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";
449 return "Unknown error";
452 static void xlate_action_set(struct xlate_ctx
*ctx
);
453 static void xlate_commit_actions(struct xlate_ctx
*ctx
);
456 patch_port_output(struct xlate_ctx
*ctx
, const struct xport
*in_dev
,
457 struct xport
*out_dev
);
460 ctx_trigger_freeze(struct xlate_ctx
*ctx
)
463 ctx
->freezing
= true;
467 ctx_trigger_recirculate_with_hash(struct xlate_ctx
*ctx
, uint32_t type
,
471 ctx
->freezing
= true;
472 ctx
->recirc_update_dp_hash
= true;
473 ctx
->dp_hash_alg
= type
;
474 ctx
->dp_hash_basis
= basis
;
478 ctx_first_frozen_action(const struct xlate_ctx
*ctx
)
480 return !ctx
->frozen_actions
.size
;
484 ctx_cancel_freeze(struct xlate_ctx
*ctx
)
487 ctx
->freezing
= false;
488 ctx
->recirc_update_dp_hash
= false;
489 ofpbuf_clear(&ctx
->frozen_actions
);
490 ctx
->frozen_actions
.header
= NULL
;
495 static void finish_freezing(struct xlate_ctx
*ctx
);
497 /* A controller may use OFPP_NONE as the ingress port to indicate that
498 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
499 * when an input bundle is needed for validation (e.g., mirroring or
500 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
501 * any 'port' structs, so care must be taken when dealing with it. */
502 static struct xbundle ofpp_none_bundle
= {
504 .vlan_mode
= PORT_VLAN_TRUNK
507 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
508 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
509 * traffic egressing the 'ofport' with that priority should be marked with. */
510 struct skb_priority_to_dscp
{
511 struct hmap_node hmap_node
; /* Node in 'ofport_dpif''s 'skb_priorities'. */
512 uint32_t skb_priority
; /* Priority of this queue (see struct flow). */
514 uint8_t dscp
; /* DSCP bits to mark outgoing traffic with. */
517 /* Xlate config contains hash maps of all bridges, bundles and ports.
518 * Xcfgp contains the pointer to the current xlate configuration.
519 * When the main thread needs to change the configuration, it copies xcfgp to
520 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
521 * does not block handler and revalidator threads. */
523 struct hmap xbridges
;
524 struct hmap xbundles
;
526 struct hmap xports_uuid
;
528 static OVSRCU_TYPE(struct xlate_cfg
*) xcfgp
= OVSRCU_INITIALIZER(NULL
);
529 static struct xlate_cfg
*new_xcfg
= NULL
;
531 typedef void xlate_actions_handler(const struct ofpact
*, size_t ofpacts_len
,
532 struct xlate_ctx
*, bool, bool);
533 static bool may_receive(const struct xport
*, struct xlate_ctx
*);
534 static void do_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
535 struct xlate_ctx
*, bool, bool);
536 static void clone_xlate_actions(const struct ofpact
*, size_t ofpacts_len
,
537 struct xlate_ctx
*, bool, bool);
538 static void xlate_normal(struct xlate_ctx
*);
539 static void xlate_normal_flood(struct xlate_ctx
*ct
,
540 struct xbundle
*in_xbundle
, struct xvlan
*);
541 static void xlate_table_action(struct xlate_ctx
*, ofp_port_t in_port
,
542 uint8_t table_id
, bool may_packet_in
,
543 bool honor_table_miss
, bool with_ct_orig
,
544 bool is_last_action
, xlate_actions_handler
*);
546 static bool input_vid_is_valid(const struct xlate_ctx
*,
547 uint16_t vid
, struct xbundle
*);
548 static void xvlan_copy(struct xvlan
*dst
, const struct xvlan
*src
);
549 static void xvlan_pop(struct xvlan
*src
);
550 static void xvlan_push_uninit(struct xvlan
*src
);
551 static void xvlan_extract(const struct flow
*, struct xvlan
*);
552 static void xvlan_put(struct flow
*, const struct xvlan
*,
553 enum port_priority_tags_mode
);
554 static void xvlan_input_translate(const struct xbundle
*,
555 const struct xvlan
*in
,
556 struct xvlan
*xvlan
);
557 static void xvlan_output_translate(const struct xbundle
*,
558 const struct xvlan
*xvlan
,
560 static void output_normal(struct xlate_ctx
*, const struct xbundle
*,
561 const struct xvlan
*);
563 /* Optional bond recirculation parameter to compose_output_action(). */
564 struct xlate_bond_recirc
{
565 uint32_t recirc_id
; /* !0 Use recirculation instead of output. */
566 uint8_t hash_alg
; /* !0 Compute hash for recirc before. */
567 uint32_t hash_basis
; /* Compute hash for recirc before. */
570 static void compose_output_action(struct xlate_ctx
*, ofp_port_t ofp_port
,
571 const struct xlate_bond_recirc
*xr
,
572 bool is_last_action
, bool truncate
);
574 static struct xbridge
*xbridge_lookup(struct xlate_cfg
*,
575 const struct ofproto_dpif
*);
576 static struct xbridge
*xbridge_lookup_by_uuid(struct xlate_cfg
*,
577 const struct uuid
*);
578 static struct xbundle
*xbundle_lookup(struct xlate_cfg
*,
579 const struct ofbundle
*);
580 static struct xport
*xport_lookup(struct xlate_cfg
*,
581 const struct ofport_dpif
*);
582 static struct xport
*xport_lookup_by_uuid(struct xlate_cfg
*,
583 const struct uuid
*);
584 static struct xport
*get_ofp_port(const struct xbridge
*, ofp_port_t ofp_port
);
585 static struct skb_priority_to_dscp
*get_skb_priority(const struct xport
*,
586 uint32_t skb_priority
);
587 static void clear_skb_priorities(struct xport
*);
588 static size_t count_skb_priorities(const struct xport
*);
589 static bool dscp_from_skb_priority(const struct xport
*, uint32_t skb_priority
,
592 static void xlate_xbridge_init(struct xlate_cfg
*, struct xbridge
*);
593 static void xlate_xbundle_init(struct xlate_cfg
*, struct xbundle
*);
594 static void xlate_xport_init(struct xlate_cfg
*, struct xport
*);
595 static void xlate_xbridge_set(struct xbridge
*, struct dpif
*,
596 const struct mac_learning
*, struct stp
*,
597 struct rstp
*, const struct mcast_snooping
*,
598 const struct mbridge
*,
599 const struct dpif_sflow
*,
600 const struct dpif_ipfix
*,
601 const struct netflow
*,
602 bool forward_bpdu
, bool has_in_band
,
603 const struct dpif_backer_support
*,
604 const struct xbridge_addr
*);
605 static void xlate_xbundle_set(struct xbundle
*xbundle
,
606 enum port_vlan_mode vlan_mode
,
607 uint16_t qinq_ethtype
, int vlan
,
608 unsigned long *trunks
, unsigned long *cvlans
,
609 enum port_priority_tags_mode
,
610 const struct bond
*bond
, const struct lacp
*lacp
,
611 bool floodable
, bool protected);
612 static void xlate_xport_set(struct xport
*xport
, odp_port_t odp_port
,
613 const struct netdev
*netdev
, const struct cfm
*cfm
,
614 const struct bfd
*bfd
, const struct lldp
*lldp
,
615 int stp_port_no
, const struct rstp_port
*rstp_port
,
616 enum ofputil_port_config config
,
617 enum ofputil_port_state state
, bool is_tunnel
,
619 static void xlate_xbridge_remove(struct xlate_cfg
*, struct xbridge
*);
620 static void xlate_xbundle_remove(struct xlate_cfg
*, struct xbundle
*);
621 static void xlate_xport_remove(struct xlate_cfg
*, struct xport
*);
622 static void xlate_xbridge_copy(struct xbridge
*);
623 static void xlate_xbundle_copy(struct xbridge
*, struct xbundle
*);
624 static void xlate_xport_copy(struct xbridge
*, struct xbundle
*,
626 static void xlate_xcfg_free(struct xlate_cfg
*);
628 /* Tracing helpers. */
630 /* If tracing is enabled in 'ctx', creates a new trace node and appends it to
631 * the list of nodes maintained in ctx->xin. The new node has type 'type' and
632 * its text is created from 'format' by treating it as a printf format string.
633 * Returns the list of nodes embedded within the new trace node; ordinarily,
634 * the calleer can ignore this, but it is useful if the caller needs to nest
635 * more trace nodes within the new node.
637 * If tracing is not enabled, does nothing and returns NULL. */
638 static struct ovs_list
* OVS_PRINTF_FORMAT(3, 4)
639 xlate_report(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
640 const char *format
, ...)
642 struct ovs_list
*subtrace
= NULL
;
643 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
645 va_start(args
, format
);
646 char *text
= xvasprintf(format
, args
);
647 subtrace
= &oftrace_report(ctx
->xin
->trace
, type
, text
)->subs
;
654 /* This is like xlate_report() for errors that are serious enough that we
655 * should log them even if we are not tracing. */
656 static void OVS_PRINTF_FORMAT(2, 3)
657 xlate_report_error(const struct xlate_ctx
*ctx
, const char *format
, ...)
659 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
660 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
661 && (!ctx
->xin
->packet
|| VLOG_DROP_WARN(&rl
))) {
665 struct ds s
= DS_EMPTY_INITIALIZER
;
667 va_start(args
, format
);
668 ds_put_format_valist(&s
, format
, args
);
671 if (ctx
->xin
->trace
) {
672 oftrace_report(ctx
->xin
->trace
, OFT_ERROR
, ds_cstr(&s
));
674 ds_put_format(&s
, " on bridge %s while processing ",
676 flow_format(&s
, &ctx
->base_flow
, NULL
);
677 VLOG_WARN("%s", ds_cstr(&s
));
682 /* This is like xlate_report() for messages that should be logged
683 at the info level (even when not tracing). */
684 static void OVS_PRINTF_FORMAT(2, 3)
685 xlate_report_info(const struct xlate_ctx
*ctx
, const char *format
, ...)
687 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
688 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
689 && (!ctx
->xin
->packet
|| VLOG_DROP_INFO(&rl
))) {
693 struct ds s
= DS_EMPTY_INITIALIZER
;
695 va_start(args
, format
);
696 ds_put_format_valist(&s
, format
, args
);
699 if (ctx
->xin
->trace
) {
700 oftrace_report(ctx
->xin
->trace
, OFT_WARN
, ds_cstr(&s
));
702 ds_put_format(&s
, " on bridge %s while processing ",
704 flow_format(&s
, &ctx
->base_flow
, NULL
);
705 VLOG_INFO("%s", ds_cstr(&s
));
710 /* This is like xlate_report() for messages that should be logged at debug
711 * level (even if we are not tracing) because they can be valuable for
713 static void OVS_PRINTF_FORMAT(3, 4)
714 xlate_report_debug(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
715 const char *format
, ...)
717 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
718 if (!OVS_UNLIKELY(ctx
->xin
->trace
)
719 && (!ctx
->xin
->packet
|| VLOG_DROP_DBG(&rl
))) {
723 struct ds s
= DS_EMPTY_INITIALIZER
;
725 va_start(args
, format
);
726 ds_put_format_valist(&s
, format
, args
);
729 if (ctx
->xin
->trace
) {
730 oftrace_report(ctx
->xin
->trace
, type
, ds_cstr(&s
));
732 VLOG_DBG("bridge %s: %s", ctx
->xbridge
->name
, ds_cstr(&s
));
737 /* If tracing is enabled in 'ctx', appends a node of the given 'type' to the
738 * trace, whose text is 'title' followed by a formatted version of the
739 * 'ofpacts_len' OpenFlow actions in 'ofpacts'.
741 * If tracing is not enabled, does nothing. */
743 xlate_report_actions(const struct xlate_ctx
*ctx
, enum oftrace_node_type type
,
745 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
747 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
748 struct ds s
= DS_EMPTY_INITIALIZER
;
749 ds_put_format(&s
, "%s: ", title
);
750 struct ofpact_format_params fp
= { .s
= &s
};
751 ofpacts_format(ofpacts
, ofpacts_len
, &fp
);
752 oftrace_report(ctx
->xin
->trace
, type
, ds_cstr(&s
));
757 /* If tracing is enabled in 'ctx', appends a node of type OFT_DETAIL to the
758 * trace, whose the message is a formatted version of the OpenFlow action set.
759 * 'verb' should be "was" or "is", depending on whether the action set reported
760 * is the new action set or the old one.
762 * If tracing is not enabled, does nothing. */
764 xlate_report_action_set(const struct xlate_ctx
*ctx
, const char *verb
)
766 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
767 struct ofpbuf action_list
;
768 ofpbuf_init(&action_list
, 0);
769 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
770 if (action_list
.size
) {
771 struct ds s
= DS_EMPTY_INITIALIZER
;
772 struct ofpact_format_params fp
= { .s
= &s
};
773 ofpacts_format(action_list
.data
, action_list
.size
, &fp
);
774 xlate_report(ctx
, OFT_DETAIL
, "action set %s: %s",
778 xlate_report(ctx
, OFT_DETAIL
, "action set %s empty", verb
);
780 ofpbuf_uninit(&action_list
);
785 /* If tracing is enabled in 'ctx', appends a node representing 'rule' (in
786 * OpenFlow table 'table_id') to the trace and makes this node the parent for
787 * future trace nodes. The caller should save ctx->xin->trace before calling
788 * this function, then after tracing all of the activities under the table,
789 * restore its previous value.
791 * If tracing is not enabled, does nothing. */
793 xlate_report_table(const struct xlate_ctx
*ctx
, struct rule_dpif
*rule
,
796 if (OVS_LIKELY(!ctx
->xin
->trace
)) {
800 struct ds s
= DS_EMPTY_INITIALIZER
;
801 ds_put_format(&s
, "%2d. ", table_id
);
802 if (rule
== ctx
->xin
->ofproto
->miss_rule
) {
803 ds_put_cstr(&s
, "No match, and a \"packet-in\" is called for.");
804 } else if (rule
== ctx
->xin
->ofproto
->no_packet_in_rule
) {
805 ds_put_cstr(&s
, "No match.");
806 } else if (rule
== ctx
->xin
->ofproto
->drop_frags_rule
) {
807 ds_put_cstr(&s
, "Packets are IP fragments and "
808 "the fragment handling mode is \"drop\".");
810 minimatch_format(&rule
->up
.cr
.match
,
811 ofproto_get_tun_tab(&ctx
->xin
->ofproto
->up
),
812 NULL
, &s
, OFP_DEFAULT_PRIORITY
);
813 if (ds_last(&s
) != ' ') {
814 ds_put_cstr(&s
, ", ");
816 ds_put_format(&s
, "priority %d", rule
->up
.cr
.priority
);
817 if (rule
->up
.flow_cookie
) {
818 ds_put_format(&s
, ", cookie %#"PRIx64
,
819 ntohll(rule
->up
.flow_cookie
));
822 ctx
->xin
->trace
= &oftrace_report(ctx
->xin
->trace
, OFT_TABLE
,
827 /* If tracing is enabled in 'ctx', adds an OFT_DETAIL trace node to 'ctx'
828 * reporting the value of subfield 'sf'.
830 * If tracing is not enabled, does nothing. */
832 xlate_report_subfield(const struct xlate_ctx
*ctx
,
833 const struct mf_subfield
*sf
)
835 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
836 struct ds s
= DS_EMPTY_INITIALIZER
;
837 mf_format_subfield(sf
, &s
);
838 ds_put_cstr(&s
, " is now ");
840 if (sf
->ofs
== 0 && sf
->n_bits
>= sf
->field
->n_bits
) {
841 union mf_value value
;
842 mf_get_value(sf
->field
, &ctx
->xin
->flow
, &value
);
843 mf_format(sf
->field
, &value
, NULL
, NULL
, &s
);
845 union mf_subvalue cst
;
846 mf_read_subfield(sf
, &ctx
->xin
->flow
, &cst
);
847 ds_put_hex(&s
, &cst
, sizeof cst
);
850 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
857 xlate_xbridge_init(struct xlate_cfg
*xcfg
, struct xbridge
*xbridge
)
859 ovs_list_init(&xbridge
->xbundles
);
860 hmap_init(&xbridge
->xports
);
861 hmap_insert(&xcfg
->xbridges
, &xbridge
->hmap_node
,
862 hash_pointer(xbridge
->ofproto
, 0));
866 xlate_xbundle_init(struct xlate_cfg
*xcfg
, struct xbundle
*xbundle
)
868 ovs_list_init(&xbundle
->xports
);
869 ovs_list_insert(&xbundle
->xbridge
->xbundles
, &xbundle
->list_node
);
870 hmap_insert(&xcfg
->xbundles
, &xbundle
->hmap_node
,
871 hash_pointer(xbundle
->ofbundle
, 0));
875 xlate_xport_init(struct xlate_cfg
*xcfg
, struct xport
*xport
)
877 hmap_init(&xport
->skb_priorities
);
878 hmap_insert(&xcfg
->xports
, &xport
->hmap_node
,
879 hash_pointer(xport
->ofport
, 0));
880 hmap_insert(&xport
->xbridge
->xports
, &xport
->ofp_node
,
881 hash_ofp_port(xport
->ofp_port
));
882 hmap_insert(&xcfg
->xports_uuid
, &xport
->uuid_node
,
883 uuid_hash(&xport
->uuid
));
886 static struct xbridge_addr
*
887 xbridge_addr_create(struct xbridge
*xbridge
)
889 struct xbridge_addr
*xbridge_addr
= xbridge
->addr
;
890 struct in6_addr
*addr
= NULL
, *mask
= NULL
;
894 err
= netdev_open(xbridge
->name
, NULL
, &dev
);
896 err
= netdev_get_addr_list(dev
, &addr
, &mask
, &n_addr
);
898 if (!xbridge
->addr
||
899 n_addr
!= xbridge
->addr
->n_addr
||
900 (xbridge
->addr
->addr
&& memcmp(addr
, xbridge
->addr
->addr
,
901 sizeof(*addr
) * n_addr
))) {
902 xbridge_addr
= xzalloc(sizeof *xbridge_addr
);
903 xbridge_addr
->addr
= addr
;
904 xbridge_addr
->n_addr
= n_addr
;
905 ovs_refcount_init(&xbridge_addr
->ref_cnt
);
917 static struct xbridge_addr
*
918 xbridge_addr_ref(const struct xbridge_addr
*addr_
)
920 struct xbridge_addr
*addr
= CONST_CAST(struct xbridge_addr
*, addr_
);
922 ovs_refcount_ref(&addr
->ref_cnt
);
928 xbridge_addr_unref(struct xbridge_addr
*addr
)
930 if (addr
&& ovs_refcount_unref_relaxed(&addr
->ref_cnt
) == 1) {
937 xlate_xbridge_set(struct xbridge
*xbridge
,
939 const struct mac_learning
*ml
, struct stp
*stp
,
940 struct rstp
*rstp
, const struct mcast_snooping
*ms
,
941 const struct mbridge
*mbridge
,
942 const struct dpif_sflow
*sflow
,
943 const struct dpif_ipfix
*ipfix
,
944 const struct netflow
*netflow
,
945 bool forward_bpdu
, bool has_in_band
,
946 const struct dpif_backer_support
*support
,
947 const struct xbridge_addr
*addr
)
949 if (xbridge
->ml
!= ml
) {
950 mac_learning_unref(xbridge
->ml
);
951 xbridge
->ml
= mac_learning_ref(ml
);
954 if (xbridge
->ms
!= ms
) {
955 mcast_snooping_unref(xbridge
->ms
);
956 xbridge
->ms
= mcast_snooping_ref(ms
);
959 if (xbridge
->mbridge
!= mbridge
) {
960 mbridge_unref(xbridge
->mbridge
);
961 xbridge
->mbridge
= mbridge_ref(mbridge
);
964 if (xbridge
->sflow
!= sflow
) {
965 dpif_sflow_unref(xbridge
->sflow
);
966 xbridge
->sflow
= dpif_sflow_ref(sflow
);
969 if (xbridge
->ipfix
!= ipfix
) {
970 dpif_ipfix_unref(xbridge
->ipfix
);
971 xbridge
->ipfix
= dpif_ipfix_ref(ipfix
);
974 if (xbridge
->stp
!= stp
) {
975 stp_unref(xbridge
->stp
);
976 xbridge
->stp
= stp_ref(stp
);
979 if (xbridge
->rstp
!= rstp
) {
980 rstp_unref(xbridge
->rstp
);
981 xbridge
->rstp
= rstp_ref(rstp
);
984 if (xbridge
->netflow
!= netflow
) {
985 netflow_unref(xbridge
->netflow
);
986 xbridge
->netflow
= netflow_ref(netflow
);
989 if (xbridge
->addr
!= addr
) {
990 xbridge_addr_unref(xbridge
->addr
);
991 xbridge
->addr
= xbridge_addr_ref(addr
);
994 xbridge
->dpif
= dpif
;
995 xbridge
->forward_bpdu
= forward_bpdu
;
996 xbridge
->has_in_band
= has_in_band
;
997 xbridge
->support
= *support
;
1001 xlate_xbundle_set(struct xbundle
*xbundle
,
1002 enum port_vlan_mode vlan_mode
, uint16_t qinq_ethtype
,
1003 int vlan
, unsigned long *trunks
, unsigned long *cvlans
,
1004 enum port_priority_tags_mode use_priority_tags
,
1005 const struct bond
*bond
, const struct lacp
*lacp
,
1006 bool floodable
, bool protected)
1008 ovs_assert(xbundle
->xbridge
);
1010 xbundle
->vlan_mode
= vlan_mode
;
1011 xbundle
->qinq_ethtype
= qinq_ethtype
;
1012 xbundle
->vlan
= vlan
;
1013 xbundle
->trunks
= trunks
;
1014 xbundle
->cvlans
= cvlans
;
1015 xbundle
->use_priority_tags
= use_priority_tags
;
1016 xbundle
->floodable
= floodable
;
1017 xbundle
->protected = protected;
1019 if (xbundle
->bond
!= bond
) {
1020 bond_unref(xbundle
->bond
);
1021 xbundle
->bond
= bond_ref(bond
);
1024 if (xbundle
->lacp
!= lacp
) {
1025 lacp_unref(xbundle
->lacp
);
1026 xbundle
->lacp
= lacp_ref(lacp
);
1031 xlate_xport_set(struct xport
*xport
, odp_port_t odp_port
,
1032 const struct netdev
*netdev
, const struct cfm
*cfm
,
1033 const struct bfd
*bfd
, const struct lldp
*lldp
, int stp_port_no
,
1034 const struct rstp_port
* rstp_port
,
1035 enum ofputil_port_config config
, enum ofputil_port_state state
,
1036 bool is_tunnel
, bool may_enable
)
1038 xport
->config
= config
;
1039 xport
->state
= state
;
1040 xport
->stp_port_no
= stp_port_no
;
1041 xport
->is_tunnel
= is_tunnel
;
1042 xport
->pt_mode
= netdev_get_pt_mode(netdev
);
1043 xport
->may_enable
= may_enable
;
1044 xport
->odp_port
= odp_port
;
1046 if (xport
->rstp_port
!= rstp_port
) {
1047 rstp_port_unref(xport
->rstp_port
);
1048 xport
->rstp_port
= rstp_port_ref(rstp_port
);
1051 if (xport
->cfm
!= cfm
) {
1052 cfm_unref(xport
->cfm
);
1053 xport
->cfm
= cfm_ref(cfm
);
1056 if (xport
->bfd
!= bfd
) {
1057 bfd_unref(xport
->bfd
);
1058 xport
->bfd
= bfd_ref(bfd
);
1061 if (xport
->lldp
!= lldp
) {
1062 lldp_unref(xport
->lldp
);
1063 xport
->lldp
= lldp_ref(lldp
);
1066 if (xport
->netdev
!= netdev
) {
1067 netdev_close(xport
->netdev
);
1068 xport
->netdev
= netdev_ref(netdev
);
1073 xlate_xbridge_copy(struct xbridge
*xbridge
)
1075 struct xbundle
*xbundle
;
1076 struct xport
*xport
;
1077 struct xbridge
*new_xbridge
= xzalloc(sizeof *xbridge
);
1078 new_xbridge
->ofproto
= xbridge
->ofproto
;
1079 new_xbridge
->name
= xstrdup(xbridge
->name
);
1080 xlate_xbridge_init(new_xcfg
, new_xbridge
);
1082 xlate_xbridge_set(new_xbridge
,
1083 xbridge
->dpif
, xbridge
->ml
, xbridge
->stp
,
1084 xbridge
->rstp
, xbridge
->ms
, xbridge
->mbridge
,
1085 xbridge
->sflow
, xbridge
->ipfix
, xbridge
->netflow
,
1086 xbridge
->forward_bpdu
, xbridge
->has_in_band
,
1087 &xbridge
->support
, xbridge
->addr
);
1088 LIST_FOR_EACH (xbundle
, list_node
, &xbridge
->xbundles
) {
1089 xlate_xbundle_copy(new_xbridge
, xbundle
);
1092 /* Copy xports which are not part of a xbundle */
1093 HMAP_FOR_EACH (xport
, ofp_node
, &xbridge
->xports
) {
1094 if (!xport
->xbundle
) {
1095 xlate_xport_copy(new_xbridge
, NULL
, xport
);
1101 xlate_xbundle_copy(struct xbridge
*xbridge
, struct xbundle
*xbundle
)
1103 struct xport
*xport
;
1104 struct xbundle
*new_xbundle
= xzalloc(sizeof *xbundle
);
1105 new_xbundle
->ofbundle
= xbundle
->ofbundle
;
1106 new_xbundle
->xbridge
= xbridge
;
1107 new_xbundle
->name
= xstrdup(xbundle
->name
);
1108 xlate_xbundle_init(new_xcfg
, new_xbundle
);
1110 xlate_xbundle_set(new_xbundle
, xbundle
->vlan_mode
, xbundle
->qinq_ethtype
,
1111 xbundle
->vlan
, xbundle
->trunks
, xbundle
->cvlans
,
1112 xbundle
->use_priority_tags
, xbundle
->bond
, xbundle
->lacp
,
1113 xbundle
->floodable
, xbundle
->protected);
1114 LIST_FOR_EACH (xport
, bundle_node
, &xbundle
->xports
) {
1115 xlate_xport_copy(xbridge
, new_xbundle
, xport
);
1120 xlate_xport_copy(struct xbridge
*xbridge
, struct xbundle
*xbundle
,
1121 struct xport
*xport
)
1123 struct skb_priority_to_dscp
*pdscp
, *new_pdscp
;
1124 struct xport
*new_xport
= xzalloc(sizeof *xport
);
1125 new_xport
->ofport
= xport
->ofport
;
1126 new_xport
->ofp_port
= xport
->ofp_port
;
1127 new_xport
->xbridge
= xbridge
;
1128 new_xport
->uuid
= xport
->uuid
;
1129 xlate_xport_init(new_xcfg
, new_xport
);
1131 xlate_xport_set(new_xport
, xport
->odp_port
, xport
->netdev
, xport
->cfm
,
1132 xport
->bfd
, xport
->lldp
, xport
->stp_port_no
,
1133 xport
->rstp_port
, xport
->config
, xport
->state
,
1134 xport
->is_tunnel
, xport
->may_enable
);
1137 struct xport
*peer
= xport_lookup(new_xcfg
, xport
->peer
->ofport
);
1139 new_xport
->peer
= peer
;
1140 new_xport
->peer
->peer
= new_xport
;
1145 new_xport
->xbundle
= xbundle
;
1146 ovs_list_insert(&new_xport
->xbundle
->xports
, &new_xport
->bundle_node
);
1149 HMAP_FOR_EACH (pdscp
, hmap_node
, &xport
->skb_priorities
) {
1150 new_pdscp
= xmalloc(sizeof *pdscp
);
1151 new_pdscp
->skb_priority
= pdscp
->skb_priority
;
1152 new_pdscp
->dscp
= pdscp
->dscp
;
1153 hmap_insert(&new_xport
->skb_priorities
, &new_pdscp
->hmap_node
,
1154 hash_int(new_pdscp
->skb_priority
, 0));
1158 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
1159 * configuration in xcfgp.
1161 * This needs to be called after editing the xlate configuration.
1163 * Functions that edit the new xlate configuration are
1164 * xlate_<ofproto/bundle/ofport>_set and xlate_<ofproto/bundle/ofport>_remove.
1166 * A sample workflow:
1168 * xlate_txn_start();
1170 * edit_xlate_configuration();
1172 * xlate_txn_commit(); */
1174 xlate_txn_commit(void)
1176 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1178 ovsrcu_set(&xcfgp
, new_xcfg
);
1179 ovsrcu_synchronize();
1180 xlate_xcfg_free(xcfg
);
1184 /* Copies the current xlate configuration in xcfgp to new_xcfg.
1186 * This needs to be called prior to editing the xlate configuration. */
1188 xlate_txn_start(void)
1190 struct xbridge
*xbridge
;
1191 struct xlate_cfg
*xcfg
;
1193 ovs_assert(!new_xcfg
);
1195 new_xcfg
= xmalloc(sizeof *new_xcfg
);
1196 hmap_init(&new_xcfg
->xbridges
);
1197 hmap_init(&new_xcfg
->xbundles
);
1198 hmap_init(&new_xcfg
->xports
);
1199 hmap_init(&new_xcfg
->xports_uuid
);
1201 xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1206 HMAP_FOR_EACH (xbridge
, hmap_node
, &xcfg
->xbridges
) {
1207 xlate_xbridge_copy(xbridge
);
1213 xlate_xcfg_free(struct xlate_cfg
*xcfg
)
1215 struct xbridge
*xbridge
, *next_xbridge
;
1221 HMAP_FOR_EACH_SAFE (xbridge
, next_xbridge
, hmap_node
, &xcfg
->xbridges
) {
1222 xlate_xbridge_remove(xcfg
, xbridge
);
1225 hmap_destroy(&xcfg
->xbridges
);
1226 hmap_destroy(&xcfg
->xbundles
);
1227 hmap_destroy(&xcfg
->xports
);
1228 hmap_destroy(&xcfg
->xports_uuid
);
1233 xlate_ofproto_set(struct ofproto_dpif
*ofproto
, const char *name
,
1235 const struct mac_learning
*ml
, struct stp
*stp
,
1236 struct rstp
*rstp
, const struct mcast_snooping
*ms
,
1237 const struct mbridge
*mbridge
,
1238 const struct dpif_sflow
*sflow
,
1239 const struct dpif_ipfix
*ipfix
,
1240 const struct netflow
*netflow
,
1241 bool forward_bpdu
, bool has_in_band
,
1242 const struct dpif_backer_support
*support
)
1244 struct xbridge
*xbridge
;
1245 struct xbridge_addr
*xbridge_addr
, *old_addr
;
1247 ovs_assert(new_xcfg
);
1249 xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1251 xbridge
= xzalloc(sizeof *xbridge
);
1252 xbridge
->ofproto
= ofproto
;
1254 xlate_xbridge_init(new_xcfg
, xbridge
);
1257 free(xbridge
->name
);
1258 xbridge
->name
= xstrdup(name
);
1260 xbridge_addr
= xbridge_addr_create(xbridge
);
1261 old_addr
= xbridge
->addr
;
1263 xlate_xbridge_set(xbridge
, dpif
, ml
, stp
, rstp
, ms
, mbridge
, sflow
, ipfix
,
1264 netflow
, forward_bpdu
, has_in_band
, support
,
1267 if (xbridge_addr
!= old_addr
) {
1268 xbridge_addr_unref(xbridge_addr
);
1273 xlate_xbridge_remove(struct xlate_cfg
*xcfg
, struct xbridge
*xbridge
)
1275 struct xbundle
*xbundle
, *next_xbundle
;
1276 struct xport
*xport
, *next_xport
;
1282 HMAP_FOR_EACH_SAFE (xport
, next_xport
, ofp_node
, &xbridge
->xports
) {
1283 xlate_xport_remove(xcfg
, xport
);
1286 LIST_FOR_EACH_SAFE (xbundle
, next_xbundle
, list_node
, &xbridge
->xbundles
) {
1287 xlate_xbundle_remove(xcfg
, xbundle
);
1290 hmap_remove(&xcfg
->xbridges
, &xbridge
->hmap_node
);
1291 mac_learning_unref(xbridge
->ml
);
1292 mcast_snooping_unref(xbridge
->ms
);
1293 mbridge_unref(xbridge
->mbridge
);
1294 dpif_sflow_unref(xbridge
->sflow
);
1295 dpif_ipfix_unref(xbridge
->ipfix
);
1296 netflow_unref(xbridge
->netflow
);
1297 stp_unref(xbridge
->stp
);
1298 rstp_unref(xbridge
->rstp
);
1299 xbridge_addr_unref(xbridge
->addr
);
1300 hmap_destroy(&xbridge
->xports
);
1301 free(xbridge
->name
);
1306 xlate_remove_ofproto(struct ofproto_dpif
*ofproto
)
1308 struct xbridge
*xbridge
;
1310 ovs_assert(new_xcfg
);
1312 xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1313 xlate_xbridge_remove(new_xcfg
, xbridge
);
1317 xlate_bundle_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
1318 const char *name
, enum port_vlan_mode vlan_mode
,
1319 uint16_t qinq_ethtype
, int vlan
,
1320 unsigned long *trunks
, unsigned long *cvlans
,
1321 enum port_priority_tags_mode use_priority_tags
,
1322 const struct bond
*bond
, const struct lacp
*lacp
,
1323 bool floodable
, bool protected)
1325 struct xbundle
*xbundle
;
1327 ovs_assert(new_xcfg
);
1329 xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1331 xbundle
= xzalloc(sizeof *xbundle
);
1332 xbundle
->ofbundle
= ofbundle
;
1333 xbundle
->xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1335 xlate_xbundle_init(new_xcfg
, xbundle
);
1338 free(xbundle
->name
);
1339 xbundle
->name
= xstrdup(name
);
1341 xlate_xbundle_set(xbundle
, vlan_mode
, qinq_ethtype
, vlan
, trunks
, cvlans
,
1342 use_priority_tags
, bond
, lacp
, floodable
, protected);
1346 xlate_xbundle_remove(struct xlate_cfg
*xcfg
, struct xbundle
*xbundle
)
1348 struct xport
*xport
;
1354 LIST_FOR_EACH_POP (xport
, bundle_node
, &xbundle
->xports
) {
1355 xport
->xbundle
= NULL
;
1358 hmap_remove(&xcfg
->xbundles
, &xbundle
->hmap_node
);
1359 ovs_list_remove(&xbundle
->list_node
);
1360 bond_unref(xbundle
->bond
);
1361 lacp_unref(xbundle
->lacp
);
1362 free(xbundle
->name
);
1367 xlate_bundle_remove(struct ofbundle
*ofbundle
)
1369 struct xbundle
*xbundle
;
1371 ovs_assert(new_xcfg
);
1373 xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1374 xlate_xbundle_remove(new_xcfg
, xbundle
);
1378 xlate_ofport_set(struct ofproto_dpif
*ofproto
, struct ofbundle
*ofbundle
,
1379 struct ofport_dpif
*ofport
, ofp_port_t ofp_port
,
1380 odp_port_t odp_port
, const struct netdev
*netdev
,
1381 const struct cfm
*cfm
, const struct bfd
*bfd
,
1382 const struct lldp
*lldp
, struct ofport_dpif
*peer
,
1383 int stp_port_no
, const struct rstp_port
*rstp_port
,
1384 const struct ofproto_port_queue
*qdscp_list
, size_t n_qdscp
,
1385 enum ofputil_port_config config
,
1386 enum ofputil_port_state state
, bool is_tunnel
,
1390 struct xport
*xport
;
1392 ovs_assert(new_xcfg
);
1394 xport
= xport_lookup(new_xcfg
, ofport
);
1396 xport
= xzalloc(sizeof *xport
);
1397 xport
->ofport
= ofport
;
1398 xport
->xbridge
= xbridge_lookup(new_xcfg
, ofproto
);
1399 xport
->ofp_port
= ofp_port
;
1400 uuid_generate(&xport
->uuid
);
1402 xlate_xport_init(new_xcfg
, xport
);
1405 ovs_assert(xport
->ofp_port
== ofp_port
);
1407 xlate_xport_set(xport
, odp_port
, netdev
, cfm
, bfd
, lldp
,
1408 stp_port_no
, rstp_port
, config
, state
, is_tunnel
,
1412 xport
->peer
->peer
= NULL
;
1414 xport
->peer
= xport_lookup(new_xcfg
, peer
);
1416 xport
->peer
->peer
= xport
;
1419 if (xport
->xbundle
) {
1420 ovs_list_remove(&xport
->bundle_node
);
1422 xport
->xbundle
= xbundle_lookup(new_xcfg
, ofbundle
);
1423 if (xport
->xbundle
) {
1424 ovs_list_insert(&xport
->xbundle
->xports
, &xport
->bundle_node
);
1427 clear_skb_priorities(xport
);
1428 for (i
= 0; i
< n_qdscp
; i
++) {
1429 struct skb_priority_to_dscp
*pdscp
;
1430 uint32_t skb_priority
;
1432 if (dpif_queue_to_priority(xport
->xbridge
->dpif
, qdscp_list
[i
].queue
,
1437 pdscp
= xmalloc(sizeof *pdscp
);
1438 pdscp
->skb_priority
= skb_priority
;
1439 pdscp
->dscp
= (qdscp_list
[i
].dscp
<< 2) & IP_DSCP_MASK
;
1440 hmap_insert(&xport
->skb_priorities
, &pdscp
->hmap_node
,
1441 hash_int(pdscp
->skb_priority
, 0));
1446 xlate_xport_remove(struct xlate_cfg
*xcfg
, struct xport
*xport
)
1453 xport
->peer
->peer
= NULL
;
1457 if (xport
->xbundle
) {
1458 ovs_list_remove(&xport
->bundle_node
);
1461 clear_skb_priorities(xport
);
1462 hmap_destroy(&xport
->skb_priorities
);
1464 hmap_remove(&xcfg
->xports
, &xport
->hmap_node
);
1465 hmap_remove(&xcfg
->xports_uuid
, &xport
->uuid_node
);
1466 hmap_remove(&xport
->xbridge
->xports
, &xport
->ofp_node
);
1468 netdev_close(xport
->netdev
);
1469 rstp_port_unref(xport
->rstp_port
);
1470 cfm_unref(xport
->cfm
);
1471 bfd_unref(xport
->bfd
);
1472 lldp_unref(xport
->lldp
);
1477 xlate_ofport_remove(struct ofport_dpif
*ofport
)
1479 struct xport
*xport
;
1481 ovs_assert(new_xcfg
);
1483 xport
= xport_lookup(new_xcfg
, ofport
);
1484 xlate_xport_remove(new_xcfg
, xport
);
1487 static struct ofproto_dpif
*
1488 xlate_lookup_ofproto_(const struct dpif_backer
*backer
,
1489 const struct flow
*flow
,
1490 ofp_port_t
*ofp_in_port
, const struct xport
**xportp
,
1493 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
1494 const struct xport
*xport
;
1496 /* If packet is recirculated, xport can be retrieved from frozen state. */
1497 if (flow
->recirc_id
) {
1498 const struct recirc_id_node
*recirc_id_node
;
1500 recirc_id_node
= recirc_id_node_find(flow
->recirc_id
);
1502 if (OVS_UNLIKELY(!recirc_id_node
)) {
1504 *errorp
= xasprintf("no recirculation data for recirc_id "
1505 "%"PRIu32
, flow
->recirc_id
);
1510 /* If recirculation was initiated due to bond (in_port = OFPP_NONE)
1511 * then frozen state is static and xport_uuid is not defined, so xport
1512 * cannot be restored from frozen state. */
1513 if (recirc_id_node
->state
.metadata
.in_port
!= OFPP_NONE
) {
1514 struct uuid xport_uuid
= recirc_id_node
->state
.xport_uuid
;
1515 xport
= xport_lookup_by_uuid(xcfg
, &xport_uuid
);
1516 if (xport
&& xport
->xbridge
&& xport
->xbridge
->ofproto
) {
1522 xport
= xport_lookup(xcfg
, tnl_port_should_receive(flow
)
1523 ? tnl_port_receive(flow
)
1524 : odp_port_to_ofport(backer
, flow
->in_port
.odp_port
));
1525 if (OVS_UNLIKELY(!xport
)) {
1527 *errorp
= (tnl_port_should_receive(flow
)
1528 ? xstrdup("no OpenFlow tunnel port for this packet")
1529 : xasprintf("no OpenFlow tunnel port for datapath "
1530 "port %"PRIu32
, flow
->in_port
.odp_port
));
1541 *ofp_in_port
= xport
->ofp_port
;
1543 return xport
->xbridge
->ofproto
;
1546 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1547 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1548 struct ofproto_dpif
*
1549 xlate_lookup_ofproto(const struct dpif_backer
*backer
, const struct flow
*flow
,
1550 ofp_port_t
*ofp_in_port
, char **errorp
)
1552 const struct xport
*xport
;
1554 return xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, errorp
);
1557 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1558 * optionally populates 'ofprotop' with the ofproto_dpif, 'ofp_in_port' with the
1559 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1560 * handles for those protocols if they're enabled. Caller may use the returned
1561 * pointers until quiescing, for longer term use additional references must
1564 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1567 xlate_lookup(const struct dpif_backer
*backer
, const struct flow
*flow
,
1568 struct ofproto_dpif
**ofprotop
, struct dpif_ipfix
**ipfix
,
1569 struct dpif_sflow
**sflow
, struct netflow
**netflow
,
1570 ofp_port_t
*ofp_in_port
)
1572 struct ofproto_dpif
*ofproto
;
1573 const struct xport
*xport
;
1575 ofproto
= xlate_lookup_ofproto_(backer
, flow
, ofp_in_port
, &xport
, NULL
);
1582 *ofprotop
= ofproto
;
1586 *ipfix
= xport
? xport
->xbridge
->ipfix
: NULL
;
1590 *sflow
= xport
? xport
->xbridge
->sflow
: NULL
;
1594 *netflow
= xport
? xport
->xbridge
->netflow
: NULL
;
1600 static struct xbridge
*
1601 xbridge_lookup(struct xlate_cfg
*xcfg
, const struct ofproto_dpif
*ofproto
)
1603 struct hmap
*xbridges
;
1604 struct xbridge
*xbridge
;
1606 if (!ofproto
|| !xcfg
) {
1610 xbridges
= &xcfg
->xbridges
;
1612 HMAP_FOR_EACH_IN_BUCKET (xbridge
, hmap_node
, hash_pointer(ofproto
, 0),
1614 if (xbridge
->ofproto
== ofproto
) {
1621 static struct xbridge
*
1622 xbridge_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1624 struct xbridge
*xbridge
;
1626 HMAP_FOR_EACH (xbridge
, hmap_node
, &xcfg
->xbridges
) {
1627 if (uuid_equals(&xbridge
->ofproto
->uuid
, uuid
)) {
1634 static struct xbundle
*
1635 xbundle_lookup(struct xlate_cfg
*xcfg
, const struct ofbundle
*ofbundle
)
1637 struct hmap
*xbundles
;
1638 struct xbundle
*xbundle
;
1640 if (!ofbundle
|| !xcfg
) {
1644 xbundles
= &xcfg
->xbundles
;
1646 HMAP_FOR_EACH_IN_BUCKET (xbundle
, hmap_node
, hash_pointer(ofbundle
, 0),
1648 if (xbundle
->ofbundle
== ofbundle
) {
1655 static struct xport
*
1656 xport_lookup(struct xlate_cfg
*xcfg
, const struct ofport_dpif
*ofport
)
1658 struct hmap
*xports
;
1659 struct xport
*xport
;
1661 if (!ofport
|| !xcfg
) {
1665 xports
= &xcfg
->xports
;
1667 HMAP_FOR_EACH_IN_BUCKET (xport
, hmap_node
, hash_pointer(ofport
, 0),
1669 if (xport
->ofport
== ofport
) {
1676 static struct xport
*
1677 xport_lookup_by_uuid(struct xlate_cfg
*xcfg
, const struct uuid
*uuid
)
1679 struct hmap
*xports
;
1680 struct xport
*xport
;
1682 if (uuid_is_zero(uuid
) || !xcfg
) {
1686 xports
= &xcfg
->xports_uuid
;
1688 HMAP_FOR_EACH_IN_BUCKET (xport
, uuid_node
, uuid_hash(uuid
), xports
) {
1689 if (uuid_equals(&xport
->uuid
, uuid
)) {
1696 static struct stp_port
*
1697 xport_get_stp_port(const struct xport
*xport
)
1699 return xport
->xbridge
->stp
&& xport
->stp_port_no
!= -1
1700 ? stp_get_port(xport
->xbridge
->stp
, xport
->stp_port_no
)
1705 xport_stp_learn_state(const struct xport
*xport
)
1707 struct stp_port
*sp
= xport_get_stp_port(xport
);
1709 ? stp_learn_in_state(stp_port_get_state(sp
))
1714 xport_stp_forward_state(const struct xport
*xport
)
1716 struct stp_port
*sp
= xport_get_stp_port(xport
);
1718 ? stp_forward_in_state(stp_port_get_state(sp
))
1723 xport_stp_should_forward_bpdu(const struct xport
*xport
)
1725 struct stp_port
*sp
= xport_get_stp_port(xport
);
1726 return stp_should_forward_bpdu(sp
? stp_port_get_state(sp
) : STP_DISABLED
);
1729 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1730 * were used to make the determination.*/
1732 stp_should_process_flow(const struct flow
*flow
, struct flow_wildcards
*wc
)
1734 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1735 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
1736 return is_stp(flow
);
1740 stp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1742 struct stp_port
*sp
= xport_get_stp_port(xport
);
1743 struct dp_packet payload
= *packet
;
1744 struct eth_header
*eth
= dp_packet_data(&payload
);
1746 /* Sink packets on ports that have STP disabled when the bridge has
1748 if (!sp
|| stp_port_get_state(sp
) == STP_DISABLED
) {
1752 /* Trim off padding on payload. */
1753 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1754 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1757 if (dp_packet_try_pull(&payload
, ETH_HEADER_LEN
+ LLC_HEADER_LEN
)) {
1758 stp_received_bpdu(sp
, dp_packet_data(&payload
), dp_packet_size(&payload
));
1762 static enum rstp_state
1763 xport_get_rstp_port_state(const struct xport
*xport
)
1765 return xport
->rstp_port
1766 ? rstp_port_get_state(xport
->rstp_port
)
1771 xport_rstp_learn_state(const struct xport
*xport
)
1773 return xport
->xbridge
->rstp
&& xport
->rstp_port
1774 ? rstp_learn_in_state(xport_get_rstp_port_state(xport
))
1779 xport_rstp_forward_state(const struct xport
*xport
)
1781 return xport
->xbridge
->rstp
&& xport
->rstp_port
1782 ? rstp_forward_in_state(xport_get_rstp_port_state(xport
))
1787 xport_rstp_should_manage_bpdu(const struct xport
*xport
)
1789 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport
));
1793 rstp_process_packet(const struct xport
*xport
, const struct dp_packet
*packet
)
1795 struct dp_packet payload
= *packet
;
1796 struct eth_header
*eth
= dp_packet_data(&payload
);
1798 /* Sink packets on ports that have no RSTP. */
1799 if (!xport
->rstp_port
) {
1803 /* Trim off padding on payload. */
1804 if (dp_packet_size(&payload
) > ntohs(eth
->eth_type
) + ETH_HEADER_LEN
) {
1805 dp_packet_set_size(&payload
, ntohs(eth
->eth_type
) + ETH_HEADER_LEN
);
1808 int len
= ETH_HEADER_LEN
+ LLC_HEADER_LEN
;
1809 if (eth
->eth_type
== htons(ETH_TYPE_VLAN
)) {
1810 len
+= VLAN_HEADER_LEN
;
1812 if (dp_packet_try_pull(&payload
, len
)) {
1813 rstp_port_received_bpdu(xport
->rstp_port
, dp_packet_data(&payload
),
1814 dp_packet_size(&payload
));
1818 static struct xport
*
1819 get_ofp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1821 struct xport
*xport
;
1823 HMAP_FOR_EACH_IN_BUCKET (xport
, ofp_node
, hash_ofp_port(ofp_port
),
1825 if (xport
->ofp_port
== ofp_port
) {
1833 ofp_port_to_odp_port(const struct xbridge
*xbridge
, ofp_port_t ofp_port
)
1835 const struct xport
*xport
= get_ofp_port(xbridge
, ofp_port
);
1836 return xport
? xport
->odp_port
: ODPP_NONE
;
1840 odp_port_is_alive(const struct xlate_ctx
*ctx
, ofp_port_t ofp_port
)
1842 struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
1843 return xport
&& xport
->may_enable
;
1846 static struct ofputil_bucket
*
1847 group_first_live_bucket(const struct xlate_ctx
*, const struct group_dpif
*,
1851 group_is_alive(const struct xlate_ctx
*ctx
, uint32_t group_id
, int depth
)
1853 struct group_dpif
*group
;
1855 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
1856 ctx
->xin
->tables_version
, false);
1858 return group_first_live_bucket(ctx
, group
, depth
) != NULL
;
1864 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1867 bucket_is_alive(const struct xlate_ctx
*ctx
,
1868 struct ofputil_bucket
*bucket
, int depth
)
1870 if (depth
>= MAX_LIVENESS_RECURSION
) {
1871 xlate_report_error(ctx
, "bucket chaining exceeded %d links",
1872 MAX_LIVENESS_RECURSION
);
1876 return (!ofputil_bucket_has_liveness(bucket
)
1877 || (bucket
->watch_port
!= OFPP_ANY
1878 && odp_port_is_alive(ctx
, bucket
->watch_port
))
1879 || (bucket
->watch_group
!= OFPG_ANY
1880 && group_is_alive(ctx
, bucket
->watch_group
, depth
+ 1)));
1884 xlate_report_bucket_not_live(const struct xlate_ctx
*ctx
,
1885 const struct ofputil_bucket
*bucket
)
1887 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
1888 struct ds s
= DS_EMPTY_INITIALIZER
;
1889 if (bucket
->watch_port
!= OFPP_ANY
) {
1890 ds_put_cstr(&s
, "port ");
1891 ofputil_format_port(bucket
->watch_port
, NULL
, &s
);
1893 if (bucket
->watch_group
!= OFPG_ANY
) {
1895 ds_put_cstr(&s
, " and ");
1897 ds_put_format(&s
, "port %"PRIu32
, bucket
->watch_group
);
1900 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": not live due to %s",
1901 bucket
->bucket_id
, ds_cstr(&s
));
1907 static struct ofputil_bucket
*
1908 group_first_live_bucket(const struct xlate_ctx
*ctx
,
1909 const struct group_dpif
*group
, int depth
)
1911 struct ofputil_bucket
*bucket
;
1912 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1913 if (bucket_is_alive(ctx
, bucket
, depth
)) {
1916 xlate_report_bucket_not_live(ctx
, bucket
);
1922 static struct ofputil_bucket
*
1923 group_best_live_bucket(const struct xlate_ctx
*ctx
,
1924 const struct group_dpif
*group
,
1927 struct ofputil_bucket
*best_bucket
= NULL
;
1928 uint32_t best_score
= 0;
1930 struct ofputil_bucket
*bucket
;
1931 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
1932 if (bucket_is_alive(ctx
, bucket
, 0)) {
1934 (hash_int(bucket
->bucket_id
, basis
) & 0xffff) * bucket
->weight
;
1935 if (score
>= best_score
) {
1936 best_bucket
= bucket
;
1939 xlate_report(ctx
, OFT_DETAIL
, "bucket %"PRIu32
": score %"PRIu32
,
1940 bucket
->bucket_id
, score
);
1942 xlate_report_bucket_not_live(ctx
, bucket
);
1950 xbundle_trunks_vlan(const struct xbundle
*bundle
, uint16_t vlan
)
1952 return (bundle
->vlan_mode
!= PORT_VLAN_ACCESS
1953 && (!bundle
->trunks
|| bitmap_is_set(bundle
->trunks
, vlan
)));
1957 xbundle_allows_cvlan(const struct xbundle
*bundle
, uint16_t vlan
)
1959 return (!bundle
->cvlans
|| bitmap_is_set(bundle
->cvlans
, vlan
));
1963 xbundle_includes_vlan(const struct xbundle
*xbundle
, const struct xvlan
*xvlan
)
1965 switch (xbundle
->vlan_mode
) {
1966 case PORT_VLAN_ACCESS
:
1967 return xvlan
->v
[0].vid
== xbundle
->vlan
&& xvlan
->v
[1].vid
== 0;
1969 case PORT_VLAN_TRUNK
:
1970 case PORT_VLAN_NATIVE_UNTAGGED
:
1971 case PORT_VLAN_NATIVE_TAGGED
:
1972 return xbundle_trunks_vlan(xbundle
, xvlan
->v
[0].vid
);
1974 case PORT_VLAN_DOT1Q_TUNNEL
:
1975 return xvlan
->v
[0].vid
== xbundle
->vlan
&&
1976 xbundle_allows_cvlan(xbundle
, xvlan
->v
[1].vid
);
1983 static mirror_mask_t
1984 xbundle_mirror_out(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
1986 return xbundle
!= &ofpp_none_bundle
1987 ? mirror_bundle_out(xbridge
->mbridge
, xbundle
->ofbundle
)
1991 static mirror_mask_t
1992 xbundle_mirror_src(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
1994 return xbundle
!= &ofpp_none_bundle
1995 ? mirror_bundle_src(xbridge
->mbridge
, xbundle
->ofbundle
)
1999 static mirror_mask_t
2000 xbundle_mirror_dst(const struct xbridge
*xbridge
, struct xbundle
*xbundle
)
2002 return xbundle
!= &ofpp_none_bundle
2003 ? mirror_bundle_dst(xbridge
->mbridge
, xbundle
->ofbundle
)
2007 static struct xbundle
*
2008 lookup_input_bundle__(const struct xbridge
*xbridge
,
2009 ofp_port_t in_port
, struct xport
**in_xportp
)
2011 struct xport
*xport
;
2013 /* Find the port and bundle for the received packet. */
2014 xport
= get_ofp_port(xbridge
, in_port
);
2018 if (xport
&& xport
->xbundle
) {
2019 return xport
->xbundle
;
2022 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
2023 * which a controller may use as the ingress port for traffic that
2024 * it is sourcing. */
2025 if (in_port
== OFPP_CONTROLLER
|| in_port
== OFPP_NONE
) {
2026 return &ofpp_none_bundle
;
2031 static struct xbundle
*
2032 lookup_input_bundle(const struct xlate_ctx
*ctx
,
2033 ofp_port_t in_port
, struct xport
**in_xportp
)
2035 struct xbundle
*xbundle
= lookup_input_bundle__(ctx
->xbridge
,
2036 in_port
, in_xportp
);
2038 /* Odd. A few possible reasons here:
2040 * - We deleted a port but there are still a few packets queued up
2043 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
2044 * we don't know about.
2046 * - The ofproto client didn't configure the port as part of a bundle.
2047 * This is particularly likely to happen if a packet was received on
2048 * the port after it was created, but before the client had a chance
2049 * to configure its bundle.
2051 xlate_report_error(ctx
, "received packet on unknown port %"PRIu32
,
2057 /* Mirrors the packet represented by 'ctx' to appropriate mirror destinations,
2058 * given the packet is ingressing or egressing on 'xbundle', which has ingress
2059 * or egress (as appropriate) mirrors 'mirrors'. */
2061 mirror_packet(struct xlate_ctx
*ctx
, struct xbundle
*xbundle
,
2062 mirror_mask_t mirrors
)
2064 struct xvlan in_xvlan
;
2067 /* Figure out what VLAN the packet is in (because mirrors can select
2068 * packets on basis of VLAN). */
2069 xvlan_extract(&ctx
->xin
->flow
, &in_xvlan
);
2070 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, xbundle
)) {
2073 xvlan_input_translate(xbundle
, &in_xvlan
, &xvlan
);
2075 const struct xbridge
*xbridge
= ctx
->xbridge
;
2077 /* Don't mirror to destinations that we've already mirrored to. */
2078 mirrors
&= ~ctx
->mirrors
;
2083 /* 'mirrors' is a bit-mask of candidates for mirroring. Iterate through
2084 * the candidates, adding the ones that really should be mirrored to
2085 * 'used_mirrors', as long as some candidates remain. */
2086 mirror_mask_t used_mirrors
= 0;
2088 const unsigned long *vlans
;
2089 mirror_mask_t dup_mirrors
;
2090 struct ofbundle
*out
;
2094 /* Get the details of the mirror represented by the rightmost 1-bit. */
2095 ovs_assert(mirror_get(xbridge
->mbridge
, raw_ctz(mirrors
),
2096 &vlans
, &dup_mirrors
,
2097 &out
, &snaplen
, &out_vlan
));
2100 /* If this mirror selects on the basis of VLAN, and it does not select
2101 * 'vlan', then discard this mirror and go on to the next one. */
2103 ctx
->wc
->masks
.vlans
[0].tci
|= htons(VLAN_CFI
| VLAN_VID_MASK
);
2105 if (vlans
&& !bitmap_is_set(vlans
, xvlan
.v
[0].vid
)) {
2106 mirrors
= zero_rightmost_1bit(mirrors
);
2110 /* We sent a packet to this mirror. */
2111 used_mirrors
|= rightmost_1bit(mirrors
);
2113 /* Record the mirror, and the mirrors that output to the same
2114 * destination, so that we don't mirror to them again. This must be
2115 * done now to ensure that output_normal(), below, doesn't recursively
2116 * output to the same mirrors. */
2117 ctx
->mirrors
|= dup_mirrors
;
2118 ctx
->mirror_snaplen
= snaplen
;
2120 /* Send the packet to the mirror. */
2122 struct xbundle
*out_xbundle
= xbundle_lookup(ctx
->xcfg
, out
);
2124 output_normal(ctx
, out_xbundle
, &xvlan
);
2126 } else if (xvlan
.v
[0].vid
!= out_vlan
2127 && !eth_addr_is_reserved(ctx
->xin
->flow
.dl_dst
)) {
2129 uint16_t old_vid
= xvlan
.v
[0].vid
;
2131 xvlan
.v
[0].vid
= out_vlan
;
2132 LIST_FOR_EACH (xb
, list_node
, &xbridge
->xbundles
) {
2133 if (xbundle_includes_vlan(xb
, &xvlan
)
2134 && !xbundle_mirror_out(xbridge
, xb
)) {
2135 output_normal(ctx
, xb
, &xvlan
);
2138 xvlan
.v
[0].vid
= old_vid
;
2141 /* output_normal() could have recursively output (to different
2142 * mirrors), so make sure that we don't send duplicates. */
2143 mirrors
&= ~ctx
->mirrors
;
2144 ctx
->mirror_snaplen
= 0;
2148 if (ctx
->xin
->resubmit_stats
) {
2149 mirror_update_stats(xbridge
->mbridge
, used_mirrors
,
2150 ctx
->xin
->resubmit_stats
->n_packets
,
2151 ctx
->xin
->resubmit_stats
->n_bytes
);
2153 if (ctx
->xin
->xcache
) {
2154 struct xc_entry
*entry
;
2156 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_MIRROR
);
2157 entry
->mirror
.mbridge
= mbridge_ref(xbridge
->mbridge
);
2158 entry
->mirror
.mirrors
= used_mirrors
;
2164 mirror_ingress_packet(struct xlate_ctx
*ctx
)
2166 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
)) {
2167 struct xbundle
*xbundle
= lookup_input_bundle(
2168 ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
);
2170 mirror_packet(ctx
, xbundle
,
2171 xbundle_mirror_src(ctx
->xbridge
, xbundle
));
2176 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
2177 * If so, returns true. Otherwise, returns false.
2179 * 'vid' should be the VID obtained from the 802.1Q header that was received as
2180 * part of a packet (specify 0 if there was no 802.1Q header), in the range
2183 input_vid_is_valid(const struct xlate_ctx
*ctx
,
2184 uint16_t vid
, struct xbundle
*in_xbundle
)
2186 /* Allow any VID on the OFPP_NONE port. */
2187 if (in_xbundle
== &ofpp_none_bundle
) {
2191 switch (in_xbundle
->vlan_mode
) {
2192 case PORT_VLAN_ACCESS
:
2194 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" tagged "
2195 "packet received on port %s configured as VLAN "
2196 "%d access port", vid
, in_xbundle
->name
,
2202 case PORT_VLAN_NATIVE_UNTAGGED
:
2203 case PORT_VLAN_NATIVE_TAGGED
:
2205 /* Port must always carry its native VLAN. */
2209 case PORT_VLAN_TRUNK
:
2210 if (!xbundle_trunks_vlan(in_xbundle
, vid
)) {
2211 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet "
2212 "received on port %s not configured for "
2213 "trunking VLAN %"PRIu16
,
2214 vid
, in_xbundle
->name
, vid
);
2219 case PORT_VLAN_DOT1Q_TUNNEL
:
2220 if (!xbundle_allows_cvlan(in_xbundle
, vid
)) {
2221 xlate_report_error(ctx
, "dropping VLAN %"PRIu16
" packet received "
2222 "on dot1q-tunnel port %s that excludes this "
2223 "VLAN", vid
, in_xbundle
->name
);
2235 xvlan_copy(struct xvlan
*dst
, const struct xvlan
*src
)
2241 xvlan_pop(struct xvlan
*src
)
2243 memmove(&src
->v
[0], &src
->v
[1], sizeof(src
->v
) - sizeof(src
->v
[0]));
2244 memset(&src
->v
[FLOW_MAX_VLAN_HEADERS
- 1], 0,
2245 sizeof(src
->v
[FLOW_MAX_VLAN_HEADERS
- 1]));
2249 xvlan_push_uninit(struct xvlan
*src
)
2251 memmove(&src
->v
[1], &src
->v
[0], sizeof(src
->v
) - sizeof(src
->v
[0]));
2252 memset(&src
->v
[0], 0, sizeof(src
->v
[0]));
2255 /* Extract VLAN information (headers) from flow */
2257 xvlan_extract(const struct flow
*flow
, struct xvlan
*xvlan
)
2260 memset(xvlan
, 0, sizeof(*xvlan
));
2261 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2262 if (!eth_type_vlan(flow
->vlans
[i
].tpid
) ||
2263 !(flow
->vlans
[i
].tci
& htons(VLAN_CFI
))) {
2266 xvlan
->v
[i
].tpid
= ntohs(flow
->vlans
[i
].tpid
);
2267 xvlan
->v
[i
].vid
= vlan_tci_to_vid(flow
->vlans
[i
].tci
);
2268 xvlan
->v
[i
].pcp
= ntohs(flow
->vlans
[i
].tci
) & VLAN_PCP_MASK
;
2272 /* Put VLAN information (headers) to flow */
2274 xvlan_put(struct flow
*flow
, const struct xvlan
*xvlan
,
2275 enum port_priority_tags_mode use_priority_tags
)
2279 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
2280 tci
= htons(xvlan
->v
[i
].vid
| (xvlan
->v
[i
].pcp
& VLAN_PCP_MASK
));
2281 if (tci
|| ((use_priority_tags
== PORT_PRIORITY_TAGS_ALWAYS
) &&
2282 xvlan
->v
[i
].tpid
)) {
2283 tci
|= htons(VLAN_CFI
);
2284 flow
->vlans
[i
].tpid
= xvlan
->v
[i
].tpid
?
2285 htons(xvlan
->v
[i
].tpid
) :
2286 htons(ETH_TYPE_VLAN_8021Q
);
2288 flow
->vlans
[i
].tci
= tci
;
2292 /* Given 'in_xvlan', extracted from the input 802.1Q headers received as part
2293 * of a packet, and 'in_xbundle', the bundle on which the packet was received,
2294 * returns the VLANs of the packet during bridge internal processing. */
2296 xvlan_input_translate(const struct xbundle
*in_xbundle
,
2297 const struct xvlan
*in_xvlan
, struct xvlan
*xvlan
)
2300 switch (in_xbundle
->vlan_mode
) {
2301 case PORT_VLAN_ACCESS
:
2302 memset(xvlan
, 0, sizeof(*xvlan
));
2303 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2304 ETH_TYPE_VLAN_8021Q
;
2305 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2306 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2309 case PORT_VLAN_TRUNK
:
2310 xvlan_copy(xvlan
, in_xvlan
);
2313 case PORT_VLAN_NATIVE_UNTAGGED
:
2314 case PORT_VLAN_NATIVE_TAGGED
:
2315 xvlan_copy(xvlan
, in_xvlan
);
2316 if (!in_xvlan
->v
[0].vid
) {
2317 xvlan
->v
[0].tpid
= in_xvlan
->v
[0].tpid
? in_xvlan
->v
[0].tpid
:
2318 ETH_TYPE_VLAN_8021Q
;
2319 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2320 xvlan
->v
[0].pcp
= in_xvlan
->v
[0].pcp
;
2324 case PORT_VLAN_DOT1Q_TUNNEL
:
2325 xvlan_copy(xvlan
, in_xvlan
);
2326 xvlan_push_uninit(xvlan
);
2327 xvlan
->v
[0].tpid
= in_xbundle
->qinq_ethtype
;
2328 xvlan
->v
[0].vid
= in_xbundle
->vlan
;
2329 xvlan
->v
[0].pcp
= 0;
2337 /* Given 'xvlan', the VLANs of a packet during internal processing, and
2338 * 'out_xbundle', a bundle on which the packet is to be output, returns the
2339 * VLANs that should be included in output packet. */
2341 xvlan_output_translate(const struct xbundle
*out_xbundle
,
2342 const struct xvlan
*xvlan
, struct xvlan
*out_xvlan
)
2344 switch (out_xbundle
->vlan_mode
) {
2345 case PORT_VLAN_ACCESS
:
2346 memset(out_xvlan
, 0, sizeof(*out_xvlan
));
2349 case PORT_VLAN_TRUNK
:
2350 case PORT_VLAN_NATIVE_TAGGED
:
2351 xvlan_copy(out_xvlan
, xvlan
);
2354 case PORT_VLAN_NATIVE_UNTAGGED
:
2355 xvlan_copy(out_xvlan
, xvlan
);
2356 if (xvlan
->v
[0].vid
== out_xbundle
->vlan
) {
2357 xvlan_pop(out_xvlan
);
2361 case PORT_VLAN_DOT1Q_TUNNEL
:
2362 xvlan_copy(out_xvlan
, xvlan
);
2363 xvlan_pop(out_xvlan
);
2371 /* If output xbundle is dot1q-tunnel, set mask bits of cvlan */
2373 check_and_set_cvlan_mask(struct flow_wildcards
*wc
,
2374 const struct xbundle
*xbundle
)
2376 if (xbundle
->vlan_mode
== PORT_VLAN_DOT1Q_TUNNEL
&& xbundle
->cvlans
) {
2377 wc
->masks
.vlans
[1].tci
= htons(0xffff);
2382 output_normal(struct xlate_ctx
*ctx
, const struct xbundle
*out_xbundle
,
2383 const struct xvlan
*xvlan
)
2386 union flow_vlan_hdr old_vlans
[FLOW_MAX_VLAN_HEADERS
];
2387 struct xport
*xport
;
2388 struct xlate_bond_recirc xr
;
2389 bool use_recirc
= false;
2390 struct xvlan out_xvlan
;
2392 check_and_set_cvlan_mask(ctx
->wc
, out_xbundle
);
2394 xvlan_output_translate(out_xbundle
, xvlan
, &out_xvlan
);
2395 if (out_xbundle
->use_priority_tags
) {
2396 out_xvlan
.v
[0].pcp
= ntohs(ctx
->xin
->flow
.vlans
[0].tci
) &
2399 vid
= out_xvlan
.v
[0].vid
;
2400 if (ovs_list_is_empty(&out_xbundle
->xports
)) {
2401 /* Partially configured bundle with no slaves. Drop the packet. */
2403 } else if (!out_xbundle
->bond
) {
2404 xport
= CONTAINER_OF(ovs_list_front(&out_xbundle
->xports
), struct xport
,
2407 struct flow_wildcards
*wc
= ctx
->wc
;
2408 struct ofport_dpif
*ofport
;
2410 if (ctx
->xbridge
->support
.odp
.recirc
) {
2411 /* In case recirculation is not actually in use, 'xr.recirc_id'
2412 * will be set to '0', since a valid 'recirc_id' can
2414 bond_update_post_recirc_rules(out_xbundle
->bond
,
2418 /* Use recirculation instead of output. */
2420 xr
.hash_alg
= OVS_HASH_ALG_L4
;
2421 /* Recirculation does not require unmasking hash fields. */
2426 ofport
= bond_choose_output_slave(out_xbundle
->bond
,
2427 &ctx
->xin
->flow
, wc
, vid
);
2428 xport
= xport_lookup(ctx
->xcfg
, ofport
);
2431 /* No slaves enabled, so drop packet. */
2435 /* If use_recirc is set, the main thread will handle stats
2436 * accounting for this bond. */
2438 if (ctx
->xin
->resubmit_stats
) {
2439 bond_account(out_xbundle
->bond
, &ctx
->xin
->flow
, vid
,
2440 ctx
->xin
->resubmit_stats
->n_bytes
);
2442 if (ctx
->xin
->xcache
) {
2443 struct xc_entry
*entry
;
2446 flow
= &ctx
->xin
->flow
;
2447 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_BOND
);
2448 entry
->bond
.bond
= bond_ref(out_xbundle
->bond
);
2449 entry
->bond
.flow
= xmemdup(flow
, sizeof *flow
);
2450 entry
->bond
.vid
= vid
;
2455 memcpy(&old_vlans
, &ctx
->xin
->flow
.vlans
, sizeof(old_vlans
));
2456 xvlan_put(&ctx
->xin
->flow
, &out_xvlan
, out_xbundle
->use_priority_tags
);
2458 compose_output_action(ctx
, xport
->ofp_port
, use_recirc
? &xr
: NULL
,
2460 memcpy(&ctx
->xin
->flow
.vlans
, &old_vlans
, sizeof(old_vlans
));
2463 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
2464 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
2465 * indicate this; newer upstream kernels use gratuitous ARP requests. */
2467 is_gratuitous_arp(const struct flow
*flow
, struct flow_wildcards
*wc
)
2469 if (flow
->dl_type
!= htons(ETH_TYPE_ARP
)) {
2473 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2474 if (!eth_addr_is_broadcast(flow
->dl_dst
)) {
2478 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
2479 if (flow
->nw_proto
== ARP_OP_REPLY
) {
2481 } else if (flow
->nw_proto
== ARP_OP_REQUEST
) {
2482 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
2483 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2485 return flow
->nw_src
== flow
->nw_dst
;
2491 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
2492 * dropped. Returns true if they may be forwarded, false if they should be
2495 * 'in_port' must be the xport that corresponds to flow->in_port.
2496 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
2498 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
2499 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
2500 * checked by input_vid_is_valid().
2502 * May also add tags to '*tags', although the current implementation only does
2503 * so in one special case.
2506 is_admissible(struct xlate_ctx
*ctx
, struct xport
*in_port
,
2509 struct xbundle
*in_xbundle
= in_port
->xbundle
;
2510 const struct xbridge
*xbridge
= ctx
->xbridge
;
2511 struct flow
*flow
= &ctx
->xin
->flow
;
2513 /* Drop frames for reserved multicast addresses
2514 * only if forward_bpdu option is absent. */
2515 if (!xbridge
->forward_bpdu
&& eth_addr_is_reserved(flow
->dl_dst
)) {
2516 xlate_report(ctx
, OFT_DETAIL
,
2517 "packet has reserved destination MAC, dropping");
2521 if (in_xbundle
->bond
) {
2522 struct mac_entry
*mac
;
2524 switch (bond_check_admissibility(in_xbundle
->bond
, in_port
->ofport
,
2530 xlate_report(ctx
, OFT_DETAIL
,
2531 "bonding refused admissibility, dropping");
2534 case BV_DROP_IF_MOVED
:
2535 ovs_rwlock_rdlock(&xbridge
->ml
->rwlock
);
2536 mac
= mac_learning_lookup(xbridge
->ml
, flow
->dl_src
, vlan
);
2538 && mac_entry_get_port(xbridge
->ml
, mac
) != in_xbundle
->ofbundle
2539 && (!is_gratuitous_arp(flow
, ctx
->wc
)
2540 || mac_entry_is_grat_arp_locked(mac
))) {
2541 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2542 xlate_report(ctx
, OFT_DETAIL
,
2543 "SLB bond thinks this packet looped back, "
2547 ovs_rwlock_unlock(&xbridge
->ml
->rwlock
);
2556 update_learning_table__(const struct xbridge
*xbridge
,
2557 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2558 int vlan
, bool is_grat_arp
)
2560 return (in_xbundle
== &ofpp_none_bundle
2561 || !mac_learning_update(xbridge
->ml
, dl_src
, vlan
,
2563 in_xbundle
->bond
!= NULL
,
2564 in_xbundle
->ofbundle
));
2568 update_learning_table(const struct xlate_ctx
*ctx
,
2569 struct xbundle
*in_xbundle
, struct eth_addr dl_src
,
2570 int vlan
, bool is_grat_arp
)
2572 if (!update_learning_table__(ctx
->xbridge
, in_xbundle
, dl_src
, vlan
,
2574 xlate_report_debug(ctx
, OFT_DETAIL
, "learned that "ETH_ADDR_FMT
" is "
2575 "on port %s in VLAN %d",
2576 ETH_ADDR_ARGS(dl_src
), in_xbundle
->name
, vlan
);
2580 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2581 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2583 update_mcast_snooping_table4__(const struct xlate_ctx
*ctx
,
2584 const struct flow
*flow
,
2585 struct mcast_snooping
*ms
, int vlan
,
2586 struct xbundle
*in_xbundle
,
2587 const struct dp_packet
*packet
)
2588 OVS_REQ_WRLOCK(ms
->rwlock
)
2590 const struct igmp_header
*igmp
;
2593 ovs_be32 ip4
= flow
->igmp_group_ip4
;
2595 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2596 igmp
= dp_packet_at(packet
, offset
, IGMP_HEADER_LEN
);
2597 if (!igmp
|| csum(igmp
, dp_packet_l4_size(packet
)) != 0) {
2598 xlate_report_debug(ctx
, OFT_DETAIL
,
2599 "multicast snooping received bad IGMP "
2600 "checksum on port %s in VLAN %d",
2601 in_xbundle
->name
, vlan
);
2605 switch (ntohs(flow
->tp_src
)) {
2606 case IGMP_HOST_MEMBERSHIP_REPORT
:
2607 case IGMPV2_HOST_MEMBERSHIP_REPORT
:
2608 if (mcast_snooping_add_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2609 xlate_report_debug(ctx
, OFT_DETAIL
,
2610 "multicast snooping learned that "
2611 IP_FMT
" is on port %s in VLAN %d",
2612 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2615 case IGMP_HOST_LEAVE_MESSAGE
:
2616 if (mcast_snooping_leave_group4(ms
, ip4
, vlan
, in_xbundle
->ofbundle
)) {
2617 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping leaving "
2618 IP_FMT
" is on port %s in VLAN %d",
2619 IP_ARGS(ip4
), in_xbundle
->name
, vlan
);
2622 case IGMP_HOST_MEMBERSHIP_QUERY
:
2623 if (flow
->nw_src
&& mcast_snooping_add_mrouter(ms
, vlan
,
2624 in_xbundle
->ofbundle
)) {
2625 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query "
2626 "from "IP_FMT
" is on port %s in VLAN %d",
2627 IP_ARGS(flow
->nw_src
), in_xbundle
->name
, vlan
);
2630 case IGMPV3_HOST_MEMBERSHIP_REPORT
:
2631 count
= mcast_snooping_add_report(ms
, packet
, vlan
,
2632 in_xbundle
->ofbundle
);
2634 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2635 "%d addresses on port %s in VLAN %d",
2636 count
, in_xbundle
->name
, vlan
);
2643 update_mcast_snooping_table6__(const struct xlate_ctx
*ctx
,
2644 const struct flow
*flow
,
2645 struct mcast_snooping
*ms
, int vlan
,
2646 struct xbundle
*in_xbundle
,
2647 const struct dp_packet
*packet
)
2648 OVS_REQ_WRLOCK(ms
->rwlock
)
2650 const struct mld_header
*mld
;
2654 offset
= (char *) dp_packet_l4(packet
) - (char *) dp_packet_data(packet
);
2655 mld
= dp_packet_at(packet
, offset
, MLD_HEADER_LEN
);
2658 packet_csum_upperlayer6(dp_packet_l3(packet
),
2659 mld
, IPPROTO_ICMPV6
,
2660 dp_packet_l4_size(packet
)) != 0) {
2661 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping received "
2662 "bad MLD checksum on port %s in VLAN %d",
2663 in_xbundle
->name
, vlan
);
2667 switch (ntohs(flow
->tp_src
)) {
2669 if (!ipv6_addr_equals(&flow
->ipv6_src
, &in6addr_any
)
2670 && mcast_snooping_add_mrouter(ms
, vlan
, in_xbundle
->ofbundle
)) {
2671 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping query on "
2672 "port %s in VLAN %d", in_xbundle
->name
, vlan
);
2678 count
= mcast_snooping_add_mld(ms
, packet
, vlan
, in_xbundle
->ofbundle
);
2680 xlate_report_debug(ctx
, OFT_DETAIL
, "multicast snooping processed "
2681 "%d addresses on port %s in VLAN %d",
2682 count
, in_xbundle
->name
, vlan
);
2688 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2689 * was received on 'in_xbundle' in 'vlan'. */
2691 update_mcast_snooping_table(const struct xlate_ctx
*ctx
,
2692 const struct flow
*flow
, int vlan
,
2693 struct xbundle
*in_xbundle
,
2694 const struct dp_packet
*packet
)
2696 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
2697 struct xbundle
*mcast_xbundle
;
2698 struct mcast_port_bundle
*fport
;
2700 /* Don't learn the OFPP_NONE port. */
2701 if (in_xbundle
== &ofpp_none_bundle
) {
2705 /* Don't learn from flood ports */
2706 mcast_xbundle
= NULL
;
2707 ovs_rwlock_wrlock(&ms
->rwlock
);
2708 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2709 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2710 if (mcast_xbundle
== in_xbundle
) {
2715 if (!mcast_xbundle
|| mcast_xbundle
!= in_xbundle
) {
2716 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2717 update_mcast_snooping_table4__(ctx
, flow
, ms
, vlan
,
2718 in_xbundle
, packet
);
2720 update_mcast_snooping_table6__(ctx
, flow
, ms
, vlan
,
2721 in_xbundle
, packet
);
2724 ovs_rwlock_unlock(&ms
->rwlock
);
2727 /* A list of multicast output ports.
2729 * We accumulate output ports and then do all the outputs afterward. It would
2730 * be more natural to do the outputs one at a time as we discover the need for
2731 * each one, but this can cause a deadlock because we need to take the
2732 * mcast_snooping's rwlock for reading to iterate through the port lists and
2733 * doing an output, if it goes to a patch port, can eventually come back to the
2734 * same mcast_snooping and attempt to take the write lock (see
2735 * https://github.com/openvswitch/ovs-issues/issues/153). */
2736 struct mcast_output
{
2737 /* Discrete ports. */
2738 struct xbundle
**xbundles
;
2739 size_t n
, allocated
;
2741 /* If set, flood to all ports. */
2744 #define MCAST_OUTPUT_INIT { NULL, 0, 0, false }
2746 /* Add 'mcast_bundle' to 'out'. */
2748 mcast_output_add(struct mcast_output
*out
, struct xbundle
*mcast_xbundle
)
2750 if (out
->n
>= out
->allocated
) {
2751 out
->xbundles
= x2nrealloc(out
->xbundles
, &out
->allocated
,
2752 sizeof *out
->xbundles
);
2754 out
->xbundles
[out
->n
++] = mcast_xbundle
;
2757 /* Outputs the packet in 'ctx' to all of the output ports in 'out', given input
2758 * bundle 'in_xbundle' and the current 'xvlan'. */
2760 mcast_output_finish(struct xlate_ctx
*ctx
, struct mcast_output
*out
,
2761 struct xbundle
*in_xbundle
, struct xvlan
*xvlan
)
2764 xlate_normal_flood(ctx
, in_xbundle
, xvlan
);
2766 for (size_t i
= 0; i
< out
->n
; i
++) {
2767 output_normal(ctx
, out
->xbundles
[i
], xvlan
);
2771 free(out
->xbundles
);
2774 /* send the packet to ports having the multicast group learned */
2776 xlate_normal_mcast_send_group(struct xlate_ctx
*ctx
,
2777 struct mcast_snooping
*ms OVS_UNUSED
,
2778 struct mcast_group
*grp
,
2779 struct xbundle
*in_xbundle
,
2780 struct mcast_output
*out
)
2781 OVS_REQ_RDLOCK(ms
->rwlock
)
2783 struct mcast_group_bundle
*b
;
2784 struct xbundle
*mcast_xbundle
;
2786 LIST_FOR_EACH(b
, bundle_node
, &grp
->bundle_lru
) {
2787 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, b
->port
);
2788 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2789 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast group port");
2790 mcast_output_add(out
, mcast_xbundle
);
2791 } else if (!mcast_xbundle
) {
2792 xlate_report(ctx
, OFT_WARN
,
2793 "mcast group port is unknown, dropping");
2795 xlate_report(ctx
, OFT_DETAIL
,
2796 "mcast group port is input port, dropping");
2801 /* send the packet to ports connected to multicast routers */
2803 xlate_normal_mcast_send_mrouters(struct xlate_ctx
*ctx
,
2804 struct mcast_snooping
*ms
,
2805 struct xbundle
*in_xbundle
,
2806 const struct xvlan
*xvlan
,
2807 struct mcast_output
*out
)
2808 OVS_REQ_RDLOCK(ms
->rwlock
)
2810 struct mcast_mrouter_bundle
*mrouter
;
2811 struct xbundle
*mcast_xbundle
;
2813 LIST_FOR_EACH(mrouter
, mrouter_node
, &ms
->mrouter_lru
) {
2814 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, mrouter
->port
);
2815 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
2816 && mrouter
->vlan
== xvlan
->v
[0].vid
) {
2817 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast router port");
2818 mcast_output_add(out
, mcast_xbundle
);
2819 } else if (!mcast_xbundle
) {
2820 xlate_report(ctx
, OFT_WARN
,
2821 "mcast router port is unknown, dropping");
2822 } else if (mrouter
->vlan
!= xvlan
->v
[0].vid
) {
2823 xlate_report(ctx
, OFT_DETAIL
,
2824 "mcast router is on another vlan, dropping");
2826 xlate_report(ctx
, OFT_DETAIL
,
2827 "mcast router port is input port, dropping");
2832 /* send the packet to ports flagged to be flooded */
2834 xlate_normal_mcast_send_fports(struct xlate_ctx
*ctx
,
2835 struct mcast_snooping
*ms
,
2836 struct xbundle
*in_xbundle
,
2837 struct mcast_output
*out
)
2838 OVS_REQ_RDLOCK(ms
->rwlock
)
2840 struct mcast_port_bundle
*fport
;
2841 struct xbundle
*mcast_xbundle
;
2843 LIST_FOR_EACH(fport
, node
, &ms
->fport_list
) {
2844 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, fport
->port
);
2845 if (mcast_xbundle
&& mcast_xbundle
!= in_xbundle
) {
2846 xlate_report(ctx
, OFT_DETAIL
, "forwarding to mcast flood port");
2847 mcast_output_add(out
, mcast_xbundle
);
2848 } else if (!mcast_xbundle
) {
2849 xlate_report(ctx
, OFT_WARN
,
2850 "mcast flood port is unknown, dropping");
2852 xlate_report(ctx
, OFT_DETAIL
,
2853 "mcast flood port is input port, dropping");
2858 /* forward the Reports to configured ports */
2860 xlate_normal_mcast_send_rports(struct xlate_ctx
*ctx
,
2861 struct mcast_snooping
*ms
,
2862 struct xbundle
*in_xbundle
,
2863 struct mcast_output
*out
)
2864 OVS_REQ_RDLOCK(ms
->rwlock
)
2866 struct mcast_port_bundle
*rport
;
2867 struct xbundle
*mcast_xbundle
;
2869 LIST_FOR_EACH(rport
, node
, &ms
->rport_list
) {
2870 mcast_xbundle
= xbundle_lookup(ctx
->xcfg
, rport
->port
);
2872 && mcast_xbundle
!= in_xbundle
2873 && mcast_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
2874 xlate_report(ctx
, OFT_DETAIL
,
2875 "forwarding report to mcast flagged port");
2876 mcast_output_add(out
, mcast_xbundle
);
2877 } else if (!mcast_xbundle
) {
2878 xlate_report(ctx
, OFT_WARN
,
2879 "mcast port is unknown, dropping the report");
2881 xlate_report(ctx
, OFT_DETAIL
,
2882 "mcast port is input port, dropping the Report");
2888 xlate_normal_flood(struct xlate_ctx
*ctx
, struct xbundle
*in_xbundle
,
2889 struct xvlan
*xvlan
)
2891 struct xbundle
*xbundle
;
2893 LIST_FOR_EACH (xbundle
, list_node
, &ctx
->xbridge
->xbundles
) {
2894 if (xbundle
!= in_xbundle
2895 && xbundle
->ofbundle
!= in_xbundle
->ofbundle
2896 && xbundle_includes_vlan(xbundle
, xvlan
)
2897 && xbundle
->floodable
2898 && !xbundle_mirror_out(ctx
->xbridge
, xbundle
)) {
2899 output_normal(ctx
, xbundle
, xvlan
);
2902 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2906 is_ip_local_multicast(const struct flow
*flow
, struct flow_wildcards
*wc
)
2908 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
2909 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
2910 return ip_is_local_multicast(flow
->nw_dst
);
2911 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
2912 memset(&wc
->masks
.ipv6_dst
, 0xff, sizeof wc
->masks
.ipv6_dst
);
2913 return ipv6_is_all_hosts(&flow
->ipv6_dst
);
2920 xlate_normal(struct xlate_ctx
*ctx
)
2922 struct flow_wildcards
*wc
= ctx
->wc
;
2923 struct flow
*flow
= &ctx
->xin
->flow
;
2924 struct xbundle
*in_xbundle
;
2925 struct xport
*in_port
;
2926 struct mac_entry
*mac
;
2928 struct xvlan in_xvlan
;
2932 memset(&wc
->masks
.dl_src
, 0xff, sizeof wc
->masks
.dl_src
);
2933 memset(&wc
->masks
.dl_dst
, 0xff, sizeof wc
->masks
.dl_dst
);
2934 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
2936 in_xbundle
= lookup_input_bundle(ctx
, flow
->in_port
.ofp_port
, &in_port
);
2938 xlate_report(ctx
, OFT_WARN
, "no input bundle, dropping");
2942 /* Drop malformed frames. */
2943 if (eth_type_vlan(flow
->dl_type
) &&
2944 !(flow
->vlans
[0].tci
& htons(VLAN_CFI
))) {
2945 if (ctx
->xin
->packet
!= NULL
) {
2946 xlate_report_error(ctx
, "dropping packet with partial "
2947 "VLAN tag received on port %s",
2950 xlate_report(ctx
, OFT_WARN
, "partial VLAN tag, dropping");
2954 /* Drop frames on bundles reserved for mirroring. */
2955 if (xbundle_mirror_out(ctx
->xbridge
, in_xbundle
)) {
2956 if (ctx
->xin
->packet
!= NULL
) {
2957 xlate_report_error(ctx
, "dropping packet received on port %s, "
2958 "which is reserved exclusively for mirroring",
2961 xlate_report(ctx
, OFT_WARN
,
2962 "input port is mirror output port, dropping");
2967 xvlan_extract(flow
, &in_xvlan
);
2968 if (!input_vid_is_valid(ctx
, in_xvlan
.v
[0].vid
, in_xbundle
)) {
2969 xlate_report(ctx
, OFT_WARN
,
2970 "disallowed VLAN VID for this input port, dropping");
2973 xvlan_input_translate(in_xbundle
, &in_xvlan
, &xvlan
);
2974 vlan
= xvlan
.v
[0].vid
;
2976 /* Check other admissibility requirements. */
2977 if (in_port
&& !is_admissible(ctx
, in_port
, vlan
)) {
2981 /* Learn source MAC. */
2982 bool is_grat_arp
= is_gratuitous_arp(flow
, wc
);
2983 if (ctx
->xin
->allow_side_effects
2984 && flow
->packet_type
== htonl(PT_ETH
)
2985 && in_port
->pt_mode
!= NETDEV_PT_LEGACY_L3
2987 update_learning_table(ctx
, in_xbundle
, flow
->dl_src
, vlan
,
2990 if (ctx
->xin
->xcache
&& in_xbundle
!= &ofpp_none_bundle
) {
2991 struct xc_entry
*entry
;
2993 /* Save just enough info to update mac learning table later. */
2994 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NORMAL
);
2995 entry
->normal
.ofproto
= ctx
->xbridge
->ofproto
;
2996 entry
->normal
.in_port
= flow
->in_port
.ofp_port
;
2997 entry
->normal
.dl_src
= flow
->dl_src
;
2998 entry
->normal
.vlan
= vlan
;
2999 entry
->normal
.is_gratuitous_arp
= is_grat_arp
;
3002 /* Determine output bundle. */
3003 if (mcast_snooping_enabled(ctx
->xbridge
->ms
)
3004 && !eth_addr_is_broadcast(flow
->dl_dst
)
3005 && eth_addr_is_multicast(flow
->dl_dst
)
3006 && is_ip_any(flow
)) {
3007 struct mcast_snooping
*ms
= ctx
->xbridge
->ms
;
3008 struct mcast_group
*grp
= NULL
;
3010 if (is_igmp(flow
, wc
)) {
3012 * IGMP packets need to take the slow path, in order to be
3013 * processed for mdb updates. That will prevent expires
3014 * firing off even after hosts have sent reports.
3016 ctx
->xout
->slow
|= SLOW_ACTION
;
3018 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
3019 if (mcast_snooping_is_membership(flow
->tp_src
) ||
3020 mcast_snooping_is_query(flow
->tp_src
)) {
3021 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3022 update_mcast_snooping_table(ctx
, flow
, vlan
,
3023 in_xbundle
, ctx
->xin
->packet
);
3027 if (mcast_snooping_is_membership(flow
->tp_src
)) {
3028 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3030 ovs_rwlock_rdlock(&ms
->rwlock
);
3031 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3033 /* RFC4541: section 2.1.1, item 1: A snooping switch should
3034 * forward IGMP Membership Reports only to those ports where
3035 * multicast routers are attached. Alternatively stated: a
3036 * snooping switch should not forward IGMP Membership Reports
3037 * to ports on which only hosts are attached.
3038 * An administrative control may be provided to override this
3039 * restriction, allowing the report messages to be flooded to
3041 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3042 ovs_rwlock_unlock(&ms
->rwlock
);
3044 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3046 xlate_report(ctx
, OFT_DETAIL
, "multicast traffic, flooding");
3047 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3050 } else if (is_mld(flow
, wc
)) {
3051 ctx
->xout
->slow
|= SLOW_ACTION
;
3052 if (ctx
->xin
->allow_side_effects
&& ctx
->xin
->packet
) {
3053 update_mcast_snooping_table(ctx
, flow
, vlan
,
3054 in_xbundle
, ctx
->xin
->packet
);
3056 if (is_mld_report(flow
, wc
)) {
3057 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3059 ovs_rwlock_rdlock(&ms
->rwlock
);
3060 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3062 xlate_normal_mcast_send_rports(ctx
, ms
, in_xbundle
, &out
);
3063 ovs_rwlock_unlock(&ms
->rwlock
);
3065 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3067 xlate_report(ctx
, OFT_DETAIL
, "MLD query, flooding");
3068 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3071 if (is_ip_local_multicast(flow
, wc
)) {
3072 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
3073 * address in the 224.0.0.x range which are not IGMP must
3074 * be forwarded on all ports */
3075 xlate_report(ctx
, OFT_DETAIL
,
3076 "RFC4541: section 2.1.2, item 2, flooding");
3077 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3082 /* forwarding to group base ports */
3083 struct mcast_output out
= MCAST_OUTPUT_INIT
;
3085 ovs_rwlock_rdlock(&ms
->rwlock
);
3086 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
3087 grp
= mcast_snooping_lookup4(ms
, flow
->nw_dst
, vlan
);
3088 } else if (flow
->dl_type
== htons(ETH_TYPE_IPV6
)) {
3089 grp
= mcast_snooping_lookup(ms
, &flow
->ipv6_dst
, vlan
);
3092 xlate_normal_mcast_send_group(ctx
, ms
, grp
, in_xbundle
, &out
);
3093 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3094 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3097 if (mcast_snooping_flood_unreg(ms
)) {
3098 xlate_report(ctx
, OFT_DETAIL
,
3099 "unregistered multicast, flooding");
3102 xlate_normal_mcast_send_mrouters(ctx
, ms
, in_xbundle
, &xvlan
,
3104 xlate_normal_mcast_send_fports(ctx
, ms
, in_xbundle
, &out
);
3107 ovs_rwlock_unlock(&ms
->rwlock
);
3109 mcast_output_finish(ctx
, &out
, in_xbundle
, &xvlan
);
3111 ovs_rwlock_rdlock(&ctx
->xbridge
->ml
->rwlock
);
3112 mac
= mac_learning_lookup(ctx
->xbridge
->ml
, flow
->dl_dst
, vlan
);
3113 mac_port
= mac
? mac_entry_get_port(ctx
->xbridge
->ml
, mac
) : NULL
;
3114 ovs_rwlock_unlock(&ctx
->xbridge
->ml
->rwlock
);
3117 struct xbundle
*mac_xbundle
= xbundle_lookup(ctx
->xcfg
, mac_port
);
3119 && mac_xbundle
!= in_xbundle
3120 && mac_xbundle
->ofbundle
!= in_xbundle
->ofbundle
) {
3121 xlate_report(ctx
, OFT_DETAIL
, "forwarding to learned port");
3122 output_normal(ctx
, mac_xbundle
, &xvlan
);
3123 } else if (!mac_xbundle
) {
3124 xlate_report(ctx
, OFT_WARN
,
3125 "learned port is unknown, dropping");
3127 xlate_report(ctx
, OFT_DETAIL
,
3128 "learned port is input port, dropping");
3131 xlate_report(ctx
, OFT_DETAIL
,
3132 "no learned MAC for destination, flooding");
3133 xlate_normal_flood(ctx
, in_xbundle
, &xvlan
);
3138 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
3139 * 'probability' is the number of packets out of UINT32_MAX to sample. The
3140 * 'cookie' is passed back in the callback for each sampled packet.
3141 * 'tunnel_out_port', if not ODPP_NONE, is added as the
3142 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions',
3143 * an OVS_USERSPACE_ATTR_ACTIONS attribute is added. If
3144 * 'emit_set_tunnel', sample(sampling_port=1) would translate into
3145 * datapath sample action set(tunnel(...)), sample(...) and it is used
3146 * for sampling egress tunnel information.
3149 compose_sample_action(struct xlate_ctx
*ctx
,
3150 const uint32_t probability
,
3151 const struct user_action_cookie
*cookie
,
3152 const odp_port_t tunnel_out_port
,
3153 bool include_actions
)
3155 if (probability
== 0) {
3156 /* No need to generate sampling or the inner action. */
3160 /* If the slow path meter is configured by the controller,
3161 * insert a meter action before the user space action. */
3162 struct ofproto
*ofproto
= &ctx
->xin
->ofproto
->up
;
3163 uint32_t meter_id
= ofproto
->slowpath_meter_id
;
3165 /* When meter action is not required, avoid generate sample action
3166 * for 100% sampling rate. */
3167 bool is_sample
= probability
< UINT32_MAX
|| meter_id
!= UINT32_MAX
;
3168 size_t sample_offset
= 0, actions_offset
= 0;
3170 sample_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3171 OVS_ACTION_ATTR_SAMPLE
);
3172 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
3174 actions_offset
= nl_msg_start_nested(ctx
->odp_actions
,
3175 OVS_SAMPLE_ATTR_ACTIONS
);
3178 if (meter_id
!= UINT32_MAX
) {
3179 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
3182 odp_port_t odp_port
= ofp_port_to_odp_port(
3183 ctx
->xbridge
, ctx
->xin
->flow
.in_port
.ofp_port
);
3184 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
3185 size_t cookie_offset
= odp_put_userspace_action(pid
, cookie
,
3192 nl_msg_end_nested(ctx
->odp_actions
, actions_offset
);
3193 nl_msg_end_nested(ctx
->odp_actions
, sample_offset
);
3196 return cookie_offset
;
3199 /* If sFLow is not enabled, returns 0 without doing anything.
3201 * If sFlow is enabled, appends a template "sample" action to the ODP actions
3202 * in 'ctx'. This action is a template because some of the information needed
3203 * to fill it out is not available until flow translation is complete. In this
3204 * case, this functions returns an offset, which is always nonzero, to pass
3205 * later to fix_sflow_action() to fill in the rest of the template. */
3207 compose_sflow_action(struct xlate_ctx
*ctx
)
3209 struct dpif_sflow
*sflow
= ctx
->xbridge
->sflow
;
3210 if (!sflow
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3214 struct user_action_cookie cookie
= {
3215 .type
= USER_ACTION_COOKIE_SFLOW
,
3216 .ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
,
3217 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
3219 return compose_sample_action(ctx
, dpif_sflow_get_probability(sflow
),
3220 &cookie
, ODPP_NONE
, true);
3223 /* If flow IPFIX is enabled, make sure IPFIX flow sample action
3224 * at egress point of tunnel port is just in front of corresponding
3225 * output action. If bridge IPFIX is enabled, this appends an IPFIX
3226 * sample action to 'ctx->odp_actions'. */
3228 compose_ipfix_action(struct xlate_ctx
*ctx
, odp_port_t output_odp_port
)
3230 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
3231 odp_port_t tunnel_out_port
= ODPP_NONE
;
3233 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
3237 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
3239 if (output_odp_port
== ODPP_NONE
&&
3240 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix
)) {
3244 /* For output case, output_odp_port is valid. */
3245 if (output_odp_port
!= ODPP_NONE
) {
3246 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix
)) {
3249 /* If tunnel sampling is enabled, put an additional option attribute:
3250 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
3252 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix
) &&
3253 dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
) ) {
3254 tunnel_out_port
= output_odp_port
;
3258 struct user_action_cookie cookie
= {
3259 .type
= USER_ACTION_COOKIE_IPFIX
,
3260 .ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
,
3261 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
3262 .ipfix
.output_odp_port
= output_odp_port
3264 compose_sample_action(ctx
,
3265 dpif_ipfix_get_bridge_exporter_probability(ipfix
),
3266 &cookie
, tunnel_out_port
, false);
3269 /* Fix "sample" action according to data collected while composing ODP actions,
3270 * as described in compose_sflow_action().
3272 * 'user_cookie_offset' must be the offset returned by
3273 * compose_sflow_action(). */
3275 fix_sflow_action(struct xlate_ctx
*ctx
, unsigned int user_cookie_offset
)
3277 const struct flow
*base
= &ctx
->base_flow
;
3278 struct user_action_cookie
*cookie
;
3280 cookie
= ofpbuf_at(ctx
->odp_actions
, user_cookie_offset
, sizeof *cookie
);
3281 ovs_assert(cookie
->type
== USER_ACTION_COOKIE_SFLOW
);
3283 cookie
->sflow
.vlan_tci
= base
->vlans
[0].tci
;
3285 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
3286 * port information") for the interpretation of cookie->output. */
3287 switch (ctx
->sflow_n_outputs
) {
3289 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
3290 cookie
->sflow
.output
= 0x40000000 | 256;
3294 cookie
->sflow
.output
= dpif_sflow_odp_port_to_ifindex(
3295 ctx
->xbridge
->sflow
, ctx
->sflow_odp_port
);
3296 if (cookie
->sflow
.output
) {
3301 /* 0x80000000 means "multiple output ports. */
3302 cookie
->sflow
.output
= 0x80000000 | ctx
->sflow_n_outputs
;
3308 process_special(struct xlate_ctx
*ctx
, const struct xport
*xport
)
3310 const struct flow
*flow
= &ctx
->xin
->flow
;
3311 struct flow_wildcards
*wc
= ctx
->wc
;
3312 const struct xbridge
*xbridge
= ctx
->xbridge
;
3313 const struct dp_packet
*packet
= ctx
->xin
->packet
;
3314 enum slow_path_reason slow
;
3315 bool lacp_may_enable
;
3319 } else if (xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
)) {
3321 cfm_process_heartbeat(xport
->cfm
, packet
);
3324 } else if (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
, wc
)) {
3326 bfd_process_packet(xport
->bfd
, flow
, packet
);
3327 /* If POLL received, immediately sends FINAL back. */
3328 if (bfd_should_send_packet(xport
->bfd
)) {
3329 ofproto_dpif_monitor_port_send_soon(xport
->ofport
);
3333 } else if (xport
->xbundle
&& xport
->xbundle
->lacp
3334 && flow
->dl_type
== htons(ETH_TYPE_LACP
)) {
3336 lacp_may_enable
= lacp_process_packet(xport
->xbundle
->lacp
,
3337 xport
->ofport
, packet
);
3338 /* Update LACP status in bond-slave to avoid packet-drops until
3339 * LACP state machine is run by the main thread. */
3340 if (xport
->xbundle
->bond
&& lacp_may_enable
) {
3341 bond_slave_set_may_enable(xport
->xbundle
->bond
, xport
->ofport
,
3346 } else if ((xbridge
->stp
|| xbridge
->rstp
) &&
3347 stp_should_process_flow(flow
, wc
)) {
3350 ? stp_process_packet(xport
, packet
)
3351 : rstp_process_packet(xport
, packet
);
3354 } else if (xport
->lldp
&& lldp_should_process_flow(xport
->lldp
, flow
)) {
3356 lldp_process_packet(xport
->lldp
, packet
);
3364 ctx
->xout
->slow
|= slow
;
3372 tnl_route_lookup_flow(const struct xlate_ctx
*ctx
,
3373 const struct flow
*oflow
,
3374 struct in6_addr
*ip
, struct in6_addr
*src
,
3375 struct xport
**out_port
)
3377 char out_dev
[IFNAMSIZ
];
3378 struct xbridge
*xbridge
;
3380 struct in6_addr dst
;
3382 dst
= flow_tnl_dst(&oflow
->tunnel
);
3383 if (!ovs_router_lookup(oflow
->pkt_mark
, &dst
, out_dev
, src
, &gw
)) {
3387 if (ipv6_addr_is_set(&gw
) &&
3388 (!IN6_IS_ADDR_V4MAPPED(&gw
) || in6_addr_get_mapped_ipv4(&gw
))) {
3394 HMAP_FOR_EACH (xbridge
, hmap_node
, &ctx
->xcfg
->xbridges
) {
3395 if (!strncmp(xbridge
->name
, out_dev
, IFNAMSIZ
)) {
3398 HMAP_FOR_EACH (port
, ofp_node
, &xbridge
->xports
) {
3399 if (!strncmp(netdev_get_name(port
->netdev
), out_dev
, IFNAMSIZ
)) {
3410 compose_table_xlate(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3411 struct dp_packet
*packet
)
3413 struct xbridge
*xbridge
= out_dev
->xbridge
;
3414 struct ofpact_output output
;
3417 ofpact_init(&output
.ofpact
, OFPACT_OUTPUT
, sizeof output
);
3418 flow_extract(packet
, &flow
);
3419 flow
.in_port
.ofp_port
= out_dev
->ofp_port
;
3420 output
.port
= OFPP_TABLE
;
3423 return ofproto_dpif_execute_actions__(xbridge
->ofproto
,
3424 ctx
->xin
->tables_version
, &flow
,
3425 NULL
, &output
.ofpact
, sizeof output
,
3426 ctx
->depth
, ctx
->resubmits
, packet
);
3430 tnl_send_nd_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3431 const struct eth_addr eth_src
,
3432 struct in6_addr
* ipv6_src
, struct in6_addr
* ipv6_dst
)
3434 struct dp_packet packet
;
3436 dp_packet_init(&packet
, 0);
3437 compose_nd_ns(&packet
, eth_src
, ipv6_src
, ipv6_dst
);
3438 compose_table_xlate(ctx
, out_dev
, &packet
);
3439 dp_packet_uninit(&packet
);
3443 tnl_send_arp_request(struct xlate_ctx
*ctx
, const struct xport
*out_dev
,
3444 const struct eth_addr eth_src
,
3445 ovs_be32 ip_src
, ovs_be32 ip_dst
)
3447 struct dp_packet packet
;
3449 dp_packet_init(&packet
, 0);
3450 compose_arp(&packet
, ARP_OP_REQUEST
,
3451 eth_src
, eth_addr_zero
, true, ip_src
, ip_dst
);
3453 compose_table_xlate(ctx
, out_dev
, &packet
);
3454 dp_packet_uninit(&packet
);
3458 propagate_tunnel_data_to_flow__(struct flow
*dst_flow
,
3459 const struct flow
*src_flow
,
3460 struct eth_addr dmac
, struct eth_addr smac
,
3461 struct in6_addr s_ip6
, ovs_be32 s_ip
,
3462 bool is_tnl_ipv6
, uint8_t nw_proto
)
3464 dst_flow
->dl_dst
= dmac
;
3465 dst_flow
->dl_src
= smac
;
3467 dst_flow
->packet_type
= htonl(PT_ETH
);
3468 dst_flow
->nw_dst
= src_flow
->tunnel
.ip_dst
;
3469 dst_flow
->nw_src
= src_flow
->tunnel
.ip_src
;
3470 dst_flow
->ipv6_dst
= src_flow
->tunnel
.ipv6_dst
;
3471 dst_flow
->ipv6_src
= src_flow
->tunnel
.ipv6_src
;
3473 dst_flow
->nw_frag
= 0; /* Tunnel packets are unfragmented. */
3474 dst_flow
->nw_tos
= src_flow
->tunnel
.ip_tos
;
3475 dst_flow
->nw_ttl
= src_flow
->tunnel
.ip_ttl
;
3476 dst_flow
->tp_dst
= src_flow
->tunnel
.tp_dst
;
3477 dst_flow
->tp_src
= src_flow
->tunnel
.tp_src
;
3480 dst_flow
->dl_type
= htons(ETH_TYPE_IPV6
);
3481 if (ipv6_mask_is_any(&dst_flow
->ipv6_src
)
3482 && !ipv6_mask_is_any(&s_ip6
)) {
3483 dst_flow
->ipv6_src
= s_ip6
;
3486 dst_flow
->dl_type
= htons(ETH_TYPE_IP
);
3487 if (dst_flow
->nw_src
== 0 && s_ip
) {
3488 dst_flow
->nw_src
= s_ip
;
3491 dst_flow
->nw_proto
= nw_proto
;
3495 * Populate the 'flow' and 'base_flow' L3 fields to do the post tunnel push
3499 propagate_tunnel_data_to_flow(struct xlate_ctx
*ctx
, struct eth_addr dmac
,
3500 struct eth_addr smac
, struct in6_addr s_ip6
,
3501 ovs_be32 s_ip
, bool is_tnl_ipv6
,
3502 enum ovs_vport_type tnl_type
)
3504 struct flow
*base_flow
, *flow
;
3505 flow
= &ctx
->xin
->flow
;
3506 base_flow
= &ctx
->base_flow
;
3507 uint8_t nw_proto
= 0;
3510 case OVS_VPORT_TYPE_GRE
:
3511 case OVS_VPORT_TYPE_ERSPAN
:
3512 case OVS_VPORT_TYPE_IP6ERSPAN
:
3513 case OVS_VPORT_TYPE_IP6GRE
:
3514 nw_proto
= IPPROTO_GRE
;
3516 case OVS_VPORT_TYPE_VXLAN
:
3517 case OVS_VPORT_TYPE_GENEVE
:
3518 nw_proto
= IPPROTO_UDP
;
3520 case OVS_VPORT_TYPE_LISP
:
3521 case OVS_VPORT_TYPE_STT
:
3522 case OVS_VPORT_TYPE_UNSPEC
:
3523 case OVS_VPORT_TYPE_NETDEV
:
3524 case OVS_VPORT_TYPE_INTERNAL
:
3525 case __OVS_VPORT_TYPE_MAX
:
3530 * Update base_flow first followed by flow as the dst_flow gets modified
3533 propagate_tunnel_data_to_flow__(base_flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3534 is_tnl_ipv6
, nw_proto
);
3535 propagate_tunnel_data_to_flow__(flow
, flow
, dmac
, smac
, s_ip6
, s_ip
,
3536 is_tnl_ipv6
, nw_proto
);
3540 native_tunnel_output(struct xlate_ctx
*ctx
, const struct xport
*xport
,
3541 const struct flow
*flow
, odp_port_t tunnel_odp_port
,
3544 struct netdev_tnl_build_header_params tnl_params
;
3545 struct ovs_action_push_tnl tnl_push_data
;
3546 struct xport
*out_dev
= NULL
;
3547 ovs_be32 s_ip
= 0, d_ip
= 0;
3548 struct in6_addr s_ip6
= in6addr_any
;
3549 struct in6_addr d_ip6
= in6addr_any
;
3550 struct eth_addr smac
;
3551 struct eth_addr dmac
;
3553 char buf_sip6
[INET6_ADDRSTRLEN
];
3554 char buf_dip6
[INET6_ADDRSTRLEN
];
3556 /* Store sFlow data. */
3557 uint32_t sflow_n_outputs
= ctx
->sflow_n_outputs
;
3559 /* Structures to backup Ethernet and IP of base_flow. */
3560 struct flow old_base_flow
;
3561 struct flow old_flow
;
3563 /* Backup flow & base_flow data. */
3564 memcpy(&old_base_flow
, &ctx
->base_flow
, sizeof old_base_flow
);
3565 memcpy(&old_flow
, &ctx
->xin
->flow
, sizeof old_flow
);
3567 if (flow
->tunnel
.ip_src
) {
3568 in6_addr_set_mapped_ipv4(&s_ip6
, flow
->tunnel
.ip_src
);
3571 err
= tnl_route_lookup_flow(ctx
, flow
, &d_ip6
, &s_ip6
, &out_dev
);
3573 xlate_report(ctx
, OFT_WARN
, "native tunnel routing failed");
3577 xlate_report(ctx
, OFT_DETAIL
, "tunneling to %s via %s",
3578 ipv6_string_mapped(buf_dip6
, &d_ip6
),
3579 netdev_get_name(out_dev
->netdev
));
3581 /* Use mac addr of bridge port of the peer. */
3582 err
= netdev_get_etheraddr(out_dev
->netdev
, &smac
);
3584 xlate_report(ctx
, OFT_WARN
,
3585 "tunnel output device lacks Ethernet address");
3589 d_ip
= in6_addr_get_mapped_ipv4(&d_ip6
);
3591 s_ip
= in6_addr_get_mapped_ipv4(&s_ip6
);
3594 err
= tnl_neigh_lookup(out_dev
->xbridge
->name
, &d_ip6
, &dmac
);
3596 xlate_report(ctx
, OFT_DETAIL
,
3597 "neighbor cache miss for %s on bridge %s, "
3598 "sending %s request",
3599 buf_dip6
, out_dev
->xbridge
->name
, d_ip
? "ARP" : "ND");
3601 tnl_send_arp_request(ctx
, out_dev
, smac
, s_ip
, d_ip
);
3603 tnl_send_nd_request(ctx
, out_dev
, smac
, &s_ip6
, &d_ip6
);
3608 if (ctx
->xin
->xcache
) {
3609 struct xc_entry
*entry
;
3611 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TNL_NEIGH
);
3612 ovs_strlcpy(entry
->tnl_neigh_cache
.br_name
, out_dev
->xbridge
->name
,
3613 sizeof entry
->tnl_neigh_cache
.br_name
);
3614 entry
->tnl_neigh_cache
.d_ipv6
= d_ip6
;
3617 xlate_report(ctx
, OFT_DETAIL
, "tunneling from "ETH_ADDR_FMT
" %s"
3618 " to "ETH_ADDR_FMT
" %s",
3619 ETH_ADDR_ARGS(smac
), ipv6_string_mapped(buf_sip6
, &s_ip6
),
3620 ETH_ADDR_ARGS(dmac
), buf_dip6
);
3622 netdev_init_tnl_build_header_params(&tnl_params
, flow
, &s_ip6
, dmac
, smac
);
3623 err
= tnl_port_build_header(xport
->ofport
, &tnl_push_data
, &tnl_params
);
3627 tnl_push_data
.tnl_port
= tunnel_odp_port
;
3628 tnl_push_data
.out_port
= out_dev
->odp_port
;
3630 /* After tunnel header has been added, MAC and IP data of flow and
3631 * base_flow need to be set properly, since there is not recirculation
3632 * any more when sending packet to tunnel. */
3634 propagate_tunnel_data_to_flow(ctx
, dmac
, smac
, s_ip6
,
3635 s_ip
, tnl_params
.is_ipv6
,
3636 tnl_push_data
.tnl_type
);
3638 size_t clone_ofs
= 0;
3639 size_t push_action_size
;
3641 clone_ofs
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
3642 odp_put_tnl_push_action(ctx
->odp_actions
, &tnl_push_data
);
3643 push_action_size
= ctx
->odp_actions
->size
;
3646 const struct dpif_flow_stats
*backup_resubmit_stats
;
3647 struct xlate_cache
*backup_xcache
;
3648 struct flow_wildcards
*backup_wc
, wc
;
3649 bool backup_side_effects
;
3650 const struct dp_packet
*backup_packet
;
3652 memset(&wc
, 0 , sizeof wc
);
3653 backup_wc
= ctx
->wc
;
3655 ctx
->xin
->wc
= NULL
;
3656 backup_resubmit_stats
= ctx
->xin
->resubmit_stats
;
3657 backup_xcache
= ctx
->xin
->xcache
;
3658 backup_side_effects
= ctx
->xin
->allow_side_effects
;
3659 backup_packet
= ctx
->xin
->packet
;
3661 ctx
->xin
->resubmit_stats
= NULL
;
3662 ctx
->xin
->xcache
= xlate_cache_new(); /* Use new temporary cache. */
3663 ctx
->xin
->allow_side_effects
= false;
3664 ctx
->xin
->packet
= NULL
;
3666 /* Push the cache entry for the tunnel first. */
3667 struct xc_entry
*entry
;
3668 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_TUNNEL_HEADER
);
3669 entry
->tunnel_hdr
.hdr_size
= tnl_push_data
.header_len
;
3670 entry
->tunnel_hdr
.operation
= ADD
;
3672 patch_port_output(ctx
, xport
, out_dev
);
3674 /* Similar to the stats update in revalidation, the x_cache entries
3675 * are populated by the previous translation are used to update the
3678 if (backup_resubmit_stats
) {
3679 struct dpif_flow_stats stats
= *backup_resubmit_stats
;
3680 xlate_push_stats(ctx
->xin
->xcache
, &stats
);
3682 xlate_cache_steal_entries(backup_xcache
, ctx
->xin
->xcache
);
3684 if (ctx
->odp_actions
->size
> push_action_size
) {
3685 nl_msg_end_non_empty_nested(ctx
->odp_actions
, clone_ofs
);
3687 nl_msg_cancel_nested(ctx
->odp_actions
, clone_ofs
);
3690 /* Restore context status. */
3691 ctx
->xin
->resubmit_stats
= backup_resubmit_stats
;
3692 xlate_cache_delete(ctx
->xin
->xcache
);
3693 ctx
->xin
->xcache
= backup_xcache
;
3694 ctx
->xin
->allow_side_effects
= backup_side_effects
;
3695 ctx
->xin
->packet
= backup_packet
;
3696 ctx
->wc
= backup_wc
;
3698 /* In order to maintain accurate stats, use recirc for
3699 * natvie tunneling. */
3700 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, 0);
3701 nl_msg_end_nested(ctx
->odp_actions
, clone_ofs
);
3704 /* Restore the flows after the translation. */
3705 memcpy(&ctx
->xin
->flow
, &old_flow
, sizeof ctx
->xin
->flow
);
3706 memcpy(&ctx
->base_flow
, &old_base_flow
, sizeof ctx
->base_flow
);
3708 /* Restore sFlow data. */
3709 ctx
->sflow_n_outputs
= sflow_n_outputs
;
3715 xlate_commit_actions(struct xlate_ctx
*ctx
)
3717 bool use_masked
= ctx
->xbridge
->support
.masked_set_action
;
3719 ctx
->xout
->slow
|= commit_odp_actions(&ctx
->xin
->flow
, &ctx
->base_flow
,
3720 ctx
->odp_actions
, ctx
->wc
,
3721 use_masked
, ctx
->pending_encap
,
3722 ctx
->pending_decap
, ctx
->encap_data
);
3723 ctx
->pending_encap
= false;
3724 ctx
->pending_decap
= false;
3725 ofpbuf_delete(ctx
->encap_data
);
3726 ctx
->encap_data
= NULL
;
3730 clear_conntrack(struct xlate_ctx
*ctx
)
3732 ctx
->conntracked
= false;
3733 flow_clear_conntrack(&ctx
->xin
->flow
);
3737 xlate_flow_is_protected(const struct xlate_ctx
*ctx
, const struct flow
*flow
, const struct xport
*xport_out
)
3739 const struct xport
*xport_in
;
3745 xport_in
= get_ofp_port(ctx
->xbridge
, flow
->in_port
.ofp_port
);
3747 return (xport_in
&& xport_in
->xbundle
&& xport_out
->xbundle
&&
3748 xport_in
->xbundle
->protected && xport_out
->xbundle
->protected);
3751 /* Function handles when a packet is sent from one bridge to another bridge.
3753 * The bridges are internally connected, either with patch ports or with
3756 * The output action to another bridge causes translation to continue within
3757 * the next bridge. This process can be recursive; the next bridge can
3758 * output yet to another bridge.
3760 * The translated actions from the second bridge onwards are enclosed within
3761 * the clone action, so that any modification to the packet will not be visible
3762 * to the remaining actions of the originating bridge.
3765 patch_port_output(struct xlate_ctx
*ctx
, const struct xport
*in_dev
,
3766 struct xport
*out_dev
)
3768 struct flow
*flow
= &ctx
->xin
->flow
;
3769 struct flow old_flow
= ctx
->xin
->flow
;
3770 struct flow_tnl old_flow_tnl_wc
= ctx
->wc
->masks
.tunnel
;
3771 bool old_conntrack
= ctx
->conntracked
;
3772 bool old_was_mpls
= ctx
->was_mpls
;
3773 ovs_version_t old_version
= ctx
->xin
->tables_version
;
3774 struct ofpbuf old_stack
= ctx
->stack
;
3775 uint8_t new_stack
[1024];
3776 struct ofpbuf old_action_set
= ctx
->action_set
;
3777 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
3778 uint64_t actset_stub
[1024 / 8];
3780 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
3781 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
3782 flow
->in_port
.ofp_port
= out_dev
->ofp_port
;
3783 flow
->metadata
= htonll(0);
3784 memset(&flow
->tunnel
, 0, sizeof flow
->tunnel
);
3785 memset(&ctx
->wc
->masks
.tunnel
, 0, sizeof ctx
->wc
->masks
.tunnel
);
3786 flow
->tunnel
.metadata
.tab
=
3787 ofproto_get_tun_tab(&out_dev
->xbridge
->ofproto
->up
);
3788 ctx
->wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
3789 memset(flow
->regs
, 0, sizeof flow
->regs
);
3790 flow
->actset_output
= OFPP_UNSET
;
3791 clear_conntrack(ctx
);
3792 ctx
->xin
->trace
= xlate_report(ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
3793 out_dev
->xbridge
->name
);
3794 mirror_mask_t old_mirrors
= ctx
->mirrors
;
3795 bool independent_mirrors
= out_dev
->xbridge
!= ctx
->xbridge
;
3796 if (independent_mirrors
) {
3799 ctx
->xbridge
= out_dev
->xbridge
;
3801 /* The bridge is now known so obtain its table version. */
3802 ctx
->xin
->tables_version
3803 = ofproto_dpif_get_tables_version(ctx
->xbridge
->ofproto
);
3805 if (!process_special(ctx
, out_dev
) && may_receive(out_dev
, ctx
)) {
3806 if (xport_stp_forward_state(out_dev
) &&
3807 xport_rstp_forward_state(out_dev
)) {
3808 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3809 false, true, clone_xlate_actions
);
3810 if (!ctx
->freezing
) {
3811 xlate_action_set(ctx
);
3813 if (ctx
->freezing
) {
3814 finish_freezing(ctx
);
3817 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
3818 * the learning action look at the packet, then drop it. */
3819 struct flow old_base_flow
= ctx
->base_flow
;
3820 size_t old_size
= ctx
->odp_actions
->size
;
3821 mirror_mask_t old_mirrors2
= ctx
->mirrors
;
3823 xlate_table_action(ctx
, flow
->in_port
.ofp_port
, 0, true, true,
3824 false, true, clone_xlate_actions
);
3825 ctx
->mirrors
= old_mirrors2
;
3826 ctx
->base_flow
= old_base_flow
;
3827 ctx
->odp_actions
->size
= old_size
;
3829 /* Undo changes that may have been done for freezing. */
3830 ctx_cancel_freeze(ctx
);
3834 ctx
->xin
->trace
= old_trace
;
3835 if (independent_mirrors
) {
3836 ctx
->mirrors
= old_mirrors
;
3838 ctx
->xin
->flow
= old_flow
;
3839 ctx
->xbridge
= in_dev
->xbridge
;
3840 ofpbuf_uninit(&ctx
->action_set
);
3841 ctx
->action_set
= old_action_set
;
3842 ofpbuf_uninit(&ctx
->stack
);
3843 ctx
->stack
= old_stack
;
3845 /* Restore calling bridge's lookup version. */
3846 ctx
->xin
->tables_version
= old_version
;
3848 /* Restore to calling bridge tunneling information */
3849 ctx
->wc
->masks
.tunnel
= old_flow_tnl_wc
;
3851 /* The out bridge popping MPLS should have no effect on the original
3853 ctx
->was_mpls
= old_was_mpls
;
3855 /* The out bridge's conntrack execution should have no effect on the
3856 * original bridge. */
3857 ctx
->conntracked
= old_conntrack
;
3859 /* The fact that the out bridge exits (for any reason) does not mean
3860 * that the original bridge should exit. Specifically, if the out
3861 * bridge freezes translation, the original bridge must continue
3862 * processing with the original, not the frozen packet! */
3865 /* Out bridge errors do not propagate back. */
3866 ctx
->error
= XLATE_OK
;
3868 if (ctx
->xin
->resubmit_stats
) {
3869 netdev_vport_inc_tx(in_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3870 netdev_vport_inc_rx(out_dev
->netdev
, ctx
->xin
->resubmit_stats
);
3872 bfd_account_rx(out_dev
->bfd
, ctx
->xin
->resubmit_stats
);
3875 if (ctx
->xin
->xcache
) {
3876 struct xc_entry
*entry
;
3878 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
3879 entry
->dev
.tx
= netdev_ref(in_dev
->netdev
);
3880 entry
->dev
.rx
= netdev_ref(out_dev
->netdev
);
3881 entry
->dev
.bfd
= bfd_ref(out_dev
->bfd
);
3886 check_output_prerequisites(struct xlate_ctx
*ctx
,
3887 const struct xport
*xport
,
3891 struct flow_wildcards
*wc
= ctx
->wc
;
3894 xlate_report(ctx
, OFT_WARN
, "Nonexistent output port");
3896 } else if (xport
->config
& OFPUTIL_PC_NO_FWD
) {
3897 xlate_report(ctx
, OFT_DETAIL
, "OFPPC_NO_FWD set, skipping output");
3899 } else if (ctx
->mirror_snaplen
!= 0 && xport
->odp_port
== ODPP_NONE
) {
3900 xlate_report(ctx
, OFT_WARN
,
3901 "Mirror truncate to ODPP_NONE, skipping output");
3903 } else if (xlate_flow_is_protected(ctx
, flow
, xport
)) {
3904 xlate_report(ctx
, OFT_WARN
,
3905 "Flow is between protected ports, skipping output.");
3907 } else if (check_stp
) {
3908 if (is_stp(&ctx
->base_flow
)) {
3909 if (!xport_stp_should_forward_bpdu(xport
) &&
3910 !xport_rstp_should_manage_bpdu(xport
)) {
3911 if (ctx
->xbridge
->stp
!= NULL
) {
3912 xlate_report(ctx
, OFT_WARN
,
3913 "STP not in listening state, "
3914 "skipping bpdu output");
3915 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3916 xlate_report(ctx
, OFT_WARN
,
3917 "RSTP not managing BPDU in this state, "
3918 "skipping bpdu output");
3922 } else if ((xport
->cfm
&& cfm_should_process_flow(xport
->cfm
, flow
, wc
))
3923 || (xport
->bfd
&& bfd_should_process_flow(xport
->bfd
, flow
,
3925 /* Pass; STP should not block link health detection. */
3926 } else if (!xport_stp_forward_state(xport
) ||
3927 !xport_rstp_forward_state(xport
)) {
3928 if (ctx
->xbridge
->stp
!= NULL
) {
3929 xlate_report(ctx
, OFT_WARN
,
3930 "STP not in forwarding state, skipping output");
3931 } else if (ctx
->xbridge
->rstp
!= NULL
) {
3932 xlate_report(ctx
, OFT_WARN
,
3933 "RSTP not in forwarding state, skipping output");
3939 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L2
&&
3940 flow
->packet_type
!= htonl(PT_ETH
)) {
3941 xlate_report(ctx
, OFT_WARN
, "Trying to send non-Ethernet packet "
3942 "through legacy L2 port. Dropping packet.");
3949 /* Function verifies if destination address of received Neighbor Advertisement
3950 * message stored in 'flow' is correct. It should be either FF02::1:FFXX:XXXX
3951 * where XX:XXXX stands for the last 24 bits of 'ipv6_addr' or it should match
3954 is_nd_dst_correct(const struct flow
*flow
, const struct in6_addr
*ipv6_addr
)
3956 const uint8_t *flow_ipv6_addr
= (uint8_t *) &flow
->ipv6_dst
;
3957 const uint8_t *addr
= (uint8_t *) ipv6_addr
;
3959 return (IN6_IS_ADDR_MC_LINKLOCAL(&flow
->ipv6_dst
) &&
3960 flow_ipv6_addr
[11] == 0x01 &&
3961 flow_ipv6_addr
[12] == 0xff &&
3962 flow_ipv6_addr
[13] == addr
[13] &&
3963 flow_ipv6_addr
[14] == addr
[14] &&
3964 flow_ipv6_addr
[15] == addr
[15]) ||
3965 IN6_ARE_ADDR_EQUAL(&flow
->ipv6_dst
, ipv6_addr
);
3968 /* Function verifies if the ARP reply or Neighbor Advertisement represented by
3969 * 'flow' addresses the 'xbridge' of 'ctx'. Returns true if the ARP TA or
3970 * neighbor discovery destination is in the list of configured IP addresses of
3971 * the bridge. Otherwise, it returns false. */
3973 is_neighbor_reply_correct(const struct xlate_ctx
*ctx
, const struct flow
*flow
)
3977 struct xbridge_addr
*xbridge_addr
= xbridge_addr_ref(ctx
->xbridge
->addr
);
3979 /* Verify if 'nw_dst' of ARP or 'ipv6_dst' of ICMPV6 is in the list. */
3980 for (i
= 0; xbridge_addr
&& i
< xbridge_addr
->n_addr
; i
++) {
3981 struct in6_addr
*ip_addr
= &xbridge_addr
->addr
[i
];
3982 if ((IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
3983 flow
->dl_type
== htons(ETH_TYPE_ARP
) &&
3984 in6_addr_get_mapped_ipv4(ip_addr
) == flow
->nw_dst
) ||
3985 (!IN6_IS_ADDR_V4MAPPED(ip_addr
) &&
3986 is_nd_dst_correct(flow
, ip_addr
))) {
3987 /* Found a match. */
3993 xbridge_addr_unref(xbridge_addr
);
3998 terminate_native_tunnel(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
3999 struct flow
*flow
, struct flow_wildcards
*wc
,
4000 odp_port_t
*tnl_port
)
4002 *tnl_port
= ODPP_NONE
;
4004 /* XXX: Write better Filter for tunnel port. We can use in_port
4005 * in tunnel-port flow to avoid these checks completely. */
4006 if (ofp_port
== OFPP_LOCAL
&&
4007 ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4008 *tnl_port
= tnl_port_map_lookup(flow
, wc
);
4010 /* If no tunnel port was found and it's about an ARP or ICMPv6 packet,
4011 * do tunnel neighbor snooping. */
4012 if (*tnl_port
== ODPP_NONE
&&
4013 (flow
->dl_type
== htons(ETH_TYPE_ARP
) ||
4014 flow
->nw_proto
== IPPROTO_ICMPV6
) &&
4015 is_neighbor_reply_correct(ctx
, flow
)) {
4016 tnl_neigh_snoop(flow
, wc
, ctx
->xbridge
->name
);
4020 return *tnl_port
!= ODPP_NONE
;
4024 compose_output_action__(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4025 const struct xlate_bond_recirc
*xr
, bool check_stp
,
4026 bool is_last_action OVS_UNUSED
, bool truncate
)
4028 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, ofp_port
);
4029 struct flow_wildcards
*wc
= ctx
->wc
;
4030 struct flow
*flow
= &ctx
->xin
->flow
;
4031 struct flow_tnl flow_tnl
;
4032 union flow_vlan_hdr flow_vlans
[FLOW_MAX_VLAN_HEADERS
];
4033 uint8_t flow_nw_tos
;
4034 odp_port_t out_port
, odp_port
, odp_tnl_port
;
4035 bool is_native_tunnel
= false;
4037 struct eth_addr flow_dl_dst
= flow
->dl_dst
;
4038 struct eth_addr flow_dl_src
= flow
->dl_src
;
4039 ovs_be32 flow_packet_type
= flow
->packet_type
;
4040 ovs_be16 flow_dl_type
= flow
->dl_type
;
4042 /* If 'struct flow' gets additional metadata, we'll need to zero it out
4043 * before traversing a patch port. */
4044 BUILD_ASSERT_DECL(FLOW_WC_SEQ
== 41);
4045 memset(&flow_tnl
, 0, sizeof flow_tnl
);
4047 if (!check_output_prerequisites(ctx
, xport
, flow
, check_stp
)) {
4051 if (flow
->packet_type
== htonl(PT_ETH
)) {
4052 /* Strip Ethernet header for legacy L3 port. */
4053 if (xport
->pt_mode
== NETDEV_PT_LEGACY_L3
) {
4054 flow
->packet_type
= PACKET_TYPE_BE(OFPHTN_ETHERTYPE
,
4055 ntohs(flow
->dl_type
));
4061 xlate_report_error(ctx
, "Cannot truncate output to patch port");
4063 patch_port_output(ctx
, xport
, xport
->peer
);
4067 memcpy(flow_vlans
, flow
->vlans
, sizeof flow_vlans
);
4068 flow_nw_tos
= flow
->nw_tos
;
4070 if (count_skb_priorities(xport
)) {
4071 memset(&wc
->masks
.skb_priority
, 0xff, sizeof wc
->masks
.skb_priority
);
4072 if (dscp_from_skb_priority(xport
, flow
->skb_priority
, &dscp
)) {
4073 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
4074 flow
->nw_tos
&= ~IP_DSCP_MASK
;
4075 flow
->nw_tos
|= dscp
;
4079 if (xport
->is_tunnel
) {
4080 struct in6_addr dst
;
4081 /* Save tunnel metadata so that changes made due to
4082 * the Logical (tunnel) Port are not visible for any further
4083 * matches, while explicit set actions on tunnel metadata are.
4085 flow_tnl
= flow
->tunnel
;
4086 odp_port
= tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
4087 if (odp_port
== ODPP_NONE
) {
4088 xlate_report(ctx
, OFT_WARN
, "Tunneling decided against output");
4089 goto out
; /* restore flow_nw_tos */
4091 dst
= flow_tnl_dst(&flow
->tunnel
);
4092 if (ipv6_addr_equals(&dst
, &ctx
->orig_tunnel_ipv6_dst
)) {
4093 xlate_report(ctx
, OFT_WARN
, "Not tunneling to our own address");
4094 goto out
; /* restore flow_nw_tos */
4096 if (ctx
->xin
->resubmit_stats
) {
4097 netdev_vport_inc_tx(xport
->netdev
, ctx
->xin
->resubmit_stats
);
4099 if (ctx
->xin
->xcache
) {
4100 struct xc_entry
*entry
;
4102 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_NETDEV
);
4103 entry
->dev
.tx
= netdev_ref(xport
->netdev
);
4105 out_port
= odp_port
;
4106 if (ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
4107 xlate_report(ctx
, OFT_DETAIL
, "output to native tunnel");
4108 is_native_tunnel
= true;
4110 const char *tnl_type
;
4112 xlate_report(ctx
, OFT_DETAIL
, "output to kernel tunnel");
4113 tnl_type
= tnl_port_get_type(xport
->ofport
);
4114 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
4115 ctx
->odp_actions
, tnl_type
);
4116 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4119 odp_port
= xport
->odp_port
;
4120 out_port
= odp_port
;
4123 if (out_port
!= ODPP_NONE
) {
4124 /* Commit accumulated flow updates before output. */
4125 xlate_commit_actions(ctx
);
4128 /* Recirculate the packet. */
4129 struct ovs_action_hash
*act_hash
;
4132 enum ovs_hash_alg hash_alg
= xr
->hash_alg
;
4133 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4134 /* Algorithm supported by all datapaths. */
4135 hash_alg
= OVS_HASH_ALG_L4
;
4137 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4138 OVS_ACTION_ATTR_HASH
,
4140 act_hash
->hash_alg
= hash_alg
;
4141 act_hash
->hash_basis
= xr
->hash_basis
;
4143 /* Recirc action. */
4144 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
,
4146 } else if (is_native_tunnel
) {
4147 /* Output to native tunnel port. */
4148 native_tunnel_output(ctx
, xport
, flow
, odp_port
, truncate
);
4149 flow
->tunnel
= flow_tnl
; /* Restore tunnel metadata */
4151 } else if (terminate_native_tunnel(ctx
, ofp_port
, flow
, wc
,
4153 /* Intercept packet to be received on native tunnel port. */
4154 nl_msg_put_odp_port(ctx
->odp_actions
, OVS_ACTION_ATTR_TUNNEL_POP
,
4158 /* Tunnel push-pop action is not compatible with
4160 compose_ipfix_action(ctx
, out_port
);
4162 /* Handle truncation of the mirrored packet. */
4163 if (ctx
->mirror_snaplen
> 0 &&
4164 ctx
->mirror_snaplen
< UINT16_MAX
) {
4165 struct ovs_action_trunc
*trunc
;
4167 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4168 OVS_ACTION_ATTR_TRUNC
,
4170 trunc
->max_len
= ctx
->mirror_snaplen
;
4171 if (!ctx
->xbridge
->support
.trunc
) {
4172 ctx
->xout
->slow
|= SLOW_ACTION
;
4176 nl_msg_put_odp_port(ctx
->odp_actions
,
4177 OVS_ACTION_ATTR_OUTPUT
,
4181 ctx
->sflow_odp_port
= odp_port
;
4182 ctx
->sflow_n_outputs
++;
4183 ctx
->nf_output_iface
= ofp_port
;
4186 if (mbridge_has_mirrors(ctx
->xbridge
->mbridge
) && xport
->xbundle
) {
4187 mirror_packet(ctx
, xport
->xbundle
,
4188 xbundle_mirror_dst(xport
->xbundle
->xbridge
,
4194 memcpy(flow
->vlans
, flow_vlans
, sizeof flow
->vlans
);
4195 flow
->nw_tos
= flow_nw_tos
;
4196 flow
->dl_dst
= flow_dl_dst
;
4197 flow
->dl_src
= flow_dl_src
;
4198 flow
->packet_type
= flow_packet_type
;
4199 flow
->dl_type
= flow_dl_type
;
4203 compose_output_action(struct xlate_ctx
*ctx
, ofp_port_t ofp_port
,
4204 const struct xlate_bond_recirc
*xr
,
4205 bool is_last_action
, bool truncate
)
4207 compose_output_action__(ctx
, ofp_port
, xr
, true,
4208 is_last_action
, truncate
);
4212 xlate_recursively(struct xlate_ctx
*ctx
, struct rule_dpif
*rule
,
4213 bool deepens
, bool is_last_action
,
4214 xlate_actions_handler
*actions_xlator
)
4216 struct rule_dpif
*old_rule
= ctx
->rule
;
4217 ovs_be64 old_cookie
= ctx
->rule_cookie
;
4218 const struct rule_actions
*actions
;
4220 if (ctx
->xin
->resubmit_stats
) {
4221 rule_dpif_credit_stats(rule
, ctx
->xin
->resubmit_stats
);
4226 ctx
->depth
+= deepens
;
4228 ctx
->rule_cookie
= rule
->up
.flow_cookie
;
4229 actions
= rule_get_actions(&rule
->up
);
4230 actions_xlator(actions
->ofpacts
, actions
->ofpacts_len
, ctx
,
4231 is_last_action
, false);
4232 ctx
->rule_cookie
= old_cookie
;
4233 ctx
->rule
= old_rule
;
4234 ctx
->depth
-= deepens
;
4238 xlate_resubmit_resource_check(struct xlate_ctx
*ctx
)
4240 if (ctx
->depth
>= MAX_DEPTH
) {
4241 xlate_report_error(ctx
, "over max translation depth %d", MAX_DEPTH
);
4242 ctx
->error
= XLATE_RECURSION_TOO_DEEP
;
4243 } else if (ctx
->resubmits
>= MAX_RESUBMITS
) {
4244 xlate_report_error(ctx
, "over %d resubmit actions", MAX_RESUBMITS
);
4245 ctx
->error
= XLATE_TOO_MANY_RESUBMITS
;
4246 } else if (ctx
->odp_actions
->size
> UINT16_MAX
) {
4247 xlate_report_error(ctx
, "resubmits yielded over 64 kB of actions");
4248 /* NOT an error, as we'll be slow-pathing the flow in this case? */
4249 ctx
->exit
= true; /* XXX: translation still terminated! */
4250 } else if (ctx
->stack
.size
>= 65536) {
4251 xlate_report_error(ctx
, "resubmits yielded over 64 kB of stack");
4252 ctx
->error
= XLATE_STACK_TOO_DEEP
;
4261 tuple_swap_flow(struct flow
*flow
, bool ipv4
)
4263 uint8_t nw_proto
= flow
->nw_proto
;
4264 flow
->nw_proto
= flow
->ct_nw_proto
;
4265 flow
->ct_nw_proto
= nw_proto
;
4268 ovs_be32 nw_src
= flow
->nw_src
;
4269 flow
->nw_src
= flow
->ct_nw_src
;
4270 flow
->ct_nw_src
= nw_src
;
4272 ovs_be32 nw_dst
= flow
->nw_dst
;
4273 flow
->nw_dst
= flow
->ct_nw_dst
;
4274 flow
->ct_nw_dst
= nw_dst
;
4276 struct in6_addr ipv6_src
= flow
->ipv6_src
;
4277 flow
->ipv6_src
= flow
->ct_ipv6_src
;
4278 flow
->ct_ipv6_src
= ipv6_src
;
4280 struct in6_addr ipv6_dst
= flow
->ipv6_dst
;
4281 flow
->ipv6_dst
= flow
->ct_ipv6_dst
;
4282 flow
->ct_ipv6_dst
= ipv6_dst
;
4285 ovs_be16 tp_src
= flow
->tp_src
;
4286 flow
->tp_src
= flow
->ct_tp_src
;
4287 flow
->ct_tp_src
= tp_src
;
4289 ovs_be16 tp_dst
= flow
->tp_dst
;
4290 flow
->tp_dst
= flow
->ct_tp_dst
;
4291 flow
->ct_tp_dst
= tp_dst
;
4295 tuple_swap(struct flow
*flow
, struct flow_wildcards
*wc
)
4297 bool ipv4
= (flow
->dl_type
== htons(ETH_TYPE_IP
));
4299 tuple_swap_flow(flow
, ipv4
);
4300 tuple_swap_flow(&wc
->masks
, ipv4
);
4304 xlate_table_action(struct xlate_ctx
*ctx
, ofp_port_t in_port
, uint8_t table_id
,
4305 bool may_packet_in
, bool honor_table_miss
,
4306 bool with_ct_orig
, bool is_last_action
,
4307 xlate_actions_handler
*xlator
)
4309 /* Check if we need to recirculate before matching in a table. */
4310 if (ctx
->was_mpls
) {
4311 ctx_trigger_freeze(ctx
);
4314 if (xlate_resubmit_resource_check(ctx
)) {
4315 uint8_t old_table_id
= ctx
->table_id
;
4316 struct rule_dpif
*rule
;
4318 ctx
->table_id
= table_id
;
4320 /* Swap packet fields with CT 5-tuple if requested. */
4322 /* Do not swap if there is no CT tuple, or if key is not IP. */
4323 if (ctx
->xin
->flow
.ct_nw_proto
== 0 ||
4324 !is_ip_any(&ctx
->xin
->flow
)) {
4325 xlate_report_error(ctx
,
4326 "resubmit(ct) with non-tracked or non-IP packet!");
4329 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4331 rule
= rule_dpif_lookup_from_table(ctx
->xbridge
->ofproto
,
4332 ctx
->xin
->tables_version
,
4333 &ctx
->xin
->flow
, ctx
->wc
,
4334 ctx
->xin
->resubmit_stats
,
4335 &ctx
->table_id
, in_port
,
4336 may_packet_in
, honor_table_miss
,
4340 tuple_swap(&ctx
->xin
->flow
, ctx
->wc
);
4344 /* Fill in the cache entry here instead of xlate_recursively
4345 * to make the reference counting more explicit. We take a
4346 * reference in the lookups above if we are going to cache the
4348 if (ctx
->xin
->xcache
) {
4349 struct xc_entry
*entry
;
4351 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_RULE
);
4353 ofproto_rule_ref(&rule
->up
);
4356 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4357 xlate_report_table(ctx
, rule
, table_id
);
4358 xlate_recursively(ctx
, rule
, table_id
<= old_table_id
,
4359 is_last_action
, xlator
);
4360 ctx
->xin
->trace
= old_trace
;
4363 ctx
->table_id
= old_table_id
;
4368 /* Consumes the group reference, which is only taken if xcache exists. */
4370 xlate_group_stats(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4371 struct ofputil_bucket
*bucket
)
4373 if (ctx
->xin
->resubmit_stats
) {
4374 group_dpif_credit_stats(group
, bucket
, ctx
->xin
->resubmit_stats
);
4376 if (ctx
->xin
->xcache
) {
4377 struct xc_entry
*entry
;
4379 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_GROUP
);
4380 entry
->group
.group
= group
;
4381 entry
->group
.bucket
= bucket
;
4386 xlate_group_bucket(struct xlate_ctx
*ctx
, struct ofputil_bucket
*bucket
,
4387 bool is_last_action
)
4389 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
4390 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
4391 char *s
= xasprintf("bucket %"PRIu32
, bucket
->bucket_id
);
4392 ctx
->xin
->trace
= &oftrace_report(ctx
->xin
->trace
, OFT_BUCKET
,
4397 uint64_t action_list_stub
[1024 / 8];
4398 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
4399 struct ofpbuf action_set
= ofpbuf_const_initializer(bucket
->ofpacts
,
4400 bucket
->ofpacts_len
);
4401 struct flow old_flow
= ctx
->xin
->flow
;
4402 bool old_was_mpls
= ctx
->was_mpls
;
4404 ofpacts_execute_action_set(&action_list
, &action_set
);
4406 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, is_last_action
,
4410 ofpbuf_uninit(&action_list
);
4412 /* Check if need to freeze. */
4413 if (ctx
->freezing
) {
4414 finish_freezing(ctx
);
4417 /* Roll back flow to previous state.
4418 * This is equivalent to cloning the packet for each bucket.
4420 * As a side effect any subsequently applied actions will
4421 * also effectively be applied to a clone of the packet taken
4422 * just before applying the all or indirect group.
4424 * Note that group buckets are action sets, hence they cannot modify the
4425 * main action set. Also any stack actions are ignored when executing an
4426 * action set, so group buckets cannot change the stack either.
4427 * However, we do allow resubmit actions in group buckets, which could
4428 * break the above assumptions. It is up to the controller to not mess up
4429 * with the action_set and stack in the tables resubmitted to from
4431 ctx
->xin
->flow
= old_flow
;
4433 /* The group bucket popping MPLS should have no effect after bucket
4435 ctx
->was_mpls
= old_was_mpls
;
4437 /* The fact that the group bucket exits (for any reason) does not mean that
4438 * the translation after the group action should exit. Specifically, if
4439 * the group bucket freezes translation, the actions after the group action
4440 * must continue processing with the original, not the frozen packet! */
4443 /* Context error in a bucket should not impact processing of other buckets
4444 * or actions. This is similar to cloning a packet for group buckets.
4445 * There is no need to restore the error back to old value due to the fact
4446 * that we actually processed group action which can happen only when there
4447 * is no previous context error.
4449 * Exception to above is errors which are system limits to protect
4450 * translation from running too long or occupy too much space. These errors
4451 * should not be masked. XLATE_RECURSION_TOO_DEEP, XLATE_TOO_MANY_RESUBMITS
4452 * and XLATE_STACK_TOO_DEEP fall in this category. */
4453 if (ctx
->error
== XLATE_TOO_MANY_MPLS_LABELS
||
4454 ctx
->error
== XLATE_UNSUPPORTED_PACKET_TYPE
) {
4455 /* reset the error and continue processing other buckets */
4456 ctx
->error
= XLATE_OK
;
4459 ctx
->xin
->trace
= old_trace
;
4462 static struct ofputil_bucket
*
4463 pick_ff_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4465 return group_first_live_bucket(ctx
, group
, 0);
4468 static struct ofputil_bucket
*
4469 pick_default_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4471 flow_mask_hash_fields(&ctx
->xin
->flow
, ctx
->wc
,
4472 NX_HASH_FIELDS_SYMMETRIC_L4
);
4473 return group_best_live_bucket(ctx
, group
,
4474 flow_hash_symmetric_l4(&ctx
->xin
->flow
, 0));
4477 static struct ofputil_bucket
*
4478 pick_hash_fields_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4480 const struct field_array
*fields
= &group
->up
.props
.fields
;
4481 const uint8_t *mask_values
= fields
->values
;
4482 uint32_t basis
= hash_uint64(group
->up
.props
.selection_method_param
);
4485 BITMAP_FOR_EACH_1 (i
, MFF_N_IDS
, fields
->used
.bm
) {
4486 const struct mf_field
*mf
= mf_from_id(i
);
4488 /* Skip fields for which prerequisites are not met. */
4489 if (!mf_are_prereqs_ok(mf
, &ctx
->xin
->flow
, ctx
->wc
)) {
4490 /* Skip the mask bytes for this field. */
4491 mask_values
+= mf
->n_bytes
;
4495 union mf_value value
;
4496 union mf_value mask
;
4498 mf_get_value(mf
, &ctx
->xin
->flow
, &value
);
4499 /* Mask the value. */
4500 for (int j
= 0; j
< mf
->n_bytes
; j
++) {
4501 mask
.b
[j
] = *mask_values
++;
4502 value
.b
[j
] &= mask
.b
[j
];
4504 basis
= hash_bytes(&value
, mf
->n_bytes
, basis
);
4506 /* For tunnels, hash in whether the field is present. */
4507 if (mf_is_tun_metadata(mf
)) {
4508 basis
= hash_boolean(mf_is_set(mf
, &ctx
->xin
->flow
), basis
);
4511 mf_mask_field_masked(mf
, &mask
, ctx
->wc
);
4514 return group_best_live_bucket(ctx
, group
, basis
);
4517 static struct ofputil_bucket
*
4518 pick_dp_hash_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4520 uint32_t dp_hash
= ctx
->xin
->flow
.dp_hash
;
4522 /* dp_hash value 0 is special since it means that the dp_hash has not been
4523 * computed, as all computed dp_hash values are non-zero. Therefore
4524 * compare to zero can be used to decide if the dp_hash value is valid
4525 * without masking the dp_hash field. */
4527 enum ovs_hash_alg hash_alg
= group
->hash_alg
;
4528 if (hash_alg
> ctx
->xbridge
->support
.max_hash_alg
) {
4529 /* Algorithm supported by all datapaths. */
4530 hash_alg
= OVS_HASH_ALG_L4
;
4532 ctx_trigger_recirculate_with_hash(ctx
, hash_alg
, group
->hash_basis
);
4535 uint32_t hash_mask
= group
->hash_mask
;
4536 ctx
->wc
->masks
.dp_hash
|= hash_mask
;
4538 /* Starting from the original masked dp_hash value iterate over the
4539 * hash mapping table to find the first live bucket. As the buckets
4540 * are quasi-randomly spread over the hash values, this maintains
4541 * a distribution according to bucket weights even when some buckets
4543 for (int i
= 0; i
<= hash_mask
; i
++) {
4544 struct ofputil_bucket
*b
=
4545 group
->hash_map
[(dp_hash
+ i
) & hash_mask
];
4546 if (bucket_is_alive(ctx
, b
, 0)) {
4555 static struct ofputil_bucket
*
4556 pick_select_group(struct xlate_ctx
*ctx
, struct group_dpif
*group
)
4558 /* Select groups may access flow keys beyond L2 in order to
4559 * select a bucket. Recirculate as appropriate to make this possible.
4561 if (ctx
->was_mpls
) {
4562 ctx_trigger_freeze(ctx
);
4566 switch (group
->selection_method
) {
4567 case SEL_METHOD_DEFAULT
:
4568 return pick_default_select_group(ctx
, group
);
4570 case SEL_METHOD_HASH
:
4571 return pick_hash_fields_select_group(ctx
, group
);
4573 case SEL_METHOD_DP_HASH
:
4574 return pick_dp_hash_select_group(ctx
, group
);
4577 /* Parsing of groups ensures this never happens */
4585 xlate_group_action__(struct xlate_ctx
*ctx
, struct group_dpif
*group
,
4586 bool is_last_action
)
4588 if (group
->up
.type
== OFPGT11_ALL
|| group
->up
.type
== OFPGT11_INDIRECT
) {
4589 struct ovs_list
*last_bucket
= group
->up
.buckets
.prev
;
4590 struct ofputil_bucket
*bucket
;
4591 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
4592 bool is_last_bucket
= &bucket
->list_node
== last_bucket
;
4593 xlate_group_bucket(ctx
, bucket
, is_last_action
&& is_last_bucket
);
4595 xlate_group_stats(ctx
, group
, NULL
);
4597 struct ofputil_bucket
*bucket
;
4598 if (group
->up
.type
== OFPGT11_SELECT
) {
4599 bucket
= pick_select_group(ctx
, group
);
4600 } else if (group
->up
.type
== OFPGT11_FF
) {
4601 bucket
= pick_ff_group(ctx
, group
);
4607 xlate_report(ctx
, OFT_DETAIL
, "using bucket %"PRIu32
,
4609 xlate_group_bucket(ctx
, bucket
, is_last_action
);
4610 xlate_group_stats(ctx
, group
, bucket
);
4612 xlate_report(ctx
, OFT_DETAIL
, "no live bucket");
4613 if (ctx
->xin
->xcache
) {
4614 ofproto_group_unref(&group
->up
);
4621 xlate_group_action(struct xlate_ctx
*ctx
, uint32_t group_id
,
4622 bool is_last_action
)
4624 if (xlate_resubmit_resource_check(ctx
)) {
4625 struct group_dpif
*group
;
4627 /* Take ref only if xcache exists. */
4628 group
= group_dpif_lookup(ctx
->xbridge
->ofproto
, group_id
,
4629 ctx
->xin
->tables_version
, ctx
->xin
->xcache
);
4631 /* XXX: Should set ctx->error ? */
4632 xlate_report(ctx
, OFT_WARN
, "output to nonexistent group %"PRIu32
,
4636 xlate_group_action__(ctx
, group
, is_last_action
);
4643 xlate_ofpact_resubmit(struct xlate_ctx
*ctx
,
4644 const struct ofpact_resubmit
*resubmit
,
4645 bool is_last_action
)
4649 bool may_packet_in
= false;
4650 bool honor_table_miss
= false;
4652 if (ctx
->rule
&& rule_dpif_is_internal(ctx
->rule
)) {
4653 /* Still allow missed packets to be sent to the controller
4654 * if resubmitting from an internal table. */
4655 may_packet_in
= true;
4656 honor_table_miss
= true;
4659 in_port
= resubmit
->in_port
;
4660 if (in_port
== OFPP_IN_PORT
) {
4661 in_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
4664 table_id
= resubmit
->table_id
;
4665 if (table_id
== 255) {
4666 table_id
= ctx
->table_id
;
4669 xlate_table_action(ctx
, in_port
, table_id
, may_packet_in
,
4670 honor_table_miss
, resubmit
->with_ct_orig
,
4671 is_last_action
, do_xlate_actions
);
4675 flood_packet_to_port(struct xlate_ctx
*ctx
, const struct xport
*xport
,
4676 bool all
, bool is_last_action
)
4683 compose_output_action__(ctx
, xport
->ofp_port
, NULL
, false,
4684 is_last_action
, false);
4686 compose_output_action(ctx
, xport
->ofp_port
, NULL
, is_last_action
,
4692 flood_packets(struct xlate_ctx
*ctx
, bool all
, bool is_last_action
)
4694 const struct xport
*xport
, *last
= NULL
;
4696 /* Use 'last' the keep track of the last output port. */
4697 HMAP_FOR_EACH (xport
, ofp_node
, &ctx
->xbridge
->xports
) {
4698 if (xport
->ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
4702 if (all
|| !(xport
->config
& OFPUTIL_PC_NO_FLOOD
)) {
4703 /* 'last' is not the last port, send a packet out, and
4705 flood_packet_to_port(ctx
, last
, all
, false);
4710 /* Send the packet to the 'last' port. */
4711 flood_packet_to_port(ctx
, last
, all
, is_last_action
);
4712 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
4716 put_controller_user_action(struct xlate_ctx
*ctx
,
4717 bool dont_send
, bool continuation
,
4718 uint32_t recirc_id
, int len
,
4719 enum ofp_packet_in_reason reason
,
4720 uint16_t controller_id
)
4722 struct user_action_cookie cookie
;
4724 memset(&cookie
, 0, sizeof cookie
);
4725 cookie
.type
= USER_ACTION_COOKIE_CONTROLLER
;
4726 cookie
.ofp_in_port
= OFPP_NONE
,
4727 cookie
.ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
;
4728 cookie
.controller
.dont_send
= dont_send
;
4729 cookie
.controller
.continuation
= continuation
;
4730 cookie
.controller
.reason
= reason
;
4731 cookie
.controller
.recirc_id
= recirc_id
;
4732 put_32aligned_be64(&cookie
.controller
.rule_cookie
, ctx
->rule_cookie
);
4733 cookie
.controller
.controller_id
= controller_id
;
4734 cookie
.controller
.max_len
= len
;
4736 odp_port_t odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
4737 ctx
->xin
->flow
.in_port
.ofp_port
);
4738 uint32_t pid
= dpif_port_get_pid(ctx
->xbridge
->dpif
, odp_port
);
4739 odp_put_userspace_action(pid
, &cookie
, sizeof cookie
, ODPP_NONE
,
4740 false, ctx
->odp_actions
);
4744 xlate_controller_action(struct xlate_ctx
*ctx
, int len
,
4745 enum ofp_packet_in_reason reason
,
4746 uint16_t controller_id
,
4747 uint32_t provider_meter_id
,
4748 const uint8_t *userdata
, size_t userdata_len
)
4750 xlate_commit_actions(ctx
);
4752 /* A packet sent by an action in a table-miss rule is considered an
4753 * explicit table miss. OpenFlow before 1.3 doesn't have that concept so
4754 * it will get translated back to OFPR_ACTION for those versions. */
4755 if (reason
== OFPR_ACTION
4756 && ctx
->rule
&& rule_is_table_miss(&ctx
->rule
->up
)) {
4757 reason
= OFPR_EXPLICIT_MISS
;
4760 struct frozen_state state
= {
4761 .table_id
= ctx
->table_id
,
4762 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4763 .stack
= ctx
->stack
.data
,
4764 .stack_size
= ctx
->stack
.size
,
4765 .mirrors
= ctx
->mirrors
,
4766 .conntracked
= ctx
->conntracked
,
4770 .action_set_len
= 0,
4771 .userdata
= CONST_CAST(uint8_t *, userdata
),
4772 .userdata_len
= userdata_len
,
4774 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4776 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4778 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4779 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4782 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4784 /* If the controller action didn't request a meter (indicated by a
4785 * 'meter_id' argument other than NX_CTLR_NO_METER), see if one was
4786 * configured through the "controller" virtual meter.
4788 * Internally, ovs-vswitchd uses UINT32_MAX to indicate no meter is
4791 if (provider_meter_id
== UINT32_MAX
) {
4792 meter_id
= ctx
->xbridge
->ofproto
->up
.controller_meter_id
;
4794 meter_id
= provider_meter_id
;
4799 if (meter_id
!= UINT32_MAX
) {
4800 /* If controller meter is configured, generate clone(meter, userspace)
4802 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
4803 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
4805 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
4806 OVS_SAMPLE_ATTR_ACTIONS
);
4807 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
, meter_id
);
4810 /* Generate the datapath flows even if we don't send the packet-in
4811 * so that debugging more closely represents normal state. */
4812 bool dont_send
= false;
4813 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4816 put_controller_user_action(ctx
, dont_send
, false, recirc_id
, len
,
4817 reason
, controller_id
);
4819 if (meter_id
!= UINT32_MAX
) {
4820 nl_msg_end_nested(ctx
->odp_actions
, ac_offset
);
4821 nl_msg_end_nested(ctx
->odp_actions
, offset
);
4825 /* Creates a frozen state, and allocates a unique recirc id for the given
4826 * state. Returns a non-zero recirc id if it is allocated successfully.
4827 * Returns 0 otherwise.
4830 finish_freezing__(struct xlate_ctx
*ctx
, uint8_t table
)
4832 ovs_assert(ctx
->freezing
);
4834 struct frozen_state state
= {
4836 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
4837 .stack
= ctx
->stack
.data
,
4838 .stack_size
= ctx
->stack
.size
,
4839 .mirrors
= ctx
->mirrors
,
4840 .conntracked
= ctx
->conntracked
,
4841 .xport_uuid
= ctx
->xin
->xport_uuid
,
4842 .ofpacts
= ctx
->frozen_actions
.data
,
4843 .ofpacts_len
= ctx
->frozen_actions
.size
,
4844 .action_set
= ctx
->action_set
.data
,
4845 .action_set_len
= ctx
->action_set
.size
,
4846 .userdata
= ctx
->pause
? CONST_CAST(uint8_t *,ctx
->pause
->userdata
)
4848 .userdata_len
= ctx
->pause
? ctx
->pause
->userdata_len
: 0,
4850 frozen_metadata_from_flow(&state
.metadata
, &ctx
->xin
->flow
);
4852 /* Allocate a unique recirc id for the given metadata state in the
4853 * flow. An existing id, with a new reference to the corresponding
4854 * recirculation context, will be returned if possible.
4855 * The life-cycle of this recirc id is managed by associating it
4856 * with the udpif key ('ukey') created for each new datapath flow. */
4857 uint32_t recirc_id
= recirc_alloc_id_ctx(&state
);
4859 xlate_report_error(ctx
, "Failed to allocate recirculation id");
4860 ctx
->error
= XLATE_NO_RECIRCULATION_CONTEXT
;
4863 recirc_refs_add(&ctx
->xout
->recircs
, recirc_id
);
4866 if (!ctx
->xin
->allow_side_effects
&& !ctx
->xin
->xcache
) {
4870 put_controller_user_action(ctx
, false, true, recirc_id
,
4871 ctx
->pause
->max_len
,
4873 ctx
->pause
->controller_id
);
4875 if (ctx
->recirc_update_dp_hash
) {
4876 struct ovs_action_hash
*act_hash
;
4879 act_hash
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
4880 OVS_ACTION_ATTR_HASH
,
4882 act_hash
->hash_alg
= ctx
->dp_hash_alg
;
4883 act_hash
->hash_basis
= ctx
->dp_hash_basis
;
4885 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_RECIRC
, recirc_id
);
4888 /* Undo changes done by freezing. */
4889 ctx_cancel_freeze(ctx
);
4893 /* Called only when we're freezing. */
4895 finish_freezing(struct xlate_ctx
*ctx
)
4897 xlate_commit_actions(ctx
);
4898 finish_freezing__(ctx
, 0);
4901 /* Fork the pipeline here. The current packet will continue processing the
4902 * current action list. A clone of the current packet will recirculate, skip
4903 * the remainder of the current action list and asynchronously resume pipeline
4904 * processing in 'table' with the current metadata and action set. */
4906 compose_recirculate_and_fork(struct xlate_ctx
*ctx
, uint8_t table
,
4907 const uint16_t zone
)
4910 ctx
->freezing
= true;
4911 recirc_id
= finish_freezing__(ctx
, table
);
4913 if (OVS_UNLIKELY(ctx
->xin
->trace
) && recirc_id
) {
4914 if (oftrace_add_recirc_node(ctx
->xin
->recirc_queue
,
4915 OFT_RECIRC_CONNTRACK
, &ctx
->xin
->flow
,
4916 ctx
->xin
->packet
, recirc_id
, zone
)) {
4917 xlate_report(ctx
, OFT_DETAIL
, "A clone of the packet is forked to "
4918 "recirculate. The forked pipeline will be resumed at "
4919 "table %u.", table
);
4921 xlate_report(ctx
, OFT_DETAIL
, "Failed to trace the conntrack "
4922 "forked pipeline with recirc_id = %d.", recirc_id
);
4928 compose_mpls_push_action(struct xlate_ctx
*ctx
, struct ofpact_push_mpls
*mpls
)
4930 struct flow
*flow
= &ctx
->xin
->flow
;
4933 ovs_assert(eth_type_mpls(mpls
->ethertype
));
4935 n
= flow_count_mpls_labels(flow
, ctx
->wc
);
4937 xlate_commit_actions(ctx
);
4938 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
4939 if (ctx
->xin
->packet
!= NULL
) {
4940 xlate_report_error(ctx
, "dropping packet on which an MPLS push "
4941 "action can't be performed as it would have "
4942 "more MPLS LSEs than the %d supported.",
4943 FLOW_MAX_MPLS_LABELS
);
4945 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
4949 /* Update flow's MPLS stack, and clear L3/4 fields to mark them invalid. */
4950 flow_push_mpls(flow
, n
, mpls
->ethertype
, ctx
->wc
, true);
4954 compose_mpls_pop_action(struct xlate_ctx
*ctx
, ovs_be16 eth_type
)
4956 struct flow
*flow
= &ctx
->xin
->flow
;
4957 int n
= flow_count_mpls_labels(flow
, ctx
->wc
);
4959 if (flow_pop_mpls(flow
, n
, eth_type
, ctx
->wc
)) {
4960 if (!eth_type_mpls(eth_type
) && ctx
->xbridge
->support
.odp
.recirc
) {
4961 ctx
->was_mpls
= true;
4963 } else if (n
>= FLOW_MAX_MPLS_LABELS
) {
4964 if (ctx
->xin
->packet
!= NULL
) {
4965 xlate_report_error(ctx
, "dropping packet on which an "
4966 "MPLS pop action can't be performed as it has "
4967 "more MPLS LSEs than the %d supported.",
4968 FLOW_MAX_MPLS_LABELS
);
4970 ctx
->error
= XLATE_TOO_MANY_MPLS_LABELS
;
4971 ofpbuf_clear(ctx
->odp_actions
);
4976 compose_dec_ttl(struct xlate_ctx
*ctx
, struct ofpact_cnt_ids
*ids
)
4978 struct flow
*flow
= &ctx
->xin
->flow
;
4980 if (!is_ip_any(flow
)) {
4984 ctx
->wc
->masks
.nw_ttl
= 0xff;
4985 if (flow
->nw_ttl
> 1) {
4991 for (i
= 0; i
< ids
->n_controllers
; i
++) {
4992 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
4993 ids
->cnt_ids
[i
], UINT32_MAX
, NULL
, 0);
4996 /* Stop processing for current table. */
4997 xlate_report(ctx
, OFT_WARN
, "IPv%d decrement TTL exception",
4998 flow
->dl_type
== htons(ETH_TYPE_IP
) ? 4 : 6);
5004 compose_set_mpls_label_action(struct xlate_ctx
*ctx
, ovs_be32 label
)
5006 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5007 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_LABEL_MASK
);
5008 set_mpls_lse_label(&ctx
->xin
->flow
.mpls_lse
[0], label
);
5013 compose_set_mpls_tc_action(struct xlate_ctx
*ctx
, uint8_t tc
)
5015 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5016 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TC_MASK
);
5017 set_mpls_lse_tc(&ctx
->xin
->flow
.mpls_lse
[0], tc
);
5022 compose_dec_nsh_ttl_action(struct xlate_ctx
*ctx
)
5024 struct flow
*flow
= &ctx
->xin
->flow
;
5026 if ((flow
->packet_type
== htonl(PT_NSH
)) ||
5027 (flow
->dl_type
== htons(ETH_TYPE_NSH
))) {
5028 ctx
->wc
->masks
.nsh
.ttl
= 0xff;
5029 if (flow
->nsh
.ttl
> 1) {
5033 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
,
5034 0, UINT32_MAX
, NULL
, 0);
5038 /* Stop processing for current table. */
5039 xlate_report(ctx
, OFT_WARN
, "NSH decrement TTL exception");
5044 compose_set_mpls_ttl_action(struct xlate_ctx
*ctx
, uint8_t ttl
)
5046 if (eth_type_mpls(ctx
->xin
->flow
.dl_type
)) {
5047 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5048 set_mpls_lse_ttl(&ctx
->xin
->flow
.mpls_lse
[0], ttl
);
5053 compose_dec_mpls_ttl_action(struct xlate_ctx
*ctx
)
5055 struct flow
*flow
= &ctx
->xin
->flow
;
5057 if (eth_type_mpls(flow
->dl_type
)) {
5058 uint8_t ttl
= mpls_lse_to_ttl(flow
->mpls_lse
[0]);
5060 ctx
->wc
->masks
.mpls_lse
[0] |= htonl(MPLS_TTL_MASK
);
5063 set_mpls_lse_ttl(&flow
->mpls_lse
[0], ttl
);
5066 xlate_controller_action(ctx
, UINT16_MAX
, OFPR_INVALID_TTL
, 0,
5067 UINT32_MAX
, NULL
, 0);
5071 /* Stop processing for current table. */
5072 xlate_report(ctx
, OFT_WARN
, "MPLS decrement TTL exception");
5076 /* Emits an action that outputs to 'port', within 'ctx'.
5078 * 'controller_len' affects only packets sent to an OpenFlow controller. It
5079 * is the maximum number of bytes of the packet to send. UINT16_MAX means to
5080 * send the whole packet (and 0 means to omit the packet entirely).
5082 * 'may_packet_in' determines whether the packet may be sent to an OpenFlow
5083 * controller. If it is false, then the packet is never sent to the OpenFlow
5086 * 'is_last_action' should be true if this output is the last OpenFlow action
5087 * to be processed, which enables certain optimizations.
5089 * 'truncate' should be true if the packet to be output is being truncated,
5090 * which suppresses certain optimizations. */
5092 xlate_output_action(struct xlate_ctx
*ctx
, ofp_port_t port
,
5093 uint16_t controller_len
, bool may_packet_in
,
5094 bool is_last_action
, bool truncate
,
5095 bool group_bucket_action
)
5097 ofp_port_t prev_nf_output_iface
= ctx
->nf_output_iface
;
5099 ctx
->nf_output_iface
= NF_OUT_DROP
;
5103 compose_output_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
, NULL
,
5104 is_last_action
, truncate
);
5107 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
5108 0, may_packet_in
, true, false, false,
5115 flood_packets(ctx
, false, is_last_action
);
5118 flood_packets(ctx
, true, is_last_action
);
5120 case OFPP_CONTROLLER
:
5121 xlate_controller_action(ctx
, controller_len
,
5122 (ctx
->in_packet_out
? OFPR_PACKET_OUT
5123 : group_bucket_action
? OFPR_GROUP
5124 : ctx
->in_action_set
? OFPR_ACTION_SET
5126 0, UINT32_MAX
, NULL
, 0);
5132 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5133 compose_output_action(ctx
, port
, NULL
, is_last_action
, truncate
);
5135 xlate_report_info(ctx
, "skipping output to input port");
5140 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
5141 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
5142 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5143 ctx
->nf_output_iface
= prev_nf_output_iface
;
5144 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
5145 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5146 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5151 xlate_output_reg_action(struct xlate_ctx
*ctx
,
5152 const struct ofpact_output_reg
*or,
5153 bool is_last_action
,
5154 bool group_bucket_action
)
5156 uint64_t port
= mf_get_subfield(&or->src
, &ctx
->xin
->flow
);
5157 if (port
<= UINT16_MAX
) {
5158 xlate_report(ctx
, OFT_DETAIL
, "output port is %"PRIu64
, port
);
5160 union mf_subvalue value
;
5162 memset(&value
, 0xff, sizeof value
);
5163 mf_write_subfield_flow(&or->src
, &value
, &ctx
->wc
->masks
);
5164 xlate_output_action(ctx
, u16_to_ofp(port
), or->max_len
,
5165 false, is_last_action
, false,
5166 group_bucket_action
);
5168 xlate_report(ctx
, OFT_WARN
, "output port %"PRIu64
" is out of range",
5174 xlate_output_trunc_action(struct xlate_ctx
*ctx
,
5175 ofp_port_t port
, uint32_t max_len
,
5176 bool is_last_action
,
5177 bool group_bucket_action
)
5179 bool support_trunc
= ctx
->xbridge
->support
.trunc
;
5180 struct ovs_action_trunc
*trunc
;
5181 char name
[OFP_MAX_PORT_NAME_LEN
];
5188 case OFPP_CONTROLLER
:
5190 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5191 xlate_report(ctx
, OFT_WARN
,
5192 "output_trunc does not support port: %s", name
);
5197 if (port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5198 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
, port
);
5200 if (xport
== NULL
|| xport
->odp_port
== ODPP_NONE
) {
5201 /* Since truncate happens at its following output action, if
5202 * the output port is a patch port, the behavior is somehow
5203 * unpredictable. For simplicity, disallow this case. */
5204 ofputil_port_to_string(port
, NULL
, name
, sizeof name
);
5205 xlate_report_error(ctx
, "output_trunc does not support "
5206 "patch port %s", name
);
5210 trunc
= nl_msg_put_unspec_uninit(ctx
->odp_actions
,
5211 OVS_ACTION_ATTR_TRUNC
,
5213 trunc
->max_len
= max_len
;
5214 xlate_output_action(ctx
, port
, 0, false, is_last_action
, true,
5215 group_bucket_action
);
5216 if (!support_trunc
) {
5217 ctx
->xout
->slow
|= SLOW_ACTION
;
5220 xlate_report_info(ctx
, "skipping output to input port");
5227 xlate_enqueue_action(struct xlate_ctx
*ctx
,
5228 const struct ofpact_enqueue
*enqueue
,
5229 bool is_last_action
,
5230 bool group_bucket_action
)
5232 ofp_port_t ofp_port
= enqueue
->port
;
5233 uint32_t queue_id
= enqueue
->queue
;
5234 uint32_t flow_priority
, priority
;
5237 /* Translate queue to priority. */
5238 error
= dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &priority
);
5240 /* Fall back to ordinary output action. */
5241 xlate_output_action(ctx
, enqueue
->port
, 0, false,
5242 is_last_action
, false,
5243 group_bucket_action
);
5247 /* Check output port. */
5248 if (ofp_port
== OFPP_IN_PORT
) {
5249 ofp_port
= ctx
->xin
->flow
.in_port
.ofp_port
;
5250 } else if (ofp_port
== ctx
->xin
->flow
.in_port
.ofp_port
) {
5254 /* Add datapath actions. */
5255 flow_priority
= ctx
->xin
->flow
.skb_priority
;
5256 ctx
->xin
->flow
.skb_priority
= priority
;
5257 compose_output_action(ctx
, ofp_port
, NULL
, is_last_action
, false);
5258 ctx
->xin
->flow
.skb_priority
= flow_priority
;
5260 /* Update NetFlow output port. */
5261 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
5262 ctx
->nf_output_iface
= ofp_port
;
5263 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
5264 ctx
->nf_output_iface
= NF_OUT_MULTI
;
5269 xlate_set_queue_action(struct xlate_ctx
*ctx
, uint32_t queue_id
)
5271 uint32_t skb_priority
;
5273 if (!dpif_queue_to_priority(ctx
->xbridge
->dpif
, queue_id
, &skb_priority
)) {
5274 ctx
->xin
->flow
.skb_priority
= skb_priority
;
5276 /* Couldn't translate queue to a priority. Nothing to do. A warning
5277 * has already been logged. */
5282 slave_enabled_cb(ofp_port_t ofp_port
, void *xbridge_
)
5284 const struct xbridge
*xbridge
= xbridge_
;
5295 case OFPP_CONTROLLER
: /* Not supported by the bundle action. */
5298 port
= get_ofp_port(xbridge
, ofp_port
);
5299 return port
? port
->may_enable
: false;
5304 xlate_bundle_action(struct xlate_ctx
*ctx
,
5305 const struct ofpact_bundle
*bundle
,
5306 bool is_last_action
,
5307 bool group_bucket_action
)
5311 port
= bundle_execute(bundle
, &ctx
->xin
->flow
, ctx
->wc
, slave_enabled_cb
,
5312 CONST_CAST(struct xbridge
*, ctx
->xbridge
));
5313 if (bundle
->dst
.field
) {
5314 nxm_reg_load(&bundle
->dst
, ofp_to_u16(port
), &ctx
->xin
->flow
, ctx
->wc
);
5315 xlate_report_subfield(ctx
, &bundle
->dst
);
5317 xlate_output_action(ctx
, port
, 0, false, is_last_action
, false,
5318 group_bucket_action
);
5323 xlate_learn_action(struct xlate_ctx
*ctx
, const struct ofpact_learn
*learn
)
5325 learn_mask(learn
, ctx
->wc
);
5327 if (ctx
->xin
->xcache
|| ctx
->xin
->allow_side_effects
) {
5328 uint64_t ofpacts_stub
[1024 / 8];
5329 struct ofputil_flow_mod fm
;
5330 struct ofproto_flow_mod ofm__
, *ofm
;
5331 struct ofpbuf ofpacts
;
5334 if (ctx
->xin
->xcache
) {
5335 ofm
= xmalloc(sizeof *ofm
);
5340 ofpbuf_use_stub(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
5341 learn_execute(learn
, &ctx
->xin
->flow
, &fm
, &ofpacts
);
5342 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
5343 struct ds s
= DS_EMPTY_INITIALIZER
;
5344 ds_put_format(&s
, "table=%"PRIu8
" ", fm
.table_id
);
5345 minimatch_format(&fm
.match
,
5346 ofproto_get_tun_tab(&ctx
->xin
->ofproto
->up
),
5347 NULL
, &s
, OFP_DEFAULT_PRIORITY
);
5349 ds_put_format(&s
, " priority=%d", fm
.priority
);
5350 if (fm
.new_cookie
) {
5351 ds_put_format(&s
, " cookie=%#"PRIx64
, ntohll(fm
.new_cookie
));
5353 if (fm
.idle_timeout
!= OFP_FLOW_PERMANENT
) {
5354 ds_put_format(&s
, " idle=%"PRIu16
, fm
.idle_timeout
);
5356 if (fm
.hard_timeout
!= OFP_FLOW_PERMANENT
) {
5357 ds_put_format(&s
, " hard=%"PRIu16
, fm
.hard_timeout
);
5359 if (fm
.flags
& NX_LEARN_F_SEND_FLOW_REM
) {
5360 ds_put_cstr(&s
, " send_flow_rem");
5362 ds_put_cstr(&s
, " actions=");
5363 struct ofpact_format_params fp
= { .s
= &s
};
5364 ofpacts_format(fm
.ofpacts
, fm
.ofpacts_len
, &fp
);
5365 xlate_report(ctx
, OFT_DETAIL
, "%s", ds_cstr(&s
));
5368 error
= ofproto_dpif_flow_mod_init_for_learn(ctx
->xbridge
->ofproto
,
5370 ofpbuf_uninit(&ofpacts
);
5373 bool success
= true;
5374 if (ctx
->xin
->allow_side_effects
) {
5375 error
= ofproto_flow_mod_learn(ofm
, ctx
->xin
->xcache
!= NULL
,
5376 learn
->limit
, &success
);
5377 } else if (learn
->limit
) {
5379 || ofm
->temp_rule
->state
!= RULE_INSERTED
) {
5380 /* The learned rule expired and there are no packets, so
5381 * we cannot learn again. Since the translated actions
5382 * depend on the result of learning, we tell the caller
5383 * that there's no point in caching this result. */
5384 ctx
->xout
->avoid_caching
= true;
5388 if (learn
->flags
& NX_LEARN_F_WRITE_RESULT
) {
5389 nxm_reg_load(&learn
->result_dst
, success
? 1 : 0,
5390 &ctx
->xin
->flow
, ctx
->wc
);
5391 xlate_report_subfield(ctx
, &learn
->result_dst
);
5394 if (success
&& ctx
->xin
->xcache
) {
5395 struct xc_entry
*entry
;
5397 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_LEARN
);
5398 entry
->learn
.ofm
= ofm
;
5399 entry
->learn
.limit
= learn
->limit
;
5402 ofproto_flow_mod_uninit(ofm
);
5405 if (OVS_UNLIKELY(ctx
->xin
->trace
&& !success
)) {
5406 xlate_report(ctx
, OFT_DETAIL
, "Limit exceeded, learn failed");
5410 if (ofm
!= &ofm__
) {
5415 xlate_report_error(ctx
, "LEARN action execution failed (%s).",
5416 ofperr_to_string(error
));
5419 minimatch_destroy(&fm
.match
);
5421 xlate_report(ctx
, OFT_WARN
,
5422 "suppressing side effects, so learn action ignored");
5427 xlate_fin_timeout__(struct rule_dpif
*rule
, uint16_t tcp_flags
,
5428 uint16_t idle_timeout
, uint16_t hard_timeout
)
5430 if (tcp_flags
& (TCP_FIN
| TCP_RST
)) {
5431 ofproto_rule_reduce_timeouts(&rule
->up
, idle_timeout
, hard_timeout
);
5436 xlate_fin_timeout(struct xlate_ctx
*ctx
,
5437 const struct ofpact_fin_timeout
*oft
)
5440 if (ctx
->xin
->allow_side_effects
) {
5441 xlate_fin_timeout__(ctx
->rule
, ctx
->xin
->tcp_flags
,
5442 oft
->fin_idle_timeout
, oft
->fin_hard_timeout
);
5444 if (ctx
->xin
->xcache
) {
5445 struct xc_entry
*entry
;
5447 entry
= xlate_cache_add_entry(ctx
->xin
->xcache
, XC_FIN_TIMEOUT
);
5448 /* XC_RULE already holds a reference on the rule, none is taken
5450 entry
->fin
.rule
= ctx
->rule
;
5451 entry
->fin
.idle
= oft
->fin_idle_timeout
;
5452 entry
->fin
.hard
= oft
->fin_hard_timeout
;
5458 xlate_sample_action(struct xlate_ctx
*ctx
,
5459 const struct ofpact_sample
*os
)
5461 odp_port_t output_odp_port
= ODPP_NONE
;
5462 odp_port_t tunnel_out_port
= ODPP_NONE
;
5463 struct dpif_ipfix
*ipfix
= ctx
->xbridge
->ipfix
;
5464 bool emit_set_tunnel
= false;
5466 if (!ipfix
|| ctx
->xin
->flow
.in_port
.ofp_port
== OFPP_NONE
) {
5470 /* Scale the probability from 16-bit to 32-bit while representing
5471 * the same percentage. */
5472 uint32_t probability
= (os
->probability
<< 16) | os
->probability
;
5474 /* If ofp_port in flow sample action is equel to ofp_port,
5475 * this sample action is a input port action. */
5476 if (os
->sampling_port
!= OFPP_NONE
&&
5477 os
->sampling_port
!= ctx
->xin
->flow
.in_port
.ofp_port
) {
5478 output_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
,
5480 if (output_odp_port
== ODPP_NONE
) {
5481 xlate_report_error(ctx
, "can't use unknown port %d in flow sample "
5482 "action", os
->sampling_port
);
5486 if (dpif_ipfix_get_flow_exporter_tunnel_sampling(ipfix
,
5487 os
->collector_set_id
)
5488 && dpif_ipfix_is_tunnel_port(ipfix
, output_odp_port
)) {
5489 tunnel_out_port
= output_odp_port
;
5490 emit_set_tunnel
= true;
5494 xlate_commit_actions(ctx
);
5495 /* If 'emit_set_tunnel', sample(sampling_port=1) would translate
5496 * into datapath sample action set(tunnel(...)), sample(...) and
5497 * it is used for sampling egress tunnel information. */
5498 if (emit_set_tunnel
) {
5499 const struct xport
*xport
= get_ofp_port(ctx
->xbridge
,
5502 if (xport
&& xport
->is_tunnel
) {
5503 struct flow
*flow
= &ctx
->xin
->flow
;
5504 tnl_port_send(xport
->ofport
, flow
, ctx
->wc
);
5505 if (!ovs_native_tunneling_is_on(ctx
->xbridge
->ofproto
)) {
5506 struct flow_tnl flow_tnl
= flow
->tunnel
;
5507 const char *tnl_type
;
5509 tnl_type
= tnl_port_get_type(xport
->ofport
);
5510 commit_odp_tunnel_action(flow
, &ctx
->base_flow
,
5511 ctx
->odp_actions
, tnl_type
);
5512 flow
->tunnel
= flow_tnl
;
5515 xlate_report_error(ctx
,
5516 "sampling_port:%d should be a tunnel port.",
5521 struct user_action_cookie cookie
= {
5522 .type
= USER_ACTION_COOKIE_FLOW_SAMPLE
,
5523 .ofp_in_port
= ctx
->xin
->flow
.in_port
.ofp_port
,
5524 .ofproto_uuid
= ctx
->xbridge
->ofproto
->uuid
,
5526 .probability
= os
->probability
,
5527 .collector_set_id
= os
->collector_set_id
,
5528 .obs_domain_id
= os
->obs_domain_id
,
5529 .obs_point_id
= os
->obs_point_id
,
5530 .output_odp_port
= output_odp_port
,
5531 .direction
= os
->direction
,
5534 compose_sample_action(ctx
, probability
, &cookie
, tunnel_out_port
, false);
5537 /* Determine if an datapath action translated from the openflow action
5538 * can be reversed by another datapath action.
5540 * Openflow actions that do not emit datapath actions are trivially
5541 * reversible. Reversiblity of other actions depends on nature of
5542 * action and their translation. */
5544 reversible_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5546 const struct ofpact
*a
;
5548 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5551 case OFPACT_CLEAR_ACTIONS
:
5553 case OFPACT_CONJUNCTION
:
5554 case OFPACT_CONTROLLER
:
5555 case OFPACT_CT_CLEAR
:
5556 case OFPACT_DEBUG_RECIRC
:
5557 case OFPACT_DEBUG_SLOW
:
5558 case OFPACT_DEC_MPLS_TTL
:
5559 case OFPACT_DEC_TTL
:
5560 case OFPACT_ENQUEUE
:
5562 case OFPACT_FIN_TIMEOUT
:
5563 case OFPACT_GOTO_TABLE
:
5566 case OFPACT_MULTIPATH
:
5569 case OFPACT_OUTPUT_REG
:
5570 case OFPACT_POP_MPLS
:
5571 case OFPACT_POP_QUEUE
:
5572 case OFPACT_PUSH_MPLS
:
5573 case OFPACT_PUSH_VLAN
:
5574 case OFPACT_REG_MOVE
:
5575 case OFPACT_RESUBMIT
:
5577 case OFPACT_SET_ETH_DST
:
5578 case OFPACT_SET_ETH_SRC
:
5579 case OFPACT_SET_FIELD
:
5580 case OFPACT_SET_IP_DSCP
:
5581 case OFPACT_SET_IP_ECN
:
5582 case OFPACT_SET_IP_TTL
:
5583 case OFPACT_SET_IPV4_DST
:
5584 case OFPACT_SET_IPV4_SRC
:
5585 case OFPACT_SET_L4_DST_PORT
:
5586 case OFPACT_SET_L4_SRC_PORT
:
5587 case OFPACT_SET_MPLS_LABEL
:
5588 case OFPACT_SET_MPLS_TC
:
5589 case OFPACT_SET_MPLS_TTL
:
5590 case OFPACT_SET_QUEUE
:
5591 case OFPACT_SET_TUNNEL
:
5592 case OFPACT_SET_VLAN_PCP
:
5593 case OFPACT_SET_VLAN_VID
:
5594 case OFPACT_STACK_POP
:
5595 case OFPACT_STACK_PUSH
:
5596 case OFPACT_STRIP_VLAN
:
5597 case OFPACT_UNROLL_XLATE
:
5598 case OFPACT_WRITE_ACTIONS
:
5599 case OFPACT_WRITE_METADATA
:
5600 case OFPACT_CHECK_PKT_LARGER
:
5606 case OFPACT_OUTPUT_TRUNC
:
5609 case OFPACT_DEC_NSH_TTL
:
5617 clone_xlate_actions(const struct ofpact
*actions
, size_t actions_len
,
5618 struct xlate_ctx
*ctx
, bool is_last_action
,
5619 bool group_bucket_action OVS_UNUSED
)
5621 struct ofpbuf old_stack
= ctx
->stack
;
5622 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
5623 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
5624 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
5626 struct ofpbuf old_action_set
= ctx
->action_set
;
5627 uint64_t actset_stub
[1024 / 8];
5628 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
5629 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
5631 size_t offset
, ac_offset
;
5632 struct flow old_flow
= ctx
->xin
->flow
;
5634 if (reversible_actions(actions
, actions_len
) || is_last_action
) {
5635 old_flow
= ctx
->xin
->flow
;
5636 do_xlate_actions(actions
, actions_len
, ctx
, is_last_action
, false);
5637 if (!ctx
->freezing
) {
5638 xlate_action_set(ctx
);
5640 if (ctx
->freezing
) {
5641 finish_freezing(ctx
);
5646 /* Commit datapath actions before emitting the clone action to
5647 * avoid emitting those actions twice. Once inside
5648 * the clone, another time for the action after clone. */
5649 xlate_commit_actions(ctx
);
5650 struct flow old_base
= ctx
->base_flow
;
5651 bool old_was_mpls
= ctx
->was_mpls
;
5652 bool old_conntracked
= ctx
->conntracked
;
5654 /* The actions are not reversible, a datapath clone action is
5655 * required to encode the translation. Select the clone action
5656 * based on datapath capabilities. */
5657 if (ctx
->xbridge
->support
.clone
) { /* Use clone action */
5658 /* Use clone action as datapath clone. */
5659 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CLONE
);
5660 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5661 if (!ctx
->freezing
) {
5662 xlate_action_set(ctx
);
5664 if (ctx
->freezing
) {
5665 finish_freezing(ctx
);
5667 nl_msg_end_non_empty_nested(ctx
->odp_actions
, offset
);
5671 if (ctx
->xbridge
->support
.sample_nesting
> 3) {
5672 /* Use sample action as datapath clone. */
5673 offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_SAMPLE
);
5674 ac_offset
= nl_msg_start_nested(ctx
->odp_actions
,
5675 OVS_SAMPLE_ATTR_ACTIONS
);
5676 do_xlate_actions(actions
, actions_len
, ctx
, true, false);
5677 if (!ctx
->freezing
) {
5678 xlate_action_set(ctx
);
5680 if (ctx
->freezing
) {
5681 finish_freezing(ctx
);
5683 if (nl_msg_end_non_empty_nested(ctx
->odp_actions
, ac_offset
)) {
5684 nl_msg_cancel_nested(ctx
->odp_actions
, offset
);
5686 nl_msg_put_u32(ctx
->odp_actions
, OVS_SAMPLE_ATTR_PROBABILITY
,
5687 UINT32_MAX
); /* 100% probability. */
5688 nl_msg_end_nested(ctx
->odp_actions
, offset
);
5693 /* Datapath does not support clone, skip xlate 'oc' and
5694 * report an error */
5695 xlate_report_error(ctx
, "Failed to compose clone action");
5698 /* The clone's conntrack execution should have no effect on the original
5700 ctx
->conntracked
= old_conntracked
;
5702 /* Popping MPLS from the clone should have no effect on the original
5704 ctx
->was_mpls
= old_was_mpls
;
5706 /* Restore the 'base_flow' for the next action. */
5707 ctx
->base_flow
= old_base
;
5710 ofpbuf_uninit(&ctx
->action_set
);
5711 ctx
->action_set
= old_action_set
;
5712 ofpbuf_uninit(&ctx
->stack
);
5713 ctx
->stack
= old_stack
;
5714 ctx
->xin
->flow
= old_flow
;
5718 compose_clone(struct xlate_ctx
*ctx
, const struct ofpact_nest
*oc
,
5719 bool is_last_action
)
5721 size_t oc_actions_len
= ofpact_nest_get_action_len(oc
);
5723 clone_xlate_actions(oc
->actions
, oc_actions_len
, ctx
, is_last_action
,
5728 xlate_meter_action(struct xlate_ctx
*ctx
, const struct ofpact_meter
*meter
)
5730 if (meter
->provider_meter_id
!= UINT32_MAX
) {
5731 nl_msg_put_u32(ctx
->odp_actions
, OVS_ACTION_ATTR_METER
,
5732 meter
->provider_meter_id
);
5737 may_receive(const struct xport
*xport
, struct xlate_ctx
*ctx
)
5739 if (xport
->config
& (is_stp(&ctx
->xin
->flow
)
5740 ? OFPUTIL_PC_NO_RECV_STP
5741 : OFPUTIL_PC_NO_RECV
)) {
5745 /* Only drop packets here if both forwarding and learning are
5746 * disabled. If just learning is enabled, we need to have
5747 * OFPP_NORMAL and the learning action have a look at the packet
5748 * before we can drop it. */
5749 if ((!xport_stp_forward_state(xport
) && !xport_stp_learn_state(xport
)) ||
5750 (!xport_rstp_forward_state(xport
) && !xport_rstp_learn_state(xport
))) {
5758 xlate_write_actions__(struct xlate_ctx
*ctx
,
5759 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
5761 /* Maintain actset_output depending on the contents of the action set:
5763 * - OFPP_UNSET, if there is no "output" action.
5765 * - The output port, if there is an "output" action and no "group"
5768 * - OFPP_UNSET, if there is a "group" action.
5770 if (!ctx
->action_set_has_group
) {
5771 const struct ofpact
*a
;
5772 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
5773 if (a
->type
== OFPACT_OUTPUT
) {
5774 ctx
->xin
->flow
.actset_output
= ofpact_get_OUTPUT(a
)->port
;
5775 } else if (a
->type
== OFPACT_GROUP
) {
5776 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
5777 ctx
->action_set_has_group
= true;
5783 ofpbuf_put(&ctx
->action_set
, ofpacts
, ofpacts_len
);
5787 xlate_write_actions(struct xlate_ctx
*ctx
, const struct ofpact_nest
*a
)
5789 xlate_write_actions__(ctx
, a
->actions
, ofpact_nest_get_action_len(a
));
5793 xlate_action_set(struct xlate_ctx
*ctx
)
5795 uint64_t action_list_stub
[1024 / 8];
5796 struct ofpbuf action_list
= OFPBUF_STUB_INITIALIZER(action_list_stub
);
5797 ofpacts_execute_action_set(&action_list
, &ctx
->action_set
);
5798 /* Clear the action set, as it is not needed any more. */
5799 ofpbuf_clear(&ctx
->action_set
);
5800 if (action_list
.size
) {
5801 ctx
->in_action_set
= true;
5803 struct ovs_list
*old_trace
= ctx
->xin
->trace
;
5804 ctx
->xin
->trace
= xlate_report(ctx
, OFT_TABLE
,
5805 "--. Executing action set:");
5806 do_xlate_actions(action_list
.data
, action_list
.size
, ctx
, true, false);
5807 ctx
->xin
->trace
= old_trace
;
5809 ctx
->in_action_set
= false;
5811 ofpbuf_uninit(&action_list
);
5815 freeze_put_unroll_xlate(struct xlate_ctx
*ctx
)
5817 struct ofpact_unroll_xlate
*unroll
= ctx
->frozen_actions
.header
;
5819 /* Restore the table_id and rule cookie for a potential PACKET
5822 (ctx
->table_id
!= unroll
->rule_table_id
5823 || ctx
->rule_cookie
!= unroll
->rule_cookie
)) {
5824 unroll
= ofpact_put_UNROLL_XLATE(&ctx
->frozen_actions
);
5825 unroll
->rule_table_id
= ctx
->table_id
;
5826 unroll
->rule_cookie
= ctx
->rule_cookie
;
5827 ctx
->frozen_actions
.header
= unroll
;
5832 /* Copy actions 'a' through 'end' to ctx->frozen_actions, which will be
5833 * executed after thawing. Inserts an UNROLL_XLATE action, if none is already
5834 * present, before any action that may depend on the current table ID or flow
5837 freeze_unroll_actions(const struct ofpact
*a
, const struct ofpact
*end
,
5838 struct xlate_ctx
*ctx
)
5840 for (; a
< end
; a
= ofpact_next(a
)) {
5842 case OFPACT_OUTPUT_REG
:
5843 case OFPACT_OUTPUT_TRUNC
:
5846 case OFPACT_CONTROLLER
:
5847 case OFPACT_DEC_MPLS_TTL
:
5848 case OFPACT_DEC_NSH_TTL
:
5849 case OFPACT_DEC_TTL
:
5850 /* These actions may generate asynchronous messages, which include
5851 * table ID and flow cookie information. */
5852 freeze_put_unroll_xlate(ctx
);
5855 case OFPACT_RESUBMIT
:
5856 if (ofpact_get_RESUBMIT(a
)->table_id
== 0xff) {
5857 /* This resubmit action is relative to the current table, so we
5858 * need to track what table that is.*/
5859 freeze_put_unroll_xlate(ctx
);
5863 case OFPACT_SET_TUNNEL
:
5864 case OFPACT_REG_MOVE
:
5865 case OFPACT_SET_FIELD
:
5866 case OFPACT_STACK_PUSH
:
5867 case OFPACT_STACK_POP
:
5869 case OFPACT_WRITE_METADATA
:
5870 case OFPACT_GOTO_TABLE
:
5871 case OFPACT_ENQUEUE
:
5872 case OFPACT_SET_VLAN_VID
:
5873 case OFPACT_SET_VLAN_PCP
:
5874 case OFPACT_STRIP_VLAN
:
5875 case OFPACT_PUSH_VLAN
:
5876 case OFPACT_SET_ETH_SRC
:
5877 case OFPACT_SET_ETH_DST
:
5878 case OFPACT_SET_IPV4_SRC
:
5879 case OFPACT_SET_IPV4_DST
:
5880 case OFPACT_SET_IP_DSCP
:
5881 case OFPACT_SET_IP_ECN
:
5882 case OFPACT_SET_IP_TTL
:
5883 case OFPACT_SET_L4_SRC_PORT
:
5884 case OFPACT_SET_L4_DST_PORT
:
5885 case OFPACT_SET_QUEUE
:
5886 case OFPACT_POP_QUEUE
:
5887 case OFPACT_PUSH_MPLS
:
5888 case OFPACT_POP_MPLS
:
5889 case OFPACT_SET_MPLS_LABEL
:
5890 case OFPACT_SET_MPLS_TC
:
5891 case OFPACT_SET_MPLS_TTL
:
5892 case OFPACT_MULTIPATH
:
5895 case OFPACT_UNROLL_XLATE
:
5896 case OFPACT_FIN_TIMEOUT
:
5897 case OFPACT_CLEAR_ACTIONS
:
5898 case OFPACT_WRITE_ACTIONS
:
5904 case OFPACT_DEBUG_RECIRC
:
5905 case OFPACT_DEBUG_SLOW
:
5907 case OFPACT_CT_CLEAR
:
5909 case OFPACT_CHECK_PKT_LARGER
:
5910 /* These may not generate PACKET INs. */
5914 case OFPACT_CONJUNCTION
:
5915 /* These need not be copied for restoration. */
5918 /* Copy the action over. */
5919 ofpbuf_put(&ctx
->frozen_actions
, a
, OFPACT_ALIGN(a
->len
));
5924 put_ct_mark(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
5925 struct flow_wildcards
*wc
)
5927 if (wc
->masks
.ct_mark
) {
5933 odp_ct_mark
= nl_msg_put_unspec_uninit(odp_actions
, OVS_CT_ATTR_MARK
,
5934 sizeof(*odp_ct_mark
));
5935 odp_ct_mark
->key
= flow
->ct_mark
& wc
->masks
.ct_mark
;
5936 odp_ct_mark
->mask
= wc
->masks
.ct_mark
;
5941 put_ct_label(const struct flow
*flow
, struct ofpbuf
*odp_actions
,
5942 struct flow_wildcards
*wc
)
5944 if (!ovs_u128_is_zero(wc
->masks
.ct_label
)) {
5950 odp_ct_label
.key
= ovs_u128_and(flow
->ct_label
, wc
->masks
.ct_label
);
5951 odp_ct_label
.mask
= wc
->masks
.ct_label
;
5952 nl_msg_put_unspec(odp_actions
, OVS_CT_ATTR_LABELS
,
5953 &odp_ct_label
, sizeof odp_ct_label
);
5958 put_ct_helper(struct xlate_ctx
*ctx
,
5959 struct ofpbuf
*odp_actions
, struct ofpact_conntrack
*ofc
)
5964 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "ftp");
5967 nl_msg_put_string(odp_actions
, OVS_CT_ATTR_HELPER
, "tftp");
5970 xlate_report_error(ctx
, "cannot serialize ct_helper %d", ofc
->alg
);
5977 put_ct_nat(struct xlate_ctx
*ctx
)
5979 struct ofpact_nat
*ofn
= ctx
->ct_nat_action
;
5986 nat_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_CT_ATTR_NAT
);
5987 if (ofn
->flags
& NX_NAT_F_SRC
|| ofn
->flags
& NX_NAT_F_DST
) {
5988 nl_msg_put_flag(ctx
->odp_actions
, ofn
->flags
& NX_NAT_F_SRC
5989 ? OVS_NAT_ATTR_SRC
: OVS_NAT_ATTR_DST
);
5990 if (ofn
->flags
& NX_NAT_F_PERSISTENT
) {
5991 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PERSISTENT
);
5993 if (ofn
->flags
& NX_NAT_F_PROTO_HASH
) {
5994 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_HASH
);
5995 } else if (ofn
->flags
& NX_NAT_F_PROTO_RANDOM
) {
5996 nl_msg_put_flag(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_RANDOM
);
5998 if (ofn
->range_af
== AF_INET
) {
5999 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6000 ofn
->range
.addr
.ipv4
.min
);
6001 if (ofn
->range
.addr
.ipv4
.max
&&
6002 (ntohl(ofn
->range
.addr
.ipv4
.max
)
6003 > ntohl(ofn
->range
.addr
.ipv4
.min
))) {
6004 nl_msg_put_be32(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6005 ofn
->range
.addr
.ipv4
.max
);
6007 } else if (ofn
->range_af
== AF_INET6
) {
6008 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MIN
,
6009 &ofn
->range
.addr
.ipv6
.min
,
6010 sizeof ofn
->range
.addr
.ipv6
.min
);
6011 if (!ipv6_mask_is_any(&ofn
->range
.addr
.ipv6
.max
) &&
6012 memcmp(&ofn
->range
.addr
.ipv6
.max
, &ofn
->range
.addr
.ipv6
.min
,
6013 sizeof ofn
->range
.addr
.ipv6
.max
) > 0) {
6014 nl_msg_put_unspec(ctx
->odp_actions
, OVS_NAT_ATTR_IP_MAX
,
6015 &ofn
->range
.addr
.ipv6
.max
,
6016 sizeof ofn
->range
.addr
.ipv6
.max
);
6019 if (ofn
->range_af
!= AF_UNSPEC
&& ofn
->range
.proto
.min
) {
6020 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MIN
,
6021 ofn
->range
.proto
.min
);
6022 if (ofn
->range
.proto
.max
&&
6023 ofn
->range
.proto
.max
> ofn
->range
.proto
.min
) {
6024 nl_msg_put_u16(ctx
->odp_actions
, OVS_NAT_ATTR_PROTO_MAX
,
6025 ofn
->range
.proto
.max
);
6029 nl_msg_end_nested(ctx
->odp_actions
, nat_offset
);
6033 compose_conntrack_action(struct xlate_ctx
*ctx
, struct ofpact_conntrack
*ofc
,
6034 bool is_last_action
)
6036 ovs_u128 old_ct_label_mask
= ctx
->wc
->masks
.ct_label
;
6037 uint32_t old_ct_mark_mask
= ctx
->wc
->masks
.ct_mark
;
6041 /* Ensure that any prior actions are applied before composing the new
6042 * conntrack action. */
6043 xlate_commit_actions(ctx
);
6045 /* Process nested actions first, to populate the key. */
6046 ctx
->ct_nat_action
= NULL
;
6047 ctx
->wc
->masks
.ct_mark
= 0;
6048 ctx
->wc
->masks
.ct_label
= OVS_U128_ZERO
;
6049 do_xlate_actions(ofc
->actions
, ofpact_ct_get_action_len(ofc
), ctx
,
6050 is_last_action
, false);
6052 if (ofc
->zone_src
.field
) {
6053 zone
= mf_get_subfield(&ofc
->zone_src
, &ctx
->xin
->flow
);
6055 zone
= ofc
->zone_imm
;
6058 ct_offset
= nl_msg_start_nested(ctx
->odp_actions
, OVS_ACTION_ATTR_CT
);
6059 if (ofc
->flags
& NX_CT_F_COMMIT
) {
6060 nl_msg_put_flag(ctx
->odp_actions
, ofc
->flags
& NX_CT_F_FORCE
?
6061 OVS_CT_ATTR_FORCE_COMMIT
: OVS_CT_ATTR_COMMIT
);
6062 if (ctx
->xbridge
->support
.ct_eventmask
) {
6063 nl_msg_put_u32(ctx
->odp_actions
, OVS_CT_ATTR_EVENTMASK
,
6064 OVS_CT_EVENTMASK_DEFAULT
);
6067 nl_msg_put_u16(ctx
->odp_actions
, OVS_CT_ATTR_ZONE
, zone
);
6068 put_ct_mark(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6069 put_ct_label(&ctx
->xin
->flow
, ctx
->odp_actions
, ctx
->wc
);
6070 put_ct_helper(ctx
, ctx
->odp_actions
, ofc
);
6072 ctx
->ct_nat_action
= NULL
;
6073 nl_msg_end_nested(ctx
->odp_actions
, ct_offset
);
6075 ctx
->wc
->masks
.ct_mark
= old_ct_mark_mask
;
6076 ctx
->wc
->masks
.ct_label
= old_ct_label_mask
;
6078 if (ofc
->recirc_table
!= NX_CT_RECIRC_NONE
) {
6079 ctx
->conntracked
= true;
6080 compose_recirculate_and_fork(ctx
, ofc
->recirc_table
, zone
);
6083 /* The ct_* fields are only available in the scope of the 'recirc_table'
6085 flow_clear_conntrack(&ctx
->xin
->flow
);
6086 xlate_report(ctx
, OFT_DETAIL
, "Sets the packet to an untracked state, "
6087 "and clears all the conntrack fields.");
6088 ctx
->conntracked
= false;
6092 compose_ct_clear_action(struct xlate_ctx
*ctx
)
6094 clear_conntrack(ctx
);
6095 /* This action originally existed without dpif support. So to preserve
6096 * compatibility, only append it if the dpif supports it. */
6097 if (ctx
->xbridge
->support
.ct_clear
) {
6098 nl_msg_put_flag(ctx
->odp_actions
, OVS_ACTION_ATTR_CT_CLEAR
);
6102 /* check_pkt_larger action checks the packet length and stores the
6103 * result in the register bit. We translate this action to the
6104 * datapath action - 'check_pkt_len' whose format
6105 * is: 'check_pkt_len(pkt_len, ge(actions), le(actions))'.
6107 * We first set the destination register bit to 1 and call
6108 * 'do_xlate_actions' for the case - packet len greater than
6109 * the specified packet length.
6111 * We then set the destination register bit to 0 and call
6112 * 'do_xlate_actions' for the case - packet length is lesser or
6113 * equal to the specified packet length.
6115 * It is possible for freezing to happen for both the cases.
6118 xlate_check_pkt_larger(struct xlate_ctx
*ctx
,
6119 struct ofpact_check_pkt_larger
*check_pkt_larger
,
6120 const struct ofpact
*remaining_acts
,
6121 size_t remaining_acts_len
)
6123 union mf_subvalue value
;
6124 memset(&value
, 0, sizeof value
);
6125 if (!ctx
->xbridge
->support
.check_pkt_len
) {
6126 uint8_t is_pkt_larger
= 0;
6127 if (ctx
->xin
->packet
) {
6129 dp_packet_size(ctx
->xin
->packet
) > check_pkt_larger
->pkt_len
;
6131 value
.u8_val
= is_pkt_larger
;
6132 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
,
6134 /* If datapath doesn't support check_pkt_len action, then set the
6135 * SLOW_ACTION flag. If we don't set SLOW_ACTION, we
6136 * will push a flow to the datapath based on the packet length
6137 * in ctx->xin->packet. For subsequent patches which match the
6138 * same flow, datapath will apply the actions without considering
6139 * the packet length. This results in wrong actions being applied.
6141 ctx
->xout
->slow
|= SLOW_ACTION
;
6145 struct ofpbuf old_stack
= ctx
->stack
;
6146 union mf_subvalue new_stack
[1024 / sizeof(union mf_subvalue
)];
6147 ofpbuf_use_stub(&ctx
->stack
, new_stack
, sizeof new_stack
);
6148 ofpbuf_put(&ctx
->stack
, old_stack
.data
, old_stack
.size
);
6150 struct ofpbuf old_action_set
= ctx
->action_set
;
6151 uint64_t actset_stub
[1024 / 8];
6152 ofpbuf_use_stub(&ctx
->action_set
, actset_stub
, sizeof actset_stub
);
6153 ofpbuf_put(&ctx
->action_set
, old_action_set
.data
, old_action_set
.size
);
6155 struct flow old_flow
= ctx
->xin
->flow
;
6156 xlate_commit_actions(ctx
);
6157 struct flow old_base
= ctx
->base_flow
;
6158 bool old_was_mpls
= ctx
->was_mpls
;
6159 bool old_conntracked
= ctx
->conntracked
;
6161 size_t offset
= nl_msg_start_nested(ctx
->odp_actions
,
6162 OVS_ACTION_ATTR_CHECK_PKT_LEN
);
6163 nl_msg_put_u16(ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_PKT_LEN
,
6164 check_pkt_larger
->pkt_len
);
6165 size_t offset_attr
= nl_msg_start_nested(
6166 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_GREATER
);
6168 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6169 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6170 if (!ctx
->freezing
) {
6171 xlate_action_set(ctx
);
6173 if (ctx
->freezing
) {
6174 finish_freezing(ctx
);
6176 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6178 ctx
->base_flow
= old_base
;
6179 ctx
->was_mpls
= old_was_mpls
;
6180 ctx
->conntracked
= old_conntracked
;
6181 ctx
->xin
->flow
= old_flow
;
6183 /* If the flow translation for the IF_GREATER case requires freezing,
6184 * then ctx->exit would be true. Reset to false so that we can
6185 * do flow translation for 'IF_LESS_EQUAL' case. finish_freezing()
6186 * would have taken care of Undoing the changes done for freeze. */
6189 offset_attr
= nl_msg_start_nested(
6190 ctx
->odp_actions
, OVS_CHECK_PKT_LEN_ATTR_ACTIONS_IF_LESS_EQUAL
);
6192 mf_write_subfield_flow(&check_pkt_larger
->dst
, &value
, &ctx
->xin
->flow
);
6193 do_xlate_actions(remaining_acts
, remaining_acts_len
, ctx
, true, false);
6194 if (!ctx
->freezing
) {
6195 xlate_action_set(ctx
);
6197 if (ctx
->freezing
) {
6198 finish_freezing(ctx
);
6200 nl_msg_end_nested(ctx
->odp_actions
, offset_attr
);
6201 nl_msg_end_nested(ctx
->odp_actions
, offset
);
6203 ofpbuf_uninit(&ctx
->action_set
);
6204 ctx
->action_set
= old_action_set
;
6205 ofpbuf_uninit(&ctx
->stack
);
6206 ctx
->stack
= old_stack
;
6207 ctx
->base_flow
= old_base
;
6208 ctx
->was_mpls
= old_was_mpls
;
6209 ctx
->conntracked
= old_conntracked
;
6210 ctx
->xin
->flow
= old_flow
;
6215 rewrite_flow_encap_ethernet(struct xlate_ctx
*ctx
,
6217 struct flow_wildcards
*wc
)
6219 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6220 if (pt_ns(flow
->packet_type
) == OFPHTN_ETHERTYPE
) {
6221 /* Only adjust the packet_type and zero the dummy Ethernet addresses. */
6222 ovs_be16 ethertype
= pt_ns_type_be(flow
->packet_type
);
6223 flow
->packet_type
= htonl(PT_ETH
);
6224 flow
->dl_src
= eth_addr_zero
;
6225 flow
->dl_dst
= eth_addr_zero
;
6226 flow
->dl_type
= ethertype
;
6228 /* Error handling: drop packet. */
6229 xlate_report_debug(ctx
, OFT_ACTION
,
6230 "Dropping packet as encap(ethernet) is not "
6231 "supported for packet type ethernet.");
6232 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6236 /* For an MD2 NSH header returns a pointer to an ofpbuf with the encoded
6237 * MD2 TLVs provided as encap properties to the encap operation. This
6238 * will be stored as encap_data in the ctx and copied into the push_nsh
6239 * action at the next commit. */
6240 static struct ofpbuf
*
6241 rewrite_flow_push_nsh(struct xlate_ctx
*ctx
,
6242 const struct ofpact_encap
*encap
,
6244 struct flow_wildcards
*wc
)
6246 ovs_be32 packet_type
= flow
->packet_type
;
6247 const char *ptr
= (char *) encap
->props
;
6248 struct ofpbuf
*buf
= ofpbuf_new(NSH_CTX_HDRS_MAX_LEN
);
6249 uint8_t md_type
= NSH_M_TYPE1
;
6253 /* Scan the optional NSH encap TLV properties, if any. */
6254 for (i
= 0; i
< encap
->n_props
; i
++) {
6255 struct ofpact_ed_prop
*prop_ptr
=
6256 ALIGNED_CAST(struct ofpact_ed_prop
*, ptr
);
6257 if (prop_ptr
->prop_class
== OFPPPC_NSH
) {
6258 switch (prop_ptr
->type
) {
6259 case OFPPPT_PROP_NSH_MDTYPE
: {
6260 struct ofpact_ed_prop_nsh_md_type
*prop_md_type
=
6261 ALIGNED_CAST(struct ofpact_ed_prop_nsh_md_type
*,
6263 md_type
= prop_md_type
->md_type
;
6266 case OFPPPT_PROP_NSH_TLV
: {
6267 struct ofpact_ed_prop_nsh_tlv
*tlv_prop
=
6268 ALIGNED_CAST(struct ofpact_ed_prop_nsh_tlv
*,
6270 struct nsh_md2_tlv
*md2_ctx
=
6271 ofpbuf_put_uninit(buf
, sizeof(*md2_ctx
));
6272 md2_ctx
->md_class
= tlv_prop
->tlv_class
;
6273 md2_ctx
->type
= tlv_prop
->tlv_type
;
6274 md2_ctx
->length
= tlv_prop
->tlv_len
;
6275 size_t len
= ROUND_UP(md2_ctx
->length
, 4);
6276 size_t padding
= len
- md2_ctx
->length
;
6277 ofpbuf_put(buf
, tlv_prop
->data
, md2_ctx
->length
);
6278 ofpbuf_put_zeros(buf
, padding
);
6282 /* No other NSH encap properties defined yet. */
6286 ptr
+= ROUND_UP(prop_ptr
->len
, 8);
6288 if (buf
->size
== 0 || buf
->size
> NSH_CTX_HDRS_MAX_LEN
) {
6293 /* Determine the Next Protocol field for NSH header. */
6294 switch (ntohl(packet_type
)) {
6296 np
= NSH_P_ETHERNET
;
6308 /* Error handling: drop packet. */
6309 xlate_report_debug(ctx
, OFT_ACTION
,
6310 "Dropping packet as encap(nsh) is not "
6311 "supported for packet type (%d,0x%x)",
6312 pt_ns(packet_type
), pt_ns_type(packet_type
));
6313 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6316 /* Note that we have matched on packet_type! */
6317 wc
->masks
.packet_type
= OVS_BE32_MAX
;
6319 /* Reset all current flow packet headers. */
6320 memset(&flow
->dl_dst
, 0,
6321 sizeof(struct flow
) - offsetof(struct flow
, dl_dst
));
6323 /* Populate the flow with the new NSH header. */
6324 flow
->packet_type
= htonl(PT_NSH
);
6325 flow
->dl_type
= htons(ETH_TYPE_NSH
);
6326 flow
->nsh
.flags
= 0;
6329 flow
->nsh
.path_hdr
= htonl(255);
6331 if (md_type
== NSH_M_TYPE1
) {
6332 flow
->nsh
.mdtype
= NSH_M_TYPE1
;
6333 memset(flow
->nsh
.context
, 0, sizeof flow
->nsh
.context
);
6335 /* Drop any MD2 context TLVs. */
6339 } else if (md_type
== NSH_M_TYPE2
) {
6340 flow
->nsh
.mdtype
= NSH_M_TYPE2
;
6342 flow
->nsh
.mdtype
&= NSH_MDTYPE_MASK
;
6348 xlate_generic_encap_action(struct xlate_ctx
*ctx
,
6349 const struct ofpact_encap
*encap
)
6351 struct flow
*flow
= &ctx
->xin
->flow
;
6352 struct flow_wildcards
*wc
= ctx
->wc
;
6353 struct ofpbuf
*encap_data
= NULL
;
6355 /* Ensure that any pending actions on the inner packet are applied before
6356 * rewriting the flow */
6357 xlate_commit_actions(ctx
);
6359 /* Rewrite the flow to reflect the effect of pushing the new encap header. */
6360 switch (ntohl(encap
->new_pkt_type
)) {
6362 rewrite_flow_encap_ethernet(ctx
, flow
, wc
);
6365 encap_data
= rewrite_flow_push_nsh(ctx
, encap
, flow
, wc
);
6368 /* New packet type was checked during decoding. */
6373 /* The actual encap datapath action will be generated at next commit. */
6374 ctx
->pending_encap
= true;
6375 ctx
->encap_data
= encap_data
;
6379 /* Returns true if packet must be recirculated after decapsulation. */
6381 xlate_generic_decap_action(struct xlate_ctx
*ctx
,
6382 const struct ofpact_decap
*decap OVS_UNUSED
)
6384 struct flow
*flow
= &ctx
->xin
->flow
;
6386 /* Ensure that any pending actions on the current packet are applied
6387 * before generating the decap action. */
6388 xlate_commit_actions(ctx
);
6390 /* We assume for now that the new_pkt_type is PT_USE_NEXT_PROTO. */
6391 switch (ntohl(flow
->packet_type
)) {
6393 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
)) {
6394 /* Error handling: drop packet. */
6395 xlate_report_debug(ctx
, OFT_ACTION
, "Dropping packet, cannot "
6396 "decap Ethernet if VLAN is present.");
6397 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6399 /* Just change the packet_type.
6400 * Delay generating pop_eth to the next commit. */
6401 flow
->packet_type
= htonl(PACKET_TYPE(OFPHTN_ETHERTYPE
,
6402 ntohs(flow
->dl_type
)));
6403 ctx
->wc
->masks
.dl_type
= OVS_BE16_MAX
;
6407 /* The pop_nsh action is generated at the commit executed as
6408 * part of freezing the ctx for recirculation. Here we just set
6409 * the new packet type based on the NSH next protocol field. */
6410 switch (flow
->nsh
.np
) {
6411 case NSH_P_ETHERNET
:
6412 flow
->packet_type
= htonl(PT_ETH
);
6415 flow
->packet_type
= htonl(PT_IPV4
);
6418 flow
->packet_type
= htonl(PT_IPV6
);
6421 flow
->packet_type
= htonl(PT_NSH
);
6424 /* Error handling: drop packet. */
6425 xlate_report_debug(ctx
, OFT_ACTION
,
6426 "Dropping packet as NSH next protocol %d "
6427 "is not supported", flow
->nsh
.np
);
6428 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6432 ctx
->wc
->masks
.nsh
.np
= UINT8_MAX
;
6433 ctx
->pending_decap
= true;
6434 /* Trigger recirculation. */
6437 /* Error handling: drop packet. */
6440 "Dropping packet as the decap() does not support "
6441 "packet type (%d,0x%x)",
6442 pt_ns(flow
->packet_type
), pt_ns_type(flow
->packet_type
));
6443 ctx
->error
= XLATE_UNSUPPORTED_PACKET_TYPE
;
6449 recirc_for_mpls(const struct ofpact
*a
, struct xlate_ctx
*ctx
)
6451 /* No need to recirculate if already exiting. */
6456 /* Do not consider recirculating unless the packet was previously MPLS. */
6457 if (!ctx
->was_mpls
) {
6461 /* Special case these actions, only recirculating if necessary.
6462 * This avoids the overhead of recirculation in common use-cases.
6466 /* Output actions do not require recirculation. */
6468 case OFPACT_OUTPUT_TRUNC
:
6469 case OFPACT_ENQUEUE
:
6470 case OFPACT_OUTPUT_REG
:
6471 /* Set actions that don't touch L3+ fields do not require recirculation. */
6472 case OFPACT_SET_VLAN_VID
:
6473 case OFPACT_SET_VLAN_PCP
:
6474 case OFPACT_SET_ETH_SRC
:
6475 case OFPACT_SET_ETH_DST
:
6476 case OFPACT_SET_TUNNEL
:
6477 case OFPACT_SET_QUEUE
:
6478 /* If actions of a group require recirculation that can be detected
6479 * when translating them. */
6483 /* Set field that don't touch L3+ fields don't require recirculation. */
6484 case OFPACT_SET_FIELD
:
6485 if (mf_is_l3_or_higher(ofpact_get_SET_FIELD(a
)->field
)) {
6490 /* For simplicity, recirculate in all other cases. */
6491 case OFPACT_CONTROLLER
:
6493 case OFPACT_STRIP_VLAN
:
6494 case OFPACT_PUSH_VLAN
:
6495 case OFPACT_SET_IPV4_SRC
:
6496 case OFPACT_SET_IPV4_DST
:
6497 case OFPACT_SET_IP_DSCP
:
6498 case OFPACT_SET_IP_ECN
:
6499 case OFPACT_SET_IP_TTL
:
6500 case OFPACT_SET_L4_SRC_PORT
:
6501 case OFPACT_SET_L4_DST_PORT
:
6502 case OFPACT_REG_MOVE
:
6503 case OFPACT_STACK_PUSH
:
6504 case OFPACT_STACK_POP
:
6505 case OFPACT_DEC_TTL
:
6506 case OFPACT_SET_MPLS_LABEL
:
6507 case OFPACT_SET_MPLS_TC
:
6508 case OFPACT_SET_MPLS_TTL
:
6509 case OFPACT_DEC_MPLS_TTL
:
6510 case OFPACT_PUSH_MPLS
:
6511 case OFPACT_POP_MPLS
:
6512 case OFPACT_POP_QUEUE
:
6513 case OFPACT_FIN_TIMEOUT
:
6514 case OFPACT_RESUBMIT
:
6516 case OFPACT_CONJUNCTION
:
6517 case OFPACT_MULTIPATH
:
6524 case OFPACT_DEC_NSH_TTL
:
6525 case OFPACT_UNROLL_XLATE
:
6527 case OFPACT_CT_CLEAR
:
6529 case OFPACT_DEBUG_RECIRC
:
6530 case OFPACT_DEBUG_SLOW
:
6532 case OFPACT_CLEAR_ACTIONS
:
6533 case OFPACT_WRITE_ACTIONS
:
6534 case OFPACT_WRITE_METADATA
:
6535 case OFPACT_GOTO_TABLE
:
6536 case OFPACT_CHECK_PKT_LARGER
:
6542 ctx_trigger_freeze(ctx
);
6546 xlate_ofpact_reg_move(struct xlate_ctx
*ctx
, const struct ofpact_reg_move
*a
)
6548 mf_subfield_copy(&a
->src
, &a
->dst
, &ctx
->xin
->flow
, ctx
->wc
);
6549 xlate_report_subfield(ctx
, &a
->dst
);
6553 xlate_ofpact_stack_pop(struct xlate_ctx
*ctx
, const struct ofpact_stack
*a
)
6555 if (nxm_execute_stack_pop(a
, &ctx
->xin
->flow
, ctx
->wc
, &ctx
->stack
)) {
6556 xlate_report_subfield(ctx
, &a
->subfield
);
6558 xlate_report_error(ctx
, "stack underflow");
6562 /* Restore translation context data that was stored earlier. */
6564 xlate_ofpact_unroll_xlate(struct xlate_ctx
*ctx
,
6565 const struct ofpact_unroll_xlate
*a
)
6567 ctx
->table_id
= a
->rule_table_id
;
6568 ctx
->rule_cookie
= a
->rule_cookie
;
6569 xlate_report(ctx
, OFT_THAW
, "restored state: table=%"PRIu8
", "
6570 "cookie=%#"PRIx64
, a
->rule_table_id
, a
->rule_cookie
);
6574 do_xlate_actions(const struct ofpact
*ofpacts
, size_t ofpacts_len
,
6575 struct xlate_ctx
*ctx
, bool is_last_action
,
6576 bool group_bucket_action
)
6578 struct flow_wildcards
*wc
= ctx
->wc
;
6579 struct flow
*flow
= &ctx
->xin
->flow
;
6580 const struct ofpact
*a
;
6582 /* dl_type already in the mask, not set below. */
6585 xlate_report(ctx
, OFT_ACTION
, "drop");
6589 OFPACT_FOR_EACH (a
, ofpacts
, ofpacts_len
) {
6590 struct ofpact_controller
*controller
;
6591 const struct ofpact_metadata
*metadata
;
6592 const struct ofpact_set_field
*set_field
;
6593 const struct mf_field
*mf
;
6594 bool last
= is_last_action
&& ofpact_last(a
, ofpacts
, ofpacts_len
)
6595 && ctx
->action_set
.size
;
6601 recirc_for_mpls(a
, ctx
);
6604 /* Check if need to store the remaining actions for later
6606 if (ctx
->freezing
) {
6607 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
),
6613 if (OVS_UNLIKELY(ctx
->xin
->trace
)) {
6614 struct ds s
= DS_EMPTY_INITIALIZER
;
6615 struct ofpact_format_params fp
= { .s
= &s
};
6616 ofpacts_format(a
, OFPACT_ALIGN(a
->len
), &fp
);
6617 xlate_report(ctx
, OFT_ACTION
, "%s", ds_cstr(&s
));
6623 xlate_output_action(ctx
, ofpact_get_OUTPUT(a
)->port
,
6624 ofpact_get_OUTPUT(a
)->max_len
, true, last
,
6625 false, group_bucket_action
);
6629 if (xlate_group_action(ctx
, ofpact_get_GROUP(a
)->group_id
, last
)) {
6630 /* Group could not be found. */
6632 /* XXX: Terminates action list translation, but does not
6633 * terminate the pipeline. */
6638 case OFPACT_CONTROLLER
:
6639 controller
= ofpact_get_CONTROLLER(a
);
6640 if (controller
->pause
) {
6641 ctx
->pause
= controller
;
6642 ctx_trigger_freeze(ctx
);
6645 xlate_controller_action(ctx
, controller
->max_len
,
6647 controller
->controller_id
,
6648 controller
->provider_meter_id
,
6649 controller
->userdata
,
6650 controller
->userdata_len
);
6654 case OFPACT_ENQUEUE
:
6655 memset(&wc
->masks
.skb_priority
, 0xff,
6656 sizeof wc
->masks
.skb_priority
);
6657 xlate_enqueue_action(ctx
, ofpact_get_ENQUEUE(a
), last
,
6658 group_bucket_action
);
6661 case OFPACT_SET_VLAN_VID
:
6662 wc
->masks
.vlans
[0].tci
|= htons(VLAN_VID_MASK
| VLAN_CFI
);
6663 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6664 ofpact_get_SET_VLAN_VID(a
)->push_vlan_if_needed
) {
6665 if (!flow
->vlans
[0].tpid
) {
6666 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6668 flow
->vlans
[0].tci
&= ~htons(VLAN_VID_MASK
);
6669 flow
->vlans
[0].tci
|=
6670 (htons(ofpact_get_SET_VLAN_VID(a
)->vlan_vid
) |
6675 case OFPACT_SET_VLAN_PCP
:
6676 wc
->masks
.vlans
[0].tci
|= htons(VLAN_PCP_MASK
| VLAN_CFI
);
6677 if (flow
->vlans
[0].tci
& htons(VLAN_CFI
) ||
6678 ofpact_get_SET_VLAN_PCP(a
)->push_vlan_if_needed
) {
6679 if (!flow
->vlans
[0].tpid
) {
6680 flow
->vlans
[0].tpid
= htons(ETH_TYPE_VLAN
);
6682 flow
->vlans
[0].tci
&= ~htons(VLAN_PCP_MASK
);
6683 flow
->vlans
[0].tci
|=
6684 htons((ofpact_get_SET_VLAN_PCP(a
)->vlan_pcp
6685 << VLAN_PCP_SHIFT
) | VLAN_CFI
);
6689 case OFPACT_STRIP_VLAN
:
6690 flow_pop_vlan(flow
, wc
);
6693 case OFPACT_PUSH_VLAN
:
6694 flow_push_vlan_uninit(flow
, wc
);
6695 flow
->vlans
[0].tpid
= ofpact_get_PUSH_VLAN(a
)->ethertype
;
6696 flow
->vlans
[0].tci
= htons(VLAN_CFI
);
6699 case OFPACT_SET_ETH_SRC
:
6700 WC_MASK_FIELD(wc
, dl_src
);
6701 flow
->dl_src
= ofpact_get_SET_ETH_SRC(a
)->mac
;
6704 case OFPACT_SET_ETH_DST
:
6705 WC_MASK_FIELD(wc
, dl_dst
);
6706 flow
->dl_dst
= ofpact_get_SET_ETH_DST(a
)->mac
;
6709 case OFPACT_SET_IPV4_SRC
:
6710 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6711 memset(&wc
->masks
.nw_src
, 0xff, sizeof wc
->masks
.nw_src
);
6712 flow
->nw_src
= ofpact_get_SET_IPV4_SRC(a
)->ipv4
;
6716 case OFPACT_SET_IPV4_DST
:
6717 if (flow
->dl_type
== htons(ETH_TYPE_IP
)) {
6718 memset(&wc
->masks
.nw_dst
, 0xff, sizeof wc
->masks
.nw_dst
);
6719 flow
->nw_dst
= ofpact_get_SET_IPV4_DST(a
)->ipv4
;
6723 case OFPACT_SET_IP_DSCP
:
6724 if (is_ip_any(flow
)) {
6725 wc
->masks
.nw_tos
|= IP_DSCP_MASK
;
6726 flow
->nw_tos
&= ~IP_DSCP_MASK
;
6727 flow
->nw_tos
|= ofpact_get_SET_IP_DSCP(a
)->dscp
;
6731 case OFPACT_SET_IP_ECN
:
6732 if (is_ip_any(flow
)) {
6733 wc
->masks
.nw_tos
|= IP_ECN_MASK
;
6734 flow
->nw_tos
&= ~IP_ECN_MASK
;
6735 flow
->nw_tos
|= ofpact_get_SET_IP_ECN(a
)->ecn
;
6739 case OFPACT_SET_IP_TTL
:
6740 if (is_ip_any(flow
)) {
6741 wc
->masks
.nw_ttl
= 0xff;
6742 flow
->nw_ttl
= ofpact_get_SET_IP_TTL(a
)->ttl
;
6746 case OFPACT_SET_L4_SRC_PORT
:
6747 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6748 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6749 memset(&wc
->masks
.tp_src
, 0xff, sizeof wc
->masks
.tp_src
);
6750 flow
->tp_src
= htons(ofpact_get_SET_L4_SRC_PORT(a
)->port
);
6754 case OFPACT_SET_L4_DST_PORT
:
6755 if (is_ip_any(flow
) && !(flow
->nw_frag
& FLOW_NW_FRAG_LATER
)) {
6756 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6757 memset(&wc
->masks
.tp_dst
, 0xff, sizeof wc
->masks
.tp_dst
);
6758 flow
->tp_dst
= htons(ofpact_get_SET_L4_DST_PORT(a
)->port
);
6762 case OFPACT_RESUBMIT
:
6763 /* Freezing complicates resubmit. Some action in the flow
6764 * entry found by resubmit might trigger freezing. If that
6765 * happens, then we do not want to execute the resubmit again after
6766 * during thawing, so we want to skip back to the head of the loop
6767 * to avoid that, only adding any actions that follow the resubmit
6768 * to the frozen actions.
6770 xlate_ofpact_resubmit(ctx
, ofpact_get_RESUBMIT(a
), last
);
6773 case OFPACT_SET_TUNNEL
:
6774 flow
->tunnel
.tun_id
= htonll(ofpact_get_SET_TUNNEL(a
)->tun_id
);
6777 case OFPACT_SET_QUEUE
:
6778 memset(&wc
->masks
.skb_priority
, 0xff,
6779 sizeof wc
->masks
.skb_priority
);
6780 xlate_set_queue_action(ctx
, ofpact_get_SET_QUEUE(a
)->queue_id
);
6783 case OFPACT_POP_QUEUE
:
6784 memset(&wc
->masks
.skb_priority
, 0xff,
6785 sizeof wc
->masks
.skb_priority
);
6786 if (flow
->skb_priority
!= ctx
->orig_skb_priority
) {
6787 flow
->skb_priority
= ctx
->orig_skb_priority
;
6788 xlate_report(ctx
, OFT_DETAIL
, "queue = %#"PRIx32
,
6789 flow
->skb_priority
);
6793 case OFPACT_REG_MOVE
:
6794 xlate_ofpact_reg_move(ctx
, ofpact_get_REG_MOVE(a
));
6797 case OFPACT_SET_FIELD
:
6798 set_field
= ofpact_get_SET_FIELD(a
);
6799 mf
= set_field
->field
;
6801 /* Set the field only if the packet actually has it. */
6802 if (mf_are_prereqs_ok(mf
, flow
, wc
)) {
6803 mf_mask_field_masked(mf
, ofpact_set_field_mask(set_field
), wc
);
6804 mf_set_flow_value_masked(mf
, set_field
->value
,
6805 ofpact_set_field_mask(set_field
),
6808 xlate_report(ctx
, OFT_WARN
,
6809 "unmet prerequisites for %s, set_field ignored",
6815 case OFPACT_STACK_PUSH
:
6816 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a
), flow
, wc
,
6820 case OFPACT_STACK_POP
:
6821 xlate_ofpact_stack_pop(ctx
, ofpact_get_STACK_POP(a
));
6824 case OFPACT_PUSH_MPLS
:
6825 compose_mpls_push_action(ctx
, ofpact_get_PUSH_MPLS(a
));
6828 case OFPACT_POP_MPLS
:
6829 compose_mpls_pop_action(ctx
, ofpact_get_POP_MPLS(a
)->ethertype
);
6832 case OFPACT_SET_MPLS_LABEL
:
6833 compose_set_mpls_label_action(
6834 ctx
, ofpact_get_SET_MPLS_LABEL(a
)->label
);
6837 case OFPACT_SET_MPLS_TC
:
6838 compose_set_mpls_tc_action(ctx
, ofpact_get_SET_MPLS_TC(a
)->tc
);
6841 case OFPACT_SET_MPLS_TTL
:
6842 compose_set_mpls_ttl_action(ctx
, ofpact_get_SET_MPLS_TTL(a
)->ttl
);
6845 case OFPACT_DEC_MPLS_TTL
:
6846 if (compose_dec_mpls_ttl_action(ctx
)) {
6851 case OFPACT_DEC_NSH_TTL
:
6852 if (compose_dec_nsh_ttl_action(ctx
)) {
6857 case OFPACT_DEC_TTL
:
6858 wc
->masks
.nw_ttl
= 0xff;
6859 if (compose_dec_ttl(ctx
, ofpact_get_DEC_TTL(a
))) {
6865 /* Nothing to do. */
6868 case OFPACT_MULTIPATH
:
6869 multipath_execute(ofpact_get_MULTIPATH(a
), flow
, wc
);
6870 xlate_report_subfield(ctx
, &ofpact_get_MULTIPATH(a
)->dst
);
6874 xlate_bundle_action(ctx
, ofpact_get_BUNDLE(a
), last
,
6875 group_bucket_action
);
6878 case OFPACT_OUTPUT_REG
:
6879 xlate_output_reg_action(ctx
, ofpact_get_OUTPUT_REG(a
), last
,
6880 group_bucket_action
);
6883 case OFPACT_OUTPUT_TRUNC
:
6884 xlate_output_trunc_action(ctx
, ofpact_get_OUTPUT_TRUNC(a
)->port
,
6885 ofpact_get_OUTPUT_TRUNC(a
)->max_len
, last
,
6886 group_bucket_action
);
6890 xlate_learn_action(ctx
, ofpact_get_LEARN(a
));
6893 case OFPACT_CONJUNCTION
:
6894 /* A flow with a "conjunction" action represents part of a special
6895 * kind of "set membership match". Such a flow should not actually
6896 * get executed, but it could via, say, a "packet-out", even though
6897 * that wouldn't be useful. Log it to help debugging. */
6898 xlate_report_error(ctx
, "executing no-op conjunction action");
6905 case OFPACT_UNROLL_XLATE
:
6906 xlate_ofpact_unroll_xlate(ctx
, ofpact_get_UNROLL_XLATE(a
));
6909 case OFPACT_FIN_TIMEOUT
:
6910 memset(&wc
->masks
.nw_proto
, 0xff, sizeof wc
->masks
.nw_proto
);
6911 xlate_fin_timeout(ctx
, ofpact_get_FIN_TIMEOUT(a
));
6914 case OFPACT_CLEAR_ACTIONS
:
6915 xlate_report_action_set(ctx
, "was");
6916 ofpbuf_clear(&ctx
->action_set
);
6917 ctx
->xin
->flow
.actset_output
= OFPP_UNSET
;
6918 ctx
->action_set_has_group
= false;
6921 case OFPACT_WRITE_ACTIONS
:
6922 xlate_write_actions(ctx
, ofpact_get_WRITE_ACTIONS(a
));
6923 xlate_report_action_set(ctx
, "is");
6926 case OFPACT_WRITE_METADATA
:
6927 metadata
= ofpact_get_WRITE_METADATA(a
);
6928 flow
->metadata
&= ~metadata
->mask
;
6929 flow
->metadata
|= metadata
->metadata
& metadata
->mask
;
6933 xlate_meter_action(ctx
, ofpact_get_METER(a
));
6936 case OFPACT_GOTO_TABLE
: {
6937 struct ofpact_goto_table
*ogt
= ofpact_get_GOTO_TABLE(a
);
6939 ovs_assert(ctx
->table_id
< ogt
->table_id
);
6941 xlate_table_action(ctx
, ctx
->xin
->flow
.in_port
.ofp_port
,
6942 ogt
->table_id
, true, true, false, last
,
6948 xlate_sample_action(ctx
, ofpact_get_SAMPLE(a
));
6952 compose_clone(ctx
, ofpact_get_CLONE(a
), last
);
6956 xlate_generic_encap_action(ctx
, ofpact_get_ENCAP(a
));
6959 case OFPACT_DECAP
: {
6960 bool recirc_needed
=
6961 xlate_generic_decap_action(ctx
, ofpact_get_DECAP(a
));
6962 if (!ctx
->error
&& recirc_needed
) {
6963 /* Recirculate for parsing of inner packet. */
6964 ctx_trigger_freeze(ctx
);
6965 /* Then continue with next action. */
6972 compose_conntrack_action(ctx
, ofpact_get_CT(a
), last
);
6975 case OFPACT_CT_CLEAR
:
6976 compose_ct_clear_action(ctx
);
6980 /* This will be processed by compose_conntrack_action(). */
6981 ctx
->ct_nat_action
= ofpact_get_NAT(a
);
6984 case OFPACT_DEBUG_RECIRC
:
6985 ctx_trigger_freeze(ctx
);
6989 case OFPACT_DEBUG_SLOW
:
6990 ctx
->xout
->slow
|= SLOW_ACTION
;
6993 case OFPACT_CHECK_PKT_LARGER
: {
6995 /* If this is last action, then there is no need to
6996 * translate the action. */
6999 const struct ofpact
*remaining_acts
= ofpact_next(a
);
7000 size_t remaining_acts_len
= ofpact_remaining_len(remaining_acts
,
7003 xlate_check_pkt_larger(ctx
, ofpact_get_CHECK_PKT_LARGER(a
),
7004 remaining_acts
, remaining_acts_len
);
7009 /* Check if need to store this and the remaining actions for later
7011 if (!ctx
->error
&& ctx
->exit
&& ctx_first_frozen_action(ctx
)) {
7012 freeze_unroll_actions(a
, ofpact_end(ofpacts
, ofpacts_len
), ctx
);
7019 xlate_in_init(struct xlate_in
*xin
, struct ofproto_dpif
*ofproto
,
7020 ovs_version_t version
, const struct flow
*flow
,
7021 ofp_port_t in_port
, struct rule_dpif
*rule
, uint16_t tcp_flags
,
7022 const struct dp_packet
*packet
, struct flow_wildcards
*wc
,
7023 struct ofpbuf
*odp_actions
)
7025 xin
->ofproto
= ofproto
;
7026 xin
->tables_version
= version
;
7028 xin
->upcall_flow
= flow
;
7029 xin
->flow
.in_port
.ofp_port
= in_port
;
7030 xin
->flow
.actset_output
= OFPP_UNSET
;
7031 xin
->packet
= packet
;
7032 xin
->allow_side_effects
= packet
!= NULL
;
7035 xin
->ofpacts
= NULL
;
7036 xin
->ofpacts_len
= 0;
7037 xin
->tcp_flags
= tcp_flags
;
7039 xin
->resubmit_stats
= NULL
;
7043 xin
->odp_actions
= odp_actions
;
7044 xin
->in_packet_out
= false;
7045 xin
->recirc_queue
= NULL
;
7046 xin
->xport_uuid
= UUID_ZERO
;
7048 /* Do recirc lookup. */
7049 xin
->frozen_state
= NULL
;
7050 if (flow
->recirc_id
) {
7051 const struct recirc_id_node
*node
7052 = recirc_id_node_find(flow
->recirc_id
);
7054 xin
->frozen_state
= &node
->state
;
7060 xlate_out_uninit(struct xlate_out
*xout
)
7063 recirc_refs_unref(&xout
->recircs
);
7067 static struct skb_priority_to_dscp
*
7068 get_skb_priority(const struct xport
*xport
, uint32_t skb_priority
)
7070 struct skb_priority_to_dscp
*pdscp
;
7073 hash
= hash_int(skb_priority
, 0);
7074 HMAP_FOR_EACH_IN_BUCKET (pdscp
, hmap_node
, hash
, &xport
->skb_priorities
) {
7075 if (pdscp
->skb_priority
== skb_priority
) {
7083 dscp_from_skb_priority(const struct xport
*xport
, uint32_t skb_priority
,
7086 struct skb_priority_to_dscp
*pdscp
= get_skb_priority(xport
, skb_priority
);
7087 *dscp
= pdscp
? pdscp
->dscp
: 0;
7088 return pdscp
!= NULL
;
7092 count_skb_priorities(const struct xport
*xport
)
7094 return hmap_count(&xport
->skb_priorities
);
7098 clear_skb_priorities(struct xport
*xport
)
7100 struct skb_priority_to_dscp
*pdscp
;
7102 HMAP_FOR_EACH_POP (pdscp
, hmap_node
, &xport
->skb_priorities
) {
7108 actions_output_to_local_port(const struct xlate_ctx
*ctx
)
7110 odp_port_t local_odp_port
= ofp_port_to_odp_port(ctx
->xbridge
, OFPP_LOCAL
);
7111 const struct nlattr
*a
;
7114 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, ctx
->odp_actions
->data
,
7115 ctx
->odp_actions
->size
) {
7116 if (nl_attr_type(a
) == OVS_ACTION_ATTR_OUTPUT
7117 && nl_attr_get_odp_port(a
) == local_odp_port
) {
7124 #if defined(__linux__)
7125 /* Returns the maximum number of packets that the Linux kernel is willing to
7126 * queue up internally to certain kinds of software-implemented ports, or the
7127 * default (and rarely modified) value if it cannot be determined. */
7129 netdev_max_backlog(void)
7131 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
7132 static int max_backlog
= 1000; /* The normal default value. */
7134 if (ovsthread_once_start(&once
)) {
7135 static const char filename
[] = "/proc/sys/net/core/netdev_max_backlog";
7139 stream
= fopen(filename
, "r");
7141 VLOG_INFO("%s: open failed (%s)", filename
, ovs_strerror(errno
));
7143 if (fscanf(stream
, "%d", &n
) != 1) {
7144 VLOG_WARN("%s: read error", filename
);
7145 } else if (n
<= 100) {
7146 VLOG_WARN("%s: unexpectedly small value %d", filename
, n
);
7152 ovsthread_once_done(&once
);
7154 VLOG_DBG("%s: using %d max_backlog", filename
, max_backlog
);
7160 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
7163 count_output_actions(const struct ofpbuf
*odp_actions
)
7165 const struct nlattr
*a
;
7169 NL_ATTR_FOR_EACH_UNSAFE (a
, left
, odp_actions
->data
, odp_actions
->size
) {
7170 if (a
->nla_type
== OVS_ACTION_ATTR_OUTPUT
) {
7176 #endif /* defined(__linux__) */
7178 /* Returns true if 'odp_actions' contains more output actions than the datapath
7179 * can reliably handle in one go. On Linux, this is the value of the
7180 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
7181 * packets that the kernel is willing to queue up for processing while the
7182 * datapath is processing a set of actions. */
7184 too_many_output_actions(const struct ofpbuf
*odp_actions OVS_UNUSED
)
7187 return (odp_actions
->size
/ NL_A_U32_SIZE
> netdev_max_backlog()
7188 && count_output_actions(odp_actions
) > netdev_max_backlog());
7190 /* OSes other than Linux might have similar limits, but we don't know how
7191 * to determine them.*/
7197 xlate_wc_init(struct xlate_ctx
*ctx
)
7199 flow_wildcards_init_catchall(ctx
->wc
);
7201 /* Some fields we consider to always be examined. */
7202 WC_MASK_FIELD(ctx
->wc
, packet_type
);
7203 WC_MASK_FIELD(ctx
->wc
, in_port
);
7204 WC_MASK_FIELD(ctx
->wc
, dl_type
);
7205 if (is_ip_any(&ctx
->xin
->flow
)) {
7206 WC_MASK_FIELD_MASK(ctx
->wc
, nw_frag
, FLOW_NW_FRAG_MASK
);
7209 if (ctx
->xbridge
->support
.odp
.recirc
) {
7210 /* Always exactly match recirc_id when datapath supports
7212 WC_MASK_FIELD(ctx
->wc
, recirc_id
);
7215 if (ctx
->xbridge
->netflow
) {
7216 netflow_mask_wc(&ctx
->xin
->flow
, ctx
->wc
);
7219 tnl_wc_init(&ctx
->xin
->flow
, ctx
->wc
);
7223 xlate_wc_finish(struct xlate_ctx
*ctx
)
7227 /* Clear the metadata and register wildcard masks, because we won't
7228 * use non-header fields as part of the cache. */
7229 flow_wildcards_clear_non_packet_fields(ctx
->wc
);
7231 /* Wildcard Ethernet address fields if the original packet type was not
7234 * (The Ethertype field is used even when the original packet type is not
7236 if (ctx
->xin
->upcall_flow
->packet_type
!= htonl(PT_ETH
)) {
7237 ctx
->wc
->masks
.dl_dst
= eth_addr_zero
;
7238 ctx
->wc
->masks
.dl_src
= eth_addr_zero
;
7241 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
7242 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
7243 * represent these fields. The datapath interface, on the other hand,
7244 * represents them with just 8 bits each. This means that if the high
7245 * 8 bits of the masks for these fields somehow become set, then they
7246 * will get chopped off by a round trip through the datapath, and
7247 * revalidation will spot that as an inconsistency and delete the flow.
7248 * Avoid the problem here by making sure that only the low 8 bits of
7249 * either field can be unwildcarded for ICMP.
7251 if (is_icmpv4(&ctx
->xin
->flow
, NULL
) || is_icmpv6(&ctx
->xin
->flow
, NULL
)) {
7252 ctx
->wc
->masks
.tp_src
&= htons(UINT8_MAX
);
7253 ctx
->wc
->masks
.tp_dst
&= htons(UINT8_MAX
);
7255 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
7256 for (i
= 0; i
< FLOW_MAX_VLAN_HEADERS
; i
++) {
7257 if (ctx
->wc
->masks
.vlans
[i
].tci
) {
7258 ctx
->wc
->masks
.vlans
[i
].tci
|= htons(VLAN_CFI
);
7262 /* The classifier might return masks that match on tp_src and tp_dst even
7263 * for later fragments. This happens because there might be flows that
7264 * match on tp_src or tp_dst without matching on the frag bits, because
7265 * it is not a prerequisite for OpenFlow. Since it is a prerequisite for
7266 * datapath flows and since tp_src and tp_dst are always going to be 0,
7267 * wildcard the fields here. */
7268 if (ctx
->xin
->flow
.nw_frag
& FLOW_NW_FRAG_LATER
) {
7269 ctx
->wc
->masks
.tp_src
= 0;
7270 ctx
->wc
->masks
.tp_dst
= 0;
7274 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
7276 * The caller must take responsibility for eventually freeing 'xout', with
7277 * xlate_out_uninit().
7278 * Returns 'XLATE_OK' if translation was successful. In case of an error an
7279 * empty set of actions will be returned in 'xin->odp_actions' (if non-NULL),
7280 * so that most callers may ignore the return value and transparently install a
7281 * drop flow when the translation fails. */
7283 xlate_actions(struct xlate_in
*xin
, struct xlate_out
*xout
)
7285 *xout
= (struct xlate_out
) {
7287 .recircs
= RECIRC_REFS_EMPTY_INITIALIZER
,
7290 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7291 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, xin
->ofproto
);
7293 return XLATE_BRIDGE_NOT_FOUND
;
7296 struct flow
*flow
= &xin
->flow
;
7298 uint8_t stack_stub
[1024];
7299 uint64_t action_set_stub
[1024 / 8];
7300 uint64_t frozen_actions_stub
[1024 / 8];
7301 uint64_t actions_stub
[256 / 8];
7302 struct ofpbuf scratch_actions
= OFPBUF_STUB_INITIALIZER(actions_stub
);
7303 struct xlate_ctx ctx
= {
7307 .orig_tunnel_ipv6_dst
= flow_tnl_dst(&flow
->tunnel
),
7310 .stack
= OFPBUF_STUB_INITIALIZER(stack_stub
),
7314 : &(struct flow_wildcards
) { .masks
= { .dl_type
= 0 } }),
7315 .odp_actions
= xin
->odp_actions
? xin
->odp_actions
: &scratch_actions
,
7317 .depth
= xin
->depth
,
7318 .resubmits
= xin
->resubmits
,
7319 .in_action_set
= false,
7320 .in_packet_out
= xin
->in_packet_out
,
7321 .pending_encap
= false,
7322 .pending_decap
= false,
7326 .rule_cookie
= OVS_BE64_MAX
,
7327 .orig_skb_priority
= flow
->skb_priority
,
7328 .sflow_n_outputs
= 0,
7329 .sflow_odp_port
= 0,
7330 .nf_output_iface
= NF_OUT_DROP
,
7336 .recirc_update_dp_hash
= false,
7337 .frozen_actions
= OFPBUF_STUB_INITIALIZER(frozen_actions_stub
),
7341 .conntracked
= false,
7343 .ct_nat_action
= NULL
,
7345 .action_set_has_group
= false,
7346 .action_set
= OFPBUF_STUB_INITIALIZER(action_set_stub
),
7349 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
7350 * the packet as the datapath will treat it for output actions. Our
7351 * datapath doesn't retain tunneling information without us re-setting
7352 * it, so clear the tunnel data.
7355 memset(&ctx
.base_flow
.tunnel
, 0, sizeof ctx
.base_flow
.tunnel
);
7357 ofpbuf_reserve(ctx
.odp_actions
, NL_A_U32_SIZE
);
7358 xlate_wc_init(&ctx
);
7360 COVERAGE_INC(xlate_actions
);
7362 xin
->trace
= xlate_report(&ctx
, OFT_BRIDGE
, "bridge(\"%s\")",
7364 if (xin
->frozen_state
) {
7365 const struct frozen_state
*state
= xin
->frozen_state
;
7367 struct ovs_list
*old_trace
= xin
->trace
;
7368 xin
->trace
= xlate_report(&ctx
, OFT_THAW
, "thaw");
7370 if (xin
->ofpacts_len
> 0 || ctx
.rule
) {
7371 xlate_report_error(&ctx
, "Recirculation conflict (%s)!",
7372 xin
->ofpacts_len
? "actions" : "rule");
7373 ctx
.error
= XLATE_RECIRCULATION_CONFLICT
;
7377 /* Set the bridge for post-recirculation processing if needed. */
7378 if (!uuid_equals(&ctx
.xbridge
->ofproto
->uuid
, &state
->ofproto_uuid
)) {
7379 const struct xbridge
*new_bridge
7380 = xbridge_lookup_by_uuid(xcfg
, &state
->ofproto_uuid
);
7382 if (OVS_UNLIKELY(!new_bridge
)) {
7383 /* Drop the packet if the bridge cannot be found. */
7384 xlate_report_error(&ctx
, "Frozen bridge no longer exists.");
7385 ctx
.error
= XLATE_BRIDGE_NOT_FOUND
;
7386 xin
->trace
= old_trace
;
7389 ctx
.xbridge
= new_bridge
;
7390 /* The bridge is now known so obtain its table version. */
7391 ctx
.xin
->tables_version
7392 = ofproto_dpif_get_tables_version(ctx
.xbridge
->ofproto
);
7395 /* Set the thawed table id. Note: A table lookup is done only if there
7396 * are no frozen actions. */
7397 ctx
.table_id
= state
->table_id
;
7398 xlate_report(&ctx
, OFT_THAW
,
7399 "Resuming from table %"PRIu8
, ctx
.table_id
);
7401 ctx
.conntracked
= state
->conntracked
;
7402 if (!state
->conntracked
) {
7403 clear_conntrack(&ctx
);
7406 /* Restore pipeline metadata. May change flow's in_port and other
7407 * metadata to the values that existed when freezing was triggered. */
7408 frozen_metadata_to_flow(&state
->metadata
, flow
);
7410 /* Restore stack, if any. */
7412 ofpbuf_put(&ctx
.stack
, state
->stack
, state
->stack_size
);
7415 /* Restore mirror state. */
7416 ctx
.mirrors
= state
->mirrors
;
7418 /* Restore action set, if any. */
7419 if (state
->action_set_len
) {
7420 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring action set",
7421 state
->action_set
, state
->action_set_len
);
7423 flow
->actset_output
= OFPP_UNSET
;
7424 xlate_write_actions__(&ctx
, state
->action_set
,
7425 state
->action_set_len
);
7428 /* Restore frozen actions. If there are no actions, processing will
7429 * start with a lookup in the table set above. */
7430 xin
->ofpacts
= state
->ofpacts
;
7431 xin
->ofpacts_len
= state
->ofpacts_len
;
7432 if (state
->ofpacts_len
) {
7433 xlate_report_actions(&ctx
, OFT_THAW
, "Restoring actions",
7434 xin
->ofpacts
, xin
->ofpacts_len
);
7437 xin
->trace
= old_trace
;
7438 } else if (OVS_UNLIKELY(flow
->recirc_id
)) {
7439 xlate_report_error(&ctx
,
7440 "Recirculation context not found for ID %"PRIx32
,
7442 ctx
.error
= XLATE_NO_RECIRCULATION_CONTEXT
;
7446 /* Tunnel metadata in udpif format must be normalized before translation. */
7447 if (flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7448 const struct tun_table
*tun_tab
= ofproto_get_tun_tab(
7449 &ctx
.xbridge
->ofproto
->up
);
7452 err
= tun_metadata_from_geneve_udpif(tun_tab
, &xin
->upcall_flow
->tunnel
,
7453 &xin
->upcall_flow
->tunnel
,
7456 xlate_report_error(&ctx
, "Invalid Geneve tunnel metadata");
7457 ctx
.error
= XLATE_INVALID_TUNNEL_METADATA
;
7460 } else if (!flow
->tunnel
.metadata
.tab
|| xin
->frozen_state
) {
7461 /* If the original flow did not come in on a tunnel, then it won't have
7462 * FLOW_TNL_F_UDPIF set. However, we still need to have a metadata
7463 * table in case we generate tunnel actions. */
7464 /* If the translation is from a frozen state, we use the latest
7465 * TLV map to avoid segmentation fault in case the old TLV map is
7466 * replaced by a new one.
7467 * XXX: It is better to abort translation if the table is changed. */
7468 flow
->tunnel
.metadata
.tab
= ofproto_get_tun_tab(
7469 &ctx
.xbridge
->ofproto
->up
);
7471 ctx
.wc
->masks
.tunnel
.metadata
.tab
= flow
->tunnel
.metadata
.tab
;
7473 /* Get the proximate input port of the packet. (If xin->frozen_state,
7474 * flow->in_port is the ultimate input port of the packet.) */
7475 struct xport
*in_port
= get_ofp_port(xbridge
,
7476 ctx
.base_flow
.in_port
.ofp_port
);
7477 if (in_port
&& !in_port
->peer
) {
7478 ctx
.xin
->xport_uuid
= in_port
->uuid
;
7481 if (flow
->packet_type
!= htonl(PT_ETH
) && in_port
&&
7482 in_port
->pt_mode
== NETDEV_PT_LEGACY_L3
&& ctx
.table_id
== 0) {
7483 /* Add dummy Ethernet header to non-L2 packet if it's coming from a
7484 * L3 port. So all packets will be L2 packets for lookup.
7485 * The dl_type has already been set from the packet_type. */
7486 flow
->packet_type
= htonl(PT_ETH
);
7487 flow
->dl_src
= eth_addr_zero
;
7488 flow
->dl_dst
= eth_addr_zero
;
7489 ctx
.pending_encap
= true;
7492 if (!xin
->ofpacts
&& !ctx
.rule
) {
7493 ctx
.rule
= rule_dpif_lookup_from_table(
7494 ctx
.xbridge
->ofproto
, ctx
.xin
->tables_version
, flow
, ctx
.wc
,
7495 ctx
.xin
->resubmit_stats
, &ctx
.table_id
,
7496 flow
->in_port
.ofp_port
, true, true, ctx
.xin
->xcache
);
7497 if (ctx
.xin
->resubmit_stats
) {
7498 rule_dpif_credit_stats(ctx
.rule
, ctx
.xin
->resubmit_stats
);
7500 if (ctx
.xin
->xcache
) {
7501 struct xc_entry
*entry
;
7503 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_RULE
);
7504 entry
->rule
= ctx
.rule
;
7505 ofproto_rule_ref(&ctx
.rule
->up
);
7508 xlate_report_table(&ctx
, ctx
.rule
, ctx
.table_id
);
7511 /* Tunnel stats only for not-thawed packets. */
7512 if (!xin
->frozen_state
&& in_port
&& in_port
->is_tunnel
) {
7513 if (ctx
.xin
->resubmit_stats
) {
7514 netdev_vport_inc_rx(in_port
->netdev
, ctx
.xin
->resubmit_stats
);
7516 bfd_account_rx(in_port
->bfd
, ctx
.xin
->resubmit_stats
);
7519 if (ctx
.xin
->xcache
) {
7520 struct xc_entry
*entry
;
7522 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETDEV
);
7523 entry
->dev
.rx
= netdev_ref(in_port
->netdev
);
7524 entry
->dev
.bfd
= bfd_ref(in_port
->bfd
);
7528 if (!xin
->frozen_state
&& process_special(&ctx
, in_port
)) {
7529 /* process_special() did all the processing for this packet.
7531 * We do not perform special processing on thawed packets, since that
7532 * was done before they were frozen and should not be redone. */
7533 mirror_ingress_packet(&ctx
);
7534 } else if (in_port
&& in_port
->xbundle
7535 && xbundle_mirror_out(xbridge
, in_port
->xbundle
)) {
7536 xlate_report_error(&ctx
, "dropping packet received on port "
7537 "%s, which is reserved exclusively for mirroring",
7538 in_port
->xbundle
->name
);
7540 /* Sampling is done on initial reception; don't redo after thawing. */
7541 unsigned int user_cookie_offset
= 0;
7542 if (!xin
->frozen_state
) {
7543 user_cookie_offset
= compose_sflow_action(&ctx
);
7544 compose_ipfix_action(&ctx
, ODPP_NONE
);
7546 size_t sample_actions_len
= ctx
.odp_actions
->size
;
7548 if (tnl_process_ecn(flow
)
7549 && (!in_port
|| may_receive(in_port
, &ctx
))) {
7550 const struct ofpact
*ofpacts
;
7554 ofpacts
= xin
->ofpacts
;
7555 ofpacts_len
= xin
->ofpacts_len
;
7556 } else if (ctx
.rule
) {
7557 const struct rule_actions
*actions
7558 = rule_get_actions(&ctx
.rule
->up
);
7559 ofpacts
= actions
->ofpacts
;
7560 ofpacts_len
= actions
->ofpacts_len
;
7561 ctx
.rule_cookie
= ctx
.rule
->up
.flow_cookie
;
7566 mirror_ingress_packet(&ctx
);
7567 do_xlate_actions(ofpacts
, ofpacts_len
, &ctx
, true, false);
7572 /* We've let OFPP_NORMAL and the learning action look at the
7573 * packet, so cancel all actions and freezing if forwarding is
7575 if (in_port
&& (!xport_stp_forward_state(in_port
) ||
7576 !xport_rstp_forward_state(in_port
))) {
7577 ctx
.odp_actions
->size
= sample_actions_len
;
7578 ctx_cancel_freeze(&ctx
);
7579 ofpbuf_clear(&ctx
.action_set
);
7582 if (!ctx
.freezing
) {
7583 xlate_action_set(&ctx
);
7586 finish_freezing(&ctx
);
7590 /* Output only fully processed packets. */
7592 && xbridge
->has_in_band
7593 && in_band_must_output_to_local_port(flow
)
7594 && !actions_output_to_local_port(&ctx
)) {
7595 WC_MASK_FIELD(ctx
.wc
, nw_proto
);
7596 WC_MASK_FIELD(ctx
.wc
, tp_src
);
7597 WC_MASK_FIELD(ctx
.wc
, tp_dst
);
7598 WC_MASK_FIELD(ctx
.wc
, dl_type
);
7599 xlate_report(&ctx
, OFT_DETAIL
, "outputting DHCP packet "
7600 "to local port for in-band control");
7601 compose_output_action(&ctx
, OFPP_LOCAL
, NULL
, false, false);
7604 if (user_cookie_offset
) {
7605 fix_sflow_action(&ctx
, user_cookie_offset
);
7609 if (nl_attr_oversized(ctx
.odp_actions
->size
)) {
7610 /* These datapath actions are too big for a Netlink attribute, so we
7611 * can't hand them to the kernel directly. dpif_execute() can execute
7612 * them one by one with help, so just mark the result as SLOW_ACTION to
7613 * prevent the flow from being installed. */
7614 COVERAGE_INC(xlate_actions_oversize
);
7615 ctx
.xout
->slow
|= SLOW_ACTION
;
7616 } else if (too_many_output_actions(ctx
.odp_actions
)) {
7617 COVERAGE_INC(xlate_actions_too_many_output
);
7618 ctx
.xout
->slow
|= SLOW_ACTION
;
7621 /* Update NetFlow for non-frozen traffic. */
7622 if (xbridge
->netflow
&& !xin
->frozen_state
) {
7623 if (ctx
.xin
->resubmit_stats
) {
7624 netflow_flow_update(xbridge
->netflow
, flow
,
7625 ctx
.nf_output_iface
,
7626 ctx
.xin
->resubmit_stats
);
7628 if (ctx
.xin
->xcache
) {
7629 struct xc_entry
*entry
;
7631 entry
= xlate_cache_add_entry(ctx
.xin
->xcache
, XC_NETFLOW
);
7632 entry
->nf
.netflow
= netflow_ref(xbridge
->netflow
);
7633 entry
->nf
.flow
= xmemdup(flow
, sizeof *flow
);
7634 entry
->nf
.iface
= ctx
.nf_output_iface
;
7638 /* Translate tunnel metadata masks to udpif format if necessary. */
7639 if (xin
->upcall_flow
->tunnel
.flags
& FLOW_TNL_F_UDPIF
) {
7640 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7641 const struct flow_tnl
*upcall_tnl
= &xin
->upcall_flow
->tunnel
;
7642 struct geneve_opt opts
[TLV_TOT_OPT_SIZE
/
7643 sizeof(struct geneve_opt
)];
7645 tun_metadata_to_geneve_udpif_mask(&flow
->tunnel
,
7646 &ctx
.wc
->masks
.tunnel
,
7647 upcall_tnl
->metadata
.opts
.gnv
,
7648 upcall_tnl
->metadata
.present
.len
,
7650 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7651 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7652 memcpy(&ctx
.wc
->masks
.tunnel
.metadata
.opts
.gnv
, opts
,
7653 upcall_tnl
->metadata
.present
.len
);
7655 ctx
.wc
->masks
.tunnel
.metadata
.present
.len
= 0xff;
7656 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7657 ctx
.wc
->masks
.tunnel
.flags
|= FLOW_TNL_F_UDPIF
;
7658 } else if (!xin
->upcall_flow
->tunnel
.metadata
.tab
) {
7659 /* If we didn't have options in UDPIF format and didn't have an existing
7660 * metadata table, then it means that there were no options at all when
7661 * we started processing and any wildcards we picked up were from
7662 * action generation. Without options on the incoming packet, wildcards
7663 * aren't meaningful. To avoid them possibly getting misinterpreted,
7664 * just clear everything. */
7665 if (ctx
.wc
->masks
.tunnel
.metadata
.present
.map
) {
7666 memset(&ctx
.wc
->masks
.tunnel
.metadata
, 0,
7667 sizeof ctx
.wc
->masks
.tunnel
.metadata
);
7669 ctx
.wc
->masks
.tunnel
.metadata
.tab
= NULL
;
7673 xlate_wc_finish(&ctx
);
7676 /* Reset the table to what it was when we came in. If we only fetched
7677 * it locally, then it has no meaning outside of flow translation. */
7678 flow
->tunnel
.metadata
.tab
= xin
->upcall_flow
->tunnel
.metadata
.tab
;
7680 ofpbuf_uninit(&ctx
.stack
);
7681 ofpbuf_uninit(&ctx
.action_set
);
7682 ofpbuf_uninit(&ctx
.frozen_actions
);
7683 ofpbuf_uninit(&scratch_actions
);
7684 ofpbuf_delete(ctx
.encap_data
);
7686 /* Make sure we return a "drop flow" in case of an error. */
7689 if (xin
->odp_actions
) {
7690 ofpbuf_clear(xin
->odp_actions
);
7697 xlate_resume(struct ofproto_dpif
*ofproto
,
7698 const struct ofputil_packet_in_private
*pin
,
7699 struct ofpbuf
*odp_actions
,
7700 enum slow_path_reason
*slow
,
7702 struct xlate_cache
*xcache
)
7704 struct dp_packet packet
;
7705 dp_packet_use_const(&packet
, pin
->base
.packet
,
7706 pin
->base
.packet_len
);
7708 pkt_metadata_from_flow(&packet
.md
, &pin
->base
.flow_metadata
.flow
);
7709 flow_extract(&packet
, flow
);
7711 struct xlate_in xin
;
7712 xlate_in_init(&xin
, ofproto
, ofproto_dpif_get_tables_version(ofproto
),
7713 flow
, 0, NULL
, ntohs(flow
->tcp_flags
),
7714 &packet
, NULL
, odp_actions
);
7715 xin
.xcache
= xcache
;
7717 struct ofpact_note noop
;
7718 ofpact_init_NOTE(&noop
);
7721 bool any_actions
= pin
->actions_len
> 0;
7722 struct frozen_state state
= {
7723 .table_id
= 0, /* Not the table where NXAST_PAUSE was executed. */
7724 .ofproto_uuid
= pin
->bridge
,
7725 .stack
= pin
->stack
,
7726 .stack_size
= pin
->stack_size
,
7727 .mirrors
= pin
->mirrors
,
7728 .conntracked
= pin
->conntracked
,
7729 .xport_uuid
= UUID_ZERO
,
7731 /* When there are no actions, xlate_actions() will search the flow
7732 * table. We don't want it to do that (we want it to resume), so
7733 * supply a no-op action if there aren't any.
7735 * (We can't necessarily avoid translating actions entirely if there
7736 * aren't any actions, because there might be some finishing-up to do
7737 * at the end of the pipeline, and we don't check for those
7739 .ofpacts
= any_actions
? pin
->actions
: &noop
.ofpact
,
7740 .ofpacts_len
= any_actions
? pin
->actions_len
: sizeof noop
,
7742 .action_set
= pin
->action_set
,
7743 .action_set_len
= pin
->action_set_len
,
7745 frozen_metadata_from_flow(&state
.metadata
,
7746 &pin
->base
.flow_metadata
.flow
);
7747 xin
.frozen_state
= &state
;
7749 struct xlate_out xout
;
7750 enum xlate_error error
= xlate_actions(&xin
, &xout
);
7752 xlate_out_uninit(&xout
);
7754 /* xlate_actions() can generate a number of errors, but only
7755 * XLATE_BRIDGE_NOT_FOUND really stands out to me as one that we should be
7756 * sure to report over OpenFlow. The others could come up in packet-outs
7757 * or regular flow translation and I don't think that it's going to be too
7758 * useful to report them to the controller. */
7759 return error
== XLATE_BRIDGE_NOT_FOUND
? OFPERR_NXR_STALE
: 0;
7762 /* Sends 'packet' out 'ofport'. If 'port' is a tunnel and that tunnel type
7763 * supports a notion of an OAM flag, sets it if 'oam' is true.
7764 * May modify 'packet'.
7765 * Returns 0 if successful, otherwise a positive errno value. */
7767 xlate_send_packet(const struct ofport_dpif
*ofport
, bool oam
,
7768 struct dp_packet
*packet
)
7770 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7771 struct xport
*xport
;
7772 uint64_t ofpacts_stub
[1024 / 8];
7773 struct ofpbuf ofpacts
;
7776 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
7777 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
7778 flow_extract(packet
, &flow
);
7779 flow
.in_port
.ofp_port
= OFPP_NONE
;
7781 xport
= xport_lookup(xcfg
, ofport
);
7787 const ovs_be16 flag
= htons(NX_TUN_FLAG_OAM
);
7788 ofpact_put_set_field(&ofpacts
, mf_from_id(MFF_TUN_FLAGS
),
7792 ofpact_put_OUTPUT(&ofpacts
)->port
= xport
->ofp_port
;
7794 /* Actions here are not referring to anything versionable (flow tables or
7795 * groups) so we don't need to worry about the version here. */
7796 return ofproto_dpif_execute_actions(xport
->xbridge
->ofproto
,
7797 OVS_VERSION_MAX
, &flow
, NULL
,
7798 ofpacts
.data
, ofpacts
.size
, packet
);
7802 xlate_mac_learning_update(const struct ofproto_dpif
*ofproto
,
7803 ofp_port_t in_port
, struct eth_addr dl_src
,
7804 int vlan
, bool is_grat_arp
)
7806 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7807 struct xbridge
*xbridge
;
7808 struct xbundle
*xbundle
;
7810 xbridge
= xbridge_lookup(xcfg
, ofproto
);
7815 xbundle
= lookup_input_bundle__(xbridge
, in_port
, NULL
);
7820 update_learning_table__(xbridge
, xbundle
, dl_src
, vlan
, is_grat_arp
);
7824 xlate_set_support(const struct ofproto_dpif
*ofproto
,
7825 const struct dpif_backer_support
*support
)
7827 struct xlate_cfg
*xcfg
= ovsrcu_get(struct xlate_cfg
*, &xcfgp
);
7828 struct xbridge
*xbridge
= xbridge_lookup(xcfg
, ofproto
);
7831 xbridge
->support
= *support
;