dropping the Report");
2901 xlate_normal_flood(struct xlate_ctx
*ctx
, struct xbundle
*in_xbundle
,
2902 struct xvlan
*xvlan
)
2904 struct xbundle
*xbundle
;
2906 LIST_FOR_EACH (xbundle
, list_node
, &ctx
->xbridge
->xbundles
) {
2907 if (xbundle
!= in_xbundle
2908 && xbundle
->ofbundle
!= in_xbundle
->ofbundle
2909 && xbundle_includes_vlan(xbundle
, xvlan
)
2910 && xbundle
->floodable
2911 && !xbundle_mirror_out(ctx
->xbridge
, xbundle
)) {
2912 output_normal(ctx
, xbundle
, xvlan
);
2915 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2919 is_ip_local_multicast(const struct flow
*flow
, struct flow_wildcards
*wc
)
2921 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2922 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2923 return ip_is_local_multicast(flow
->nw_dst
);
2924 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2925 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
2926 return ipv6_is_all_hosts(&flow
->ipv6_dst
);
2933 xlate_normal(struct xlate_ctx
*ctx
)
2935 struct flow_wildcards
*wc
= ctx
->wc
;
2936 struct flow
*flow
= &ctx
->xin
->flow
;
2937 struct xbundle
*in_xbundle
;
2938 struct xport
*in_port
;
2939 struct mac_entry
*mac
;
2941 struct xvlan in_xvlan
;
2945 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
2946 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2947 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
2949 in_xbundle
= lookup_input_bundle(ctx
, flow
->in_port
.ofp_port
, &in_port
);
2951 xlate_report(ctx
, OFT_WARN
, "no input bundle, dropping");
2955 /* Drop malformed frames. */
2956 if (eth_type_vlan(flow
->dl_type
) &&
2957 !(flow
->vlans
[0].tci
& htons(VLAN_CFI
))) {
2958 if (ctx
->xin
->packet
!= NULL
) {
2959 xlate_report_error(ctx
, "dropping packet with partial "
2960 "VLAN tag received on port %s",
2963 xlate_report(ctx
, OFT_WARN
, "partial VLAN tag, dropping");
2967 /* Drop frames on bundles reserved for mirroring. */
2968 if (xbundle_mirror_out(ctx
->xbridge
, in_xbundle
)) {
2969 if (ctx
->xin
->packet
!= NULL
) {
2970 xlate_report_error(ctx
, "dropping packet received on port %s, "
2971 "which is reserved exclusively for mirroring",
2974 xlate_report(ctx
, OFT_WARN
,
2975 "input port is mirror output port, dropping");
2980 xvlan_extract(flow
, &in_xvlan
);
2981 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, in_xbundle
)) {
2982 xlate_report(ctx
, OFT_WARN
,
2983 "disallowed VLAN VID for this input port, dropping");
2986 xvlan_input_translate(in_xbundle
, &in_xvlan
, &xvlan
);
2987 vlan
= xvlan
.v
[0].vid
;
2989 /* Check other admissibility requirements. */
2990 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
2994 /* Learn source MAC. */
2995 bool is_grat_arp
= is_gratuitous_arp(flow
, wc
);
2996 if (ctx
->xin
->allow_side_effects
2997 && flow
->packet_type
== htonl(PT_ETH
)
2998 && in_port
->pt_mode
!= NETDEV_PT_LEGACY_L3
3000 update_learning_table(ctx
, in_xbundle
, flow
->dl_src
, vlan
,
3003 if (ctx
->xin
->xcache
&& in_xbundle
!= &ofpp_none_bundle
) {
3004 struct xc_entry
*entry
;
3006 /* Save just enough info to update mac learning table later. */
3007 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NORMAL
);
3008 entry
->normal
.ofproto
= ctx
->xbridge
->ofproto
;
3009 entry
->normal
.in_port
= flow
->in_port
.ofp_port
;
3010 entry
->normal
.dl_src
= flow
->dl_src
;
3011 entry
->normal
.vlan
= vlan
;
3012 entry
->normal
.is_gratuitous_arp
= is_grat_arp
;
3015 /* Determine output bundle. */
3016 if (mcast_snooping_enabled(ctx
->xbridge
->ms
)
3017 && !eth_addr_is_broadcast(flow
->dl_dst
)
3018 && eth_addr_is_multicast(flow
->dl_dst
)
3019 && is_ip_any(flow
)) {
3020 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
3021 struct mcast_group
*grp
= NULL
;
3023 if (is_igmp(flow
, wc
)) {
3025 * IGMP packets need to take the slow path, in order to be
3026 * processed for mdb updates. That will prevent expires
3027 * firing off even after hosts have sent reports.
3029 ctx
->xout
->slow
|= SLOW_ACTION
;
3031 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
3032 if (mcast_snooping_is_membership(flow
->tp_src
) ||
3033 mcast_snooping_is_query(flow
->tp_src
)) {
3034 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3035 update_mcast_snooping_table(ctx
, flow
, vlan
,
3036 in_xbundle
, ctx
->xin
->packet
);
3040 if (mcast_snooping_is_membership(flow
->tp_src
)) {
3041 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3043 ovs_rwlock_rdlock(&ms
->rwlock
);
3044 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3046 /* RFC4541: section 2.1.1, item 1: A snooping switch should
3047 * forward IGMP Membership Reports only to those ports where
3048 * multicast routers are attached. Alternatively stated: a
3049 * snooping switch should not forward IGMP Membership Reports
3050 * to ports on which only hosts are attached.
3051 * An administrative control may be provided to override this
3052 * restriction, allowing the report messages to be flooded to
3054 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3055 ovs_rwlock_unlock(&ms
->rwlock
);
3057 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3059 xlate_report(ctx
, OFT_DETAIL
, "multicast traffic, flooding");
3060 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3063 } else if (is_mld(flow
, wc
)) {
3064 ctx
->xout
->slow
|= SLOW_ACTION
;
3065 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3066 update_mcast_snooping_table(ctx
, flow
, vlan
,
3067 in_xbundle
, ctx
->xin
->packet
);
3069 if (is_mld_report(flow
, wc
)) {
3070 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3072 ovs_rwlock_rdlock(&ms
->rwlock
);
3073 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3075 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3076 ovs_rwlock_unlock(&ms
->rwlock
);
3078 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3080 xlate_report(ctx
, OFT_DETAIL
, "MLD query, flooding");
3081 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3084 if (is_ip_local_multicast(flow
, wc
)) {
3085 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
3086 * address in the 224.0.0.x range which are not IGMP must
3087 * be forwarded on all ports */
3088 xlate_report(ctx
, OFT_DETAIL
,
3089 "RFC4541: section 2.1.2, item 2, flooding");
3090 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3095 /* forwarding to group base ports */
3096 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3098 ovs_rwlock_rdlock(&ms
->rwlock
);
3099 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3100 grp
= mcast_snooping_lookup4(ms
, flow
->nw_dst
, vlan
);
3101 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3102 grp
= mcast_snooping_lookup(ms
, &flow
->ipv6_dst
, vlan
);
3105 xlate_normal_mcast_send_group(ctx
, ms
, grp
, in_xbundle
, &out
);
3106 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3107 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3110 if (mcast_snooping_flood_unreg(ms
)) {
3111 xlate_report(ctx
, OFT_DETAIL
,
3112 "unregistered multicast, flooding");
3115 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3117 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3120 ovs_rwlock_unlock(&ms
->rwlock
);
3122 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3124 ovs_rwlock_rdlock(&ctx
->xbridge
->ml
->rwlock
);
3125 mac
= mac_learning_lookup(ctx
->xbridge
->ml
, flow
->dl_dst
, vlan
);
3126 mac_port
= mac
? mac_entry_get_port(ctx
->xbridge
->ml
, mac
) : NULL
;
3127 ovs_rwlock_unlock(&ctx
->xbridge
->ml
->rwlock
);
3130 struct xbundle
*mac_xbundle
= xbundle_lookup(ctx
->xcfg
, mac_port
);
3132 if (mac_xbundle
&& xbundle_mirror_out(ctx
->xbridge
, mac_xbundle
)) {
3133 xlate_report(ctx
, OFT_WARN
,
3134 "learned port is a mirror port, dropping");
3139 && mac_xbundle
!= in_xbundle
3140 && mac_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
3141 xlate_report(ctx
, OFT_DETAIL
, "forwarding to learned port");
3142 output_normal(ctx
, mac_xbundle
, &xvlan
);
3143 } else if (!mac_xbundle
) {
3144 xlate_report(ctx
, OFT_WARN
,
3145 "learned port is unknown, dropping");
3147 xlate_report(ctx
, OFT_DETAIL
,
3148 "learned port is input port, dropping");
3151 xlate_report(ctx
, OFT_DETAIL
,
3152 "no learned MAC for destination, flooding");
3153 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3158 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
3159 * 'probability' is the number of packets out of UINT32_MAX to sample. The
3160 * 'cookie' is passed back in the callback for each sampled packet.
3161 * 'tunnel_out_port', if not ODPP_NONE, is added as the
3162 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions',
3163 * an OVS_USERSPACE_ATTR_ACTIONS attribute is added. If
3164 * 'emit_set_tunnel', sample(sampling_port=1) would translate into
3165 * datapath sample action set(tunnel(...)), sample(...) and it is used
3166 * for sampling egress tunnel information.
3169 compose_sample_action(struct xlate_ctx
*ctx
,
3170 const uint32_t probability
,
3171 const struct user_action_cookie
*cookie
,
3172 const odp_port_t tunnel_out_port
,
3173 bool include_actions
)
3175 if (probability
== 0) {
3176 /* No need to generate sampling or the inner action. */
3180 /* If the slow path meter is configured by the controller,
3181 * insert a meter action before the user space action. */
3182 struct ofproto
*ofproto
= &ctx
->xin
->ofproto
->up
;
3183 uint32_t meter_id
= ofproto
->slowpath_meter_id
;
3185 /* When meter action is not required, avoid generate sample action
3186 * for 100% sampling rate. */
3187 bool is_sample
= probability
< UINT32_MAX
|| meter_id
!= UINT32_MAX
;
3188 size_t sample_offset
= 0, actions_offset
= 0;
3190 sample_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3191 OVS_ACTION_ATTR_SAMPLE
);
3192 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
3194 actions_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3195 OVS_SAMPLE_ATTR_ACTIONS
);
3198 if (meter_id
!= UINT32_MAX
) {
3199 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
3202 odp_port_t odp_port
= ofp_port_to_odp_port(
3203 ctx
->xbridge
, ctx
->xin
->flow
.in_port
.ofp_port
);
3204 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
3205 size_t cookie_offset
= odp_put_userspace_action(pid
, cookie
,
3212 nl_msg_end_nested(ctx
->odp_actions
, actions_offset
);
3213 nl_msg_end_nested(ctx
->odp_actions
, sample_offset
);
3216 return cookie_offset
;
3219 /* If sFLow is not enabled, returns 0 without doing anything.
3221 * If sFlow is enabled, appends a template "sample" action to the ODP actions
3222 * in 'ctx'. This action is a template because some of the information needed
3223 * to fill it out is not available until flow translation is complete. In this
3224 * case, this functions returns an offset, which is always nonzero, to pass
3225 * later to fix_sflow_action() to fill in the rest of the template. */
3227 compose_sflow_action(struct xlate_ctx
*ctx
)
3229 struct dpif_sflow
*sflow
= ctx
->xbridge
->sflow
;
3230 if (!sflow
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3234 struct user_action_cookie cookie
;
3236 memset(&cookie
, 0, sizeof cookie
);
3237 cookie
.type
= USER_ACTION_COOKIE_SFLOW
;
3238 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
3239 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
3241 return compose_sample_action(ctx
, dpif_sflow_get_probability(sflow
),
3242 &cookie
, ODPP_NONE
, true);
3245 /* If flow IPFIX is enabled, make sure IPFIX flow sample action
3246 * at egress point of tunnel port is just in front of corresponding
3247 * output action. If bridge IPFIX is enabled, this appends an IPFIX
3248 * sample action to 'ctx->odp_actions'. */
3250 compose_ipfix_action(struct xlate_ctx
*ctx
, odp_port_t output_odp_port
)
3252 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
3253 odp_port_t tunnel_out_port
= ODPP_NONE
;
3255 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3259 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
3261 if (output_odp_port
== ODPP_NONE
&&
3262 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix
)) {
3266 /* For output case, output_odp_port is valid. */
3267 if (output_odp_port
!= ODPP_NONE
) {
3268 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix
)) {
3271 /* If tunnel sampling is enabled, put an additional option attribute:
3272 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
3274 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix
) &&
3275 dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
) ) {
3276 tunnel_out_port
= output_odp_port
;
3280 struct user_action_cookie cookie
;
3282 memset(&cookie
, 0, sizeof cookie
);
3283 cookie
.type
= USER_ACTION_COOKIE_IPFIX
;
3284 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
3285 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
3286 cookie
.ipfix
.output_odp_port
= output_odp_port
;
3288 compose_sample_action(ctx
,
3289 dpif_ipfix_get_bridge_exporter_probability(ipfix
),
3290 &cookie
, tunnel_out_port
, false);
3293 /* Fix "sample" action according to data collected while composing ODP actions,
3294 * as described in compose_sflow_action().
3296 * 'user_cookie_offset' must be the offset returned by
3297 * compose_sflow_action(). */
3299 fix_sflow_action(struct xlate_ctx
*ctx
, unsigned int user_cookie_offset
)
3301 const struct flow
*base
= &ctx
->base_flow
;
3302 struct user_action_cookie
*cookie
;
3304 cookie
= ofpbuf_at(ctx
->odp_actions
, user_cookie_offset
, sizeof *cookie
);
3305 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
3307 cookie
->sflow
.vlan_tci
= base
->vlans
[0].tci
;
3309 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
3310 * port information") for the interpretation of cookie->output. */
3311 switch (ctx
->sflow_n_outputs
) {
3313 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
3314 cookie
->sflow
.output
= 0x40000000 | 256;
3318 cookie
->sflow
.output
= dpif_sflow_odp_port_to_ifindex(
3319 ctx
->xbridge
->sflow
, ctx
->sflow_odp_port
);
3320 if (cookie
->sflow
.output
) {
3325 /* 0x80000000 means "multiple output ports. */
3326 cookie
->sflow
.output
= 0x80000000 | ctx
->sflow_n_outputs
;
3332 process_special(struct xlate_ctx
*ctx
, const struct xport
*xport
)
3334 const struct flow
*flow
= &ctx
->xin
->flow
;
3335 struct flow_wildcards
*wc
= ctx
->wc
;
3336 const struct xbridge
*xbridge
= ctx
->xbridge
;
3337 const struct dp_packet
*packet
= ctx
->xin
->packet
;
3338 enum slow_path_reason slow
;
3339 bool lacp_may_enable
;
3343 } else if (xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
)) {
3345 cfm_process_heartbeat(xport
->cfm
, packet
);
3348 } else if (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
, wc
)) {
3350 bfd_process_packet(xport
->bfd
, flow
, packet
);
3351 /* If POLL received, immediately sends FINAL back. */
3352 if (bfd_should_send_packet(xport
->bfd
)) {
3353 ofproto_dpif_monitor_port_send_soon(xport
->ofport
);
3357 } else if (xport
->xbundle
&& xport
->xbundle
->lacp
3358 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
3360 lacp_may_enable
= lacp_process_packet(xport
->xbundle
->lacp
,
3361 xport
->ofport
, packet
);
3362 /* Update LACP status in bond-slave to avoid packet-drops until
3363 * LACP state machine is run by the main thread. */
3364 if (xport
->xbundle
->bond
&& lacp_may_enable
) {
3365 bond_slave_set_may_enable(xport
->xbundle
->bond
, xport
->ofport
,
3370 } else if ((xbridge
->stp
|| xbridge
->rstp
) &&
3371 stp_should_process_flow(flow
, wc
)) {
3374 ? stp_process_packet(xport
, packet
)
3375 : rstp_process_packet(xport
, packet
);
3378 } else if (xport
->lldp
&& lldp_should_process_flow(xport
->lldp
, flow
)) {
3380 lldp_process_packet(xport
->lldp
, packet
);
3388 ctx
->xout
->slow
|= slow
;
3396 tnl_route_lookup_flow(const struct xlate_ctx
*ctx
,
3397 const struct flow
*oflow
,
3398 struct in6_addr
*ip
, struct in6_addr
*src
,
3399 struct xport
**out_port
)
3401 char out_dev
[IFNAMSIZ
];
3402 struct xbridge
*xbridge
;
3404 struct in6_addr dst
;
3406 dst
= flow_tnl_dst(&oflow
->tunnel
);
3407 if (!ovs_router_lookup(oflow
->pkt_mark
, &dst
, out_dev
, src
, &gw
)) {
3411 if (ipv6_addr_is_set(&gw
) &&
3412 (!IN6_IS_ADDR_V4MAPPED(&gw
) || in6_addr_get_mapped_ipv4(&gw
))) {
3418 HMAP_FOR_EACH (xbridge
, hmap_node
, &ctx
->xcfg
->xbridges
) {
3419 if (!strncmp(xbridge
->name
, out_dev
, IFNAMSIZ
)) {
3422 HMAP_FOR_EACH (port
, ofp_node
, &xbridge
->xports
) {
3423 if (!strncmp(netdev_get_name(port
->netdev
), out_dev
, IFNAMSIZ
)) {
3431 /* If tunnel IP isn't configured on bridges, then we search all ports. */
3432 HMAP_FOR_EACH (xbridge
, hmap_node
, &ctx
->xcfg
->xbridges
) {
3435 HMAP_FOR_EACH (port
, ofp_node
, &xbridge
->xports
) {
3436 if (!strncmp(netdev_get_name(port
->netdev
),
3437 out_dev
, IFNAMSIZ
)) {
3447 compose_table_xlate(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3448 struct dp_packet
*packet
)
3450 struct xbridge
*xbridge
= out_dev
->xbridge
;
3451 ovs_version_t version
= ofproto_dpif_get_tables_version(xbridge
->ofproto
);
3452 struct ofpact_output output
;
3455 ofpact_init(&output
.ofpact
, OFPACT_OUTPUT
, sizeof output
);
3456 flow_extract(packet
, &flow
);
3457 flow
.in_port
.ofp_port
= out_dev
->ofp_port
;
3458 output
.port
= OFPP_TABLE
;
3461 return ofproto_dpif_execute_actions__(xbridge
->ofproto
, version
, &flow
,
3462 NULL
, &output
.ofpact
, sizeof output
,
3463 ctx
->depth
, ctx
->resubmits
, packet
);
3467 tnl_send_nd_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3468 const struct eth_addr eth_src
,
3469 struct in6_addr
* ipv6_src
, struct in6_addr
* ipv6_dst
)
3471 struct dp_packet packet
;
3473 dp_packet_init(&packet
, 0);
3474 compose_nd_ns(&packet
, eth_src
, ipv6_src
, ipv6_dst
);
3475 compose_table_xlate(ctx
, out_dev
, &packet
);
3476 dp_packet_uninit(&packet
);
3480 tnl_send_arp_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3481 const struct eth_addr eth_src
,
3482 ovs_be32 ip_src
, ovs_be32 ip_dst
)
3484 struct dp_packet packet
;
3486 dp_packet_init(&packet
, 0);
3487 compose_arp(&packet
, ARP_OP_REQUEST
,
3488 eth_src
, eth_addr_zero
, true, ip_src
, ip_dst
);
3490 compose_table_xlate(ctx
, out_dev
, &packet
);
3491 dp_packet_uninit(&packet
);
3495 propagate_tunnel_data_to_flow__(struct flow
*dst_flow
,
3496 const struct flow
*src_flow
,
3497 struct eth_addr dmac
, struct eth_addr smac
,
3498 struct in6_addr s_ip6
, ovs_be32 s_ip
,
3499 bool is_tnl_ipv6
, uint8_t nw_proto
)
3501 dst_flow
->dl_dst
= dmac
;
3502 dst_flow
->dl_src
= smac
;
3504 dst_flow
->packet_type
= htonl(PT_ETH
);
3505 dst_flow
->nw_dst
= src_flow
->tunnel
.ip_dst
;
3506 dst_flow
->nw_src
= src_flow
->tunnel
.ip_src
;
3507 dst_flow
->ipv6_dst
= src_flow
->tunnel
.ipv6_dst
;
3508 dst_flow
->ipv6_src
= src_flow
->tunnel
.ipv6_src
;
3510 dst_flow
->nw_frag
= 0; /* Tunnel packets are unfragmented. */
3511 dst_flow
->nw_tos
= src_flow
->tunnel
.ip_tos
;
3512 dst_flow
->nw_ttl
= src_flow
->tunnel
.ip_ttl
;
3513 dst_flow
->tp_dst
= src_flow
->tunnel
.tp_dst
;
3514 dst_flow
->tp_src
= src_flow
->tunnel
.tp_src
;
3517 dst_flow
->dl_type
= htons(ETH_TYPE_IPV6
);
3518 if (ipv6_mask_is_any(&dst_flow
->ipv6_src
)
3519 && !ipv6_mask_is_any(&s_ip6
)) {
3520 dst_flow
->ipv6_src
= s_ip6
;
3523 dst_flow
->dl_type
= htons(ETH_TYPE_IP
);
3524 if (dst_flow
->nw_src
== 0 && s_ip
) {
3525 dst_flow
->nw_src
= s_ip
;
3528 dst_flow
->nw_proto
= nw_proto
;
3532 * Populate the 'flow' and 'base_flow' L3 fields to do the post tunnel push
3536 propagate_tunnel_data_to_flow(struct xlate_ctx
*ctx
, struct eth_addr dmac
,
3537 struct eth_addr smac
, struct in6_addr s_ip6
,
3538 ovs_be32 s_ip
, bool is_tnl_ipv6
,
3539 enum ovs_vport_type tnl_type
)
3541 struct flow
*base_flow
, *flow
;
3542 flow
= &ctx
->xin
->flow
;
3543 base_flow
= &ctx
->base_flow
;
3544 uint8_t nw_proto
= 0;
3547 case OVS_VPORT_TYPE_GRE
:
3548 case OVS_VPORT_TYPE_ERSPAN
:
3549 case OVS_VPORT_TYPE_IP6ERSPAN
:
3550 case OVS_VPORT_TYPE_IP6GRE
:
3551 nw_proto
= IPPROTO_GRE
;
3553 case OVS_VPORT_TYPE_VXLAN
:
3554 case OVS_VPORT_TYPE_GENEVE
:
3555 nw_proto
= IPPROTO_UDP
;
3557 case OVS_VPORT_TYPE_LISP
:
3558 case OVS_VPORT_TYPE_STT
:
3559 case OVS_VPORT_TYPE_UNSPEC
:
3560 case OVS_VPORT_TYPE_NETDEV
:
3561 case OVS_VPORT_TYPE_INTERNAL
:
3562 case __OVS_VPORT_TYPE_MAX
:
3567 * Update base_flow first followed by flow as the dst_flow gets modified
3570 propagate_tunnel_data_to_flow__(base_flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3571 is_tnl_ipv6
, nw_proto
);
3572 propagate_tunnel_data_to_flow__(flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3573 is_tnl_ipv6
, nw_proto
);
3577 native_tunnel_output(struct xlate_ctx
*ctx
, const struct xport
*xport
,
3578 const struct flow
*flow
, odp_port_t tunnel_odp_port
,
3581 struct netdev_tnl_build_header_params tnl_params
;
3582 struct ovs_action_push_tnl tnl_push_data
;
3583 struct xport
*out_dev
= NULL
;
3584 ovs_be32 s_ip
= 0, d_ip
= 0;
3585 struct in6_addr s_ip6
= in6addr_any
;
3586 struct in6_addr d_ip6
= in6addr_any
;
3587 struct eth_addr smac
;
3588 struct eth_addr dmac
;
3590 char buf_sip6
[INET6_ADDRSTRLEN
];
3591 char buf_dip6
[INET6_ADDRSTRLEN
];
3593 /* Store sFlow data. */
3594 uint32_t sflow_n_outputs
= ctx
->sflow_n_outputs
;
3596 /* Structures to backup Ethernet and IP of base_flow. */
3597 struct flow old_base_flow
;
3598 struct flow old_flow
;
3600 /* Backup flow & base_flow data. */
3601 memcpy(&old_base_flow
, &ctx
->base_flow
, sizeof old_base_flow
);
3602 memcpy(&old_flow
, &ctx
->xin
->flow
, sizeof old_flow
);
3604 if (flow
->tunnel
.ip_src
) {
3605 in6_addr_set_mapped_ipv4(&s_ip6
, flow
->tunnel
.ip_src
);
3608 err
= tnl_route_lookup_flow(ctx
, flow
, &d_ip6
, &s_ip6
, &out_dev
);
3610 xlate_report(ctx
, OFT_WARN
, "native tunnel routing failed");
3614 xlate_report(ctx
, OFT_DETAIL
, "tunneling to %s via %s",
3615 ipv6_string_mapped(buf_dip6
, &d_ip6
),
3616 netdev_get_name(out_dev
->netdev
));
3618 /* Use mac addr of bridge port of the peer. */
3619 err
= netdev_get_etheraddr(out_dev
->netdev
, &smac
);
3621 xlate_report(ctx
, OFT_WARN
,
3622 "tunnel output device lacks Ethernet address");
3626 d_ip
= in6_addr_get_mapped_ipv4(&d_ip6
);
3628 s_ip
= in6_addr_get_mapped_ipv4(&s_ip6
);
3631 err
= tnl_neigh_lookup(out_dev
->xbridge
->name
, &d_ip6
, &dmac
);
3633 xlate_report(ctx
, OFT_DETAIL
,
3634 "neighbor cache miss for %s on bridge %s, "
3635 "sending %s request",
3636 buf_dip6
, out_dev
->xbridge
->name
, d_ip
? "ARP" : "ND");
3638 tnl_send_arp_request(ctx
, out_dev
, smac
, s_ip
, d_ip
);
3640 tnl_send_nd_request(ctx
, out_dev
, smac
, &s_ip6
, &d_ip6
);
3645 if (ctx
->xin
->xcache
) {
3646 struct xc_entry
*entry
;
3648 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TNL_NEIGH
);
3649 ovs_strlcpy(entry
->tnl_neigh_cache
.br_name
, out_dev
->xbridge
->name
,
3650 sizeof entry
->tnl_neigh_cache
.br_name
);
3651 entry
->tnl_neigh_cache
.d_ipv6
= d_ip6
;
3654 xlate_report(ctx
, OFT_DETAIL
, "tunneling from "ETH_ADDR_FMT
" %s"
3655 " to "ETH_ADDR_FMT
" %s",
3656 ETH_ADDR_ARGS(smac
), ipv6_string_mapped(buf_sip6
, &s_ip6
),
3657 ETH_ADDR_ARGS(dmac
), buf_dip6
);
3659 netdev_init_tnl_build_header_params(&tnl_params
, flow
, &s_ip6
, dmac
, smac
);
3660 err
= tnl_port_build_header(xport
->ofport
, &tnl_push_data
, &tnl_params
);
3664 tnl_push_data
.tnl_port
= tunnel_odp_port
;
3665 tnl_push_data
.out_port
= out_dev
->odp_port
;
3667 /* After tunnel header has been added, MAC and IP data of flow and
3668 * base_flow need to be set properly, since there is not recirculation
3669 * any more when sending packet to tunnel. */
3671 propagate_tunnel_data_to_flow(ctx
, dmac
, smac
, s_ip6
,
3672 s_ip
, tnl_params
.is_ipv6
,
3673 tnl_push_data
.tnl_type
);
3675 size_t clone_ofs
= 0;
3676 size_t push_action_size
;
3678 clone_ofs
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
3679 odp_put_tnl_push_action(ctx
->odp_actions
, &tnl_push_data
);
3680 push_action_size
= ctx
->odp_actions
->size
;
3683 const struct dpif_flow_stats
*backup_resubmit_stats
;
3684 struct xlate_cache
*backup_xcache
;
3685 struct flow_wildcards
*backup_wc
, wc
;
3686 bool backup_side_effects
;
3687 const struct dp_packet
*backup_packet
;
3689 memset(&wc
, 0 , sizeof wc
);
3690 backup_wc
= ctx
->wc
;
3692 ctx
->xin
->wc
= NULL
;
3693 backup_resubmit_stats
= ctx
->xin
->resubmit_stats
;
3694 backup_xcache
= ctx
->xin
->xcache
;
3695 backup_side_effects
= ctx
->xin
->allow_side_effects
;
3696 backup_packet
= ctx
->xin
->packet
;
3698 ctx
->xin
->resubmit_stats
= NULL
;
3699 ctx
->xin
->xcache
= xlate_cache_new(); /* Use new temporary cache. */
3700 ctx
->xin
->allow_side_effects
= false;
3701 ctx
->xin
->packet
= NULL
;
3703 /* Push the cache entry for the tunnel first. */
3704 struct xc_entry
*entry
;
3705 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TUNNEL_HEADER
);
3706 entry
->tunnel_hdr
.hdr_size
= tnl_push_data
.header_len
;
3707 entry
->tunnel_hdr
.operation
= ADD
;
3709 patch_port_output(ctx
, xport
, out_dev
);
3711 /* Similar to the stats update in revalidation, the x_cache entries
3712 * are populated by the previous translation are used to update the
3715 if (backup_resubmit_stats
) {
3716 struct dpif_flow_stats stats
= *backup_resubmit_stats
;
3717 xlate_push_stats(ctx
->xin
->xcache
, &stats
, false);
3719 xlate_cache_steal_entries(backup_xcache
, ctx
->xin
->xcache
);
3721 if (ctx
->odp_actions
->size
> push_action_size
) {
3722 nl_msg_end_non_empty_nested(ctx
->odp_actions
, clone_ofs
);
3724 nl_msg_cancel_nested(ctx
->odp_actions
, clone_ofs
);
3727 /* Restore context status. */
3728 ctx
->xin
->resubmit_stats
= backup_resubmit_stats
;
3729 xlate_cache_delete(ctx
->xin
->xcache
);
3730 ctx
->xin
->xcache
= backup_xcache
;
3731 ctx
->xin
->allow_side_effects
= backup_side_effects
;
3732 ctx
->xin
->packet
= backup_packet
;
3733 ctx
->wc
= backup_wc
;
3735 /* In order to maintain accurate stats, use recirc for
3736 * natvie tunneling. */
3737 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, 0);
3738 nl_msg_end_nested(ctx
->odp_actions
, clone_ofs
);
3741 /* Restore the flows after the translation. */
3742 memcpy(&ctx
->xin
->flow
, &old_flow
, sizeof ctx
->xin
->flow
);
3743 memcpy(&ctx
->base_flow
, &old_base_flow
, sizeof ctx
->base_flow
);
3745 /* Restore sFlow data. */
3746 ctx
->sflow_n_outputs
= sflow_n_outputs
;
3752 xlate_commit_actions(struct xlate_ctx
*ctx
)
3754 bool use_masked
= ctx
->xbridge
->support
.masked_set_action
;
3756 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
3757 ctx
->odp_actions
, ctx
->wc
,
3758 use_masked
, ctx
->pending_encap
,
3759 ctx
->pending_decap
, ctx
->encap_data
);
3760 ctx
->pending_encap
= false;
3761 ctx
->pending_decap
= false;
3762 ofpbuf_delete(ctx
->encap_data
);
3763 ctx
->encap_data
= NULL
;
3767 clear_conntrack(struct xlate_ctx
*ctx
)
3769 ctx
->conntracked
= false;
3770 flow_clear_conntrack(&ctx
->xin
->flow
);
3774 xlate_flow_is_protected(const struct xlate_ctx
*ctx
, const struct flow
*flow
, const struct xport
*xport_out
)
3776 const struct xport
*xport_in
;
3782 xport_in
= get_ofp_port(ctx
->xbridge
, flow
->in_port
.ofp_port
);
3784 return (xport_in
&& xport_in
->xbundle
&& xport_out
->xbundle
&&
3785 xport_in
->xbundle
->protected && xport_out
->xbundle
->protected);
3788 /* Function handles when a packet is sent from one bridge to another bridge.
3790 * The bridges are internally connected, either with patch ports or with
3793 * The output action to another bridge causes translation to continue within
3794 * the next bridge. This process can be recursive; the next bridge can
3795 * output yet to another bridge.
3797 * The translated actions from the second bridge onwards are enclosed within
3798 * the clone action, so that any modification to the packet will not be visible
3799 * to the remaining actions of the originating bridge.
3802 patch_port_output(struct xlate_ctx
*ctx
, const struct xport
*in_dev
,
3803 struct xport
*out_dev
)
3805 struct flow
*flow
= &ctx
->xin
->flow
;
3806 struct flow old_flow
= ctx
->xin
->flow
;
3807 struct flow_tnl old_flow_tnl_wc
= ctx
->wc
->masks
.tunnel
;
3808 bool old_conntrack
= ctx
->conntracked
;
3809 bool old_was_mpls
= ctx
->was_mpls
;
3810 ovs_version_t old_version
= ctx
->xin
->tables_version
;
3811 struct ofpbuf old_stack
= ctx
->stack
;
3812 uint8_t new_stack
[1024];
3813 struct ofpbuf old_action_set
= ctx
->action_set
;
3814 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
3815 uint64_t actset_stub
[1024 / 8];
3817 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
3818 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
3819 flow
->in_port
.ofp_port
= out_dev
->ofp_port
;
3820 flow
->metadata
= htonll(0);
3821 memset(&flow
->tunnel
, 0, sizeof flow
->tunnel
);
3822 memset(&ctx
->wc
->masks
.tunnel
, 0, sizeof ctx
->wc
->masks
.tunnel
);
3823 flow
->tunnel
.metadata
.tab
=
3824 ofproto_get_tun_tab(&out_dev
->xbridge
->ofproto
->up
);
3825 ctx
->wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
3826 memset(flow
->regs
, 0, sizeof flow
->regs
);
3827 flow
->actset_output
= OFPP_UNSET
;
3828 clear_conntrack(ctx
);
3829 ctx
->xin
->trace
= xlate_report(ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
3830 out_dev
->xbridge
->name
);
3831 mirror_mask_t old_mirrors
= ctx
->mirrors
;
3832 bool independent_mirrors
= out_dev
->xbridge
!= ctx
->xbridge
;
3833 if (independent_mirrors
) {
3836 ctx
->xbridge
= out_dev
->xbridge
;
3838 /* The bridge is now known so obtain its table version. */
3839 ctx
->xin
->tables_version
3840 = ofproto_dpif_get_tables_version(ctx
->xbridge
->ofproto
);
3842 if (!process_special(ctx
, out_dev
) && may_receive(out_dev
, ctx
)) {
3843 if (xport_stp_forward_state(out_dev
) &&
3844 xport_rstp_forward_state(out_dev
)) {
3845 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3846 false, true, clone_xlate_actions
);
3847 if (!ctx
->freezing
) {
3848 xlate_action_set(ctx
);
3850 if (ctx
->freezing
) {
3851 finish_freezing(ctx
);
3854 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
3855 * the learning action look at the packet, then drop it. */
3856 struct flow old_base_flow
= ctx
->base_flow
;
3857 size_t old_size
= ctx
->odp_actions
->size
;
3858 mirror_mask_t old_mirrors2
= ctx
->mirrors
;
3860 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3861 false, true, clone_xlate_actions
);
3862 ctx
->mirrors
= old_mirrors2
;
3863 ctx
->base_flow
= old_base_flow
;
3864 ctx
->odp_actions
->size
= old_size
;
3866 /* Undo changes that may have been done for freezing. */
3867 ctx_cancel_freeze(ctx
);
3871 ctx
->xin
->trace
= old_trace
;
3872 if (independent_mirrors
) {
3873 ctx
->mirrors
= old_mirrors
;
3875 ctx
->xin
->flow
= old_flow
;
3876 ctx
->xbridge
= in_dev
->xbridge
;
3877 ofpbuf_uninit(&ctx
->action_set
);
3878 ctx
->action_set
= old_action_set
;
3879 ofpbuf_uninit(&ctx
->stack
);
3880 ctx
->stack
= old_stack
;
3882 /* Restore calling bridge's lookup version. */
3883 ctx
->xin
->tables_version
= old_version
;
3885 /* Restore to calling bridge tunneling information */
3886 ctx
->wc
->masks
.tunnel
= old_flow_tnl_wc
;
3888 /* The out bridge popping MPLS should have no effect on the original
3890 ctx
->was_mpls
= old_was_mpls
;
3892 /* The out bridge's conntrack execution should have no effect on the
3893 * original bridge. */
3894 ctx
->conntracked
= old_conntrack
;
3896 /* The fact that the out bridge exits (for any reason) does not mean
3897 * that the original bridge should exit. Specifically, if the out
3898 * bridge freezes translation, the original bridge must continue
3899 * processing with the original, not the frozen packet! */
3902 /* Out bridge errors do not propagate back. */
3903 ctx
->error
= XLATE_OK
;
3905 if (ctx
->xin
->resubmit_stats
) {
3906 netdev_vport_inc_tx(in_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3907 netdev_vport_inc_rx(out_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3909 bfd_account_rx(out_dev
->bfd
, ctx
->xin
->resubmit_stats
);
3912 if (ctx
->xin
->xcache
) {
3913 struct xc_entry
*entry
;
3915 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
3916 entry
->dev
.tx
= netdev_ref(in_dev
->netdev
);
3917 entry
->dev
.rx
= netdev_ref(out_dev
->netdev
);
3918 entry
->dev
.bfd
= bfd_ref(out_dev
->bfd
);
3923 check_output_prerequisites(struct xlate_ctx
*ctx
,
3924 const struct xport
*xport
,
3928 struct flow_wildcards
*wc
= ctx
->wc
;
3931 xlate_report(ctx
, OFT_WARN
, "Nonexistent output port");
3933 } else if (xport
->config
& OFPUTIL_PC_NO_FWD
) {
3934 xlate_report(ctx
, OFT_DETAIL
, "OFPPC_NO_FWD set, skipping output");
3936 } else if (ctx
->mirror_snaplen
!= 0 && xport
->odp_port
== ODPP_NONE
) {
3937 xlate_report(ctx
, OFT_WARN
,
3938 "Mirror truncate to ODPP_NONE, skipping output");
3940 } else if (xlate_flow_is_protected(ctx
, flow
, xport
)) {
3941 xlate_report(ctx
, OFT_WARN
,
3942 "Flow is between protected ports, skipping output.");
3944 } else if (check_stp
) {
3945 if (is_stp(&ctx
->base_flow
)) {
3946 if (!xport_stp_should_forward_bpdu(xport
) &&
3947 !xport_rstp_should_manage_bpdu(xport
)) {
3948 if (ctx
->xbridge
->stp
!= NULL
) {
3949 xlate_report(ctx
, OFT_WARN
,
3950 "STP not in listening state, "
3951 "skipping bpdu output");
3952 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3953 xlate_report(ctx
, OFT_WARN
,
3954 "RSTP not managing BPDU in this state, "
3955 "skipping bpdu output");
3959 } else if ((xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
))
3960 || (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
,
3962 /* Pass; STP should not block link health detection. */
3963 } else if (!xport_stp_forward_state(xport
) ||
3964 !xport_rstp_forward_state(xport
)) {
3965 if (ctx
->xbridge
->stp
!= NULL
) {
3966 xlate_report(ctx
, OFT_WARN
,
3967 "STP not in forwarding state, skipping output");
3968 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3969 xlate_report(ctx
, OFT_WARN
,
3970 "RSTP not in forwarding state, skipping output");
3976 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L2
&&
3977 flow
->packet_type
!= htonl(PT_ETH
)) {
3978 xlate_report(ctx
, OFT_WARN
, "Trying to send non-Ethernet packet "
3979 "through legacy L2 port. Dropping packet.");
3986 /* Function verifies if destination address of received Neighbor Advertisement
3987 * message stored in 'flow' is correct. It should be either FF02::1:FFXX:XXXX
3988 * where XX:XXXX stands for the last 24 bits of 'ipv6_addr' or it should match
3991 is_nd_dst_correct(const struct flow
*flow
, const struct in6_addr
*ipv6_addr
)
3993 const uint8_t *flow_ipv6_addr
= (uint8_t *) &flow
->ipv6_dst
;
3994 const uint8_t *addr
= (uint8_t *) ipv6_addr
;
3996 return (IN6_IS_ADDR_MC_LINKLOCAL(&flow
->ipv6_dst
) &&
3997 flow_ipv6_addr
[11] == 0x01 &&
3998 flow_ipv6_addr
[12] == 0xff &&
3999 flow_ipv6_addr
[13] == addr
[13] &&
4000 flow_ipv6_addr
[14] == addr
[14] &&
4001 flow_ipv6_addr
[15] == addr
[15]) ||
4002 IN6_ARE_ADDR_EQUAL(&flow
->ipv6_dst
, ipv6_addr
);
4006 is_neighbor_reply_matched(const struct flow
*flow
, struct in6_addr
*ip_addr
)
4008 return ((IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
4009 flow
->dl_type
== htons(ETH_TYPE_ARP
) &&
4010 in6_addr_get_mapped_ipv4(ip_addr
) == flow
->nw_dst
) ||
4011 (!IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
4012 is_nd_dst_correct(flow
, ip_addr
)));
4015 /* Function verifies if the ARP reply or Neighbor Advertisement represented by
4016 * 'flow' addresses the 'xbridge' of 'ctx'. Returns true if the ARP TA or
4017 * neighbor discovery destination is in the list of configured IP addresses of
4018 * the bridge. Otherwise, it returns false. */
4020 is_neighbor_reply_correct(const struct xlate_ctx
*ctx
, const struct flow
*flow
)
4024 struct xbridge_addr
*xbridge_addr
= xbridge_addr_ref(ctx
->xbridge
->addr
);
4026 /* Verify if 'nw_dst' of ARP or 'ipv6_dst' of ICMPV6 is in the list. */
4027 for (i
= 0; xbridge_addr
&& i
< xbridge_addr
->n_addr
; i
++) {
4028 struct in6_addr
*ip_addr
= &xbridge_addr
->addr
[i
];
4029 if (is_neighbor_reply_matched(flow
, ip_addr
)) {
4030 /* Found a match. */
4036 xbridge_addr_unref(xbridge_addr
);
4038 /* If not found in bridge's IPs, search in its ports. */
4040 struct in6_addr
*ip_addr
, *mask
;
4044 HMAP_FOR_EACH (port
, ofp_node
, &ctx
->xbridge
->xports
) {
4045 error
= netdev_get_addr_list(port
->netdev
, &ip_addr
,
4048 ret
= is_neighbor_reply_matched(flow
, ip_addr
);
4052 /* Found a match. */
4062 terminate_native_tunnel(struct xlate_ctx
*ctx
, struct flow
*flow
,
4063 struct flow_wildcards
*wc
, odp_port_t
*tnl_port
)
4065 *tnl_port
= ODPP_NONE
;
4067 /* XXX: Write better Filter for tunnel port. We can use in_port
4068 * in tunnel-port flow to avoid these checks completely. */
4069 if (ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4070 *tnl_port
= tnl_port_map_lookup(flow
, wc
);
4072 /* If no tunnel port was found and it's about an ARP or ICMPv6 packet,
4073 * do tunnel neighbor snooping. */
4074 if (*tnl_port
== ODPP_NONE
&&
4075 (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
4076 flow
->nw_proto
== IPPROTO_ICMPV6
) &&
4077 is_neighbor_reply_correct(ctx
, flow
)) {
4078 tnl_neigh_snoop(flow
, wc
, ctx
->xbridge
->name
);
4082 return *tnl_port
!= ODPP_NONE
;
4086 compose_output_action__(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4087 const struct xlate_bond_recirc
*xr
, bool check_stp
,
4088 bool is_last_action OVS_UNUSED
, bool truncate
)
4090 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
4091 struct flow_wildcards
*wc
= ctx
->wc
;
4092 struct flow
*flow
= &ctx
->xin
->flow
;
4093 struct flow_tnl flow_tnl
;
4094 union flow_vlan_hdr flow_vlans
[FLOW_MAX_VLAN_HEADERS
];
4095 uint8_t flow_nw_tos
;
4096 odp_port_t out_port
, odp_port
, odp_tnl_port
;
4097 bool is_native_tunnel
= false;
4099 struct eth_addr flow_dl_dst
= flow
->dl_dst
;
4100 struct eth_addr flow_dl_src
= flow
->dl_src
;
4101 ovs_be32 flow_packet_type
= flow
->packet_type
;
4102 ovs_be16 flow_dl_type
= flow
->dl_type
;
4104 /* If 'struct flow' gets additional metadata, we'll need to zero it out
4105 * before traversing a patch port. */
4106 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 41);
4107 memset(&flow_tnl
, 0, sizeof flow_tnl
);
4109 if (!check_output_prerequisites(ctx
, xport
, flow
, check_stp
)) {
4113 if (flow
->packet_type
== htonl(PT_ETH
)) {
4114 /* Strip Ethernet header for legacy L3 port. */
4115 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L3
) {
4116 flow
->packet_type
= PACKET_TYPE_BE(OFPHTN_ETHERTYPE
,
4117 ntohs(flow
->dl_type
));
4123 xlate_report_error(ctx
, "Cannot truncate output to patch port");
4125 patch_port_output(ctx
, xport
, xport
->peer
);
4129 memcpy(flow_vlans
, flow
->vlans
, sizeof flow_vlans
);
4130 flow_nw_tos
= flow
->nw_tos
;
4132 if (count_skb_priorities(xport
)) {
4133 memset(&wc
->masks
.skb_priority
, 0xff, sizeof wc
->masks
.skb_priority
);
4134 if (dscp_from_skb_priority(xport
, flow
->skb_priority
, &dscp
)) {
4135 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
4136 flow
->nw_tos
&= ~IP_DSCP_MASK
;
4137 flow
->nw_tos
|= dscp
;
4141 if (xport
->is_tunnel
) {
4142 struct in6_addr dst
;
4143 /* Save tunnel metadata so that changes made due to
4144 * the Logical (tunnel) Port are not visible for any further
4145 * matches, while explicit set actions on tunnel metadata are.
4147 flow_tnl
= flow
->tunnel
;
4148 odp_port
= tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
4149 if (odp_port
== ODPP_NONE
) {
4150 xlate_report(ctx
, OFT_WARN
, "Tunneling decided against output");
4151 goto out
; /* restore flow_nw_tos */
4153 dst
= flow_tnl_dst(&flow
->tunnel
);
4154 if (ipv6_addr_equals(&dst
, &ctx
->orig_tunnel_ipv6_dst
)) {
4155 xlate_report(ctx
, OFT_WARN
, "Not tunneling to our own address");
4156 goto out
; /* restore flow_nw_tos */
4158 if (ctx
->xin
->resubmit_stats
) {
4159 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
4161 if (ctx
->xin
->xcache
) {
4162 struct xc_entry
*entry
;
4164 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
4165 entry
->dev
.tx
= netdev_ref(xport
->netdev
);
4167 out_port
= odp_port
;
4168 if (ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4169 xlate_report(ctx
, OFT_DETAIL
, "output to native tunnel");
4170 is_native_tunnel
= true;
4172 const char *tnl_type
;
4174 xlate_report(ctx
, OFT_DETAIL
, "output to kernel tunnel");
4175 tnl_type
= tnl_port_get_type(xport
->ofport
);
4176 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
4177 ctx
->odp_actions
, tnl_type
);
4178 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4181 odp_port
= xport
->odp_port
;
4182 out_port
= odp_port
;
4185 if (out_port
!= ODPP_NONE
) {
4186 /* Commit accumulated flow updates before output. */
4187 xlate_commit_actions(ctx
);
4190 /* Recirculate the packet. */
4191 struct ovs_action_hash
*act_hash
;
4194 enum ovs_hash_alg hash_alg
= xr
->hash_alg
;
4195 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4196 /* Algorithm supported by all datapaths. */
4197 hash_alg
= OVS_HASH_ALG_L4
;
4199 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4200 OVS_ACTION_ATTR_HASH
,
4202 act_hash
->hash_alg
= hash_alg
;
4203 act_hash
->hash_basis
= xr
->hash_basis
;
4205 /* Recirc action. */
4206 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
,
4208 } else if (is_native_tunnel
) {
4209 /* Output to native tunnel port. */
4210 native_tunnel_output(ctx
, xport
, flow
, odp_port
, truncate
);
4211 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4213 } else if (terminate_native_tunnel(ctx
, flow
, wc
,
4215 /* Intercept packet to be received on native tunnel port. */
4216 nl_msg_put_odp_port(ctx
->odp_actions
, OVS_ACTION_ATTR_TUNNEL_POP
,
4220 /* Tunnel push-pop action is not compatible with
4222 compose_ipfix_action(ctx
, out_port
);
4224 /* Handle truncation of the mirrored packet. */
4225 if (ctx
->mirror_snaplen
> 0 &&
4226 ctx
->mirror_snaplen
< UINT16_MAX
) {
4227 struct ovs_action_trunc
*trunc
;
4229 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4230 OVS_ACTION_ATTR_TRUNC
,
4232 trunc
->max_len
= ctx
->mirror_snaplen
;
4233 if (!ctx
->xbridge
->support
.trunc
) {
4234 ctx
->xout
->slow
|= SLOW_ACTION
;
4238 nl_msg_put_odp_port(ctx
->odp_actions
,
4239 OVS_ACTION_ATTR_OUTPUT
,
4243 ctx
->sflow_odp_port
= odp_port
;
4244 ctx
->sflow_n_outputs
++;
4245 ctx
->nf_output_iface
= ofp_port
;
4248 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
) && xport
->xbundle
) {
4249 mirror_packet(ctx
, xport
->xbundle
,
4250 xbundle_mirror_dst(xport
->xbundle
->xbridge
,
4256 memcpy(flow
->vlans
, flow_vlans
, sizeof flow
->vlans
);
4257 flow
->nw_tos
= flow_nw_tos
;
4258 flow
->dl_dst
= flow_dl_dst
;
4259 flow
->dl_src
= flow_dl_src
;
4260 flow
->packet_type
= flow_packet_type
;
4261 flow
->dl_type
= flow_dl_type
;
4265 compose_output_action(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4266 const struct xlate_bond_recirc
*xr
,
4267 bool is_last_action
, bool truncate
)
4269 compose_output_action__(ctx
, ofp_port
, xr
, true,
4270 is_last_action
, truncate
);
4274 xlate_recursively(struct xlate_ctx
*ctx
, struct rule_dpif
*rule
,
4275 bool deepens
, bool is_last_action
,
4276 xlate_actions_handler
*actions_xlator
)
4278 struct rule_dpif
*old_rule
= ctx
->rule
;
4279 ovs_be64 old_cookie
= ctx
->rule_cookie
;
4280 const struct rule_actions
*actions
;
4282 if (ctx
->xin
->resubmit_stats
) {
4283 rule_dpif_credit_stats(rule
, ctx
->xin
->resubmit_stats
, false);
4288 ctx
->depth
+= deepens
;
4290 ctx
->rule_cookie
= rule
->up
.flow_cookie
;
4291 actions
= rule_get_actions(&rule
->up
);
4292 actions_xlator(actions
->ofpacts
, actions
->ofpacts_len
, ctx
,
4293 is_last_action
, false);
4294 ctx
->rule_cookie
= old_cookie
;
4295 ctx
->rule
= old_rule
;
4296 ctx
->depth
-= deepens
;
4300 xlate_resubmit_resource_check(struct xlate_ctx
*ctx
)
4302 if (ctx
->depth
>= MAX_DEPTH
) {
4303 xlate_report_error(ctx
, "over max translation depth %d", MAX_DEPTH
);
4304 ctx
->error
= XLATE_RECURSION_TOO_DEEP
;
4305 } else if (ctx
->resubmits
>= MAX_RESUBMITS
) {
4306 xlate_report_error(ctx
, "over %d resubmit actions", MAX_RESUBMITS
);
4307 ctx
->error
= XLATE_TOO_MANY_RESUBMITS
;
4308 } else if (ctx
->odp_actions
->size
> UINT16_MAX
) {
4309 xlate_report_error(ctx
, "resubmits yielded over 64 kB of actions");
4310 /* NOT an error, as we'll be slow-pathing the flow in this case? */
4311 ctx
->exit
= true; /* XXX: translation still terminated! */
4312 } else if (ctx
->stack
.size
>= 65536) {
4313 xlate_report_error(ctx
, "resubmits yielded over 64 kB of stack");
4314 ctx
->error
= XLATE_STACK_TOO_DEEP
;
4323 tuple_swap_flow(struct flow
*flow
, bool ipv4
)
4325 uint8_t nw_proto
= flow
->nw_proto
;
4326 flow
->nw_proto
= flow
->ct_nw_proto
;
4327 flow
->ct_nw_proto
= nw_proto
;
4330 ovs_be32 nw_src
= flow
->nw_src
;
4331 flow
->nw_src
= flow
->ct_nw_src
;
4332 flow
->ct_nw_src
= nw_src
;
4334 ovs_be32 nw_dst
= flow
->nw_dst
;
4335 flow
->nw_dst
= flow
->ct_nw_dst
;
4336 flow
->ct_nw_dst
= nw_dst
;
4338 struct in6_addr ipv6_src
= flow
->ipv6_src
;
4339 flow
->ipv6_src
= flow
->ct_ipv6_src
;
4340 flow
->ct_ipv6_src
= ipv6_src
;
4342 struct in6_addr ipv6_dst
= flow
->ipv6_dst
;
4343 flow
->ipv6_dst
= flow
->ct_ipv6_dst
;
4344 flow
->ct_ipv6_dst
= ipv6_dst
;
4347 ovs_be16 tp_src
= flow
->tp_src
;
4348 flow
->tp_src
= flow
->ct_tp_src
;
4349 flow
->ct_tp_src
= tp_src
;
4351 ovs_be16 tp_dst
= flow
->tp_dst
;
4352 flow
->tp_dst
= flow
->ct_tp_dst
;
4353 flow
->ct_tp_dst
= tp_dst
;
4357 tuple_swap(struct flow
*flow
, struct flow_wildcards
*wc
)
4359 bool ipv4
= (flow
->dl_type
== htons(ETH_TYPE_IP
));
4361 tuple_swap_flow(flow
, ipv4
);
4362 tuple_swap_flow(&wc
->masks
, ipv4
);
4366 xlate_table_action(struct xlate_ctx
*ctx
, ofp_port_t in_port
, uint8_t table_id
,
4367 bool may_packet_in
, bool honor_table_miss
,
4368 bool with_ct_orig
, bool is_last_action
,
4369 xlate_actions_handler
*xlator
)
4371 /* Check if we need to recirculate before matching in a table. */
4372 if (ctx
->was_mpls
) {
4373 ctx_trigger_freeze(ctx
);
4376 if (xlate_resubmit_resource_check(ctx
)) {
4377 uint8_t old_table_id
= ctx
->table_id
;
4378 struct rule_dpif
*rule
;
4380 ctx
->table_id
= table_id
;
4382 /* Swap packet fields with CT 5-tuple if requested. */
4384 /* Do not swap if there is no CT tuple, or if key is not IP. */
4385 if (ctx
->xin
->flow
.ct_nw_proto
== 0 ||
4386 !is_ip_any(&ctx
->xin
->flow
)) {
4387 xlate_report_error(ctx
,
4388 "resubmit(ct) with non-tracked or non-IP packet!");
4389 ctx
->table_id
= old_table_id
;
4392 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4394 rule
= rule_dpif_lookup_from_table(ctx
->xbridge
->ofproto
,
4395 ctx
->xin
->tables_version
,
4396 &ctx
->xin
->flow
, ctx
->wc
,
4397 ctx
->xin
->resubmit_stats
,
4398 &ctx
->table_id
, in_port
,
4399 may_packet_in
, honor_table_miss
,
4403 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4407 /* Fill in the cache entry here instead of xlate_recursively
4408 * to make the reference counting more explicit. We take a
4409 * reference in the lookups above if we are going to cache the
4411 if (ctx
->xin
->xcache
) {
4412 struct xc_entry
*entry
;
4414 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_RULE
);
4416 ofproto_rule_ref(&rule
->up
);
4419 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4420 xlate_report_table(ctx
, rule
, table_id
);
4421 xlate_recursively(ctx
, rule
, table_id
<= old_table_id
,
4422 is_last_action
, xlator
);
4423 ctx
->xin
->trace
= old_trace
;
4426 ctx
->table_id
= old_table_id
;
4431 /* Consumes the group reference, which is only taken if xcache exists. */
4433 xlate_group_stats(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4434 struct ofputil_bucket
*bucket
)
4436 if (ctx
->xin
->resubmit_stats
) {
4437 group_dpif_credit_stats(group
, bucket
, ctx
->xin
->resubmit_stats
);
4439 if (ctx
->xin
->xcache
) {
4440 struct xc_entry
*entry
;
4442 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_GROUP
);
4443 entry
->group
.group
= group
;
4444 entry
->group
.bucket
= bucket
;
4449 xlate_group_bucket(struct xlate_ctx
*ctx
, struct ofputil_bucket
*bucket
,
4450 bool is_last_action
)
4452 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4453 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
4454 char *s
= xasprintf("bucket %"PRIu32
, bucket
->bucket_id
);
4455 ctx
->xin
->trace
= &oftrace_report(ctx
->xin
->trace
, OFT_BUCKET
,
4460 uint64_t action_list_stub
[1024 / 8];
4461 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
4462 struct ofpbuf action_set
= ofpbuf_const_initializer(bucket
->ofpacts
,
4463 bucket
->ofpacts_len
);
4464 struct flow old_flow
= ctx
->xin
->flow
;
4465 bool old_was_mpls
= ctx
->was_mpls
;
4467 ofpacts_execute_action_set(&action_list
, &action_set
);
4469 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, is_last_action
,
4473 ofpbuf_uninit(&action_list
);
4475 /* Check if need to freeze. */
4476 if (ctx
->freezing
) {
4477 finish_freezing(ctx
);
4480 /* Roll back flow to previous state.
4481 * This is equivalent to cloning the packet for each bucket.
4483 * As a side effect any subsequently applied actions will
4484 * also effectively be applied to a clone of the packet taken
4485 * just before applying the all or indirect group.
4487 * Note that group buckets are action sets, hence they cannot modify the
4488 * main action set. Also any stack actions are ignored when executing an
4489 * action set, so group buckets cannot directly change the stack either.
4490 * However, we do allow resubmit actions in group buckets, which could
4491 * recursively execute actions that do modify the action set or change the
4492 * stack. The controller must be careful about what it does to the
4493 * action_set and stack in the tables resubmitted to from group buckets. */
4494 ctx
->xin
->flow
= old_flow
;
4496 /* The group bucket popping MPLS should have no effect after bucket
4498 ctx
->was_mpls
= old_was_mpls
;
4500 /* The fact that the group bucket exits (for any reason) does not mean that
4501 * the translation after the group action should exit. Specifically, if
4502 * the group bucket freezes translation, the actions after the group action
4503 * must continue processing with the original, not the frozen packet! */
4506 /* Context error in a bucket should not impact processing of other buckets
4507 * or actions. This is similar to cloning a packet for group buckets.
4508 * There is no need to restore the error back to old value due to the fact
4509 * that we actually processed group action which can happen only when there
4510 * is no previous context error.
4512 * Exception to above is errors which are system limits to protect
4513 * translation from running too long or occupy too much space. These errors
4514 * should not be masked. XLATE_RECURSION_TOO_DEEP, XLATE_TOO_MANY_RESUBMITS
4515 * and XLATE_STACK_TOO_DEEP fall in this category. */
4516 if (ctx
->error
== XLATE_TOO_MANY_MPLS_LABELS
||
4517 ctx
->error
== XLATE_UNSUPPORTED_PACKET_TYPE
) {
4518 /* reset the error and continue processing other buckets */
4519 ctx
->error
= XLATE_OK
;
4522 ctx
->xin
->trace
= old_trace
;
4525 static struct ofputil_bucket
*
4526 pick_ff_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4528 return group_first_live_bucket(ctx
, group
, 0);
4531 static struct ofputil_bucket
*
4532 pick_default_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4534 flow_mask_hash_fields(&ctx
->xin
->flow
, ctx
->wc
,
4535 NX_HASH_FIELDS_SYMMETRIC_L4
);
4536 return group_best_live_bucket(ctx
, group
,
4537 flow_hash_symmetric_l4(&ctx
->xin
->flow
, 0));
4540 static struct ofputil_bucket
*
4541 pick_hash_fields_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4543 const struct field_array
*fields
= &group
->up
.props
.fields
;
4544 const uint8_t *mask_values
= fields
->values
;
4545 uint32_t basis
= hash_uint64(group
->up
.props
.selection_method_param
);
4548 BITMAP_FOR_EACH_1 (i
, MFF_N_IDS
, fields
->used
.bm
) {
4549 const struct mf_field
*mf
= mf_from_id(i
);
4551 /* Skip fields for which prerequisites are not met. */
4552 if (!mf_are_prereqs_ok(mf
, &ctx
->xin
->flow
, ctx
->wc
)) {
4553 /* Skip the mask bytes for this field. */
4554 mask_values
+= mf
->n_bytes
;
4558 union mf_value value
;
4559 union mf_value mask
;
4561 mf_get_value(mf
, &ctx
->xin
->flow
, &value
);
4562 /* Mask the value. */
4563 for (int j
= 0; j
< mf
->n_bytes
; j
++) {
4564 mask
.b
[j
] = *mask_values
++;
4565 value
.b
[j
] &= mask
.b
[j
];
4567 basis
= hash_bytes(&value
, mf
->n_bytes
, basis
);
4569 /* For tunnels, hash in whether the field is present. */
4570 if (mf_is_tun_metadata(mf
)) {
4571 basis
= hash_boolean(mf_is_set(mf
, &ctx
->xin
->flow
), basis
);
4574 mf_mask_field_masked(mf
, &mask
, ctx
->wc
);
4577 return group_best_live_bucket(ctx
, group
, basis
);
4580 static struct ofputil_bucket
*
4581 pick_dp_hash_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4583 uint32_t dp_hash
= ctx
->xin
->flow
.dp_hash
;
4585 /* dp_hash value 0 is special since it means that the dp_hash has not been
4586 * computed, as all computed dp_hash values are non-zero. Therefore
4587 * compare to zero can be used to decide if the dp_hash value is valid
4588 * without masking the dp_hash field. */
4590 enum ovs_hash_alg hash_alg
= group
->hash_alg
;
4591 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4592 /* Algorithm supported by all datapaths. */
4593 hash_alg
= OVS_HASH_ALG_L4
;
4595 ctx_trigger_recirculate_with_hash(ctx
, hash_alg
, group
->hash_basis
);
4598 uint32_t hash_mask
= group
->hash_mask
;
4599 ctx
->wc
->masks
.dp_hash
|= hash_mask
;
4601 /* Starting from the original masked dp_hash value iterate over the
4602 * hash mapping table to find the first live bucket. As the buckets
4603 * are quasi-randomly spread over the hash values, this maintains
4604 * a distribution according to bucket weights even when some buckets
4606 for (int i
= 0; i
<= hash_mask
; i
++) {
4607 struct ofputil_bucket
*b
=
4608 group
->hash_map
[(dp_hash
+ i
) & hash_mask
];
4609 if (bucket_is_alive(ctx
, b
, 0)) {
4618 static struct ofputil_bucket
*
4619 pick_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4621 /* Select groups may access flow keys beyond L2 in order to
4622 * select a bucket. Recirculate as appropriate to make this possible.
4624 if (ctx
->was_mpls
) {
4625 ctx_trigger_freeze(ctx
);
4629 switch (group
->selection_method
) {
4630 case SEL_METHOD_DEFAULT
:
4631 return pick_default_select_group(ctx
, group
);
4633 case SEL_METHOD_HASH
:
4634 return pick_hash_fields_select_group(ctx
, group
);
4636 case SEL_METHOD_DP_HASH
:
4637 return pick_dp_hash_select_group(ctx
, group
);
4640 /* Parsing of groups ensures this never happens */
4648 xlate_group_action__(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4649 bool is_last_action
)
4651 if (group
->up
.type
== OFPGT11_ALL
|| group
->up
.type
== OFPGT11_INDIRECT
) {
4652 struct ovs_list
*last_bucket
= group
->up
.buckets
.prev
;
4653 struct ofputil_bucket
*bucket
;
4654 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
4655 bool is_last_bucket
= &bucket
->list_node
== last_bucket
;
4656 xlate_group_bucket(ctx
, bucket
, is_last_action
&& is_last_bucket
);
4658 xlate_group_stats(ctx
, group
, NULL
);
4660 struct ofputil_bucket
*bucket
;
4661 if (group
->up
.type
== OFPGT11_SELECT
) {
4662 bucket
= pick_select_group(ctx
, group
);
4663 } else if (group
->up
.type
== OFPGT11_FF
) {
4664 bucket
= pick_ff_group(ctx
, group
);
4670 xlate_report(ctx
, OFT_DETAIL
, "using bucket %"PRIu32
,
4672 xlate_group_bucket(ctx
, bucket
, is_last_action
);
4673 xlate_group_stats(ctx
, group
, bucket
);
4675 xlate_report(ctx
, OFT_DETAIL
, "no live bucket");
4676 if (ctx
->xin
->xcache
) {
4677 ofproto_group_unref(&group
->up
);
4684 xlate_group_action(struct xlate_ctx
*ctx
, uint32_t group_id
,
4685 bool is_last_action
)
4687 if (xlate_resubmit_resource_check(ctx
)) {
4688 struct group_dpif
*group
;
4690 /* Take ref only if xcache exists. */
4691 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
4692 ctx
->xin
->tables_version
, ctx
->xin
->xcache
);
4694 /* XXX: Should set ctx->error ? */
4695 xlate_report(ctx
, OFT_WARN
, "output to nonexistent group %"PRIu32
,
4699 xlate_group_action__(ctx
, group
, is_last_action
);
4706 xlate_ofpact_resubmit(struct xlate_ctx
*ctx
,
4707 const struct ofpact_resubmit
*resubmit
,
4708 bool is_last_action
)
4712 bool may_packet_in
= false;
4713 bool honor_table_miss
= false;
4715 if (ctx
->rule
&& rule_dpif_is_internal(ctx
->rule
)) {
4716 /* Still allow missed packets to be sent to the controller
4717 * if resubmitting from an internal table. */
4718 may_packet_in
= true;
4719 honor_table_miss
= true;
4722 in_port
= resubmit
->in_port
;
4723 if (in_port
== OFPP_IN_PORT
) {
4724 in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
4727 table_id
= resubmit
->table_id
;
4728 if (table_id
== 255) {
4729 table_id
= ctx
->table_id
;
4732 xlate_table_action(ctx
, in_port
, table_id
, may_packet_in
,
4733 honor_table_miss
, resubmit
->with_ct_orig
,
4734 is_last_action
, do_xlate_actions
);
4738 flood_packet_to_port(struct xlate_ctx
*ctx
, const struct xport
*xport
,
4739 bool all
, bool is_last_action
)
4746 compose_output_action__(ctx
, xport
->ofp_port
, NULL
, false,
4747 is_last_action
, false);
4749 compose_output_action(ctx
, xport
->ofp_port
, NULL
, is_last_action
,
4755 flood_packets(struct xlate_ctx
*ctx
, bool all
, bool is_last_action
)
4757 const struct xport
*xport
, *last
= NULL
;
4759 /* Use 'last' the keep track of the last output port. */
4760 HMAP_FOR_EACH (xport
, ofp_node
, &ctx
->xbridge
->xports
) {
4761 if (xport
->ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
4765 if (all
|| !(xport
->config
& OFPUTIL_PC_NO_FLOOD
)) {
4766 /* 'last' is not the last port, send a packet out, and
4768 flood_packet_to_port(ctx
, last
, all
, false);
4773 /* Send the packet to the 'last' port. */
4774 flood_packet_to_port(ctx
, last
, all
, is_last_action
);
4775 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
4779 put_controller_user_action(struct xlate_ctx
*ctx
,
4780 bool dont_send
, bool continuation
,
4781 uint32_t recirc_id
, int len
,
4782 enum ofp_packet_in_reason reason
,
4783 uint16_t controller_id
)
4785 struct user_action_cookie cookie
;
4787 memset(&cookie
, 0, sizeof cookie
);
4788 cookie
.type
= USER_ACTION_COOKIE_CONTROLLER
;
4789 cookie
.ofp_in_port
= OFPP_NONE
,
4790 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
4791 cookie
.controller
.dont_send
= dont_send
;
4792 cookie
.controller
.continuation
= continuation
;
4793 cookie
.controller
.reason
= reason
;
4794 cookie
.controller
.recirc_id
= recirc_id
;
4795 put_32aligned_be64(&cookie
.controller
.rule_cookie
, ctx
->rule_cookie
);
4796 cookie
.controller
.controller_id
= controller_id
;
4797 cookie
.controller
.max_len
= len
;
4799 odp_port_t odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
4800 ctx
->xin
->flow
.in_port
.ofp_port
);
4801 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
4802 odp_put_userspace_action(pid
, &cookie
, sizeof cookie
, ODPP_NONE
,
4803 false, ctx
->odp_actions
);
4807 xlate_controller_action(struct xlate_ctx
*ctx
, int len
,
4808 enum ofp_packet_in_reason reason
,
4809 uint16_t controller_id
,
4810 uint32_t provider_meter_id
,
4811 const uint8_t *userdata
, size_t userdata_len
)
4813 xlate_commit_actions(ctx
);
4815 /* A packet sent by an action in a table-miss rule is considered an
4816 * explicit table miss. OpenFlow before 1.3 doesn't have that concept so
4817 * it will get translated back to OFPR_ACTION for those versions. */
4818 if (reason
== OFPR_ACTION
4819 && ctx
->rule
&& rule_is_table_miss(&ctx
->rule
->up
)) {
4820 reason
= OFPR_EXPLICIT_MISS
;
4823 struct frozen_state state
= {
4824 .table_id
= ctx
->table_id
,
4825 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4826 .stack
= ctx
->stack
.data
,
4827 .stack_size
= ctx
->stack
.size
,
4828 .mirrors
= ctx
->mirrors
,
4829 .conntracked
= ctx
->conntracked
,
4830 .was_mpls
= ctx
->was_mpls
,
4834 .action_set_len
= 0,
4835 .userdata
= CONST_CAST(uint8_t *, userdata
),
4836 .userdata_len
= userdata_len
,
4838 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4840 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4842 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4843 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4846 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4848 /* If the controller action didn't request a meter (indicated by a
4849 * 'meter_id' argument other than NX_CTLR_NO_METER), see if one was
4850 * configured through the "controller" virtual meter.
4852 * Internally, ovs-vswitchd uses UINT32_MAX to indicate no meter is
4855 if (provider_meter_id
== UINT32_MAX
) {
4856 meter_id
= ctx
->xbridge
->ofproto
->up
.controller_meter_id
;
4858 meter_id
= provider_meter_id
;
4863 if (meter_id
!= UINT32_MAX
) {
4864 /* If controller meter is configured, generate clone(meter, userspace)
4866 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
4867 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
4869 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
4870 OVS_SAMPLE_ATTR_ACTIONS
);
4871 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
4874 /* Generate the datapath flows even if we don't send the packet-in
4875 * so that debugging more closely represents normal state. */
4876 bool dont_send
= false;
4877 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4880 put_controller_user_action(ctx
, dont_send
, false, recirc_id
, len
,
4881 reason
, controller_id
);
4883 if (meter_id
!= UINT32_MAX
) {
4884 nl_msg_end_nested(ctx
->odp_actions
, ac_offset
);
4885 nl_msg_end_nested(ctx
->odp_actions
, offset
);
4889 /* Creates a frozen state, and allocates a unique recirc id for the given
4890 * state. Returns a non-zero recirc id if it is allocated successfully.
4891 * Returns 0 otherwise.
4894 finish_freezing__(struct xlate_ctx
*ctx
, uint8_t table
)
4896 ovs_assert(ctx
->freezing
);
4898 struct frozen_state state
= {
4900 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4901 .stack
= ctx
->stack
.data
,
4902 .stack_size
= ctx
->stack
.size
,
4903 .mirrors
= ctx
->mirrors
,
4904 .conntracked
= ctx
->conntracked
,
4905 .was_mpls
= ctx
->was_mpls
,
4906 .xport_uuid
= ctx
->xin
->xport_uuid
,
4907 .ofpacts
= ctx
->frozen_actions
.data
,
4908 .ofpacts_len
= ctx
->frozen_actions
.size
,
4909 .action_set
= ctx
->action_set
.data
,
4910 .action_set_len
= ctx
->action_set
.size
,
4911 .userdata
= ctx
->pause
? CONST_CAST(uint8_t *,ctx
->pause
->userdata
)
4913 .userdata_len
= ctx
->pause
? ctx
->pause
->userdata_len
: 0,
4915 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4917 /* Allocate a unique recirc id for the given metadata state in the
4918 * flow. An existing id, with a new reference to the corresponding
4919 * recirculation context, will be returned if possible.
4920 * The life-cycle of this recirc id is managed by associating it
4921 * with the udpif key ('ukey') created for each new datapath flow. */
4922 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4924 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4925 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4928 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4931 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4935 put_controller_user_action(ctx
, false, true, recirc_id
,
4936 ctx
->pause
->max_len
,
4938 ctx
->pause
->controller_id
);
4940 if (ctx
->recirc_update_dp_hash
) {
4941 struct ovs_action_hash
*act_hash
;
4944 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4945 OVS_ACTION_ATTR_HASH
,
4947 act_hash
->hash_alg
= ctx
->dp_hash_alg
;
4948 act_hash
->hash_basis
= ctx
->dp_hash_basis
;
4950 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, recirc_id
);
4953 /* Undo changes done by freezing. */
4954 ctx_cancel_freeze(ctx
);
4958 /* Called only when we're freezing. */
4960 finish_freezing(struct xlate_ctx
*ctx
)
4962 xlate_commit_actions(ctx
);
4963 finish_freezing__(ctx
, 0);
4966 /* Fork the pipeline here. The current packet will continue processing the
4967 * current action list. A clone of the current packet will recirculate, skip
4968 * the remainder of the current action list and asynchronously resume pipeline
4969 * processing in 'table' with the current metadata and action set. */
4971 compose_recirculate_and_fork(struct xlate_ctx
*ctx
, uint8_t table
,
4972 const uint16_t zone
)
4975 ctx
->freezing
= true;
4976 recirc_id
= finish_freezing__(ctx
, table
);
4978 if (OVS_UNLIKELY(ctx
->xin
->trace
) && recirc_id
) {
4979 if (oftrace_add_recirc_node(ctx
->xin
->recirc_queue
,
4980 OFT_RECIRC_CONNTRACK
, &ctx
->xin
->flow
,
4981 ctx
->xin
->packet
, recirc_id
, zone
)) {
4982 xlate_report(ctx
, OFT_DETAIL
, "A clone of the packet is forked to "
4983 "recirculate. The forked pipeline will be resumed at "
4984 "table %u.", table
);
4986 xlate_report(ctx
, OFT_DETAIL
, "Failed to trace the conntrack "
4987 "forked pipeline with recirc_id = %d.", recirc_id
);
4993 compose_mpls_push_action(struct xlate_ctx
*ctx
, struct ofpact_push_mpls
*mpls
)
4995 struct flow
*flow
= &ctx
->xin
->flow
;
4998 ovs_assert(eth_type_mpls(mpls
->ethertype
));
5000 n
= flow_count_mpls_labels(flow
, ctx
->wc
);
5002 xlate_commit_actions(ctx
);
5003 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
5004 if (ctx
->xin
->packet
!= NULL
) {
5005 xlate_report_error(ctx
, "dropping packet on which an MPLS push "
5006 "action can't be performed as it would have "
5007 "more MPLS LSEs than the %d supported.",
5008 FLOW_MAX_MPLS_LABELS
);
5010 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
5014 /* Update flow's MPLS stack, and clear L3/4 fields to mark them invalid. */
5015 flow_push_mpls(flow
, n
, mpls
->ethertype
, ctx
->wc
, true);
5019 compose_mpls_pop_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
5021 struct flow
*flow
= &ctx
->xin
->flow
;
5022 int n
= flow_count_mpls_labels(flow
, ctx
->wc
);
5024 if (flow_pop_mpls(flow
, n
, eth_type
, ctx
->wc
)) {
5025 if (!eth_type_mpls(eth_type
) && ctx
->xbridge
->support
.odp
.recirc
) {
5026 ctx
->was_mpls
= true;
5028 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
5029 if (ctx
->xin
->packet
!= NULL
) {
5030 xlate_report_error(ctx
, "dropping packet on which an "
5031 "MPLS pop action can't be performed as it has "
5032 "more MPLS LSEs than the %d supported.",
5033 FLOW_MAX_MPLS_LABELS
);
5035 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
5036 ofpbuf_clear(ctx
->odp_actions
);
5041 compose_dec_ttl(struct xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
5043 struct flow
*flow
= &ctx
->xin
->flow
;
5045 if (!is_ip_any(flow
)) {
5049 ctx
->wc
->masks
.nw_ttl
= 0xff;
5050 if (flow
->nw_ttl
> 1) {
5056 for (i
= 0; i
< ids
->n_controllers
; i
++) {
5057 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5058 ids
->cnt_ids
[i
], UINT32_MAX
, NULL
, 0);
5061 /* Stop processing for current table. */
5062 xlate_report(ctx
, OFT_WARN
, "IPv%d decrement TTL exception",
5063 flow
->dl_type
== htons(ETH_TYPE_IP
) ? 4 : 6);
5069 compose_set_mpls_label_action(struct xlate_ctx
*ctx
, ovs_be32 label
)
5071 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5072 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_LABEL_MASK
);
5073 set_mpls_lse_label(&ctx
->xin
->flow
.mpls_lse
[0], label
);
5078 compose_set_mpls_tc_action(struct xlate_ctx
*ctx
, uint8_t tc
)
5080 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5081 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TC_MASK
);
5082 set_mpls_lse_tc(&ctx
->xin
->flow
.mpls_lse
[0], tc
);
5087 compose_dec_nsh_ttl_action(struct xlate_ctx
*ctx
)
5089 struct flow
*flow
= &ctx
->xin
->flow
;
5091 if ((flow
->packet_type
== htonl(PT_NSH
)) ||
5092 (flow
->dl_type
== htons(ETH_TYPE_NSH
))) {
5093 ctx
->wc
->masks
.nsh
.ttl
= 0xff;
5094 if (flow
->nsh
.ttl
> 1) {
5098 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5099 0, UINT32_MAX
, NULL
, 0);
5103 /* Stop processing for current table. */
5104 xlate_report(ctx
, OFT_WARN
, "NSH decrement TTL exception");
5109 compose_set_mpls_ttl_action(struct xlate_ctx
*ctx
, uint8_t ttl
)
5111 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5112 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5113 set_mpls_lse_ttl(&ctx
->xin
->flow
.mpls_lse
[0], ttl
);
5118 compose_dec_mpls_ttl_action(struct xlate_ctx
*ctx
)
5120 struct flow
*flow
= &ctx
->xin
->flow
;
5122 if (eth_type_mpls(flow
->dl_type
)) {
5123 uint8_t ttl
= mpls_lse_to_ttl(flow
->mpls_lse
[0]);
5125 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5128 set_mpls_lse_ttl(&flow
->mpls_lse
[0], ttl
);
5131 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0,
5132 UINT32_MAX
, NULL
, 0);
5136 /* Stop processing for current table. */
5137 xlate_report(ctx
, OFT_WARN
, "MPLS decrement TTL exception");
5141 /* Emits an action that outputs to 'port', within 'ctx'.
5143 * 'controller_len' affects only packets sent to an OpenFlow controller. It
5144 * is the maximum number of bytes of the packet to send. UINT16_MAX means to
5145 * send the whole packet (and 0 means to omit the packet entirely).
5147 * 'may_packet_in' determines whether the packet may be sent to an OpenFlow
5148 * controller. If it is false, then the packet is never sent to the OpenFlow
5151 * 'is_last_action' should be true if this output is the last OpenFlow action
5152 * to be processed, which enables certain optimizations.
5154 * 'truncate' should be true if the packet to be output is being truncated,
5155 * which suppresses certain optimizations. */
5157 xlate_output_action(struct xlate_ctx
*ctx
, ofp_port_t port
,
5158 uint16_t controller_len
, bool may_packet_in
,
5159 bool is_last_action
, bool truncate
,
5160 bool group_bucket_action
)
5162 ofp_port_t prev_nf_output_iface
= ctx
->nf_output_iface
;
5164 ctx
->nf_output_iface
= NF_OUT_DROP
;
5168 compose_output_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
,
5169 is_last_action
, truncate
);
5172 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
5173 0, may_packet_in
, true, false, false,
5180 flood_packets(ctx
, false, is_last_action
);
5183 flood_packets(ctx
, true, is_last_action
);
5185 case OFPP_CONTROLLER
:
5186 xlate_controller_action(ctx
, controller_len
,
5187 (ctx
->in_packet_out
? OFPR_PACKET_OUT
5188 : group_bucket_action
? OFPR_GROUP
5189 : ctx
->in_action_set
? OFPR_ACTION_SET
5191 0, UINT32_MAX
, NULL
, 0);
5197 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5198 compose_output_action(ctx
, port
, NULL
, is_last_action
, truncate
);
5200 xlate_report_info(ctx
, "skipping output to input port");
5205 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
5206 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
5207 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5208 ctx
->nf_output_iface
= prev_nf_output_iface
;
5209 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
5210 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5211 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5216 xlate_output_reg_action(struct xlate_ctx
*ctx
,
5217 const struct ofpact_output_reg
*or,
5218 bool is_last_action
,
5219 bool group_bucket_action
)
5221 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->xin
->flow
);
5222 if (port
<= UINT16_MAX
) {
5223 xlate_report(ctx
, OFT_DETAIL
, "output port is %"PRIu64
, port
);
5225 union mf_subvalue value
;
5227 memset(&value
, 0xff, sizeof value
);
5228 mf_write_subfield_flow(&or->src
, &value
, &ctx
->wc
->masks
);
5229 xlate_output_action(ctx
, u16_to_ofp(port
), or->max_len
,
5230 false, is_last_action
, false,
5231 group_bucket_action
);
5233 xlate_report(ctx
, OFT_WARN
, "output port %"PRIu64
" is out of range",
5239 xlate_output_trunc_action(struct xlate_ctx
*ctx
,
5240 ofp_port_t port
, uint32_t max_len
,
5241 bool is_last_action
,
5242 bool group_bucket_action
)
5244 bool support_trunc
= ctx
->xbridge
->support
.trunc
;
5245 struct ovs_action_trunc
*trunc
;
5246 char name
[OFP_MAX_PORT_NAME_LEN
];
5253 case OFPP_CONTROLLER
:
5255 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5256 xlate_report(ctx
, OFT_WARN
,
5257 "output_trunc does not support port: %s", name
);
5262 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5263 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, port
);
5265 if (xport
== NULL
|| xport
->odp_port
== ODPP_NONE
) {
5266 /* Since truncate happens at its following output action, if
5267 * the output port is a patch port, the behavior is somehow
5268 * unpredictable. For simplicity, disallow this case. */
5269 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5270 xlate_report_error(ctx
, "output_trunc does not support "
5271 "patch port %s", name
);
5275 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
5276 OVS_ACTION_ATTR_TRUNC
,
5278 trunc
->max_len
= max_len
;
5279 xlate_output_action(ctx
, port
, 0, false, is_last_action
, true,
5280 group_bucket_action
);
5281 if (!support_trunc
) {
5282 ctx
->xout
->slow
|= SLOW_ACTION
;
5285 xlate_report_info(ctx
, "skipping output to input port");
5292 xlate_enqueue_action(struct xlate_ctx
*ctx
,
5293 const struct ofpact_enqueue
*enqueue
,
5294 bool is_last_action
,
5295 bool group_bucket_action
)
5297 ofp_port_t ofp_port
= enqueue
->port
;
5298 uint32_t queue_id
= enqueue
->queue
;
5299 uint32_t flow_priority
, priority
;
5302 /* Translate queue to priority. */
5303 error
= dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &priority
);
5305 /* Fall back to ordinary output action. */
5306 xlate_output_action(ctx
, enqueue
->port
, 0, false,
5307 is_last_action
, false,
5308 group_bucket_action
);
5312 /* Check output port. */
5313 if (ofp_port
== OFPP_IN_PORT
) {
5314 ofp_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
5315 } else if (ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
5319 /* Add datapath actions. */
5320 flow_priority
= ctx
->xin
->flow
.skb_priority
;
5321 ctx
->xin
->flow
.skb_priority
= priority
;
5322 compose_output_action(ctx
, ofp_port
, NULL
, is_last_action
, false);
5323 ctx
->xin
->flow
.skb_priority
= flow_priority
;
5325 /* Update NetFlow output port. */
5326 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5327 ctx
->nf_output_iface
= ofp_port
;
5328 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5329 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5334 xlate_set_queue_action(struct xlate_ctx
*ctx
, uint32_t queue_id
)
5336 uint32_t skb_priority
;
5338 if (!dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &skb_priority
)) {
5339 ctx
->xin
->flow
.skb_priority
= skb_priority
;
5341 /* Couldn't translate queue to a priority. Nothing to do. A warning
5342 * has already been logged. */
5347 slave_enabled_cb(ofp_port_t ofp_port
, void *xbridge_
)
5349 const struct xbridge
*xbridge
= xbridge_
;
5360 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
5363 port
= get_ofp_port(xbridge
, ofp_port
);
5364 return port
? port
->may_enable
: false;
5369 xlate_bundle_action(struct xlate_ctx
*ctx
,
5370 const struct ofpact_bundle
*bundle
,
5371 bool is_last_action
,
5372 bool group_bucket_action
)
5376 port
= bundle_execute(bundle
, &ctx
->xin
->flow
, ctx
->wc
, slave_enabled_cb
,
5377 CONST_CAST(struct xbridge
*, ctx
->xbridge
));
5378 if (bundle
->dst
.field
) {
5379 nxm_reg_load(&bundle
->dst
, ofp_to_u16(port
), &ctx
->xin
->flow
, ctx
->wc
);
5380 xlate_report_subfield(ctx
, &bundle
->dst
);
5382 xlate_output_action(ctx
, port
, 0, false, is_last_action
, false,
5383 group_bucket_action
);
5388 xlate_learn_action(struct xlate_ctx
*ctx
, const struct ofpact_learn
*learn
)
5390 learn_mask(learn
, ctx
->wc
);
5392 if (ctx
->xin
->xcache
|| ctx
->xin
->allow_side_effects
) {
5393 uint64_t ofpacts_stub
[1024 / 8];
5394 struct ofputil_flow_mod fm
;
5395 struct ofproto_flow_mod ofm__
, *ofm
;
5396 struct ofpbuf ofpacts
;
5399 if (ctx
->xin
->xcache
) {
5400 ofm
= xmalloc(sizeof *ofm
);
5405 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
5406 learn_execute(learn
, &ctx
->xin
->flow
, &fm
, &ofpacts
);
5407 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
5408 struct ds s
= DS_EMPTY_INITIALIZER
;
5409 ds_put_format(&s
, "table=%"PRIu8
" ", fm
.table_id
);
5410 minimatch_format(&fm
.match
,
5411 ofproto_get_tun_tab(&ctx
->xin
->ofproto
->up
),
5412 NULL
, &s
, OFP_DEFAULT_PRIORITY
);
5414 ds_put_format(&s
, " priority=%d", fm
.priority
);
5415 if (fm
.new_cookie
) {
5416 ds_put_format(&s
, " cookie=%#"PRIx64
, ntohll(fm
.new_cookie
));
5418 if (fm
.idle_timeout
!= OFP_FLOW_PERMANENT
) {
5419 ds_put_format(&s
, " idle=%"PRIu16
, fm
.idle_timeout
);
5421 if (fm
.hard_timeout
!= OFP_FLOW_PERMANENT
) {
5422 ds_put_format(&s
, " hard=%"PRIu16
, fm
.hard_timeout
);
5424 if (fm
.flags
& NX_LEARN_F_SEND_FLOW_REM
) {
5425 ds_put_cstr(&s
, " send_flow_rem");
5427 ds_put_cstr(&s
, " actions=");
5428 struct ofpact_format_params fp
= { .s
= &s
};
5429 ofpacts_format(fm
.ofpacts
, fm
.ofpacts_len
, &fp
);
5430 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
5433 error
= ofproto_dpif_flow_mod_init_for_learn(ctx
->xbridge
->ofproto
,
5435 ofpbuf_uninit(&ofpacts
);
5438 bool success
= true;
5439 if (ctx
->xin
->allow_side_effects
) {
5440 error
= ofproto_flow_mod_learn(ofm
, ctx
->xin
->xcache
!= NULL
,
5441 learn
->limit
, &success
);
5442 } else if (learn
->limit
) {
5444 || ofm
->temp_rule
->state
!= RULE_INSERTED
) {
5445 /* The learned rule expired and there are no packets, so
5446 * we cannot learn again. Since the translated actions
5447 * depend on the result of learning, we tell the caller
5448 * that there's no point in caching this result. */
5449 ctx
->xout
->avoid_caching
= true;
5453 if (learn
->flags
& NX_LEARN_F_WRITE_RESULT
) {
5454 nxm_reg_load(&learn
->result_dst
, success
? 1 : 0,
5455 &ctx
->xin
->flow
, ctx
->wc
);
5456 xlate_report_subfield(ctx
, &learn
->result_dst
);
5459 if (success
&& ctx
->xin
->xcache
) {
5460 struct xc_entry
*entry
;
5462 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_LEARN
);
5463 entry
->learn
.ofm
= ofm
;
5464 entry
->learn
.limit
= learn
->limit
;
5467 ofproto_flow_mod_uninit(ofm
);
5470 if (OVS_UNLIKELY(ctx
->xin
->trace
&& !success
)) {
5471 xlate_report(ctx
, OFT_DETAIL
, "Limit exceeded, learn failed");
5475 if (ofm
!= &ofm__
) {
5480 xlate_report_error(ctx
, "LEARN action execution failed (%s).",
5481 ofperr_to_string(error
));
5484 minimatch_destroy(&fm
.match
);
5486 xlate_report(ctx
, OFT_WARN
,
5487 "suppressing side effects, so learn action ignored");
5492 xlate_fin_timeout__(struct rule_dpif
*rule
, uint16_t tcp_flags
,
5493 uint16_t idle_timeout
, uint16_t hard_timeout
)
5495 if (tcp_flags
& (TCP_FIN
| TCP_RST
)) {
5496 ofproto_rule_reduce_timeouts(&rule
->up
, idle_timeout
, hard_timeout
);
5501 xlate_fin_timeout(struct xlate_ctx
*ctx
,
5502 const struct ofpact_fin_timeout
*oft
)
5505 if (ctx
->xin
->allow_side_effects
) {
5506 xlate_fin_timeout__(ctx
->rule
, ctx
->xin
->tcp_flags
,
5507 oft
->fin_idle_timeout
, oft
->fin_hard_timeout
);
5509 if (ctx
->xin
->xcache
) {
5510 struct xc_entry
*entry
;
5512 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_FIN_TIMEOUT
);
5513 /* XC_RULE already holds a reference on the rule, none is taken
5515 entry
->fin
.rule
= ctx
->rule
;
5516 entry
->fin
.idle
= oft
->fin_idle_timeout
;
5517 entry
->fin
.hard
= oft
->fin_hard_timeout
;
5523 xlate_sample_action(struct xlate_ctx
*ctx
,
5524 const struct ofpact_sample
*os
)
5526 odp_port_t output_odp_port
= ODPP_NONE
;
5527 odp_port_t tunnel_out_port
= ODPP_NONE
;
5528 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
5529 bool emit_set_tunnel
= false;
5531 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
5535 /* Scale the probability from 16-bit to 32-bit while representing
5536 * the same percentage. */
5537 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
5539 /* If ofp_port in flow sample action is equel to ofp_port,
5540 * this sample action is a input port action. */
5541 if (os
->sampling_port
!= OFPP_NONE
&&
5542 os
->sampling_port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5543 output_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
5545 if (output_odp_port
== ODPP_NONE
) {
5546 xlate_report_error(ctx
, "can't use unknown port %d in flow sample "
5547 "action", os
->sampling_port
);
5551 if (dpif_ipfix_get_flow_exporter_tunnel_sampling(ipfix
,
5552 os
->collector_set_id
)
5553 && dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
)) {
5554 tunnel_out_port
= output_odp_port
;
5555 emit_set_tunnel
= true;
5559 xlate_commit_actions(ctx
);
5560 /* If 'emit_set_tunnel', sample(sampling_port=1) would translate
5561 * into datapath sample action set(tunnel(...)), sample(...) and
5562 * it is used for sampling egress tunnel information. */
5563 if (emit_set_tunnel
) {
5564 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
,
5567 if (xport
&& xport
->is_tunnel
) {
5568 struct flow
*flow
= &ctx
->xin
->flow
;
5569 tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
5570 if (!ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
5571 struct flow_tnl flow_tnl
= flow
->tunnel
;
5572 const char *tnl_type
;
5574 tnl_type
= tnl_port_get_type(xport
->ofport
);
5575 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
5576 ctx
->odp_actions
, tnl_type
);
5577 flow
->tunnel
= flow_tnl
;
5580 xlate_report_error(ctx
,
5581 "sampling_port:%d should be a tunnel port.",
5586 struct user_action_cookie cookie
;
5588 memset(&cookie
, 0, sizeof cookie
);
5589 cookie
.type
= USER_ACTION_COOKIE_FLOW_SAMPLE
;
5590 cookie
.ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
5591 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
5592 cookie
.flow_sample
.probability
= os
->probability
;
5593 cookie
.flow_sample
.collector_set_id
= os
->collector_set_id
;
5594 cookie
.flow_sample
.obs_domain_id
= os
->obs_domain_id
;
5595 cookie
.flow_sample
.obs_point_id
= os
->obs_point_id
;
5596 cookie
.flow_sample
.output_odp_port
= output_odp_port
;
5597 cookie
.flow_sample
.direction
= os
->direction
;
5599 compose_sample_action(ctx
, probability
, &cookie
, tunnel_out_port
, false);
5602 /* Determine if an datapath action translated from the openflow action
5603 * can be reversed by another datapath action.
5605 * Openflow actions that do not emit datapath actions are trivially
5606 * reversible. Reversiblity of other actions depends on nature of
5607 * action and their translation. */
5609 reversible_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5611 const struct ofpact
*a
;
5613 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5616 case OFPACT_CLEAR_ACTIONS
:
5618 case OFPACT_CONJUNCTION
:
5619 case OFPACT_CONTROLLER
:
5620 case OFPACT_CT_CLEAR
:
5621 case OFPACT_DEBUG_RECIRC
:
5622 case OFPACT_DEBUG_SLOW
:
5623 case OFPACT_DEC_MPLS_TTL
:
5624 case OFPACT_DEC_TTL
:
5625 case OFPACT_ENQUEUE
:
5627 case OFPACT_FIN_TIMEOUT
:
5628 case OFPACT_GOTO_TABLE
:
5631 case OFPACT_MULTIPATH
:
5634 case OFPACT_OUTPUT_REG
:
5635 case OFPACT_POP_MPLS
:
5636 case OFPACT_POP_QUEUE
:
5637 case OFPACT_PUSH_MPLS
:
5638 case OFPACT_PUSH_VLAN
:
5639 case OFPACT_REG_MOVE
:
5640 case OFPACT_RESUBMIT
:
5642 case OFPACT_SET_ETH_DST
:
5643 case OFPACT_SET_ETH_SRC
:
5644 case OFPACT_SET_FIELD
:
5645 case OFPACT_SET_IP_DSCP
:
5646 case OFPACT_SET_IP_ECN
:
5647 case OFPACT_SET_IP_TTL
:
5648 case OFPACT_SET_IPV4_DST
:
5649 case OFPACT_SET_IPV4_SRC
:
5650 case OFPACT_SET_L4_DST_PORT
:
5651 case OFPACT_SET_L4_SRC_PORT
:
5652 case OFPACT_SET_MPLS_LABEL
:
5653 case OFPACT_SET_MPLS_TC
:
5654 case OFPACT_SET_MPLS_TTL
:
5655 case OFPACT_SET_QUEUE
:
5656 case OFPACT_SET_TUNNEL
:
5657 case OFPACT_SET_VLAN_PCP
:
5658 case OFPACT_SET_VLAN_VID
:
5659 case OFPACT_STACK_POP
:
5660 case OFPACT_STACK_PUSH
:
5661 case OFPACT_STRIP_VLAN
:
5662 case OFPACT_UNROLL_XLATE
:
5663 case OFPACT_WRITE_ACTIONS
:
5664 case OFPACT_WRITE_METADATA
:
5665 case OFPACT_CHECK_PKT_LARGER
:
5671 case OFPACT_OUTPUT_TRUNC
:
5674 case OFPACT_DEC_NSH_TTL
:
5682 clone_xlate_actions(const struct ofpact
*actions
, size_t actions_len
,
5683 struct xlate_ctx
*ctx
, bool is_last_action
,
5684 bool group_bucket_action OVS_UNUSED
)
5686 struct ofpbuf old_stack
= ctx
->stack
;
5687 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
5688 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
5689 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
5691 struct ofpbuf old_action_set
= ctx
->action_set
;
5692 uint64_t actset_stub
[1024 / 8];
5693 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
5694 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
5696 size_t offset
, ac_offset
;
5697 struct flow old_flow
= ctx
->xin
->flow
;
5699 if (reversible_actions(actions
, actions_len
) || is_last_action
) {
5700 old_flow
= ctx
->xin
->flow
;
5701 do_xlate_actions(actions
, actions_len
, ctx
, is_last_action
, false);
5702 if (!ctx
->freezing
) {
5703 xlate_action_set(ctx
);
5705 if (ctx
->freezing
) {
5706 finish_freezing(ctx
);
5711 /* Commit datapath actions before emitting the clone action to
5712 * avoid emitting those actions twice. Once inside
5713 * the clone, another time for the action after clone. */
5714 xlate_commit_actions(ctx
);
5715 struct flow old_base
= ctx
->base_flow
;
5716 bool old_was_mpls
= ctx
->was_mpls
;
5717 bool old_conntracked
= ctx
->conntracked
;
5719 /* The actions are not reversible, a datapath clone action is
5720 * required to encode the translation. Select the clone action
5721 * based on datapath capabilities. */
5722 if (ctx
->xbridge
->support
.clone
) { /* Use clone action */
5723 /* Use clone action as datapath clone. */
5724 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
5725 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5726 if (!ctx
->freezing
) {
5727 xlate_action_set(ctx
);
5729 if (ctx
->freezing
) {
5730 finish_freezing(ctx
);
5732 nl_msg_end_non_empty_nested(ctx
->odp_actions
, offset
);
5736 if (ctx
->xbridge
->support
.sample_nesting
> 3) {
5737 /* Use sample action as datapath clone. */
5738 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
5739 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
5740 OVS_SAMPLE_ATTR_ACTIONS
);
5741 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5742 if (!ctx
->freezing
) {
5743 xlate_action_set(ctx
);
5745 if (ctx
->freezing
) {
5746 finish_freezing(ctx
);
5748 if (nl_msg_end_non_empty_nested(ctx
->odp_actions
, ac_offset
)) {
5749 nl_msg_cancel_nested(ctx
->odp_actions
, offset
);
5751 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
5752 UINT32_MAX
); /* 100% probability. */
5753 nl_msg_end_nested(ctx
->odp_actions
, offset
);
5758 /* Datapath does not support clone, skip xlate 'oc' and
5759 * report an error */
5760 xlate_report_error(ctx
, "Failed to compose clone action");
5763 /* The clone's conntrack execution should have no effect on the original
5765 ctx
->conntracked
= old_conntracked
;
5767 /* Popping MPLS from the clone should have no effect on the original
5769 ctx
->was_mpls
= old_was_mpls
;
5771 /* Restore the 'base_flow' for the next action. */
5772 ctx
->base_flow
= old_base
;
5775 ofpbuf_uninit(&ctx
->action_set
);
5776 ctx
->action_set
= old_action_set
;
5777 ofpbuf_uninit(&ctx
->stack
);
5778 ctx
->stack
= old_stack
;
5779 ctx
->xin
->flow
= old_flow
;
5783 compose_clone(struct xlate_ctx
*ctx
, const struct ofpact_nest
*oc
,
5784 bool is_last_action
)
5786 size_t oc_actions_len
= ofpact_nest_get_action_len(oc
);
5788 clone_xlate_actions(oc
->actions
, oc_actions_len
, ctx
, is_last_action
,
5793 xlate_meter_action(struct xlate_ctx
*ctx
, const struct ofpact_meter
*meter
)
5795 if (meter
->provider_meter_id
!= UINT32_MAX
) {
5796 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
,
5797 meter
->provider_meter_id
);
5802 may_receive(const struct xport
*xport
, struct xlate_ctx
*ctx
)
5804 if (xport
->config
& (is_stp(&ctx
->xin
->flow
)
5805 ? OFPUTIL_PC_NO_RECV_STP
5806 : OFPUTIL_PC_NO_RECV
)) {
5810 /* Only drop packets here if both forwarding and learning are
5811 * disabled. If just learning is enabled, we need to have
5812 * OFPP_NORMAL and the learning action have a look at the packet
5813 * before we can drop it. */
5814 if ((!xport_stp_forward_state(xport
) && !xport_stp_learn_state(xport
)) ||
5815 (!xport_rstp_forward_state(xport
) && !xport_rstp_learn_state(xport
))) {
5823 xlate_write_actions__(struct xlate_ctx
*ctx
,
5824 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5826 /* Maintain actset_output depending on the contents of the action set:
5828 * - OFPP_UNSET, if there is no "output" action.
5830 * - The output port, if there is an "output" action and no "group"
5833 * - OFPP_UNSET, if there is a "group" action.
5835 if (!ctx
->action_set_has_group
) {
5836 const struct ofpact
*a
;
5837 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5838 if (a
->type
== OFPACT_OUTPUT
) {
5839 ctx
->xin
->flow
.actset_output
= ofpact_get_OUTPUT(a
)->port
;
5840 } else if (a
->type
== OFPACT_GROUP
) {
5841 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
5842 ctx
->action_set_has_group
= true;
5848 ofpbuf_put(&ctx
->action_set
, ofpacts
, ofpacts_len
);
5852 xlate_write_actions(struct xlate_ctx
*ctx
, const struct ofpact_nest
*a
)
5854 xlate_write_actions__(ctx
, a
->actions
, ofpact_nest_get_action_len(a
));
5858 xlate_action_set(struct xlate_ctx
*ctx
)
5860 uint64_t action_list_stub
[1024 / 8];
5861 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
5862 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
5863 /* Clear the action set, as it is not needed any more. */
5864 ofpbuf_clear(&ctx
->action_set
);
5865 if (action_list
.size
) {
5866 ctx
->in_action_set
= true;
5868 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
5869 ctx
->xin
->trace
= xlate_report(ctx
, OFT_TABLE
,
5870 "--. Executing action set:");
5871 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, true, false);
5872 ctx
->xin
->trace
= old_trace
;
5874 ctx
->in_action_set
= false;
5876 ofpbuf_uninit(&action_list
);
5880 freeze_put_unroll_xlate(struct xlate_ctx
*ctx
)
5882 struct ofpact_unroll_xlate
*unroll
= ctx
->frozen_actions
.header
;
5884 /* Restore the table_id and rule cookie for a potential PACKET
5887 (ctx
->table_id
!= unroll
->rule_table_id
5888 || ctx
->rule_cookie
!= unroll
->rule_cookie
)) {
5889 unroll
= ofpact_put_UNROLL_XLATE(&ctx
->frozen_actions
);
5890 unroll
->rule_table_id
= ctx
->table_id
;
5891 unroll
->rule_cookie
= ctx
->rule_cookie
;
5892 ctx
->frozen_actions
.header
= unroll
;
5897 /* Copy actions 'a' through 'end' to ctx->frozen_actions, which will be
5898 * executed after thawing. Inserts an UNROLL_XLATE action, if none is already
5899 * present, before any action that may depend on the current table ID or flow
5902 freeze_unroll_actions(const struct ofpact
*a
, const struct ofpact
*end
,
5903 struct xlate_ctx
*ctx
)
5905 for (; a
< end
; a
= ofpact_next(a
)) {
5907 case OFPACT_OUTPUT_REG
:
5908 case OFPACT_OUTPUT_TRUNC
:
5911 case OFPACT_CONTROLLER
:
5912 case OFPACT_DEC_MPLS_TTL
:
5913 case OFPACT_DEC_NSH_TTL
:
5914 case OFPACT_DEC_TTL
:
5915 /* These actions may generate asynchronous messages, which include
5916 * table ID and flow cookie information. */
5917 freeze_put_unroll_xlate(ctx
);
5920 case OFPACT_RESUBMIT
:
5921 if (ofpact_get_RESUBMIT(a
)->table_id
== 0xff) {
5922 /* This resubmit action is relative to the current table, so we
5923 * need to track what table that is.*/
5924 freeze_put_unroll_xlate(ctx
);
5928 case OFPACT_SET_TUNNEL
:
5929 case OFPACT_REG_MOVE
:
5930 case OFPACT_SET_FIELD
:
5931 case OFPACT_STACK_PUSH
:
5932 case OFPACT_STACK_POP
:
5934 case OFPACT_WRITE_METADATA
:
5935 case OFPACT_GOTO_TABLE
:
5936 case OFPACT_ENQUEUE
:
5937 case OFPACT_SET_VLAN_VID
:
5938 case OFPACT_SET_VLAN_PCP
:
5939 case OFPACT_STRIP_VLAN
:
5940 case OFPACT_PUSH_VLAN
:
5941 case OFPACT_SET_ETH_SRC
:
5942 case OFPACT_SET_ETH_DST
:
5943 case OFPACT_SET_IPV4_SRC
:
5944 case OFPACT_SET_IPV4_DST
:
5945 case OFPACT_SET_IP_DSCP
:
5946 case OFPACT_SET_IP_ECN
:
5947 case OFPACT_SET_IP_TTL
:
5948 case OFPACT_SET_L4_SRC_PORT
:
5949 case OFPACT_SET_L4_DST_PORT
:
5950 case OFPACT_SET_QUEUE
:
5951 case OFPACT_POP_QUEUE
:
5952 case OFPACT_PUSH_MPLS
:
5953 case OFPACT_POP_MPLS
:
5954 case OFPACT_SET_MPLS_LABEL
:
5955 case OFPACT_SET_MPLS_TC
:
5956 case OFPACT_SET_MPLS_TTL
:
5957 case OFPACT_MULTIPATH
:
5960 case OFPACT_UNROLL_XLATE
:
5961 case OFPACT_FIN_TIMEOUT
:
5962 case OFPACT_CLEAR_ACTIONS
:
5963 case OFPACT_WRITE_ACTIONS
:
5969 case OFPACT_DEBUG_RECIRC
:
5970 case OFPACT_DEBUG_SLOW
:
5972 case OFPACT_CT_CLEAR
:
5974 case OFPACT_CHECK_PKT_LARGER
:
5975 /* These may not generate PACKET INs. */
5979 case OFPACT_CONJUNCTION
:
5980 /* These need not be copied for restoration. */
5983 /* Copy the action over. */
5984 ofpbuf_put(&ctx
->frozen_actions
, a
, OFPACT_ALIGN(a
->len
));
5989 put_ct_mark(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
5990 struct flow_wildcards
*wc
)
5992 if (wc
->masks
.ct_mark
) {
5998 odp_ct_mark
= nl_msg_put_unspec_uninit(odp_actions
, OVS_CT_ATTR_MARK
,
5999 sizeof(*odp_ct_mark
));
6000 odp_ct_mark
->key
= flow
->ct_mark
& wc
->masks
.ct_mark
;
6001 odp_ct_mark
->mask
= wc
->masks
.ct_mark
;
6006 put_ct_label(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
6007 struct flow_wildcards
*wc
)
6009 if (!ovs_u128_is_zero(wc
->masks
.ct_label
)) {
6015 odp_ct_label
.key
= ovs_u128_and(flow
->ct_label
, wc
->masks
.ct_label
);
6016 odp_ct_label
.mask
= wc
->masks
.ct_label
;
6017 nl_msg_put_unspec(odp_actions
, OVS_CT_ATTR_LABELS
,
6018 &odp_ct_label
, sizeof odp_ct_label
);
6023 put_drop_action(struct ofpbuf
*odp_actions
, enum xlate_error error
)
6025 nl_msg_put_u32(odp_actions
, OVS_ACTION_ATTR_DROP
, error
);
6029 put_ct_helper(struct xlate_ctx
*ctx
,
6030 struct ofpbuf
*odp_actions
, struct ofpact_conntrack
*ofc
)
6035 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "ftp");
6038 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "tftp");
6041 xlate_report_error(ctx
, "cannot serialize ct_helper %d", ofc
->alg
);
6048 put_ct_timeout(struct ofpbuf
*odp_actions
, const struct dpif_backer
*backer
,
6049 const struct flow
*flow
, struct flow_wildcards
*wc
,
6053 char *tp_name
= NULL
;
6055 if (ofproto_dpif_ct_zone_timeout_policy_get_name(backer
, zone_id
,
6056 ntohs(flow
->dl_type
), flow
->nw_proto
, &tp_name
, &unwildcard
)) {
6057 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_TIMEOUT
, tp_name
);
6060 /* The underlying datapath requires separate timeout
6061 * policies for different Ethertypes and IP protocols. We
6062 * don't need to unwildcard 'wc->masks.dl_type' since that
6063 * field is always unwildcarded in megaflows. */
6064 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6071 put_ct_nat(struct xlate_ctx
*ctx
)
6073 struct ofpact_nat
*ofn
= ctx
->ct_nat_action
;
6080 nat_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_CT_ATTR_NAT
);
6081 if (ofn
->flags
& NX_NAT_F_SRC
|| ofn
->flags
& NX_NAT_F_DST
) {
6082 nl_msg_put_flag(ctx
->odp_actions
, ofn
->flags
& NX_NAT_F_SRC
6083 ? OVS_NAT_ATTR_SRC
: OVS_NAT_ATTR_DST
);
6084 if (ofn
->flags
& NX_NAT_F_PERSISTENT
) {
6085 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PERSISTENT
);
6087 if (ofn
->flags
& NX_NAT_F_PROTO_HASH
) {
6088 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_HASH
);
6089 } else if (ofn
->flags
& NX_NAT_F_PROTO_RANDOM
) {
6090 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_RANDOM
);
6092 if (ofn
->range_af
== AF_INET
) {
6093 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6094 ofn
->range
.addr
.ipv4
.min
);
6095 if (ofn
->range
.addr
.ipv4
.max
&&
6096 (ntohl(ofn
->range
.addr
.ipv4
.max
)
6097 > ntohl(ofn
->range
.addr
.ipv4
.min
))) {
6098 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6099 ofn
->range
.addr
.ipv4
.max
);
6101 } else if (ofn
->range_af
== AF_INET6
) {
6102 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6103 &ofn
->range
.addr
.ipv6
.min
,
6104 sizeof ofn
->range
.addr
.ipv6
.min
);
6105 if (!ipv6_mask_is_any(&ofn
->range
.addr
.ipv6
.max
) &&
6106 memcmp(&ofn
->range
.addr
.ipv6
.max
, &ofn
->range
.addr
.ipv6
.min
,
6107 sizeof ofn
->range
.addr
.ipv6
.max
) > 0) {
6108 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6109 &ofn
->range
.addr
.ipv6
.max
,
6110 sizeof ofn
->range
.addr
.ipv6
.max
);
6113 if (ofn
->range_af
!= AF_UNSPEC
&& ofn
->range
.proto
.min
) {
6114 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MIN
,
6115 ofn
->range
.proto
.min
);
6116 if (ofn
->range
.proto
.max
&&
6117 ofn
->range
.proto
.max
> ofn
->range
.proto
.min
) {
6118 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MAX
,
6119 ofn
->range
.proto
.max
);
6123 nl_msg_end_nested(ctx
->odp_actions
, nat_offset
);
6127 compose_conntrack_action(struct xlate_ctx
*ctx
, struct ofpact_conntrack
*ofc
,
6128 bool is_last_action
)
6130 ovs_u128 old_ct_label_mask
= ctx
->wc
->masks
.ct_label
;
6131 uint32_t old_ct_mark_mask
= ctx
->wc
->masks
.ct_mark
;
6135 /* Ensure that any prior actions are applied before composing the new
6136 * conntrack action. */
6137 xlate_commit_actions(ctx
);
6139 /* Process nested actions first, to populate the key. */
6140 ctx
->ct_nat_action
= NULL
;
6141 ctx
->wc
->masks
.ct_mark
= 0;
6142 ctx
->wc
->masks
.ct_label
= OVS_U128_ZERO
;
6143 do_xlate_actions(ofc
->actions
, ofpact_ct_get_action_len(ofc
), ctx
,
6144 is_last_action
, false);
6146 if (ofc
->zone_src
.field
) {
6147 zone
= mf_get_subfield(&ofc
->zone_src
, &ctx
->xin
->flow
);
6149 zone
= ofc
->zone_imm
;
6152 ct_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CT
);
6153 if (ofc
->flags
& NX_CT_F_COMMIT
) {
6154 nl_msg_put_flag(ctx
->odp_actions
, ofc
->flags
& NX_CT_F_FORCE
?
6155 OVS_CT_ATTR_FORCE_COMMIT
: OVS_CT_ATTR_COMMIT
);
6156 if (ctx
->xbridge
->support
.ct_eventmask
) {
6157 nl_msg_put_u32(ctx
->odp_actions
, OVS_CT_ATTR_EVENTMASK
,
6158 OVS_CT_EVENTMASK_DEFAULT
);
6160 if (ctx
->xbridge
->support
.ct_timeout
) {
6161 put_ct_timeout(ctx
->odp_actions
, ctx
->xbridge
->ofproto
->backer
,
6162 &ctx
->xin
->flow
, ctx
->wc
, zone
);
6165 nl_msg_put_u16(ctx
->odp_actions
, OVS_CT_ATTR_ZONE
, zone
);
6166 put_ct_mark(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6167 put_ct_label(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6168 put_ct_helper(ctx
, ctx
->odp_actions
, ofc
);
6170 ctx
->ct_nat_action
= NULL
;
6171 nl_msg_end_nested(ctx
->odp_actions
, ct_offset
);
6173 ctx
->wc
->masks
.ct_mark
= old_ct_mark_mask
;
6174 ctx
->wc
->masks
.ct_label
= old_ct_label_mask
;
6176 if (ofc
->recirc_table
!= NX_CT_RECIRC_NONE
) {
6177 ctx
->conntracked
= true;
6178 compose_recirculate_and_fork(ctx
, ofc
->recirc_table
, zone
);
6181 /* The ct_* fields are only available in the scope of the 'recirc_table'
6183 flow_clear_conntrack(&ctx
->xin
->flow
);
6184 xlate_report(ctx
, OFT_DETAIL
, "Sets the packet to an untracked state, "
6185 "and clears all the conntrack fields.");
6186 ctx
->conntracked
= false;
6190 compose_ct_clear_action(struct xlate_ctx
*ctx
)
6192 clear_conntrack(ctx
);
6193 /* This action originally existed without dpif support. So to preserve
6194 * compatibility, only append it if the dpif supports it. */
6195 if (ctx
->xbridge
->support
.ct_clear
) {
6196 nl_msg_put_flag(ctx
->odp_actions
, OVS_ACTION_ATTR_CT_CLEAR
);
6200 /* check_pkt_larger action checks the packet length and stores the
6201 * result in the register bit. We translate this action to the
6202 * datapath action - 'check_pkt_len' whose format
6203 * is: 'check_pkt_len(pkt_len, ge(actions), le(actions))'.
6205 * We first set the destination register bit to 1 and call
6206 * 'do_xlate_actions' for the case - packet len greater than
6207 * the specified packet length.
6209 * We then set the destination register bit to 0 and call
6210 * 'do_xlate_actions' for the case - packet length is lesser or
6211 * equal to the specified packet length.
6213 * It is possible for freezing to happen for both the cases.
6216 xlate_check_pkt_larger(struct xlate_ctx
*ctx
,
6217 struct ofpact_check_pkt_larger
*check_pkt_larger
,
6218 const struct ofpact
*remaining_acts
,
6219 size_t remaining_acts_len
)
6221 union mf_subvalue value
;
6222 memset(&value
, 0, sizeof value
);
6223 if (!ctx
->xbridge
->support
.check_pkt_len
) {
6224 uint8_t is_pkt_larger
= 0;
6225 if (ctx
->xin
->packet
) {
6227 dp_packet_size(ctx
->xin
->packet
) > check_pkt_larger
->pkt_len
;
6229 value
.u8_val
= is_pkt_larger
;
6230 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
,
6232 /* If datapath doesn't support check_pkt_len action, then set the
6233 * SLOW_ACTION flag. If we don't set SLOW_ACTION, we
6234 * will push a flow to the datapath based on the packet length
6235 * in ctx->xin->packet. For subsequent patches which match the
6236 * same flow, datapath will apply the actions without considering
6237 * the packet length. This results in wrong actions being applied.
6239 ctx
->xout
->slow
|= SLOW_ACTION
;
6243 struct ofpbuf old_stack
= ctx
->stack
;
6244 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
6245 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
6246 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
6248 struct ofpbuf old_action_set
= ctx
->action_set
;
6249 uint64_t actset_stub
[1024 / 8];
6250 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
6251 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
6253 struct flow old_flow
= ctx
->xin
->flow
;
6254 xlate_commit_actions(ctx
);
6255 struct flow old_base
= ctx
->base_flow
;
6256 bool old_was_mpls
= ctx
->was_mpls
;
6257 bool old_conntracked
= ctx
->conntracked
;
6259 size_t offset
= nl_msg_start_nested(ctx
->odp_actions
,
6260 OVS_ACTION_ATTR_CHECK_PKT_LEN
);
6261 nl_msg_put_u16(ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN
,
6262 check_pkt_larger
->pkt_len
);
6263 size_t offset_attr
= nl_msg_start_nested(
6264 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER
);
6266 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6267 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6268 if (!ctx
->freezing
) {
6269 xlate_action_set(ctx
);
6271 if (ctx
->freezing
) {
6272 finish_freezing(ctx
);
6274 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6276 ctx
->base_flow
= old_base
;
6277 ctx
->was_mpls
= old_was_mpls
;
6278 ctx
->conntracked
= old_conntracked
;
6279 ctx
->xin
->flow
= old_flow
;
6281 /* If the flow translation for the IF_GREATER case requires freezing,
6282 * then ctx->exit would be true. Reset to false so that we can
6283 * do flow translation for 'IF_LESS_EQUAL' case. finish_freezing()
6284 * would have taken care of Undoing the changes done for freeze. */
6287 offset_attr
= nl_msg_start_nested(
6288 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL
);
6290 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6291 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6292 if (!ctx
->freezing
) {
6293 xlate_action_set(ctx
);
6295 if (ctx
->freezing
) {
6296 finish_freezing(ctx
);
6298 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6299 nl_msg_end_nested(ctx
->odp_actions
, offset
);
6301 ofpbuf_uninit(&ctx
->action_set
);
6302 ctx
->action_set
= old_action_set
;
6303 ofpbuf_uninit(&ctx
->stack
);
6304 ctx
->stack
= old_stack
;
6305 ctx
->base_flow
= old_base
;
6306 ctx
->was_mpls
= old_was_mpls
;
6307 ctx
->conntracked
= old_conntracked
;
6308 ctx
->xin
->flow
= old_flow
;
6313 rewrite_flow_encap_ethernet(struct xlate_ctx
*ctx
,
6315 struct flow_wildcards
*wc
)
6317 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6318 if (pt_ns(flow
->packet_type
) == OFPHTN_ETHERTYPE
) {
6319 /* Only adjust the packet_type and zero the dummy Ethernet addresses. */
6320 ovs_be16 ethertype
= pt_ns_type_be(flow
->packet_type
);
6321 flow
->packet_type
= htonl(PT_ETH
);
6322 flow
->dl_src
= eth_addr_zero
;
6323 flow
->dl_dst
= eth_addr_zero
;
6324 flow
->dl_type
= ethertype
;
6326 /* Error handling: drop packet. */
6327 xlate_report_debug(ctx
, OFT_ACTION
,
6328 "Dropping packet as encap(ethernet) is not "
6329 "supported for packet type ethernet.");
6330 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6334 /* For an MD2 NSH header returns a pointer to an ofpbuf with the encoded
6335 * MD2 TLVs provided as encap properties to the encap operation. This
6336 * will be stored as encap_data in the ctx and copied into the push_nsh
6337 * action at the next commit. */
6338 static struct ofpbuf
*
6339 rewrite_flow_push_nsh(struct xlate_ctx
*ctx
,
6340 const struct ofpact_encap
*encap
,
6342 struct flow_wildcards
*wc
)
6344 ovs_be32 packet_type
= flow
->packet_type
;
6345 const char *ptr
= (char *) encap
->props
;
6346 struct ofpbuf
*buf
= ofpbuf_new(NSH_CTX_HDRS_MAX_LEN
);
6347 uint8_t md_type
= NSH_M_TYPE1
;
6351 /* Scan the optional NSH encap TLV properties, if any. */
6352 for (i
= 0; i
< encap
->n_props
; i
++) {
6353 struct ofpact_ed_prop
*prop_ptr
=
6354 ALIGNED_CAST(struct ofpact_ed_prop
*, ptr
);
6355 if (prop_ptr
->prop_class
== OFPPPC_NSH
) {
6356 switch (prop_ptr
->type
) {
6357 case OFPPPT_PROP_NSH_MDTYPE
: {
6358 struct ofpact_ed_prop_nsh_md_type
*prop_md_type
=
6359 ALIGNED_CAST(struct ofpact_ed_prop_nsh_md_type
*,
6361 md_type
= prop_md_type
->md_type
;
6364 case OFPPPT_PROP_NSH_TLV
: {
6365 struct ofpact_ed_prop_nsh_tlv
*tlv_prop
=
6366 ALIGNED_CAST(struct ofpact_ed_prop_nsh_tlv
*,
6368 struct nsh_md2_tlv
*md2_ctx
=
6369 ofpbuf_put_uninit(buf
, sizeof(*md2_ctx
));
6370 md2_ctx
->md_class
= tlv_prop
->tlv_class
;
6371 md2_ctx
->type
= tlv_prop
->tlv_type
;
6372 md2_ctx
->length
= tlv_prop
->tlv_len
;
6373 size_t len
= ROUND_UP(md2_ctx
->length
, 4);
6374 size_t padding
= len
- md2_ctx
->length
;
6375 ofpbuf_put(buf
, tlv_prop
->data
, md2_ctx
->length
);
6376 ofpbuf_put_zeros(buf
, padding
);
6380 /* No other NSH encap properties defined yet. */
6384 ptr
+= ROUND_UP(prop_ptr
->len
, 8);
6386 if (buf
->size
== 0 || buf
->size
> NSH_CTX_HDRS_MAX_LEN
) {
6391 /* Determine the Next Protocol field for NSH header. */
6392 switch (ntohl(packet_type
)) {
6394 np
= NSH_P_ETHERNET
;
6406 /* Error handling: drop packet. */
6407 xlate_report_debug(ctx
, OFT_ACTION
,
6408 "Dropping packet as encap(nsh) is not "
6409 "supported for packet type (%d,0x%x)",
6410 pt_ns(packet_type
), pt_ns_type(packet_type
));
6411 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6414 /* Note that we have matched on packet_type! */
6415 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6417 /* Reset all current flow packet headers. */
6418 memset(&flow
->dl_dst
, 0,
6419 sizeof(struct flow
) - offsetof(struct flow
, dl_dst
));
6421 /* Populate the flow with the new NSH header. */
6422 flow
->packet_type
= htonl(PT_NSH
);
6423 flow
->dl_type
= htons(ETH_TYPE_NSH
);
6424 flow
->nsh
.flags
= 0;
6427 flow
->nsh
.path_hdr
= htonl(255);
6429 if (md_type
== NSH_M_TYPE1
) {
6430 flow
->nsh
.mdtype
= NSH_M_TYPE1
;
6431 memset(flow
->nsh
.context
, 0, sizeof flow
->nsh
.context
);
6433 /* Drop any MD2 context TLVs. */
6437 } else if (md_type
== NSH_M_TYPE2
) {
6438 flow
->nsh
.mdtype
= NSH_M_TYPE2
;
6440 flow
->nsh
.mdtype
&= NSH_MDTYPE_MASK
;
6446 xlate_generic_encap_action(struct xlate_ctx
*ctx
,
6447 const struct ofpact_encap
*encap
)
6449 struct flow
*flow
= &ctx
->xin
->flow
;
6450 struct flow_wildcards
*wc
= ctx
->wc
;
6451 struct ofpbuf
*encap_data
= NULL
;
6453 /* Ensure that any pending actions on the inner packet are applied before
6454 * rewriting the flow */
6455 xlate_commit_actions(ctx
);
6457 /* Rewrite the flow to reflect the effect of pushing the new encap header. */
6458 switch (ntohl(encap
->new_pkt_type
)) {
6460 rewrite_flow_encap_ethernet(ctx
, flow
, wc
);
6463 encap_data
= rewrite_flow_push_nsh(ctx
, encap
, flow
, wc
);
6466 /* New packet type was checked during decoding. */
6471 /* The actual encap datapath action will be generated at next commit. */
6472 ctx
->pending_encap
= true;
6473 ctx
->encap_data
= encap_data
;
6477 /* Returns true if packet must be recirculated after decapsulation. */
6479 xlate_generic_decap_action(struct xlate_ctx
*ctx
,
6480 const struct ofpact_decap
*decap OVS_UNUSED
)
6482 struct flow
*flow
= &ctx
->xin
->flow
;
6484 /* Ensure that any pending actions on the current packet are applied
6485 * before generating the decap action. */
6486 xlate_commit_actions(ctx
);
6488 /* We assume for now that the new_pkt_type is PT_USE_NEXT_PROTO. */
6489 switch (ntohl(flow
->packet_type
)) {
6491 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
)) {
6492 /* Error handling: drop packet. */
6493 xlate_report_debug(ctx
, OFT_ACTION
, "Dropping packet, cannot "
6494 "decap Ethernet if VLAN is present.");
6495 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6497 /* Just change the packet_type.
6498 * Delay generating pop_eth to the next commit. */
6499 flow
->packet_type
= htonl(PACKET_TYPE(OFPHTN_ETHERTYPE
,
6500 ntohs(flow
->dl_type
)));
6501 ctx
->wc
->masks
.dl_type
= OVS_BE16_MAX
;
6505 /* The pop_nsh action is generated at the commit executed as
6506 * part of freezing the ctx for recirculation. Here we just set
6507 * the new packet type based on the NSH next protocol field. */
6508 switch (flow
->nsh
.np
) {
6509 case NSH_P_ETHERNET
:
6510 flow
->packet_type
= htonl(PT_ETH
);
6513 flow
->packet_type
= htonl(PT_IPV4
);
6516 flow
->packet_type
= htonl(PT_IPV6
);
6519 flow
->packet_type
= htonl(PT_NSH
);
6522 /* Error handling: drop packet. */
6523 xlate_report_debug(ctx
, OFT_ACTION
,
6524 "Dropping packet as NSH next protocol %d "
6525 "is not supported", flow
->nsh
.np
);
6526 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6530 ctx
->wc
->masks
.nsh
.np
= UINT8_MAX
;
6531 ctx
->pending_decap
= true;
6532 /* Trigger recirculation. */
6535 /* Error handling: drop packet. */
6538 "Dropping packet as the decap() does not support "
6539 "packet type (%d,0x%x)",
6540 pt_ns(flow
->packet_type
), pt_ns_type(flow
->packet_type
));
6541 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6547 recirc_for_mpls(const struct ofpact
*a
, struct xlate_ctx
*ctx
)
6549 /* No need to recirculate if already exiting. */
6554 /* Do not consider recirculating unless the packet was previously MPLS. */
6555 if (!ctx
->was_mpls
) {
6559 /* Special case these actions, only recirculating if necessary.
6560 * This avoids the overhead of recirculation in common use-cases.
6564 /* Output actions do not require recirculation. */
6566 case OFPACT_OUTPUT_TRUNC
:
6567 case OFPACT_ENQUEUE
:
6568 case OFPACT_OUTPUT_REG
:
6569 /* Set actions that don't touch L3+ fields do not require recirculation. */
6570 case OFPACT_SET_VLAN_VID
:
6571 case OFPACT_SET_VLAN_PCP
:
6572 case OFPACT_SET_ETH_SRC
:
6573 case OFPACT_SET_ETH_DST
:
6574 case OFPACT_SET_TUNNEL
:
6575 case OFPACT_SET_QUEUE
:
6576 /* If actions of a group require recirculation that can be detected
6577 * when translating them. */
6581 /* Set field that don't touch L3+ fields don't require recirculation. */
6582 case OFPACT_SET_FIELD
:
6583 if (mf_is_l3_or_higher(ofpact_get_SET_FIELD(a
)->field
)) {
6588 /* For simplicity, recirculate in all other cases. */
6589 case OFPACT_CONTROLLER
:
6591 case OFPACT_STRIP_VLAN
:
6592 case OFPACT_PUSH_VLAN
:
6593 case OFPACT_SET_IPV4_SRC
:
6594 case OFPACT_SET_IPV4_DST
:
6595 case OFPACT_SET_IP_DSCP
:
6596 case OFPACT_SET_IP_ECN
:
6597 case OFPACT_SET_IP_TTL
:
6598 case OFPACT_SET_L4_SRC_PORT
:
6599 case OFPACT_SET_L4_DST_PORT
:
6600 case OFPACT_REG_MOVE
:
6601 case OFPACT_STACK_PUSH
:
6602 case OFPACT_STACK_POP
:
6603 case OFPACT_DEC_TTL
:
6604 case OFPACT_SET_MPLS_LABEL
:
6605 case OFPACT_SET_MPLS_TC
:
6606 case OFPACT_SET_MPLS_TTL
:
6607 case OFPACT_DEC_MPLS_TTL
:
6608 case OFPACT_PUSH_MPLS
:
6609 case OFPACT_POP_MPLS
:
6610 case OFPACT_POP_QUEUE
:
6611 case OFPACT_FIN_TIMEOUT
:
6612 case OFPACT_RESUBMIT
:
6614 case OFPACT_CONJUNCTION
:
6615 case OFPACT_MULTIPATH
:
6622 case OFPACT_DEC_NSH_TTL
:
6623 case OFPACT_UNROLL_XLATE
:
6625 case OFPACT_CT_CLEAR
:
6627 case OFPACT_DEBUG_RECIRC
:
6628 case OFPACT_DEBUG_SLOW
:
6630 case OFPACT_CLEAR_ACTIONS
:
6631 case OFPACT_WRITE_ACTIONS
:
6632 case OFPACT_WRITE_METADATA
:
6633 case OFPACT_GOTO_TABLE
:
6634 case OFPACT_CHECK_PKT_LARGER
:
6640 ctx_trigger_freeze(ctx
);
6644 xlate_ofpact_reg_move(struct xlate_ctx
*ctx
, const struct ofpact_reg_move
*a
)
6646 mf_subfield_copy(&a
->src
, &a
->dst
, &ctx
->xin
->flow
, ctx
->wc
);
6647 xlate_report_subfield(ctx
, &a
->dst
);
6651 xlate_ofpact_stack_pop(struct xlate_ctx
*ctx
, const struct ofpact_stack
*a
)
6653 if (nxm_execute_stack_pop(a
, &ctx
->xin
->flow
, ctx
->wc
, &ctx
->stack
)) {
6654 xlate_report_subfield(ctx
, &a
->subfield
);
6656 xlate_report_error(ctx
, "stack underflow");
6660 /* Restore translation context data that was stored earlier. */
6662 xlate_ofpact_unroll_xlate(struct xlate_ctx
*ctx
,
6663 const struct ofpact_unroll_xlate
*a
)
6665 ctx
->table_id
= a
->rule_table_id
;
6666 ctx
->rule_cookie
= a
->rule_cookie
;
6667 xlate_report(ctx
, OFT_THAW
, "restored state: table=%"PRIu8
", "
6668 "cookie=%#"PRIx64
, a
->rule_table_id
, a
->rule_cookie
);
6672 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
6673 struct xlate_ctx
*ctx
, bool is_last_action
,
6674 bool group_bucket_action
)
6676 struct flow_wildcards
*wc
= ctx
->wc
;
6677 struct flow
*flow
= &ctx
->xin
->flow
;
6678 const struct ofpact
*a
;
6680 /* dl_type already in the mask, not set below. */
6683 xlate_report(ctx
, OFT_ACTION
, "drop");
6687 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
6688 struct ofpact_controller
*controller
;
6689 const struct ofpact_metadata
*metadata
;
6690 const struct ofpact_set_field
*set_field
;
6691 const struct mf_field
*mf
;
6692 bool last
= is_last_action
&& ofpact_last(a
, ofpacts
, ofpacts_len
)
6693 && ctx
->action_set
.size
;
6699 recirc_for_mpls(a
, ctx
);
6702 /* Check if need to store the remaining actions for later
6704 if (ctx
->freezing
) {
6705 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
),
6711 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
6712 struct ds s
= DS_EMPTY_INITIALIZER
;
6713 struct ofpact_format_params fp
= { .s
= &s
};
6714 ofpacts_format(a
, OFPACT_ALIGN(a
->len
), &fp
);
6715 xlate_report(ctx
, OFT_ACTION
, "%s", ds_cstr(&s
));
6721 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
6722 ofpact_get_OUTPUT(a
)->max_len
, true, last
,
6723 false, group_bucket_action
);
6727 if (xlate_group_action(ctx
, ofpact_get_GROUP(a
)->group_id
, last
)) {
6728 /* Group could not be found. */
6730 /* XXX: Terminates action list translation, but does not
6731 * terminate the pipeline. */
6736 case OFPACT_CONTROLLER
:
6737 controller
= ofpact_get_CONTROLLER(a
);
6738 if (controller
->pause
) {
6739 ctx
->pause
= controller
;
6740 ctx_trigger_freeze(ctx
);
6743 xlate_controller_action(ctx
, controller
->max_len
,
6745 controller
->controller_id
,
6746 controller
->provider_meter_id
,
6747 controller
->userdata
,
6748 controller
->userdata_len
);
6752 case OFPACT_ENQUEUE
:
6753 memset(&wc
->masks
.skb_priority
, 0xff,
6754 sizeof wc
->masks
.skb_priority
);
6755 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
), last
,
6756 group_bucket_action
);
6759 case OFPACT_SET_VLAN_VID
:
6760 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
6761 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6762 ofpact_get_SET_VLAN_VID(a
)->push_vlan_if_needed
) {
6763 if (!flow
->vlans
[0].tpid
) {
6764 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6766 flow
->vlans
[0].tci
&= ~htons(VLAN_VID_MASK
);
6767 flow
->vlans
[0].tci
|=
6768 (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
) |
6773 case OFPACT_SET_VLAN_PCP
:
6774 wc
->masks
.vlans
[0].tci
|= htons(VLAN_PCP_MASK
| VLAN_CFI
);
6775 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6776 ofpact_get_SET_VLAN_PCP(a
)->push_vlan_if_needed
) {
6777 if (!flow
->vlans
[0].tpid
) {
6778 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6780 flow
->vlans
[0].tci
&= ~htons(VLAN_PCP_MASK
);
6781 flow
->vlans
[0].tci
|=
6782 htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
6783 << VLAN_PCP_SHIFT
) | VLAN_CFI
);
6787 case OFPACT_STRIP_VLAN
:
6788 flow_pop_vlan(flow
, wc
);
6791 case OFPACT_PUSH_VLAN
:
6792 flow_push_vlan_uninit(flow
, wc
);
6793 flow
->vlans
[0].tpid
= ofpact_get_PUSH_VLAN(a
)->ethertype
;
6794 flow
->vlans
[0].tci
= htons(VLAN_CFI
);
6797 case OFPACT_SET_ETH_SRC
:
6798 WC_MASK_FIELD(wc
, dl_src
);
6799 flow
->dl_src
= ofpact_get_SET_ETH_SRC(a
)->mac
;
6802 case OFPACT_SET_ETH_DST
:
6803 WC_MASK_FIELD(wc
, dl_dst
);
6804 flow
->dl_dst
= ofpact_get_SET_ETH_DST(a
)->mac
;
6807 case OFPACT_SET_IPV4_SRC
:
6808 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6809 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
6810 flow
->nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
6814 case OFPACT_SET_IPV4_DST
:
6815 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6816 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
6817 flow
->nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
6821 case OFPACT_SET_IP_DSCP
:
6822 if (is_ip_any(flow
)) {
6823 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
6824 flow
->nw_tos
&= ~IP_DSCP_MASK
;
6825 flow
->nw_tos
|= ofpact_get_SET_IP_DSCP(a
)->dscp
;
6829 case OFPACT_SET_IP_ECN
:
6830 if (is_ip_any(flow
)) {
6831 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
6832 flow
->nw_tos
&= ~IP_ECN_MASK
;
6833 flow
->nw_tos
|= ofpact_get_SET_IP_ECN(a
)->ecn
;
6837 case OFPACT_SET_IP_TTL
:
6838 if (is_ip_any(flow
)) {
6839 wc
->masks
.nw_ttl
= 0xff;
6840 flow
->nw_ttl
= ofpact_get_SET_IP_TTL(a
)->ttl
;
6844 case OFPACT_SET_L4_SRC_PORT
:
6845 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6846 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6847 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
6848 flow
->tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
6852 case OFPACT_SET_L4_DST_PORT
:
6853 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6854 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6855 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
6856 flow
->tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
6860 case OFPACT_RESUBMIT
:
6861 /* Freezing complicates resubmit. Some action in the flow
6862 * entry found by resubmit might trigger freezing. If that
6863 * happens, then we do not want to execute the resubmit again after
6864 * during thawing, so we want to skip back to the head of the loop
6865 * to avoid that, only adding any actions that follow the resubmit
6866 * to the frozen actions.
6868 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
), last
);
6871 case OFPACT_SET_TUNNEL
:
6872 flow
->tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
6875 case OFPACT_SET_QUEUE
:
6876 memset(&wc
->masks
.skb_priority
, 0xff,
6877 sizeof wc
->masks
.skb_priority
);
6878 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
6881 case OFPACT_POP_QUEUE
:
6882 memset(&wc
->masks
.skb_priority
, 0xff,
6883 sizeof wc
->masks
.skb_priority
);
6884 if (flow
->skb_priority
!= ctx
->orig_skb_priority
) {
6885 flow
->skb_priority
= ctx
->orig_skb_priority
;
6886 xlate_report(ctx
, OFT_DETAIL
, "queue = %#"PRIx32
,
6887 flow
->skb_priority
);
6891 case OFPACT_REG_MOVE
:
6892 xlate_ofpact_reg_move(ctx
, ofpact_get_REG_MOVE(a
));
6895 case OFPACT_SET_FIELD
:
6896 set_field
= ofpact_get_SET_FIELD(a
);
6897 mf
= set_field
->field
;
6899 /* Set the field only if the packet actually has it. */
6900 if (mf_are_prereqs_ok(mf
, flow
, wc
)) {
6901 mf_mask_field_masked(mf
, ofpact_set_field_mask(set_field
), wc
);
6902 mf_set_flow_value_masked(mf
, set_field
->value
,
6903 ofpact_set_field_mask(set_field
),
6906 xlate_report(ctx
, OFT_WARN
,
6907 "unmet prerequisites for %s, set_field ignored",
6913 case OFPACT_STACK_PUSH
:
6914 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), flow
, wc
,
6918 case OFPACT_STACK_POP
:
6919 xlate_ofpact_stack_pop(ctx
, ofpact_get_STACK_POP(a
));
6922 case OFPACT_PUSH_MPLS
:
6923 compose_mpls_push_action(ctx
, ofpact_get_PUSH_MPLS(a
));
6926 case OFPACT_POP_MPLS
:
6927 compose_mpls_pop_action(ctx
, ofpact_get_POP_MPLS(a
)->ethertype
);
6930 case OFPACT_SET_MPLS_LABEL
:
6931 compose_set_mpls_label_action(
6932 ctx
, ofpact_get_SET_MPLS_LABEL(a
)->label
);
6935 case OFPACT_SET_MPLS_TC
:
6936 compose_set_mpls_tc_action(ctx
, ofpact_get_SET_MPLS_TC(a
)->tc
);
6939 case OFPACT_SET_MPLS_TTL
:
6940 compose_set_mpls_ttl_action(ctx
, ofpact_get_SET_MPLS_TTL(a
)->ttl
);
6943 case OFPACT_DEC_MPLS_TTL
:
6944 if (compose_dec_mpls_ttl_action(ctx
)) {
6949 case OFPACT_DEC_NSH_TTL
:
6950 if (compose_dec_nsh_ttl_action(ctx
)) {
6955 case OFPACT_DEC_TTL
:
6956 wc
->masks
.nw_ttl
= 0xff;
6957 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
6963 /* Nothing to do. */
6966 case OFPACT_MULTIPATH
:
6967 multipath_execute(ofpact_get_MULTIPATH(a
), flow
, wc
);
6968 xlate_report_subfield(ctx
, &ofpact_get_MULTIPATH(a
)->dst
);
6972 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
), last
,
6973 group_bucket_action
);
6976 case OFPACT_OUTPUT_REG
:
6977 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
), last
,
6978 group_bucket_action
);
6981 case OFPACT_OUTPUT_TRUNC
:
6982 xlate_output_trunc_action(ctx
, ofpact_get_OUTPUT_TRUNC(a
)->port
,
6983 ofpact_get_OUTPUT_TRUNC(a
)->max_len
, last
,
6984 group_bucket_action
);
6988 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
6991 case OFPACT_CONJUNCTION
:
6992 /* A flow with a "conjunction" action represents part of a special
6993 * kind of "set membership match". Such a flow should not actually
6994 * get executed, but it could via, say, a "packet-out", even though
6995 * that wouldn't be useful. Log it to help debugging. */
6996 xlate_report_error(ctx
, "executing no-op conjunction action");
7003 case OFPACT_UNROLL_XLATE
:
7004 xlate_ofpact_unroll_xlate(ctx
, ofpact_get_UNROLL_XLATE(a
));
7007 case OFPACT_FIN_TIMEOUT
:
7008 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
7009 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
7012 case OFPACT_CLEAR_ACTIONS
:
7013 xlate_report_action_set(ctx
, "was");
7014 ofpbuf_clear(&ctx
->action_set
);
7015 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
7016 ctx
->action_set_has_group
= false;
7019 case OFPACT_WRITE_ACTIONS
:
7020 xlate_write_actions(ctx
, ofpact_get_WRITE_ACTIONS(a
));
7021 xlate_report_action_set(ctx
, "is");
7024 case OFPACT_WRITE_METADATA
:
7025 metadata
= ofpact_get_WRITE_METADATA(a
);
7026 flow
->metadata
&= ~metadata
->mask
;
7027 flow
->metadata
|= metadata
->metadata
& metadata
->mask
;
7031 xlate_meter_action(ctx
, ofpact_get_METER(a
));
7034 case OFPACT_GOTO_TABLE
: {
7035 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
7037 ovs_assert(ctx
->table_id
< ogt
->table_id
);
7039 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
7040 ogt
->table_id
, true, true, false, last
,
7046 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
7050 compose_clone(ctx
, ofpact_get_CLONE(a
), last
);
7054 xlate_generic_encap_action(ctx
, ofpact_get_ENCAP(a
));
7057 case OFPACT_DECAP
: {
7058 bool recirc_needed
=
7059 xlate_generic_decap_action(ctx
, ofpact_get_DECAP(a
));
7060 if (!ctx
->error
&& recirc_needed
) {
7061 /* Recirculate for parsing of inner packet. */
7062 ctx_trigger_freeze(ctx
);
7063 /* Then continue with next action. */
7070 compose_conntrack_action(ctx
, ofpact_get_CT(a
), last
);
7073 case OFPACT_CT_CLEAR
:
7074 compose_ct_clear_action(ctx
);
7078 /* This will be processed by compose_conntrack_action(). */
7079 ctx
->ct_nat_action
= ofpact_get_NAT(a
);
7082 case OFPACT_DEBUG_RECIRC
:
7083 ctx_trigger_freeze(ctx
);
7087 case OFPACT_DEBUG_SLOW
:
7088 ctx
->xout
->slow
|= SLOW_ACTION
;
7091 case OFPACT_CHECK_PKT_LARGER
: {
7093 /* If this is last action, then there is no need to
7094 * translate the action. */
7097 const struct ofpact
*remaining_acts
= ofpact_next(a
);
7098 size_t remaining_acts_len
= ofpact_remaining_len(remaining_acts
,
7101 xlate_check_pkt_larger(ctx
, ofpact_get_CHECK_PKT_LARGER(a
),
7102 remaining_acts
, remaining_acts_len
);
7107 /* Check if need to store this and the remaining actions for later
7109 if (!ctx
->error
&& ctx
->exit
&& ctx_first_frozen_action(ctx
)) {
7110 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
), ctx
);
7117 xlate_in_init(struct xlate_in
*xin
, struct ofproto_dpif
*ofproto
,
7118 ovs_version_t version
, const struct flow
*flow
,
7119 ofp_port_t in_port
, struct rule_dpif
*rule
, uint16_t tcp_flags
,
7120 const struct dp_packet
*packet
, struct flow_wildcards
*wc
,
7121 struct ofpbuf
*odp_actions
)
7123 xin
->ofproto
= ofproto
;
7124 xin
->tables_version
= version
;
7126 xin
->upcall_flow
= flow
;
7127 xin
->flow
.in_port
.ofp_port
= in_port
;
7128 xin
->flow
.actset_output
= OFPP_UNSET
;
7129 xin
->packet
= packet
;
7130 xin
->allow_side_effects
= packet
!= NULL
;
7133 xin
->ofpacts
= NULL
;
7134 xin
->ofpacts_len
= 0;
7135 xin
->tcp_flags
= tcp_flags
;
7137 xin
->resubmit_stats
= NULL
;
7141 xin
->odp_actions
= odp_actions
;
7142 xin
->in_packet_out
= false;
7143 xin
->recirc_queue
= NULL
;
7144 xin
->xport_uuid
= UUID_ZERO
;
7146 /* Do recirc lookup. */
7147 xin
->frozen_state
= NULL
;
7148 if (flow
->recirc_id
) {
7149 const struct recirc_id_node
*node
7150 = recirc_id_node_find(flow
->recirc_id
);
7152 xin
->frozen_state
= &node
->state
;
7158 xlate_out_uninit(struct xlate_out
*xout
)
7161 recirc_refs_unref(&xout
->recircs
);
7165 static struct skb_priority_to_dscp
*
7166 get_skb_priority(const struct xport
*xport
, uint32_t skb_priority
)
7168 struct skb_priority_to_dscp
*pdscp
;
7171 hash
= hash_int(skb_priority
, 0);
7172 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &xport
->skb_priorities
) {
7173 if (pdscp
->skb_priority
== skb_priority
) {
7181 dscp_from_skb_priority(const struct xport
*xport
, uint32_t skb_priority
,
7184 struct skb_priority_to_dscp
*pdscp
= get_skb_priority(xport
, skb_priority
);
7185 *dscp
= pdscp
? pdscp
->dscp
: 0;
7186 return pdscp
!= NULL
;
7190 count_skb_priorities(const struct xport
*xport
)
7192 return hmap_count(&xport
->skb_priorities
);
7196 clear_skb_priorities(struct xport
*xport
)
7198 struct skb_priority_to_dscp
*pdscp
;
7200 HMAP_FOR_EACH_POP (pdscp
, hmap_node
, &xport
->skb_priorities
) {
7206 actions_output_to_local_port(const struct xlate_ctx
*ctx
)
7208 odp_port_t local_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
, OFPP_LOCAL
);
7209 const struct nlattr
*a
;
7212 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, ctx
->odp_actions
->data
,
7213 ctx
->odp_actions
->size
) {
7214 if (nl_attr_type(a
) == OVS_ACTION_ATTR_OUTPUT
7215 && nl_attr_get_odp_port(a
) == local_odp_port
) {
7222 #if defined(__linux__)
7223 /* Returns the maximum number of packets that the Linux kernel is willing to
7224 * queue up internally to certain kinds of software-implemented ports, or the
7225 * default (and rarely modified) value if it cannot be determined. */
7227 netdev_max_backlog(void)
7229 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
7230 static int max_backlog
= 1000; /* The normal default value. */
7232 if (ovsthread_once_start(&once
)) {
7233 static const char filename
[] = "/proc/sys/net/core/netdev_max_backlog";
7237 stream
= fopen(filename
, "r");
7239 VLOG_INFO("%s: open failed (%s)", filename
, ovs_strerror(errno
));
7241 if (fscanf(stream
, "%d", &n
) != 1) {
7242 VLOG_WARN("%s: read error", filename
);
7243 } else if (n
<= 100) {
7244 VLOG_WARN("%s: unexpectedly small value %d", filename
, n
);
7250 ovsthread_once_done(&once
);
7252 VLOG_DBG("%s: using %d max_backlog", filename
, max_backlog
);
7258 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
7261 count_output_actions(const struct ofpbuf
*odp_actions
)
7263 const struct nlattr
*a
;
7267 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, odp_actions
->data
, odp_actions
->size
) {
7268 if (a
->nla_type
== OVS_ACTION_ATTR_OUTPUT
) {
7274 #endif /* defined(__linux__) */
7276 /* Returns true if 'odp_actions' contains more output actions than the datapath
7277 * can reliably handle in one go. On Linux, this is the value of the
7278 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
7279 * packets that the kernel is willing to queue up for processing while the
7280 * datapath is processing a set of actions. */
7282 too_many_output_actions(const struct ofpbuf
*odp_actions OVS_UNUSED
)
7285 return (odp_actions
->size
/ NL_A_U32_SIZE
> netdev_max_backlog()
7286 && count_output_actions(odp_actions
) > netdev_max_backlog());
7288 /* OSes other than Linux might have similar limits, but we don't know how
7289 * to determine them.*/
7295 xlate_wc_init(struct xlate_ctx
*ctx
)
7297 flow_wildcards_init_catchall(ctx
->wc
);
7299 /* Some fields we consider to always be examined. */
7300 WC_MASK_FIELD(ctx
->wc
, packet_type
);
7301 WC_MASK_FIELD(ctx
->wc
, in_port
);
7302 WC_MASK_FIELD(ctx
->wc
, dl_type
);
7303 if (is_ip_any(&ctx
->xin
->flow
)) {
7304 WC_MASK_FIELD_MASK(ctx
->wc
, nw_frag
, FLOW_NW_FRAG_MASK
);
7307 if (ctx
->xbridge
->support
.odp
.recirc
) {
7308 /* Always exactly match recirc_id when datapath supports
7310 WC_MASK_FIELD(ctx
->wc
, recirc_id
);
7313 if (ctx
->xbridge
->netflow
) {
7314 netflow_mask_wc(&ctx
->xin
->flow
, ctx
->wc
);
7317 tnl_wc_init(&ctx
->xin
->flow
, ctx
->wc
);
7321 xlate_wc_finish(struct xlate_ctx
*ctx
)
7325 /* Clear the metadata and register wildcard masks, because we won't
7326 * use non-header fields as part of the cache. */
7327 flow_wildcards_clear_non_packet_fields(ctx
->wc
);
7329 /* Wildcard Ethernet address fields if the original packet type was not
7332 * (The Ethertype field is used even when the original packet type is not
7334 if (ctx
->xin
->upcall_flow
->packet_type
!= htonl(PT_ETH
)) {
7335 ctx
->wc
->masks
.dl_dst
= eth_addr_zero
;
7336 ctx
->wc
->masks
.dl_src
= eth_addr_zero
;
7339 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
7340 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
7341 * represent these fields. The datapath interface, on the other hand,
7342 * represents them with just 8 bits each. This means that if the high
7343 * 8 bits of the masks for these fields somehow become set, then they
7344 * will get chopped off by a round trip through the datapath, and
7345 * revalidation will spot that as an inconsistency and delete the flow.
7346 * Avoid the problem here by making sure that only the low 8 bits of
7347 * either field can be unwildcarded for ICMP.
7349 if (is_icmpv4(&ctx
->xin
->flow
, NULL
) || is_icmpv6(&ctx
->xin
->flow
, NULL
)) {
7350 ctx
->wc
->masks
.tp_src
&= htons(UINT8_MAX
);
7351 ctx
->wc
->masks
.tp_dst
&= htons(UINT8_MAX
);
7353 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
7354 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
7355 if (ctx
->wc
->masks
.vlans
[i
].tci
) {
7356 ctx
->wc
->masks
.vlans
[i
].tci
|= htons(VLAN_CFI
);
7360 /* The classifier might return masks that match on tp_src and tp_dst even
7361 * for later fragments. This happens because there might be flows that
7362 * match on tp_src or tp_dst without matching on the frag bits, because
7363 * it is not a prerequisite for OpenFlow. Since it is a prerequisite for
7364 * datapath flows and since tp_src and tp_dst are always going to be 0,
7365 * wildcard the fields here. */
7366 if (ctx
->xin
->flow
.nw_frag
& FLOW_NW_FRAG_LATER
) {
7367 ctx
->wc
->masks
.tp_src
= 0;
7368 ctx
->wc
->masks
.tp_dst
= 0;
7371 /* Clear flow wildcard bits for fields which are not present
7372 * in the original packet header. These wildcards may get set
7373 * due to push/set_field actions. This results into frequent
7374 * invalidation of datapath flows by revalidator thread. */
7376 /* Clear mpls label wc bits if original packet is non-mpls. */
7377 if (!eth_type_mpls(ctx
->xin
->upcall_flow
->dl_type
)) {
7378 for (i
= 0; i
< FLOW_MAX_MPLS_LABELS
; i
++) {
7379 ctx
->wc
->masks
.mpls_lse
[i
] = 0;
7382 /* Clear vlan header wc bits if original packet does not have
7384 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
7385 if (!eth_type_vlan(ctx
->xin
->upcall_flow
->vlans
[i
].tpid
)) {
7386 ctx
->wc
->masks
.vlans
[i
].tpid
= 0;
7387 ctx
->wc
->masks
.vlans
[i
].tci
= 0;
7392 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
7394 * The caller must take responsibility for eventually freeing 'xout', with
7395 * xlate_out_uninit().
7396 * Returns 'XLATE_OK' if translation was successful. In case of an error an
7397 * empty set of actions will be returned in 'xin->odp_actions' (if non-NULL),
7398 * so that most callers may ignore the return value and transparently install a
7399 * drop flow when the translation fails. */
7401 xlate_actions(struct xlate_in
*xin
, struct xlate_out
*xout
)
7403 *xout
= (struct xlate_out
) {
7405 .recircs
= RECIRC_REFS_EMPTY_INITIALIZER
,
7408 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7409 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, xin
->ofproto
);
7411 return XLATE_BRIDGE_NOT_FOUND
;
7414 struct flow
*flow
= &xin
->flow
;
7416 uint8_t stack_stub
[1024];
7417 uint64_t action_set_stub
[1024 / 8];
7418 uint64_t frozen_actions_stub
[1024 / 8];
7419 uint64_t actions_stub
[256 / 8];
7420 struct ofpbuf scratch_actions
= OFPBUF_STUB_INITIALIZER(actions_stub
);
7421 struct xlate_ctx ctx
= {
7425 .orig_tunnel_ipv6_dst
= flow_tnl_dst(&flow
->tunnel
),
7428 .stack
= OFPBUF_STUB_INITIALIZER(stack_stub
),
7432 : &(struct flow_wildcards
) { .masks
= { .dl_type
= 0 } }),
7433 .odp_actions
= xin
->odp_actions
? xin
->odp_actions
: &scratch_actions
,
7435 .depth
= xin
->depth
,
7436 .resubmits
= xin
->resubmits
,
7437 .in_action_set
= false,
7438 .in_packet_out
= xin
->in_packet_out
,
7439 .pending_encap
= false,
7440 .pending_decap
= false,
7444 .rule_cookie
= OVS_BE64_MAX
,
7445 .orig_skb_priority
= flow
->skb_priority
,
7446 .sflow_n_outputs
= 0,
7447 .sflow_odp_port
= 0,
7448 .nf_output_iface
= NF_OUT_DROP
,
7454 .recirc_update_dp_hash
= false,
7455 .frozen_actions
= OFPBUF_STUB_INITIALIZER(frozen_actions_stub
),
7459 .conntracked
= false,
7461 .ct_nat_action
= NULL
,
7463 .action_set_has_group
= false,
7464 .action_set
= OFPBUF_STUB_INITIALIZER(action_set_stub
),
7467 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
7468 * the packet as the datapath will treat it for output actions. Our
7469 * datapath doesn't retain tunneling information without us re-setting
7470 * it, so clear the tunnel data.
7473 memset(&ctx
.base_flow
.tunnel
, 0, sizeof ctx
.base_flow
.tunnel
);
7475 ofpbuf_reserve(ctx
.odp_actions
, NL_A_U32_SIZE
);
7476 xlate_wc_init(&ctx
);
7478 COVERAGE_INC(xlate_actions
);
7480 xin
->trace
= xlate_report(&ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
7482 if (xin
->frozen_state
) {
7483 const struct frozen_state
*state
= xin
->frozen_state
;
7485 struct ovs_list
*old_trace
= xin
->trace
;
7486 xin
->trace
= xlate_report(&ctx
, OFT_THAW
, "thaw");
7488 if (xin
->ofpacts_len
> 0 || ctx
.rule
) {
7489 xlate_report_error(&ctx
, "Recirculation conflict (%s)!",
7490 xin
->ofpacts_len
? "actions" : "rule");
7491 ctx
.error
= XLATE_RECIRCULATION_CONFLICT
;
7495 /* Set the bridge for post-recirculation processing if needed. */
7496 if (!uuid_equals(&ctx
.xbridge
->ofproto
->uuid
, &state
->ofproto_uuid
)) {
7497 const struct xbridge
*new_bridge
7498 = xbridge_lookup_by_uuid(xcfg
, &state
->ofproto_uuid
);
7500 if (OVS_UNLIKELY(!new_bridge
)) {
7501 /* Drop the packet if the bridge cannot be found. */
7502 xlate_report_error(&ctx
, "Frozen bridge no longer exists.");
7503 ctx
.error
= XLATE_BRIDGE_NOT_FOUND
;
7504 xin
->trace
= old_trace
;
7507 ctx
.xbridge
= new_bridge
;
7508 /* The bridge is now known so obtain its table version. */
7509 ctx
.xin
->tables_version
7510 = ofproto_dpif_get_tables_version(ctx
.xbridge
->ofproto
);
7513 /* Set the thawed table id. Note: A table lookup is done only if there
7514 * are no frozen actions. */
7515 ctx
.table_id
= state
->table_id
;
7516 xlate_report(&ctx
, OFT_THAW
,
7517 "Resuming from table %"PRIu8
, ctx
.table_id
);
7519 ctx
.conntracked
= state
->conntracked
;
7520 if (!state
->conntracked
) {
7521 clear_conntrack(&ctx
);
7524 /* Restore pipeline metadata. May change flow's in_port and other
7525 * metadata to the values that existed when freezing was triggered. */
7526 frozen_metadata_to_flow(&state
->metadata
, flow
);
7528 /* Restore stack, if any. */
7530 ofpbuf_put(&ctx
.stack
, state
->stack
, state
->stack_size
);
7533 /* Restore mirror state. */
7534 ctx
.mirrors
= state
->mirrors
;
7536 /* Restore action set, if any. */
7537 if (state
->action_set_len
) {
7538 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring action set",
7539 state
->action_set
, state
->action_set_len
);
7541 flow
->actset_output
= OFPP_UNSET
;
7542 xlate_write_actions__(&ctx
, state
->action_set
,
7543 state
->action_set_len
);
7546 /* Restore frozen actions. If there are no actions, processing will
7547 * start with a lookup in the table set above. */
7548 xin
->ofpacts
= state
->ofpacts
;
7549 xin
->ofpacts_len
= state
->ofpacts_len
;
7550 if (state
->ofpacts_len
) {
7551 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring actions",
7552 xin
->ofpacts
, xin
->ofpacts_len
);
7555 xin
->trace
= old_trace
;
7556 } else if (OVS_UNLIKELY(flow
->recirc_id
)) {
7557 xlate_report_error(&ctx
,
7558 "Recirculation context not found for ID %"PRIx32
,
7560 ctx
.error
= XLATE_NO_RECIRCULATION_CONTEXT
;
7564 /* Tunnel metadata in udpif format must be normalized before translation. */
7565 if (flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7566 const struct tun_table
*tun_tab
= ofproto_get_tun_tab(
7567 &ctx
.xbridge
->ofproto
->up
);
7570 err
= tun_metadata_from_geneve_udpif(tun_tab
, &xin
->upcall_flow
->tunnel
,
7571 &xin
->upcall_flow
->tunnel
,
7574 xlate_report_error(&ctx
, "Invalid Geneve tunnel metadata");
7575 ctx
.error
= XLATE_INVALID_TUNNEL_METADATA
;
7578 } else if (!flow
->tunnel
.metadata
.tab
|| xin
->frozen_state
) {
7579 /* If the original flow did not come in on a tunnel, then it won't have
7580 * FLOW_TNL_F_UDPIF set. However, we still need to have a metadata
7581 * table in case we generate tunnel actions. */
7582 /* If the translation is from a frozen state, we use the latest
7583 * TLV map to avoid segmentation fault in case the old TLV map is
7584 * replaced by a new one.
7585 * XXX: It is better to abort translation if the table is changed. */
7586 flow
->tunnel
.metadata
.tab
= ofproto_get_tun_tab(
7587 &ctx
.xbridge
->ofproto
->up
);
7589 ctx
.wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
7591 /* Get the proximate input port of the packet. (If xin->frozen_state,
7592 * flow->in_port is the ultimate input port of the packet.) */
7593 struct xport
*in_port
= get_ofp_port(xbridge
,
7594 ctx
.base_flow
.in_port
.ofp_port
);
7595 if (in_port
&& !in_port
->peer
) {
7596 ctx
.xin
->xport_uuid
= in_port
->uuid
;
7599 if (flow
->packet_type
!= htonl(PT_ETH
) && in_port
&&
7600 in_port
->pt_mode
== NETDEV_PT_LEGACY_L3
&& ctx
.table_id
== 0) {
7601 /* Add dummy Ethernet header to non-L2 packet if it's coming from a
7602 * L3 port. So all packets will be L2 packets for lookup.
7603 * The dl_type has already been set from the packet_type. */
7604 flow
->packet_type
= htonl(PT_ETH
);
7605 flow
->dl_src
= eth_addr_zero
;
7606 flow
->dl_dst
= eth_addr_zero
;
7607 ctx
.pending_encap
= true;
7610 if (!xin
->ofpacts
&& !ctx
.rule
) {
7611 ctx
.rule
= rule_dpif_lookup_from_table(
7612 ctx
.xbridge
->ofproto
, ctx
.xin
->tables_version
, flow
, ctx
.wc
,
7613 ctx
.xin
->resubmit_stats
, &ctx
.table_id
,
7614 flow
->in_port
.ofp_port
, true, true, ctx
.xin
->xcache
);
7615 if (ctx
.xin
->resubmit_stats
) {
7616 rule_dpif_credit_stats(ctx
.rule
, ctx
.xin
->resubmit_stats
, false);
7618 if (ctx
.xin
->xcache
) {
7619 struct xc_entry
*entry
;
7621 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_RULE
);
7622 entry
->rule
= ctx
.rule
;
7623 ofproto_rule_ref(&ctx
.rule
->up
);
7626 xlate_report_table(&ctx
, ctx
.rule
, ctx
.table_id
);
7629 /* Tunnel stats only for not-thawed packets. */
7630 if (!xin
->frozen_state
&& in_port
&& in_port
->is_tunnel
) {
7631 if (ctx
.xin
->resubmit_stats
) {
7632 netdev_vport_inc_rx(in_port
->netdev
, ctx
.xin
->resubmit_stats
);
7634 bfd_account_rx(in_port
->bfd
, ctx
.xin
->resubmit_stats
);
7637 if (ctx
.xin
->xcache
) {
7638 struct xc_entry
*entry
;
7640 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETDEV
);
7641 entry
->dev
.rx
= netdev_ref(in_port
->netdev
);
7642 entry
->dev
.bfd
= bfd_ref(in_port
->bfd
);
7646 if (!xin
->frozen_state
&& process_special(&ctx
, in_port
)) {
7647 /* process_special() did all the processing for this packet.
7649 * We do not perform special processing on thawed packets, since that
7650 * was done before they were frozen and should not be redone. */
7651 mirror_ingress_packet(&ctx
);
7652 } else if (in_port
&& in_port
->xbundle
7653 && xbundle_mirror_out(xbridge
, in_port
->xbundle
)) {
7654 xlate_report_error(&ctx
, "dropping packet received on port "
7655 "%s, which is reserved exclusively for mirroring",
7656 in_port
->xbundle
->name
);
7658 /* Sampling is done on initial reception; don't redo after thawing. */
7659 unsigned int user_cookie_offset
= 0;
7660 if (!xin
->frozen_state
) {
7661 user_cookie_offset
= compose_sflow_action(&ctx
);
7662 compose_ipfix_action(&ctx
, ODPP_NONE
);
7664 size_t sample_actions_len
= ctx
.odp_actions
->size
;
7665 bool ecn_drop
= !tnl_process_ecn(flow
);
7668 && (!in_port
|| may_receive(in_port
, &ctx
))) {
7669 const struct ofpact
*ofpacts
;
7673 ofpacts
= xin
->ofpacts
;
7674 ofpacts_len
= xin
->ofpacts_len
;
7675 } else if (ctx
.rule
) {
7676 const struct rule_actions
*actions
7677 = rule_get_actions(&ctx
.rule
->up
);
7678 ofpacts
= actions
->ofpacts
;
7679 ofpacts_len
= actions
->ofpacts_len
;
7680 ctx
.rule_cookie
= ctx
.rule
->up
.flow_cookie
;
7685 mirror_ingress_packet(&ctx
);
7686 do_xlate_actions(ofpacts
, ofpacts_len
, &ctx
, true, false);
7691 /* We've let OFPP_NORMAL and the learning action look at the
7692 * packet, so cancel all actions and freezing if forwarding is
7694 if (in_port
&& (!xport_stp_forward_state(in_port
) ||
7695 !xport_rstp_forward_state(in_port
))) {
7696 ctx
.odp_actions
->size
= sample_actions_len
;
7697 ctx_cancel_freeze(&ctx
);
7698 ofpbuf_clear(&ctx
.action_set
);
7699 ctx
.error
= XLATE_FORWARDING_DISABLED
;
7702 if (!ctx
.freezing
) {
7703 xlate_action_set(&ctx
);
7706 finish_freezing(&ctx
);
7708 } else if (ecn_drop
) {
7709 ctx
.error
= XLATE_CONGESTION_DROP
;
7712 /* Output only fully processed packets. */
7714 && xbridge
->has_in_band
7715 && in_band_must_output_to_local_port(flow
)
7716 && !actions_output_to_local_port(&ctx
)) {
7717 WC_MASK_FIELD(ctx
.wc
, nw_proto
);
7718 WC_MASK_FIELD(ctx
.wc
, tp_src
);
7719 WC_MASK_FIELD(ctx
.wc
, tp_dst
);
7720 WC_MASK_FIELD(ctx
.wc
, dl_type
);
7721 xlate_report(&ctx
, OFT_DETAIL
, "outputting DHCP packet "
7722 "to local port for in-band control");
7723 compose_output_action(&ctx
, OFPP_LOCAL
, NULL
, false, false);
7726 if (user_cookie_offset
) {
7727 fix_sflow_action(&ctx
, user_cookie_offset
);
7731 if (nl_attr_oversized(ctx
.odp_actions
->size
)) {
7732 /* These datapath actions are too big for a Netlink attribute, so we
7733 * can't hand them to the kernel directly. dpif_execute() can execute
7734 * them one by one with help, so just mark the result as SLOW_ACTION to
7735 * prevent the flow from being installed. */
7736 COVERAGE_INC(xlate_actions_oversize
);
7737 ctx
.xout
->slow
|= SLOW_ACTION
;
7738 } else if (too_many_output_actions(ctx
.odp_actions
)) {
7739 COVERAGE_INC(xlate_actions_too_many_output
);
7740 ctx
.xout
->slow
|= SLOW_ACTION
;
7743 /* Update NetFlow for non-frozen traffic. */
7744 if (xbridge
->netflow
&& !xin
->frozen_state
) {
7745 if (ctx
.xin
->resubmit_stats
) {
7746 netflow_flow_update(xbridge
->netflow
, flow
,
7747 ctx
.nf_output_iface
,
7748 ctx
.xin
->resubmit_stats
);
7750 if (ctx
.xin
->xcache
) {
7751 struct xc_entry
*entry
;
7753 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETFLOW
);
7754 entry
->nf
.netflow
= netflow_ref(xbridge
->netflow
);
7755 entry
->nf
.flow
= xmemdup(flow
, sizeof *flow
);
7756 entry
->nf
.iface
= ctx
.nf_output_iface
;
7760 /* Translate tunnel metadata masks to udpif format if necessary. */
7761 if (xin
->upcall_flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7762 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7763 const struct flow_tnl
*upcall_tnl
= &xin
->upcall_flow
->tunnel
;
7764 struct geneve_opt opts
[TLV_TOT_OPT_SIZE
/
7765 sizeof(struct geneve_opt
)];
7767 tun_metadata_to_geneve_udpif_mask(&flow
->tunnel
,
7768 &ctx
.wc
->masks
.tunnel
,
7769 upcall_tnl
->metadata
.opts
.gnv
,
7770 upcall_tnl
->metadata
.present
.len
,
7772 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7773 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7774 memcpy(&ctx
.wc
->masks
.tunnel
.metadata
.opts
.gnv
, opts
,
7775 upcall_tnl
->metadata
.present
.len
);
7777 ctx
.wc
->masks
.tunnel
.metadata
.present
.len
= 0xff;
7778 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7779 ctx
.wc
->masks
.tunnel
.flags
|= FLOW_TNL_F_UDPIF
;
7780 } else if (!xin
->upcall_flow
->tunnel
.metadata
.tab
) {
7781 /* If we didn't have options in UDPIF format and didn't have an existing
7782 * metadata table, then it means that there were no options at all when
7783 * we started processing and any wildcards we picked up were from
7784 * action generation. Without options on the incoming packet, wildcards
7785 * aren't meaningful. To avoid them possibly getting misinterpreted,
7786 * just clear everything. */
7787 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7788 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7789 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7791 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7795 xlate_wc_finish(&ctx
);
7798 /* Reset the table to what it was when we came in. If we only fetched
7799 * it locally, then it has no meaning outside of flow translation. */
7800 flow
->tunnel
.metadata
.tab
= xin
->upcall_flow
->tunnel
.metadata
.tab
;
7802 ofpbuf_uninit(&ctx
.stack
);
7803 ofpbuf_uninit(&ctx
.action_set
);
7804 ofpbuf_uninit(&ctx
.frozen_actions
);
7805 ofpbuf_uninit(&scratch_actions
);
7806 ofpbuf_delete(ctx
.encap_data
);
7808 /* Make sure we return a "drop flow" in case of an error. */
7811 if (xin
->odp_actions
) {
7812 ofpbuf_clear(xin
->odp_actions
);
7816 /* Install drop action if datapath supports explicit drop action. */
7817 if (xin
->odp_actions
&& !xin
->odp_actions
->size
&&
7818 ovs_explicit_drop_action_supported(ctx
.xbridge
->ofproto
)) {
7819 put_drop_action(xin
->odp_actions
, ctx
.error
);
7822 /* Since congestion drop and forwarding drop are not exactly
7823 * translation error, we are resetting the translation error.
7825 if (ctx
.error
== XLATE_CONGESTION_DROP
||
7826 ctx
.error
== XLATE_FORWARDING_DISABLED
) {
7827 ctx
.error
= XLATE_OK
;
7834 xlate_resume(struct ofproto_dpif
*ofproto
,
7835 const struct ofputil_packet_in_private
*pin
,
7836 struct ofpbuf
*odp_actions
,
7837 enum slow_path_reason
*slow
,
7839 struct xlate_cache
*xcache
)
7841 struct dp_packet packet
;
7842 dp_packet_use_const(&packet
, pin
->base
.packet
,
7843 pin
->base
.packet_len
);
7845 pkt_metadata_from_flow(&packet
.md
, &pin
->base
.flow_metadata
.flow
);
7846 flow_extract(&packet
, flow
);
7848 struct xlate_in xin
;
7849 xlate_in_init(&xin
, ofproto
, ofproto_dpif_get_tables_version(ofproto
),
7850 flow
, 0, NULL
, ntohs(flow
->tcp_flags
),
7851 &packet
, NULL
, odp_actions
);
7852 xin
.xcache
= xcache
;
7854 struct ofpact_note noop
;
7855 ofpact_init_NOTE(&noop
);
7858 bool any_actions
= pin
->actions_len
> 0;
7859 struct frozen_state state
= {
7860 .table_id
= 0, /* Not the table where NXAST_PAUSE was executed. */
7861 .ofproto_uuid
= pin
->bridge
,
7862 .stack
= pin
->stack
,
7863 .stack_size
= pin
->stack_size
,
7864 .mirrors
= pin
->mirrors
,
7865 .conntracked
= pin
->conntracked
,
7866 .xport_uuid
= UUID_ZERO
,
7868 /* When there are no actions, xlate_actions() will search the flow
7869 * table. We don't want it to do that (we want it to resume), so
7870 * supply a no-op action if there aren't any.
7872 * (We can't necessarily avoid translating actions entirely if there
7873 * aren't any actions, because there might be some finishing-up to do
7874 * at the end of the pipeline, and we don't check for those
7876 .ofpacts
= any_actions
? pin
->actions
: &noop
.ofpact
,
7877 .ofpacts_len
= any_actions
? pin
->actions_len
: sizeof noop
,
7879 .action_set
= pin
->action_set
,
7880 .action_set_len
= pin
->action_set_len
,
7882 frozen_metadata_from_flow(&state
.metadata
,
7883 &pin
->base
.flow_metadata
.flow
);
7884 xin
.frozen_state
= &state
;
7886 struct xlate_out xout
;
7887 enum xlate_error error
= xlate_actions(&xin
, &xout
);
7889 xlate_out_uninit(&xout
);
7891 /* xlate_actions() can generate a number of errors, but only
7892 * XLATE_BRIDGE_NOT_FOUND really stands out to me as one that we should be
7893 * sure to report over OpenFlow. The others could come up in packet-outs
7894 * or regular flow translation and I don't think that it's going to be too
7895 * useful to report them to the controller. */
7896 return error
== XLATE_BRIDGE_NOT_FOUND
? OFPERR_NXR_STALE
: 0;
7899 /* Sends 'packet' out 'ofport'. If 'port' is a tunnel and that tunnel type
7900 * supports a notion of an OAM flag, sets it if 'oam' is true.
7901 * May modify 'packet'.
7902 * Returns 0 if successful, otherwise a positive errno value. */
7904 xlate_send_packet(const struct ofport_dpif
*ofport
, bool oam
,
7905 struct dp_packet
*packet
)
7907 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7908 struct xport
*xport
;
7909 uint64_t ofpacts_stub
[1024 / 8];
7910 struct ofpbuf ofpacts
;
7913 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
7914 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
7915 flow_extract(packet
, &flow
);
7916 flow
.in_port
.ofp_port
= OFPP_NONE
;
7918 xport
= xport_lookup(xcfg
, ofport
);
7924 const ovs_be16 flag
= htons(NX_TUN_FLAG_OAM
);
7925 ofpact_put_set_field(&ofpacts
, mf_from_id(MFF_TUN_FLAGS
),
7929 ofpact_put_OUTPUT(&ofpacts
)->port
= xport
->ofp_port
;
7931 /* Actions here are not referring to anything versionable (flow tables or
7932 * groups) so we don't need to worry about the version here. */
7933 return ofproto_dpif_execute_actions(xport
->xbridge
->ofproto
,
7934 OVS_VERSION_MAX
, &flow
, NULL
,
7935 ofpacts
.data
, ofpacts
.size
, packet
);
7939 xlate_mac_learning_update(const struct ofproto_dpif
*ofproto
,
7940 ofp_port_t in_port
, struct eth_addr dl_src
,
7941 int vlan
, bool is_grat_arp
)
7943 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7944 struct xbridge
*xbridge
;
7945 struct xbundle
*xbundle
;
7947 xbridge
= xbridge_lookup(xcfg
, ofproto
);
7952 xbundle
= lookup_input_bundle__(xbridge
, in_port
, NULL
);
7957 update_learning_table__(xbridge
, xbundle
, dl_src
, vlan
, is_grat_arp
);
7961 xlate_set_support(const struct ofproto_dpif
*ofproto
,
7962 const struct dpif_backer_support
*support
)
7964 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7965 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, ofproto
);
7968 xbridge
->support
= *support
;