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>
25 #include "tnl-neigh-cache.h"
30 #include "byte-order.h"
34 #include "dp-packet.h"
36 #include "dynamic-string.h"
42 #include "mac-learning.h"
43 #include "mcast-snooping.h"
44 #include "meta-flow.h"
45 #include "multipath.h"
46 #include "netdev-vport.h"
49 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofproto/ofproto-dpif-ipfix.h"
52 #include "ofproto/ofproto-dpif-mirror.h"
53 #include "ofproto/ofproto-dpif-monitor.h"
54 #include "ofproto/ofproto-dpif-sflow.h"
55 #include "ofproto/ofproto-dpif.h"
56 #include "ofproto/ofproto-provider.h"
58 #include "ovs-router.h"
59 #include "tnl-ports.h"
61 #include "openvswitch/vlog.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
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
72 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
75 /* Maximum number of resubmit actions in a flow translation, whether they are
76 * recursive or not. */
77 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
80 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
81 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
83 struct ovs_list xbundles; /* Owned xbundles. */
84 struct hmap xports; /* Indexed by ofp_port. */
86 char *name; /* Name used in log messages. */
87 struct dpif *dpif; /* Datapath interface. */
88 struct mac_learning *ml; /* Mac learning handle. */
89 struct mcast_snooping *ms; /* Multicast Snooping handle. */
90 struct mbridge *mbridge; /* Mirroring. */
91 struct dpif_sflow *sflow; /* SFlow handle, or null. */
92 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
93 struct netflow *netflow; /* Netflow handle, or null. */
94 struct stp *stp; /* STP or null if disabled. */
95 struct rstp *rstp; /* RSTP or null if disabled. */
97 bool has_in_band; /* Bridge has in band control? */
98 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
100 /* Datapath feature support. */
101 struct dpif_backer_support support;
105 struct hmap_node hmap_node; /* In global 'xbundles' map. */
106 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
108 struct ovs_list list_node; /* In parent 'xbridges' list. */
109 struct xbridge *xbridge; /* Parent xbridge. */
111 struct ovs_list xports; /* Contains "struct xport"s. */
113 char *name; /* Name used in log messages. */
114 struct bond *bond; /* Nonnull iff more than one port. */
115 struct lacp *lacp; /* LACP handle or null. */
117 enum port_vlan_mode vlan_mode; /* VLAN mode. */
118 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
119 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
120 * NULL if all VLANs are trunked. */
121 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
122 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
126 struct hmap_node hmap_node; /* Node in global 'xports' map. */
127 struct ofport_dpif *ofport; /* Key in global 'xports map. */
129 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
130 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
132 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
134 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
135 struct xbundle *xbundle; /* Parent xbundle or null. */
137 struct netdev *netdev; /* 'ofport''s netdev. */
139 struct xbridge *xbridge; /* Parent bridge. */
140 struct xport *peer; /* Patch port peer or null. */
142 enum ofputil_port_config config; /* OpenFlow port configuration. */
143 enum ofputil_port_state state; /* OpenFlow port state. */
144 int stp_port_no; /* STP port number or -1 if not in use. */
145 struct rstp_port *rstp_port; /* RSTP port or null. */
147 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
149 bool may_enable; /* May be enabled in bonds. */
150 bool is_tunnel; /* Is a tunnel port. */
152 struct cfm *cfm; /* CFM handle or null. */
153 struct bfd *bfd; /* BFD handle or null. */
154 struct lldp *lldp; /* LLDP handle or null. */
158 struct xlate_in *xin;
159 struct xlate_out *xout;
161 const struct xbridge *xbridge;
163 /* Flow tables version at the beginning of the translation. */
164 cls_version_t tables_version;
166 /* Flow at the last commit. */
167 struct flow base_flow;
169 /* Tunnel IP destination address as received. This is stored separately
170 * as the base_flow.tunnel is cleared on init to reflect the datapath
171 * behavior. Used to make sure not to send tunneled output to ourselves,
172 * which might lead to an infinite loop. This could happen easily
173 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
174 * actually set the tun_dst field. */
175 struct in6_addr orig_tunnel_ipv6_dst;
177 /* Stack for the push and pop actions. Each stack element is of type
178 * "union mf_subvalue". */
181 /* The rule that we are currently translating, or NULL. */
182 struct rule_dpif *rule;
184 /* Flow translation populates this with wildcards relevant in translation.
185 * When 'xin->wc' is nonnull, this is the same pointer. When 'xin->wc' is
186 * null, this is a pointer to uninitialized scratch memory. This allows
187 * code to blindly write to 'ctx->wc' without worrying about whether the
188 * caller really wants wildcards. */
189 struct flow_wildcards *wc;
191 /* Output buffer for datapath actions. When 'xin->odp_actions' is nonnull,
192 * this is the same pointer. When 'xin->odp_actions' is null, this points
193 * to a scratch ofpbuf. This allows code to add actions to
194 * 'ctx->odp_actions' without worrying about whether the caller really
196 struct ofpbuf *odp_actions;
198 /* Resubmit statistics, via xlate_table_action(). */
199 int recurse; /* Current resubmit nesting depth. */
200 int resubmits; /* Total number of resubmits. */
201 bool in_group; /* Currently translating ofgroup, if true. */
202 bool in_action_set; /* Currently translating action_set, if true. */
204 uint8_t table_id; /* OpenFlow table ID where flow was found. */
205 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
206 uint32_t orig_skb_priority; /* Priority when packet arrived. */
207 uint32_t sflow_n_outputs; /* Number of output ports. */
208 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
209 ofp_port_t nf_output_iface; /* Output interface index for NetFlow. */
210 bool exit; /* No further actions should be processed. */
211 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
213 /* Freezing Translation
214 * ====================
216 * At some point during translation, the code may recognize the need to halt
217 * and checkpoint the translation in a way that it can be restarted again
218 * later. We call the checkpointing process "freezing" and the restarting
221 * The use cases for freezing are:
223 * - "Recirculation", where the translation process discovers that it
224 * doesn't have enough information to complete translation without
225 * actually executing the actions that have already been translated,
226 * which provides the additionally needed information. In these
227 * situations, translation freezes translation and assigns the frozen
228 * data a unique "recirculation ID", which it associates with the data
229 * in a table in userspace (see ofproto-dpif-rid.h). It also adds a
230 * OVS_ACTION_ATTR_RECIRC action specifying that ID to the datapath
231 * actions. When a packet hits that action, the datapath looks its
232 * flow up again using the ID. If there's a miss, it comes back to
233 * userspace, which find the recirculation table entry for the ID,
234 * thaws the associated frozen data, and continues translation from
235 * that point given the additional information that is now known.
237 * The archetypal example is MPLS. As MPLS is implemented in
238 * OpenFlow, the protocol that follows the last MPLS label becomes
239 * known only when that label is popped by an OpenFlow action. That
240 * means that Open vSwitch can't extract the headers beyond the MPLS
241 * labels until the pop action is executed. Thus, at that point
242 * translation uses the recirculation process to extract the headers
243 * beyond the MPLS labels.
245 * (OVS also uses OVS_ACTION_ATTR_RECIRC to implement hashing for
246 * output to bonds. OVS pre-populates all the datapath flows for bond
247 * output in the datapath, though, which means that the elaborate
248 * process of coming back to userspace for a second round of
249 * translation isn't needed, and so bonds don't follow the above
253 * The main problem of freezing translation is preserving state, so that
254 * when the translation is thawed later it resumes from where it left off,
255 * without disruption. In particular, actions must be preserved as follows:
257 * - If we're freezing because an action needed more information, the
258 * action that prompted it.
260 * - Any actions remaining to be translated within the current flow.
262 * - If translation was frozen within a NXAST_RESUBMIT, then any actions
263 * following the resubmit action. Resubmit actions can be nested, so
264 * this has to go all the way up the control stack.
266 * - The OpenFlow 1.1+ action set.
268 * State that actions and flow table lookups can depend on, such as the
269 * following, must also be preserved:
271 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
273 * - The stack used by NXAST_STACK_PUSH and NXAST_STACK_POP actions.
275 * - The table ID and cookie of the flow being translated at each level
276 * of the control stack, because these can become visible through
277 * OFPAT_CONTROLLER actions (and other ways).
279 * Translation allows for the control of this state preservation via these
280 * members. When a need to freeze translation is identified, the
281 * translation process:
283 * 1. Sets 'freezing' to true.
285 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
286 * translation process.
288 * 3. Adds an OFPACT_UNROLL_XLATE action to 'frozen_actions', and points
289 * frozen_actions.header to the action to make it easy to find it later.
290 * This action holds the current table ID and cookie so that they can be
291 * restored during a post-recirculation upcall translation.
293 * 4. Adds the action that prompted recirculation and any actions following
294 * it within the same flow to 'frozen_actions', so that they can be
295 * executed during a post-recirculation upcall translation.
299 * 6. The action that prompted recirculation might be nested in a stack of
300 * nested "resubmit"s that have actions remaining. Each of these notices
301 * that we're exiting and freezing and responds by adding more
302 * OFPACT_UNROLL_XLATE actions to 'frozen_actions', as necessary,
303 * followed by any actions that were yet unprocessed.
305 * If we're freezing because of recirculation, the caller generates a
306 * recirculation ID and associates all the state produced by this process
307 * with it. For post-recirculation upcall translation, the caller passes it
308 * back in for the new translation to execute. The process yielded a set of
309 * ofpacts that can be translated directly, so it is not much of a special
310 * case at that point.
313 struct ofpbuf frozen_actions;
315 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
316 * This is a trigger for recirculation in cases where translating an action
317 * or looking up a flow requires access to the fields of the packet after
318 * the MPLS label stack that was originally present. */
321 /* True if conntrack has been performed on this packet during processing
322 * on the current bridge. This is used to determine whether conntrack
323 * state from the datapath should be honored after thawing. */
326 /* Pointer to an embedded NAT action in a conntrack action, or NULL. */
327 struct ofpact_nat *ct_nat_action;
329 /* OpenFlow 1.1+ action set.
331 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
332 * When translation is otherwise complete, ofpacts_execute_action_set()
333 * converts it to a set of "struct ofpact"s that can be translated into
334 * datapath actions. */
335 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
336 struct ofpbuf action_set; /* Action set. */
338 enum xlate_error error; /* Translation failed. */
341 const char *xlate_strerror(enum xlate_error error)
346 case XLATE_BRIDGE_NOT_FOUND:
347 return "Bridge not found";
348 case XLATE_RECURSION_TOO_DEEP:
349 return "Recursion too deep";
350 case XLATE_TOO_MANY_RESUBMITS:
351 return "Too many resubmits";
352 case XLATE_STACK_TOO_DEEP:
353 return "Stack too deep";
354 case XLATE_NO_RECIRCULATION_CONTEXT:
355 return "No recirculation context";
356 case XLATE_RECIRCULATION_CONFLICT:
357 return "Recirculation conflict";
358 case XLATE_TOO_MANY_MPLS_LABELS:
359 return "Too many MPLS labels";
361 return "Unknown error";
364 static void xlate_action_set(struct xlate_ctx *ctx);
365 static void xlate_commit_actions(struct xlate_ctx *ctx);
368 ctx_trigger_freeze(struct xlate_ctx *ctx)
371 ctx->freezing = true;
375 ctx_first_frozen_action(const struct xlate_ctx *ctx)
377 return !ctx->frozen_actions.size;
381 ctx_cancel_freeze(struct xlate_ctx *ctx)
384 ctx->freezing = false;
385 ofpbuf_clear(&ctx->frozen_actions);
386 ctx->frozen_actions.header = NULL;
390 static void compose_recirculate_action(struct xlate_ctx *ctx);
392 /* A controller may use OFPP_NONE as the ingress port to indicate that
393 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
394 * when an input bundle is needed for validation (e.g., mirroring or
395 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
396 * any 'port' structs, so care must be taken when dealing with it. */
397 static struct xbundle ofpp_none_bundle = {
399 .vlan_mode = PORT_VLAN_TRUNK
402 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
403 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
404 * traffic egressing the 'ofport' with that priority should be marked with. */
405 struct skb_priority_to_dscp {
406 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
407 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
409 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
425 /* xlate_cache entries hold enough information to perform the side effects of
426 * xlate_actions() for a rule, without needing to perform rule translation
427 * from scratch. The primary usage of these is to submit statistics to objects
428 * that a flow relates to, although they may be used for other effects as well
429 * (for instance, refreshing hard timeouts for learned flows). */
433 struct rule_dpif *rule;
440 struct netflow *netflow;
445 struct mbridge *mbridge;
446 mirror_mask_t mirrors;
454 struct ofproto_dpif *ofproto;
455 struct ofputil_flow_mod *fm;
456 struct ofpbuf *ofpacts;
459 struct ofproto_dpif *ofproto;
464 struct rule_dpif *rule;
469 struct group_dpif *group;
470 struct ofputil_bucket *bucket;
473 char br_name[IFNAMSIZ];
474 struct in6_addr d_ipv6;
479 #define XC_ENTRY_FOR_EACH(ENTRY, ENTRIES, XCACHE) \
480 ENTRIES = XCACHE->entries; \
481 for (ENTRY = ofpbuf_try_pull(&ENTRIES, sizeof *ENTRY); \
483 ENTRY = ofpbuf_try_pull(&ENTRIES, sizeof *ENTRY))
486 struct ofpbuf entries;
489 /* Xlate config contains hash maps of all bridges, bundles and ports.
490 * Xcfgp contains the pointer to the current xlate configuration.
491 * When the main thread needs to change the configuration, it copies xcfgp to
492 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
493 * does not block handler and revalidator threads. */
495 struct hmap xbridges;
496 struct hmap xbundles;
499 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
500 static struct xlate_cfg *new_xcfg = NULL;
502 static bool may_receive(const struct xport *, struct xlate_ctx *);
503 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
505 static void xlate_normal(struct xlate_ctx *);
506 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
507 OVS_PRINTF_FORMAT(2, 3);
508 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
509 uint8_t table_id, bool may_packet_in,
510 bool honor_table_miss);
511 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
512 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
513 static void output_normal(struct xlate_ctx *, const struct xbundle *,
516 /* Optional bond recirculation parameter to compose_output_action(). */
517 struct xlate_bond_recirc {
518 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
519 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
520 uint32_t hash_basis; /* Compute hash for recirc before. */
523 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
524 const struct xlate_bond_recirc *xr);
526 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
527 const struct ofproto_dpif *);
528 static struct xbridge *xbridge_lookup_by_uuid(struct xlate_cfg *,
529 const struct uuid *);
530 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
531 const struct ofbundle *);
532 static struct xport *xport_lookup(struct xlate_cfg *,
533 const struct ofport_dpif *);
534 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
535 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
536 uint32_t skb_priority);
537 static void clear_skb_priorities(struct xport *);
538 static size_t count_skb_priorities(const struct xport *);
539 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
542 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
544 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
545 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
546 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
547 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
548 const struct mac_learning *, struct stp *,
549 struct rstp *, const struct mcast_snooping *,
550 const struct mbridge *,
551 const struct dpif_sflow *,
552 const struct dpif_ipfix *,
553 const struct netflow *,
554 bool forward_bpdu, bool has_in_band,
555 const struct dpif_backer_support *);
556 static void xlate_xbundle_set(struct xbundle *xbundle,
557 enum port_vlan_mode vlan_mode, int vlan,
558 unsigned long *trunks, bool use_priority_tags,
559 const struct bond *bond, const struct lacp *lacp,
561 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
562 const struct netdev *netdev, const struct cfm *cfm,
563 const struct bfd *bfd, const struct lldp *lldp,
564 int stp_port_no, const struct rstp_port *rstp_port,
565 enum ofputil_port_config config,
566 enum ofputil_port_state state, bool is_tunnel,
568 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
569 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
570 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
571 static void xlate_xbridge_copy(struct xbridge *);
572 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
573 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
575 static void xlate_xcfg_free(struct xlate_cfg *);
578 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
580 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
583 va_start(args, format);
584 ctx->xin->report_hook(ctx->xin, ctx->recurse, format, args);
589 static struct vlog_rate_limit error_report_rl = VLOG_RATE_LIMIT_INIT(1, 5);
591 #define XLATE_REPORT_ERROR(CTX, ...) \
593 if (OVS_UNLIKELY((CTX)->xin->report_hook)) { \
594 xlate_report(CTX, __VA_ARGS__); \
596 VLOG_ERR_RL(&error_report_rl, __VA_ARGS__); \
601 xlate_report_actions(struct xlate_ctx *ctx, const char *title,
602 const struct ofpact *ofpacts, size_t ofpacts_len)
604 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
605 struct ds s = DS_EMPTY_INITIALIZER;
606 ofpacts_format(ofpacts, ofpacts_len, &s);
607 xlate_report(ctx, "%s: %s", title, ds_cstr(&s));
613 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
615 list_init(&xbridge->xbundles);
616 hmap_init(&xbridge->xports);
617 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
618 hash_pointer(xbridge->ofproto, 0));
622 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
624 list_init(&xbundle->xports);
625 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
626 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
627 hash_pointer(xbundle->ofbundle, 0));
631 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
633 hmap_init(&xport->skb_priorities);
634 hmap_insert(&xcfg->xports, &xport->hmap_node,
635 hash_pointer(xport->ofport, 0));
636 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
637 hash_ofp_port(xport->ofp_port));
641 xlate_xbridge_set(struct xbridge *xbridge,
643 const struct mac_learning *ml, struct stp *stp,
644 struct rstp *rstp, const struct mcast_snooping *ms,
645 const struct mbridge *mbridge,
646 const struct dpif_sflow *sflow,
647 const struct dpif_ipfix *ipfix,
648 const struct netflow *netflow,
649 bool forward_bpdu, bool has_in_band,
650 const struct dpif_backer_support *support)
652 if (xbridge->ml != ml) {
653 mac_learning_unref(xbridge->ml);
654 xbridge->ml = mac_learning_ref(ml);
657 if (xbridge->ms != ms) {
658 mcast_snooping_unref(xbridge->ms);
659 xbridge->ms = mcast_snooping_ref(ms);
662 if (xbridge->mbridge != mbridge) {
663 mbridge_unref(xbridge->mbridge);
664 xbridge->mbridge = mbridge_ref(mbridge);
667 if (xbridge->sflow != sflow) {
668 dpif_sflow_unref(xbridge->sflow);
669 xbridge->sflow = dpif_sflow_ref(sflow);
672 if (xbridge->ipfix != ipfix) {
673 dpif_ipfix_unref(xbridge->ipfix);
674 xbridge->ipfix = dpif_ipfix_ref(ipfix);
677 if (xbridge->stp != stp) {
678 stp_unref(xbridge->stp);
679 xbridge->stp = stp_ref(stp);
682 if (xbridge->rstp != rstp) {
683 rstp_unref(xbridge->rstp);
684 xbridge->rstp = rstp_ref(rstp);
687 if (xbridge->netflow != netflow) {
688 netflow_unref(xbridge->netflow);
689 xbridge->netflow = netflow_ref(netflow);
692 xbridge->dpif = dpif;
693 xbridge->forward_bpdu = forward_bpdu;
694 xbridge->has_in_band = has_in_band;
695 xbridge->support = *support;
699 xlate_xbundle_set(struct xbundle *xbundle,
700 enum port_vlan_mode vlan_mode, int vlan,
701 unsigned long *trunks, bool use_priority_tags,
702 const struct bond *bond, const struct lacp *lacp,
705 ovs_assert(xbundle->xbridge);
707 xbundle->vlan_mode = vlan_mode;
708 xbundle->vlan = vlan;
709 xbundle->trunks = trunks;
710 xbundle->use_priority_tags = use_priority_tags;
711 xbundle->floodable = floodable;
713 if (xbundle->bond != bond) {
714 bond_unref(xbundle->bond);
715 xbundle->bond = bond_ref(bond);
718 if (xbundle->lacp != lacp) {
719 lacp_unref(xbundle->lacp);
720 xbundle->lacp = lacp_ref(lacp);
725 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
726 const struct netdev *netdev, const struct cfm *cfm,
727 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
728 const struct rstp_port* rstp_port,
729 enum ofputil_port_config config, enum ofputil_port_state state,
730 bool is_tunnel, bool may_enable)
732 xport->config = config;
733 xport->state = state;
734 xport->stp_port_no = stp_port_no;
735 xport->is_tunnel = is_tunnel;
736 xport->may_enable = may_enable;
737 xport->odp_port = odp_port;
739 if (xport->rstp_port != rstp_port) {
740 rstp_port_unref(xport->rstp_port);
741 xport->rstp_port = rstp_port_ref(rstp_port);
744 if (xport->cfm != cfm) {
745 cfm_unref(xport->cfm);
746 xport->cfm = cfm_ref(cfm);
749 if (xport->bfd != bfd) {
750 bfd_unref(xport->bfd);
751 xport->bfd = bfd_ref(bfd);
754 if (xport->lldp != lldp) {
755 lldp_unref(xport->lldp);
756 xport->lldp = lldp_ref(lldp);
759 if (xport->netdev != netdev) {
760 netdev_close(xport->netdev);
761 xport->netdev = netdev_ref(netdev);
766 xlate_xbridge_copy(struct xbridge *xbridge)
768 struct xbundle *xbundle;
770 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
771 new_xbridge->ofproto = xbridge->ofproto;
772 new_xbridge->name = xstrdup(xbridge->name);
773 xlate_xbridge_init(new_xcfg, new_xbridge);
775 xlate_xbridge_set(new_xbridge,
776 xbridge->dpif, xbridge->ml, xbridge->stp,
777 xbridge->rstp, xbridge->ms, xbridge->mbridge,
778 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
779 xbridge->forward_bpdu, xbridge->has_in_band,
781 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
782 xlate_xbundle_copy(new_xbridge, xbundle);
785 /* Copy xports which are not part of a xbundle */
786 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
787 if (!xport->xbundle) {
788 xlate_xport_copy(new_xbridge, NULL, xport);
794 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
797 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
798 new_xbundle->ofbundle = xbundle->ofbundle;
799 new_xbundle->xbridge = xbridge;
800 new_xbundle->name = xstrdup(xbundle->name);
801 xlate_xbundle_init(new_xcfg, new_xbundle);
803 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
804 xbundle->vlan, xbundle->trunks,
805 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
807 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
808 xlate_xport_copy(xbridge, new_xbundle, xport);
813 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
816 struct skb_priority_to_dscp *pdscp, *new_pdscp;
817 struct xport *new_xport = xzalloc(sizeof *xport);
818 new_xport->ofport = xport->ofport;
819 new_xport->ofp_port = xport->ofp_port;
820 new_xport->xbridge = xbridge;
821 xlate_xport_init(new_xcfg, new_xport);
823 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
824 xport->bfd, xport->lldp, xport->stp_port_no,
825 xport->rstp_port, xport->config, xport->state,
826 xport->is_tunnel, xport->may_enable);
829 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
831 new_xport->peer = peer;
832 new_xport->peer->peer = new_xport;
837 new_xport->xbundle = xbundle;
838 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
841 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
842 new_pdscp = xmalloc(sizeof *pdscp);
843 new_pdscp->skb_priority = pdscp->skb_priority;
844 new_pdscp->dscp = pdscp->dscp;
845 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
846 hash_int(new_pdscp->skb_priority, 0));
850 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
851 * configuration in xcfgp.
853 * This needs to be called after editing the xlate configuration.
855 * Functions that edit the new xlate configuration are
856 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
862 * edit_xlate_configuration();
864 * xlate_txn_commit(); */
866 xlate_txn_commit(void)
868 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
870 ovsrcu_set(&xcfgp, new_xcfg);
871 ovsrcu_synchronize();
872 xlate_xcfg_free(xcfg);
876 /* Copies the current xlate configuration in xcfgp to new_xcfg.
878 * This needs to be called prior to editing the xlate configuration. */
880 xlate_txn_start(void)
882 struct xbridge *xbridge;
883 struct xlate_cfg *xcfg;
885 ovs_assert(!new_xcfg);
887 new_xcfg = xmalloc(sizeof *new_xcfg);
888 hmap_init(&new_xcfg->xbridges);
889 hmap_init(&new_xcfg->xbundles);
890 hmap_init(&new_xcfg->xports);
892 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
897 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
898 xlate_xbridge_copy(xbridge);
904 xlate_xcfg_free(struct xlate_cfg *xcfg)
906 struct xbridge *xbridge, *next_xbridge;
912 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
913 xlate_xbridge_remove(xcfg, xbridge);
916 hmap_destroy(&xcfg->xbridges);
917 hmap_destroy(&xcfg->xbundles);
918 hmap_destroy(&xcfg->xports);
923 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
925 const struct mac_learning *ml, struct stp *stp,
926 struct rstp *rstp, const struct mcast_snooping *ms,
927 const struct mbridge *mbridge,
928 const struct dpif_sflow *sflow,
929 const struct dpif_ipfix *ipfix,
930 const struct netflow *netflow,
931 bool forward_bpdu, bool has_in_band,
932 const struct dpif_backer_support *support)
934 struct xbridge *xbridge;
936 ovs_assert(new_xcfg);
938 xbridge = xbridge_lookup(new_xcfg, ofproto);
940 xbridge = xzalloc(sizeof *xbridge);
941 xbridge->ofproto = ofproto;
943 xlate_xbridge_init(new_xcfg, xbridge);
947 xbridge->name = xstrdup(name);
949 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
950 netflow, forward_bpdu, has_in_band, support);
954 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
956 struct xbundle *xbundle, *next_xbundle;
957 struct xport *xport, *next_xport;
963 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
964 xlate_xport_remove(xcfg, xport);
967 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
968 xlate_xbundle_remove(xcfg, xbundle);
971 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
972 mac_learning_unref(xbridge->ml);
973 mcast_snooping_unref(xbridge->ms);
974 mbridge_unref(xbridge->mbridge);
975 dpif_sflow_unref(xbridge->sflow);
976 dpif_ipfix_unref(xbridge->ipfix);
977 stp_unref(xbridge->stp);
978 rstp_unref(xbridge->rstp);
979 hmap_destroy(&xbridge->xports);
985 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
987 struct xbridge *xbridge;
989 ovs_assert(new_xcfg);
991 xbridge = xbridge_lookup(new_xcfg, ofproto);
992 xlate_xbridge_remove(new_xcfg, xbridge);
996 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
997 const char *name, enum port_vlan_mode vlan_mode, int vlan,
998 unsigned long *trunks, bool use_priority_tags,
999 const struct bond *bond, const struct lacp *lacp,
1002 struct xbundle *xbundle;
1004 ovs_assert(new_xcfg);
1006 xbundle = xbundle_lookup(new_xcfg, ofbundle);
1008 xbundle = xzalloc(sizeof *xbundle);
1009 xbundle->ofbundle = ofbundle;
1010 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
1012 xlate_xbundle_init(new_xcfg, xbundle);
1015 free(xbundle->name);
1016 xbundle->name = xstrdup(name);
1018 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
1019 use_priority_tags, bond, lacp, floodable);
1023 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
1025 struct xport *xport;
1031 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
1032 xport->xbundle = NULL;
1035 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
1036 list_remove(&xbundle->list_node);
1037 bond_unref(xbundle->bond);
1038 lacp_unref(xbundle->lacp);
1039 free(xbundle->name);
1044 xlate_bundle_remove(struct ofbundle *ofbundle)
1046 struct xbundle *xbundle;
1048 ovs_assert(new_xcfg);
1050 xbundle = xbundle_lookup(new_xcfg, ofbundle);
1051 xlate_xbundle_remove(new_xcfg, xbundle);
1055 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
1056 struct ofport_dpif *ofport, ofp_port_t ofp_port,
1057 odp_port_t odp_port, const struct netdev *netdev,
1058 const struct cfm *cfm, const struct bfd *bfd,
1059 const struct lldp *lldp, struct ofport_dpif *peer,
1060 int stp_port_no, const struct rstp_port *rstp_port,
1061 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
1062 enum ofputil_port_config config,
1063 enum ofputil_port_state state, bool is_tunnel,
1067 struct xport *xport;
1069 ovs_assert(new_xcfg);
1071 xport = xport_lookup(new_xcfg, ofport);
1073 xport = xzalloc(sizeof *xport);
1074 xport->ofport = ofport;
1075 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
1076 xport->ofp_port = ofp_port;
1078 xlate_xport_init(new_xcfg, xport);
1081 ovs_assert(xport->ofp_port == ofp_port);
1083 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1084 stp_port_no, rstp_port, config, state, is_tunnel,
1088 xport->peer->peer = NULL;
1090 xport->peer = xport_lookup(new_xcfg, peer);
1092 xport->peer->peer = xport;
1095 if (xport->xbundle) {
1096 list_remove(&xport->bundle_node);
1098 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1099 if (xport->xbundle) {
1100 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1103 clear_skb_priorities(xport);
1104 for (i = 0; i < n_qdscp; i++) {
1105 struct skb_priority_to_dscp *pdscp;
1106 uint32_t skb_priority;
1108 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1113 pdscp = xmalloc(sizeof *pdscp);
1114 pdscp->skb_priority = skb_priority;
1115 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1116 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1117 hash_int(pdscp->skb_priority, 0));
1122 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1129 xport->peer->peer = NULL;
1133 if (xport->xbundle) {
1134 list_remove(&xport->bundle_node);
1137 clear_skb_priorities(xport);
1138 hmap_destroy(&xport->skb_priorities);
1140 hmap_remove(&xcfg->xports, &xport->hmap_node);
1141 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1143 netdev_close(xport->netdev);
1144 rstp_port_unref(xport->rstp_port);
1145 cfm_unref(xport->cfm);
1146 bfd_unref(xport->bfd);
1147 lldp_unref(xport->lldp);
1152 xlate_ofport_remove(struct ofport_dpif *ofport)
1154 struct xport *xport;
1156 ovs_assert(new_xcfg);
1158 xport = xport_lookup(new_xcfg, ofport);
1159 xlate_xport_remove(new_xcfg, xport);
1162 static struct ofproto_dpif *
1163 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1164 ofp_port_t *ofp_in_port, const struct xport **xportp)
1166 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1167 const struct xport *xport;
1169 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1170 ? tnl_port_receive(flow)
1171 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1172 if (OVS_UNLIKELY(!xport)) {
1177 *ofp_in_port = xport->ofp_port;
1179 return xport->xbridge->ofproto;
1182 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1183 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1184 struct ofproto_dpif *
1185 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1186 ofp_port_t *ofp_in_port)
1188 const struct xport *xport;
1190 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1193 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1194 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1195 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1196 * handles for those protocols if they're enabled. Caller may use the returned
1197 * pointers until quiescing, for longer term use additional references must
1200 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1203 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1204 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1205 struct dpif_sflow **sflow, struct netflow **netflow,
1206 ofp_port_t *ofp_in_port)
1208 struct ofproto_dpif *ofproto;
1209 const struct xport *xport;
1211 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1218 *ofprotop = ofproto;
1222 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1226 *sflow = xport ? xport->xbridge->sflow : NULL;
1230 *netflow = xport ? xport->xbridge->netflow : NULL;
1236 static struct xbridge *
1237 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1239 struct hmap *xbridges;
1240 struct xbridge *xbridge;
1242 if (!ofproto || !xcfg) {
1246 xbridges = &xcfg->xbridges;
1248 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1250 if (xbridge->ofproto == ofproto) {
1257 static struct xbridge *
1258 xbridge_lookup_by_uuid(struct xlate_cfg *xcfg, const struct uuid *uuid)
1260 struct xbridge *xbridge;
1262 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
1263 if (uuid_equals(ofproto_dpif_get_uuid(xbridge->ofproto), uuid)) {
1270 static struct xbundle *
1271 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1273 struct hmap *xbundles;
1274 struct xbundle *xbundle;
1276 if (!ofbundle || !xcfg) {
1280 xbundles = &xcfg->xbundles;
1282 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1284 if (xbundle->ofbundle == ofbundle) {
1291 static struct xport *
1292 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1294 struct hmap *xports;
1295 struct xport *xport;
1297 if (!ofport || !xcfg) {
1301 xports = &xcfg->xports;
1303 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1305 if (xport->ofport == ofport) {
1312 static struct stp_port *
1313 xport_get_stp_port(const struct xport *xport)
1315 return xport->xbridge->stp && xport->stp_port_no != -1
1316 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1321 xport_stp_learn_state(const struct xport *xport)
1323 struct stp_port *sp = xport_get_stp_port(xport);
1325 ? stp_learn_in_state(stp_port_get_state(sp))
1330 xport_stp_forward_state(const struct xport *xport)
1332 struct stp_port *sp = xport_get_stp_port(xport);
1334 ? stp_forward_in_state(stp_port_get_state(sp))
1339 xport_stp_should_forward_bpdu(const struct xport *xport)
1341 struct stp_port *sp = xport_get_stp_port(xport);
1342 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1345 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1346 * were used to make the determination.*/
1348 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1350 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1351 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1352 return is_stp(flow);
1356 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1358 struct stp_port *sp = xport_get_stp_port(xport);
1359 struct dp_packet payload = *packet;
1360 struct eth_header *eth = dp_packet_data(&payload);
1362 /* Sink packets on ports that have STP disabled when the bridge has
1364 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1368 /* Trim off padding on payload. */
1369 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1370 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1373 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1374 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1378 static enum rstp_state
1379 xport_get_rstp_port_state(const struct xport *xport)
1381 return xport->rstp_port
1382 ? rstp_port_get_state(xport->rstp_port)
1387 xport_rstp_learn_state(const struct xport *xport)
1389 return xport->xbridge->rstp && xport->rstp_port
1390 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1395 xport_rstp_forward_state(const struct xport *xport)
1397 return xport->xbridge->rstp && xport->rstp_port
1398 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1403 xport_rstp_should_manage_bpdu(const struct xport *xport)
1405 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1409 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1411 struct dp_packet payload = *packet;
1412 struct eth_header *eth = dp_packet_data(&payload);
1414 /* Sink packets on ports that have no RSTP. */
1415 if (!xport->rstp_port) {
1419 /* Trim off padding on payload. */
1420 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1421 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1424 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1425 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1426 dp_packet_size(&payload));
1430 static struct xport *
1431 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1433 struct xport *xport;
1435 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1437 if (xport->ofp_port == ofp_port) {
1445 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1447 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1448 return xport ? xport->odp_port : ODPP_NONE;
1452 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1454 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1455 return xport && xport->may_enable;
1458 static struct ofputil_bucket *
1459 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1463 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1465 struct group_dpif *group;
1467 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1468 struct ofputil_bucket *bucket;
1470 bucket = group_first_live_bucket(ctx, group, depth);
1471 group_dpif_unref(group);
1472 return bucket == NULL;
1478 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1481 bucket_is_alive(const struct xlate_ctx *ctx,
1482 struct ofputil_bucket *bucket, int depth)
1484 if (depth >= MAX_LIVENESS_RECURSION) {
1485 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1487 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1488 MAX_LIVENESS_RECURSION);
1492 return (!ofputil_bucket_has_liveness(bucket)
1493 || (bucket->watch_port != OFPP_ANY
1494 && odp_port_is_alive(ctx, bucket->watch_port))
1495 || (bucket->watch_group != OFPG_ANY
1496 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1499 static struct ofputil_bucket *
1500 group_first_live_bucket(const struct xlate_ctx *ctx,
1501 const struct group_dpif *group, int depth)
1503 struct ofputil_bucket *bucket;
1504 const struct ovs_list *buckets;
1506 group_dpif_get_buckets(group, &buckets);
1507 LIST_FOR_EACH (bucket, list_node, buckets) {
1508 if (bucket_is_alive(ctx, bucket, depth)) {
1516 static struct ofputil_bucket *
1517 group_best_live_bucket(const struct xlate_ctx *ctx,
1518 const struct group_dpif *group,
1521 struct ofputil_bucket *best_bucket = NULL;
1522 uint32_t best_score = 0;
1525 struct ofputil_bucket *bucket;
1526 const struct ovs_list *buckets;
1528 group_dpif_get_buckets(group, &buckets);
1529 LIST_FOR_EACH (bucket, list_node, buckets) {
1530 if (bucket_is_alive(ctx, bucket, 0)) {
1531 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1532 if (score >= best_score) {
1533 best_bucket = bucket;
1544 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1546 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1547 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1551 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1553 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1556 static mirror_mask_t
1557 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1559 return xbundle != &ofpp_none_bundle
1560 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1564 static mirror_mask_t
1565 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1567 return xbundle != &ofpp_none_bundle
1568 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1572 static mirror_mask_t
1573 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1575 return xbundle != &ofpp_none_bundle
1576 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1580 static struct xbundle *
1581 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1582 bool warn, struct xport **in_xportp)
1584 struct xport *xport;
1586 /* Find the port and bundle for the received packet. */
1587 xport = get_ofp_port(xbridge, in_port);
1591 if (xport && xport->xbundle) {
1592 return xport->xbundle;
1595 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1596 * which a controller may use as the ingress port for traffic that
1597 * it is sourcing. */
1598 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1599 return &ofpp_none_bundle;
1602 /* Odd. A few possible reasons here:
1604 * - We deleted a port but there are still a few packets queued up
1607 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1608 * we don't know about.
1610 * - The ofproto client didn't configure the port as part of a bundle.
1611 * This is particularly likely to happen if a packet was received on the
1612 * port after it was created, but before the client had a chance to
1613 * configure its bundle.
1616 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1618 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1619 "port %"PRIu16, xbridge->name, in_port);
1625 mirror_packet(struct xlate_ctx *ctx, struct xbundle *xbundle,
1626 mirror_mask_t mirrors)
1628 bool warn = ctx->xin->packet != NULL;
1629 uint16_t vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
1630 if (!input_vid_is_valid(vid, xbundle, warn)) {
1633 uint16_t vlan = input_vid_to_vlan(xbundle, vid);
1635 const struct xbridge *xbridge = ctx->xbridge;
1637 /* Don't mirror to destinations that we've already mirrored to. */
1638 mirrors &= ~ctx->mirrors;
1643 /* Record these mirrors so that we don't mirror to them again. */
1644 ctx->mirrors |= mirrors;
1646 if (ctx->xin->resubmit_stats) {
1647 mirror_update_stats(xbridge->mbridge, mirrors,
1648 ctx->xin->resubmit_stats->n_packets,
1649 ctx->xin->resubmit_stats->n_bytes);
1651 if (ctx->xin->xcache) {
1652 struct xc_entry *entry;
1654 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_MIRROR);
1655 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
1656 entry->u.mirror.mirrors = mirrors;
1660 const unsigned long *vlans;
1661 mirror_mask_t dup_mirrors;
1662 struct ofbundle *out;
1665 bool has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1666 &vlans, &dup_mirrors, &out, &out_vlan);
1667 ovs_assert(has_mirror);
1670 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1673 if (vlans && !bitmap_is_set(vlans, vlan)) {
1674 mirrors = zero_rightmost_1bit(mirrors);
1678 mirrors &= ~dup_mirrors;
1679 ctx->mirrors |= dup_mirrors;
1681 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1682 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1684 output_normal(ctx, out_xbundle, vlan);
1686 } else if (vlan != out_vlan
1687 && !eth_addr_is_reserved(ctx->xin->flow.dl_dst)) {
1688 struct xbundle *xbundle;
1690 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1691 if (xbundle_includes_vlan(xbundle, out_vlan)
1692 && !xbundle_mirror_out(xbridge, xbundle)) {
1693 output_normal(ctx, xbundle, out_vlan);
1701 mirror_ingress_packet(struct xlate_ctx *ctx)
1703 if (mbridge_has_mirrors(ctx->xbridge->mbridge)) {
1704 bool warn = ctx->xin->packet != NULL;
1705 struct xbundle *xbundle = lookup_input_bundle(
1706 ctx->xbridge, ctx->xin->flow.in_port.ofp_port, warn, NULL);
1708 mirror_packet(ctx, xbundle,
1709 xbundle_mirror_src(ctx->xbridge, xbundle));
1714 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1715 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1716 * the bundle on which the packet was received, returns the VLAN to which the
1719 * Both 'vid' and the return value are in the range 0...4095. */
1721 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1723 switch (in_xbundle->vlan_mode) {
1724 case PORT_VLAN_ACCESS:
1725 return in_xbundle->vlan;
1728 case PORT_VLAN_TRUNK:
1731 case PORT_VLAN_NATIVE_UNTAGGED:
1732 case PORT_VLAN_NATIVE_TAGGED:
1733 return vid ? vid : in_xbundle->vlan;
1740 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1741 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1744 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1745 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1748 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1750 /* Allow any VID on the OFPP_NONE port. */
1751 if (in_xbundle == &ofpp_none_bundle) {
1755 switch (in_xbundle->vlan_mode) {
1756 case PORT_VLAN_ACCESS:
1759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1760 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1761 "packet received on port %s configured as VLAN "
1762 "%"PRIu16" access port", vid, in_xbundle->name,
1769 case PORT_VLAN_NATIVE_UNTAGGED:
1770 case PORT_VLAN_NATIVE_TAGGED:
1772 /* Port must always carry its native VLAN. */
1776 case PORT_VLAN_TRUNK:
1777 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1779 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1780 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1781 "received on port %s not configured for trunking "
1782 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1794 /* Given 'vlan', the VLAN that a packet belongs to, and
1795 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1796 * that should be included in the 802.1Q header. (If the return value is 0,
1797 * then the 802.1Q header should only be included in the packet if there is a
1800 * Both 'vlan' and the return value are in the range 0...4095. */
1802 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1804 switch (out_xbundle->vlan_mode) {
1805 case PORT_VLAN_ACCESS:
1808 case PORT_VLAN_TRUNK:
1809 case PORT_VLAN_NATIVE_TAGGED:
1812 case PORT_VLAN_NATIVE_UNTAGGED:
1813 return vlan == out_xbundle->vlan ? 0 : vlan;
1821 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1824 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1826 ovs_be16 tci, old_tci;
1827 struct xport *xport;
1828 struct xlate_bond_recirc xr;
1829 bool use_recirc = false;
1831 vid = output_vlan_to_vid(out_xbundle, vlan);
1832 if (list_is_empty(&out_xbundle->xports)) {
1833 /* Partially configured bundle with no slaves. Drop the packet. */
1835 } else if (!out_xbundle->bond) {
1836 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1839 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1840 struct flow_wildcards *wc = ctx->wc;
1841 struct ofport_dpif *ofport;
1843 if (ctx->xbridge->support.odp.recirc) {
1844 use_recirc = bond_may_recirc(
1845 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1848 /* Only TCP mode uses recirculation. */
1849 xr.hash_alg = OVS_HASH_ALG_L4;
1850 bond_update_post_recirc_rules(out_xbundle->bond, false);
1852 /* Recirculation does not require unmasking hash fields. */
1857 ofport = bond_choose_output_slave(out_xbundle->bond,
1858 &ctx->xin->flow, wc, vid);
1859 xport = xport_lookup(xcfg, ofport);
1862 /* No slaves enabled, so drop packet. */
1866 /* If use_recirc is set, the main thread will handle stats
1867 * accounting for this bond. */
1869 if (ctx->xin->resubmit_stats) {
1870 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1871 ctx->xin->resubmit_stats->n_bytes);
1873 if (ctx->xin->xcache) {
1874 struct xc_entry *entry;
1877 flow = &ctx->xin->flow;
1878 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1879 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1880 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1881 entry->u.bond.vid = vid;
1886 old_tci = *flow_tci;
1888 if (tci || out_xbundle->use_priority_tags) {
1889 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1891 tci |= htons(VLAN_CFI);
1896 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1897 *flow_tci = old_tci;
1900 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1901 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1902 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1904 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1906 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1910 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1911 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1915 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1916 if (flow->nw_proto == ARP_OP_REPLY) {
1918 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1919 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1920 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1922 return flow->nw_src == flow->nw_dst;
1928 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1929 * dropped. Returns true if they may be forwarded, false if they should be
1932 * 'in_port' must be the xport that corresponds to flow->in_port.
1933 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1935 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1936 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1937 * checked by input_vid_is_valid().
1939 * May also add tags to '*tags', although the current implementation only does
1940 * so in one special case.
1943 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1946 struct xbundle *in_xbundle = in_port->xbundle;
1947 const struct xbridge *xbridge = ctx->xbridge;
1948 struct flow *flow = &ctx->xin->flow;
1950 /* Drop frames for reserved multicast addresses
1951 * only if forward_bpdu option is absent. */
1952 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1953 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1957 if (in_xbundle->bond) {
1958 struct mac_entry *mac;
1960 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1966 xlate_report(ctx, "bonding refused admissibility, dropping");
1969 case BV_DROP_IF_MOVED:
1970 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1971 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1973 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1974 && (!is_gratuitous_arp(flow, ctx->wc)
1975 || mac_entry_is_grat_arp_locked(mac))) {
1976 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1977 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1981 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1989 /* Checks whether a MAC learning update is necessary for MAC learning table
1990 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1993 * Most packets processed through the MAC learning table do not actually
1994 * change it in any way. This function requires only a read lock on the MAC
1995 * learning table, so it is much cheaper in this common case.
1997 * Keep the code here synchronized with that in update_learning_table__()
2000 is_mac_learning_update_needed(const struct mac_learning *ml,
2001 const struct flow *flow,
2002 struct flow_wildcards *wc,
2003 int vlan, struct xbundle *in_xbundle)
2004 OVS_REQ_RDLOCK(ml->rwlock)
2006 struct mac_entry *mac;
2008 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
2012 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
2013 if (!mac || mac_entry_age(ml, mac)) {
2017 if (is_gratuitous_arp(flow, wc)) {
2018 /* We don't want to learn from gratuitous ARP packets that are
2019 * reflected back over bond slaves so we lock the learning table. */
2020 if (!in_xbundle->bond) {
2022 } else if (mac_entry_is_grat_arp_locked(mac)) {
2027 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
2031 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
2032 * received on 'in_xbundle' in 'vlan'.
2034 * This code repeats all the checks in is_mac_learning_update_needed() because
2035 * the lock was released between there and here and thus the MAC learning state
2036 * could have changed.
2038 * Keep the code here synchronized with that in is_mac_learning_update_needed()
2041 update_learning_table__(const struct xbridge *xbridge,
2042 const struct flow *flow, struct flow_wildcards *wc,
2043 int vlan, struct xbundle *in_xbundle)
2044 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
2046 struct mac_entry *mac;
2048 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
2052 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
2053 if (is_gratuitous_arp(flow, wc)) {
2054 /* We don't want to learn from gratuitous ARP packets that are
2055 * reflected back over bond slaves so we lock the learning table. */
2056 if (!in_xbundle->bond) {
2057 mac_entry_set_grat_arp_lock(mac);
2058 } else if (mac_entry_is_grat_arp_locked(mac)) {
2063 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
2064 /* The log messages here could actually be useful in debugging,
2065 * so keep the rate limit relatively high. */
2066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
2068 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
2069 "on port %s in VLAN %d",
2070 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
2071 in_xbundle->name, vlan);
2073 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
2078 update_learning_table(const struct xbridge *xbridge,
2079 const struct flow *flow, struct flow_wildcards *wc,
2080 int vlan, struct xbundle *in_xbundle)
2084 /* Don't learn the OFPP_NONE port. */
2085 if (in_xbundle == &ofpp_none_bundle) {
2089 /* First try the common case: no change to MAC learning table. */
2090 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
2091 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
2093 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2096 /* Slow path: MAC learning table might need an update. */
2097 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2098 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2099 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2103 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2104 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2106 update_mcast_snooping_table4__(const struct xbridge *xbridge,
2107 const struct flow *flow,
2108 struct mcast_snooping *ms, int vlan,
2109 struct xbundle *in_xbundle,
2110 const struct dp_packet *packet)
2111 OVS_REQ_WRLOCK(ms->rwlock)
2113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2115 ovs_be32 ip4 = flow->igmp_group_ip4;
2117 switch (ntohs(flow->tp_src)) {
2118 case IGMP_HOST_MEMBERSHIP_REPORT:
2119 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2120 if (mcast_snooping_add_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2121 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2122 IP_FMT" is on port %s in VLAN %d",
2123 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2126 case IGMP_HOST_LEAVE_MESSAGE:
2127 if (mcast_snooping_leave_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2128 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2129 IP_FMT" is on port %s in VLAN %d",
2130 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2133 case IGMP_HOST_MEMBERSHIP_QUERY:
2134 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2135 in_xbundle->ofbundle)) {
2136 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2137 IP_FMT" is on port %s in VLAN %d",
2138 xbridge->name, IP_ARGS(flow->nw_src),
2139 in_xbundle->name, vlan);
2142 case IGMPV3_HOST_MEMBERSHIP_REPORT:
2143 if ((count = mcast_snooping_add_report(ms, packet, vlan,
2144 in_xbundle->ofbundle))) {
2145 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2146 "addresses on port %s in VLAN %d",
2147 xbridge->name, count, in_xbundle->name, vlan);
2154 update_mcast_snooping_table6__(const struct xbridge *xbridge,
2155 const struct flow *flow,
2156 struct mcast_snooping *ms, int vlan,
2157 struct xbundle *in_xbundle,
2158 const struct dp_packet *packet)
2159 OVS_REQ_WRLOCK(ms->rwlock)
2161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2164 switch (ntohs(flow->tp_src)) {
2166 if (!ipv6_addr_equals(&flow->ipv6_src, &in6addr_any)
2167 && mcast_snooping_add_mrouter(ms, vlan, in_xbundle->ofbundle)) {
2168 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query on port %s"
2170 xbridge->name, in_xbundle->name, vlan);
2176 count = mcast_snooping_add_mld(ms, packet, vlan, in_xbundle->ofbundle);
2178 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2179 "addresses on port %s in VLAN %d",
2180 xbridge->name, count, in_xbundle->name, vlan);
2186 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2187 * was received on 'in_xbundle' in 'vlan'. */
2189 update_mcast_snooping_table(const struct xbridge *xbridge,
2190 const struct flow *flow, int vlan,
2191 struct xbundle *in_xbundle,
2192 const struct dp_packet *packet)
2194 struct mcast_snooping *ms = xbridge->ms;
2195 struct xlate_cfg *xcfg;
2196 struct xbundle *mcast_xbundle;
2197 struct mcast_port_bundle *fport;
2199 /* Don't learn the OFPP_NONE port. */
2200 if (in_xbundle == &ofpp_none_bundle) {
2204 /* Don't learn from flood ports */
2205 mcast_xbundle = NULL;
2206 ovs_rwlock_wrlock(&ms->rwlock);
2207 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2208 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2209 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2210 if (mcast_xbundle == in_xbundle) {
2215 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2216 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2217 update_mcast_snooping_table4__(xbridge, flow, ms, vlan,
2218 in_xbundle, packet);
2220 update_mcast_snooping_table6__(xbridge, flow, ms, vlan,
2221 in_xbundle, packet);
2224 ovs_rwlock_unlock(&ms->rwlock);
2227 /* send the packet to ports having the multicast group learned */
2229 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2230 struct mcast_snooping *ms OVS_UNUSED,
2231 struct mcast_group *grp,
2232 struct xbundle *in_xbundle, uint16_t vlan)
2233 OVS_REQ_RDLOCK(ms->rwlock)
2235 struct xlate_cfg *xcfg;
2236 struct mcast_group_bundle *b;
2237 struct xbundle *mcast_xbundle;
2239 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2240 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2241 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2242 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2243 xlate_report(ctx, "forwarding to mcast group port");
2244 output_normal(ctx, mcast_xbundle, vlan);
2245 } else if (!mcast_xbundle) {
2246 xlate_report(ctx, "mcast group port is unknown, dropping");
2248 xlate_report(ctx, "mcast group port is input port, dropping");
2253 /* send the packet to ports connected to multicast routers */
2255 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2256 struct mcast_snooping *ms,
2257 struct xbundle *in_xbundle, uint16_t vlan)
2258 OVS_REQ_RDLOCK(ms->rwlock)
2260 struct xlate_cfg *xcfg;
2261 struct mcast_mrouter_bundle *mrouter;
2262 struct xbundle *mcast_xbundle;
2264 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2265 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2266 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2267 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2268 xlate_report(ctx, "forwarding to mcast router port");
2269 output_normal(ctx, mcast_xbundle, vlan);
2270 } else if (!mcast_xbundle) {
2271 xlate_report(ctx, "mcast router port is unknown, dropping");
2273 xlate_report(ctx, "mcast router port is input port, dropping");
2278 /* send the packet to ports flagged to be flooded */
2280 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2281 struct mcast_snooping *ms,
2282 struct xbundle *in_xbundle, uint16_t vlan)
2283 OVS_REQ_RDLOCK(ms->rwlock)
2285 struct xlate_cfg *xcfg;
2286 struct mcast_port_bundle *fport;
2287 struct xbundle *mcast_xbundle;
2289 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2290 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2291 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2292 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2293 xlate_report(ctx, "forwarding to mcast flood port");
2294 output_normal(ctx, mcast_xbundle, vlan);
2295 } else if (!mcast_xbundle) {
2296 xlate_report(ctx, "mcast flood port is unknown, dropping");
2298 xlate_report(ctx, "mcast flood port is input port, dropping");
2303 /* forward the Reports to configured ports */
2305 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2306 struct mcast_snooping *ms,
2307 struct xbundle *in_xbundle, uint16_t vlan)
2308 OVS_REQ_RDLOCK(ms->rwlock)
2310 struct xlate_cfg *xcfg;
2311 struct mcast_port_bundle *rport;
2312 struct xbundle *mcast_xbundle;
2314 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2315 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2316 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2317 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2318 xlate_report(ctx, "forwarding Report to mcast flagged port");
2319 output_normal(ctx, mcast_xbundle, vlan);
2320 } else if (!mcast_xbundle) {
2321 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2323 xlate_report(ctx, "mcast port is input port, dropping the Report");
2329 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2332 struct xbundle *xbundle;
2334 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2335 if (xbundle != in_xbundle
2336 && xbundle_includes_vlan(xbundle, vlan)
2337 && xbundle->floodable
2338 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2339 output_normal(ctx, xbundle, vlan);
2342 ctx->nf_output_iface = NF_OUT_FLOOD;
2346 xlate_normal(struct xlate_ctx *ctx)
2348 struct flow_wildcards *wc = ctx->wc;
2349 struct flow *flow = &ctx->xin->flow;
2350 struct xbundle *in_xbundle;
2351 struct xport *in_port;
2352 struct mac_entry *mac;
2357 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2358 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2359 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2361 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2362 ctx->xin->packet != NULL, &in_port);
2364 xlate_report(ctx, "no input bundle, dropping");
2368 /* Drop malformed frames. */
2369 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2370 !(flow->vlan_tci & htons(VLAN_CFI))) {
2371 if (ctx->xin->packet != NULL) {
2372 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2373 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2374 "VLAN tag received on port %s",
2375 ctx->xbridge->name, in_xbundle->name);
2377 xlate_report(ctx, "partial VLAN tag, dropping");
2381 /* Drop frames on bundles reserved for mirroring. */
2382 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2383 if (ctx->xin->packet != NULL) {
2384 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2385 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2386 "%s, which is reserved exclusively for mirroring",
2387 ctx->xbridge->name, in_xbundle->name);
2389 xlate_report(ctx, "input port is mirror output port, dropping");
2394 vid = vlan_tci_to_vid(flow->vlan_tci);
2395 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2396 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2399 vlan = input_vid_to_vlan(in_xbundle, vid);
2401 /* Check other admissibility requirements. */
2402 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2406 /* Learn source MAC. */
2407 if (ctx->xin->may_learn) {
2408 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2410 if (ctx->xin->xcache) {
2411 struct xc_entry *entry;
2413 /* Save enough info to update mac learning table later. */
2414 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2415 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2416 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2417 entry->u.normal.vlan = vlan;
2420 /* Determine output bundle. */
2421 if (mcast_snooping_enabled(ctx->xbridge->ms)
2422 && !eth_addr_is_broadcast(flow->dl_dst)
2423 && eth_addr_is_multicast(flow->dl_dst)
2424 && is_ip_any(flow)) {
2425 struct mcast_snooping *ms = ctx->xbridge->ms;
2426 struct mcast_group *grp = NULL;
2428 if (is_igmp(flow)) {
2429 if (mcast_snooping_is_membership(flow->tp_src) ||
2430 mcast_snooping_is_query(flow->tp_src)) {
2431 if (ctx->xin->may_learn) {
2432 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2433 in_xbundle, ctx->xin->packet);
2436 * IGMP packets need to take the slow path, in order to be
2437 * processed for mdb updates. That will prevent expires
2438 * firing off even after hosts have sent reports.
2440 ctx->xout->slow |= SLOW_ACTION;
2443 if (mcast_snooping_is_membership(flow->tp_src)) {
2444 ovs_rwlock_rdlock(&ms->rwlock);
2445 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2446 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2447 * forward IGMP Membership Reports only to those ports where
2448 * multicast routers are attached. Alternatively stated: a
2449 * snooping switch should not forward IGMP Membership Reports
2450 * to ports on which only hosts are attached.
2451 * An administrative control may be provided to override this
2452 * restriction, allowing the report messages to be flooded to
2454 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2455 ovs_rwlock_unlock(&ms->rwlock);
2457 xlate_report(ctx, "multicast traffic, flooding");
2458 xlate_normal_flood(ctx, in_xbundle, vlan);
2461 } else if (is_mld(flow)) {
2462 ctx->xout->slow |= SLOW_ACTION;
2463 if (ctx->xin->may_learn) {
2464 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2465 in_xbundle, ctx->xin->packet);
2467 if (is_mld_report(flow)) {
2468 ovs_rwlock_rdlock(&ms->rwlock);
2469 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2470 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2471 ovs_rwlock_unlock(&ms->rwlock);
2473 xlate_report(ctx, "MLD query, flooding");
2474 xlate_normal_flood(ctx, in_xbundle, vlan);
2477 if ((flow->dl_type == htons(ETH_TYPE_IP)
2478 && ip_is_local_multicast(flow->nw_dst))
2479 || (flow->dl_type == htons(ETH_TYPE_IPV6)
2480 && ipv6_is_all_hosts(&flow->ipv6_dst))) {
2481 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2482 * address in the 224.0.0.x range which are not IGMP must
2483 * be forwarded on all ports */
2484 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2485 xlate_normal_flood(ctx, in_xbundle, vlan);
2490 /* forwarding to group base ports */
2491 ovs_rwlock_rdlock(&ms->rwlock);
2492 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2493 grp = mcast_snooping_lookup4(ms, flow->nw_dst, vlan);
2494 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2495 grp = mcast_snooping_lookup(ms, &flow->ipv6_dst, vlan);
2498 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2499 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2500 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2502 if (mcast_snooping_flood_unreg(ms)) {
2503 xlate_report(ctx, "unregistered multicast, flooding");
2504 xlate_normal_flood(ctx, in_xbundle, vlan);
2506 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2507 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2510 ovs_rwlock_unlock(&ms->rwlock);
2512 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2513 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2514 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2515 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2518 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2519 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2520 if (mac_xbundle && mac_xbundle != in_xbundle) {
2521 xlate_report(ctx, "forwarding to learned port");
2522 output_normal(ctx, mac_xbundle, vlan);
2523 } else if (!mac_xbundle) {
2524 xlate_report(ctx, "learned port is unknown, dropping");
2526 xlate_report(ctx, "learned port is input port, dropping");
2529 xlate_report(ctx, "no learned MAC for destination, flooding");
2530 xlate_normal_flood(ctx, in_xbundle, vlan);
2535 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
2536 * 'probability' is the number of packets out of UINT32_MAX to sample. The
2537 * 'cookie' (of length 'cookie_size' bytes) is passed back in the callback for
2538 * each sampled packet. 'tunnel_out_port', if not ODPP_NONE, is added as the
2539 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions', an
2540 * OVS_USERSPACE_ATTR_ACTIONS attribute is added.
2543 compose_sample_action(struct xlate_ctx *ctx,
2544 const uint32_t probability,
2545 const union user_action_cookie *cookie,
2546 const size_t cookie_size,
2547 const odp_port_t tunnel_out_port,
2548 bool include_actions)
2550 size_t sample_offset = nl_msg_start_nested(ctx->odp_actions,
2551 OVS_ACTION_ATTR_SAMPLE);
2553 nl_msg_put_u32(ctx->odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2555 size_t actions_offset = nl_msg_start_nested(ctx->odp_actions,
2556 OVS_SAMPLE_ATTR_ACTIONS);
2558 odp_port_t odp_port = ofp_port_to_odp_port(
2559 ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
2560 uint32_t pid = dpif_port_get_pid(ctx->xbridge->dpif, odp_port,
2561 flow_hash_5tuple(&ctx->xin->flow, 0));
2562 int cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2567 nl_msg_end_nested(ctx->odp_actions, actions_offset);
2568 nl_msg_end_nested(ctx->odp_actions, sample_offset);
2570 return cookie_offset;
2573 /* If sFLow is not enabled, returns 0 without doing anything.
2575 * If sFlow is enabled, appends a template "sample" action to the ODP actions
2576 * in 'ctx'. This action is a template because some of the information needed
2577 * to fill it out is not available until flow translation is complete. In this
2578 * case, this functions returns an offset, which is always nonzero, to pass
2579 * later to fix_sflow_action() to fill in the rest of the template. */
2581 compose_sflow_action(struct xlate_ctx *ctx)
2583 struct dpif_sflow *sflow = ctx->xbridge->sflow;
2584 if (!sflow || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2588 union user_action_cookie cookie = { .type = USER_ACTION_COOKIE_SFLOW };
2589 return compose_sample_action(ctx, dpif_sflow_get_probability(sflow),
2590 &cookie, sizeof cookie.sflow, ODPP_NONE,
2594 /* If IPFIX is enabled, this appends a "sample" action to implement IPFIX to
2595 * 'ctx->odp_actions'. */
2597 compose_ipfix_action(struct xlate_ctx *ctx, odp_port_t output_odp_port)
2599 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
2600 odp_port_t tunnel_out_port = ODPP_NONE;
2602 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2606 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2608 if (output_odp_port == ODPP_NONE &&
2609 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix)) {
2613 /* For output case, output_odp_port is valid*/
2614 if (output_odp_port != ODPP_NONE) {
2615 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix)) {
2618 /* If tunnel sampling is enabled, put an additional option attribute:
2619 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2621 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix) &&
2622 dpif_ipfix_get_tunnel_port(ipfix, output_odp_port) ) {
2623 tunnel_out_port = output_odp_port;
2627 union user_action_cookie cookie = {
2629 .type = USER_ACTION_COOKIE_IPFIX,
2630 .output_odp_port = output_odp_port,
2633 compose_sample_action(ctx,
2634 dpif_ipfix_get_bridge_exporter_probability(ipfix),
2635 &cookie, sizeof cookie.ipfix, tunnel_out_port,
2639 /* Fix "sample" action according to data collected while composing ODP actions,
2640 * as described in compose_sflow_action().
2642 * 'user_cookie_offset' must be the offset returned by add_sflow_action(). */
2644 fix_sflow_action(struct xlate_ctx *ctx, unsigned int user_cookie_offset)
2646 const struct flow *base = &ctx->base_flow;
2647 union user_action_cookie *cookie;
2649 cookie = ofpbuf_at(ctx->odp_actions, user_cookie_offset,
2650 sizeof cookie->sflow);
2651 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2653 cookie->type = USER_ACTION_COOKIE_SFLOW;
2654 cookie->sflow.vlan_tci = base->vlan_tci;
2656 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2657 * port information") for the interpretation of cookie->output. */
2658 switch (ctx->sflow_n_outputs) {
2660 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2661 cookie->sflow.output = 0x40000000 | 256;
2665 cookie->sflow.output = dpif_sflow_odp_port_to_ifindex(
2666 ctx->xbridge->sflow, ctx->sflow_odp_port);
2667 if (cookie->sflow.output) {
2672 /* 0x80000000 means "multiple output ports. */
2673 cookie->sflow.output = 0x80000000 | ctx->sflow_n_outputs;
2679 process_special(struct xlate_ctx *ctx, const struct xport *xport)
2681 const struct flow *flow = &ctx->xin->flow;
2682 struct flow_wildcards *wc = ctx->wc;
2683 const struct xbridge *xbridge = ctx->xbridge;
2684 const struct dp_packet *packet = ctx->xin->packet;
2685 enum slow_path_reason slow;
2689 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2691 cfm_process_heartbeat(xport->cfm, packet);
2694 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2696 bfd_process_packet(xport->bfd, flow, packet);
2697 /* If POLL received, immediately sends FINAL back. */
2698 if (bfd_should_send_packet(xport->bfd)) {
2699 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2703 } else if (xport->xbundle && xport->xbundle->lacp
2704 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2706 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2709 } else if ((xbridge->stp || xbridge->rstp) &&
2710 stp_should_process_flow(flow, wc)) {
2713 ? stp_process_packet(xport, packet)
2714 : rstp_process_packet(xport, packet);
2717 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2719 lldp_process_packet(xport->lldp, packet);
2727 ctx->xout->slow |= slow;
2735 tnl_route_lookup_flow(const struct flow *oflow,
2736 struct in6_addr *ip, struct xport **out_port)
2738 char out_dev[IFNAMSIZ];
2739 struct xbridge *xbridge;
2740 struct xlate_cfg *xcfg;
2742 struct in6_addr dst;
2744 dst = flow_tnl_dst(&oflow->tunnel);
2745 if (!ovs_router_lookup(&dst, out_dev, &gw)) {
2749 if (ipv6_addr_is_set(&gw) &&
2750 (!IN6_IS_ADDR_V4MAPPED(&gw) || in6_addr_get_mapped_ipv4(&gw))) {
2756 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2759 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2760 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2763 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2764 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2775 compose_table_xlate(struct xlate_ctx *ctx, const struct xport *out_dev,
2776 struct dp_packet *packet)
2778 struct xbridge *xbridge = out_dev->xbridge;
2779 struct ofpact_output output;
2782 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2783 flow_extract(packet, &flow);
2784 flow.in_port.ofp_port = out_dev->ofp_port;
2785 output.port = OFPP_TABLE;
2788 return ofproto_dpif_execute_actions__(xbridge->ofproto, &flow, NULL,
2789 &output.ofpact, sizeof output,
2790 ctx->recurse, ctx->resubmits, packet);
2794 tnl_send_nd_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2795 const struct eth_addr eth_src,
2796 struct in6_addr * ipv6_src, struct in6_addr * ipv6_dst)
2798 struct dp_packet packet;
2800 dp_packet_init(&packet, 0);
2801 compose_nd(&packet, eth_src, ipv6_src, ipv6_dst);
2802 compose_table_xlate(ctx, out_dev, &packet);
2803 dp_packet_uninit(&packet);
2807 tnl_send_arp_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2808 const struct eth_addr eth_src,
2809 ovs_be32 ip_src, ovs_be32 ip_dst)
2811 struct dp_packet packet;
2813 dp_packet_init(&packet, 0);
2814 compose_arp(&packet, ARP_OP_REQUEST,
2815 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2817 compose_table_xlate(ctx, out_dev, &packet);
2818 dp_packet_uninit(&packet);
2822 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2823 const struct flow *flow, odp_port_t tunnel_odp_port)
2825 struct ovs_action_push_tnl tnl_push_data;
2826 struct xport *out_dev = NULL;
2827 ovs_be32 s_ip = 0, d_ip = 0;
2828 struct in6_addr s_ip6 = in6addr_any;
2829 struct in6_addr d_ip6 = in6addr_any;
2830 struct eth_addr smac;
2831 struct eth_addr dmac;
2833 char buf_sip6[INET6_ADDRSTRLEN];
2834 char buf_dip6[INET6_ADDRSTRLEN];
2836 err = tnl_route_lookup_flow(flow, &d_ip6, &out_dev);
2838 xlate_report(ctx, "native tunnel routing failed");
2842 xlate_report(ctx, "tunneling to %s via %s",
2843 ipv6_string_mapped(buf_dip6, &d_ip6),
2844 netdev_get_name(out_dev->netdev));
2846 /* Use mac addr of bridge port of the peer. */
2847 err = netdev_get_etheraddr(out_dev->netdev, &smac);
2849 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2853 d_ip = in6_addr_get_mapped_ipv4(&d_ip6);
2855 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2857 xlate_report(ctx, "tunnel output device lacks IPv4 address");
2860 in6_addr_set_mapped_ipv4(&s_ip6, s_ip);
2862 err = netdev_get_in6(out_dev->netdev, &s_ip6);
2864 xlate_report(ctx, "tunnel output device lacks IPv6 address");
2869 err = tnl_neigh_lookup(out_dev->xbridge->name, &d_ip6, &dmac);
2871 xlate_report(ctx, "neighbor cache miss for %s on bridge %s, "
2872 "sending %s request",
2873 buf_dip6, out_dev->xbridge->name, d_ip ? "ARP" : "ND");
2875 tnl_send_arp_request(ctx, out_dev, smac, s_ip, d_ip);
2877 tnl_send_nd_request(ctx, out_dev, smac, &s_ip6, &d_ip6);
2882 if (ctx->xin->xcache) {
2883 struct xc_entry *entry;
2885 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_NEIGH);
2886 ovs_strlcpy(entry->u.tnl_neigh_cache.br_name, out_dev->xbridge->name,
2887 sizeof entry->u.tnl_neigh_cache.br_name);
2888 entry->u.tnl_neigh_cache.d_ipv6 = d_ip6;
2891 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" %s"
2892 " to "ETH_ADDR_FMT" %s",
2893 ETH_ADDR_ARGS(smac), ipv6_string_mapped(buf_sip6, &s_ip6),
2894 ETH_ADDR_ARGS(dmac), buf_dip6);
2896 err = tnl_port_build_header(xport->ofport, flow,
2897 dmac, smac, &s_ip6, &tnl_push_data);
2901 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2902 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2903 odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
2908 xlate_commit_actions(struct xlate_ctx *ctx)
2910 bool use_masked = ctx->xbridge->support.masked_set_action;
2912 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2913 ctx->odp_actions, ctx->wc,
2918 clear_conntrack(struct flow *flow)
2923 memset(&flow->ct_label, 0, sizeof flow->ct_label);
2927 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2928 const struct xlate_bond_recirc *xr, bool check_stp)
2930 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2931 struct flow_wildcards *wc = ctx->wc;
2932 struct flow *flow = &ctx->xin->flow;
2933 struct flow_tnl flow_tnl;
2934 ovs_be16 flow_vlan_tci;
2935 uint32_t flow_pkt_mark;
2936 uint8_t flow_nw_tos;
2937 odp_port_t out_port, odp_port;
2938 bool tnl_push_pop_send = false;
2941 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2942 * before traversing a patch port. */
2943 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 35);
2944 memset(&flow_tnl, 0, sizeof flow_tnl);
2947 xlate_report(ctx, "Nonexistent output port");
2949 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2950 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2952 } else if (check_stp) {
2953 if (is_stp(&ctx->base_flow)) {
2954 if (!xport_stp_should_forward_bpdu(xport) &&
2955 !xport_rstp_should_manage_bpdu(xport)) {
2956 if (ctx->xbridge->stp != NULL) {
2957 xlate_report(ctx, "STP not in listening state, "
2958 "skipping bpdu output");
2959 } else if (ctx->xbridge->rstp != NULL) {
2960 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2961 "skipping bpdu output");
2965 } else if (!xport_stp_forward_state(xport) ||
2966 !xport_rstp_forward_state(xport)) {
2967 if (ctx->xbridge->stp != NULL) {
2968 xlate_report(ctx, "STP not in forwarding state, "
2970 } else if (ctx->xbridge->rstp != NULL) {
2971 xlate_report(ctx, "RSTP not in forwarding state, "
2979 const struct xport *peer = xport->peer;
2980 struct flow old_flow = ctx->xin->flow;
2981 bool old_conntrack = ctx->conntracked;
2982 bool old_was_mpls = ctx->was_mpls;
2983 cls_version_t old_version = ctx->tables_version;
2984 struct ofpbuf old_stack = ctx->stack;
2985 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2986 struct ofpbuf old_action_set = ctx->action_set;
2987 uint64_t actset_stub[1024 / 8];
2989 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2990 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2991 ctx->xbridge = peer->xbridge;
2992 flow->in_port.ofp_port = peer->ofp_port;
2993 flow->metadata = htonll(0);
2994 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2995 memset(flow->regs, 0, sizeof flow->regs);
2996 flow->actset_output = OFPP_UNSET;
2997 ctx->conntracked = false;
2998 clear_conntrack(flow);
3000 /* The bridge is now known so obtain its table version. */
3002 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
3004 if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
3005 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
3006 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
3007 if (!ctx->freezing) {
3008 xlate_action_set(ctx);
3010 if (ctx->freezing) {
3011 compose_recirculate_action(ctx);
3014 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
3015 * the learning action look at the packet, then drop it. */
3016 struct flow old_base_flow = ctx->base_flow;
3017 size_t old_size = ctx->odp_actions->size;
3018 mirror_mask_t old_mirrors = ctx->mirrors;
3020 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
3021 ctx->mirrors = old_mirrors;
3022 ctx->base_flow = old_base_flow;
3023 ctx->odp_actions->size = old_size;
3025 /* Undo changes that may have been done for freezing. */
3026 ctx_cancel_freeze(ctx);
3030 ctx->xin->flow = old_flow;
3031 ctx->xbridge = xport->xbridge;
3032 ofpbuf_uninit(&ctx->action_set);
3033 ctx->action_set = old_action_set;
3034 ofpbuf_uninit(&ctx->stack);
3035 ctx->stack = old_stack;
3037 /* Restore calling bridge's lookup version. */
3038 ctx->tables_version = old_version;
3040 /* The peer bridge popping MPLS should have no effect on the original
3042 ctx->was_mpls = old_was_mpls;
3044 /* The peer bridge's conntrack execution should have no effect on the
3045 * original bridge. */
3046 ctx->conntracked = old_conntrack;
3048 /* The fact that the peer bridge exits (for any reason) does not mean
3049 * that the original bridge should exit. Specifically, if the peer
3050 * bridge freezes translation, the original bridge must continue
3051 * processing with the original, not the frozen packet! */
3054 /* Peer bridge errors do not propagate back. */
3055 ctx->error = XLATE_OK;
3057 if (ctx->xin->resubmit_stats) {
3058 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
3059 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
3061 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
3064 if (ctx->xin->xcache) {
3065 struct xc_entry *entry;
3067 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3068 entry->u.dev.tx = netdev_ref(xport->netdev);
3069 entry->u.dev.rx = netdev_ref(peer->netdev);
3070 entry->u.dev.bfd = bfd_ref(peer->bfd);
3075 flow_vlan_tci = flow->vlan_tci;
3076 flow_pkt_mark = flow->pkt_mark;
3077 flow_nw_tos = flow->nw_tos;
3079 if (count_skb_priorities(xport)) {
3080 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3081 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
3082 wc->masks.nw_tos |= IP_DSCP_MASK;
3083 flow->nw_tos &= ~IP_DSCP_MASK;
3084 flow->nw_tos |= dscp;
3088 if (xport->is_tunnel) {
3089 struct in6_addr dst;
3090 /* Save tunnel metadata so that changes made due to
3091 * the Logical (tunnel) Port are not visible for any further
3092 * matches, while explicit set actions on tunnel metadata are.
3094 flow_tnl = flow->tunnel;
3095 odp_port = tnl_port_send(xport->ofport, flow, ctx->wc);
3096 if (odp_port == ODPP_NONE) {
3097 xlate_report(ctx, "Tunneling decided against output");
3098 goto out; /* restore flow_nw_tos */
3100 dst = flow_tnl_dst(&flow->tunnel);
3101 if (ipv6_addr_equals(&dst, &ctx->orig_tunnel_ipv6_dst)) {
3102 xlate_report(ctx, "Not tunneling to our own address");
3103 goto out; /* restore flow_nw_tos */
3105 if (ctx->xin->resubmit_stats) {
3106 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
3108 if (ctx->xin->xcache) {
3109 struct xc_entry *entry;
3111 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3112 entry->u.dev.tx = netdev_ref(xport->netdev);
3114 out_port = odp_port;
3115 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3116 xlate_report(ctx, "output to native tunnel");
3117 tnl_push_pop_send = true;
3119 xlate_report(ctx, "output to kernel tunnel");
3120 commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions);
3121 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3124 odp_port = xport->odp_port;
3125 out_port = odp_port;
3126 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
3127 ofp_port_t vlandev_port;
3129 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3130 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
3131 ofp_port, flow->vlan_tci);
3132 if (vlandev_port != ofp_port) {
3133 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
3134 flow->vlan_tci = htons(0);
3139 if (out_port != ODPP_NONE) {
3140 xlate_commit_actions(ctx);
3143 struct ovs_action_hash *act_hash;
3146 act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
3147 OVS_ACTION_ATTR_HASH,
3149 act_hash->hash_alg = xr->hash_alg;
3150 act_hash->hash_basis = xr->hash_basis;
3152 /* Recirc action. */
3153 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
3157 if (tnl_push_pop_send) {
3158 build_tunnel_send(ctx, xport, flow, odp_port);
3159 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3161 odp_port_t odp_tnl_port = ODPP_NONE;
3163 /* XXX: Write better Filter for tunnel port. We can use inport
3164 * int tunnel-port flow to avoid these checks completely. */
3165 if (ofp_port == OFPP_LOCAL &&
3166 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3168 odp_tnl_port = tnl_port_map_lookup(flow, wc);
3171 if (odp_tnl_port != ODPP_NONE) {
3172 nl_msg_put_odp_port(ctx->odp_actions,
3173 OVS_ACTION_ATTR_TUNNEL_POP,
3176 /* Tunnel push-pop action is not compatible with
3178 compose_ipfix_action(ctx, out_port);
3179 nl_msg_put_odp_port(ctx->odp_actions,
3180 OVS_ACTION_ATTR_OUTPUT,
3186 ctx->sflow_odp_port = odp_port;
3187 ctx->sflow_n_outputs++;
3188 ctx->nf_output_iface = ofp_port;
3191 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
3192 mirror_packet(ctx, xport->xbundle,
3193 xbundle_mirror_dst(xport->xbundle->xbridge,
3199 flow->vlan_tci = flow_vlan_tci;
3200 flow->pkt_mark = flow_pkt_mark;
3201 flow->nw_tos = flow_nw_tos;
3205 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3206 const struct xlate_bond_recirc *xr)
3208 compose_output_action__(ctx, ofp_port, xr, true);
3212 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3214 struct rule_dpif *old_rule = ctx->rule;
3215 ovs_be64 old_cookie = ctx->rule_cookie;
3216 const struct rule_actions *actions;
3218 if (ctx->xin->resubmit_stats) {
3219 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3225 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3226 actions = rule_dpif_get_actions(rule);
3227 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3228 ctx->rule_cookie = old_cookie;
3229 ctx->rule = old_rule;
3234 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3236 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3237 XLATE_REPORT_ERROR(ctx, "resubmit actions recursed over %d times",
3238 MAX_RESUBMIT_RECURSION);
3239 ctx->error = XLATE_RECURSION_TOO_DEEP;
3240 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3241 XLATE_REPORT_ERROR(ctx, "over %d resubmit actions", MAX_RESUBMITS);
3242 ctx->error = XLATE_TOO_MANY_RESUBMITS;
3243 } else if (ctx->odp_actions->size > UINT16_MAX) {
3244 XLATE_REPORT_ERROR(ctx, "resubmits yielded over 64 kB of actions");
3245 /* NOT an error, as we'll be slow-pathing the flow in this case? */
3246 ctx->exit = true; /* XXX: translation still terminated! */
3247 } else if (ctx->stack.size >= 65536) {
3248 XLATE_REPORT_ERROR(ctx, "resubmits yielded over 64 kB of stack");
3249 ctx->error = XLATE_STACK_TOO_DEEP;
3258 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3259 bool may_packet_in, bool honor_table_miss)
3261 /* Check if we need to recirculate before matching in a table. */
3262 if (ctx->was_mpls) {
3263 ctx_trigger_freeze(ctx);
3266 if (xlate_resubmit_resource_check(ctx)) {
3267 uint8_t old_table_id = ctx->table_id;
3268 struct rule_dpif *rule;
3270 ctx->table_id = table_id;
3272 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3273 ctx->tables_version,
3274 &ctx->xin->flow, ctx->xin->wc,
3275 ctx->xin->resubmit_stats,
3276 &ctx->table_id, in_port,
3277 may_packet_in, honor_table_miss);
3279 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3280 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3284 /* Fill in the cache entry here instead of xlate_recursively
3285 * to make the reference counting more explicit. We take a
3286 * reference in the lookups above if we are going to cache the
3288 if (ctx->xin->xcache) {
3289 struct xc_entry *entry;
3291 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3292 entry->u.rule = rule;
3293 rule_dpif_ref(rule);
3295 xlate_recursively(ctx, rule);
3298 ctx->table_id = old_table_id;
3304 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3305 struct ofputil_bucket *bucket)
3307 if (ctx->xin->resubmit_stats) {
3308 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3310 if (ctx->xin->xcache) {
3311 struct xc_entry *entry;
3313 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3314 entry->u.group.group = group_dpif_ref(group);
3315 entry->u.group.bucket = bucket;
3320 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3322 uint64_t action_list_stub[1024 / 8];
3323 struct ofpbuf action_list, action_set;
3324 struct flow old_flow = ctx->xin->flow;
3325 bool old_was_mpls = ctx->was_mpls;
3327 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3328 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3330 ofpacts_execute_action_set(&action_list, &action_set);
3332 do_xlate_actions(action_list.data, action_list.size, ctx);
3335 ofpbuf_uninit(&action_set);
3336 ofpbuf_uninit(&action_list);
3338 /* Check if need to recirculate. */
3339 if (ctx->freezing) {
3340 compose_recirculate_action(ctx);
3343 /* Roll back flow to previous state.
3344 * This is equivalent to cloning the packet for each bucket.
3346 * As a side effect any subsequently applied actions will
3347 * also effectively be applied to a clone of the packet taken
3348 * just before applying the all or indirect group.
3350 * Note that group buckets are action sets, hence they cannot modify the
3351 * main action set. Also any stack actions are ignored when executing an
3352 * action set, so group buckets cannot change the stack either.
3353 * However, we do allow resubmit actions in group buckets, which could
3354 * break the above assumptions. It is up to the controller to not mess up
3355 * with the action_set and stack in the tables resubmitted to from
3357 ctx->xin->flow = old_flow;
3359 /* The group bucket popping MPLS should have no effect after bucket
3361 ctx->was_mpls = old_was_mpls;
3363 /* The fact that the group bucket exits (for any reason) does not mean that
3364 * the translation after the group action should exit. Specifically, if
3365 * the group bucket freezes translation, the actions after the group action
3366 * must continue processing with the original, not the frozen packet! */
3371 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3373 struct ofputil_bucket *bucket;
3374 const struct ovs_list *buckets;
3376 group_dpif_get_buckets(group, &buckets);
3378 LIST_FOR_EACH (bucket, list_node, buckets) {
3379 xlate_group_bucket(ctx, bucket);
3381 xlate_group_stats(ctx, group, NULL);
3385 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3387 struct ofputil_bucket *bucket;
3389 bucket = group_first_live_bucket(ctx, group, 0);
3391 xlate_group_bucket(ctx, bucket);
3392 xlate_group_stats(ctx, group, bucket);
3397 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3399 struct flow_wildcards *wc = ctx->wc;
3400 struct ofputil_bucket *bucket;
3403 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3404 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3405 bucket = group_best_live_bucket(ctx, group, basis);
3407 xlate_group_bucket(ctx, bucket);
3408 xlate_group_stats(ctx, group, bucket);
3413 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3415 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3416 const struct field_array *fields;
3417 struct ofputil_bucket *bucket;
3421 fields = group_dpif_get_fields(group);
3422 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3424 /* Determine which fields to hash */
3425 for (i = 0; i < MFF_N_IDS; i++) {
3426 if (bitmap_is_set(fields->used.bm, i)) {
3427 const struct mf_field *mf;
3429 /* If the field is already present in 'hash_fields' then
3430 * this loop has already checked that it and its pre-requisites
3431 * are present in the flow and its pre-requisites have
3432 * already been added to 'hash_fields'. There is nothing more
3433 * to do here and as an optimisation the loop can continue. */
3434 if (bitmap_is_set(hash_fields.bm, i)) {
3440 /* Only hash a field if it and its pre-requisites are present
3442 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3446 /* Hash both the field and its pre-requisites */
3447 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3451 /* Hash the fields */
3452 for (i = 0; i < MFF_N_IDS; i++) {
3453 if (bitmap_is_set(hash_fields.bm, i)) {
3454 const struct mf_field *mf = mf_from_id(i);
3455 union mf_value value;
3458 mf_get_value(mf, &ctx->xin->flow, &value);
3459 /* This seems inefficient but so does apply_mask() */
3460 for (j = 0; j < mf->n_bytes; j++) {
3461 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3463 basis = hash_bytes(&value, mf->n_bytes, basis);
3465 /* For tunnels, hash in whether the field is present. */
3466 if (mf_is_tun_metadata(mf)) {
3467 basis = hash_boolean(mf_is_set(mf, &ctx->xin->flow), basis);
3470 mf_mask_field(mf, &ctx->wc->masks);
3474 bucket = group_best_live_bucket(ctx, group, basis);
3476 xlate_group_bucket(ctx, bucket);
3477 xlate_group_stats(ctx, group, bucket);
3482 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3484 const char *selection_method = group_dpif_get_selection_method(group);
3486 if (selection_method[0] == '\0') {
3487 xlate_default_select_group(ctx, group);
3488 } else if (!strcasecmp("hash", selection_method)) {
3489 xlate_hash_fields_select_group(ctx, group);
3491 /* Parsing of groups should ensure this never happens */
3497 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3499 bool was_in_group = ctx->in_group;
3500 ctx->in_group = true;
3502 switch (group_dpif_get_type(group)) {
3504 case OFPGT11_INDIRECT:
3505 xlate_all_group(ctx, group);
3507 case OFPGT11_SELECT:
3508 xlate_select_group(ctx, group);
3511 xlate_ff_group(ctx, group);
3516 group_dpif_unref(group);
3518 ctx->in_group = was_in_group;
3522 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3524 if (xlate_resubmit_resource_check(ctx)) {
3525 struct group_dpif *group;
3528 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3530 xlate_group_action__(ctx, group);
3540 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3541 const struct ofpact_resubmit *resubmit)
3545 bool may_packet_in = false;
3546 bool honor_table_miss = false;
3548 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3549 /* Still allow missed packets to be sent to the controller
3550 * if resubmitting from an internal table. */
3551 may_packet_in = true;
3552 honor_table_miss = true;
3555 in_port = resubmit->in_port;
3556 if (in_port == OFPP_IN_PORT) {
3557 in_port = ctx->xin->flow.in_port.ofp_port;
3560 table_id = resubmit->table_id;
3561 if (table_id == 255) {
3562 table_id = ctx->table_id;
3565 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3570 flood_packets(struct xlate_ctx *ctx, bool all)
3572 const struct xport *xport;
3574 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3575 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3580 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3581 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3582 compose_output_action(ctx, xport->ofp_port, NULL);
3586 ctx->nf_output_iface = NF_OUT_FLOOD;
3590 execute_controller_action(struct xlate_ctx *ctx, int len,
3591 enum ofp_packet_in_reason reason,
3592 uint16_t controller_id)
3594 struct dp_packet *packet;
3596 ctx->xout->slow |= SLOW_CONTROLLER;
3597 xlate_commit_actions(ctx);
3598 if (!ctx->xin->packet) {
3602 packet = dp_packet_clone(ctx->xin->packet);
3604 odp_execute_actions(NULL, &packet, 1, false,
3605 ctx->odp_actions->data, ctx->odp_actions->size, NULL);
3607 /* A packet sent by an action in a table-miss rule is considered an
3608 * explicit table miss. OpenFlow before 1.3 doesn't have that concept so
3609 * it will get translated back to OFPR_ACTION for those versions. */
3610 if (reason == OFPR_ACTION
3611 && ctx->rule && rule_dpif_is_table_miss(ctx->rule)) {
3612 reason = OFPR_EXPLICIT_MISS;
3615 size_t packet_len = dp_packet_size(packet);
3617 struct ofproto_async_msg *am = xmalloc(sizeof *am);
3618 *am = (struct ofproto_async_msg) {
3619 .controller_id = controller_id,
3620 .oam = OAM_PACKET_IN,
3623 .packet = dp_packet_steal_data(packet),
3626 .table_id = ctx->table_id,
3627 .cookie = ctx->rule_cookie,
3632 flow_get_metadata(&ctx->xin->flow, &am->pin.up.flow_metadata);
3634 ofproto_dpif_send_async_msg(ctx->xbridge->ofproto, am);
3635 dp_packet_delete(packet);
3639 compose_recirculate_action__(struct xlate_ctx *ctx, uint8_t table)
3641 struct frozen_metadata md;
3644 frozen_metadata_from_flow(&md, &ctx->xin->flow);
3646 ovs_assert(ctx->freezing);
3648 struct frozen_state state = {
3650 .ofproto_uuid = *ofproto_dpif_get_uuid(ctx->xbridge->ofproto),
3652 .stack = ctx->stack.data,
3653 .n_stack = ctx->stack.size / sizeof(union mf_subvalue),
3654 .mirrors = ctx->mirrors,
3655 .conntracked = ctx->conntracked,
3656 .ofpacts = ctx->frozen_actions.data,
3657 .ofpacts_len = ctx->frozen_actions.size,
3658 .action_set = ctx->action_set.data,
3659 .action_set_len = ctx->action_set.size,
3662 /* Allocate a unique recirc id for the given metadata state in the
3663 * flow. An existing id, with a new reference to the corresponding
3664 * recirculation context, will be returned if possible.
3665 * The life-cycle of this recirc id is managed by associating it
3666 * with the udpif key ('ukey') created for each new datapath flow. */
3667 id = recirc_alloc_id_ctx(&state);
3669 XLATE_REPORT_ERROR(ctx, "Failed to allocate recirculation id");
3670 ctx->error = XLATE_NO_RECIRCULATION_CONTEXT;
3673 recirc_refs_add(&ctx->xout->recircs, id);
3675 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3677 /* Undo changes done by freezing. */
3678 ctx_cancel_freeze(ctx);
3681 /* Called only when we're freezing. */
3683 compose_recirculate_action(struct xlate_ctx *ctx)
3685 xlate_commit_actions(ctx);
3686 compose_recirculate_action__(ctx, 0);
3689 /* Fork the pipeline here. The current packet will continue processing the
3690 * current action list. A clone of the current packet will recirculate, skip
3691 * the remainder of the current action list and asynchronously resume pipeline
3692 * processing in 'table' with the current metadata and action set. */
3694 compose_recirculate_and_fork(struct xlate_ctx *ctx, uint8_t table)
3696 ctx->freezing = true;
3697 compose_recirculate_action__(ctx, table);
3701 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3703 struct flow *flow = &ctx->xin->flow;
3706 ovs_assert(eth_type_mpls(mpls->ethertype));
3708 n = flow_count_mpls_labels(flow, ctx->wc);
3710 xlate_commit_actions(ctx);
3711 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3712 if (ctx->xin->packet != NULL) {
3713 XLATE_REPORT_ERROR(ctx, "bridge %s: dropping packet on which an "
3714 "MPLS push action can't be performed as it would "
3715 "have more MPLS LSEs than the %d supported.",
3716 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3718 ctx->error = XLATE_TOO_MANY_MPLS_LABELS;
3722 flow_push_mpls(flow, n, mpls->ethertype, ctx->wc);
3726 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3728 struct flow *flow = &ctx->xin->flow;
3729 int n = flow_count_mpls_labels(flow, ctx->wc);
3731 if (flow_pop_mpls(flow, n, eth_type, ctx->wc)) {
3732 if (ctx->xbridge->support.odp.recirc) {
3733 ctx->was_mpls = true;
3735 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3736 if (ctx->xin->packet != NULL) {
3737 XLATE_REPORT_ERROR(ctx, "bridge %s: dropping packet on which an "
3738 "MPLS pop action can't be performed as it has "
3739 "more MPLS LSEs than the %d supported.",
3740 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3742 ctx->error = XLATE_TOO_MANY_MPLS_LABELS;
3743 ofpbuf_clear(ctx->odp_actions);
3748 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3750 struct flow *flow = &ctx->xin->flow;
3752 if (!is_ip_any(flow)) {
3756 ctx->wc->masks.nw_ttl = 0xff;
3757 if (flow->nw_ttl > 1) {
3763 for (i = 0; i < ids->n_controllers; i++) {
3764 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3768 /* Stop processing for current table. */
3774 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3776 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3777 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3778 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3783 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3785 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3786 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3787 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3792 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3794 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3795 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3796 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3801 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3803 struct flow *flow = &ctx->xin->flow;
3805 if (eth_type_mpls(flow->dl_type)) {
3806 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3808 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3811 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3814 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3818 /* Stop processing for current table. */
3823 xlate_output_action(struct xlate_ctx *ctx,
3824 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3826 ofp_port_t prev_nf_output_iface = ctx->nf_output_iface;
3828 ctx->nf_output_iface = NF_OUT_DROP;
3832 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3835 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3836 0, may_packet_in, true);
3842 flood_packets(ctx, false);
3845 flood_packets(ctx, true);
3847 case OFPP_CONTROLLER:
3848 execute_controller_action(ctx, max_len,
3849 (ctx->in_group ? OFPR_GROUP
3850 : ctx->in_action_set ? OFPR_ACTION_SET
3858 if (port != ctx->xin->flow.in_port.ofp_port) {
3859 compose_output_action(ctx, port, NULL);
3861 xlate_report(ctx, "skipping output to input port");
3866 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3867 ctx->nf_output_iface = NF_OUT_FLOOD;
3868 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3869 ctx->nf_output_iface = prev_nf_output_iface;
3870 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3871 ctx->nf_output_iface != NF_OUT_FLOOD) {
3872 ctx->nf_output_iface = NF_OUT_MULTI;
3877 xlate_output_reg_action(struct xlate_ctx *ctx,
3878 const struct ofpact_output_reg *or)
3880 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3881 if (port <= UINT16_MAX) {
3882 union mf_subvalue value;
3884 memset(&value, 0xff, sizeof value);
3885 mf_write_subfield_flow(&or->src, &value, &ctx->wc->masks);
3886 xlate_output_action(ctx, u16_to_ofp(port),
3887 or->max_len, false);
3892 xlate_enqueue_action(struct xlate_ctx *ctx,
3893 const struct ofpact_enqueue *enqueue)
3895 ofp_port_t ofp_port = enqueue->port;
3896 uint32_t queue_id = enqueue->queue;
3897 uint32_t flow_priority, priority;
3900 /* Translate queue to priority. */
3901 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3903 /* Fall back to ordinary output action. */
3904 xlate_output_action(ctx, enqueue->port, 0, false);
3908 /* Check output port. */
3909 if (ofp_port == OFPP_IN_PORT) {
3910 ofp_port = ctx->xin->flow.in_port.ofp_port;
3911 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3915 /* Add datapath actions. */
3916 flow_priority = ctx->xin->flow.skb_priority;
3917 ctx->xin->flow.skb_priority = priority;
3918 compose_output_action(ctx, ofp_port, NULL);
3919 ctx->xin->flow.skb_priority = flow_priority;
3921 /* Update NetFlow output port. */
3922 if (ctx->nf_output_iface == NF_OUT_DROP) {
3923 ctx->nf_output_iface = ofp_port;
3924 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3925 ctx->nf_output_iface = NF_OUT_MULTI;
3930 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3932 uint32_t skb_priority;
3934 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3935 ctx->xin->flow.skb_priority = skb_priority;
3937 /* Couldn't translate queue to a priority. Nothing to do. A warning
3938 * has already been logged. */
3943 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3945 const struct xbridge *xbridge = xbridge_;
3956 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3959 port = get_ofp_port(xbridge, ofp_port);
3960 return port ? port->may_enable : false;
3965 xlate_bundle_action(struct xlate_ctx *ctx,
3966 const struct ofpact_bundle *bundle)
3970 port = bundle_execute(bundle, &ctx->xin->flow, ctx->wc, slave_enabled_cb,
3971 CONST_CAST(struct xbridge *, ctx->xbridge));
3972 if (bundle->dst.field) {
3973 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow, ctx->wc);
3975 xlate_output_action(ctx, port, 0, false);
3980 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3981 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3983 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3984 if (ctx->xin->may_learn) {
3985 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3990 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3992 learn_mask(learn, ctx->wc);
3994 if (ctx->xin->xcache) {
3995 struct xc_entry *entry;
3997 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3998 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3999 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
4000 entry->u.learn.ofpacts = ofpbuf_new(64);
4001 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
4002 entry->u.learn.ofpacts);
4003 } else if (ctx->xin->may_learn) {
4004 uint64_t ofpacts_stub[1024 / 8];
4005 struct ofputil_flow_mod fm;
4006 struct ofpbuf ofpacts;
4008 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4009 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
4010 ofpbuf_uninit(&ofpacts);
4015 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
4016 uint16_t idle_timeout, uint16_t hard_timeout)
4018 if (tcp_flags & (TCP_FIN | TCP_RST)) {
4019 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
4024 xlate_fin_timeout(struct xlate_ctx *ctx,
4025 const struct ofpact_fin_timeout *oft)
4028 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
4029 oft->fin_idle_timeout, oft->fin_hard_timeout);
4030 if (ctx->xin->xcache) {
4031 struct xc_entry *entry;
4033 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
4034 /* XC_RULE already holds a reference on the rule, none is taken
4036 entry->u.fin.rule = ctx->rule;
4037 entry->u.fin.idle = oft->fin_idle_timeout;
4038 entry->u.fin.hard = oft->fin_hard_timeout;
4044 xlate_sample_action(struct xlate_ctx *ctx,
4045 const struct ofpact_sample *os)
4047 /* Scale the probability from 16-bit to 32-bit while representing
4048 * the same percentage. */
4049 uint32_t probability = (os->probability << 16) | os->probability;
4051 if (!ctx->xbridge->support.variable_length_userdata) {
4052 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4054 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
4055 "lacks support (needs Linux 3.10+ or kernel module from "
4060 xlate_commit_actions(ctx);
4062 union user_action_cookie cookie = {
4064 .type = USER_ACTION_COOKIE_FLOW_SAMPLE,
4065 .probability = os->probability,
4066 .collector_set_id = os->collector_set_id,
4067 .obs_domain_id = os->obs_domain_id,
4068 .obs_point_id = os->obs_point_id,
4071 compose_sample_action(ctx, probability, &cookie, sizeof cookie.flow_sample,
4076 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
4078 if (xport->config & (is_stp(&ctx->xin->flow)
4079 ? OFPUTIL_PC_NO_RECV_STP
4080 : OFPUTIL_PC_NO_RECV)) {
4084 /* Only drop packets here if both forwarding and learning are
4085 * disabled. If just learning is enabled, we need to have
4086 * OFPP_NORMAL and the learning action have a look at the packet
4087 * before we can drop it. */
4088 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
4089 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
4097 xlate_write_actions__(struct xlate_ctx *ctx,
4098 const struct ofpact *ofpacts, size_t ofpacts_len)
4100 /* Maintain actset_output depending on the contents of the action set:
4102 * - OFPP_UNSET, if there is no "output" action.
4104 * - The output port, if there is an "output" action and no "group"
4107 * - OFPP_UNSET, if there is a "group" action.
4109 if (!ctx->action_set_has_group) {
4110 const struct ofpact *a;
4111 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4112 if (a->type == OFPACT_OUTPUT) {
4113 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(a)->port;
4114 } else if (a->type == OFPACT_GROUP) {
4115 ctx->xin->flow.actset_output = OFPP_UNSET;
4116 ctx->action_set_has_group = true;
4122 ofpbuf_put(&ctx->action_set, ofpacts, ofpacts_len);
4126 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact_nest *a)
4128 xlate_write_actions__(ctx, a->actions, ofpact_nest_get_action_len(a));
4132 xlate_action_set(struct xlate_ctx *ctx)
4134 uint64_t action_list_stub[1024 / 64];
4135 struct ofpbuf action_list;
4137 ctx->in_action_set = true;
4138 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
4139 ofpacts_execute_action_set(&action_list, &ctx->action_set);
4140 /* Clear the action set, as it is not needed any more. */
4141 ofpbuf_clear(&ctx->action_set);
4142 do_xlate_actions(action_list.data, action_list.size, ctx);
4143 ctx->in_action_set = false;
4144 ofpbuf_uninit(&action_list);
4148 freeze_put_unroll_xlate(struct xlate_ctx *ctx)
4150 struct ofpact_unroll_xlate *unroll = ctx->frozen_actions.header;
4152 /* Restore the table_id and rule cookie for a potential PACKET
4155 (ctx->table_id != unroll->rule_table_id
4156 || ctx->rule_cookie != unroll->rule_cookie)) {
4157 unroll = ofpact_put_UNROLL_XLATE(&ctx->frozen_actions);
4158 unroll->rule_table_id = ctx->table_id;
4159 unroll->rule_cookie = ctx->rule_cookie;
4160 ctx->frozen_actions.header = unroll;
4165 /* Copy actions 'a' through 'end' to ctx->frozen_actions, which will be
4166 * executed after thawing. Inserts an UNROLL_XLATE action, if none is already
4167 * present, before any action that may depend on the current table ID or flow
4170 freeze_unroll_actions(const struct ofpact *a, const struct ofpact *end,
4171 struct xlate_ctx *ctx)
4173 for (; a < end; a = ofpact_next(a)) {
4175 case OFPACT_OUTPUT_REG:
4178 case OFPACT_CONTROLLER:
4179 case OFPACT_DEC_MPLS_TTL:
4180 case OFPACT_DEC_TTL:
4181 /* These actions may generate asynchronous messages, which include
4182 * table ID and flow cookie information. */
4183 freeze_put_unroll_xlate(ctx);
4186 case OFPACT_RESUBMIT:
4187 if (ofpact_get_RESUBMIT(a)->table_id == 0xff) {
4188 /* This resubmit action is relative to the current table, so we
4189 * need to track what table that is.*/
4190 freeze_put_unroll_xlate(ctx);
4194 case OFPACT_SET_TUNNEL:
4195 case OFPACT_REG_MOVE:
4196 case OFPACT_SET_FIELD:
4197 case OFPACT_STACK_PUSH:
4198 case OFPACT_STACK_POP:
4200 case OFPACT_WRITE_METADATA:
4201 case OFPACT_GOTO_TABLE:
4202 case OFPACT_ENQUEUE:
4203 case OFPACT_SET_VLAN_VID:
4204 case OFPACT_SET_VLAN_PCP:
4205 case OFPACT_STRIP_VLAN:
4206 case OFPACT_PUSH_VLAN:
4207 case OFPACT_SET_ETH_SRC:
4208 case OFPACT_SET_ETH_DST:
4209 case OFPACT_SET_IPV4_SRC:
4210 case OFPACT_SET_IPV4_DST:
4211 case OFPACT_SET_IP_DSCP:
4212 case OFPACT_SET_IP_ECN:
4213 case OFPACT_SET_IP_TTL:
4214 case OFPACT_SET_L4_SRC_PORT:
4215 case OFPACT_SET_L4_DST_PORT:
4216 case OFPACT_SET_QUEUE:
4217 case OFPACT_POP_QUEUE:
4218 case OFPACT_PUSH_MPLS:
4219 case OFPACT_POP_MPLS:
4220 case OFPACT_SET_MPLS_LABEL:
4221 case OFPACT_SET_MPLS_TC:
4222 case OFPACT_SET_MPLS_TTL:
4223 case OFPACT_MULTIPATH:
4226 case OFPACT_UNROLL_XLATE:
4227 case OFPACT_FIN_TIMEOUT:
4228 case OFPACT_CLEAR_ACTIONS:
4229 case OFPACT_WRITE_ACTIONS:
4232 case OFPACT_DEBUG_RECIRC:
4235 /* These may not generate PACKET INs. */
4239 case OFPACT_CONJUNCTION:
4240 /* These need not be copied for restoration. */
4243 /* Copy the action over. */
4244 ofpbuf_put(&ctx->frozen_actions, a, OFPACT_ALIGN(a->len));
4248 #define CHECK_MPLS_RECIRCULATION() \
4249 if (ctx->was_mpls) { \
4250 ctx_trigger_freeze(ctx); \
4253 #define CHECK_MPLS_RECIRCULATION_IF(COND) \
4255 CHECK_MPLS_RECIRCULATION(); \
4259 put_ct_mark(const struct flow *flow, struct flow *base_flow,
4260 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4267 odp_attr.key = flow->ct_mark;
4268 odp_attr.mask = wc->masks.ct_mark;
4270 if (odp_attr.mask && odp_attr.key != base_flow->ct_mark) {
4271 nl_msg_put_unspec(odp_actions, OVS_CT_ATTR_MARK, &odp_attr,
4277 put_ct_label(const struct flow *flow, struct flow *base_flow,
4278 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4280 if (!ovs_u128_is_zero(&wc->masks.ct_label)
4281 && !ovs_u128_equals(&flow->ct_label, &base_flow->ct_label)) {
4287 odp_ct_label = nl_msg_put_unspec_uninit(odp_actions,
4289 sizeof(*odp_ct_label));
4290 odp_ct_label->key = flow->ct_label;
4291 odp_ct_label->mask = wc->masks.ct_label;
4296 put_ct_helper(struct ofpbuf *odp_actions, struct ofpact_conntrack *ofc)
4299 if (ofc->alg == IPPORT_FTP) {
4300 nl_msg_put_string(odp_actions, OVS_CT_ATTR_HELPER, "ftp");
4302 VLOG_WARN("Cannot serialize ct_helper %d\n", ofc->alg);
4308 put_ct_nat(struct xlate_ctx *ctx)
4310 struct ofpact_nat *ofn = ctx->ct_nat_action;
4317 nat_offset = nl_msg_start_nested(ctx->odp_actions, OVS_CT_ATTR_NAT);
4318 if (ofn->flags & NX_NAT_F_SRC || ofn->flags & NX_NAT_F_DST) {
4319 nl_msg_put_flag(ctx->odp_actions, ofn->flags & NX_NAT_F_SRC
4320 ? OVS_NAT_ATTR_SRC : OVS_NAT_ATTR_DST);
4321 if (ofn->flags & NX_NAT_F_PERSISTENT) {
4322 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PERSISTENT);
4324 if (ofn->flags & NX_NAT_F_PROTO_HASH) {
4325 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PROTO_HASH);
4326 } else if (ofn->flags & NX_NAT_F_PROTO_RANDOM) {
4327 nl_msg_put_flag(ctx->odp_actions, OVS_NAT_ATTR_PROTO_RANDOM);
4329 if (ofn->range_af == AF_INET) {
4330 nl_msg_put_be32(ctx->odp_actions, OVS_NAT_ATTR_IP_MIN,
4331 ofn->range.addr.ipv4.min);
4332 if (ofn->range.addr.ipv4.max &&
4333 (ntohl(ofn->range.addr.ipv4.max)
4334 > ntohl(ofn->range.addr.ipv4.min))) {
4335 nl_msg_put_be32(ctx->odp_actions, OVS_NAT_ATTR_IP_MAX,
4336 ofn->range.addr.ipv4.max);
4338 } else if (ofn->range_af == AF_INET6) {
4339 nl_msg_put_unspec(ctx->odp_actions, OVS_NAT_ATTR_IP_MIN,
4340 &ofn->range.addr.ipv6.min,
4341 sizeof ofn->range.addr.ipv6.min);
4342 if (!ipv6_mask_is_any(&ofn->range.addr.ipv6.max) &&
4343 memcmp(&ofn->range.addr.ipv6.max, &ofn->range.addr.ipv6.min,
4344 sizeof ofn->range.addr.ipv6.max) > 0) {
4345 nl_msg_put_unspec(ctx->odp_actions, OVS_NAT_ATTR_IP_MAX,
4346 &ofn->range.addr.ipv6.max,
4347 sizeof ofn->range.addr.ipv6.max);
4350 if (ofn->range_af != AF_UNSPEC && ofn->range.proto.min) {
4351 nl_msg_put_u16(ctx->odp_actions, OVS_NAT_ATTR_PROTO_MIN,
4352 ofn->range.proto.min);
4353 if (ofn->range.proto.max &&
4354 ofn->range.proto.max > ofn->range.proto.min) {
4355 nl_msg_put_u16(ctx->odp_actions, OVS_NAT_ATTR_PROTO_MAX,
4356 ofn->range.proto.max);
4360 nl_msg_end_nested(ctx->odp_actions, nat_offset);
4364 compose_conntrack_action(struct xlate_ctx *ctx, struct ofpact_conntrack *ofc)
4366 ovs_u128 old_ct_label = ctx->base_flow.ct_label;
4367 uint32_t old_ct_mark = ctx->base_flow.ct_mark;
4371 /* Ensure that any prior actions are applied before composing the new
4372 * conntrack action. */
4373 xlate_commit_actions(ctx);
4375 /* Process nested actions first, to populate the key. */
4376 ctx->ct_nat_action = NULL;
4377 do_xlate_actions(ofc->actions, ofpact_ct_get_action_len(ofc), ctx);
4379 if (ofc->zone_src.field) {
4380 zone = mf_get_subfield(&ofc->zone_src, &ctx->xin->flow);
4382 zone = ofc->zone_imm;
4385 ct_offset = nl_msg_start_nested(ctx->odp_actions, OVS_ACTION_ATTR_CT);
4386 if (ofc->flags & NX_CT_F_COMMIT) {
4387 nl_msg_put_flag(ctx->odp_actions, OVS_CT_ATTR_COMMIT);
4389 nl_msg_put_u16(ctx->odp_actions, OVS_CT_ATTR_ZONE, zone);
4390 put_ct_mark(&ctx->xin->flow, &ctx->base_flow, ctx->odp_actions, ctx->wc);
4391 put_ct_label(&ctx->xin->flow, &ctx->base_flow, ctx->odp_actions, ctx->wc);
4392 put_ct_helper(ctx->odp_actions, ofc);
4394 ctx->ct_nat_action = NULL;
4395 nl_msg_end_nested(ctx->odp_actions, ct_offset);
4397 /* Restore the original ct fields in the key. These should only be exposed
4398 * after recirculation to another table. */
4399 ctx->base_flow.ct_mark = old_ct_mark;
4400 ctx->base_flow.ct_label = old_ct_label;
4402 if (ofc->recirc_table == NX_CT_RECIRC_NONE) {
4403 /* If we do not recirculate as part of this action, hide the results of
4404 * connection tracking from subsequent recirculations. */
4405 ctx->conntracked = false;
4407 /* Use ct_* fields from datapath during recirculation upcall. */
4408 ctx->conntracked = true;
4409 compose_recirculate_and_fork(ctx, ofc->recirc_table);
4414 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4415 struct xlate_ctx *ctx)
4417 struct flow_wildcards *wc = ctx->wc;
4418 struct flow *flow = &ctx->xin->flow;
4419 const struct ofpact *a;
4421 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4422 tnl_neigh_snoop(flow, wc, ctx->xbridge->name);
4424 /* dl_type already in the mask, not set below. */
4426 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4427 struct ofpact_controller *controller;
4428 const struct ofpact_metadata *metadata;
4429 const struct ofpact_set_field *set_field;
4430 const struct mf_field *mf;
4437 /* Check if need to store the remaining actions for later
4439 if (ctx->freezing) {
4440 freeze_unroll_actions(a, ofpact_end(ofpacts, ofpacts_len),
4448 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4449 ofpact_get_OUTPUT(a)->max_len, true);
4453 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4454 /* Group could not be found. */
4459 case OFPACT_CONTROLLER:
4460 controller = ofpact_get_CONTROLLER(a);
4461 execute_controller_action(ctx, controller->max_len,
4463 controller->controller_id);
4466 case OFPACT_ENQUEUE:
4467 memset(&wc->masks.skb_priority, 0xff,
4468 sizeof wc->masks.skb_priority);
4469 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4472 case OFPACT_SET_VLAN_VID:
4473 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4474 if (flow->vlan_tci & htons(VLAN_CFI) ||
4475 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4476 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4477 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4482 case OFPACT_SET_VLAN_PCP:
4483 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4484 if (flow->vlan_tci & htons(VLAN_CFI) ||
4485 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4486 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4487 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4488 << VLAN_PCP_SHIFT) | VLAN_CFI);
4492 case OFPACT_STRIP_VLAN:
4493 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4494 flow->vlan_tci = htons(0);
4497 case OFPACT_PUSH_VLAN:
4498 /* XXX 802.1AD(QinQ) */
4499 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4500 flow->vlan_tci = htons(VLAN_CFI);
4503 case OFPACT_SET_ETH_SRC:
4504 WC_MASK_FIELD(wc, dl_src);
4505 flow->dl_src = ofpact_get_SET_ETH_SRC(a)->mac;
4508 case OFPACT_SET_ETH_DST:
4509 WC_MASK_FIELD(wc, dl_dst);
4510 flow->dl_dst = ofpact_get_SET_ETH_DST(a)->mac;
4513 case OFPACT_SET_IPV4_SRC:
4514 CHECK_MPLS_RECIRCULATION();
4515 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4516 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4517 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4521 case OFPACT_SET_IPV4_DST:
4522 CHECK_MPLS_RECIRCULATION();
4523 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4524 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4525 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4529 case OFPACT_SET_IP_DSCP:
4530 CHECK_MPLS_RECIRCULATION();
4531 if (is_ip_any(flow)) {
4532 wc->masks.nw_tos |= IP_DSCP_MASK;
4533 flow->nw_tos &= ~IP_DSCP_MASK;
4534 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4538 case OFPACT_SET_IP_ECN:
4539 CHECK_MPLS_RECIRCULATION();
4540 if (is_ip_any(flow)) {
4541 wc->masks.nw_tos |= IP_ECN_MASK;
4542 flow->nw_tos &= ~IP_ECN_MASK;
4543 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4547 case OFPACT_SET_IP_TTL:
4548 CHECK_MPLS_RECIRCULATION();
4549 if (is_ip_any(flow)) {
4550 wc->masks.nw_ttl = 0xff;
4551 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4555 case OFPACT_SET_L4_SRC_PORT:
4556 CHECK_MPLS_RECIRCULATION();
4557 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4558 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4559 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4560 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4564 case OFPACT_SET_L4_DST_PORT:
4565 CHECK_MPLS_RECIRCULATION();
4566 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4567 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4568 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4569 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4573 case OFPACT_RESUBMIT:
4574 /* Freezing complicates resubmit. There are two cases:
4576 * - If mpls_pop has been executed, then the flow table lookup
4577 * as part of resubmit might depend on fields that can only
4578 * be obtained via recirculation, so the resubmit itself
4579 * triggers recirculation and we need to make sure that the
4580 * resubmit is executed again after recirculation.
4581 * Therefore, in this case we trigger recirculation and let
4582 * the code following this "switch" append the resubmit to
4583 * the post-recirculation actions.
4585 * - Otherwise, some action in the flow entry found by resubmit
4586 * might trigger freezing. If that happens, then we do not
4587 * want to execute the resubmit again during thawing, so we
4588 * want to skip back to the head of the loop to avoid that,
4589 * only adding any actions that follow the resubmit to the
4592 if (ctx->was_mpls) {
4593 ctx_trigger_freeze(ctx);
4596 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4599 case OFPACT_SET_TUNNEL:
4600 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4603 case OFPACT_SET_QUEUE:
4604 memset(&wc->masks.skb_priority, 0xff,
4605 sizeof wc->masks.skb_priority);
4606 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4609 case OFPACT_POP_QUEUE:
4610 memset(&wc->masks.skb_priority, 0xff,
4611 sizeof wc->masks.skb_priority);
4612 flow->skb_priority = ctx->orig_skb_priority;
4615 case OFPACT_REG_MOVE:
4616 CHECK_MPLS_RECIRCULATION_IF(
4617 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4618 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4619 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4622 case OFPACT_SET_FIELD:
4623 CHECK_MPLS_RECIRCULATION_IF(
4624 mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field));
4625 set_field = ofpact_get_SET_FIELD(a);
4626 mf = set_field->field;
4628 /* Set field action only ever overwrites packet's outermost
4629 * applicable header fields. Do nothing if no header exists. */
4630 if (mf->id == MFF_VLAN_VID) {
4631 wc->masks.vlan_tci |= htons(VLAN_CFI);
4632 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4635 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4636 /* 'dl_type' is already unwildcarded. */
4637 && !eth_type_mpls(flow->dl_type)) {
4640 /* A flow may wildcard nw_frag. Do nothing if setting a transport
4641 * header field on a packet that does not have them. */
4642 mf_mask_field_and_prereqs(mf, wc);
4643 if (mf_are_prereqs_ok(mf, flow)) {
4644 mf_set_flow_value_masked(mf, &set_field->value,
4645 &set_field->mask, flow);
4649 case OFPACT_STACK_PUSH:
4650 CHECK_MPLS_RECIRCULATION_IF(
4651 mf_is_l3_or_higher(ofpact_get_STACK_PUSH(a)->subfield.field));
4652 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4656 case OFPACT_STACK_POP:
4657 CHECK_MPLS_RECIRCULATION_IF(
4658 mf_is_l3_or_higher(ofpact_get_STACK_POP(a)->subfield.field));
4659 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4663 case OFPACT_PUSH_MPLS:
4664 /* Recirculate if it is an IP packet with a zero ttl. This may
4665 * indicate that the packet was previously MPLS and an MPLS pop
4666 * action converted it to IP. In this case recirculating should
4667 * reveal the IP TTL which is used as the basis for a new MPLS
4669 CHECK_MPLS_RECIRCULATION_IF(
4670 !flow_count_mpls_labels(flow, wc)
4671 && flow->nw_ttl == 0
4672 && is_ip_any(flow));
4673 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4676 case OFPACT_POP_MPLS:
4677 CHECK_MPLS_RECIRCULATION();
4678 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4681 case OFPACT_SET_MPLS_LABEL:
4682 CHECK_MPLS_RECIRCULATION();
4683 compose_set_mpls_label_action(
4684 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4687 case OFPACT_SET_MPLS_TC:
4688 CHECK_MPLS_RECIRCULATION();
4689 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4692 case OFPACT_SET_MPLS_TTL:
4693 CHECK_MPLS_RECIRCULATION();
4694 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4697 case OFPACT_DEC_MPLS_TTL:
4698 CHECK_MPLS_RECIRCULATION();
4699 if (compose_dec_mpls_ttl_action(ctx)) {
4704 case OFPACT_DEC_TTL:
4705 CHECK_MPLS_RECIRCULATION();
4706 wc->masks.nw_ttl = 0xff;
4707 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4713 /* Nothing to do. */
4716 case OFPACT_MULTIPATH:
4717 CHECK_MPLS_RECIRCULATION();
4718 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4722 CHECK_MPLS_RECIRCULATION();
4723 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4726 case OFPACT_OUTPUT_REG:
4727 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4731 CHECK_MPLS_RECIRCULATION();
4732 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4735 case OFPACT_CONJUNCTION: {
4736 /* A flow with a "conjunction" action represents part of a special
4737 * kind of "set membership match". Such a flow should not actually
4738 * get executed, but it could via, say, a "packet-out", even though
4739 * that wouldn't be useful. Log it to help debugging. */
4740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4741 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4749 case OFPACT_UNROLL_XLATE: {
4750 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4752 /* Restore translation context data that was stored earlier. */
4753 ctx->table_id = unroll->rule_table_id;
4754 ctx->rule_cookie = unroll->rule_cookie;
4757 case OFPACT_FIN_TIMEOUT:
4758 CHECK_MPLS_RECIRCULATION();
4759 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4760 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4763 case OFPACT_CLEAR_ACTIONS:
4764 ofpbuf_clear(&ctx->action_set);
4765 ctx->xin->flow.actset_output = OFPP_UNSET;
4766 ctx->action_set_has_group = false;
4769 case OFPACT_WRITE_ACTIONS:
4770 xlate_write_actions(ctx, ofpact_get_WRITE_ACTIONS(a));
4773 case OFPACT_WRITE_METADATA:
4774 metadata = ofpact_get_WRITE_METADATA(a);
4775 flow->metadata &= ~metadata->mask;
4776 flow->metadata |= metadata->metadata & metadata->mask;
4780 /* Not implemented yet. */
4783 case OFPACT_GOTO_TABLE: {
4784 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4786 ovs_assert(ctx->table_id < ogt->table_id);
4788 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4789 ogt->table_id, true, true);
4794 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4798 CHECK_MPLS_RECIRCULATION();
4799 compose_conntrack_action(ctx, ofpact_get_CT(a));
4803 /* This will be processed by compose_conntrack_action(). */
4804 ctx->ct_nat_action = ofpact_get_NAT(a);
4807 case OFPACT_DEBUG_RECIRC:
4808 ctx_trigger_freeze(ctx);
4813 /* Check if need to store this and the remaining actions for later
4815 if (!ctx->error && ctx->exit && ctx_first_frozen_action(ctx)) {
4816 freeze_unroll_actions(a, ofpact_end(ofpacts, ofpacts_len), ctx);
4823 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4824 const struct flow *flow, ofp_port_t in_port,
4825 struct rule_dpif *rule, uint16_t tcp_flags,
4826 const struct dp_packet *packet, struct flow_wildcards *wc,
4827 struct ofpbuf *odp_actions)
4829 xin->ofproto = ofproto;
4831 xin->flow.in_port.ofp_port = in_port;
4832 xin->flow.actset_output = OFPP_UNSET;
4833 xin->packet = packet;
4834 xin->may_learn = packet != NULL;
4837 xin->ofpacts = NULL;
4838 xin->ofpacts_len = 0;
4839 xin->tcp_flags = tcp_flags;
4840 xin->resubmit_hook = NULL;
4841 xin->report_hook = NULL;
4842 xin->resubmit_stats = NULL;
4846 xin->odp_actions = odp_actions;
4848 /* Do recirc lookup. */
4849 xin->frozen_state = NULL;
4850 if (flow->recirc_id) {
4851 const struct recirc_id_node *node
4852 = recirc_id_node_find(flow->recirc_id);
4854 xin->frozen_state = &node->state;
4860 xlate_out_uninit(struct xlate_out *xout)
4863 recirc_refs_unref(&xout->recircs);
4867 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4868 * into datapath actions, using 'ctx', and discards the datapath actions. */
4870 xlate_actions_for_side_effects(struct xlate_in *xin)
4872 struct xlate_out xout;
4873 enum xlate_error error;
4875 error = xlate_actions(xin, &xout);
4877 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
4879 VLOG_WARN_RL(&rl, "xlate_actions failed (%s)!", xlate_strerror(error));
4882 xlate_out_uninit(&xout);
4885 static struct skb_priority_to_dscp *
4886 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4888 struct skb_priority_to_dscp *pdscp;
4891 hash = hash_int(skb_priority, 0);
4892 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4893 if (pdscp->skb_priority == skb_priority) {
4901 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4904 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4905 *dscp = pdscp ? pdscp->dscp : 0;
4906 return pdscp != NULL;
4910 count_skb_priorities(const struct xport *xport)
4912 return hmap_count(&xport->skb_priorities);
4916 clear_skb_priorities(struct xport *xport)
4918 struct skb_priority_to_dscp *pdscp, *next;
4920 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4921 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4927 actions_output_to_local_port(const struct xlate_ctx *ctx)
4929 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4930 const struct nlattr *a;
4933 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->odp_actions->data,
4934 ctx->odp_actions->size) {
4935 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4936 && nl_attr_get_odp_port(a) == local_odp_port) {
4943 #if defined(__linux__)
4944 /* Returns the maximum number of packets that the Linux kernel is willing to
4945 * queue up internally to certain kinds of software-implemented ports, or the
4946 * default (and rarely modified) value if it cannot be determined. */
4948 netdev_max_backlog(void)
4950 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4951 static int max_backlog = 1000; /* The normal default value. */
4953 if (ovsthread_once_start(&once)) {
4954 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4958 stream = fopen(filename, "r");
4960 VLOG_INFO("%s: open failed (%s)", filename, ovs_strerror(errno));
4962 if (fscanf(stream, "%d", &n) != 1) {
4963 VLOG_WARN("%s: read error", filename);
4964 } else if (n <= 100) {
4965 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4971 ovsthread_once_done(&once);
4973 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4979 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4982 count_output_actions(const struct ofpbuf *odp_actions)
4984 const struct nlattr *a;
4988 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4989 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4995 #endif /* defined(__linux__) */
4997 /* Returns true if 'odp_actions' contains more output actions than the datapath
4998 * can reliably handle in one go. On Linux, this is the value of the
4999 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
5000 * packets that the kernel is willing to queue up for processing while the
5001 * datapath is processing a set of actions. */
5003 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
5006 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
5007 && count_output_actions(odp_actions) > netdev_max_backlog());
5009 /* OSes other than Linux might have similar limits, but we don't know how
5010 * to determine them.*/
5016 xlate_wc_init(struct xlate_ctx *ctx)
5018 flow_wildcards_init_catchall(ctx->wc);
5020 /* Some fields we consider to always be examined. */
5021 WC_MASK_FIELD(ctx->wc, in_port);
5022 WC_MASK_FIELD(ctx->wc, dl_type);
5023 if (is_ip_any(&ctx->xin->flow)) {
5024 WC_MASK_FIELD_MASK(ctx->wc, nw_frag, FLOW_NW_FRAG_MASK);
5027 if (ctx->xbridge->support.odp.recirc) {
5028 /* Always exactly match recirc_id when datapath supports
5030 WC_MASK_FIELD(ctx->wc, recirc_id);
5033 if (ctx->xbridge->netflow) {
5034 netflow_mask_wc(&ctx->xin->flow, ctx->wc);
5037 tnl_wc_init(&ctx->xin->flow, ctx->wc);
5041 xlate_wc_finish(struct xlate_ctx *ctx)
5043 /* Clear the metadata and register wildcard masks, because we won't
5044 * use non-header fields as part of the cache. */
5045 flow_wildcards_clear_non_packet_fields(ctx->wc);
5047 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
5048 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
5049 * represent these fields. The datapath interface, on the other hand,
5050 * represents them with just 8 bits each. This means that if the high
5051 * 8 bits of the masks for these fields somehow become set, then they
5052 * will get chopped off by a round trip through the datapath, and
5053 * revalidation will spot that as an inconsistency and delete the flow.
5054 * Avoid the problem here by making sure that only the low 8 bits of
5055 * either field can be unwildcarded for ICMP.
5057 if (is_icmpv4(&ctx->xin->flow) || is_icmpv6(&ctx->xin->flow)) {
5058 ctx->wc->masks.tp_src &= htons(UINT8_MAX);
5059 ctx->wc->masks.tp_dst &= htons(UINT8_MAX);
5061 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
5062 if (ctx->wc->masks.vlan_tci) {
5063 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI);
5067 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
5069 * The caller must take responsibility for eventually freeing 'xout', with
5070 * xlate_out_uninit().
5071 * Returns 'XLATE_OK' if translation was successful. In case of an error an
5072 * empty set of actions will be returned in 'xin->odp_actions' (if non-NULL),
5073 * so that most callers may ignore the return value and transparently install a
5074 * drop flow when the translation fails. */
5076 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
5078 *xout = (struct xlate_out) {
5080 .recircs = RECIRC_REFS_EMPTY_INITIALIZER,
5083 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5084 struct xbridge *xbridge = xbridge_lookup(xcfg, xin->ofproto);
5086 return XLATE_BRIDGE_NOT_FOUND;
5089 struct flow *flow = &xin->flow;
5091 union mf_subvalue stack_stub[1024 / sizeof(union mf_subvalue)];
5092 uint64_t action_set_stub[1024 / 8];
5093 uint64_t frozen_actions_stub[1024 / 8];
5094 struct flow_wildcards scratch_wc;
5095 uint64_t actions_stub[256 / 8];
5096 struct ofpbuf scratch_actions = OFPBUF_STUB_INITIALIZER(actions_stub);
5097 struct xlate_ctx ctx = {
5101 .orig_tunnel_ipv6_dst = flow_tnl_dst(&flow->tunnel),
5103 .stack = OFPBUF_STUB_INITIALIZER(stack_stub),
5105 .wc = xin->wc ? xin->wc : &scratch_wc,
5106 .odp_actions = xin->odp_actions ? xin->odp_actions : &scratch_actions,
5108 .recurse = xin->recurse,
5109 .resubmits = xin->resubmits,
5111 .in_action_set = false,
5114 .rule_cookie = OVS_BE64_MAX,
5115 .orig_skb_priority = flow->skb_priority,
5116 .sflow_n_outputs = 0,
5117 .sflow_odp_port = 0,
5118 .nf_output_iface = NF_OUT_DROP,
5124 .frozen_actions = OFPBUF_STUB_INITIALIZER(frozen_actions_stub),
5127 .conntracked = false,
5129 .ct_nat_action = NULL,
5131 .action_set_has_group = false,
5132 .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
5135 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
5136 * the packet as the datapath will treat it for output actions:
5138 * - Our datapath doesn't retain tunneling information without us
5139 * re-setting it, so clear the tunnel data.
5141 * - For VLAN splinters, a higher layer may pretend that the packet
5142 * came in on 'flow->in_port.ofp_port' with 'flow->vlan_tci'
5143 * attached, because that's how we want to treat it from an OpenFlow
5144 * perspective. But from the datapath's perspective it actually came
5145 * in on a VLAN device without any VLAN attached. So here we put the
5146 * datapath's view of the VLAN information in 'base_flow' to ensure
5147 * correct treatment.
5149 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
5150 if (flow->in_port.ofp_port
5151 != vsp_realdev_to_vlandev(xbridge->ofproto,
5152 flow->in_port.ofp_port,
5154 ctx.base_flow.vlan_tci = 0;
5157 ofpbuf_reserve(ctx.odp_actions, NL_A_U32_SIZE);
5159 xlate_wc_init(&ctx);
5162 COVERAGE_INC(xlate_actions);
5164 if (xin->frozen_state) {
5165 const struct frozen_state *state = xin->frozen_state;
5167 xlate_report(&ctx, "Thawing frozen state:");
5169 if (xin->ofpacts_len > 0 || ctx.rule) {
5170 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5171 const char *conflict = xin->ofpacts_len ? "actions" : "rule";
5173 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!", conflict);
5174 xlate_report(&ctx, "- Recirculation conflict (%s)!", conflict);
5175 ctx.error = XLATE_RECIRCULATION_CONFLICT;
5179 /* Set the bridge for post-recirculation processing if needed. */
5180 if (!uuid_equals(ofproto_dpif_get_uuid(ctx.xbridge->ofproto),
5181 &state->ofproto_uuid)) {
5182 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5183 const struct xbridge *new_bridge
5184 = xbridge_lookup_by_uuid(xcfg, &state->ofproto_uuid);
5186 if (OVS_UNLIKELY(!new_bridge)) {
5187 /* Drop the packet if the bridge cannot be found. */
5188 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5189 VLOG_WARN_RL(&rl, "Frozen bridge no longer exists.");
5190 xlate_report(&ctx, "- Frozen bridge no longer exists.");
5191 ctx.error = XLATE_BRIDGE_NOT_FOUND;
5194 ctx.xbridge = new_bridge;
5197 /* Set the thawed table id. Note: A table lookup is done only if there
5198 * are no frozen actions. */
5199 ctx.table_id = state->table_id;
5200 xlate_report(&ctx, "- Resuming from table %"PRIu8, ctx.table_id);
5202 if (!state->conntracked) {
5203 clear_conntrack(flow);
5206 /* Restore pipeline metadata. May change flow's in_port and other
5207 * metadata to the values that existed when freezing was triggered. */
5208 frozen_metadata_to_flow(&state->metadata, flow);
5210 /* Restore stack, if any. */
5212 ofpbuf_put(&ctx.stack, state->stack,
5213 state->n_stack * sizeof *state->stack);
5216 /* Restore mirror state. */
5217 ctx.mirrors = state->mirrors;
5219 /* Restore action set, if any. */
5220 if (state->action_set_len) {
5221 xlate_report_actions(&ctx, "- Restoring action set",
5222 state->action_set, state->action_set_len);
5224 flow->actset_output = OFPP_UNSET;
5225 xlate_write_actions__(&ctx, state->action_set,
5226 state->action_set_len);
5229 /* Restore frozen actions. If there are no actions, processing will
5230 * start with a lookup in the table set above. */
5231 xin->ofpacts = state->ofpacts;
5232 xin->ofpacts_len = state->ofpacts_len;
5233 if (state->ofpacts_len) {
5234 xlate_report_actions(&ctx, "- Restoring actions",
5235 xin->ofpacts, xin->ofpacts_len);
5237 } else if (OVS_UNLIKELY(flow->recirc_id)) {
5238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5240 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
5242 ctx.error = XLATE_NO_RECIRCULATION_CONTEXT;
5245 /* The bridge is now known so obtain its table version. */
5246 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
5248 if (!xin->ofpacts && !ctx.rule) {
5249 ctx.rule = rule_dpif_lookup_from_table(
5250 ctx.xbridge->ofproto, ctx.tables_version, flow, xin->wc,
5251 ctx.xin->resubmit_stats, &ctx.table_id,
5252 flow->in_port.ofp_port, true, true);
5253 if (ctx.xin->resubmit_stats) {
5254 rule_dpif_credit_stats(ctx.rule, ctx.xin->resubmit_stats);
5256 if (ctx.xin->xcache) {
5257 struct xc_entry *entry;
5259 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
5260 entry->u.rule = ctx.rule;
5261 rule_dpif_ref(ctx.rule);
5264 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
5265 ctx.xin->resubmit_hook(ctx.xin, ctx.rule, 0);
5269 /* Get the proximate input port of the packet. (If xin->frozen_state,
5270 * flow->in_port is the ultimate input port of the packet.) */
5271 struct xport *in_port = get_ofp_port(xbridge,
5272 ctx.base_flow.in_port.ofp_port);
5274 /* Tunnel stats only for not-thawed packets. */
5275 if (!xin->frozen_state && in_port && in_port->is_tunnel) {
5276 if (ctx.xin->resubmit_stats) {
5277 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
5279 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
5282 if (ctx.xin->xcache) {
5283 struct xc_entry *entry;
5285 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
5286 entry->u.dev.rx = netdev_ref(in_port->netdev);
5287 entry->u.dev.bfd = bfd_ref(in_port->bfd);
5291 if (!xin->frozen_state && process_special(&ctx, in_port)) {
5292 /* process_special() did all the processing for this packet.
5294 * We do not perform special processing on thawed packets, since that
5295 * was done before they were frozen and should not be redone. */
5296 } else if (in_port && in_port->xbundle
5297 && xbundle_mirror_out(xbridge, in_port->xbundle)) {
5298 if (ctx.xin->packet != NULL) {
5299 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5300 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5301 "%s, which is reserved exclusively for mirroring",
5302 ctx.xbridge->name, in_port->xbundle->name);
5305 /* Sampling is done on initial reception; don't redo after thawing. */
5306 unsigned int user_cookie_offset = 0;
5307 if (!xin->frozen_state) {
5308 user_cookie_offset = compose_sflow_action(&ctx);
5309 compose_ipfix_action(&ctx, ODPP_NONE);
5311 size_t sample_actions_len = ctx.odp_actions->size;
5313 if (tnl_process_ecn(flow)
5314 && (!in_port || may_receive(in_port, &ctx))) {
5315 const struct ofpact *ofpacts;
5319 ofpacts = xin->ofpacts;
5320 ofpacts_len = xin->ofpacts_len;
5321 } else if (ctx.rule) {
5322 const struct rule_actions *actions
5323 = rule_dpif_get_actions(ctx.rule);
5324 ofpacts = actions->ofpacts;
5325 ofpacts_len = actions->ofpacts_len;
5326 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
5331 mirror_ingress_packet(&ctx);
5332 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
5337 /* We've let OFPP_NORMAL and the learning action look at the
5338 * packet, so cancel all actions and freezing if forwarding is
5340 if (in_port && (!xport_stp_forward_state(in_port) ||
5341 !xport_rstp_forward_state(in_port))) {
5342 ctx.odp_actions->size = sample_actions_len;
5343 ctx_cancel_freeze(&ctx);
5344 ofpbuf_clear(&ctx.action_set);
5347 if (!ctx.freezing) {
5348 xlate_action_set(&ctx);
5351 compose_recirculate_action(&ctx);
5355 /* Output only fully processed packets. */
5357 && xbridge->has_in_band
5358 && in_band_must_output_to_local_port(flow)
5359 && !actions_output_to_local_port(&ctx)) {
5360 compose_output_action(&ctx, OFPP_LOCAL, NULL);
5363 if (user_cookie_offset) {
5364 fix_sflow_action(&ctx, user_cookie_offset);
5368 if (nl_attr_oversized(ctx.odp_actions->size)) {
5369 /* These datapath actions are too big for a Netlink attribute, so we
5370 * can't hand them to the kernel directly. dpif_execute() can execute
5371 * them one by one with help, so just mark the result as SLOW_ACTION to
5372 * prevent the flow from being installed. */
5373 COVERAGE_INC(xlate_actions_oversize);
5374 ctx.xout->slow |= SLOW_ACTION;
5375 } else if (too_many_output_actions(ctx.odp_actions)) {
5376 COVERAGE_INC(xlate_actions_too_many_output);
5377 ctx.xout->slow |= SLOW_ACTION;
5380 /* Do netflow only for packets on initial reception, that are not sent to
5381 * the controller. We consider packets sent to the controller to be part
5382 * of the control plane rather than the data plane. */
5383 if (!xin->frozen_state
5385 && !(xout->slow & SLOW_CONTROLLER)) {
5386 if (ctx.xin->resubmit_stats) {
5387 netflow_flow_update(xbridge->netflow, flow,
5388 ctx.nf_output_iface,
5389 ctx.xin->resubmit_stats);
5391 if (ctx.xin->xcache) {
5392 struct xc_entry *entry;
5394 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5395 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5396 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5397 entry->u.nf.iface = ctx.nf_output_iface;
5402 xlate_wc_finish(&ctx);
5406 ofpbuf_uninit(&ctx.stack);
5407 ofpbuf_uninit(&ctx.action_set);
5408 ofpbuf_uninit(&ctx.frozen_actions);
5409 ofpbuf_uninit(&scratch_actions);
5411 /* Make sure we return a "drop flow" in case of an error. */
5414 if (xin->odp_actions) {
5415 ofpbuf_clear(xin->odp_actions);
5421 /* Sends 'packet' out 'ofport'.
5422 * May modify 'packet'.
5423 * Returns 0 if successful, otherwise a positive errno value. */
5425 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5427 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5428 struct xport *xport;
5429 struct ofpact_output output;
5432 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5433 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5434 flow_extract(packet, &flow);
5435 flow.in_port.ofp_port = OFPP_NONE;
5437 xport = xport_lookup(xcfg, ofport);
5441 output.port = xport->ofp_port;
5444 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5445 &output.ofpact, sizeof output,
5449 struct xlate_cache *
5450 xlate_cache_new(void)
5452 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5454 ofpbuf_init(&xcache->entries, 512);
5458 static struct xc_entry *
5459 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5461 struct xc_entry *entry;
5463 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5470 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5472 if (entry->u.dev.tx) {
5473 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5475 if (entry->u.dev.rx) {
5476 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5478 if (entry->u.dev.bfd) {
5479 bfd_account_rx(entry->u.dev.bfd, stats);
5484 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5486 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5487 struct xbridge *xbridge;
5488 struct xbundle *xbundle;
5489 struct flow_wildcards wc;
5491 xbridge = xbridge_lookup(xcfg, ofproto);
5496 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5502 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5505 /* Push stats and perform side effects of flow translation. */
5507 xlate_push_stats(struct xlate_cache *xcache,
5508 const struct dpif_flow_stats *stats)
5510 struct xc_entry *entry;
5511 struct ofpbuf entries = xcache->entries;
5512 struct eth_addr dmac;
5514 if (!stats->n_packets) {
5518 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5519 switch (entry->type) {
5521 rule_dpif_credit_stats(entry->u.rule, stats);
5524 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5525 entry->u.bond.vid, stats->n_bytes);
5528 xlate_cache_netdev(entry, stats);
5531 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5532 entry->u.nf.iface, stats);
5535 mirror_update_stats(entry->u.mirror.mbridge,
5536 entry->u.mirror.mirrors,
5537 stats->n_packets, stats->n_bytes);
5540 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5543 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5544 entry->u.normal.vlan);
5546 case XC_FIN_TIMEOUT:
5547 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5548 entry->u.fin.idle, entry->u.fin.hard);
5551 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5555 /* Lookup neighbor to avoid timeout. */
5556 tnl_neigh_lookup(entry->u.tnl_neigh_cache.br_name,
5557 &entry->u.tnl_neigh_cache.d_ipv6, &dmac);
5566 xlate_dev_unref(struct xc_entry *entry)
5568 if (entry->u.dev.tx) {
5569 netdev_close(entry->u.dev.tx);
5571 if (entry->u.dev.rx) {
5572 netdev_close(entry->u.dev.rx);
5574 if (entry->u.dev.bfd) {
5575 bfd_unref(entry->u.dev.bfd);
5580 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5582 netflow_flow_clear(netflow, flow);
5583 netflow_unref(netflow);
5588 xlate_cache_clear(struct xlate_cache *xcache)
5590 struct xc_entry *entry;
5591 struct ofpbuf entries;
5597 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5598 switch (entry->type) {
5600 rule_dpif_unref(entry->u.rule);
5603 free(entry->u.bond.flow);
5604 bond_unref(entry->u.bond.bond);
5607 xlate_dev_unref(entry);
5610 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5613 mbridge_unref(entry->u.mirror.mbridge);
5616 free(entry->u.learn.fm);
5617 ofpbuf_delete(entry->u.learn.ofpacts);
5620 free(entry->u.normal.flow);
5622 case XC_FIN_TIMEOUT:
5623 /* 'u.fin.rule' is always already held as a XC_RULE, which
5624 * has already released it's reference above. */
5627 group_dpif_unref(entry->u.group.group);
5636 ofpbuf_clear(&xcache->entries);
5640 xlate_cache_delete(struct xlate_cache *xcache)
5642 xlate_cache_clear(xcache);
5643 ofpbuf_uninit(&xcache->entries);