1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 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 <arpa/inet.h>
22 #include <sys/socket.h>
23 #include <netinet/in.h>
29 #include "byte-order.h"
33 #include "dp-packet.h"
38 #include "mac-learning.h"
39 #include "mcast-snooping.h"
40 #include "multipath.h"
41 #include "netdev-vport.h"
44 #include "odp-execute.h"
45 #include "ofproto/ofproto-dpif-ipfix.h"
46 #include "ofproto/ofproto-dpif-mirror.h"
47 #include "ofproto/ofproto-dpif-monitor.h"
48 #include "ofproto/ofproto-dpif-sflow.h"
49 #include "ofproto/ofproto-dpif.h"
50 #include "ofproto/ofproto-provider.h"
51 #include "openvswitch/dynamic-string.h"
52 #include "openvswitch/meta-flow.h"
53 #include "openvswitch/list.h"
54 #include "openvswitch/ofp-actions.h"
55 #include "openvswitch/vlog.h"
57 #include "ovs-router.h"
59 #include "tnl-neigh-cache.h"
60 #include "tnl-ports.h"
63 COVERAGE_DEFINE(xlate_actions);
64 COVERAGE_DEFINE(xlate_actions_oversize);
65 COVERAGE_DEFINE(xlate_actions_too_many_output);
67 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * The goal of limiting the depth of resubmits is to ensure that flow
73 * translation eventually terminates. Only resubmits to the same table or an
74 * earlier table count against the maximum depth. This is because resubmits to
75 * strictly monotonically increasing table IDs will eventually terminate, since
76 * any OpenFlow switch has a finite number of tables. OpenFlow tables are most
77 * commonly traversed in numerically increasing order, so this limit has little
78 * effect on conventionally designed OpenFlow pipelines.
80 * Outputs to patch ports and to groups also count against the depth limit. */
83 /* Maximum number of resubmit actions in a flow translation, whether they are
84 * recursive or not. */
85 #define MAX_RESUBMITS (MAX_DEPTH * MAX_DEPTH)
88 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
89 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
91 struct ovs_list xbundles; /* Owned xbundles. */
92 struct hmap xports; /* Indexed by ofp_port. */
94 char *name; /* Name used in log messages. */
95 struct dpif *dpif; /* Datapath interface. */
96 struct mac_learning *ml; /* Mac learning handle. */
97 struct mcast_snooping *ms; /* Multicast Snooping handle. */
98 struct mbridge *mbridge; /* Mirroring. */
99 struct dpif_sflow *sflow; /* SFlow handle, or null. */
100 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
101 struct netflow *netflow; /* Netflow handle, or null. */
102 struct stp *stp; /* STP or null if disabled. */
103 struct rstp *rstp; /* RSTP or null if disabled. */
105 bool has_in_band; /* Bridge has in band control? */
106 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
108 /* Datapath feature support. */
109 struct dpif_backer_support support;
113 struct hmap_node hmap_node; /* In global 'xbundles' map. */
114 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
116 struct ovs_list list_node; /* In parent 'xbridges' list. */
117 struct xbridge *xbridge; /* Parent xbridge. */
119 struct ovs_list xports; /* Contains "struct xport"s. */
121 char *name; /* Name used in log messages. */
122 struct bond *bond; /* Nonnull iff more than one port. */
123 struct lacp *lacp; /* LACP handle or null. */
125 enum port_vlan_mode vlan_mode; /* VLAN mode. */
126 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
127 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
128 * NULL if all VLANs are trunked. */
129 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
130 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
134 struct hmap_node hmap_node; /* Node in global 'xports' map. */
135 struct ofport_dpif *ofport; /* Key in global 'xports map. */
137 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
138 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
140 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
142 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
143 struct xbundle *xbundle; /* Parent xbundle or null. */
145 struct netdev *netdev; /* 'ofport''s netdev. */
147 struct xbridge *xbridge; /* Parent bridge. */
148 struct xport *peer; /* Patch port peer or null. */
150 enum ofputil_port_config config; /* OpenFlow port configuration. */
151 enum ofputil_port_state state; /* OpenFlow port state. */
152 int stp_port_no; /* STP port number or -1 if not in use. */
153 struct rstp_port *rstp_port; /* RSTP port or null. */
155 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
157 bool may_enable; /* May be enabled in bonds. */
158 bool is_tunnel; /* Is a tunnel port. */
160 struct cfm *cfm; /* CFM handle or null. */
161 struct bfd *bfd; /* BFD handle or null. */
162 struct lldp *lldp; /* LLDP handle or null. */
166 struct xlate_in *xin;
167 struct xlate_out *xout;
169 const struct xbridge *xbridge;
171 /* Flow tables version at the beginning of the translation. */
172 cls_version_t tables_version;
174 /* Flow at the last commit. */
175 struct flow base_flow;
177 /* Tunnel IP destination address as received. This is stored separately
178 * as the base_flow.tunnel is cleared on init to reflect the datapath
179 * behavior. Used to make sure not to send tunneled output to ourselves,
180 * which might lead to an infinite loop. This could happen easily
181 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
182 * actually set the tun_dst field. */
183 struct in6_addr orig_tunnel_ipv6_dst;
185 /* Stack for the push and pop actions. Each stack element is of type
186 * "union mf_subvalue". */
189 /* The rule that we are currently translating, or NULL. */
190 struct rule_dpif *rule;
192 /* Flow translation populates this with wildcards relevant in translation.
193 * When 'xin->wc' is nonnull, this is the same pointer. When 'xin->wc' is
194 * null, this is a pointer to a temporary buffer. */
195 struct flow_wildcards *wc;
197 /* Output buffer for datapath actions. When 'xin->odp_actions' is nonnull,
198 * this is the same pointer. When 'xin->odp_actions' is null, this points
199 * to a scratch ofpbuf. This allows code to add actions to
200 * 'ctx->odp_actions' without worrying about whether the caller really
202 struct ofpbuf *odp_actions;
204 /* Statistics maintained by xlate_table_action().
206 * 'indentation' is the nesting level for resubmits. It is used to indent
207 * the output of resubmit_hook (e.g. for the "ofproto/trace" feature).
209 * The other statistics limit the amount of work that a single flow
210 * translation can perform. The goal of the first of these, 'depth', is
211 * primarily to prevent translation from performing an infinite amount of
212 * work. It counts the current depth of nested "resubmit"s (and a few
213 * other activities); when a resubmit returns, it decreases. Resubmits to
214 * tables in strictly monotonically increasing order don't contribute to
215 * 'depth' because they cannot cause a flow translation to take an infinite
216 * amount of time (because the number of tables is finite). Translation
217 * aborts when 'depth' exceeds MAX_DEPTH.
219 * 'resubmits', on the other hand, prevents flow translation from
220 * performing an extraordinarily large while still finite amount of work.
221 * It counts the total number of resubmits (and a few other activities)
222 * that have been executed. Returning from a resubmit does not affect this
223 * counter. Thus, this limits the amount of work that a particular
224 * translation can perform. Translation aborts when 'resubmits' exceeds
225 * MAX_RESUBMITS (which is much larger than MAX_DEPTH).
227 int indentation; /* Indentation level for resubmit_hook. */
228 int depth; /* Current resubmit nesting depth. */
229 int resubmits; /* Total number of resubmits. */
230 bool in_group; /* Currently translating ofgroup, if true. */
231 bool in_action_set; /* Currently translating action_set, if true. */
233 uint8_t table_id; /* OpenFlow table ID where flow was found. */
234 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
235 uint32_t orig_skb_priority; /* Priority when packet arrived. */
236 uint32_t sflow_n_outputs; /* Number of output ports. */
237 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
238 ofp_port_t nf_output_iface; /* Output interface index for NetFlow. */
239 bool exit; /* No further actions should be processed. */
240 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
242 /* Freezing Translation
243 * ====================
245 * At some point during translation, the code may recognize the need to halt
246 * and checkpoint the translation in a way that it can be restarted again
247 * later. We call the checkpointing process "freezing" and the restarting
250 * The use cases for freezing are:
252 * - "Recirculation", where the translation process discovers that it
253 * doesn't have enough information to complete translation without
254 * actually executing the actions that have already been translated,
255 * which provides the additionally needed information. In these
256 * situations, translation freezes translation and assigns the frozen
257 * data a unique "recirculation ID", which it associates with the data
258 * in a table in userspace (see ofproto-dpif-rid.h). It also adds a
259 * OVS_ACTION_ATTR_RECIRC action specifying that ID to the datapath
260 * actions. When a packet hits that action, the datapath looks its
261 * flow up again using the ID. If there's a miss, it comes back to
262 * userspace, which find the recirculation table entry for the ID,
263 * thaws the associated frozen data, and continues translation from
264 * that point given the additional information that is now known.
266 * The archetypal example is MPLS. As MPLS is implemented in
267 * OpenFlow, the protocol that follows the last MPLS label becomes
268 * known only when that label is popped by an OpenFlow action. That
269 * means that Open vSwitch can't extract the headers beyond the MPLS
270 * labels until the pop action is executed. Thus, at that point
271 * translation uses the recirculation process to extract the headers
272 * beyond the MPLS labels.
274 * (OVS also uses OVS_ACTION_ATTR_RECIRC to implement hashing for
275 * output to bonds. OVS pre-populates all the datapath flows for bond
276 * output in the datapath, though, which means that the elaborate
277 * process of coming back to userspace for a second round of
278 * translation isn't needed, and so bonds don't follow the above
281 * - "Continuation". A continuation is a way for an OpenFlow controller
282 * to interpose on a packet's traversal of the OpenFlow tables. When
283 * the translation process encounters a "controller" action with the
284 * "pause" flag, it freezes translation, serializes the frozen data,
285 * and sends it to an OpenFlow controller. The controller then
286 * examines and possibly modifies the frozen data and eventually sends
287 * it back to the switch, which thaws it and continues translation.
289 * The main problem of freezing translation is preserving state, so that
290 * when the translation is thawed later it resumes from where it left off,
291 * without disruption. In particular, actions must be preserved as follows:
293 * - If we're freezing because an action needed more information, the
294 * action that prompted it.
296 * - Any actions remaining to be translated within the current flow.
298 * - If translation was frozen within a NXAST_RESUBMIT, then any actions
299 * following the resubmit action. Resubmit actions can be nested, so
300 * this has to go all the way up the control stack.
302 * - The OpenFlow 1.1+ action set.
304 * State that actions and flow table lookups can depend on, such as the
305 * following, must also be preserved:
307 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
309 * - The stack used by NXAST_STACK_PUSH and NXAST_STACK_POP actions.
311 * - The table ID and cookie of the flow being translated at each level
312 * of the control stack, because these can become visible through
313 * OFPAT_CONTROLLER actions (and other ways).
315 * Translation allows for the control of this state preservation via these
316 * members. When a need to freeze translation is identified, the
317 * translation process:
319 * 1. Sets 'freezing' to true.
321 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
322 * translation process.
324 * 3. Adds an OFPACT_UNROLL_XLATE action to 'frozen_actions', and points
325 * frozen_actions.header to the action to make it easy to find it later.
326 * This action holds the current table ID and cookie so that they can be
327 * restored during a post-recirculation upcall translation.
329 * 4. Adds the action that prompted recirculation and any actions following
330 * it within the same flow to 'frozen_actions', so that they can be
331 * executed during a post-recirculation upcall translation.
335 * 6. The action that prompted recirculation might be nested in a stack of
336 * nested "resubmit"s that have actions remaining. Each of these notices
337 * that we're exiting and freezing and responds by adding more
338 * OFPACT_UNROLL_XLATE actions to 'frozen_actions', as necessary,
339 * followed by any actions that were yet unprocessed.
341 * If we're freezing because of recirculation, the caller generates a
342 * recirculation ID and associates all the state produced by this process
343 * with it. For post-recirculation upcall translation, the caller passes it
344 * back in for the new translation to execute. The process yielded a set of
345 * ofpacts that can be translated directly, so it is not much of a special
346 * case at that point.
349 struct ofpbuf frozen_actions;
350 const struct ofpact_controller *pause;
352 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
353 * This is a trigger for recirculation in cases where translating an action
354 * or looking up a flow requires access to the fields of the packet after
355 * the MPLS label stack that was originally present. */
358 /* True if conntrack has been performed on this packet during processing
359 * on the current bridge. This is used to determine whether conntrack
360 * state from the datapath should be honored after thawing. */
363 /* Pointer to an embedded NAT action in a conntrack action, or NULL. */
364 struct ofpact_nat *ct_nat_action;
366 /* OpenFlow 1.1+ action set.
368 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
369 * When translation is otherwise complete, ofpacts_execute_action_set()
370 * converts it to a set of "struct ofpact"s that can be translated into
371 * datapath actions. */
372 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
373 struct ofpbuf action_set; /* Action set. */
375 enum xlate_error error; /* Translation failed. */
378 const char *xlate_strerror(enum xlate_error error)
383 case XLATE_BRIDGE_NOT_FOUND:
384 return "Bridge not found";
385 case XLATE_RECURSION_TOO_DEEP:
386 return "Recursion too deep";
387 case XLATE_TOO_MANY_RESUBMITS:
388 return "Too many resubmits";
389 case XLATE_STACK_TOO_DEEP:
390 return "Stack too deep";
391 case XLATE_NO_RECIRCULATION_CONTEXT:
392 return "No recirculation context";
393 case XLATE_RECIRCULATION_CONFLICT:
394 return "Recirculation conflict";
395 case XLATE_TOO_MANY_MPLS_LABELS:
396 return "Too many MPLS labels";
398 return "Unknown error";
401 static void xlate_action_set(struct xlate_ctx *ctx);
402 static void xlate_commit_actions(struct xlate_ctx *ctx);
405 ctx_trigger_freeze(struct xlate_ctx *ctx)
408 ctx->freezing = true;
412 ctx_first_frozen_action(const struct xlate_ctx *ctx)
414 return !ctx->frozen_actions.size;
418 ctx_cancel_freeze(struct xlate_ctx *ctx)
421 ctx->freezing = false;
422 ofpbuf_clear(&ctx->frozen_actions);
423 ctx->frozen_actions.header = NULL;
427 static void finish_freezing(struct xlate_ctx *ctx);
429 /* A controller may use OFPP_NONE as the ingress port to indicate that
430 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
431 * when an input bundle is needed for validation (e.g., mirroring or
432 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
433 * any 'port' structs, so care must be taken when dealing with it. */
434 static struct xbundle ofpp_none_bundle = {
436 .vlan_mode = PORT_VLAN_TRUNK
439 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
440 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
441 * traffic egressing the 'ofport' with that priority should be marked with. */
442 struct skb_priority_to_dscp {
443 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
444 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
446 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
462 /* xlate_cache entries hold enough information to perform the side effects of
463 * xlate_actions() for a rule, without needing to perform rule translation
464 * from scratch. The primary usage of these is to submit statistics to objects
465 * that a flow relates to, although they may be used for other effects as well
466 * (for instance, refreshing hard timeouts for learned flows). */
470 struct rule_dpif *rule;
477 struct netflow *netflow;
482 struct mbridge *mbridge;
483 mirror_mask_t mirrors;
491 struct ofproto_dpif *ofproto;
492 struct ofputil_flow_mod *fm;
493 struct ofpbuf *ofpacts;
496 struct ofproto_dpif *ofproto;
501 struct rule_dpif *rule;
506 struct group_dpif *group;
507 struct ofputil_bucket *bucket;
510 char br_name[IFNAMSIZ];
511 struct in6_addr d_ipv6;
516 #define XC_ENTRY_FOR_EACH(ENTRY, ENTRIES, XCACHE) \
517 ENTRIES = XCACHE->entries; \
518 for (ENTRY = ofpbuf_try_pull(&ENTRIES, sizeof *ENTRY); \
520 ENTRY = ofpbuf_try_pull(&ENTRIES, sizeof *ENTRY))
523 struct ofpbuf entries;
526 /* Xlate config contains hash maps of all bridges, bundles and ports.
527 * Xcfgp contains the pointer to the current xlate configuration.
528 * When the main thread needs to change the configuration, it copies xcfgp to
529 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
530 * does not block handler and revalidator threads. */
532 struct hmap xbridges;
533 struct hmap xbundles;
536 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
537 static struct xlate_cfg *new_xcfg = NULL;
539 static bool may_receive(const struct xport *, struct xlate_ctx *);
540 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
542 static void xlate_normal(struct xlate_ctx *);
543 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
544 OVS_PRINTF_FORMAT(2, 3);
545 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
546 uint8_t table_id, bool may_packet_in,
547 bool honor_table_miss);
548 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
549 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
550 static void output_normal(struct xlate_ctx *, const struct xbundle *,
553 /* Optional bond recirculation parameter to compose_output_action(). */
554 struct xlate_bond_recirc {
555 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
556 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
557 uint32_t hash_basis; /* Compute hash for recirc before. */
560 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
561 const struct xlate_bond_recirc *xr);
563 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
564 const struct ofproto_dpif *);
565 static struct xbridge *xbridge_lookup_by_uuid(struct xlate_cfg *,
566 const struct uuid *);
567 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
568 const struct ofbundle *);
569 static struct xport *xport_lookup(struct xlate_cfg *,
570 const struct ofport_dpif *);
571 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
572 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
573 uint32_t skb_priority);
574 static void clear_skb_priorities(struct xport *);
575 static size_t count_skb_priorities(const struct xport *);
576 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
579 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
581 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
582 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
583 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
584 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
585 const struct mac_learning *, struct stp *,
586 struct rstp *, const struct mcast_snooping *,
587 const struct mbridge *,
588 const struct dpif_sflow *,
589 const struct dpif_ipfix *,
590 const struct netflow *,
591 bool forward_bpdu, bool has_in_band,
592 const struct dpif_backer_support *);
593 static void xlate_xbundle_set(struct xbundle *xbundle,
594 enum port_vlan_mode vlan_mode, int vlan,
595 unsigned long *trunks, bool use_priority_tags,
596 const struct bond *bond, const struct lacp *lacp,
598 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
599 const struct netdev *netdev, const struct cfm *cfm,
600 const struct bfd *bfd, const struct lldp *lldp,
601 int stp_port_no, const struct rstp_port *rstp_port,
602 enum ofputil_port_config config,
603 enum ofputil_port_state state, bool is_tunnel,
605 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
606 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
607 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
608 static void xlate_xbridge_copy(struct xbridge *);
609 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
610 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
612 static void xlate_xcfg_free(struct xlate_cfg *);
615 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
617 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
620 va_start(args, format);
621 ctx->xin->report_hook(ctx->xin, ctx->indentation, format, args);
626 static struct vlog_rate_limit error_report_rl = VLOG_RATE_LIMIT_INIT(1, 5);
628 #define XLATE_REPORT_ERROR(CTX, ...) \
630 if (OVS_UNLIKELY((CTX)->xin->report_hook)) { \
631 xlate_report(CTX, __VA_ARGS__); \
633 VLOG_ERR_RL(&error_report_rl, __VA_ARGS__); \
638 xlate_report_actions(struct xlate_ctx *ctx, const char *title,
639 const struct ofpact *ofpacts, size_t ofpacts_len)
641 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
642 struct ds s = DS_EMPTY_INITIALIZER;
643 ofpacts_format(ofpacts, ofpacts_len, &s);
644 xlate_report(ctx, "%s: %s", title, ds_cstr(&s));
650 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
652 ovs_list_init(&xbridge->xbundles);
653 hmap_init(&xbridge->xports);
654 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
655 hash_pointer(xbridge->ofproto, 0));
659 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
661 ovs_list_init(&xbundle->xports);
662 ovs_list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
663 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
664 hash_pointer(xbundle->ofbundle, 0));
668 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
670 hmap_init(&xport->skb_priorities);
671 hmap_insert(&xcfg->xports, &xport->hmap_node,
672 hash_pointer(xport->ofport, 0));
673 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
674 hash_ofp_port(xport->ofp_port));
678 xlate_xbridge_set(struct xbridge *xbridge,
680 const struct mac_learning *ml, struct stp *stp,
681 struct rstp *rstp, const struct mcast_snooping *ms,
682 const struct mbridge *mbridge,
683 const struct dpif_sflow *sflow,
684 const struct dpif_ipfix *ipfix,
685 const struct netflow *netflow,
686 bool forward_bpdu, bool has_in_band,
687 const struct dpif_backer_support *support)
689 if (xbridge->ml != ml) {
690 mac_learning_unref(xbridge->ml);
691 xbridge->ml = mac_learning_ref(ml);
694 if (xbridge->ms != ms) {
695 mcast_snooping_unref(xbridge->ms);
696 xbridge->ms = mcast_snooping_ref(ms);
699 if (xbridge->mbridge != mbridge) {
700 mbridge_unref(xbridge->mbridge);
701 xbridge->mbridge = mbridge_ref(mbridge);
704 if (xbridge->sflow != sflow) {
705 dpif_sflow_unref(xbridge->sflow);
706 xbridge->sflow = dpif_sflow_ref(sflow);
709 if (xbridge->ipfix != ipfix) {
710 dpif_ipfix_unref(xbridge->ipfix);
711 xbridge->ipfix = dpif_ipfix_ref(ipfix);
714 if (xbridge->stp != stp) {
715 stp_unref(xbridge->stp);
716 xbridge->stp = stp_ref(stp);
719 if (xbridge->rstp != rstp) {
720 rstp_unref(xbridge->rstp);
721 xbridge->rstp = rstp_ref(rstp);
724 if (xbridge->netflow != netflow) {
725 netflow_unref(xbridge->netflow);
726 xbridge->netflow = netflow_ref(netflow);
729 xbridge->dpif = dpif;
730 xbridge->forward_bpdu = forward_bpdu;
731 xbridge->has_in_band = has_in_band;
732 xbridge->support = *support;
736 xlate_xbundle_set(struct xbundle *xbundle,
737 enum port_vlan_mode vlan_mode, int vlan,
738 unsigned long *trunks, bool use_priority_tags,
739 const struct bond *bond, const struct lacp *lacp,
742 ovs_assert(xbundle->xbridge);
744 xbundle->vlan_mode = vlan_mode;
745 xbundle->vlan = vlan;
746 xbundle->trunks = trunks;
747 xbundle->use_priority_tags = use_priority_tags;
748 xbundle->floodable = floodable;
750 if (xbundle->bond != bond) {
751 bond_unref(xbundle->bond);
752 xbundle->bond = bond_ref(bond);
755 if (xbundle->lacp != lacp) {
756 lacp_unref(xbundle->lacp);
757 xbundle->lacp = lacp_ref(lacp);
762 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
763 const struct netdev *netdev, const struct cfm *cfm,
764 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
765 const struct rstp_port* rstp_port,
766 enum ofputil_port_config config, enum ofputil_port_state state,
767 bool is_tunnel, bool may_enable)
769 xport->config = config;
770 xport->state = state;
771 xport->stp_port_no = stp_port_no;
772 xport->is_tunnel = is_tunnel;
773 xport->may_enable = may_enable;
774 xport->odp_port = odp_port;
776 if (xport->rstp_port != rstp_port) {
777 rstp_port_unref(xport->rstp_port);
778 xport->rstp_port = rstp_port_ref(rstp_port);
781 if (xport->cfm != cfm) {
782 cfm_unref(xport->cfm);
783 xport->cfm = cfm_ref(cfm);
786 if (xport->bfd != bfd) {
787 bfd_unref(xport->bfd);
788 xport->bfd = bfd_ref(bfd);
791 if (xport->lldp != lldp) {
792 lldp_unref(xport->lldp);
793 xport->lldp = lldp_ref(lldp);
796 if (xport->netdev != netdev) {
797 netdev_close(xport->netdev);
798 xport->netdev = netdev_ref(netdev);
803 xlate_xbridge_copy(struct xbridge *xbridge)
805 struct xbundle *xbundle;
807 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
808 new_xbridge->ofproto = xbridge->ofproto;
809 new_xbridge->name = xstrdup(xbridge->name);
810 xlate_xbridge_init(new_xcfg, new_xbridge);
812 xlate_xbridge_set(new_xbridge,
813 xbridge->dpif, xbridge->ml, xbridge->stp,
814 xbridge->rstp, xbridge->ms, xbridge->mbridge,
815 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
816 xbridge->forward_bpdu, xbridge->has_in_band,
818 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
819 xlate_xbundle_copy(new_xbridge, xbundle);
822 /* Copy xports which are not part of a xbundle */
823 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
824 if (!xport->xbundle) {
825 xlate_xport_copy(new_xbridge, NULL, xport);
831 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
834 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
835 new_xbundle->ofbundle = xbundle->ofbundle;
836 new_xbundle->xbridge = xbridge;
837 new_xbundle->name = xstrdup(xbundle->name);
838 xlate_xbundle_init(new_xcfg, new_xbundle);
840 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
841 xbundle->vlan, xbundle->trunks,
842 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
844 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
845 xlate_xport_copy(xbridge, new_xbundle, xport);
850 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
853 struct skb_priority_to_dscp *pdscp, *new_pdscp;
854 struct xport *new_xport = xzalloc(sizeof *xport);
855 new_xport->ofport = xport->ofport;
856 new_xport->ofp_port = xport->ofp_port;
857 new_xport->xbridge = xbridge;
858 xlate_xport_init(new_xcfg, new_xport);
860 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
861 xport->bfd, xport->lldp, xport->stp_port_no,
862 xport->rstp_port, xport->config, xport->state,
863 xport->is_tunnel, xport->may_enable);
866 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
868 new_xport->peer = peer;
869 new_xport->peer->peer = new_xport;
874 new_xport->xbundle = xbundle;
875 ovs_list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
878 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
879 new_pdscp = xmalloc(sizeof *pdscp);
880 new_pdscp->skb_priority = pdscp->skb_priority;
881 new_pdscp->dscp = pdscp->dscp;
882 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
883 hash_int(new_pdscp->skb_priority, 0));
887 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
888 * configuration in xcfgp.
890 * This needs to be called after editing the xlate configuration.
892 * Functions that edit the new xlate configuration are
893 * xlate_<ofproto/bundle/ofport>_set and xlate_<ofproto/bundle/ofport>_remove.
899 * edit_xlate_configuration();
901 * xlate_txn_commit(); */
903 xlate_txn_commit(void)
905 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
907 ovsrcu_set(&xcfgp, new_xcfg);
908 ovsrcu_synchronize();
909 xlate_xcfg_free(xcfg);
913 /* Copies the current xlate configuration in xcfgp to new_xcfg.
915 * This needs to be called prior to editing the xlate configuration. */
917 xlate_txn_start(void)
919 struct xbridge *xbridge;
920 struct xlate_cfg *xcfg;
922 ovs_assert(!new_xcfg);
924 new_xcfg = xmalloc(sizeof *new_xcfg);
925 hmap_init(&new_xcfg->xbridges);
926 hmap_init(&new_xcfg->xbundles);
927 hmap_init(&new_xcfg->xports);
929 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
934 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
935 xlate_xbridge_copy(xbridge);
941 xlate_xcfg_free(struct xlate_cfg *xcfg)
943 struct xbridge *xbridge, *next_xbridge;
949 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
950 xlate_xbridge_remove(xcfg, xbridge);
953 hmap_destroy(&xcfg->xbridges);
954 hmap_destroy(&xcfg->xbundles);
955 hmap_destroy(&xcfg->xports);
960 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
962 const struct mac_learning *ml, struct stp *stp,
963 struct rstp *rstp, const struct mcast_snooping *ms,
964 const struct mbridge *mbridge,
965 const struct dpif_sflow *sflow,
966 const struct dpif_ipfix *ipfix,
967 const struct netflow *netflow,
968 bool forward_bpdu, bool has_in_band,
969 const struct dpif_backer_support *support)
971 struct xbridge *xbridge;
973 ovs_assert(new_xcfg);
975 xbridge = xbridge_lookup(new_xcfg, ofproto);
977 xbridge = xzalloc(sizeof *xbridge);
978 xbridge->ofproto = ofproto;
980 xlate_xbridge_init(new_xcfg, xbridge);
984 xbridge->name = xstrdup(name);
986 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
987 netflow, forward_bpdu, has_in_band, support);
991 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
993 struct xbundle *xbundle, *next_xbundle;
994 struct xport *xport, *next_xport;
1000 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
1001 xlate_xport_remove(xcfg, xport);
1004 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
1005 xlate_xbundle_remove(xcfg, xbundle);
1008 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
1009 mac_learning_unref(xbridge->ml);
1010 mcast_snooping_unref(xbridge->ms);
1011 mbridge_unref(xbridge->mbridge);
1012 dpif_sflow_unref(xbridge->sflow);
1013 dpif_ipfix_unref(xbridge->ipfix);
1014 stp_unref(xbridge->stp);
1015 rstp_unref(xbridge->rstp);
1016 hmap_destroy(&xbridge->xports);
1017 free(xbridge->name);
1022 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
1024 struct xbridge *xbridge;
1026 ovs_assert(new_xcfg);
1028 xbridge = xbridge_lookup(new_xcfg, ofproto);
1029 xlate_xbridge_remove(new_xcfg, xbridge);
1033 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
1034 const char *name, enum port_vlan_mode vlan_mode, int vlan,
1035 unsigned long *trunks, bool use_priority_tags,
1036 const struct bond *bond, const struct lacp *lacp,
1039 struct xbundle *xbundle;
1041 ovs_assert(new_xcfg);
1043 xbundle = xbundle_lookup(new_xcfg, ofbundle);
1045 xbundle = xzalloc(sizeof *xbundle);
1046 xbundle->ofbundle = ofbundle;
1047 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
1049 xlate_xbundle_init(new_xcfg, xbundle);
1052 free(xbundle->name);
1053 xbundle->name = xstrdup(name);
1055 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
1056 use_priority_tags, bond, lacp, floodable);
1060 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
1062 struct xport *xport;
1068 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
1069 xport->xbundle = NULL;
1072 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
1073 ovs_list_remove(&xbundle->list_node);
1074 bond_unref(xbundle->bond);
1075 lacp_unref(xbundle->lacp);
1076 free(xbundle->name);
1081 xlate_bundle_remove(struct ofbundle *ofbundle)
1083 struct xbundle *xbundle;
1085 ovs_assert(new_xcfg);
1087 xbundle = xbundle_lookup(new_xcfg, ofbundle);
1088 xlate_xbundle_remove(new_xcfg, xbundle);
1092 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
1093 struct ofport_dpif *ofport, ofp_port_t ofp_port,
1094 odp_port_t odp_port, const struct netdev *netdev,
1095 const struct cfm *cfm, const struct bfd *bfd,
1096 const struct lldp *lldp, struct ofport_dpif *peer,
1097 int stp_port_no, const struct rstp_port *rstp_port,
1098 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
1099 enum ofputil_port_config config,
1100 enum ofputil_port_state state, bool is_tunnel,
1104 struct xport *xport;
1106 ovs_assert(new_xcfg);
1108 xport = xport_lookup(new_xcfg, ofport);
1110 xport = xzalloc(sizeof *xport);
1111 xport->ofport = ofport;
1112 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
1113 xport->ofp_port = ofp_port;
1115 xlate_xport_init(new_xcfg, xport);
1118 ovs_assert(xport->ofp_port == ofp_port);
1120 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1121 stp_port_no, rstp_port, config, state, is_tunnel,
1125 xport->peer->peer = NULL;
1127 xport->peer = xport_lookup(new_xcfg, peer);
1129 xport->peer->peer = xport;
1132 if (xport->xbundle) {
1133 ovs_list_remove(&xport->bundle_node);
1135 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1136 if (xport->xbundle) {
1137 ovs_list_insert(&xport->xbundle->xports, &xport->bundle_node);
1140 clear_skb_priorities(xport);
1141 for (i = 0; i < n_qdscp; i++) {
1142 struct skb_priority_to_dscp *pdscp;
1143 uint32_t skb_priority;
1145 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1150 pdscp = xmalloc(sizeof *pdscp);
1151 pdscp->skb_priority = skb_priority;
1152 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1153 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1154 hash_int(pdscp->skb_priority, 0));
1159 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1166 xport->peer->peer = NULL;
1170 if (xport->xbundle) {
1171 ovs_list_remove(&xport->bundle_node);
1174 clear_skb_priorities(xport);
1175 hmap_destroy(&xport->skb_priorities);
1177 hmap_remove(&xcfg->xports, &xport->hmap_node);
1178 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1180 netdev_close(xport->netdev);
1181 rstp_port_unref(xport->rstp_port);
1182 cfm_unref(xport->cfm);
1183 bfd_unref(xport->bfd);
1184 lldp_unref(xport->lldp);
1189 xlate_ofport_remove(struct ofport_dpif *ofport)
1191 struct xport *xport;
1193 ovs_assert(new_xcfg);
1195 xport = xport_lookup(new_xcfg, ofport);
1196 xlate_xport_remove(new_xcfg, xport);
1199 static struct ofproto_dpif *
1200 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1201 ofp_port_t *ofp_in_port, const struct xport **xportp)
1203 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1204 const struct xport *xport;
1206 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1207 ? tnl_port_receive(flow)
1208 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1209 if (OVS_UNLIKELY(!xport)) {
1214 *ofp_in_port = xport->ofp_port;
1216 return xport->xbridge->ofproto;
1219 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1220 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1221 struct ofproto_dpif *
1222 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1223 ofp_port_t *ofp_in_port)
1225 const struct xport *xport;
1227 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1230 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1231 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1232 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1233 * handles for those protocols if they're enabled. Caller may use the returned
1234 * pointers until quiescing, for longer term use additional references must
1237 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1240 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1241 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1242 struct dpif_sflow **sflow, struct netflow **netflow,
1243 ofp_port_t *ofp_in_port)
1245 struct ofproto_dpif *ofproto;
1246 const struct xport *xport;
1248 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1255 *ofprotop = ofproto;
1259 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1263 *sflow = xport ? xport->xbridge->sflow : NULL;
1267 *netflow = xport ? xport->xbridge->netflow : NULL;
1273 static struct xbridge *
1274 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1276 struct hmap *xbridges;
1277 struct xbridge *xbridge;
1279 if (!ofproto || !xcfg) {
1283 xbridges = &xcfg->xbridges;
1285 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1287 if (xbridge->ofproto == ofproto) {
1294 static struct xbridge *
1295 xbridge_lookup_by_uuid(struct xlate_cfg *xcfg, const struct uuid *uuid)
1297 struct xbridge *xbridge;
1299 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
1300 if (uuid_equals(ofproto_dpif_get_uuid(xbridge->ofproto), uuid)) {
1307 static struct xbundle *
1308 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1310 struct hmap *xbundles;
1311 struct xbundle *xbundle;
1313 if (!ofbundle || !xcfg) {
1317 xbundles = &xcfg->xbundles;
1319 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1321 if (xbundle->ofbundle == ofbundle) {
1328 static struct xport *
1329 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1331 struct hmap *xports;
1332 struct xport *xport;
1334 if (!ofport || !xcfg) {
1338 xports = &xcfg->xports;
1340 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1342 if (xport->ofport == ofport) {
1349 static struct stp_port *
1350 xport_get_stp_port(const struct xport *xport)
1352 return xport->xbridge->stp && xport->stp_port_no != -1
1353 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1358 xport_stp_learn_state(const struct xport *xport)
1360 struct stp_port *sp = xport_get_stp_port(xport);
1362 ? stp_learn_in_state(stp_port_get_state(sp))
1367 xport_stp_forward_state(const struct xport *xport)
1369 struct stp_port *sp = xport_get_stp_port(xport);
1371 ? stp_forward_in_state(stp_port_get_state(sp))
1376 xport_stp_should_forward_bpdu(const struct xport *xport)
1378 struct stp_port *sp = xport_get_stp_port(xport);
1379 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1382 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1383 * were used to make the determination.*/
1385 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1387 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1388 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1389 return is_stp(flow);
1393 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1395 struct stp_port *sp = xport_get_stp_port(xport);
1396 struct dp_packet payload = *packet;
1397 struct eth_header *eth = dp_packet_data(&payload);
1399 /* Sink packets on ports that have STP disabled when the bridge has
1401 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1405 /* Trim off padding on payload. */
1406 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1407 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1410 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1411 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1415 static enum rstp_state
1416 xport_get_rstp_port_state(const struct xport *xport)
1418 return xport->rstp_port
1419 ? rstp_port_get_state(xport->rstp_port)
1424 xport_rstp_learn_state(const struct xport *xport)
1426 return xport->xbridge->rstp && xport->rstp_port
1427 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1432 xport_rstp_forward_state(const struct xport *xport)
1434 return xport->xbridge->rstp && xport->rstp_port
1435 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1440 xport_rstp_should_manage_bpdu(const struct xport *xport)
1442 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1446 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1448 struct dp_packet payload = *packet;
1449 struct eth_header *eth = dp_packet_data(&payload);
1451 /* Sink packets on ports that have no RSTP. */
1452 if (!xport->rstp_port) {
1456 /* Trim off padding on payload. */
1457 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1458 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1461 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1462 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1463 dp_packet_size(&payload));
1467 static struct xport *
1468 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1470 struct xport *xport;
1472 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1474 if (xport->ofp_port == ofp_port) {
1482 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1484 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1485 return xport ? xport->odp_port : ODPP_NONE;
1489 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1491 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1492 return xport && xport->may_enable;
1495 static struct ofputil_bucket *
1496 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1500 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1502 struct group_dpif *group;
1504 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1505 struct ofputil_bucket *bucket;
1507 bucket = group_first_live_bucket(ctx, group, depth);
1508 group_dpif_unref(group);
1509 return bucket != NULL;
1515 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1518 bucket_is_alive(const struct xlate_ctx *ctx,
1519 struct ofputil_bucket *bucket, int depth)
1521 if (depth >= MAX_LIVENESS_RECURSION) {
1522 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1524 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1525 MAX_LIVENESS_RECURSION);
1529 return (!ofputil_bucket_has_liveness(bucket)
1530 || (bucket->watch_port != OFPP_ANY
1531 && odp_port_is_alive(ctx, bucket->watch_port))
1532 || (bucket->watch_group != OFPG_ANY
1533 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1536 static struct ofputil_bucket *
1537 group_first_live_bucket(const struct xlate_ctx *ctx,
1538 const struct group_dpif *group, int depth)
1540 struct ofputil_bucket *bucket;
1541 const struct ovs_list *buckets;
1543 group_dpif_get_buckets(group, &buckets);
1544 LIST_FOR_EACH (bucket, list_node, buckets) {
1545 if (bucket_is_alive(ctx, bucket, depth)) {
1553 static struct ofputil_bucket *
1554 group_best_live_bucket(const struct xlate_ctx *ctx,
1555 const struct group_dpif *group,
1558 struct ofputil_bucket *best_bucket = NULL;
1559 uint32_t best_score = 0;
1562 struct ofputil_bucket *bucket;
1563 const struct ovs_list *buckets;
1565 group_dpif_get_buckets(group, &buckets);
1566 LIST_FOR_EACH (bucket, list_node, buckets) {
1567 if (bucket_is_alive(ctx, bucket, 0)) {
1568 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1569 if (score >= best_score) {
1570 best_bucket = bucket;
1581 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1583 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1584 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1588 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1590 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1593 static mirror_mask_t
1594 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1596 return xbundle != &ofpp_none_bundle
1597 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1601 static mirror_mask_t
1602 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1604 return xbundle != &ofpp_none_bundle
1605 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1609 static mirror_mask_t
1610 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1612 return xbundle != &ofpp_none_bundle
1613 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1617 static struct xbundle *
1618 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1619 bool warn, struct xport **in_xportp)
1621 struct xport *xport;
1623 /* Find the port and bundle for the received packet. */
1624 xport = get_ofp_port(xbridge, in_port);
1628 if (xport && xport->xbundle) {
1629 return xport->xbundle;
1632 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1633 * which a controller may use as the ingress port for traffic that
1634 * it is sourcing. */
1635 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1636 return &ofpp_none_bundle;
1639 /* Odd. A few possible reasons here:
1641 * - We deleted a port but there are still a few packets queued up
1644 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1645 * we don't know about.
1647 * - The ofproto client didn't configure the port as part of a bundle.
1648 * This is particularly likely to happen if a packet was received on the
1649 * port after it was created, but before the client had a chance to
1650 * configure its bundle.
1653 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1655 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1656 "port %"PRIu16, xbridge->name, in_port);
1661 /* Mirrors the packet represented by 'ctx' to appropriate mirror destinations,
1662 * given the packet is ingressing or egressing on 'xbundle', which has ingress
1663 * or egress (as appropriate) mirrors 'mirrors'. */
1665 mirror_packet(struct xlate_ctx *ctx, struct xbundle *xbundle,
1666 mirror_mask_t mirrors)
1668 /* Figure out what VLAN the packet is in (because mirrors can select
1669 * packets on basis of VLAN). */
1670 bool warn = ctx->xin->packet != NULL;
1671 uint16_t vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
1672 if (!input_vid_is_valid(vid, xbundle, warn)) {
1675 uint16_t vlan = input_vid_to_vlan(xbundle, vid);
1677 const struct xbridge *xbridge = ctx->xbridge;
1679 /* Don't mirror to destinations that we've already mirrored to. */
1680 mirrors &= ~ctx->mirrors;
1685 if (ctx->xin->resubmit_stats) {
1686 mirror_update_stats(xbridge->mbridge, mirrors,
1687 ctx->xin->resubmit_stats->n_packets,
1688 ctx->xin->resubmit_stats->n_bytes);
1690 if (ctx->xin->xcache) {
1691 struct xc_entry *entry;
1693 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_MIRROR);
1694 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
1695 entry->u.mirror.mirrors = mirrors;
1698 /* 'mirrors' is a bit-mask of candidates for mirroring. Iterate as long as
1699 * some candidates remain. */
1701 const unsigned long *vlans;
1702 mirror_mask_t dup_mirrors;
1703 struct ofbundle *out;
1706 /* Get the details of the mirror represented by the rightmost 1-bit. */
1707 bool has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1708 &vlans, &dup_mirrors, &out, &out_vlan);
1709 ovs_assert(has_mirror);
1711 /* If this mirror selects on the basis of VLAN, and it does not select
1712 * 'vlan', then discard this mirror and go on to the next one. */
1714 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1716 if (vlans && !bitmap_is_set(vlans, vlan)) {
1717 mirrors = zero_rightmost_1bit(mirrors);
1721 /* Record the mirror, and the mirrors that output to the same
1722 * destination, so that we don't mirror to them again. This must be
1723 * done now to ensure that output_normal(), below, doesn't recursively
1724 * output to the same mirrors. */
1725 ctx->mirrors |= dup_mirrors;
1727 /* Send the packet to the mirror. */
1729 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1730 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1732 output_normal(ctx, out_xbundle, vlan);
1734 } else if (vlan != out_vlan
1735 && !eth_addr_is_reserved(ctx->xin->flow.dl_dst)) {
1736 struct xbundle *xbundle;
1738 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1739 if (xbundle_includes_vlan(xbundle, out_vlan)
1740 && !xbundle_mirror_out(xbridge, xbundle)) {
1741 output_normal(ctx, xbundle, out_vlan);
1746 /* output_normal() could have recursively output (to different
1747 * mirrors), so make sure that we don't send duplicates. */
1748 mirrors &= ~ctx->mirrors;
1753 mirror_ingress_packet(struct xlate_ctx *ctx)
1755 if (mbridge_has_mirrors(ctx->xbridge->mbridge)) {
1756 bool warn = ctx->xin->packet != NULL;
1757 struct xbundle *xbundle = lookup_input_bundle(
1758 ctx->xbridge, ctx->xin->flow.in_port.ofp_port, warn, NULL);
1760 mirror_packet(ctx, xbundle,
1761 xbundle_mirror_src(ctx->xbridge, xbundle));
1766 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1767 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1768 * the bundle on which the packet was received, returns the VLAN to which the
1771 * Both 'vid' and the return value are in the range 0...4095. */
1773 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1775 switch (in_xbundle->vlan_mode) {
1776 case PORT_VLAN_ACCESS:
1777 return in_xbundle->vlan;
1780 case PORT_VLAN_TRUNK:
1783 case PORT_VLAN_NATIVE_UNTAGGED:
1784 case PORT_VLAN_NATIVE_TAGGED:
1785 return vid ? vid : in_xbundle->vlan;
1792 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1793 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1796 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1797 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1800 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1802 /* Allow any VID on the OFPP_NONE port. */
1803 if (in_xbundle == &ofpp_none_bundle) {
1807 switch (in_xbundle->vlan_mode) {
1808 case PORT_VLAN_ACCESS:
1811 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1812 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1813 "packet received on port %s configured as VLAN "
1814 "%"PRIu16" access port", vid, in_xbundle->name,
1821 case PORT_VLAN_NATIVE_UNTAGGED:
1822 case PORT_VLAN_NATIVE_TAGGED:
1824 /* Port must always carry its native VLAN. */
1828 case PORT_VLAN_TRUNK:
1829 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1831 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1832 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1833 "received on port %s not configured for trunking "
1834 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1846 /* Given 'vlan', the VLAN that a packet belongs to, and
1847 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1848 * that should be included in the 802.1Q header. (If the return value is 0,
1849 * then the 802.1Q header should only be included in the packet if there is a
1852 * Both 'vlan' and the return value are in the range 0...4095. */
1854 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1856 switch (out_xbundle->vlan_mode) {
1857 case PORT_VLAN_ACCESS:
1860 case PORT_VLAN_TRUNK:
1861 case PORT_VLAN_NATIVE_TAGGED:
1864 case PORT_VLAN_NATIVE_UNTAGGED:
1865 return vlan == out_xbundle->vlan ? 0 : vlan;
1873 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1876 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1878 ovs_be16 tci, old_tci;
1879 struct xport *xport;
1880 struct xlate_bond_recirc xr;
1881 bool use_recirc = false;
1883 vid = output_vlan_to_vid(out_xbundle, vlan);
1884 if (ovs_list_is_empty(&out_xbundle->xports)) {
1885 /* Partially configured bundle with no slaves. Drop the packet. */
1887 } else if (!out_xbundle->bond) {
1888 xport = CONTAINER_OF(ovs_list_front(&out_xbundle->xports), struct xport,
1891 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1892 struct flow_wildcards *wc = ctx->wc;
1893 struct ofport_dpif *ofport;
1895 if (ctx->xbridge->support.odp.recirc) {
1896 use_recirc = bond_may_recirc(
1897 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1900 /* Only TCP mode uses recirculation. */
1901 xr.hash_alg = OVS_HASH_ALG_L4;
1902 bond_update_post_recirc_rules(out_xbundle->bond, false);
1904 /* Recirculation does not require unmasking hash fields. */
1909 ofport = bond_choose_output_slave(out_xbundle->bond,
1910 &ctx->xin->flow, wc, vid);
1911 xport = xport_lookup(xcfg, ofport);
1914 /* No slaves enabled, so drop packet. */
1918 /* If use_recirc is set, the main thread will handle stats
1919 * accounting for this bond. */
1921 if (ctx->xin->resubmit_stats) {
1922 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1923 ctx->xin->resubmit_stats->n_bytes);
1925 if (ctx->xin->xcache) {
1926 struct xc_entry *entry;
1929 flow = &ctx->xin->flow;
1930 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1931 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1932 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1933 entry->u.bond.vid = vid;
1938 old_tci = *flow_tci;
1940 if (tci || out_xbundle->use_priority_tags) {
1941 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1943 tci |= htons(VLAN_CFI);
1948 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1949 *flow_tci = old_tci;
1952 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1953 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1954 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1956 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1958 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1962 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1963 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1967 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1968 if (flow->nw_proto == ARP_OP_REPLY) {
1970 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1971 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1972 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1974 return flow->nw_src == flow->nw_dst;
1980 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1981 * dropped. Returns true if they may be forwarded, false if they should be
1984 * 'in_port' must be the xport that corresponds to flow->in_port.
1985 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1987 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1988 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1989 * checked by input_vid_is_valid().
1991 * May also add tags to '*tags', although the current implementation only does
1992 * so in one special case.
1995 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1998 struct xbundle *in_xbundle = in_port->xbundle;
1999 const struct xbridge *xbridge = ctx->xbridge;
2000 struct flow *flow = &ctx->xin->flow;
2002 /* Drop frames for reserved multicast addresses
2003 * only if forward_bpdu option is absent. */
2004 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
2005 xlate_report(ctx, "packet has reserved destination MAC, dropping");
2009 if (in_xbundle->bond) {
2010 struct mac_entry *mac;
2012 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
2018 xlate_report(ctx, "bonding refused admissibility, dropping");
2021 case BV_DROP_IF_MOVED:
2022 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
2023 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
2025 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
2026 && (!is_gratuitous_arp(flow, ctx->wc)
2027 || mac_entry_is_grat_arp_locked(mac))) {
2028 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2029 xlate_report(ctx, "SLB bond thinks this packet looped back, "
2033 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2041 /* Checks whether a MAC learning update is necessary for MAC learning table
2042 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
2045 * Most packets processed through the MAC learning table do not actually
2046 * change it in any way. This function requires only a read lock on the MAC
2047 * learning table, so it is much cheaper in this common case.
2049 * Keep the code here synchronized with that in update_learning_table__()
2052 is_mac_learning_update_needed(const struct mac_learning *ml,
2053 const struct flow *flow,
2054 struct flow_wildcards *wc,
2055 int vlan, struct xbundle *in_xbundle)
2056 OVS_REQ_RDLOCK(ml->rwlock)
2058 struct mac_entry *mac;
2060 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
2064 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
2065 if (!mac || mac_entry_age(ml, mac)) {
2069 if (is_gratuitous_arp(flow, wc)) {
2070 /* We don't want to learn from gratuitous ARP packets that are
2071 * reflected back over bond slaves so we lock the learning table. */
2072 if (!in_xbundle->bond) {
2074 } else if (mac_entry_is_grat_arp_locked(mac)) {
2079 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
2083 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
2084 * received on 'in_xbundle' in 'vlan'.
2086 * This code repeats all the checks in is_mac_learning_update_needed() because
2087 * the lock was released between there and here and thus the MAC learning state
2088 * could have changed.
2090 * Keep the code here synchronized with that in is_mac_learning_update_needed()
2093 update_learning_table__(const struct xbridge *xbridge,
2094 const struct flow *flow, struct flow_wildcards *wc,
2095 int vlan, struct xbundle *in_xbundle)
2096 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
2098 struct mac_entry *mac;
2100 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
2104 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
2105 if (is_gratuitous_arp(flow, wc)) {
2106 /* We don't want to learn from gratuitous ARP packets that are
2107 * reflected back over bond slaves so we lock the learning table. */
2108 if (!in_xbundle->bond) {
2109 mac_entry_set_grat_arp_lock(mac);
2110 } else if (mac_entry_is_grat_arp_locked(mac)) {
2115 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
2116 /* The log messages here could actually be useful in debugging,
2117 * so keep the rate limit relatively high. */
2118 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2120 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2121 "on port %s in VLAN %d",
2122 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
2123 in_xbundle->name, vlan);
2125 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
2130 update_learning_table(const struct xbridge *xbridge,
2131 const struct flow *flow, struct flow_wildcards *wc,
2132 int vlan, struct xbundle *in_xbundle)
2136 /* Don't learn the OFPP_NONE port. */
2137 if (in_xbundle == &ofpp_none_bundle) {
2141 /* First try the common case: no change to MAC learning table. */
2142 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
2143 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
2145 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2148 /* Slow path: MAC learning table might need an update. */
2149 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2150 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2151 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2155 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2156 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2158 update_mcast_snooping_table4__(const struct xbridge *xbridge,
2159 const struct flow *flow,
2160 struct mcast_snooping *ms, int vlan,
2161 struct xbundle *in_xbundle,
2162 const struct dp_packet *packet)
2163 OVS_REQ_WRLOCK(ms->rwlock)
2165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2167 ovs_be32 ip4 = flow->igmp_group_ip4;
2169 switch (ntohs(flow->tp_src)) {
2170 case IGMP_HOST_MEMBERSHIP_REPORT:
2171 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2172 if (mcast_snooping_add_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2173 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2174 IP_FMT" is on port %s in VLAN %d",
2175 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2178 case IGMP_HOST_LEAVE_MESSAGE:
2179 if (mcast_snooping_leave_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2180 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2181 IP_FMT" is on port %s in VLAN %d",
2182 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2185 case IGMP_HOST_MEMBERSHIP_QUERY:
2186 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2187 in_xbundle->ofbundle)) {
2188 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2189 IP_FMT" is on port %s in VLAN %d",
2190 xbridge->name, IP_ARGS(flow->nw_src),
2191 in_xbundle->name, vlan);
2194 case IGMPV3_HOST_MEMBERSHIP_REPORT:
2195 if ((count = mcast_snooping_add_report(ms, packet, vlan,
2196 in_xbundle->ofbundle))) {
2197 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2198 "addresses on port %s in VLAN %d",
2199 xbridge->name, count, in_xbundle->name, vlan);
2206 update_mcast_snooping_table6__(const struct xbridge *xbridge,
2207 const struct flow *flow,
2208 struct mcast_snooping *ms, int vlan,
2209 struct xbundle *in_xbundle,
2210 const struct dp_packet *packet)
2211 OVS_REQ_WRLOCK(ms->rwlock)
2213 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2216 switch (ntohs(flow->tp_src)) {
2218 if (!ipv6_addr_equals(&flow->ipv6_src, &in6addr_any)
2219 && mcast_snooping_add_mrouter(ms, vlan, in_xbundle->ofbundle)) {
2220 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query on port %s"
2222 xbridge->name, in_xbundle->name, vlan);
2228 count = mcast_snooping_add_mld(ms, packet, vlan, in_xbundle->ofbundle);
2230 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2231 "addresses on port %s in VLAN %d",
2232 xbridge->name, count, in_xbundle->name, vlan);
2238 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2239 * was received on 'in_xbundle' in 'vlan'. */
2241 update_mcast_snooping_table(const struct xbridge *xbridge,
2242 const struct flow *flow, int vlan,
2243 struct xbundle *in_xbundle,
2244 const struct dp_packet *packet)
2246 struct mcast_snooping *ms = xbridge->ms;
2247 struct xlate_cfg *xcfg;
2248 struct xbundle *mcast_xbundle;
2249 struct mcast_port_bundle *fport;
2251 /* Don't learn the OFPP_NONE port. */
2252 if (in_xbundle == &ofpp_none_bundle) {
2256 /* Don't learn from flood ports */
2257 mcast_xbundle = NULL;
2258 ovs_rwlock_wrlock(&ms->rwlock);
2259 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2260 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2261 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2262 if (mcast_xbundle == in_xbundle) {
2267 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2268 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2269 update_mcast_snooping_table4__(xbridge, flow, ms, vlan,
2270 in_xbundle, packet);
2272 update_mcast_snooping_table6__(xbridge, flow, ms, vlan,
2273 in_xbundle, packet);
2276 ovs_rwlock_unlock(&ms->rwlock);
2279 /* send the packet to ports having the multicast group learned */
2281 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2282 struct mcast_snooping *ms OVS_UNUSED,
2283 struct mcast_group *grp,
2284 struct xbundle *in_xbundle, uint16_t vlan)
2285 OVS_REQ_RDLOCK(ms->rwlock)
2287 struct xlate_cfg *xcfg;
2288 struct mcast_group_bundle *b;
2289 struct xbundle *mcast_xbundle;
2291 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2292 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2293 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2294 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2295 xlate_report(ctx, "forwarding to mcast group port");
2296 output_normal(ctx, mcast_xbundle, vlan);
2297 } else if (!mcast_xbundle) {
2298 xlate_report(ctx, "mcast group port is unknown, dropping");
2300 xlate_report(ctx, "mcast group port is input port, dropping");
2305 /* send the packet to ports connected to multicast routers */
2307 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2308 struct mcast_snooping *ms,
2309 struct xbundle *in_xbundle, uint16_t vlan)
2310 OVS_REQ_RDLOCK(ms->rwlock)
2312 struct xlate_cfg *xcfg;
2313 struct mcast_mrouter_bundle *mrouter;
2314 struct xbundle *mcast_xbundle;
2316 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2317 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2318 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2319 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2320 xlate_report(ctx, "forwarding to mcast router port");
2321 output_normal(ctx, mcast_xbundle, vlan);
2322 } else if (!mcast_xbundle) {
2323 xlate_report(ctx, "mcast router port is unknown, dropping");
2325 xlate_report(ctx, "mcast router port is input port, dropping");
2330 /* send the packet to ports flagged to be flooded */
2332 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2333 struct mcast_snooping *ms,
2334 struct xbundle *in_xbundle, uint16_t vlan)
2335 OVS_REQ_RDLOCK(ms->rwlock)
2337 struct xlate_cfg *xcfg;
2338 struct mcast_port_bundle *fport;
2339 struct xbundle *mcast_xbundle;
2341 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2342 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2343 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2344 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2345 xlate_report(ctx, "forwarding to mcast flood port");
2346 output_normal(ctx, mcast_xbundle, vlan);
2347 } else if (!mcast_xbundle) {
2348 xlate_report(ctx, "mcast flood port is unknown, dropping");
2350 xlate_report(ctx, "mcast flood port is input port, dropping");
2355 /* forward the Reports to configured ports */
2357 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2358 struct mcast_snooping *ms,
2359 struct xbundle *in_xbundle, uint16_t vlan)
2360 OVS_REQ_RDLOCK(ms->rwlock)
2362 struct xlate_cfg *xcfg;
2363 struct mcast_port_bundle *rport;
2364 struct xbundle *mcast_xbundle;
2366 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2367 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2368 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2369 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2370 xlate_report(ctx, "forwarding Report to mcast flagged port");
2371 output_normal(ctx, mcast_xbundle, vlan);
2372 } else if (!mcast_xbundle) {
2373 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2375 xlate_report(ctx, "mcast port is input port, dropping the Report");
2381 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2384 struct xbundle *xbundle;
2386 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2387 if (xbundle != in_xbundle
2388 && xbundle_includes_vlan(xbundle, vlan)
2389 && xbundle->floodable
2390 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2391 output_normal(ctx, xbundle, vlan);
2394 ctx->nf_output_iface = NF_OUT_FLOOD;
2398 is_ip_local_multicast(const struct flow *flow, struct flow_wildcards *wc)
2400 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2401 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2402 return ip_is_local_multicast(flow->nw_dst);
2403 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2404 memset(&wc->masks.ipv6_dst, 0xff, sizeof wc->masks.ipv6_dst);
2405 return ipv6_is_all_hosts(&flow->ipv6_dst);
2412 xlate_normal(struct xlate_ctx *ctx)
2414 struct flow_wildcards *wc = ctx->wc;
2415 struct flow *flow = &ctx->xin->flow;
2416 struct xbundle *in_xbundle;
2417 struct xport *in_port;
2418 struct mac_entry *mac;
2423 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2424 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2425 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2427 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2428 ctx->xin->packet != NULL, &in_port);
2430 xlate_report(ctx, "no input bundle, dropping");
2434 /* Drop malformed frames. */
2435 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2436 !(flow->vlan_tci & htons(VLAN_CFI))) {
2437 if (ctx->xin->packet != NULL) {
2438 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2439 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2440 "VLAN tag received on port %s",
2441 ctx->xbridge->name, in_xbundle->name);
2443 xlate_report(ctx, "partial VLAN tag, dropping");
2447 /* Drop frames on bundles reserved for mirroring. */
2448 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2449 if (ctx->xin->packet != NULL) {
2450 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2451 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2452 "%s, which is reserved exclusively for mirroring",
2453 ctx->xbridge->name, in_xbundle->name);
2455 xlate_report(ctx, "input port is mirror output port, dropping");
2460 vid = vlan_tci_to_vid(flow->vlan_tci);
2461 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2462 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2465 vlan = input_vid_to_vlan(in_xbundle, vid);
2467 /* Check other admissibility requirements. */
2468 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2472 /* Learn source MAC. */
2473 if (ctx->xin->may_learn) {
2474 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2476 if (ctx->xin->xcache) {
2477 struct xc_entry *entry;
2479 /* Save enough info to update mac learning table later. */
2480 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2481 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2482 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2483 entry->u.normal.vlan = vlan;
2486 /* Determine output bundle. */
2487 if (mcast_snooping_enabled(ctx->xbridge->ms)
2488 && !eth_addr_is_broadcast(flow->dl_dst)
2489 && eth_addr_is_multicast(flow->dl_dst)
2490 && is_ip_any(flow)) {
2491 struct mcast_snooping *ms = ctx->xbridge->ms;
2492 struct mcast_group *grp = NULL;
2494 if (is_igmp(flow, wc)) {
2495 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2496 if (mcast_snooping_is_membership(flow->tp_src) ||
2497 mcast_snooping_is_query(flow->tp_src)) {
2498 if (ctx->xin->may_learn && ctx->xin->packet) {
2499 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2500 in_xbundle, ctx->xin->packet);
2503 * IGMP packets need to take the slow path, in order to be
2504 * processed for mdb updates. That will prevent expires
2505 * firing off even after hosts have sent reports.
2507 ctx->xout->slow |= SLOW_ACTION;
2510 if (mcast_snooping_is_membership(flow->tp_src)) {
2511 ovs_rwlock_rdlock(&ms->rwlock);
2512 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2513 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2514 * forward IGMP Membership Reports only to those ports where
2515 * multicast routers are attached. Alternatively stated: a
2516 * snooping switch should not forward IGMP Membership Reports
2517 * to ports on which only hosts are attached.
2518 * An administrative control may be provided to override this
2519 * restriction, allowing the report messages to be flooded to
2521 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2522 ovs_rwlock_unlock(&ms->rwlock);
2524 xlate_report(ctx, "multicast traffic, flooding");
2525 xlate_normal_flood(ctx, in_xbundle, vlan);
2528 } else if (is_mld(flow, wc)) {
2529 ctx->xout->slow |= SLOW_ACTION;
2530 if (ctx->xin->may_learn && ctx->xin->packet) {
2531 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2532 in_xbundle, ctx->xin->packet);
2534 if (is_mld_report(flow, wc)) {
2535 ovs_rwlock_rdlock(&ms->rwlock);
2536 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2537 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2538 ovs_rwlock_unlock(&ms->rwlock);
2540 xlate_report(ctx, "MLD query, flooding");
2541 xlate_normal_flood(ctx, in_xbundle, vlan);
2544 if (is_ip_local_multicast(flow, wc)) {
2545 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2546 * address in the 224.0.0.x range which are not IGMP must
2547 * be forwarded on all ports */
2548 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2549 xlate_normal_flood(ctx, in_xbundle, vlan);
2554 /* forwarding to group base ports */
2555 ovs_rwlock_rdlock(&ms->rwlock);
2556 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2557 grp = mcast_snooping_lookup4(ms, flow->nw_dst, vlan);
2558 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2559 grp = mcast_snooping_lookup(ms, &flow->ipv6_dst, vlan);
2562 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2563 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2564 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2566 if (mcast_snooping_flood_unreg(ms)) {
2567 xlate_report(ctx, "unregistered multicast, flooding");
2568 xlate_normal_flood(ctx, in_xbundle, vlan);
2570 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2571 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2574 ovs_rwlock_unlock(&ms->rwlock);
2576 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2577 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2578 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2579 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2582 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2583 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2584 if (mac_xbundle && mac_xbundle != in_xbundle) {
2585 xlate_report(ctx, "forwarding to learned port");
2586 output_normal(ctx, mac_xbundle, vlan);
2587 } else if (!mac_xbundle) {
2588 xlate_report(ctx, "learned port is unknown, dropping");
2590 xlate_report(ctx, "learned port is input port, dropping");
2593 xlate_report(ctx, "no learned MAC for destination, flooding");
2594 xlate_normal_flood(ctx, in_xbundle, vlan);
2599 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
2600 * 'probability' is the number of packets out of UINT32_MAX to sample. The
2601 * 'cookie' (of length 'cookie_size' bytes) is passed back in the callback for
2602 * each sampled packet. 'tunnel_out_port', if not ODPP_NONE, is added as the
2603 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions', an
2604 * OVS_USERSPACE_ATTR_ACTIONS attribute is added. If 'emit_set_tunnel',
2605 * sample(sampling_port=1) would translate into datapath sample action
2606 * set(tunnel(...)), sample(...) and it is used for sampling egress tunnel
2610 compose_sample_action(struct xlate_ctx *ctx,
2611 const uint32_t probability,
2612 const union user_action_cookie *cookie,
2613 const size_t cookie_size,
2614 const odp_port_t tunnel_out_port,
2615 bool include_actions)
2617 size_t sample_offset = nl_msg_start_nested(ctx->odp_actions,
2618 OVS_ACTION_ATTR_SAMPLE);
2620 nl_msg_put_u32(ctx->odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2622 size_t actions_offset = nl_msg_start_nested(ctx->odp_actions,
2623 OVS_SAMPLE_ATTR_ACTIONS);
2625 odp_port_t odp_port = ofp_port_to_odp_port(
2626 ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
2627 uint32_t pid = dpif_port_get_pid(ctx->xbridge->dpif, odp_port,
2628 flow_hash_5tuple(&ctx->xin->flow, 0));
2629 int cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2634 nl_msg_end_nested(ctx->odp_actions, actions_offset);
2635 nl_msg_end_nested(ctx->odp_actions, sample_offset);
2637 return cookie_offset;
2640 /* If sFLow is not enabled, returns 0 without doing anything.
2642 * If sFlow is enabled, appends a template "sample" action to the ODP actions
2643 * in 'ctx'. This action is a template because some of the information needed
2644 * to fill it out is not available until flow translation is complete. In this
2645 * case, this functions returns an offset, which is always nonzero, to pass
2646 * later to fix_sflow_action() to fill in the rest of the template. */
2648 compose_sflow_action(struct xlate_ctx *ctx)
2650 struct dpif_sflow *sflow = ctx->xbridge->sflow;
2651 if (!sflow || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2655 union user_action_cookie cookie = { .type = USER_ACTION_COOKIE_SFLOW };
2656 return compose_sample_action(ctx, dpif_sflow_get_probability(sflow),
2657 &cookie, sizeof cookie.sflow, ODPP_NONE,
2661 /* If flow IPFIX is enabled, make sure IPFIX flow sample action
2662 * at egress point of tunnel port is just in front of corresponding
2663 * output action. If bridge IPFIX is enabled, this appends an IPFIX
2664 * sample action to 'ctx->odp_actions'. */
2666 compose_ipfix_action(struct xlate_ctx *ctx, odp_port_t output_odp_port)
2668 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
2669 odp_port_t tunnel_out_port = ODPP_NONE;
2671 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2675 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2677 if (output_odp_port == ODPP_NONE &&
2678 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix)) {
2682 /* For output case, output_odp_port is valid. */
2683 if (output_odp_port != ODPP_NONE) {
2684 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix)) {
2687 /* If tunnel sampling is enabled, put an additional option attribute:
2688 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2690 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix) &&
2691 dpif_ipfix_get_tunnel_port(ipfix, output_odp_port) ) {
2692 tunnel_out_port = output_odp_port;
2696 union user_action_cookie cookie = {
2698 .type = USER_ACTION_COOKIE_IPFIX,
2699 .output_odp_port = output_odp_port,
2702 compose_sample_action(ctx,
2703 dpif_ipfix_get_bridge_exporter_probability(ipfix),
2704 &cookie, sizeof cookie.ipfix, tunnel_out_port,
2708 /* Fix "sample" action according to data collected while composing ODP actions,
2709 * as described in compose_sflow_action().
2711 * 'user_cookie_offset' must be the offset returned by add_sflow_action(). */
2713 fix_sflow_action(struct xlate_ctx *ctx, unsigned int user_cookie_offset)
2715 const struct flow *base = &ctx->base_flow;
2716 union user_action_cookie *cookie;
2718 cookie = ofpbuf_at(ctx->odp_actions, user_cookie_offset,
2719 sizeof cookie->sflow);
2720 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2722 cookie->type = USER_ACTION_COOKIE_SFLOW;
2723 cookie->sflow.vlan_tci = base->vlan_tci;
2725 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2726 * port information") for the interpretation of cookie->output. */
2727 switch (ctx->sflow_n_outputs) {
2729 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2730 cookie->sflow.output = 0x40000000 | 256;
2734 cookie->sflow.output = dpif_sflow_odp_port_to_ifindex(
2735 ctx->xbridge->sflow, ctx->sflow_odp_port);
2736 if (cookie->sflow.output) {
2741 /* 0x80000000 means "multiple output ports. */
2742 cookie->sflow.output = 0x80000000 | ctx->sflow_n_outputs;
2748 process_special(struct xlate_ctx *ctx, const struct xport *xport)
2750 const struct flow *flow = &ctx->xin->flow;
2751 struct flow_wildcards *wc = ctx->wc;
2752 const struct xbridge *xbridge = ctx->xbridge;
2753 const struct dp_packet *packet = ctx->xin->packet;
2754 enum slow_path_reason slow;
2758 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2760 cfm_process_heartbeat(xport->cfm, packet);
2763 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2765 bfd_process_packet(xport->bfd, flow, packet);
2766 /* If POLL received, immediately sends FINAL back. */
2767 if (bfd_should_send_packet(xport->bfd)) {
2768 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2772 } else if (xport->xbundle && xport->xbundle->lacp
2773 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2775 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2778 } else if ((xbridge->stp || xbridge->rstp) &&
2779 stp_should_process_flow(flow, wc)) {
2782 ? stp_process_packet(xport, packet)
2783 : rstp_process_packet(xport, packet);
2786 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2788 lldp_process_packet(xport->lldp, packet);
2796 ctx->xout->slow |= slow;
2804 tnl_route_lookup_flow(const struct flow *oflow,
2805 struct in6_addr *ip, struct in6_addr *src,
2806 struct xport **out_port)
2808 char out_dev[IFNAMSIZ];
2809 struct xbridge *xbridge;
2810 struct xlate_cfg *xcfg;
2812 struct in6_addr dst;
2814 dst = flow_tnl_dst(&oflow->tunnel);
2815 if (!ovs_router_lookup(&dst, out_dev, src, &gw)) {
2819 if (ipv6_addr_is_set(&gw) &&
2820 (!IN6_IS_ADDR_V4MAPPED(&gw) || in6_addr_get_mapped_ipv4(&gw))) {
2826 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2829 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2830 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2833 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2834 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2845 compose_table_xlate(struct xlate_ctx *ctx, const struct xport *out_dev,
2846 struct dp_packet *packet)
2848 struct xbridge *xbridge = out_dev->xbridge;
2849 struct ofpact_output output;
2852 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2853 flow_extract(packet, &flow);
2854 flow.in_port.ofp_port = out_dev->ofp_port;
2855 output.port = OFPP_TABLE;
2858 return ofproto_dpif_execute_actions__(xbridge->ofproto, &flow, NULL,
2859 &output.ofpact, sizeof output,
2860 ctx->indentation, ctx->depth,
2861 ctx->resubmits, packet);
2865 tnl_send_nd_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2866 const struct eth_addr eth_src,
2867 struct in6_addr * ipv6_src, struct in6_addr * ipv6_dst)
2869 struct dp_packet packet;
2871 dp_packet_init(&packet, 0);
2872 compose_nd(&packet, eth_src, ipv6_src, ipv6_dst);
2873 compose_table_xlate(ctx, out_dev, &packet);
2874 dp_packet_uninit(&packet);
2878 tnl_send_arp_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2879 const struct eth_addr eth_src,
2880 ovs_be32 ip_src, ovs_be32 ip_dst)
2882 struct dp_packet packet;
2884 dp_packet_init(&packet, 0);
2885 compose_arp(&packet, ARP_OP_REQUEST,
2886 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2888 compose_table_xlate(ctx, out_dev, &packet);
2889 dp_packet_uninit(&packet);
2893 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2894 const struct flow *flow, odp_port_t tunnel_odp_port)
2896 struct netdev_tnl_build_header_params tnl_params;
2897 struct ovs_action_push_tnl tnl_push_data;
2898 struct xport *out_dev = NULL;
2899 ovs_be32 s_ip = 0, d_ip = 0;
2900 struct in6_addr s_ip6 = in6addr_any;
2901 struct in6_addr d_ip6 = in6addr_any;
2902 struct eth_addr smac;
2903 struct eth_addr dmac;
2905 char buf_sip6[INET6_ADDRSTRLEN];
2906 char buf_dip6[INET6_ADDRSTRLEN];
2908 err = tnl_route_lookup_flow(flow, &d_ip6, &s_ip6, &out_dev);
2910 xlate_report(ctx, "native tunnel routing failed");
2914 xlate_report(ctx, "tunneling to %s via %s",
2915 ipv6_string_mapped(buf_dip6, &d_ip6),
2916 netdev_get_name(out_dev->netdev));
2918 /* Use mac addr of bridge port of the peer. */
2919 err = netdev_get_etheraddr(out_dev->netdev, &smac);
2921 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2925 d_ip = in6_addr_get_mapped_ipv4(&d_ip6);
2927 s_ip = in6_addr_get_mapped_ipv4(&s_ip6);
2930 err = tnl_neigh_lookup(out_dev->xbridge->name, &d_ip6, &dmac);
2932 xlate_report(ctx, "neighbor cache miss for %s on bridge %s, "
2933 "sending %s request",
2934 buf_dip6, out_dev->xbridge->name, d_ip ? "ARP" : "ND");
2936 tnl_send_arp_request(ctx, out_dev, smac, s_ip, d_ip);
2938 tnl_send_nd_request(ctx, out_dev, smac, &s_ip6, &d_ip6);
2943 if (ctx->xin->xcache) {
2944 struct xc_entry *entry;
2946 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_NEIGH);
2947 ovs_strlcpy(entry->u.tnl_neigh_cache.br_name, out_dev->xbridge->name,
2948 sizeof entry->u.tnl_neigh_cache.br_name);
2949 entry->u.tnl_neigh_cache.d_ipv6 = d_ip6;
2952 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" %s"
2953 " to "ETH_ADDR_FMT" %s",
2954 ETH_ADDR_ARGS(smac), ipv6_string_mapped(buf_sip6, &s_ip6),
2955 ETH_ADDR_ARGS(dmac), buf_dip6);
2957 netdev_init_tnl_build_header_params(&tnl_params, flow, &s_ip6, dmac, smac);
2958 err = tnl_port_build_header(xport->ofport, &tnl_push_data, &tnl_params);
2962 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2963 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2964 odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
2969 xlate_commit_actions(struct xlate_ctx *ctx)
2971 bool use_masked = ctx->xbridge->support.masked_set_action;
2973 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2974 ctx->odp_actions, ctx->wc,
2979 clear_conntrack(struct flow *flow)
2984 memset(&flow->ct_label, 0, sizeof flow->ct_label);
2988 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2989 const struct xlate_bond_recirc *xr, bool check_stp)
2991 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2992 struct flow_wildcards *wc = ctx->wc;
2993 struct flow *flow = &ctx->xin->flow;
2994 struct flow_tnl flow_tnl;
2995 ovs_be16 flow_vlan_tci;
2996 uint32_t flow_pkt_mark;
2997 uint8_t flow_nw_tos;
2998 odp_port_t out_port, odp_port;
2999 bool tnl_push_pop_send = false;
3002 /* If 'struct flow' gets additional metadata, we'll need to zero it out
3003 * before traversing a patch port. */
3004 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
3005 memset(&flow_tnl, 0, sizeof flow_tnl);
3008 xlate_report(ctx, "Nonexistent output port");
3010 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
3011 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
3013 } else if (check_stp) {
3014 if (is_stp(&ctx->base_flow)) {
3015 if (!xport_stp_should_forward_bpdu(xport) &&
3016 !xport_rstp_should_manage_bpdu(xport)) {
3017 if (ctx->xbridge->stp != NULL) {
3018 xlate_report(ctx, "STP not in listening state, "
3019 "skipping bpdu output");
3020 } else if (ctx->xbridge->rstp != NULL) {
3021 xlate_report(ctx, "RSTP not managing BPDU in this state, "
3022 "skipping bpdu output");
3026 } else if (!xport_stp_forward_state(xport) ||
3027 !xport_rstp_forward_state(xport)) {
3028 if (ctx->xbridge->stp != NULL) {
3029 xlate_report(ctx, "STP not in forwarding state, "
3031 } else if (ctx->xbridge->rstp != NULL) {
3032 xlate_report(ctx, "RSTP not in forwarding state, "
3040 const struct xport *peer = xport->peer;
3041 struct flow old_flow = ctx->xin->flow;
3042 bool old_conntrack = ctx->conntracked;
3043 bool old_was_mpls = ctx->was_mpls;
3044 cls_version_t old_version = ctx->tables_version;
3045 struct ofpbuf old_stack = ctx->stack;
3046 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
3047 struct ofpbuf old_action_set = ctx->action_set;
3048 uint64_t actset_stub[1024 / 8];
3050 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
3051 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
3052 ctx->xbridge = peer->xbridge;
3053 flow->in_port.ofp_port = peer->ofp_port;
3054 flow->metadata = htonll(0);
3055 memset(&flow->tunnel, 0, sizeof flow->tunnel);
3056 memset(flow->regs, 0, sizeof flow->regs);
3057 flow->actset_output = OFPP_UNSET;
3058 ctx->conntracked = false;
3059 clear_conntrack(flow);
3061 /* The bridge is now known so obtain its table version. */
3063 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
3065 if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
3066 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
3067 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
3068 if (!ctx->freezing) {
3069 xlate_action_set(ctx);
3071 if (ctx->freezing) {
3072 finish_freezing(ctx);
3075 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
3076 * the learning action look at the packet, then drop it. */
3077 struct flow old_base_flow = ctx->base_flow;
3078 size_t old_size = ctx->odp_actions->size;
3079 mirror_mask_t old_mirrors = ctx->mirrors;
3081 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
3082 ctx->mirrors = old_mirrors;
3083 ctx->base_flow = old_base_flow;
3084 ctx->odp_actions->size = old_size;
3086 /* Undo changes that may have been done for freezing. */
3087 ctx_cancel_freeze(ctx);
3091 ctx->xin->flow = old_flow;
3092 ctx->xbridge = xport->xbridge;
3093 ofpbuf_uninit(&ctx->action_set);
3094 ctx->action_set = old_action_set;
3095 ofpbuf_uninit(&ctx->stack);
3096 ctx->stack = old_stack;
3098 /* Restore calling bridge's lookup version. */
3099 ctx->tables_version = old_version;
3101 /* The peer bridge popping MPLS should have no effect on the original
3103 ctx->was_mpls = old_was_mpls;
3105 /* The peer bridge's conntrack execution should have no effect on the
3106 * original bridge. */
3107 ctx->conntracked = old_conntrack;
3109 /* The fact that the peer bridge exits (for any reason) does not mean
3110 * that the original bridge should exit. Specifically, if the peer
3111 * bridge freezes translation, the original bridge must continue
3112 * processing with the original, not the frozen packet! */
3115 /* Peer bridge errors do not propagate back. */
3116 ctx->error = XLATE_OK;
3118 if (ctx->xin->resubmit_stats) {
3119 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
3120 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
3122 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
3125 if (ctx->xin->xcache) {
3126 struct xc_entry *entry;
3128 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3129 entry->u.dev.tx = netdev_ref(xport->netdev);
3130 entry->u.dev.rx = netdev_ref(peer->netdev);
3131 entry->u.dev.bfd = bfd_ref(peer->bfd);
3136 flow_vlan_tci = flow->vlan_tci;
3137 flow_pkt_mark = flow->pkt_mark;
3138 flow_nw_tos = flow->nw_tos;
3140 if (count_skb_priorities(xport)) {
3141 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3142 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
3143 wc->masks.nw_tos |= IP_DSCP_MASK;
3144 flow->nw_tos &= ~IP_DSCP_MASK;
3145 flow->nw_tos |= dscp;
3149 if (xport->is_tunnel) {
3150 struct in6_addr dst;
3151 /* Save tunnel metadata so that changes made due to
3152 * the Logical (tunnel) Port are not visible for any further
3153 * matches, while explicit set actions on tunnel metadata are.
3155 flow_tnl = flow->tunnel;
3156 odp_port = tnl_port_send(xport->ofport, flow, ctx->wc);
3157 if (odp_port == ODPP_NONE) {
3158 xlate_report(ctx, "Tunneling decided against output");
3159 goto out; /* restore flow_nw_tos */
3161 dst = flow_tnl_dst(&flow->tunnel);
3162 if (ipv6_addr_equals(&dst, &ctx->orig_tunnel_ipv6_dst)) {
3163 xlate_report(ctx, "Not tunneling to our own address");
3164 goto out; /* restore flow_nw_tos */
3166 if (ctx->xin->resubmit_stats) {
3167 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
3169 if (ctx->xin->xcache) {
3170 struct xc_entry *entry;
3172 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3173 entry->u.dev.tx = netdev_ref(xport->netdev);
3175 out_port = odp_port;
3176 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3177 xlate_report(ctx, "output to native tunnel");
3178 tnl_push_pop_send = true;
3180 xlate_report(ctx, "output to kernel tunnel");
3181 commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions);
3182 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3185 odp_port = xport->odp_port;
3186 out_port = odp_port;
3189 if (out_port != ODPP_NONE) {
3190 xlate_commit_actions(ctx);
3193 struct ovs_action_hash *act_hash;
3196 act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
3197 OVS_ACTION_ATTR_HASH,
3199 act_hash->hash_alg = xr->hash_alg;
3200 act_hash->hash_basis = xr->hash_basis;
3202 /* Recirc action. */
3203 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
3207 if (tnl_push_pop_send) {
3208 build_tunnel_send(ctx, xport, flow, odp_port);
3209 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3211 odp_port_t odp_tnl_port = ODPP_NONE;
3213 /* XXX: Write better Filter for tunnel port. We can use inport
3214 * int tunnel-port flow to avoid these checks completely. */
3215 if (ofp_port == OFPP_LOCAL &&
3216 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3218 odp_tnl_port = tnl_port_map_lookup(flow, wc);
3221 if (odp_tnl_port != ODPP_NONE) {
3222 nl_msg_put_odp_port(ctx->odp_actions,
3223 OVS_ACTION_ATTR_TUNNEL_POP,
3226 /* Tunnel push-pop action is not compatible with
3228 compose_ipfix_action(ctx, out_port);
3229 nl_msg_put_odp_port(ctx->odp_actions,
3230 OVS_ACTION_ATTR_OUTPUT,
3236 ctx->sflow_odp_port = odp_port;
3237 ctx->sflow_n_outputs++;
3238 ctx->nf_output_iface = ofp_port;
3241 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
3242 mirror_packet(ctx, xport->xbundle,
3243 xbundle_mirror_dst(xport->xbundle->xbridge,
3249 flow->vlan_tci = flow_vlan_tci;
3250 flow->pkt_mark = flow_pkt_mark;
3251 flow->nw_tos = flow_nw_tos;
3255 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3256 const struct xlate_bond_recirc *xr)
3258 compose_output_action__(ctx, ofp_port, xr, true);
3262 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule, bool deepens)
3264 struct rule_dpif *old_rule = ctx->rule;
3265 ovs_be64 old_cookie = ctx->rule_cookie;
3266 const struct rule_actions *actions;
3268 if (ctx->xin->resubmit_stats) {
3269 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3275 ctx->depth += deepens;
3277 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3278 actions = rule_dpif_get_actions(rule);
3279 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3280 ctx->rule_cookie = old_cookie;
3281 ctx->rule = old_rule;
3282 ctx->depth -= deepens;
3287 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3289 if (ctx->depth >= MAX_DEPTH) {
3290 XLATE_REPORT_ERROR(ctx, "over max translation depth %d", MAX_DEPTH);
3291 ctx->error = XLATE_RECURSION_TOO_DEEP;
3292 } else if (ctx->resubmits >= MAX_RESUBMITS) {
3293 XLATE_REPORT_ERROR(ctx, "over %d resubmit actions", MAX_RESUBMITS);
3294 ctx->error = XLATE_TOO_MANY_RESUBMITS;
3295 } else if (ctx->odp_actions->size > UINT16_MAX) {
3296 XLATE_REPORT_ERROR(ctx, "resubmits yielded over 64 kB of actions");
3297 /* NOT an error, as we'll be slow-pathing the flow in this case? */
3298 ctx->exit = true; /* XXX: translation still terminated! */
3299 } else if (ctx->stack.size >= 65536) {
3300 XLATE_REPORT_ERROR(ctx, "resubmits yielded over 64 kB of stack");
3301 ctx->error = XLATE_STACK_TOO_DEEP;
3310 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3311 bool may_packet_in, bool honor_table_miss)
3313 /* Check if we need to recirculate before matching in a table. */
3314 if (ctx->was_mpls) {
3315 ctx_trigger_freeze(ctx);
3318 if (xlate_resubmit_resource_check(ctx)) {
3319 uint8_t old_table_id = ctx->table_id;
3320 struct rule_dpif *rule;
3322 ctx->table_id = table_id;
3324 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3325 ctx->tables_version,
3326 &ctx->xin->flow, ctx->wc,
3327 ctx->xin->resubmit_stats,
3328 &ctx->table_id, in_port,
3329 may_packet_in, honor_table_miss);
3331 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3332 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->indentation + 1);
3336 /* Fill in the cache entry here instead of xlate_recursively
3337 * to make the reference counting more explicit. We take a
3338 * reference in the lookups above if we are going to cache the
3340 if (ctx->xin->xcache) {
3341 struct xc_entry *entry;
3343 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3344 entry->u.rule = rule;
3345 rule_dpif_ref(rule);
3347 xlate_recursively(ctx, rule, table_id <= old_table_id);
3350 ctx->table_id = old_table_id;
3356 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3357 struct ofputil_bucket *bucket)
3359 if (ctx->xin->resubmit_stats) {
3360 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3362 if (ctx->xin->xcache) {
3363 struct xc_entry *entry;
3365 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3366 entry->u.group.group = group_dpif_ref(group);
3367 entry->u.group.bucket = bucket;
3372 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3374 uint64_t action_list_stub[1024 / 8];
3375 struct ofpbuf action_list = OFPBUF_STUB_INITIALIZER(action_list_stub);
3376 struct ofpbuf action_set = ofpbuf_const_initializer(bucket->ofpacts,
3377 bucket->ofpacts_len);
3378 struct flow old_flow = ctx->xin->flow;
3379 bool old_was_mpls = ctx->was_mpls;
3381 ofpacts_execute_action_set(&action_list, &action_set);
3384 do_xlate_actions(action_list.data, action_list.size, ctx);
3388 ofpbuf_uninit(&action_list);
3390 /* Check if need to freeze. */
3391 if (ctx->freezing) {
3392 finish_freezing(ctx);
3395 /* Roll back flow to previous state.
3396 * This is equivalent to cloning the packet for each bucket.
3398 * As a side effect any subsequently applied actions will
3399 * also effectively be applied to a clone of the packet taken
3400 * just before applying the all or indirect group.
3402 * Note that group buckets are action sets, hence they cannot modify the
3403 * main action set. Also any stack actions are ignored when executing an
3404 * action set, so group buckets cannot change the stack either.
3405 * However, we do allow resubmit actions in group buckets, which could
3406 * break the above assumptions. It is up to the controller to not mess up
3407 * with the action_set and stack in the tables resubmitted to from
3409 ctx->xin->flow = old_flow;
3411 /* The group bucket popping MPLS should have no effect after bucket
3413 ctx->was_mpls = old_was_mpls;
3415 /* The fact that the group bucket exits (for any reason) does not mean that
3416 * the translation after the group action should exit. Specifically, if
3417 * the group bucket freezes translation, the actions after the group action
3418 * must continue processing with the original, not the frozen packet! */
3423 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3425 struct ofputil_bucket *bucket;
3426 const struct ovs_list *buckets;
3428 group_dpif_get_buckets(group, &buckets);
3430 LIST_FOR_EACH (bucket, list_node, buckets) {
3431 xlate_group_bucket(ctx, bucket);
3433 xlate_group_stats(ctx, group, NULL);
3437 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3439 struct ofputil_bucket *bucket;
3441 bucket = group_first_live_bucket(ctx, group, 0);
3443 xlate_group_bucket(ctx, bucket);
3444 xlate_group_stats(ctx, group, bucket);
3449 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3451 struct flow_wildcards *wc = ctx->wc;
3452 struct ofputil_bucket *bucket;
3455 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3456 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3457 bucket = group_best_live_bucket(ctx, group, basis);
3459 xlate_group_bucket(ctx, bucket);
3460 xlate_group_stats(ctx, group, bucket);
3465 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3467 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3468 const struct field_array *fields;
3469 struct ofputil_bucket *bucket;
3473 fields = group_dpif_get_fields(group);
3474 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3476 /* Determine which fields to hash */
3477 for (i = 0; i < MFF_N_IDS; i++) {
3478 if (bitmap_is_set(fields->used.bm, i)) {
3479 const struct mf_field *mf;
3481 /* If the field is already present in 'hash_fields' then
3482 * this loop has already checked that it and its pre-requisites
3483 * are present in the flow and its pre-requisites have
3484 * already been added to 'hash_fields'. There is nothing more
3485 * to do here and as an optimisation the loop can continue. */
3486 if (bitmap_is_set(hash_fields.bm, i)) {
3492 /* Only hash a field if it and its pre-requisites are present
3494 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3498 /* Hash both the field and its pre-requisites */
3499 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3503 /* Hash the fields */
3504 for (i = 0; i < MFF_N_IDS; i++) {
3505 if (bitmap_is_set(hash_fields.bm, i)) {
3506 const struct mf_field *mf = mf_from_id(i);
3507 union mf_value value;
3510 mf_get_value(mf, &ctx->xin->flow, &value);
3511 /* This seems inefficient but so does apply_mask() */
3512 for (j = 0; j < mf->n_bytes; j++) {
3513 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3515 basis = hash_bytes(&value, mf->n_bytes, basis);
3517 /* For tunnels, hash in whether the field is present. */
3518 if (mf_is_tun_metadata(mf)) {
3519 basis = hash_boolean(mf_is_set(mf, &ctx->xin->flow), basis);
3522 mf_mask_field(mf, &ctx->wc->masks);
3526 bucket = group_best_live_bucket(ctx, group, basis);
3528 xlate_group_bucket(ctx, bucket);
3529 xlate_group_stats(ctx, group, bucket);
3534 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3536 const char *selection_method = group_dpif_get_selection_method(group);
3538 /* Select groups may access flow keys beyond L2 in order to
3539 * select a bucket. Recirculate as appropriate to make this possible.
3541 if (ctx->was_mpls) {
3542 ctx_trigger_freeze(ctx);
3545 if (selection_method[0] == '\0') {
3546 xlate_default_select_group(ctx, group);
3547 } else if (!strcasecmp("hash", selection_method)) {
3548 xlate_hash_fields_select_group(ctx, group);
3550 /* Parsing of groups should ensure this never happens */
3556 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3558 bool was_in_group = ctx->in_group;
3559 ctx->in_group = true;
3561 switch (group_dpif_get_type(group)) {
3563 case OFPGT11_INDIRECT:
3564 xlate_all_group(ctx, group);
3566 case OFPGT11_SELECT:
3567 xlate_select_group(ctx, group);
3570 xlate_ff_group(ctx, group);
3575 group_dpif_unref(group);
3577 ctx->in_group = was_in_group;
3581 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3583 if (xlate_resubmit_resource_check(ctx)) {
3584 struct group_dpif *group;
3587 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3589 xlate_group_action__(ctx, group);
3599 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3600 const struct ofpact_resubmit *resubmit)
3604 bool may_packet_in = false;
3605 bool honor_table_miss = false;
3607 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3608 /* Still allow missed packets to be sent to the controller
3609 * if resubmitting from an internal table. */
3610 may_packet_in = true;
3611 honor_table_miss = true;
3614 in_port = resubmit->in_port;
3615 if (in_port == OFPP_IN_PORT) {
3616 in_port = ctx->xin->flow.in_port.ofp_port;
3619 table_id = resubmit->table_id;
3620 if (table_id == 255) {
3621 table_id = ctx->table_id;
3624 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3629 flood_packets(struct xlate_ctx *ctx, bool all)
3631 const struct xport *xport;
3633 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3634 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3639 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3640 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3641 compose_output_action(ctx, xport->ofp_port, NULL);
3645 ctx->nf_output_iface = NF_OUT_FLOOD;
3649 execute_controller_action(struct xlate_ctx *ctx, int len,
3650 enum ofp_packet_in_reason reason,
3651 uint16_t controller_id,
3652 const uint8_t *userdata, size_t userdata_len)
3654 struct dp_packet_batch batch;
3655 struct dp_packet *packet;
3657 ctx->xout->slow |= SLOW_CONTROLLER;
3658 xlate_commit_actions(ctx);
3659 if (!ctx->xin->packet) {
3663 packet = dp_packet_clone(ctx->xin->packet);
3664 packet_batch_init_packet(&batch, packet);
3665 odp_execute_actions(NULL, &batch, false,
3666 ctx->odp_actions->data, ctx->odp_actions->size, NULL);
3668 /* A packet sent by an action in a table-miss rule is considered an
3669 * explicit table miss. OpenFlow before 1.3 doesn't have that concept so
3670 * it will get translated back to OFPR_ACTION for those versions. */
3671 if (reason == OFPR_ACTION
3672 && ctx->rule && rule_dpif_is_table_miss(ctx->rule)) {
3673 reason = OFPR_EXPLICIT_MISS;
3676 size_t packet_len = dp_packet_size(packet);
3678 struct ofproto_async_msg *am = xmalloc(sizeof *am);
3679 *am = (struct ofproto_async_msg) {
3680 .controller_id = controller_id,
3681 .oam = OAM_PACKET_IN,
3685 .packet = dp_packet_steal_data(packet),
3686 .packet_len = packet_len,
3688 .table_id = ctx->table_id,
3689 .cookie = ctx->rule_cookie,
3690 .userdata = (userdata_len
3691 ? xmemdup(userdata, userdata_len)
3693 .userdata_len = userdata_len,
3699 flow_get_metadata(&ctx->xin->flow, &am->pin.up.public.flow_metadata);
3701 ofproto_dpif_send_async_msg(ctx->xbridge->ofproto, am);
3702 dp_packet_delete(packet);
3706 emit_continuation(struct xlate_ctx *ctx, const struct frozen_state *state)
3708 struct ofproto_async_msg *am = xmalloc(sizeof *am);
3709 *am = (struct ofproto_async_msg) {
3710 .controller_id = ctx->pause->controller_id,
3711 .oam = OAM_PACKET_IN,
3715 .userdata = xmemdup(ctx->pause->userdata,
3716 ctx->pause->userdata_len),
3717 .userdata_len = ctx->pause->userdata_len,
3718 .packet = xmemdup(dp_packet_data(ctx->xin->packet),
3719 dp_packet_size(ctx->xin->packet)),
3720 .packet_len = dp_packet_size(ctx->xin->packet),
3721 .reason = ctx->pause->reason,
3723 .bridge = *ofproto_dpif_get_uuid(ctx->xbridge->ofproto),
3724 .stack = xmemdup(state->stack,
3725 state->n_stack * sizeof *state->stack),
3726 .n_stack = state->n_stack,
3727 .mirrors = state->mirrors,
3728 .conntracked = state->conntracked,
3729 .actions = xmemdup(state->ofpacts, state->ofpacts_len),
3730 .actions_len = state->ofpacts_len,
3731 .action_set = xmemdup(state->action_set,
3732 state->action_set_len),
3733 .action_set_len = state->action_set_len,
3735 .max_len = UINT16_MAX,
3738 flow_get_metadata(&ctx->xin->flow, &am->pin.up.public.flow_metadata);
3739 ofproto_dpif_send_async_msg(ctx->xbridge->ofproto, am);
3743 finish_freezing__(struct xlate_ctx *ctx, uint8_t table)
3745 ovs_assert(ctx->freezing);
3747 struct frozen_state state = {
3749 .ofproto_uuid = *ofproto_dpif_get_uuid(ctx->xbridge->ofproto),
3750 .stack = ctx->stack.data,
3751 .n_stack = ctx->stack.size / sizeof(union mf_subvalue),
3752 .mirrors = ctx->mirrors,
3753 .conntracked = ctx->conntracked,
3754 .ofpacts = ctx->frozen_actions.data,
3755 .ofpacts_len = ctx->frozen_actions.size,
3756 .action_set = ctx->action_set.data,
3757 .action_set_len = ctx->action_set.size,
3759 frozen_metadata_from_flow(&state.metadata, &ctx->xin->flow);
3762 if (ctx->xin->packet) {
3763 emit_continuation(ctx, &state);
3766 /* Allocate a unique recirc id for the given metadata state in the
3767 * flow. An existing id, with a new reference to the corresponding
3768 * recirculation context, will be returned if possible.
3769 * The life-cycle of this recirc id is managed by associating it
3770 * with the udpif key ('ukey') created for each new datapath flow. */
3771 uint32_t id = recirc_alloc_id_ctx(&state);
3773 XLATE_REPORT_ERROR(ctx, "Failed to allocate recirculation id");
3774 ctx->error = XLATE_NO_RECIRCULATION_CONTEXT;
3777 recirc_refs_add(&ctx->xout->recircs, id);
3779 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3782 /* Undo changes done by freezing. */
3783 ctx_cancel_freeze(ctx);
3786 /* Called only when we're freezing. */
3788 finish_freezing(struct xlate_ctx *ctx)
3790 xlate_commit_actions(ctx);
3791 finish_freezing__(ctx, 0);
3794 /* Fork the pipeline here. The current packet will continue processing the
3795 * current action list. A clone of the current packet will recirculate, skip
3796 * the remainder of the current action list and asynchronously resume pipeline
3797 * processing in 'table' with the current metadata and action set. */
3799 compose_recirculate_and_fork(struct xlate_ctx *ctx, uint8_t table)
3801 ctx->freezing = true;
3802 finish_freezing__(ctx, table);
3806 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3808 struct flow *flow = &ctx->xin->flow;
3811 ovs_assert(eth_type_mpls(mpls->ethertype));
3813 n = flow_count_mpls_labels(flow, ctx->wc);
3815 xlate_commit_actions(ctx);
3816 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3817 if (ctx->xin->packet != NULL) {
3818 XLATE_REPORT_ERROR(ctx, "bridge %s: dropping packet on which an "
3819 "MPLS push action can't be performed as it would "
3820 "have more MPLS LSEs than the %d supported.",
3821 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3823 ctx->error = XLATE_TOO_MANY_MPLS_LABELS;
3827 flow_push_mpls(flow, n, mpls->ethertype, ctx->wc);
3831 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3833 struct flow *flow = &ctx->xin->flow;
3834 int n = flow_count_mpls_labels(flow, ctx->wc);
3836 if (flow_pop_mpls(flow, n, eth_type, ctx->wc)) {
3837 if (!eth_type_mpls(eth_type) && ctx->xbridge->support.odp.recirc) {
3838 ctx->was_mpls = true;
3840 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3841 if (ctx->xin->packet != NULL) {
3842 XLATE_REPORT_ERROR(ctx, "bridge %s: dropping packet on which an "
3843 "MPLS pop action can't be performed as it has "
3844 "more MPLS LSEs than the %d supported.",
3845 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3847 ctx->error = XLATE_TOO_MANY_MPLS_LABELS;
3848 ofpbuf_clear(ctx->odp_actions);
3853 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3855 struct flow *flow = &ctx->xin->flow;
3857 if (!is_ip_any(flow)) {
3861 ctx->wc->masks.nw_ttl = 0xff;
3862 if (flow->nw_ttl > 1) {
3868 for (i = 0; i < ids->n_controllers; i++) {
3869 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3870 ids->cnt_ids[i], NULL, 0);
3873 /* Stop processing for current table. */
3879 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3881 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3882 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3883 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3888 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3890 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3891 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3892 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3897 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3899 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3900 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3901 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3906 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3908 struct flow *flow = &ctx->xin->flow;
3910 if (eth_type_mpls(flow->dl_type)) {
3911 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3913 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3916 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3919 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0,
3924 /* Stop processing for current table. */
3929 xlate_output_action(struct xlate_ctx *ctx,
3930 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3932 ofp_port_t prev_nf_output_iface = ctx->nf_output_iface;
3934 ctx->nf_output_iface = NF_OUT_DROP;
3938 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3941 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3942 0, may_packet_in, true);
3948 flood_packets(ctx, false);
3951 flood_packets(ctx, true);
3953 case OFPP_CONTROLLER:
3954 execute_controller_action(ctx, max_len,
3955 (ctx->in_group ? OFPR_GROUP
3956 : ctx->in_action_set ? OFPR_ACTION_SET
3964 if (port != ctx->xin->flow.in_port.ofp_port) {
3965 compose_output_action(ctx, port, NULL);
3967 xlate_report(ctx, "skipping output to input port");
3972 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3973 ctx->nf_output_iface = NF_OUT_FLOOD;
3974 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3975 ctx->nf_output_iface = prev_nf_output_iface;
3976 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3977 ctx->nf_output_iface != NF_OUT_FLOOD) {
3978 ctx->nf_output_iface = NF_OUT_MULTI;
3983 xlate_output_reg_action(struct xlate_ctx *ctx,
3984 const struct ofpact_output_reg *or)
3986 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3987 if (port <= UINT16_MAX) {
3988 union mf_subvalue value;
3990 memset(&value, 0xff, sizeof value);
3991 mf_write_subfield_flow(&or->src, &value, &ctx->wc->masks);
3992 xlate_output_action(ctx, u16_to_ofp(port),
3993 or->max_len, false);
3998 xlate_enqueue_action(struct xlate_ctx *ctx,
3999 const struct ofpact_enqueue *enqueue)
4001 ofp_port_t ofp_port = enqueue->port;
4002 uint32_t queue_id = enqueue->queue;
4003 uint32_t flow_priority, priority;
4006 /* Translate queue to priority. */
4007 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
4009 /* Fall back to ordinary output action. */
4010 xlate_output_action(ctx, enqueue->port, 0, false);
4014 /* Check output port. */
4015 if (ofp_port == OFPP_IN_PORT) {
4016 ofp_port = ctx->xin->flow.in_port.ofp_port;
4017 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
4021 /* Add datapath actions. */
4022 flow_priority = ctx->xin->flow.skb_priority;
4023 ctx->xin->flow.skb_priority = priority;
4024 compose_output_action(ctx, ofp_port, NULL);
4025 ctx->xin->flow.skb_priority = flow_priority;
4027 /* Update NetFlow output port. */
4028 if (ctx->nf_output_iface == NF_OUT_DROP) {
4029 ctx->nf_output_iface = ofp_port;
4030 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
4031 ctx->nf_output_iface = NF_OUT_MULTI;
4036 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
4038 uint32_t skb_priority;
4040 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
4041 ctx->xin->flow.skb_priority = skb_priority;
4043 /* Couldn't translate queue to a priority. Nothing to do. A warning
4044 * has already been logged. */
4049 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
4051 const struct xbridge *xbridge = xbridge_;
4062 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
4065 port = get_ofp_port(xbridge, ofp_port);
4066 return port ? port->may_enable : false;
4071 xlate_bundle_action(struct xlate_ctx *ctx,
4072 const struct ofpact_bundle *bundle)
4076 port = bundle_execute(bundle, &ctx->xin->flow, ctx->wc, slave_enabled_cb,
4077 CONST_CAST(struct xbridge *, ctx->xbridge));
4078 if (bundle->dst.field) {
4079 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow, ctx->wc);
4081 xlate_output_action(ctx, port, 0, false);
4086 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
4087 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
4089 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
4090 if (ctx->xin->may_learn) {
4091 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
4096 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
4098 learn_mask(learn, ctx->wc);
4100 if (ctx->xin->xcache) {
4101 struct xc_entry *entry;
4103 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
4104 entry->u.learn.ofproto = ctx->xbridge->ofproto;
4105 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
4106 entry->u.learn.ofpacts = ofpbuf_new(64);
4107 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
4108 entry->u.learn.ofpacts);
4109 } else if (ctx->xin->may_learn) {
4110 uint64_t ofpacts_stub[1024 / 8];
4111 struct ofputil_flow_mod fm;
4112 struct ofpbuf ofpacts;
4114 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4115 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
4116 ofpbuf_uninit(&ofpacts);
4121 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
4122 uint16_t idle_timeout, uint16_t hard_timeout)
4124 if (tcp_flags & (TCP_FIN | TCP_RST)) {
4125 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
4130 xlate_fin_timeout(struct xlate_ctx *ctx,
4131 const struct ofpact_fin_timeout *oft)
4134 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
4135 oft->fin_idle_timeout, oft->fin_hard_timeout);
4136 if (ctx->xin->xcache) {
4137 struct xc_entry *entry;
4139 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
4140 /* XC_RULE already holds a reference on the rule, none is taken
4142 entry->u.fin.rule = ctx->rule;
4143 entry->u.fin.idle = oft->fin_idle_timeout;
4144 entry->u.fin.hard = oft->fin_hard_timeout;
4150 xlate_sample_action(struct xlate_ctx *ctx,
4151 const struct ofpact_sample *os)
4153 odp_port_t output_odp_port = ODPP_NONE;
4154 odp_port_t tunnel_out_port = ODPP_NONE;
4155 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
4156 bool emit_set_tunnel = false;
4158 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
4162 /* Scale the probability from 16-bit to 32-bit while representing
4163 * the same percentage. */
4164 uint32_t probability = (os->probability << 16) | os->probability;
4166 if (!ctx->xbridge->support.variable_length_userdata) {
4167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4169 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
4170 "lacks support (needs Linux 3.10+ or kernel module from "
4175 /* If ofp_port in flow sample action is equel to ofp_port,
4176 * this sample action is a input port action. */
4177 if (os->sampling_port != OFPP_NONE &&
4178 os->sampling_port != ctx->xin->flow.in_port.ofp_port) {
4179 output_odp_port = ofp_port_to_odp_port(ctx->xbridge,
4181 if (output_odp_port == ODPP_NONE) {
4182 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4183 VLOG_WARN_RL(&rl, "can't use unknown port %d in flow sample "
4184 "action", os->sampling_port);
4188 if (dpif_ipfix_get_flow_exporter_tunnel_sampling(ipfix,
4189 os->collector_set_id)
4190 && dpif_ipfix_get_tunnel_port(ipfix, output_odp_port)) {
4191 tunnel_out_port = output_odp_port;
4192 emit_set_tunnel = true;
4196 xlate_commit_actions(ctx);
4197 /* If 'emit_set_tunnel', sample(sampling_port=1) would translate
4198 * into datapath sample action set(tunnel(...)), sample(...) and
4199 * it is used for sampling egress tunnel information. */
4200 if (emit_set_tunnel) {
4201 const struct xport *xport = get_ofp_port(ctx->xbridge,
4204 if (xport && xport->is_tunnel) {
4205 struct flow *flow = &ctx->xin->flow;
4206 tnl_port_send(xport->ofport, flow, ctx->wc);
4207 if (!ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4208 struct flow_tnl flow_tnl = flow->tunnel;
4210 commit_odp_tunnel_action(flow, &ctx->base_flow,
4212 flow->tunnel = flow_tnl;
4215 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4216 VLOG_WARN_RL(&rl, "sampling_port:%d should be a tunnel port.",
4221 union user_action_cookie cookie = {
4223 .type = USER_ACTION_COOKIE_FLOW_SAMPLE,
4224 .probability = os->probability,
4225 .collector_set_id = os->collector_set_id,
4226 .obs_domain_id = os->obs_domain_id,
4227 .obs_point_id = os->obs_point_id,
4228 .output_odp_port = output_odp_port,
4231 compose_sample_action(ctx, probability, &cookie, sizeof cookie.flow_sample,
4232 tunnel_out_port, false);
4236 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
4238 if (xport->config & (is_stp(&ctx->xin->flow)
4239 ? OFPUTIL_PC_NO_RECV_STP
4240 : OFPUTIL_PC_NO_RECV)) {
4244 /* Only drop packets here if both forwarding and learning are
4245 * disabled. If just learning is enabled, we need to have
4246 * OFPP_NORMAL and the learning action have a look at the packet
4247 * before we can drop it. */
4248 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
4249 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
4257 xlate_write_actions__(struct xlate_ctx *ctx,
4258 const struct ofpact *ofpacts, size_t ofpacts_len)
4260 /* Maintain actset_output depending on the contents of the action set:
4262 * - OFPP_UNSET, if there is no "output" action.
4264 * - The output port, if there is an "output" action and no "group"
4267 * - OFPP_UNSET, if there is a "group" action.
4269 if (!ctx->action_set_has_group) {
4270 const struct ofpact *a;
4271 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4272 if (a->type == OFPACT_OUTPUT) {
4273 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(a)->port;
4274 } else if (a->type == OFPACT_GROUP) {
4275 ctx->xin->flow.actset_output = OFPP_UNSET;
4276 ctx->action_set_has_group = true;
4282 ofpbuf_put(&ctx->action_set, ofpacts, ofpacts_len);
4286 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact_nest *a)
4288 xlate_write_actions__(ctx, a->actions, ofpact_nest_get_action_len(a));
4292 xlate_action_set(struct xlate_ctx *ctx)
4294 uint64_t action_list_stub[1024 / 64];
4295 struct ofpbuf action_list;
4297 ctx->in_action_set = true;
4298 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
4299 ofpacts_execute_action_set(&action_list, &ctx->action_set);
4300 /* Clear the action set, as it is not needed any more. */
4301 ofpbuf_clear(&ctx->action_set);
4302 do_xlate_actions(action_list.data, action_list.size, ctx);
4303 ctx->in_action_set = false;
4304 ofpbuf_uninit(&action_list);
4308 freeze_put_unroll_xlate(struct xlate_ctx *ctx)
4310 struct ofpact_unroll_xlate *unroll = ctx->frozen_actions.header;
4312 /* Restore the table_id and rule cookie for a potential PACKET
4315 (ctx->table_id != unroll->rule_table_id
4316 || ctx->rule_cookie != unroll->rule_cookie)) {
4317 unroll = ofpact_put_UNROLL_XLATE(&ctx->frozen_actions);
4318 unroll->rule_table_id = ctx->table_id;
4319 unroll->rule_cookie = ctx->rule_cookie;
4320 ctx->frozen_actions.header = unroll;
4325 /* Copy actions 'a' through 'end' to ctx->frozen_actions, which will be
4326 * executed after thawing. Inserts an UNROLL_XLATE action, if none is already
4327 * present, before any action that may depend on the current table ID or flow
4330 freeze_unroll_actions(const struct ofpact *a, const struct ofpact *end,
4331 struct xlate_ctx *ctx)
4333 for (; a < end; a = ofpact_next(a)) {
4335 case OFPACT_OUTPUT_REG:
4338 case OFPACT_CONTROLLER:
4339 case OFPACT_DEC_MPLS_TTL:
4340 case OFPACT_DEC_TTL:
4341 /* These actions may generate asynchronous messages, which include
4342 * table ID and flow cookie information. */
4343 freeze_put_unroll_xlate(ctx);
4346 case OFPACT_RESUBMIT:
4347 if (ofpact_get_RESUBMIT(a)->table_id == 0xff) {
4348 /* This resubmit action is relative to the current table, so we
4349 * need to track what table that is.*/
4350 freeze_put_unroll_xlate(ctx);
4354 case OFPACT_SET_TUNNEL:
4355 case OFPACT_REG_MOVE:
4356 case OFPACT_SET_FIELD:
4357 case OFPACT_STACK_PUSH:
4358 case OFPACT_STACK_POP:
4360 case OFPACT_WRITE_METADATA:
4361 case OFPACT_GOTO_TABLE:
4362 case OFPACT_ENQUEUE:
4363 case OFPACT_SET_VLAN_VID:
4364 case OFPACT_SET_VLAN_PCP:
4365 case OFPACT_STRIP_VLAN:
4366 case OFPACT_PUSH_VLAN:
4367 case OFPACT_SET_ETH_SRC:
4368 case OFPACT_SET_ETH_DST:
4369 case OFPACT_SET_IPV4_SRC:
4370 case OFPACT_SET_IPV4_DST:
4371 case OFPACT_SET_IP_DSCP:
4372 case OFPACT_SET_IP_ECN:
4373 case OFPACT_SET_IP_TTL:
4374 case OFPACT_SET_L4_SRC_PORT:
4375 case OFPACT_SET_L4_DST_PORT:
4376 case OFPACT_SET_QUEUE:
4377 case OFPACT_POP_QUEUE:
4378 case OFPACT_PUSH_MPLS:
4379 case OFPACT_POP_MPLS:
4380 case OFPACT_SET_MPLS_LABEL:
4381 case OFPACT_SET_MPLS_TC:
4382 case OFPACT_SET_MPLS_TTL:
4383 case OFPACT_MULTIPATH:
4386 case OFPACT_UNROLL_XLATE:
4387 case OFPACT_FIN_TIMEOUT:
4388 case OFPACT_CLEAR_ACTIONS:
4389 case OFPACT_WRITE_ACTIONS:
4392 case OFPACT_DEBUG_RECIRC:
4395 /* These may not generate PACKET INs. */
4399 case OFPACT_CONJUNCTION:
4400 /* These need not be copied for restoration. */
4403 /* Copy the action over. */
4404 ofpbuf_put(&ctx->frozen_actions, a, OFPACT_ALIGN(a->len));
4409 put_ct_mark(const struct flow *flow, struct ofpbuf *odp_actions,
4410 struct flow_wildcards *wc)
4412 if (wc->masks.ct_mark) {
4418 odp_ct_mark = nl_msg_put_unspec_uninit(odp_actions, OVS_CT_ATTR_MARK,
4419 sizeof(*odp_ct_mark));
4420 odp_ct_mark->key = flow->ct_mark & wc->masks.ct_mark;
4421 odp_ct_mark->mask = wc->masks.ct_mark;
4426 put_ct_label(const struct flow *flow, struct ofpbuf *odp_actions,
4427 struct flow_wildcards *wc)
4429 if (!ovs_u128_is_zero(wc->masks.ct_label)) {
4435 odp_ct_label = nl_msg_put_unspec_uninit(odp_actions,
4437 sizeof(*odp_ct_label));
4438 odp_ct_label->key = ovs_u128_and(flow->ct_label, wc->masks.ct_label);
4439 odp_ct_label->mask = wc->masks.ct_label;
4444 put_ct_helper(struct ofpbuf *odp_actions, struct ofpact_conntrack *ofc)
4447 if (ofc->alg == IPPORT_FTP) {
4448 nl_msg_put_string(odp_actions, OVS_CT_ATTR_HELPER, "ftp");
4450 VLOG_WARN("Cannot serialize ct_helper %d\n", ofc->alg);
4456 put_ct_nat(struct xlate_ctx *ctx)
4458 struct ofpact_nat *ofn = ctx->ct_nat_action;
4465 nat_offset = nl_msg_start_nested(ctx->odp_actions, OVS_CT_ATTR_NAT);
4466 if (ofn->flags & NX_NAT_F_SRC || ofn->flags & NX_NAT_F_DST) {
4467 nl_msg_put_flag(ctx->odp_actions, ofn->flags & NX_NAT_F_SRC
4468 ? OVS_NAT_ATTR_SRC : OVS_NAT_ATTR_DST);
4469 if (ofn->flags & NX_NAT_F_PERSISTENT) {
4470 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PERSISTENT);
4472 if (ofn->flags & NX_NAT_F_PROTO_HASH) {
4473 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PROTO_HASH);
4474 } else if (ofn->flags & NX_NAT_F_PROTO_RANDOM) {
4475 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PROTO_RANDOM);
4477 if (ofn->range_af == AF_INET) {
4478 nl_msg_put_be32(ctx->odp_actions, OVS_NAT_ATTR_IP_MIN,
4479 ofn->range.addr.ipv4.min);
4480 if (ofn->range.addr.ipv4.max &&
4481 (ntohl(ofn->range.addr.ipv4.max)
4482 > ntohl(ofn->range.addr.ipv4.min))) {
4483 nl_msg_put_be32(ctx->odp_actions, OVS_NAT_ATTR_IP_MAX,
4484 ofn->range.addr.ipv4.max);
4486 } else if (ofn->range_af == AF_INET6) {
4487 nl_msg_put_unspec(ctx->odp_actions, OVS_NAT_ATTR_IP_MIN,
4488 &ofn->range.addr.ipv6.min,
4489 sizeof ofn->range.addr.ipv6.min);
4490 if (!ipv6_mask_is_any(&ofn->range.addr.ipv6.max) &&
4491 memcmp(&ofn->range.addr.ipv6.max, &ofn->range.addr.ipv6.min,
4492 sizeof ofn->range.addr.ipv6.max) > 0) {
4493 nl_msg_put_unspec(ctx->odp_actions, OVS_NAT_ATTR_IP_MAX,
4494 &ofn->range.addr.ipv6.max,
4495 sizeof ofn->range.addr.ipv6.max);
4498 if (ofn->range_af != AF_UNSPEC && ofn->range.proto.min) {
4499 nl_msg_put_u16(ctx->odp_actions, OVS_NAT_ATTR_PROTO_MIN,
4500 ofn->range.proto.min);
4501 if (ofn->range.proto.max &&
4502 ofn->range.proto.max > ofn->range.proto.min) {
4503 nl_msg_put_u16(ctx->odp_actions, OVS_NAT_ATTR_PROTO_MAX,
4504 ofn->range.proto.max);
4508 nl_msg_end_nested(ctx->odp_actions, nat_offset);
4512 compose_conntrack_action(struct xlate_ctx *ctx, struct ofpact_conntrack *ofc)
4514 ovs_u128 old_ct_label = ctx->base_flow.ct_label;
4515 ovs_u128 old_ct_label_mask = ctx->wc->masks.ct_label;
4516 uint32_t old_ct_mark = ctx->base_flow.ct_mark;
4517 uint32_t old_ct_mark_mask = ctx->wc->masks.ct_mark;
4521 /* Ensure that any prior actions are applied before composing the new
4522 * conntrack action. */
4523 xlate_commit_actions(ctx);
4525 /* Process nested actions first, to populate the key. */
4526 ctx->ct_nat_action = NULL;
4527 ctx->wc->masks.ct_mark = 0;
4528 ctx->wc->masks.ct_label.u64.hi = ctx->wc->masks.ct_label.u64.lo = 0;
4529 do_xlate_actions(ofc->actions, ofpact_ct_get_action_len(ofc), ctx);
4531 if (ofc->zone_src.field) {
4532 zone = mf_get_subfield(&ofc->zone_src, &ctx->xin->flow);
4534 zone = ofc->zone_imm;
4537 ct_offset = nl_msg_start_nested(ctx->odp_actions, OVS_ACTION_ATTR_CT);
4538 if (ofc->flags & NX_CT_F_COMMIT) {
4539 nl_msg_put_flag(ctx->odp_actions, OVS_CT_ATTR_COMMIT);
4541 nl_msg_put_u16(ctx->odp_actions, OVS_CT_ATTR_ZONE, zone);
4542 put_ct_mark(&ctx->xin->flow, ctx->odp_actions, ctx->wc);
4543 put_ct_label(&ctx->xin->flow, ctx->odp_actions, ctx->wc);
4544 put_ct_helper(ctx->odp_actions, ofc);
4546 ctx->ct_nat_action = NULL;
4547 nl_msg_end_nested(ctx->odp_actions, ct_offset);
4549 /* Restore the original ct fields in the key. These should only be exposed
4550 * after recirculation to another table. */
4551 ctx->base_flow.ct_mark = old_ct_mark;
4552 ctx->wc->masks.ct_mark = old_ct_mark_mask;
4553 ctx->base_flow.ct_label = old_ct_label;
4554 ctx->wc->masks.ct_label = old_ct_label_mask;
4556 if (ofc->recirc_table == NX_CT_RECIRC_NONE) {
4557 /* If we do not recirculate as part of this action, hide the results of
4558 * connection tracking from subsequent recirculations. */
4559 ctx->conntracked = false;
4561 /* Use ct_* fields from datapath during recirculation upcall. */
4562 ctx->conntracked = true;
4563 compose_recirculate_and_fork(ctx, ofc->recirc_table);
4568 recirc_for_mpls(const struct ofpact *a, struct xlate_ctx *ctx)
4570 /* No need to recirculate if already exiting. */
4575 /* Do not consider recirculating unless the packet was previously MPLS. */
4576 if (!ctx->was_mpls) {
4580 /* Special case these actions, only recirculating if necessary.
4581 * This avoids the overhead of recirculation in common use-cases.
4585 /* Output actions do not require recirculation. */
4587 case OFPACT_ENQUEUE:
4588 case OFPACT_OUTPUT_REG:
4589 /* Set actions that don't touch L3+ fields do not require recirculation. */
4590 case OFPACT_SET_VLAN_VID:
4591 case OFPACT_SET_VLAN_PCP:
4592 case OFPACT_SET_ETH_SRC:
4593 case OFPACT_SET_ETH_DST:
4594 case OFPACT_SET_TUNNEL:
4595 case OFPACT_SET_QUEUE:
4596 /* If actions of a group require recirculation that can be detected
4597 * when translating them. */
4601 /* Set field that don't touch L3+ fields don't require recirculation. */
4602 case OFPACT_SET_FIELD:
4603 if (mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field)) {
4608 /* For simplicity, recirculate in all other cases. */
4609 case OFPACT_CONTROLLER:
4611 case OFPACT_STRIP_VLAN:
4612 case OFPACT_PUSH_VLAN:
4613 case OFPACT_SET_IPV4_SRC:
4614 case OFPACT_SET_IPV4_DST:
4615 case OFPACT_SET_IP_DSCP:
4616 case OFPACT_SET_IP_ECN:
4617 case OFPACT_SET_IP_TTL:
4618 case OFPACT_SET_L4_SRC_PORT:
4619 case OFPACT_SET_L4_DST_PORT:
4620 case OFPACT_REG_MOVE:
4621 case OFPACT_STACK_PUSH:
4622 case OFPACT_STACK_POP:
4623 case OFPACT_DEC_TTL:
4624 case OFPACT_SET_MPLS_LABEL:
4625 case OFPACT_SET_MPLS_TC:
4626 case OFPACT_SET_MPLS_TTL:
4627 case OFPACT_DEC_MPLS_TTL:
4628 case OFPACT_PUSH_MPLS:
4629 case OFPACT_POP_MPLS:
4630 case OFPACT_POP_QUEUE:
4631 case OFPACT_FIN_TIMEOUT:
4632 case OFPACT_RESUBMIT:
4634 case OFPACT_CONJUNCTION:
4635 case OFPACT_MULTIPATH:
4639 case OFPACT_UNROLL_XLATE:
4642 case OFPACT_DEBUG_RECIRC:
4644 case OFPACT_CLEAR_ACTIONS:
4645 case OFPACT_WRITE_ACTIONS:
4646 case OFPACT_WRITE_METADATA:
4647 case OFPACT_GOTO_TABLE:
4653 ctx_trigger_freeze(ctx);
4657 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4658 struct xlate_ctx *ctx)
4660 struct flow_wildcards *wc = ctx->wc;
4661 struct flow *flow = &ctx->xin->flow;
4662 const struct ofpact *a;
4664 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4665 tnl_neigh_snoop(flow, wc, ctx->xbridge->name);
4667 /* dl_type already in the mask, not set below. */
4669 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4670 struct ofpact_controller *controller;
4671 const struct ofpact_metadata *metadata;
4672 const struct ofpact_set_field *set_field;
4673 const struct mf_field *mf;
4679 recirc_for_mpls(a, ctx);
4682 /* Check if need to store the remaining actions for later
4684 if (ctx->freezing) {
4685 freeze_unroll_actions(a, ofpact_end(ofpacts, ofpacts_len),
4693 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4694 ofpact_get_OUTPUT(a)->max_len, true);
4698 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4699 /* Group could not be found. */
4704 case OFPACT_CONTROLLER:
4705 controller = ofpact_get_CONTROLLER(a);
4706 if (controller->pause) {
4707 ctx->pause = controller;
4708 ctx->xout->slow |= SLOW_CONTROLLER;
4709 ctx_trigger_freeze(ctx);
4712 execute_controller_action(ctx, controller->max_len,
4714 controller->controller_id,
4715 controller->userdata,
4716 controller->userdata_len);
4720 case OFPACT_ENQUEUE:
4721 memset(&wc->masks.skb_priority, 0xff,
4722 sizeof wc->masks.skb_priority);
4723 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4726 case OFPACT_SET_VLAN_VID:
4727 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4728 if (flow->vlan_tci & htons(VLAN_CFI) ||
4729 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4730 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4731 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4736 case OFPACT_SET_VLAN_PCP:
4737 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4738 if (flow->vlan_tci & htons(VLAN_CFI) ||
4739 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4740 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4741 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4742 << VLAN_PCP_SHIFT) | VLAN_CFI);
4746 case OFPACT_STRIP_VLAN:
4747 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4748 flow->vlan_tci = htons(0);
4751 case OFPACT_PUSH_VLAN:
4752 /* XXX 802.1AD(QinQ) */
4753 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4754 flow->vlan_tci = htons(VLAN_CFI);
4757 case OFPACT_SET_ETH_SRC:
4758 WC_MASK_FIELD(wc, dl_src);
4759 flow->dl_src = ofpact_get_SET_ETH_SRC(a)->mac;
4762 case OFPACT_SET_ETH_DST:
4763 WC_MASK_FIELD(wc, dl_dst);
4764 flow->dl_dst = ofpact_get_SET_ETH_DST(a)->mac;
4767 case OFPACT_SET_IPV4_SRC:
4768 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4769 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4770 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4774 case OFPACT_SET_IPV4_DST:
4775 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4776 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4777 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4781 case OFPACT_SET_IP_DSCP:
4782 if (is_ip_any(flow)) {
4783 wc->masks.nw_tos |= IP_DSCP_MASK;
4784 flow->nw_tos &= ~IP_DSCP_MASK;
4785 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4789 case OFPACT_SET_IP_ECN:
4790 if (is_ip_any(flow)) {
4791 wc->masks.nw_tos |= IP_ECN_MASK;
4792 flow->nw_tos &= ~IP_ECN_MASK;
4793 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4797 case OFPACT_SET_IP_TTL:
4798 if (is_ip_any(flow)) {
4799 wc->masks.nw_ttl = 0xff;
4800 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4804 case OFPACT_SET_L4_SRC_PORT:
4805 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4806 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4807 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4808 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4812 case OFPACT_SET_L4_DST_PORT:
4813 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4814 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4815 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4816 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4820 case OFPACT_RESUBMIT:
4821 /* Freezing complicates resubmit. Some action in the flow
4822 * entry found by resubmit might trigger freezing. If that
4823 * happens, then we do not want to execute the resubmit again after
4824 * during thawing, so we want to skip back to the head of the loop
4825 * to avoid that, only adding any actions that follow the resubmit
4826 * to the frozen actions.
4828 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4831 case OFPACT_SET_TUNNEL:
4832 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4835 case OFPACT_SET_QUEUE:
4836 memset(&wc->masks.skb_priority, 0xff,
4837 sizeof wc->masks.skb_priority);
4838 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4841 case OFPACT_POP_QUEUE:
4842 memset(&wc->masks.skb_priority, 0xff,
4843 sizeof wc->masks.skb_priority);
4844 flow->skb_priority = ctx->orig_skb_priority;
4847 case OFPACT_REG_MOVE:
4848 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4851 case OFPACT_SET_FIELD:
4852 set_field = ofpact_get_SET_FIELD(a);
4853 mf = set_field->field;
4855 /* Set field action only ever overwrites packet's outermost
4856 * applicable header fields. Do nothing if no header exists. */
4857 if (mf->id == MFF_VLAN_VID) {
4858 wc->masks.vlan_tci |= htons(VLAN_CFI);
4859 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4862 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4863 /* 'dl_type' is already unwildcarded. */
4864 && !eth_type_mpls(flow->dl_type)) {
4867 /* A flow may wildcard nw_frag. Do nothing if setting a transport
4868 * header field on a packet that does not have them. */
4869 mf_mask_field_and_prereqs__(mf, &set_field->mask, wc);
4870 if (mf_are_prereqs_ok(mf, flow)) {
4871 mf_set_flow_value_masked(mf, &set_field->value,
4872 &set_field->mask, flow);
4876 case OFPACT_STACK_PUSH:
4877 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4881 case OFPACT_STACK_POP:
4882 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4886 case OFPACT_PUSH_MPLS:
4887 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4890 case OFPACT_POP_MPLS:
4891 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4894 case OFPACT_SET_MPLS_LABEL:
4895 compose_set_mpls_label_action(
4896 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4899 case OFPACT_SET_MPLS_TC:
4900 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4903 case OFPACT_SET_MPLS_TTL:
4904 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4907 case OFPACT_DEC_MPLS_TTL:
4908 if (compose_dec_mpls_ttl_action(ctx)) {
4913 case OFPACT_DEC_TTL:
4914 wc->masks.nw_ttl = 0xff;
4915 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4921 /* Nothing to do. */
4924 case OFPACT_MULTIPATH:
4925 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4929 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4932 case OFPACT_OUTPUT_REG:
4933 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4937 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4940 case OFPACT_CONJUNCTION: {
4941 /* A flow with a "conjunction" action represents part of a special
4942 * kind of "set membership match". Such a flow should not actually
4943 * get executed, but it could via, say, a "packet-out", even though
4944 * that wouldn't be useful. Log it to help debugging. */
4945 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4946 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4954 case OFPACT_UNROLL_XLATE: {
4955 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4957 /* Restore translation context data that was stored earlier. */
4958 ctx->table_id = unroll->rule_table_id;
4959 ctx->rule_cookie = unroll->rule_cookie;
4962 case OFPACT_FIN_TIMEOUT:
4963 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4964 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4967 case OFPACT_CLEAR_ACTIONS:
4968 ofpbuf_clear(&ctx->action_set);
4969 ctx->xin->flow.actset_output = OFPP_UNSET;
4970 ctx->action_set_has_group = false;
4973 case OFPACT_WRITE_ACTIONS:
4974 xlate_write_actions(ctx, ofpact_get_WRITE_ACTIONS(a));
4977 case OFPACT_WRITE_METADATA:
4978 metadata = ofpact_get_WRITE_METADATA(a);
4979 flow->metadata &= ~metadata->mask;
4980 flow->metadata |= metadata->metadata & metadata->mask;
4984 /* Not implemented yet. */
4987 case OFPACT_GOTO_TABLE: {
4988 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4990 ovs_assert(ctx->table_id < ogt->table_id);
4992 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4993 ogt->table_id, true, true);
4998 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
5002 compose_conntrack_action(ctx, ofpact_get_CT(a));
5006 /* This will be processed by compose_conntrack_action(). */
5007 ctx->ct_nat_action = ofpact_get_NAT(a);
5010 case OFPACT_DEBUG_RECIRC:
5011 ctx_trigger_freeze(ctx);
5016 /* Check if need to store this and the remaining actions for later
5018 if (!ctx->error && ctx->exit && ctx_first_frozen_action(ctx)) {
5019 freeze_unroll_actions(a, ofpact_end(ofpacts, ofpacts_len), ctx);
5026 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
5027 const struct flow *flow, ofp_port_t in_port,
5028 struct rule_dpif *rule, uint16_t tcp_flags,
5029 const struct dp_packet *packet, struct flow_wildcards *wc,
5030 struct ofpbuf *odp_actions)
5032 xin->ofproto = ofproto;
5034 xin->flow.in_port.ofp_port = in_port;
5035 xin->flow.actset_output = OFPP_UNSET;
5036 xin->packet = packet;
5037 xin->may_learn = packet != NULL;
5040 xin->ofpacts = NULL;
5041 xin->ofpacts_len = 0;
5042 xin->tcp_flags = tcp_flags;
5043 xin->resubmit_hook = NULL;
5044 xin->report_hook = NULL;
5045 xin->resubmit_stats = NULL;
5046 xin->indentation = 0;
5050 xin->odp_actions = odp_actions;
5052 /* Do recirc lookup. */
5053 xin->frozen_state = NULL;
5054 if (flow->recirc_id) {
5055 const struct recirc_id_node *node
5056 = recirc_id_node_find(flow->recirc_id);
5058 xin->frozen_state = &node->state;
5064 xlate_out_uninit(struct xlate_out *xout)
5067 recirc_refs_unref(&xout->recircs);
5071 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
5072 * into datapath actions, using 'ctx', and discards the datapath actions. */
5074 xlate_actions_for_side_effects(struct xlate_in *xin)
5076 struct xlate_out xout;
5077 enum xlate_error error;
5079 error = xlate_actions(xin, &xout);
5081 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5083 VLOG_WARN_RL(&rl, "xlate_actions failed (%s)!", xlate_strerror(error));
5086 xlate_out_uninit(&xout);
5089 static struct skb_priority_to_dscp *
5090 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
5092 struct skb_priority_to_dscp *pdscp;
5095 hash = hash_int(skb_priority, 0);
5096 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
5097 if (pdscp->skb_priority == skb_priority) {
5105 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
5108 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
5109 *dscp = pdscp ? pdscp->dscp : 0;
5110 return pdscp != NULL;
5114 count_skb_priorities(const struct xport *xport)
5116 return hmap_count(&xport->skb_priorities);
5120 clear_skb_priorities(struct xport *xport)
5122 struct skb_priority_to_dscp *pdscp;
5124 HMAP_FOR_EACH_POP (pdscp, hmap_node, &xport->skb_priorities) {
5130 actions_output_to_local_port(const struct xlate_ctx *ctx)
5132 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
5133 const struct nlattr *a;
5136 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->odp_actions->data,
5137 ctx->odp_actions->size) {
5138 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
5139 && nl_attr_get_odp_port(a) == local_odp_port) {
5146 #if defined(__linux__)
5147 /* Returns the maximum number of packets that the Linux kernel is willing to
5148 * queue up internally to certain kinds of software-implemented ports, or the
5149 * default (and rarely modified) value if it cannot be determined. */
5151 netdev_max_backlog(void)
5153 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
5154 static int max_backlog = 1000; /* The normal default value. */
5156 if (ovsthread_once_start(&once)) {
5157 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
5161 stream = fopen(filename, "r");
5163 VLOG_INFO("%s: open failed (%s)", filename, ovs_strerror(errno));
5165 if (fscanf(stream, "%d", &n) != 1) {
5166 VLOG_WARN("%s: read error", filename);
5167 } else if (n <= 100) {
5168 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
5174 ovsthread_once_done(&once);
5176 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
5182 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
5185 count_output_actions(const struct ofpbuf *odp_actions)
5187 const struct nlattr *a;
5191 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
5192 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
5198 #endif /* defined(__linux__) */
5200 /* Returns true if 'odp_actions' contains more output actions than the datapath
5201 * can reliably handle in one go. On Linux, this is the value of the
5202 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
5203 * packets that the kernel is willing to queue up for processing while the
5204 * datapath is processing a set of actions. */
5206 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
5209 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
5210 && count_output_actions(odp_actions) > netdev_max_backlog());
5212 /* OSes other than Linux might have similar limits, but we don't know how
5213 * to determine them.*/
5219 xlate_wc_init(struct xlate_ctx *ctx)
5221 flow_wildcards_init_catchall(ctx->wc);
5223 /* Some fields we consider to always be examined. */
5224 WC_MASK_FIELD(ctx->wc, in_port);
5225 WC_MASK_FIELD(ctx->wc, dl_type);
5226 if (is_ip_any(&ctx->xin->flow)) {
5227 WC_MASK_FIELD_MASK(ctx->wc, nw_frag, FLOW_NW_FRAG_MASK);
5230 if (ctx->xbridge->support.odp.recirc) {
5231 /* Always exactly match recirc_id when datapath supports
5233 WC_MASK_FIELD(ctx->wc, recirc_id);
5236 if (ctx->xbridge->netflow) {
5237 netflow_mask_wc(&ctx->xin->flow, ctx->wc);
5240 tnl_wc_init(&ctx->xin->flow, ctx->wc);
5244 xlate_wc_finish(struct xlate_ctx *ctx)
5246 /* Clear the metadata and register wildcard masks, because we won't
5247 * use non-header fields as part of the cache. */
5248 flow_wildcards_clear_non_packet_fields(ctx->wc);
5250 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
5251 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
5252 * represent these fields. The datapath interface, on the other hand,
5253 * represents them with just 8 bits each. This means that if the high
5254 * 8 bits of the masks for these fields somehow become set, then they
5255 * will get chopped off by a round trip through the datapath, and
5256 * revalidation will spot that as an inconsistency and delete the flow.
5257 * Avoid the problem here by making sure that only the low 8 bits of
5258 * either field can be unwildcarded for ICMP.
5260 if (is_icmpv4(&ctx->xin->flow, NULL) || is_icmpv6(&ctx->xin->flow, NULL)) {
5261 ctx->wc->masks.tp_src &= htons(UINT8_MAX);
5262 ctx->wc->masks.tp_dst &= htons(UINT8_MAX);
5264 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
5265 if (ctx->wc->masks.vlan_tci) {
5266 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI);
5270 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
5272 * The caller must take responsibility for eventually freeing 'xout', with
5273 * xlate_out_uninit().
5274 * Returns 'XLATE_OK' if translation was successful. In case of an error an
5275 * empty set of actions will be returned in 'xin->odp_actions' (if non-NULL),
5276 * so that most callers may ignore the return value and transparently install a
5277 * drop flow when the translation fails. */
5279 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
5281 *xout = (struct xlate_out) {
5283 .recircs = RECIRC_REFS_EMPTY_INITIALIZER,
5286 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5287 struct xbridge *xbridge = xbridge_lookup(xcfg, xin->ofproto);
5289 return XLATE_BRIDGE_NOT_FOUND;
5292 struct flow *flow = &xin->flow;
5294 union mf_subvalue stack_stub[1024 / sizeof(union mf_subvalue)];
5295 uint64_t action_set_stub[1024 / 8];
5296 uint64_t frozen_actions_stub[1024 / 8];
5297 uint64_t actions_stub[256 / 8];
5298 struct ofpbuf scratch_actions = OFPBUF_STUB_INITIALIZER(actions_stub);
5299 struct xlate_ctx ctx = {
5303 .orig_tunnel_ipv6_dst = flow_tnl_dst(&flow->tunnel),
5305 .stack = OFPBUF_STUB_INITIALIZER(stack_stub),
5309 : &(struct flow_wildcards) { .masks = { .dl_type = 0 } }),
5310 .odp_actions = xin->odp_actions ? xin->odp_actions : &scratch_actions,
5312 .indentation = xin->indentation,
5313 .depth = xin->depth,
5314 .resubmits = xin->resubmits,
5316 .in_action_set = false,
5319 .rule_cookie = OVS_BE64_MAX,
5320 .orig_skb_priority = flow->skb_priority,
5321 .sflow_n_outputs = 0,
5322 .sflow_odp_port = 0,
5323 .nf_output_iface = NF_OUT_DROP,
5329 .frozen_actions = OFPBUF_STUB_INITIALIZER(frozen_actions_stub),
5333 .conntracked = false,
5335 .ct_nat_action = NULL,
5337 .action_set_has_group = false,
5338 .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
5341 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
5342 * the packet as the datapath will treat it for output actions. Our
5343 * datapath doesn't retain tunneling information without us re-setting
5344 * it, so clear the tunnel data.
5347 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
5349 ofpbuf_reserve(ctx.odp_actions, NL_A_U32_SIZE);
5350 xlate_wc_init(&ctx);
5352 COVERAGE_INC(xlate_actions);
5354 if (xin->frozen_state) {
5355 const struct frozen_state *state = xin->frozen_state;
5357 xlate_report(&ctx, "Thawing frozen state:");
5359 if (xin->ofpacts_len > 0 || ctx.rule) {
5360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5361 const char *conflict = xin->ofpacts_len ? "actions" : "rule";
5363 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!", conflict);
5364 xlate_report(&ctx, "- Recirculation conflict (%s)!", conflict);
5365 ctx.error = XLATE_RECIRCULATION_CONFLICT;
5369 /* Set the bridge for post-recirculation processing if needed. */
5370 if (!uuid_equals(ofproto_dpif_get_uuid(ctx.xbridge->ofproto),
5371 &state->ofproto_uuid)) {
5372 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5373 const struct xbridge *new_bridge
5374 = xbridge_lookup_by_uuid(xcfg, &state->ofproto_uuid);
5376 if (OVS_UNLIKELY(!new_bridge)) {
5377 /* Drop the packet if the bridge cannot be found. */
5378 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5379 VLOG_WARN_RL(&rl, "Frozen bridge no longer exists.");
5380 xlate_report(&ctx, "- Frozen bridge no longer exists.");
5381 ctx.error = XLATE_BRIDGE_NOT_FOUND;
5384 ctx.xbridge = new_bridge;
5387 /* Set the thawed table id. Note: A table lookup is done only if there
5388 * are no frozen actions. */
5389 ctx.table_id = state->table_id;
5390 xlate_report(&ctx, "- Resuming from table %"PRIu8, ctx.table_id);
5392 if (!state->conntracked) {
5393 clear_conntrack(flow);
5396 /* Restore pipeline metadata. May change flow's in_port and other
5397 * metadata to the values that existed when freezing was triggered. */
5398 frozen_metadata_to_flow(&state->metadata, flow);
5400 /* Restore stack, if any. */
5402 ofpbuf_put(&ctx.stack, state->stack,
5403 state->n_stack * sizeof *state->stack);
5406 /* Restore mirror state. */
5407 ctx.mirrors = state->mirrors;
5409 /* Restore action set, if any. */
5410 if (state->action_set_len) {
5411 xlate_report_actions(&ctx, "- Restoring action set",
5412 state->action_set, state->action_set_len);
5414 flow->actset_output = OFPP_UNSET;
5415 xlate_write_actions__(&ctx, state->action_set,
5416 state->action_set_len);
5419 /* Restore frozen actions. If there are no actions, processing will
5420 * start with a lookup in the table set above. */
5421 xin->ofpacts = state->ofpacts;
5422 xin->ofpacts_len = state->ofpacts_len;
5423 if (state->ofpacts_len) {
5424 xlate_report_actions(&ctx, "- Restoring actions",
5425 xin->ofpacts, xin->ofpacts_len);
5427 } else if (OVS_UNLIKELY(flow->recirc_id)) {
5428 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5430 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
5432 ctx.error = XLATE_NO_RECIRCULATION_CONTEXT;
5435 /* The bridge is now known so obtain its table version. */
5436 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
5438 if (!xin->ofpacts && !ctx.rule) {
5439 ctx.rule = rule_dpif_lookup_from_table(
5440 ctx.xbridge->ofproto, ctx.tables_version, flow, ctx.wc,
5441 ctx.xin->resubmit_stats, &ctx.table_id,
5442 flow->in_port.ofp_port, true, true);
5443 if (ctx.xin->resubmit_stats) {
5444 rule_dpif_credit_stats(ctx.rule, ctx.xin->resubmit_stats);
5446 if (ctx.xin->xcache) {
5447 struct xc_entry *entry;
5449 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
5450 entry->u.rule = ctx.rule;
5451 rule_dpif_ref(ctx.rule);
5454 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
5455 ctx.xin->resubmit_hook(ctx.xin, ctx.rule, 0);
5459 /* Get the proximate input port of the packet. (If xin->frozen_state,
5460 * flow->in_port is the ultimate input port of the packet.) */
5461 struct xport *in_port = get_ofp_port(xbridge,
5462 ctx.base_flow.in_port.ofp_port);
5464 /* Tunnel stats only for not-thawed packets. */
5465 if (!xin->frozen_state && in_port && in_port->is_tunnel) {
5466 if (ctx.xin->resubmit_stats) {
5467 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
5469 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
5472 if (ctx.xin->xcache) {
5473 struct xc_entry *entry;
5475 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
5476 entry->u.dev.rx = netdev_ref(in_port->netdev);
5477 entry->u.dev.bfd = bfd_ref(in_port->bfd);
5481 if (!xin->frozen_state && process_special(&ctx, in_port)) {
5482 /* process_special() did all the processing for this packet.
5484 * We do not perform special processing on thawed packets, since that
5485 * was done before they were frozen and should not be redone. */
5486 } else if (in_port && in_port->xbundle
5487 && xbundle_mirror_out(xbridge, in_port->xbundle)) {
5488 if (ctx.xin->packet != NULL) {
5489 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5490 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5491 "%s, which is reserved exclusively for mirroring",
5492 ctx.xbridge->name, in_port->xbundle->name);
5495 /* Sampling is done on initial reception; don't redo after thawing. */
5496 unsigned int user_cookie_offset = 0;
5497 if (!xin->frozen_state) {
5498 user_cookie_offset = compose_sflow_action(&ctx);
5499 compose_ipfix_action(&ctx, ODPP_NONE);
5501 size_t sample_actions_len = ctx.odp_actions->size;
5503 if (tnl_process_ecn(flow)
5504 && (!in_port || may_receive(in_port, &ctx))) {
5505 const struct ofpact *ofpacts;
5509 ofpacts = xin->ofpacts;
5510 ofpacts_len = xin->ofpacts_len;
5511 } else if (ctx.rule) {
5512 const struct rule_actions *actions
5513 = rule_dpif_get_actions(ctx.rule);
5514 ofpacts = actions->ofpacts;
5515 ofpacts_len = actions->ofpacts_len;
5516 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
5521 mirror_ingress_packet(&ctx);
5522 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
5527 /* We've let OFPP_NORMAL and the learning action look at the
5528 * packet, so cancel all actions and freezing if forwarding is
5530 if (in_port && (!xport_stp_forward_state(in_port) ||
5531 !xport_rstp_forward_state(in_port))) {
5532 ctx.odp_actions->size = sample_actions_len;
5533 ctx_cancel_freeze(&ctx);
5534 ofpbuf_clear(&ctx.action_set);
5537 if (!ctx.freezing) {
5538 xlate_action_set(&ctx);
5541 finish_freezing(&ctx);
5545 /* Output only fully processed packets. */
5547 && xbridge->has_in_band
5548 && in_band_must_output_to_local_port(flow)
5549 && !actions_output_to_local_port(&ctx)) {
5550 compose_output_action(&ctx, OFPP_LOCAL, NULL);
5553 if (user_cookie_offset) {
5554 fix_sflow_action(&ctx, user_cookie_offset);
5558 if (nl_attr_oversized(ctx.odp_actions->size)) {
5559 /* These datapath actions are too big for a Netlink attribute, so we
5560 * can't hand them to the kernel directly. dpif_execute() can execute
5561 * them one by one with help, so just mark the result as SLOW_ACTION to
5562 * prevent the flow from being installed. */
5563 COVERAGE_INC(xlate_actions_oversize);
5564 ctx.xout->slow |= SLOW_ACTION;
5565 } else if (too_many_output_actions(ctx.odp_actions)) {
5566 COVERAGE_INC(xlate_actions_too_many_output);
5567 ctx.xout->slow |= SLOW_ACTION;
5570 /* Do netflow only for packets on initial reception, that are not sent to
5571 * the controller. We consider packets sent to the controller to be part
5572 * of the control plane rather than the data plane. */
5573 if (!xin->frozen_state
5575 && !(xout->slow & SLOW_CONTROLLER)) {
5576 if (ctx.xin->resubmit_stats) {
5577 netflow_flow_update(xbridge->netflow, flow,
5578 ctx.nf_output_iface,
5579 ctx.xin->resubmit_stats);
5581 if (ctx.xin->xcache) {
5582 struct xc_entry *entry;
5584 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5585 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5586 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5587 entry->u.nf.iface = ctx.nf_output_iface;
5591 xlate_wc_finish(&ctx);
5594 ofpbuf_uninit(&ctx.stack);
5595 ofpbuf_uninit(&ctx.action_set);
5596 ofpbuf_uninit(&ctx.frozen_actions);
5597 ofpbuf_uninit(&scratch_actions);
5599 /* Make sure we return a "drop flow" in case of an error. */
5602 if (xin->odp_actions) {
5603 ofpbuf_clear(xin->odp_actions);
5610 xlate_resume(struct ofproto_dpif *ofproto,
5611 const struct ofputil_packet_in_private *pin,
5612 struct ofpbuf *odp_actions,
5613 enum slow_path_reason *slow)
5615 struct dp_packet packet;
5616 dp_packet_use_const(&packet, pin->public.packet,
5617 pin->public.packet_len);
5620 flow_extract(&packet, &flow);
5622 struct xlate_in xin;
5623 xlate_in_init(&xin, ofproto, &flow, 0, NULL, ntohs(flow.tcp_flags),
5624 &packet, NULL, odp_actions);
5626 struct ofpact_note noop;
5627 ofpact_init_NOTE(&noop);
5630 bool any_actions = pin->actions_len > 0;
5631 struct frozen_state state = {
5632 .table_id = 0, /* Not the table where NXAST_PAUSE was executed. */
5633 .ofproto_uuid = pin->bridge,
5634 .stack = pin->stack,
5635 .n_stack = pin->n_stack,
5636 .mirrors = pin->mirrors,
5637 .conntracked = pin->conntracked,
5639 /* When there are no actions, xlate_actions() will search the flow
5640 * table. We don't want it to do that (we want it to resume), so
5641 * supply a no-op action if there aren't any.
5643 * (We can't necessarily avoid translating actions entirely if there
5644 * aren't any actions, because there might be some finishing-up to do
5645 * at the end of the pipeline, and we don't check for those
5647 .ofpacts = any_actions ? pin->actions : &noop.ofpact,
5648 .ofpacts_len = any_actions ? pin->actions_len : sizeof noop,
5650 .action_set = pin->action_set,
5651 .action_set_len = pin->action_set_len,
5653 frozen_metadata_from_flow(&state.metadata,
5654 &pin->public.flow_metadata.flow);
5655 xin.frozen_state = &state;
5657 struct xlate_out xout;
5658 enum xlate_error error = xlate_actions(&xin, &xout);
5660 xlate_out_uninit(&xout);
5662 /* xlate_actions() can generate a number of errors, but only
5663 * XLATE_BRIDGE_NOT_FOUND really stands out to me as one that we should be
5664 * sure to report over OpenFlow. The others could come up in packet-outs
5665 * or regular flow translation and I don't think that it's going to be too
5666 * useful to report them to the controller. */
5667 return error == XLATE_BRIDGE_NOT_FOUND ? OFPERR_NXR_STALE : 0;
5670 /* Sends 'packet' out 'ofport'.
5671 * May modify 'packet'.
5672 * Returns 0 if successful, otherwise a positive errno value. */
5674 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5676 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5677 struct xport *xport;
5678 struct ofpact_output output;
5681 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5682 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5683 flow_extract(packet, &flow);
5684 flow.in_port.ofp_port = OFPP_NONE;
5686 xport = xport_lookup(xcfg, ofport);
5690 output.port = xport->ofp_port;
5693 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5694 &output.ofpact, sizeof output,
5698 struct xlate_cache *
5699 xlate_cache_new(void)
5701 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5703 ofpbuf_init(&xcache->entries, 512);
5707 static struct xc_entry *
5708 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5710 struct xc_entry *entry;
5712 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5719 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5721 if (entry->u.dev.tx) {
5722 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5724 if (entry->u.dev.rx) {
5725 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5727 if (entry->u.dev.bfd) {
5728 bfd_account_rx(entry->u.dev.bfd, stats);
5733 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5735 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5736 struct xbridge *xbridge;
5737 struct xbundle *xbundle;
5738 struct flow_wildcards wc;
5740 xbridge = xbridge_lookup(xcfg, ofproto);
5745 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5751 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5754 /* Push stats and perform side effects of flow translation. */
5756 xlate_push_stats(struct xlate_cache *xcache,
5757 const struct dpif_flow_stats *stats)
5759 struct xc_entry *entry;
5760 struct ofpbuf entries = xcache->entries;
5761 struct eth_addr dmac;
5763 if (!stats->n_packets) {
5767 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5768 switch (entry->type) {
5770 rule_dpif_credit_stats(entry->u.rule, stats);
5773 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5774 entry->u.bond.vid, stats->n_bytes);
5777 xlate_cache_netdev(entry, stats);
5780 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5781 entry->u.nf.iface, stats);
5784 mirror_update_stats(entry->u.mirror.mbridge,
5785 entry->u.mirror.mirrors,
5786 stats->n_packets, stats->n_bytes);
5789 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5792 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5793 entry->u.normal.vlan);
5795 case XC_FIN_TIMEOUT:
5796 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5797 entry->u.fin.idle, entry->u.fin.hard);
5800 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5804 /* Lookup neighbor to avoid timeout. */
5805 tnl_neigh_lookup(entry->u.tnl_neigh_cache.br_name,
5806 &entry->u.tnl_neigh_cache.d_ipv6, &dmac);
5815 xlate_dev_unref(struct xc_entry *entry)
5817 if (entry->u.dev.tx) {
5818 netdev_close(entry->u.dev.tx);
5820 if (entry->u.dev.rx) {
5821 netdev_close(entry->u.dev.rx);
5823 if (entry->u.dev.bfd) {
5824 bfd_unref(entry->u.dev.bfd);
5829 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5831 netflow_flow_clear(netflow, flow);
5832 netflow_unref(netflow);
5837 xlate_cache_clear(struct xlate_cache *xcache)
5839 struct xc_entry *entry;
5840 struct ofpbuf entries;
5846 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5847 switch (entry->type) {
5849 rule_dpif_unref(entry->u.rule);
5852 free(entry->u.bond.flow);
5853 bond_unref(entry->u.bond.bond);
5856 xlate_dev_unref(entry);
5859 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5862 mbridge_unref(entry->u.mirror.mbridge);
5865 free(entry->u.learn.fm);
5866 ofpbuf_delete(entry->u.learn.ofpacts);
5869 free(entry->u.normal.flow);
5871 case XC_FIN_TIMEOUT:
5872 /* 'u.fin.rule' is always already held as a XC_RULE, which
5873 * has already released it's reference above. */
5876 group_dpif_unref(entry->u.group.group);
5885 ofpbuf_clear(&xcache->entries);
5889 xlate_cache_delete(struct xlate_cache *xcache)
5891 xlate_cache_clear(xcache);
5892 ofpbuf_uninit(&xcache->entries);