1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 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-arp-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"
57 #include "ovs-router.h"
58 #include "tnl-ports.h"
60 #include "openvswitch/vlog.h"
62 COVERAGE_DEFINE(xlate_actions);
63 COVERAGE_DEFINE(xlate_actions_oversize);
64 COVERAGE_DEFINE(xlate_actions_too_many_output);
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
71 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
74 /* Maximum number of resubmit actions in a flow translation, whether they are
75 * recursive or not. */
76 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
79 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
80 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
82 struct ovs_list xbundles; /* Owned xbundles. */
83 struct hmap xports; /* Indexed by ofp_port. */
85 char *name; /* Name used in log messages. */
86 struct dpif *dpif; /* Datapath interface. */
87 struct mac_learning *ml; /* Mac learning handle. */
88 struct mcast_snooping *ms; /* Multicast Snooping handle. */
89 struct mbridge *mbridge; /* Mirroring. */
90 struct dpif_sflow *sflow; /* SFlow handle, or null. */
91 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
92 struct netflow *netflow; /* Netflow handle, or null. */
93 struct stp *stp; /* STP or null if disabled. */
94 struct rstp *rstp; /* RSTP or null if disabled. */
96 bool has_in_band; /* Bridge has in band control? */
97 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
99 /* Datapath feature support. */
100 struct dpif_backer_support support;
104 struct hmap_node hmap_node; /* In global 'xbundles' map. */
105 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
107 struct ovs_list list_node; /* In parent 'xbridges' list. */
108 struct xbridge *xbridge; /* Parent xbridge. */
110 struct ovs_list xports; /* Contains "struct xport"s. */
112 char *name; /* Name used in log messages. */
113 struct bond *bond; /* Nonnull iff more than one port. */
114 struct lacp *lacp; /* LACP handle or null. */
116 enum port_vlan_mode vlan_mode; /* VLAN mode. */
117 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
118 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
119 * NULL if all VLANs are trunked. */
120 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
121 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
125 struct hmap_node hmap_node; /* Node in global 'xports' map. */
126 struct ofport_dpif *ofport; /* Key in global 'xports map. */
128 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
129 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
131 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
133 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
134 struct xbundle *xbundle; /* Parent xbundle or null. */
136 struct netdev *netdev; /* 'ofport''s netdev. */
138 struct xbridge *xbridge; /* Parent bridge. */
139 struct xport *peer; /* Patch port peer or null. */
141 enum ofputil_port_config config; /* OpenFlow port configuration. */
142 enum ofputil_port_state state; /* OpenFlow port state. */
143 int stp_port_no; /* STP port number or -1 if not in use. */
144 struct rstp_port *rstp_port; /* RSTP port or null. */
146 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
148 bool may_enable; /* May be enabled in bonds. */
149 bool is_tunnel; /* Is a tunnel port. */
151 struct cfm *cfm; /* CFM handle or null. */
152 struct bfd *bfd; /* BFD handle or null. */
153 struct lldp *lldp; /* LLDP handle or null. */
157 struct xlate_in *xin;
158 struct xlate_out *xout;
160 const struct xbridge *xbridge;
162 /* Flow tables version at the beginning of the translation. */
163 cls_version_t tables_version;
165 /* Flow at the last commit. */
166 struct flow base_flow;
168 /* Tunnel IP destination address as received. This is stored separately
169 * as the base_flow.tunnel is cleared on init to reflect the datapath
170 * behavior. Used to make sure not to send tunneled output to ourselves,
171 * which might lead to an infinite loop. This could happen easily
172 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
173 * actually set the tun_dst field. */
174 ovs_be32 orig_tunnel_ip_dst;
176 /* Stack for the push and pop actions. Each stack element is of type
177 * "union mf_subvalue". */
180 /* The rule that we are currently translating, or NULL. */
181 struct rule_dpif *rule;
183 /* Flow translation populates this with wildcards relevant in translation.
184 * When 'xin->wc' is nonnull, this is the same pointer. When 'xin->wc' is
185 * null, this is a pointer to uninitialized scratch memory. This allows
186 * code to blindly write to 'ctx->wc' without worrying about whether the
187 * caller really wants wildcards. */
188 struct flow_wildcards *wc;
190 /* Output buffer for datapath actions. When 'xin->odp_actions' is nonnull,
191 * this is the same pointer. When 'xin->odp_actions' is null, this points
192 * to a scratch ofpbuf. This allows code to add actions to
193 * 'ctx->odp_actions' without worrying about whether the caller really
195 struct ofpbuf *odp_actions;
197 /* Resubmit statistics, via xlate_table_action(). */
198 int recurse; /* Current resubmit nesting depth. */
199 int resubmits; /* Total number of resubmits. */
200 bool in_group; /* Currently translating ofgroup, if true. */
201 bool in_action_set; /* Currently translating action_set, if true. */
203 uint8_t table_id; /* OpenFlow table ID where flow was found. */
204 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
205 uint32_t orig_skb_priority; /* Priority when packet arrived. */
206 uint32_t sflow_n_outputs; /* Number of output ports. */
207 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
208 ofp_port_t nf_output_iface; /* Output interface index for NetFlow. */
209 bool exit; /* No further actions should be processed. */
210 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
212 /* These are used for non-bond recirculation. The recirculation IDs are
213 * stored in xout and must be associated with a datapath flow (ukey),
214 * otherwise they will be freed when the xout is uninitialized.
217 * Steps in Recirculation Translation
218 * ==================================
220 * At some point during translation, the code recognizes the need for
221 * recirculation. For example, recirculation is necessary when, after
222 * popping the last MPLS label, an action or a match tries to examine or
223 * modify a field that has been newly revealed following the MPLS label.
225 * The simplest part of the work to be done is to commit existing changes to
226 * the packet, which produces datapath actions corresponding to the changes,
227 * and after this, add an OVS_ACTION_ATTR_RECIRC datapath action.
229 * The main problem here is preserving state. When the datapath executes
230 * OVS_ACTION_ATTR_RECIRC, it will upcall to userspace to get a translation
231 * for the post-recirculation actions. At this point userspace has to
232 * resume the translation where it left off, which means that it has to
233 * execute the following:
235 * - The action that prompted recirculation, and any actions following
236 * it within the same flow.
238 * - If the action that prompted recirculation was invoked within a
239 * NXAST_RESUBMIT, then any actions following the resubmit. These
240 * "resubmit"s can be nested, so this has to go all the way up the
243 * - The OpenFlow 1.1+ action set.
245 * State that actions and flow table lookups can depend on, such as the
246 * following, must also be preserved:
248 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
250 * - Action set, stack
252 * - The table ID and cookie of the flow being translated at each level
253 * of the control stack (since OFPAT_CONTROLLER actions send these to
256 * Translation allows for the control of this state preservation via these
257 * members. When a need for recirculation is identified, the translation
260 * 1. Sets 'recirc_action_offset' to the current size of 'action_set'. The
261 * action set is part of what needs to be preserved, so this allows the
262 * action set and the additional state to share the 'action_set' buffer.
263 * Later steps can tell that setup for recirculation is in progress from
264 * the nonnegative value of 'recirc_action_offset'.
266 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
267 * translation process.
269 * 3. Adds an OFPACT_UNROLL_XLATE action to 'action_set'. This action
270 * holds the current table ID and cookie so that they can be restored
271 * during a post-recirculation upcall translation.
273 * 4. Adds the action that prompted recirculation and any actions following
274 * it within the same flow to 'action_set', so that they can be executed
275 * during a post-recirculation upcall translation.
279 * 6. The action that prompted recirculation might be nested in a stack of
280 * nested "resubmit"s that have actions remaining. Each of these notices
281 * that we're exiting (from 'exit') and that recirculation setup is in
282 * progress (from 'recirc_action_offset') and responds by adding more
283 * OFPACT_UNROLL_XLATE actions to 'action_set', as necessary, and any
284 * actions that were yet unprocessed.
286 * The caller stores all the state produced by this process associated with
287 * the recirculation ID. For post-recirculation upcall translation, the
288 * caller passes it back in for the new translation to execute. The
289 * process yielded a set of ofpacts that can be translated directly, so it
290 * is not much of a special case at that point.
292 int recirc_action_offset; /* Offset in 'action_set' to actions to be
293 * executed after recirculation, or -1. */
294 int last_unroll_offset; /* Offset in 'action_set' to the latest unroll
297 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
298 * This is a trigger for recirculation in cases where translating an action
299 * or looking up a flow requires access to the fields of the packet after
300 * the MPLS label stack that was originally present. */
303 /* True if conntrack has been performed on this packet during processing
304 * on the current bridge. This is used to determine whether conntrack
305 * state from the datapath should be honored after recirculation. */
308 /* OpenFlow 1.1+ action set.
310 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
311 * When translation is otherwise complete, ofpacts_execute_action_set()
312 * converts it to a set of "struct ofpact"s that can be translated into
313 * datapath actions. */
314 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
315 struct ofpbuf action_set; /* Action set. */
318 static void xlate_action_set(struct xlate_ctx *ctx);
319 static void xlate_commit_actions(struct xlate_ctx *ctx);
322 ctx_trigger_recirculation(struct xlate_ctx *ctx)
325 ctx->recirc_action_offset = ctx->action_set.size;
329 ctx_first_recirculation_action(const struct xlate_ctx *ctx)
331 return ctx->recirc_action_offset == ctx->action_set.size;
335 exit_recirculates(const struct xlate_ctx *ctx)
337 /* When recirculating the 'recirc_action_offset' has a non-negative value.
339 return ctx->recirc_action_offset >= 0;
342 static void compose_recirculate_action(struct xlate_ctx *ctx);
344 /* A controller may use OFPP_NONE as the ingress port to indicate that
345 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
346 * when an input bundle is needed for validation (e.g., mirroring or
347 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
348 * any 'port' structs, so care must be taken when dealing with it. */
349 static struct xbundle ofpp_none_bundle = {
351 .vlan_mode = PORT_VLAN_TRUNK
354 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
355 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
356 * traffic egressing the 'ofport' with that priority should be marked with. */
357 struct skb_priority_to_dscp {
358 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
359 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
361 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
377 /* xlate_cache entries hold enough information to perform the side effects of
378 * xlate_actions() for a rule, without needing to perform rule translation
379 * from scratch. The primary usage of these is to submit statistics to objects
380 * that a flow relates to, although they may be used for other effects as well
381 * (for instance, refreshing hard timeouts for learned flows). */
385 struct rule_dpif *rule;
392 struct netflow *netflow;
397 struct mbridge *mbridge;
398 mirror_mask_t mirrors;
406 struct ofproto_dpif *ofproto;
407 struct ofputil_flow_mod *fm;
408 struct ofpbuf *ofpacts;
411 struct ofproto_dpif *ofproto;
416 struct rule_dpif *rule;
421 struct group_dpif *group;
422 struct ofputil_bucket *bucket;
425 char br_name[IFNAMSIZ];
431 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
432 entries = xcache->entries; \
433 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
435 entry = ofpbuf_try_pull(&entries, sizeof *entry))
438 struct ofpbuf entries;
441 /* Xlate config contains hash maps of all bridges, bundles and ports.
442 * Xcfgp contains the pointer to the current xlate configuration.
443 * When the main thread needs to change the configuration, it copies xcfgp to
444 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
445 * does not block handler and revalidator threads. */
447 struct hmap xbridges;
448 struct hmap xbundles;
451 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
452 static struct xlate_cfg *new_xcfg = NULL;
454 static bool may_receive(const struct xport *, struct xlate_ctx *);
455 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
457 static void xlate_normal(struct xlate_ctx *);
458 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
459 OVS_PRINTF_FORMAT(2, 3);
460 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
461 uint8_t table_id, bool may_packet_in,
462 bool honor_table_miss);
463 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
464 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
465 static void output_normal(struct xlate_ctx *, const struct xbundle *,
468 /* Optional bond recirculation parameter to compose_output_action(). */
469 struct xlate_bond_recirc {
470 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
471 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
472 uint32_t hash_basis; /* Compute hash for recirc before. */
475 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
476 const struct xlate_bond_recirc *xr);
478 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
479 const struct ofproto_dpif *);
480 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
481 const struct ofbundle *);
482 static struct xport *xport_lookup(struct xlate_cfg *,
483 const struct ofport_dpif *);
484 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
485 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
486 uint32_t skb_priority);
487 static void clear_skb_priorities(struct xport *);
488 static size_t count_skb_priorities(const struct xport *);
489 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
492 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
494 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
495 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
496 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
497 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
498 const struct mac_learning *, struct stp *,
499 struct rstp *, const struct mcast_snooping *,
500 const struct mbridge *,
501 const struct dpif_sflow *,
502 const struct dpif_ipfix *,
503 const struct netflow *,
504 bool forward_bpdu, bool has_in_band,
505 const struct dpif_backer_support *);
506 static void xlate_xbundle_set(struct xbundle *xbundle,
507 enum port_vlan_mode vlan_mode, int vlan,
508 unsigned long *trunks, bool use_priority_tags,
509 const struct bond *bond, const struct lacp *lacp,
511 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
512 const struct netdev *netdev, const struct cfm *cfm,
513 const struct bfd *bfd, const struct lldp *lldp,
514 int stp_port_no, const struct rstp_port *rstp_port,
515 enum ofputil_port_config config,
516 enum ofputil_port_state state, bool is_tunnel,
518 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
519 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
520 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
521 static void xlate_xbridge_copy(struct xbridge *);
522 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
523 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
525 static void xlate_xcfg_free(struct xlate_cfg *);
528 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
530 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
533 va_start(args, format);
534 ctx->xin->report_hook(ctx->xin, ctx->recurse, format, args);
540 xlate_report_actions(struct xlate_ctx *ctx, const char *title,
541 const struct ofpact *ofpacts, size_t ofpacts_len)
543 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
544 struct ds s = DS_EMPTY_INITIALIZER;
545 ofpacts_format(ofpacts, ofpacts_len, &s);
546 xlate_report(ctx, "%s: %s", title, ds_cstr(&s));
552 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
554 list_init(&xbridge->xbundles);
555 hmap_init(&xbridge->xports);
556 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
557 hash_pointer(xbridge->ofproto, 0));
561 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
563 list_init(&xbundle->xports);
564 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
565 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
566 hash_pointer(xbundle->ofbundle, 0));
570 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
572 hmap_init(&xport->skb_priorities);
573 hmap_insert(&xcfg->xports, &xport->hmap_node,
574 hash_pointer(xport->ofport, 0));
575 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
576 hash_ofp_port(xport->ofp_port));
580 xlate_xbridge_set(struct xbridge *xbridge,
582 const struct mac_learning *ml, struct stp *stp,
583 struct rstp *rstp, const struct mcast_snooping *ms,
584 const struct mbridge *mbridge,
585 const struct dpif_sflow *sflow,
586 const struct dpif_ipfix *ipfix,
587 const struct netflow *netflow,
588 bool forward_bpdu, bool has_in_band,
589 const struct dpif_backer_support *support)
591 if (xbridge->ml != ml) {
592 mac_learning_unref(xbridge->ml);
593 xbridge->ml = mac_learning_ref(ml);
596 if (xbridge->ms != ms) {
597 mcast_snooping_unref(xbridge->ms);
598 xbridge->ms = mcast_snooping_ref(ms);
601 if (xbridge->mbridge != mbridge) {
602 mbridge_unref(xbridge->mbridge);
603 xbridge->mbridge = mbridge_ref(mbridge);
606 if (xbridge->sflow != sflow) {
607 dpif_sflow_unref(xbridge->sflow);
608 xbridge->sflow = dpif_sflow_ref(sflow);
611 if (xbridge->ipfix != ipfix) {
612 dpif_ipfix_unref(xbridge->ipfix);
613 xbridge->ipfix = dpif_ipfix_ref(ipfix);
616 if (xbridge->stp != stp) {
617 stp_unref(xbridge->stp);
618 xbridge->stp = stp_ref(stp);
621 if (xbridge->rstp != rstp) {
622 rstp_unref(xbridge->rstp);
623 xbridge->rstp = rstp_ref(rstp);
626 if (xbridge->netflow != netflow) {
627 netflow_unref(xbridge->netflow);
628 xbridge->netflow = netflow_ref(netflow);
631 xbridge->dpif = dpif;
632 xbridge->forward_bpdu = forward_bpdu;
633 xbridge->has_in_band = has_in_band;
634 xbridge->support = *support;
638 xlate_xbundle_set(struct xbundle *xbundle,
639 enum port_vlan_mode vlan_mode, int vlan,
640 unsigned long *trunks, bool use_priority_tags,
641 const struct bond *bond, const struct lacp *lacp,
644 ovs_assert(xbundle->xbridge);
646 xbundle->vlan_mode = vlan_mode;
647 xbundle->vlan = vlan;
648 xbundle->trunks = trunks;
649 xbundle->use_priority_tags = use_priority_tags;
650 xbundle->floodable = floodable;
652 if (xbundle->bond != bond) {
653 bond_unref(xbundle->bond);
654 xbundle->bond = bond_ref(bond);
657 if (xbundle->lacp != lacp) {
658 lacp_unref(xbundle->lacp);
659 xbundle->lacp = lacp_ref(lacp);
664 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
665 const struct netdev *netdev, const struct cfm *cfm,
666 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
667 const struct rstp_port* rstp_port,
668 enum ofputil_port_config config, enum ofputil_port_state state,
669 bool is_tunnel, bool may_enable)
671 xport->config = config;
672 xport->state = state;
673 xport->stp_port_no = stp_port_no;
674 xport->is_tunnel = is_tunnel;
675 xport->may_enable = may_enable;
676 xport->odp_port = odp_port;
678 if (xport->rstp_port != rstp_port) {
679 rstp_port_unref(xport->rstp_port);
680 xport->rstp_port = rstp_port_ref(rstp_port);
683 if (xport->cfm != cfm) {
684 cfm_unref(xport->cfm);
685 xport->cfm = cfm_ref(cfm);
688 if (xport->bfd != bfd) {
689 bfd_unref(xport->bfd);
690 xport->bfd = bfd_ref(bfd);
693 if (xport->lldp != lldp) {
694 lldp_unref(xport->lldp);
695 xport->lldp = lldp_ref(lldp);
698 if (xport->netdev != netdev) {
699 netdev_close(xport->netdev);
700 xport->netdev = netdev_ref(netdev);
705 xlate_xbridge_copy(struct xbridge *xbridge)
707 struct xbundle *xbundle;
709 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
710 new_xbridge->ofproto = xbridge->ofproto;
711 new_xbridge->name = xstrdup(xbridge->name);
712 xlate_xbridge_init(new_xcfg, new_xbridge);
714 xlate_xbridge_set(new_xbridge,
715 xbridge->dpif, xbridge->ml, xbridge->stp,
716 xbridge->rstp, xbridge->ms, xbridge->mbridge,
717 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
718 xbridge->forward_bpdu, xbridge->has_in_band,
720 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
721 xlate_xbundle_copy(new_xbridge, xbundle);
724 /* Copy xports which are not part of a xbundle */
725 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
726 if (!xport->xbundle) {
727 xlate_xport_copy(new_xbridge, NULL, xport);
733 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
736 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
737 new_xbundle->ofbundle = xbundle->ofbundle;
738 new_xbundle->xbridge = xbridge;
739 new_xbundle->name = xstrdup(xbundle->name);
740 xlate_xbundle_init(new_xcfg, new_xbundle);
742 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
743 xbundle->vlan, xbundle->trunks,
744 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
746 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
747 xlate_xport_copy(xbridge, new_xbundle, xport);
752 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
755 struct skb_priority_to_dscp *pdscp, *new_pdscp;
756 struct xport *new_xport = xzalloc(sizeof *xport);
757 new_xport->ofport = xport->ofport;
758 new_xport->ofp_port = xport->ofp_port;
759 new_xport->xbridge = xbridge;
760 xlate_xport_init(new_xcfg, new_xport);
762 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
763 xport->bfd, xport->lldp, xport->stp_port_no,
764 xport->rstp_port, xport->config, xport->state,
765 xport->is_tunnel, xport->may_enable);
768 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
770 new_xport->peer = peer;
771 new_xport->peer->peer = new_xport;
776 new_xport->xbundle = xbundle;
777 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
780 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
781 new_pdscp = xmalloc(sizeof *pdscp);
782 new_pdscp->skb_priority = pdscp->skb_priority;
783 new_pdscp->dscp = pdscp->dscp;
784 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
785 hash_int(new_pdscp->skb_priority, 0));
789 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
790 * configuration in xcfgp.
792 * This needs to be called after editing the xlate configuration.
794 * Functions that edit the new xlate configuration are
795 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
801 * edit_xlate_configuration();
803 * xlate_txn_commit(); */
805 xlate_txn_commit(void)
807 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
809 ovsrcu_set(&xcfgp, new_xcfg);
810 ovsrcu_synchronize();
811 xlate_xcfg_free(xcfg);
815 /* Copies the current xlate configuration in xcfgp to new_xcfg.
817 * This needs to be called prior to editing the xlate configuration. */
819 xlate_txn_start(void)
821 struct xbridge *xbridge;
822 struct xlate_cfg *xcfg;
824 ovs_assert(!new_xcfg);
826 new_xcfg = xmalloc(sizeof *new_xcfg);
827 hmap_init(&new_xcfg->xbridges);
828 hmap_init(&new_xcfg->xbundles);
829 hmap_init(&new_xcfg->xports);
831 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
836 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
837 xlate_xbridge_copy(xbridge);
843 xlate_xcfg_free(struct xlate_cfg *xcfg)
845 struct xbridge *xbridge, *next_xbridge;
851 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
852 xlate_xbridge_remove(xcfg, xbridge);
855 hmap_destroy(&xcfg->xbridges);
856 hmap_destroy(&xcfg->xbundles);
857 hmap_destroy(&xcfg->xports);
862 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
864 const struct mac_learning *ml, struct stp *stp,
865 struct rstp *rstp, const struct mcast_snooping *ms,
866 const struct mbridge *mbridge,
867 const struct dpif_sflow *sflow,
868 const struct dpif_ipfix *ipfix,
869 const struct netflow *netflow,
870 bool forward_bpdu, bool has_in_band,
871 const struct dpif_backer_support *support)
873 struct xbridge *xbridge;
875 ovs_assert(new_xcfg);
877 xbridge = xbridge_lookup(new_xcfg, ofproto);
879 xbridge = xzalloc(sizeof *xbridge);
880 xbridge->ofproto = ofproto;
882 xlate_xbridge_init(new_xcfg, xbridge);
886 xbridge->name = xstrdup(name);
888 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
889 netflow, forward_bpdu, has_in_band, support);
893 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
895 struct xbundle *xbundle, *next_xbundle;
896 struct xport *xport, *next_xport;
902 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
903 xlate_xport_remove(xcfg, xport);
906 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
907 xlate_xbundle_remove(xcfg, xbundle);
910 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
911 mac_learning_unref(xbridge->ml);
912 mcast_snooping_unref(xbridge->ms);
913 mbridge_unref(xbridge->mbridge);
914 dpif_sflow_unref(xbridge->sflow);
915 dpif_ipfix_unref(xbridge->ipfix);
916 stp_unref(xbridge->stp);
917 rstp_unref(xbridge->rstp);
918 hmap_destroy(&xbridge->xports);
924 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
926 struct xbridge *xbridge;
928 ovs_assert(new_xcfg);
930 xbridge = xbridge_lookup(new_xcfg, ofproto);
931 xlate_xbridge_remove(new_xcfg, xbridge);
935 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
936 const char *name, enum port_vlan_mode vlan_mode, int vlan,
937 unsigned long *trunks, bool use_priority_tags,
938 const struct bond *bond, const struct lacp *lacp,
941 struct xbundle *xbundle;
943 ovs_assert(new_xcfg);
945 xbundle = xbundle_lookup(new_xcfg, ofbundle);
947 xbundle = xzalloc(sizeof *xbundle);
948 xbundle->ofbundle = ofbundle;
949 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
951 xlate_xbundle_init(new_xcfg, xbundle);
955 xbundle->name = xstrdup(name);
957 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
958 use_priority_tags, bond, lacp, floodable);
962 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
970 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
971 xport->xbundle = NULL;
974 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
975 list_remove(&xbundle->list_node);
976 bond_unref(xbundle->bond);
977 lacp_unref(xbundle->lacp);
983 xlate_bundle_remove(struct ofbundle *ofbundle)
985 struct xbundle *xbundle;
987 ovs_assert(new_xcfg);
989 xbundle = xbundle_lookup(new_xcfg, ofbundle);
990 xlate_xbundle_remove(new_xcfg, xbundle);
994 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
995 struct ofport_dpif *ofport, ofp_port_t ofp_port,
996 odp_port_t odp_port, const struct netdev *netdev,
997 const struct cfm *cfm, const struct bfd *bfd,
998 const struct lldp *lldp, struct ofport_dpif *peer,
999 int stp_port_no, const struct rstp_port *rstp_port,
1000 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
1001 enum ofputil_port_config config,
1002 enum ofputil_port_state state, bool is_tunnel,
1006 struct xport *xport;
1008 ovs_assert(new_xcfg);
1010 xport = xport_lookup(new_xcfg, ofport);
1012 xport = xzalloc(sizeof *xport);
1013 xport->ofport = ofport;
1014 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
1015 xport->ofp_port = ofp_port;
1017 xlate_xport_init(new_xcfg, xport);
1020 ovs_assert(xport->ofp_port == ofp_port);
1022 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1023 stp_port_no, rstp_port, config, state, is_tunnel,
1027 xport->peer->peer = NULL;
1029 xport->peer = xport_lookup(new_xcfg, peer);
1031 xport->peer->peer = xport;
1034 if (xport->xbundle) {
1035 list_remove(&xport->bundle_node);
1037 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1038 if (xport->xbundle) {
1039 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1042 clear_skb_priorities(xport);
1043 for (i = 0; i < n_qdscp; i++) {
1044 struct skb_priority_to_dscp *pdscp;
1045 uint32_t skb_priority;
1047 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1052 pdscp = xmalloc(sizeof *pdscp);
1053 pdscp->skb_priority = skb_priority;
1054 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1055 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1056 hash_int(pdscp->skb_priority, 0));
1061 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1068 xport->peer->peer = NULL;
1072 if (xport->xbundle) {
1073 list_remove(&xport->bundle_node);
1076 clear_skb_priorities(xport);
1077 hmap_destroy(&xport->skb_priorities);
1079 hmap_remove(&xcfg->xports, &xport->hmap_node);
1080 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1082 netdev_close(xport->netdev);
1083 rstp_port_unref(xport->rstp_port);
1084 cfm_unref(xport->cfm);
1085 bfd_unref(xport->bfd);
1086 lldp_unref(xport->lldp);
1091 xlate_ofport_remove(struct ofport_dpif *ofport)
1093 struct xport *xport;
1095 ovs_assert(new_xcfg);
1097 xport = xport_lookup(new_xcfg, ofport);
1098 xlate_xport_remove(new_xcfg, xport);
1101 static struct ofproto_dpif *
1102 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1103 ofp_port_t *ofp_in_port, const struct xport **xportp)
1105 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1106 const struct xport *xport;
1108 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1109 ? tnl_port_receive(flow)
1110 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1111 if (OVS_UNLIKELY(!xport)) {
1116 *ofp_in_port = xport->ofp_port;
1118 return xport->xbridge->ofproto;
1121 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1122 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1123 struct ofproto_dpif *
1124 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1125 ofp_port_t *ofp_in_port)
1127 const struct xport *xport;
1129 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1132 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1133 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1134 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1135 * handles for those protocols if they're enabled. Caller may use the returned
1136 * pointers until quiescing, for longer term use additional references must
1139 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1142 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1143 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1144 struct dpif_sflow **sflow, struct netflow **netflow,
1145 ofp_port_t *ofp_in_port)
1147 struct ofproto_dpif *ofproto;
1148 const struct xport *xport;
1150 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1157 *ofprotop = ofproto;
1161 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1165 *sflow = xport ? xport->xbridge->sflow : NULL;
1169 *netflow = xport ? xport->xbridge->netflow : NULL;
1175 static struct xbridge *
1176 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1178 struct hmap *xbridges;
1179 struct xbridge *xbridge;
1181 if (!ofproto || !xcfg) {
1185 xbridges = &xcfg->xbridges;
1187 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1189 if (xbridge->ofproto == ofproto) {
1196 static struct xbundle *
1197 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1199 struct hmap *xbundles;
1200 struct xbundle *xbundle;
1202 if (!ofbundle || !xcfg) {
1206 xbundles = &xcfg->xbundles;
1208 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1210 if (xbundle->ofbundle == ofbundle) {
1217 static struct xport *
1218 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1220 struct hmap *xports;
1221 struct xport *xport;
1223 if (!ofport || !xcfg) {
1227 xports = &xcfg->xports;
1229 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1231 if (xport->ofport == ofport) {
1238 static struct stp_port *
1239 xport_get_stp_port(const struct xport *xport)
1241 return xport->xbridge->stp && xport->stp_port_no != -1
1242 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1247 xport_stp_learn_state(const struct xport *xport)
1249 struct stp_port *sp = xport_get_stp_port(xport);
1251 ? stp_learn_in_state(stp_port_get_state(sp))
1256 xport_stp_forward_state(const struct xport *xport)
1258 struct stp_port *sp = xport_get_stp_port(xport);
1260 ? stp_forward_in_state(stp_port_get_state(sp))
1265 xport_stp_should_forward_bpdu(const struct xport *xport)
1267 struct stp_port *sp = xport_get_stp_port(xport);
1268 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1271 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1272 * were used to make the determination.*/
1274 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1276 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1277 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1278 return is_stp(flow);
1282 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1284 struct stp_port *sp = xport_get_stp_port(xport);
1285 struct dp_packet payload = *packet;
1286 struct eth_header *eth = dp_packet_data(&payload);
1288 /* Sink packets on ports that have STP disabled when the bridge has
1290 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1294 /* Trim off padding on payload. */
1295 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1296 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1299 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1300 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1304 static enum rstp_state
1305 xport_get_rstp_port_state(const struct xport *xport)
1307 return xport->rstp_port
1308 ? rstp_port_get_state(xport->rstp_port)
1313 xport_rstp_learn_state(const struct xport *xport)
1315 return xport->xbridge->rstp && xport->rstp_port
1316 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1321 xport_rstp_forward_state(const struct xport *xport)
1323 return xport->xbridge->rstp && xport->rstp_port
1324 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1329 xport_rstp_should_manage_bpdu(const struct xport *xport)
1331 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1335 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1337 struct dp_packet payload = *packet;
1338 struct eth_header *eth = dp_packet_data(&payload);
1340 /* Sink packets on ports that have no RSTP. */
1341 if (!xport->rstp_port) {
1345 /* Trim off padding on payload. */
1346 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1347 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1350 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1351 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1352 dp_packet_size(&payload));
1356 static struct xport *
1357 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1359 struct xport *xport;
1361 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1363 if (xport->ofp_port == ofp_port) {
1371 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1373 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1374 return xport ? xport->odp_port : ODPP_NONE;
1378 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1380 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1381 return xport && xport->may_enable;
1384 static struct ofputil_bucket *
1385 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1389 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1391 struct group_dpif *group;
1393 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1394 struct ofputil_bucket *bucket;
1396 bucket = group_first_live_bucket(ctx, group, depth);
1397 group_dpif_unref(group);
1398 return bucket == NULL;
1404 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1407 bucket_is_alive(const struct xlate_ctx *ctx,
1408 struct ofputil_bucket *bucket, int depth)
1410 if (depth >= MAX_LIVENESS_RECURSION) {
1411 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1413 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1414 MAX_LIVENESS_RECURSION);
1418 return (!ofputil_bucket_has_liveness(bucket)
1419 || (bucket->watch_port != OFPP_ANY
1420 && odp_port_is_alive(ctx, bucket->watch_port))
1421 || (bucket->watch_group != OFPG_ANY
1422 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1425 static struct ofputil_bucket *
1426 group_first_live_bucket(const struct xlate_ctx *ctx,
1427 const struct group_dpif *group, int depth)
1429 struct ofputil_bucket *bucket;
1430 const struct ovs_list *buckets;
1432 group_dpif_get_buckets(group, &buckets);
1433 LIST_FOR_EACH (bucket, list_node, buckets) {
1434 if (bucket_is_alive(ctx, bucket, depth)) {
1442 static struct ofputil_bucket *
1443 group_best_live_bucket(const struct xlate_ctx *ctx,
1444 const struct group_dpif *group,
1447 struct ofputil_bucket *best_bucket = NULL;
1448 uint32_t best_score = 0;
1451 struct ofputil_bucket *bucket;
1452 const struct ovs_list *buckets;
1454 group_dpif_get_buckets(group, &buckets);
1455 LIST_FOR_EACH (bucket, list_node, buckets) {
1456 if (bucket_is_alive(ctx, bucket, 0)) {
1457 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1458 if (score >= best_score) {
1459 best_bucket = bucket;
1470 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1472 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1473 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1477 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1479 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1482 static mirror_mask_t
1483 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1485 return xbundle != &ofpp_none_bundle
1486 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1490 static mirror_mask_t
1491 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1493 return xbundle != &ofpp_none_bundle
1494 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1498 static mirror_mask_t
1499 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1501 return xbundle != &ofpp_none_bundle
1502 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1506 static struct xbundle *
1507 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1508 bool warn, struct xport **in_xportp)
1510 struct xport *xport;
1512 /* Find the port and bundle for the received packet. */
1513 xport = get_ofp_port(xbridge, in_port);
1517 if (xport && xport->xbundle) {
1518 return xport->xbundle;
1521 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1522 * which a controller may use as the ingress port for traffic that
1523 * it is sourcing. */
1524 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1525 return &ofpp_none_bundle;
1528 /* Odd. A few possible reasons here:
1530 * - We deleted a port but there are still a few packets queued up
1533 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1534 * we don't know about.
1536 * - The ofproto client didn't configure the port as part of a bundle.
1537 * This is particularly likely to happen if a packet was received on the
1538 * port after it was created, but before the client had a chance to
1539 * configure its bundle.
1542 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1544 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1545 "port %"PRIu16, xbridge->name, in_port);
1551 mirror_packet(struct xlate_ctx *ctx, struct xbundle *xbundle,
1552 mirror_mask_t mirrors)
1554 bool warn = ctx->xin->packet != NULL;
1555 uint16_t vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
1556 if (!input_vid_is_valid(vid, xbundle, warn)) {
1559 uint16_t vlan = input_vid_to_vlan(xbundle, vid);
1561 const struct xbridge *xbridge = ctx->xbridge;
1563 /* Don't mirror to destinations that we've already mirrored to. */
1564 mirrors &= ~ctx->mirrors;
1569 /* Record these mirrors so that we don't mirror to them again. */
1570 ctx->mirrors |= mirrors;
1572 if (ctx->xin->resubmit_stats) {
1573 mirror_update_stats(xbridge->mbridge, mirrors,
1574 ctx->xin->resubmit_stats->n_packets,
1575 ctx->xin->resubmit_stats->n_bytes);
1577 if (ctx->xin->xcache) {
1578 struct xc_entry *entry;
1580 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_MIRROR);
1581 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
1582 entry->u.mirror.mirrors = mirrors;
1586 const unsigned long *vlans;
1587 mirror_mask_t dup_mirrors;
1588 struct ofbundle *out;
1591 bool has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1592 &vlans, &dup_mirrors, &out, &out_vlan);
1593 ovs_assert(has_mirror);
1596 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1599 if (vlans && !bitmap_is_set(vlans, vlan)) {
1600 mirrors = zero_rightmost_1bit(mirrors);
1604 mirrors &= ~dup_mirrors;
1605 ctx->mirrors |= dup_mirrors;
1607 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1608 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1610 output_normal(ctx, out_xbundle, vlan);
1612 } else if (vlan != out_vlan
1613 && !eth_addr_is_reserved(ctx->xin->flow.dl_dst)) {
1614 struct xbundle *xbundle;
1616 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1617 if (xbundle_includes_vlan(xbundle, out_vlan)
1618 && !xbundle_mirror_out(xbridge, xbundle)) {
1619 output_normal(ctx, xbundle, out_vlan);
1627 mirror_ingress_packet(struct xlate_ctx *ctx)
1629 if (mbridge_has_mirrors(ctx->xbridge->mbridge)) {
1630 bool warn = ctx->xin->packet != NULL;
1631 struct xbundle *xbundle = lookup_input_bundle(
1632 ctx->xbridge, ctx->xin->flow.in_port.ofp_port, warn, NULL);
1634 mirror_packet(ctx, xbundle,
1635 xbundle_mirror_src(ctx->xbridge, xbundle));
1640 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1641 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1642 * the bundle on which the packet was received, returns the VLAN to which the
1645 * Both 'vid' and the return value are in the range 0...4095. */
1647 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1649 switch (in_xbundle->vlan_mode) {
1650 case PORT_VLAN_ACCESS:
1651 return in_xbundle->vlan;
1654 case PORT_VLAN_TRUNK:
1657 case PORT_VLAN_NATIVE_UNTAGGED:
1658 case PORT_VLAN_NATIVE_TAGGED:
1659 return vid ? vid : in_xbundle->vlan;
1666 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1667 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1670 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1671 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1674 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1676 /* Allow any VID on the OFPP_NONE port. */
1677 if (in_xbundle == &ofpp_none_bundle) {
1681 switch (in_xbundle->vlan_mode) {
1682 case PORT_VLAN_ACCESS:
1685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1686 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1687 "packet received on port %s configured as VLAN "
1688 "%"PRIu16" access port", vid, in_xbundle->name,
1695 case PORT_VLAN_NATIVE_UNTAGGED:
1696 case PORT_VLAN_NATIVE_TAGGED:
1698 /* Port must always carry its native VLAN. */
1702 case PORT_VLAN_TRUNK:
1703 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1705 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1706 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1707 "received on port %s not configured for trunking "
1708 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1720 /* Given 'vlan', the VLAN that a packet belongs to, and
1721 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1722 * that should be included in the 802.1Q header. (If the return value is 0,
1723 * then the 802.1Q header should only be included in the packet if there is a
1726 * Both 'vlan' and the return value are in the range 0...4095. */
1728 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1730 switch (out_xbundle->vlan_mode) {
1731 case PORT_VLAN_ACCESS:
1734 case PORT_VLAN_TRUNK:
1735 case PORT_VLAN_NATIVE_TAGGED:
1738 case PORT_VLAN_NATIVE_UNTAGGED:
1739 return vlan == out_xbundle->vlan ? 0 : vlan;
1747 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1750 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1752 ovs_be16 tci, old_tci;
1753 struct xport *xport;
1754 struct xlate_bond_recirc xr;
1755 bool use_recirc = false;
1757 vid = output_vlan_to_vid(out_xbundle, vlan);
1758 if (list_is_empty(&out_xbundle->xports)) {
1759 /* Partially configured bundle with no slaves. Drop the packet. */
1761 } else if (!out_xbundle->bond) {
1762 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1765 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1766 struct flow_wildcards *wc = ctx->wc;
1767 struct ofport_dpif *ofport;
1769 if (ctx->xbridge->support.odp.recirc) {
1770 use_recirc = bond_may_recirc(
1771 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1774 /* Only TCP mode uses recirculation. */
1775 xr.hash_alg = OVS_HASH_ALG_L4;
1776 bond_update_post_recirc_rules(out_xbundle->bond, false);
1778 /* Recirculation does not require unmasking hash fields. */
1783 ofport = bond_choose_output_slave(out_xbundle->bond,
1784 &ctx->xin->flow, wc, vid);
1785 xport = xport_lookup(xcfg, ofport);
1788 /* No slaves enabled, so drop packet. */
1792 /* If use_recirc is set, the main thread will handle stats
1793 * accounting for this bond. */
1795 if (ctx->xin->resubmit_stats) {
1796 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1797 ctx->xin->resubmit_stats->n_bytes);
1799 if (ctx->xin->xcache) {
1800 struct xc_entry *entry;
1803 flow = &ctx->xin->flow;
1804 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1805 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1806 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1807 entry->u.bond.vid = vid;
1812 old_tci = *flow_tci;
1814 if (tci || out_xbundle->use_priority_tags) {
1815 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1817 tci |= htons(VLAN_CFI);
1822 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1823 *flow_tci = old_tci;
1826 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1827 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1828 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1830 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1832 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1836 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1837 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1841 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1842 if (flow->nw_proto == ARP_OP_REPLY) {
1844 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1845 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1846 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1848 return flow->nw_src == flow->nw_dst;
1854 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1855 * dropped. Returns true if they may be forwarded, false if they should be
1858 * 'in_port' must be the xport that corresponds to flow->in_port.
1859 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1861 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1862 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1863 * checked by input_vid_is_valid().
1865 * May also add tags to '*tags', although the current implementation only does
1866 * so in one special case.
1869 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1872 struct xbundle *in_xbundle = in_port->xbundle;
1873 const struct xbridge *xbridge = ctx->xbridge;
1874 struct flow *flow = &ctx->xin->flow;
1876 /* Drop frames for reserved multicast addresses
1877 * only if forward_bpdu option is absent. */
1878 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1879 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1883 if (in_xbundle->bond) {
1884 struct mac_entry *mac;
1886 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1892 xlate_report(ctx, "bonding refused admissibility, dropping");
1895 case BV_DROP_IF_MOVED:
1896 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1897 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1899 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1900 && (!is_gratuitous_arp(flow, ctx->wc)
1901 || mac_entry_is_grat_arp_locked(mac))) {
1902 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1903 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1907 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1915 /* Checks whether a MAC learning update is necessary for MAC learning table
1916 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1919 * Most packets processed through the MAC learning table do not actually
1920 * change it in any way. This function requires only a read lock on the MAC
1921 * learning table, so it is much cheaper in this common case.
1923 * Keep the code here synchronized with that in update_learning_table__()
1926 is_mac_learning_update_needed(const struct mac_learning *ml,
1927 const struct flow *flow,
1928 struct flow_wildcards *wc,
1929 int vlan, struct xbundle *in_xbundle)
1930 OVS_REQ_RDLOCK(ml->rwlock)
1932 struct mac_entry *mac;
1934 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1938 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1939 if (!mac || mac_entry_age(ml, mac)) {
1943 if (is_gratuitous_arp(flow, wc)) {
1944 /* We don't want to learn from gratuitous ARP packets that are
1945 * reflected back over bond slaves so we lock the learning table. */
1946 if (!in_xbundle->bond) {
1948 } else if (mac_entry_is_grat_arp_locked(mac)) {
1953 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1957 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1958 * received on 'in_xbundle' in 'vlan'.
1960 * This code repeats all the checks in is_mac_learning_update_needed() because
1961 * the lock was released between there and here and thus the MAC learning state
1962 * could have changed.
1964 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1967 update_learning_table__(const struct xbridge *xbridge,
1968 const struct flow *flow, struct flow_wildcards *wc,
1969 int vlan, struct xbundle *in_xbundle)
1970 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1972 struct mac_entry *mac;
1974 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1978 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1979 if (is_gratuitous_arp(flow, wc)) {
1980 /* We don't want to learn from gratuitous ARP packets that are
1981 * reflected back over bond slaves so we lock the learning table. */
1982 if (!in_xbundle->bond) {
1983 mac_entry_set_grat_arp_lock(mac);
1984 } else if (mac_entry_is_grat_arp_locked(mac)) {
1989 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1990 /* The log messages here could actually be useful in debugging,
1991 * so keep the rate limit relatively high. */
1992 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1994 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1995 "on port %s in VLAN %d",
1996 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1997 in_xbundle->name, vlan);
1999 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
2004 update_learning_table(const struct xbridge *xbridge,
2005 const struct flow *flow, struct flow_wildcards *wc,
2006 int vlan, struct xbundle *in_xbundle)
2010 /* Don't learn the OFPP_NONE port. */
2011 if (in_xbundle == &ofpp_none_bundle) {
2015 /* First try the common case: no change to MAC learning table. */
2016 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
2017 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
2019 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2022 /* Slow path: MAC learning table might need an update. */
2023 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2024 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2025 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2029 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2030 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2032 update_mcast_snooping_table4__(const struct xbridge *xbridge,
2033 const struct flow *flow,
2034 struct mcast_snooping *ms, int vlan,
2035 struct xbundle *in_xbundle,
2036 const struct dp_packet *packet)
2037 OVS_REQ_WRLOCK(ms->rwlock)
2039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2041 ovs_be32 ip4 = flow->igmp_group_ip4;
2043 switch (ntohs(flow->tp_src)) {
2044 case IGMP_HOST_MEMBERSHIP_REPORT:
2045 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2046 if (mcast_snooping_add_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2047 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2048 IP_FMT" is on port %s in VLAN %d",
2049 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2052 case IGMP_HOST_LEAVE_MESSAGE:
2053 if (mcast_snooping_leave_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2054 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2055 IP_FMT" is on port %s in VLAN %d",
2056 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2059 case IGMP_HOST_MEMBERSHIP_QUERY:
2060 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2061 in_xbundle->ofbundle)) {
2062 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2063 IP_FMT" is on port %s in VLAN %d",
2064 xbridge->name, IP_ARGS(flow->nw_src),
2065 in_xbundle->name, vlan);
2068 case IGMPV3_HOST_MEMBERSHIP_REPORT:
2069 if ((count = mcast_snooping_add_report(ms, packet, vlan,
2070 in_xbundle->ofbundle))) {
2071 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2072 "addresses on port %s in VLAN %d",
2073 xbridge->name, count, in_xbundle->name, vlan);
2080 update_mcast_snooping_table6__(const struct xbridge *xbridge,
2081 const struct flow *flow,
2082 struct mcast_snooping *ms, int vlan,
2083 struct xbundle *in_xbundle,
2084 const struct dp_packet *packet)
2085 OVS_REQ_WRLOCK(ms->rwlock)
2087 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2090 switch (ntohs(flow->tp_src)) {
2092 if (!ipv6_addr_equals(&flow->ipv6_src, &in6addr_any)
2093 && mcast_snooping_add_mrouter(ms, vlan, in_xbundle->ofbundle)) {
2094 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query on port %s"
2096 xbridge->name, in_xbundle->name, vlan);
2102 count = mcast_snooping_add_mld(ms, packet, vlan, in_xbundle->ofbundle);
2104 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2105 "addresses on port %s in VLAN %d",
2106 xbridge->name, count, in_xbundle->name, vlan);
2112 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2113 * was received on 'in_xbundle' in 'vlan'. */
2115 update_mcast_snooping_table(const struct xbridge *xbridge,
2116 const struct flow *flow, int vlan,
2117 struct xbundle *in_xbundle,
2118 const struct dp_packet *packet)
2120 struct mcast_snooping *ms = xbridge->ms;
2121 struct xlate_cfg *xcfg;
2122 struct xbundle *mcast_xbundle;
2123 struct mcast_port_bundle *fport;
2125 /* Don't learn the OFPP_NONE port. */
2126 if (in_xbundle == &ofpp_none_bundle) {
2130 /* Don't learn from flood ports */
2131 mcast_xbundle = NULL;
2132 ovs_rwlock_wrlock(&ms->rwlock);
2133 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2134 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2135 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2136 if (mcast_xbundle == in_xbundle) {
2141 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2142 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2143 update_mcast_snooping_table4__(xbridge, flow, ms, vlan,
2144 in_xbundle, packet);
2146 update_mcast_snooping_table6__(xbridge, flow, ms, vlan,
2147 in_xbundle, packet);
2150 ovs_rwlock_unlock(&ms->rwlock);
2153 /* send the packet to ports having the multicast group learned */
2155 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2156 struct mcast_snooping *ms OVS_UNUSED,
2157 struct mcast_group *grp,
2158 struct xbundle *in_xbundle, uint16_t vlan)
2159 OVS_REQ_RDLOCK(ms->rwlock)
2161 struct xlate_cfg *xcfg;
2162 struct mcast_group_bundle *b;
2163 struct xbundle *mcast_xbundle;
2165 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2166 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2167 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2168 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2169 xlate_report(ctx, "forwarding to mcast group port");
2170 output_normal(ctx, mcast_xbundle, vlan);
2171 } else if (!mcast_xbundle) {
2172 xlate_report(ctx, "mcast group port is unknown, dropping");
2174 xlate_report(ctx, "mcast group port is input port, dropping");
2179 /* send the packet to ports connected to multicast routers */
2181 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2182 struct mcast_snooping *ms,
2183 struct xbundle *in_xbundle, uint16_t vlan)
2184 OVS_REQ_RDLOCK(ms->rwlock)
2186 struct xlate_cfg *xcfg;
2187 struct mcast_mrouter_bundle *mrouter;
2188 struct xbundle *mcast_xbundle;
2190 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2191 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2192 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2193 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2194 xlate_report(ctx, "forwarding to mcast router port");
2195 output_normal(ctx, mcast_xbundle, vlan);
2196 } else if (!mcast_xbundle) {
2197 xlate_report(ctx, "mcast router port is unknown, dropping");
2199 xlate_report(ctx, "mcast router port is input port, dropping");
2204 /* send the packet to ports flagged to be flooded */
2206 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2207 struct mcast_snooping *ms,
2208 struct xbundle *in_xbundle, uint16_t vlan)
2209 OVS_REQ_RDLOCK(ms->rwlock)
2211 struct xlate_cfg *xcfg;
2212 struct mcast_port_bundle *fport;
2213 struct xbundle *mcast_xbundle;
2215 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2216 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2217 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2218 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2219 xlate_report(ctx, "forwarding to mcast flood port");
2220 output_normal(ctx, mcast_xbundle, vlan);
2221 } else if (!mcast_xbundle) {
2222 xlate_report(ctx, "mcast flood port is unknown, dropping");
2224 xlate_report(ctx, "mcast flood port is input port, dropping");
2229 /* forward the Reports to configured ports */
2231 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2232 struct mcast_snooping *ms,
2233 struct xbundle *in_xbundle, uint16_t vlan)
2234 OVS_REQ_RDLOCK(ms->rwlock)
2236 struct xlate_cfg *xcfg;
2237 struct mcast_port_bundle *rport;
2238 struct xbundle *mcast_xbundle;
2240 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2241 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2242 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2243 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2244 xlate_report(ctx, "forwarding Report to mcast flagged port");
2245 output_normal(ctx, mcast_xbundle, vlan);
2246 } else if (!mcast_xbundle) {
2247 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2249 xlate_report(ctx, "mcast port is input port, dropping the Report");
2255 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2258 struct xbundle *xbundle;
2260 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2261 if (xbundle != in_xbundle
2262 && xbundle_includes_vlan(xbundle, vlan)
2263 && xbundle->floodable
2264 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2265 output_normal(ctx, xbundle, vlan);
2268 ctx->nf_output_iface = NF_OUT_FLOOD;
2272 xlate_normal(struct xlate_ctx *ctx)
2274 struct flow_wildcards *wc = ctx->wc;
2275 struct flow *flow = &ctx->xin->flow;
2276 struct xbundle *in_xbundle;
2277 struct xport *in_port;
2278 struct mac_entry *mac;
2283 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2284 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2285 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2287 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2288 ctx->xin->packet != NULL, &in_port);
2290 xlate_report(ctx, "no input bundle, dropping");
2294 /* Drop malformed frames. */
2295 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2296 !(flow->vlan_tci & htons(VLAN_CFI))) {
2297 if (ctx->xin->packet != NULL) {
2298 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2299 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2300 "VLAN tag received on port %s",
2301 ctx->xbridge->name, in_xbundle->name);
2303 xlate_report(ctx, "partial VLAN tag, dropping");
2307 /* Drop frames on bundles reserved for mirroring. */
2308 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2309 if (ctx->xin->packet != NULL) {
2310 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2311 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2312 "%s, which is reserved exclusively for mirroring",
2313 ctx->xbridge->name, in_xbundle->name);
2315 xlate_report(ctx, "input port is mirror output port, dropping");
2320 vid = vlan_tci_to_vid(flow->vlan_tci);
2321 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2322 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2325 vlan = input_vid_to_vlan(in_xbundle, vid);
2327 /* Check other admissibility requirements. */
2328 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2332 /* Learn source MAC. */
2333 if (ctx->xin->may_learn) {
2334 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2336 if (ctx->xin->xcache) {
2337 struct xc_entry *entry;
2339 /* Save enough info to update mac learning table later. */
2340 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2341 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2342 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2343 entry->u.normal.vlan = vlan;
2346 /* Determine output bundle. */
2347 if (mcast_snooping_enabled(ctx->xbridge->ms)
2348 && !eth_addr_is_broadcast(flow->dl_dst)
2349 && eth_addr_is_multicast(flow->dl_dst)
2350 && is_ip_any(flow)) {
2351 struct mcast_snooping *ms = ctx->xbridge->ms;
2352 struct mcast_group *grp = NULL;
2354 if (is_igmp(flow)) {
2355 if (mcast_snooping_is_membership(flow->tp_src) ||
2356 mcast_snooping_is_query(flow->tp_src)) {
2357 if (ctx->xin->may_learn) {
2358 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2359 in_xbundle, ctx->xin->packet);
2362 * IGMP packets need to take the slow path, in order to be
2363 * processed for mdb updates. That will prevent expires
2364 * firing off even after hosts have sent reports.
2366 ctx->xout->slow |= SLOW_ACTION;
2369 if (mcast_snooping_is_membership(flow->tp_src)) {
2370 ovs_rwlock_rdlock(&ms->rwlock);
2371 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2372 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2373 * forward IGMP Membership Reports only to those ports where
2374 * multicast routers are attached. Alternatively stated: a
2375 * snooping switch should not forward IGMP Membership Reports
2376 * to ports on which only hosts are attached.
2377 * An administrative control may be provided to override this
2378 * restriction, allowing the report messages to be flooded to
2380 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2381 ovs_rwlock_unlock(&ms->rwlock);
2383 xlate_report(ctx, "multicast traffic, flooding");
2384 xlate_normal_flood(ctx, in_xbundle, vlan);
2387 } else if (is_mld(flow)) {
2388 ctx->xout->slow |= SLOW_ACTION;
2389 if (ctx->xin->may_learn) {
2390 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2391 in_xbundle, ctx->xin->packet);
2393 if (is_mld_report(flow)) {
2394 ovs_rwlock_rdlock(&ms->rwlock);
2395 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2396 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2397 ovs_rwlock_unlock(&ms->rwlock);
2399 xlate_report(ctx, "MLD query, flooding");
2400 xlate_normal_flood(ctx, in_xbundle, vlan);
2403 if ((flow->dl_type == htons(ETH_TYPE_IP)
2404 && ip_is_local_multicast(flow->nw_dst))
2405 || (flow->dl_type == htons(ETH_TYPE_IPV6)
2406 && ipv6_is_all_hosts(&flow->ipv6_dst))) {
2407 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2408 * address in the 224.0.0.x range which are not IGMP must
2409 * be forwarded on all ports */
2410 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2411 xlate_normal_flood(ctx, in_xbundle, vlan);
2416 /* forwarding to group base ports */
2417 ovs_rwlock_rdlock(&ms->rwlock);
2418 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2419 grp = mcast_snooping_lookup4(ms, flow->nw_dst, vlan);
2420 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2421 grp = mcast_snooping_lookup(ms, &flow->ipv6_dst, vlan);
2424 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2425 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2426 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2428 if (mcast_snooping_flood_unreg(ms)) {
2429 xlate_report(ctx, "unregistered multicast, flooding");
2430 xlate_normal_flood(ctx, in_xbundle, vlan);
2432 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2433 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2436 ovs_rwlock_unlock(&ms->rwlock);
2438 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2439 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2440 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2441 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2444 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2445 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2446 if (mac_xbundle && mac_xbundle != in_xbundle) {
2447 xlate_report(ctx, "forwarding to learned port");
2448 output_normal(ctx, mac_xbundle, vlan);
2449 } else if (!mac_xbundle) {
2450 xlate_report(ctx, "learned port is unknown, dropping");
2452 xlate_report(ctx, "learned port is input port, dropping");
2455 xlate_report(ctx, "no learned MAC for destination, flooding");
2456 xlate_normal_flood(ctx, in_xbundle, vlan);
2461 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
2462 * 'probability' is the number of packets out of UINT32_MAX to sample. The
2463 * 'cookie' (of length 'cookie_size' bytes) is passed back in the callback for
2464 * each sampled packet. 'tunnel_out_port', if not ODPP_NONE, is added as the
2465 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions', an
2466 * OVS_USERSPACE_ATTR_ACTIONS attribute is added.
2469 compose_sample_action(struct xlate_ctx *ctx,
2470 const uint32_t probability,
2471 const union user_action_cookie *cookie,
2472 const size_t cookie_size,
2473 const odp_port_t tunnel_out_port,
2474 bool include_actions)
2476 size_t sample_offset = nl_msg_start_nested(ctx->odp_actions,
2477 OVS_ACTION_ATTR_SAMPLE);
2479 nl_msg_put_u32(ctx->odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2481 size_t actions_offset = nl_msg_start_nested(ctx->odp_actions,
2482 OVS_SAMPLE_ATTR_ACTIONS);
2484 odp_port_t odp_port = ofp_port_to_odp_port(
2485 ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
2486 uint32_t pid = dpif_port_get_pid(ctx->xbridge->dpif, odp_port,
2487 flow_hash_5tuple(&ctx->xin->flow, 0));
2488 int cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2493 nl_msg_end_nested(ctx->odp_actions, actions_offset);
2494 nl_msg_end_nested(ctx->odp_actions, sample_offset);
2496 return cookie_offset;
2499 /* If sFLow is not enabled, returns 0 without doing anything.
2501 * If sFlow is enabled, appends a template "sample" action to the ODP actions
2502 * in 'ctx'. This action is a template because some of the information needed
2503 * to fill it out is not available until flow translation is complete. In this
2504 * case, this functions returns an offset, which is always nonzero, to pass
2505 * later to fix_sflow_action() to fill in the rest of the template. */
2507 compose_sflow_action(struct xlate_ctx *ctx)
2509 struct dpif_sflow *sflow = ctx->xbridge->sflow;
2510 if (!sflow || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2514 union user_action_cookie cookie = { .type = USER_ACTION_COOKIE_SFLOW };
2515 return compose_sample_action(ctx, dpif_sflow_get_probability(sflow),
2516 &cookie, sizeof cookie.sflow, ODPP_NONE,
2520 /* If IPFIX is enabled, this appends a "sample" action to implement IPFIX to
2521 * 'ctx->odp_actions'. */
2523 compose_ipfix_action(struct xlate_ctx *ctx, odp_port_t output_odp_port)
2525 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
2526 odp_port_t tunnel_out_port = ODPP_NONE;
2528 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2532 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2534 if (output_odp_port == ODPP_NONE &&
2535 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix)) {
2539 /* For output case, output_odp_port is valid*/
2540 if (output_odp_port != ODPP_NONE) {
2541 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix)) {
2544 /* If tunnel sampling is enabled, put an additional option attribute:
2545 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2547 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix) &&
2548 dpif_ipfix_get_tunnel_port(ipfix, output_odp_port) ) {
2549 tunnel_out_port = output_odp_port;
2553 union user_action_cookie cookie = {
2555 .type = USER_ACTION_COOKIE_IPFIX,
2556 .output_odp_port = output_odp_port,
2559 compose_sample_action(ctx,
2560 dpif_ipfix_get_bridge_exporter_probability(ipfix),
2561 &cookie, sizeof cookie.ipfix, tunnel_out_port,
2565 /* Fix "sample" action according to data collected while composing ODP actions,
2566 * as described in compose_sflow_action().
2568 * 'user_cookie_offset' must be the offset returned by add_sflow_action(). */
2570 fix_sflow_action(struct xlate_ctx *ctx, unsigned int user_cookie_offset)
2572 const struct flow *base = &ctx->base_flow;
2573 union user_action_cookie *cookie;
2575 cookie = ofpbuf_at(ctx->odp_actions, user_cookie_offset,
2576 sizeof cookie->sflow);
2577 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2579 cookie->type = USER_ACTION_COOKIE_SFLOW;
2580 cookie->sflow.vlan_tci = base->vlan_tci;
2582 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2583 * port information") for the interpretation of cookie->output. */
2584 switch (ctx->sflow_n_outputs) {
2586 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2587 cookie->sflow.output = 0x40000000 | 256;
2591 cookie->sflow.output = dpif_sflow_odp_port_to_ifindex(
2592 ctx->xbridge->sflow, ctx->sflow_odp_port);
2593 if (cookie->sflow.output) {
2598 /* 0x80000000 means "multiple output ports. */
2599 cookie->sflow.output = 0x80000000 | ctx->sflow_n_outputs;
2605 process_special(struct xlate_ctx *ctx, const struct xport *xport)
2607 const struct flow *flow = &ctx->xin->flow;
2608 struct flow_wildcards *wc = ctx->wc;
2609 const struct xbridge *xbridge = ctx->xbridge;
2610 const struct dp_packet *packet = ctx->xin->packet;
2611 enum slow_path_reason slow;
2615 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2617 cfm_process_heartbeat(xport->cfm, packet);
2620 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2622 bfd_process_packet(xport->bfd, flow, packet);
2623 /* If POLL received, immediately sends FINAL back. */
2624 if (bfd_should_send_packet(xport->bfd)) {
2625 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2629 } else if (xport->xbundle && xport->xbundle->lacp
2630 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2632 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2635 } else if ((xbridge->stp || xbridge->rstp) &&
2636 stp_should_process_flow(flow, wc)) {
2639 ? stp_process_packet(xport, packet)
2640 : rstp_process_packet(xport, packet);
2643 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2645 lldp_process_packet(xport->lldp, packet);
2653 ctx->xout->slow |= slow;
2661 tnl_route_lookup_flow(const struct flow *oflow,
2662 ovs_be32 *ip, struct xport **out_port)
2664 char out_dev[IFNAMSIZ];
2665 struct xbridge *xbridge;
2666 struct xlate_cfg *xcfg;
2669 if (!ovs_router_lookup4(oflow->tunnel.ip_dst, out_dev, &gw)) {
2676 *ip = oflow->tunnel.ip_dst;
2679 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2682 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2683 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2686 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2687 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2698 compose_table_xlate(struct xlate_ctx *ctx, const struct xport *out_dev,
2699 struct dp_packet *packet)
2701 struct xbridge *xbridge = out_dev->xbridge;
2702 struct ofpact_output output;
2705 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2706 flow_extract(packet, &flow);
2707 flow.in_port.ofp_port = out_dev->ofp_port;
2708 output.port = OFPP_TABLE;
2711 return ofproto_dpif_execute_actions__(xbridge->ofproto, &flow, NULL,
2712 &output.ofpact, sizeof output,
2713 ctx->recurse, ctx->resubmits, packet);
2717 tnl_send_arp_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2718 const struct eth_addr eth_src,
2719 ovs_be32 ip_src, ovs_be32 ip_dst)
2721 struct dp_packet packet;
2723 dp_packet_init(&packet, 0);
2724 compose_arp(&packet, ARP_OP_REQUEST,
2725 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2727 compose_table_xlate(ctx, out_dev, &packet);
2728 dp_packet_uninit(&packet);
2732 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2733 const struct flow *flow, odp_port_t tunnel_odp_port)
2735 struct ovs_action_push_tnl tnl_push_data;
2736 struct xport *out_dev = NULL;
2737 ovs_be32 s_ip, d_ip = 0;
2738 struct eth_addr smac;
2739 struct eth_addr dmac;
2742 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2744 xlate_report(ctx, "native tunnel routing failed");
2747 xlate_report(ctx, "tunneling to "IP_FMT" via %s",
2748 IP_ARGS(d_ip), netdev_get_name(out_dev->netdev));
2750 /* Use mac addr of bridge port of the peer. */
2751 err = netdev_get_etheraddr(out_dev->netdev, &smac);
2753 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2757 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2759 xlate_report(ctx, "tunnel output device lacks IPv4 address");
2763 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, &dmac);
2765 xlate_report(ctx, "ARP cache miss for "IP_FMT" on bridge %s, "
2766 "sending ARP request",
2767 IP_ARGS(d_ip), out_dev->xbridge->name);
2768 tnl_send_arp_request(ctx, out_dev, smac, s_ip, d_ip);
2771 if (ctx->xin->xcache) {
2772 struct xc_entry *entry;
2774 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2775 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2776 sizeof entry->u.tnl_arp_cache.br_name);
2777 entry->u.tnl_arp_cache.d_ip = d_ip;
2780 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" "IP_FMT
2781 " to "ETH_ADDR_FMT" "IP_FMT,
2782 ETH_ADDR_ARGS(smac), IP_ARGS(s_ip),
2783 ETH_ADDR_ARGS(dmac), IP_ARGS(d_ip));
2784 err = tnl_port_build_header(xport->ofport, flow,
2785 dmac, smac, s_ip, &tnl_push_data);
2789 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2790 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2791 odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
2796 xlate_commit_actions(struct xlate_ctx *ctx)
2798 bool use_masked = ctx->xbridge->support.masked_set_action;
2800 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2801 ctx->odp_actions, ctx->wc,
2806 clear_conntrack(struct flow *flow)
2814 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2815 const struct xlate_bond_recirc *xr, bool check_stp)
2817 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2818 struct flow_wildcards *wc = ctx->wc;
2819 struct flow *flow = &ctx->xin->flow;
2820 struct flow_tnl flow_tnl;
2821 ovs_be16 flow_vlan_tci;
2822 uint32_t flow_pkt_mark;
2823 uint8_t flow_nw_tos;
2824 odp_port_t out_port, odp_port;
2825 bool tnl_push_pop_send = false;
2828 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2829 * before traversing a patch port. */
2830 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 34);
2831 memset(&flow_tnl, 0, sizeof flow_tnl);
2834 xlate_report(ctx, "Nonexistent output port");
2836 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2837 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2839 } else if (check_stp) {
2840 if (is_stp(&ctx->base_flow)) {
2841 if (!xport_stp_should_forward_bpdu(xport) &&
2842 !xport_rstp_should_manage_bpdu(xport)) {
2843 if (ctx->xbridge->stp != NULL) {
2844 xlate_report(ctx, "STP not in listening state, "
2845 "skipping bpdu output");
2846 } else if (ctx->xbridge->rstp != NULL) {
2847 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2848 "skipping bpdu output");
2852 } else if (!xport_stp_forward_state(xport) ||
2853 !xport_rstp_forward_state(xport)) {
2854 if (ctx->xbridge->stp != NULL) {
2855 xlate_report(ctx, "STP not in forwarding state, "
2857 } else if (ctx->xbridge->rstp != NULL) {
2858 xlate_report(ctx, "RSTP not in forwarding state, "
2866 const struct xport *peer = xport->peer;
2867 struct flow old_flow = ctx->xin->flow;
2868 bool old_conntrack = ctx->conntracked;
2869 bool old_was_mpls = ctx->was_mpls;
2870 cls_version_t old_version = ctx->tables_version;
2871 struct ofpbuf old_stack = ctx->stack;
2872 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2873 struct ofpbuf old_action_set = ctx->action_set;
2874 uint64_t actset_stub[1024 / 8];
2876 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2877 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2878 ctx->xbridge = peer->xbridge;
2879 flow->in_port.ofp_port = peer->ofp_port;
2880 flow->metadata = htonll(0);
2881 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2882 memset(flow->regs, 0, sizeof flow->regs);
2883 flow->actset_output = OFPP_UNSET;
2884 ctx->conntracked = false;
2885 clear_conntrack(flow);
2887 /* The bridge is now known so obtain its table version. */
2889 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
2891 if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
2892 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2893 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2894 if (ctx->action_set.size) {
2895 /* Translate action set only if not dropping the packet and
2896 * not recirculating. */
2897 if (!exit_recirculates(ctx)) {
2898 xlate_action_set(ctx);
2901 /* Check if need to recirculate. */
2902 if (exit_recirculates(ctx)) {
2903 compose_recirculate_action(ctx);
2906 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2907 * the learning action look at the packet, then drop it. */
2908 struct flow old_base_flow = ctx->base_flow;
2909 size_t old_size = ctx->odp_actions->size;
2910 mirror_mask_t old_mirrors = ctx->mirrors;
2912 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2913 ctx->mirrors = old_mirrors;
2914 ctx->base_flow = old_base_flow;
2915 ctx->odp_actions->size = old_size;
2917 /* Undo changes that may have been done for recirculation. */
2918 if (exit_recirculates(ctx)) {
2919 ctx->action_set.size = ctx->recirc_action_offset;
2920 ctx->recirc_action_offset = -1;
2921 ctx->last_unroll_offset = -1;
2926 ctx->xin->flow = old_flow;
2927 ctx->xbridge = xport->xbridge;
2928 ofpbuf_uninit(&ctx->action_set);
2929 ctx->action_set = old_action_set;
2930 ofpbuf_uninit(&ctx->stack);
2931 ctx->stack = old_stack;
2933 /* Restore calling bridge's lookup version. */
2934 ctx->tables_version = old_version;
2936 /* The peer bridge popping MPLS should have no effect on the original
2938 ctx->was_mpls = old_was_mpls;
2940 /* The peer bridge's conntrack execution should have no effect on the
2941 * original bridge. */
2942 ctx->conntracked = old_conntrack;
2944 /* The fact that the peer bridge exits (for any reason) does not mean
2945 * that the original bridge should exit. Specifically, if the peer
2946 * bridge recirculates (which typically modifies the packet), the
2947 * original bridge must continue processing with the original, not the
2948 * recirculated packet! */
2951 if (ctx->xin->resubmit_stats) {
2952 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2953 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2955 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2958 if (ctx->xin->xcache) {
2959 struct xc_entry *entry;
2961 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2962 entry->u.dev.tx = netdev_ref(xport->netdev);
2963 entry->u.dev.rx = netdev_ref(peer->netdev);
2964 entry->u.dev.bfd = bfd_ref(peer->bfd);
2969 flow_vlan_tci = flow->vlan_tci;
2970 flow_pkt_mark = flow->pkt_mark;
2971 flow_nw_tos = flow->nw_tos;
2973 if (count_skb_priorities(xport)) {
2974 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2975 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2976 wc->masks.nw_tos |= IP_DSCP_MASK;
2977 flow->nw_tos &= ~IP_DSCP_MASK;
2978 flow->nw_tos |= dscp;
2982 if (xport->is_tunnel) {
2983 /* Save tunnel metadata so that changes made due to
2984 * the Logical (tunnel) Port are not visible for any further
2985 * matches, while explicit set actions on tunnel metadata are.
2987 flow_tnl = flow->tunnel;
2988 odp_port = tnl_port_send(xport->ofport, flow, ctx->wc);
2989 if (odp_port == ODPP_NONE) {
2990 xlate_report(ctx, "Tunneling decided against output");
2991 goto out; /* restore flow_nw_tos */
2993 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2994 xlate_report(ctx, "Not tunneling to our own address");
2995 goto out; /* restore flow_nw_tos */
2997 if (ctx->xin->resubmit_stats) {
2998 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
3000 if (ctx->xin->xcache) {
3001 struct xc_entry *entry;
3003 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3004 entry->u.dev.tx = netdev_ref(xport->netdev);
3006 out_port = odp_port;
3007 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3008 xlate_report(ctx, "output to native tunnel");
3009 tnl_push_pop_send = true;
3011 xlate_report(ctx, "output to kernel tunnel");
3012 commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions);
3013 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3016 odp_port = xport->odp_port;
3017 out_port = odp_port;
3018 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
3019 ofp_port_t vlandev_port;
3021 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3022 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
3023 ofp_port, flow->vlan_tci);
3024 if (vlandev_port != ofp_port) {
3025 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
3026 flow->vlan_tci = htons(0);
3031 if (out_port != ODPP_NONE) {
3032 xlate_commit_actions(ctx);
3035 struct ovs_action_hash *act_hash;
3038 act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
3039 OVS_ACTION_ATTR_HASH,
3041 act_hash->hash_alg = xr->hash_alg;
3042 act_hash->hash_basis = xr->hash_basis;
3044 /* Recirc action. */
3045 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
3049 if (tnl_push_pop_send) {
3050 build_tunnel_send(ctx, xport, flow, odp_port);
3051 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3053 odp_port_t odp_tnl_port = ODPP_NONE;
3055 /* XXX: Write better Filter for tunnel port. We can use inport
3056 * int tunnel-port flow to avoid these checks completely. */
3057 if (ofp_port == OFPP_LOCAL &&
3058 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3060 odp_tnl_port = tnl_port_map_lookup(flow, wc);
3063 if (odp_tnl_port != ODPP_NONE) {
3064 nl_msg_put_odp_port(ctx->odp_actions,
3065 OVS_ACTION_ATTR_TUNNEL_POP,
3068 /* Tunnel push-pop action is not compatible with
3070 compose_ipfix_action(ctx, out_port);
3071 nl_msg_put_odp_port(ctx->odp_actions,
3072 OVS_ACTION_ATTR_OUTPUT,
3078 ctx->sflow_odp_port = odp_port;
3079 ctx->sflow_n_outputs++;
3080 ctx->nf_output_iface = ofp_port;
3083 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
3084 mirror_packet(ctx, xport->xbundle,
3085 xbundle_mirror_dst(xport->xbundle->xbridge,
3091 flow->vlan_tci = flow_vlan_tci;
3092 flow->pkt_mark = flow_pkt_mark;
3093 flow->nw_tos = flow_nw_tos;
3097 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3098 const struct xlate_bond_recirc *xr)
3100 compose_output_action__(ctx, ofp_port, xr, true);
3104 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3106 struct rule_dpif *old_rule = ctx->rule;
3107 ovs_be64 old_cookie = ctx->rule_cookie;
3108 const struct rule_actions *actions;
3110 if (ctx->xin->resubmit_stats) {
3111 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3117 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3118 actions = rule_dpif_get_actions(rule);
3119 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3120 ctx->rule_cookie = old_cookie;
3121 ctx->rule = old_rule;
3126 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3128 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3130 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3131 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
3132 MAX_RESUBMIT_RECURSION);
3133 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3134 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
3135 } else if (ctx->odp_actions->size > UINT16_MAX) {
3136 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
3137 } else if (ctx->stack.size >= 65536) {
3138 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
3147 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3148 bool may_packet_in, bool honor_table_miss)
3150 /* Check if we need to recirculate before matching in a table. */
3151 if (ctx->was_mpls) {
3152 ctx_trigger_recirculation(ctx);
3155 if (xlate_resubmit_resource_check(ctx)) {
3156 uint8_t old_table_id = ctx->table_id;
3157 struct rule_dpif *rule;
3159 ctx->table_id = table_id;
3161 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3162 ctx->tables_version,
3163 &ctx->xin->flow, ctx->xin->wc,
3164 ctx->xin->resubmit_stats,
3165 &ctx->table_id, in_port,
3166 may_packet_in, honor_table_miss);
3168 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3169 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3173 /* Fill in the cache entry here instead of xlate_recursively
3174 * to make the reference counting more explicit. We take a
3175 * reference in the lookups above if we are going to cache the
3177 if (ctx->xin->xcache) {
3178 struct xc_entry *entry;
3180 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3181 entry->u.rule = rule;
3182 rule_dpif_ref(rule);
3184 xlate_recursively(ctx, rule);
3187 ctx->table_id = old_table_id;
3195 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3196 struct ofputil_bucket *bucket)
3198 if (ctx->xin->resubmit_stats) {
3199 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3201 if (ctx->xin->xcache) {
3202 struct xc_entry *entry;
3204 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3205 entry->u.group.group = group_dpif_ref(group);
3206 entry->u.group.bucket = bucket;
3211 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3213 uint64_t action_list_stub[1024 / 8];
3214 struct ofpbuf action_list, action_set;
3215 struct flow old_flow = ctx->xin->flow;
3216 bool old_was_mpls = ctx->was_mpls;
3218 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3219 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3221 ofpacts_execute_action_set(&action_list, &action_set);
3223 do_xlate_actions(action_list.data, action_list.size, ctx);
3226 ofpbuf_uninit(&action_set);
3227 ofpbuf_uninit(&action_list);
3229 /* Check if need to recirculate. */
3230 if (exit_recirculates(ctx)) {
3231 compose_recirculate_action(ctx);
3234 /* Roll back flow to previous state.
3235 * This is equivalent to cloning the packet for each bucket.
3237 * As a side effect any subsequently applied actions will
3238 * also effectively be applied to a clone of the packet taken
3239 * just before applying the all or indirect group.
3241 * Note that group buckets are action sets, hence they cannot modify the
3242 * main action set. Also any stack actions are ignored when executing an
3243 * action set, so group buckets cannot change the stack either.
3244 * However, we do allow resubmit actions in group buckets, which could
3245 * break the above assumptions. It is up to the controller to not mess up
3246 * with the action_set and stack in the tables resubmitted to from
3248 ctx->xin->flow = old_flow;
3250 /* The group bucket popping MPLS should have no effect after bucket
3252 ctx->was_mpls = old_was_mpls;
3254 /* The fact that the group bucket exits (for any reason) does not mean that
3255 * the translation after the group action should exit. Specifically, if
3256 * the group bucket recirculates (which typically modifies the packet), the
3257 * actions after the group action must continue processing with the
3258 * original, not the recirculated packet! */
3263 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3265 struct ofputil_bucket *bucket;
3266 const struct ovs_list *buckets;
3268 group_dpif_get_buckets(group, &buckets);
3270 LIST_FOR_EACH (bucket, list_node, buckets) {
3271 xlate_group_bucket(ctx, bucket);
3273 xlate_group_stats(ctx, group, NULL);
3277 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3279 struct ofputil_bucket *bucket;
3281 bucket = group_first_live_bucket(ctx, group, 0);
3283 xlate_group_bucket(ctx, bucket);
3284 xlate_group_stats(ctx, group, bucket);
3289 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3291 struct flow_wildcards *wc = ctx->wc;
3292 struct ofputil_bucket *bucket;
3295 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3296 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3297 bucket = group_best_live_bucket(ctx, group, basis);
3299 xlate_group_bucket(ctx, bucket);
3300 xlate_group_stats(ctx, group, bucket);
3305 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3307 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3308 const struct field_array *fields;
3309 struct ofputil_bucket *bucket;
3313 fields = group_dpif_get_fields(group);
3314 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3316 /* Determine which fields to hash */
3317 for (i = 0; i < MFF_N_IDS; i++) {
3318 if (bitmap_is_set(fields->used.bm, i)) {
3319 const struct mf_field *mf;
3321 /* If the field is already present in 'hash_fields' then
3322 * this loop has already checked that it and its pre-requisites
3323 * are present in the flow and its pre-requisites have
3324 * already been added to 'hash_fields'. There is nothing more
3325 * to do here and as an optimisation the loop can continue. */
3326 if (bitmap_is_set(hash_fields.bm, i)) {
3332 /* Only hash a field if it and its pre-requisites are present
3334 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3338 /* Hash both the field and its pre-requisites */
3339 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3343 /* Hash the fields */
3344 for (i = 0; i < MFF_N_IDS; i++) {
3345 if (bitmap_is_set(hash_fields.bm, i)) {
3346 const struct mf_field *mf = mf_from_id(i);
3347 union mf_value value;
3350 mf_get_value(mf, &ctx->xin->flow, &value);
3351 /* This seems inefficient but so does apply_mask() */
3352 for (j = 0; j < mf->n_bytes; j++) {
3353 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3355 basis = hash_bytes(&value, mf->n_bytes, basis);
3357 /* For tunnels, hash in whether the field is present. */
3358 if (mf_is_tun_metadata(mf)) {
3359 basis = hash_boolean(mf_is_set(mf, &ctx->xin->flow), basis);
3362 mf_mask_field(mf, &ctx->wc->masks);
3366 bucket = group_best_live_bucket(ctx, group, basis);
3368 xlate_group_bucket(ctx, bucket);
3369 xlate_group_stats(ctx, group, bucket);
3374 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3376 const char *selection_method = group_dpif_get_selection_method(group);
3378 if (selection_method[0] == '\0') {
3379 xlate_default_select_group(ctx, group);
3380 } else if (!strcasecmp("hash", selection_method)) {
3381 xlate_hash_fields_select_group(ctx, group);
3383 /* Parsing of groups should ensure this never happens */
3389 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3391 bool was_in_group = ctx->in_group;
3392 ctx->in_group = true;
3394 switch (group_dpif_get_type(group)) {
3396 case OFPGT11_INDIRECT:
3397 xlate_all_group(ctx, group);
3399 case OFPGT11_SELECT:
3400 xlate_select_group(ctx, group);
3403 xlate_ff_group(ctx, group);
3408 group_dpif_unref(group);
3410 ctx->in_group = was_in_group;
3414 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3416 if (xlate_resubmit_resource_check(ctx)) {
3417 struct group_dpif *group;
3420 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3422 xlate_group_action__(ctx, group);
3432 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3433 const struct ofpact_resubmit *resubmit)
3437 bool may_packet_in = false;
3438 bool honor_table_miss = false;
3440 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3441 /* Still allow missed packets to be sent to the controller
3442 * if resubmitting from an internal table. */
3443 may_packet_in = true;
3444 honor_table_miss = true;
3447 in_port = resubmit->in_port;
3448 if (in_port == OFPP_IN_PORT) {
3449 in_port = ctx->xin->flow.in_port.ofp_port;
3452 table_id = resubmit->table_id;
3453 if (table_id == 255) {
3454 table_id = ctx->table_id;
3457 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3462 flood_packets(struct xlate_ctx *ctx, bool all)
3464 const struct xport *xport;
3466 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3467 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3472 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3473 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3474 compose_output_action(ctx, xport->ofp_port, NULL);
3478 ctx->nf_output_iface = NF_OUT_FLOOD;
3482 execute_controller_action(struct xlate_ctx *ctx, int len,
3483 enum ofp_packet_in_reason reason,
3484 uint16_t controller_id)
3486 struct ofproto_packet_in *pin;
3487 struct dp_packet *packet;
3489 ctx->xout->slow |= SLOW_CONTROLLER;
3490 if (!ctx->xin->packet) {
3494 packet = dp_packet_clone(ctx->xin->packet);
3496 xlate_commit_actions(ctx);
3498 odp_execute_actions(NULL, &packet, 1, false,
3499 ctx->odp_actions->data, ctx->odp_actions->size, NULL);
3501 pin = xmalloc(sizeof *pin);
3502 pin->up.packet_len = dp_packet_size(packet);
3503 pin->up.packet = dp_packet_steal_data(packet);
3504 pin->up.reason = reason;
3505 pin->up.table_id = ctx->table_id;
3506 pin->up.cookie = ctx->rule_cookie;
3508 flow_get_metadata(&ctx->xin->flow, &pin->up.flow_metadata);
3510 pin->controller_id = controller_id;
3511 pin->send_len = len;
3512 /* If a rule is a table-miss rule then this is
3513 * a table-miss handled by a table-miss rule.
3515 * Else, if rule is internal and has a controller action,
3516 * the later being implied by the rule being processed here,
3517 * then this is a table-miss handled without a table-miss rule.
3519 * Otherwise this is not a table-miss. */
3520 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3522 if (rule_dpif_is_table_miss(ctx->rule)) {
3523 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3524 } else if (rule_dpif_is_internal(ctx->rule)) {
3525 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3528 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3529 dp_packet_delete(packet);
3533 compose_recirculate_action__(struct xlate_ctx *ctx, uint8_t table)
3535 struct recirc_metadata md;
3538 recirc_metadata_from_flow(&md, &ctx->xin->flow);
3540 ovs_assert(ctx->recirc_action_offset >= 0);
3542 struct recirc_state state = {
3544 .ofproto = ctx->xbridge->ofproto,
3546 .stack = &ctx->stack,
3547 .mirrors = ctx->mirrors,
3548 .conntracked = ctx->conntracked,
3549 .action_set_len = ctx->recirc_action_offset,
3550 .ofpacts_len = ctx->action_set.size,
3551 .ofpacts = ctx->action_set.data,
3554 /* Only allocate recirculation ID if we have a packet. */
3555 if (ctx->xin->packet) {
3556 /* Allocate a unique recirc id for the given metadata state in the
3557 * flow. The life-cycle of this recirc id is managed by associating it
3558 * with the udpif key ('ukey') created for each new datapath flow. */
3559 id = recirc_alloc_id_ctx(&state);
3561 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3562 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3565 xlate_out_add_recirc(ctx->xout, id);
3567 /* Look up an existing recirc id for the given metadata state in the
3568 * flow. No new reference is taken, as the ID is RCU protected and is
3569 * only required temporarily for verification.
3571 * This might fail and return 0. We let zero 'id' to be used in the
3572 * RECIRC action below, which will fail all revalidations as zero is
3573 * not a valid recirculation ID. */
3574 id = recirc_find_id(&state);
3577 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3579 /* Undo changes done by recirculation. */
3580 ctx->action_set.size = ctx->recirc_action_offset;
3581 ctx->recirc_action_offset = -1;
3582 ctx->last_unroll_offset = -1;
3585 /* Called only when ctx->recirc_action_offset is set. */
3587 compose_recirculate_action(struct xlate_ctx *ctx)
3589 xlate_commit_actions(ctx);
3590 compose_recirculate_action__(ctx, 0);
3594 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3596 struct flow *flow = &ctx->xin->flow;
3599 ovs_assert(eth_type_mpls(mpls->ethertype));
3601 n = flow_count_mpls_labels(flow, ctx->wc);
3603 xlate_commit_actions(ctx);
3604 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3605 if (ctx->xin->packet != NULL) {
3606 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3607 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3608 "MPLS push action can't be performed as it would "
3609 "have more MPLS LSEs than the %d supported.",
3610 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3616 flow_push_mpls(flow, n, mpls->ethertype, ctx->wc);
3620 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3622 struct flow *flow = &ctx->xin->flow;
3623 int n = flow_count_mpls_labels(flow, ctx->wc);
3625 if (flow_pop_mpls(flow, n, eth_type, ctx->wc)) {
3626 if (ctx->xbridge->support.odp.recirc) {
3627 ctx->was_mpls = true;
3629 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3630 if (ctx->xin->packet != NULL) {
3631 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3632 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3633 "MPLS pop action can't be performed as it has "
3634 "more MPLS LSEs than the %d supported.",
3635 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3638 ofpbuf_clear(ctx->odp_actions);
3643 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3645 struct flow *flow = &ctx->xin->flow;
3647 if (!is_ip_any(flow)) {
3651 ctx->wc->masks.nw_ttl = 0xff;
3652 if (flow->nw_ttl > 1) {
3658 for (i = 0; i < ids->n_controllers; i++) {
3659 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3663 /* Stop processing for current table. */
3669 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3671 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3672 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3673 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3678 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3680 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3681 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3682 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3687 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3689 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3690 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3691 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3696 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3698 struct flow *flow = &ctx->xin->flow;
3700 if (eth_type_mpls(flow->dl_type)) {
3701 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3703 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3706 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3709 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3713 /* Stop processing for current table. */
3718 xlate_output_action(struct xlate_ctx *ctx,
3719 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3721 ofp_port_t prev_nf_output_iface = ctx->nf_output_iface;
3723 ctx->nf_output_iface = NF_OUT_DROP;
3727 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3730 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3731 0, may_packet_in, true);
3737 flood_packets(ctx, false);
3740 flood_packets(ctx, true);
3742 case OFPP_CONTROLLER:
3743 execute_controller_action(ctx, max_len,
3744 (ctx->in_group ? OFPR_GROUP
3745 : ctx->in_action_set ? OFPR_ACTION_SET
3753 if (port != ctx->xin->flow.in_port.ofp_port) {
3754 compose_output_action(ctx, port, NULL);
3756 xlate_report(ctx, "skipping output to input port");
3761 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3762 ctx->nf_output_iface = NF_OUT_FLOOD;
3763 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3764 ctx->nf_output_iface = prev_nf_output_iface;
3765 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3766 ctx->nf_output_iface != NF_OUT_FLOOD) {
3767 ctx->nf_output_iface = NF_OUT_MULTI;
3772 xlate_output_reg_action(struct xlate_ctx *ctx,
3773 const struct ofpact_output_reg *or)
3775 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3776 if (port <= UINT16_MAX) {
3777 union mf_subvalue value;
3779 memset(&value, 0xff, sizeof value);
3780 mf_write_subfield_flow(&or->src, &value, &ctx->wc->masks);
3781 xlate_output_action(ctx, u16_to_ofp(port),
3782 or->max_len, false);
3787 xlate_enqueue_action(struct xlate_ctx *ctx,
3788 const struct ofpact_enqueue *enqueue)
3790 ofp_port_t ofp_port = enqueue->port;
3791 uint32_t queue_id = enqueue->queue;
3792 uint32_t flow_priority, priority;
3795 /* Translate queue to priority. */
3796 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3798 /* Fall back to ordinary output action. */
3799 xlate_output_action(ctx, enqueue->port, 0, false);
3803 /* Check output port. */
3804 if (ofp_port == OFPP_IN_PORT) {
3805 ofp_port = ctx->xin->flow.in_port.ofp_port;
3806 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3810 /* Add datapath actions. */
3811 flow_priority = ctx->xin->flow.skb_priority;
3812 ctx->xin->flow.skb_priority = priority;
3813 compose_output_action(ctx, ofp_port, NULL);
3814 ctx->xin->flow.skb_priority = flow_priority;
3816 /* Update NetFlow output port. */
3817 if (ctx->nf_output_iface == NF_OUT_DROP) {
3818 ctx->nf_output_iface = ofp_port;
3819 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3820 ctx->nf_output_iface = NF_OUT_MULTI;
3825 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3827 uint32_t skb_priority;
3829 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3830 ctx->xin->flow.skb_priority = skb_priority;
3832 /* Couldn't translate queue to a priority. Nothing to do. A warning
3833 * has already been logged. */
3838 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3840 const struct xbridge *xbridge = xbridge_;
3851 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3854 port = get_ofp_port(xbridge, ofp_port);
3855 return port ? port->may_enable : false;
3860 xlate_bundle_action(struct xlate_ctx *ctx,
3861 const struct ofpact_bundle *bundle)
3865 port = bundle_execute(bundle, &ctx->xin->flow, ctx->wc, slave_enabled_cb,
3866 CONST_CAST(struct xbridge *, ctx->xbridge));
3867 if (bundle->dst.field) {
3868 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow, ctx->wc);
3870 xlate_output_action(ctx, port, 0, false);
3875 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3876 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3878 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3879 if (ctx->xin->may_learn) {
3880 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3885 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3887 learn_mask(learn, ctx->wc);
3889 if (ctx->xin->xcache) {
3890 struct xc_entry *entry;
3892 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3893 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3894 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3895 entry->u.learn.ofpacts = ofpbuf_new(64);
3896 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3897 entry->u.learn.ofpacts);
3898 } else if (ctx->xin->may_learn) {
3899 uint64_t ofpacts_stub[1024 / 8];
3900 struct ofputil_flow_mod fm;
3901 struct ofpbuf ofpacts;
3903 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3904 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3905 ofpbuf_uninit(&ofpacts);
3910 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3911 uint16_t idle_timeout, uint16_t hard_timeout)
3913 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3914 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3919 xlate_fin_timeout(struct xlate_ctx *ctx,
3920 const struct ofpact_fin_timeout *oft)
3923 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3924 oft->fin_idle_timeout, oft->fin_hard_timeout);
3925 if (ctx->xin->xcache) {
3926 struct xc_entry *entry;
3928 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3929 /* XC_RULE already holds a reference on the rule, none is taken
3931 entry->u.fin.rule = ctx->rule;
3932 entry->u.fin.idle = oft->fin_idle_timeout;
3933 entry->u.fin.hard = oft->fin_hard_timeout;
3939 xlate_sample_action(struct xlate_ctx *ctx,
3940 const struct ofpact_sample *os)
3942 /* Scale the probability from 16-bit to 32-bit while representing
3943 * the same percentage. */
3944 uint32_t probability = (os->probability << 16) | os->probability;
3946 if (!ctx->xbridge->support.variable_length_userdata) {
3947 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3949 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3950 "lacks support (needs Linux 3.10+ or kernel module from "
3955 xlate_commit_actions(ctx);
3957 union user_action_cookie cookie = {
3959 .type = USER_ACTION_COOKIE_FLOW_SAMPLE,
3960 .probability = os->probability,
3961 .collector_set_id = os->collector_set_id,
3962 .obs_domain_id = os->obs_domain_id,
3963 .obs_point_id = os->obs_point_id,
3966 compose_sample_action(ctx, probability, &cookie, sizeof cookie.flow_sample,
3971 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3973 if (xport->config & (is_stp(&ctx->xin->flow)
3974 ? OFPUTIL_PC_NO_RECV_STP
3975 : OFPUTIL_PC_NO_RECV)) {
3979 /* Only drop packets here if both forwarding and learning are
3980 * disabled. If just learning is enabled, we need to have
3981 * OFPP_NORMAL and the learning action have a look at the packet
3982 * before we can drop it. */
3983 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3984 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3992 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3994 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3995 size_t on_len = ofpact_nest_get_action_len(on);
3996 const struct ofpact *inner;
3998 /* Maintain actset_output depending on the contents of the action set:
4000 * - OFPP_UNSET, if there is no "output" action.
4002 * - The output port, if there is an "output" action and no "group"
4005 * - OFPP_UNSET, if there is a "group" action.
4007 if (!ctx->action_set_has_group) {
4008 OFPACT_FOR_EACH (inner, on->actions, on_len) {
4009 if (inner->type == OFPACT_OUTPUT) {
4010 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
4011 } else if (inner->type == OFPACT_GROUP) {
4012 ctx->xin->flow.actset_output = OFPP_UNSET;
4013 ctx->action_set_has_group = true;
4019 ofpbuf_put(&ctx->action_set, on->actions, on_len);
4020 ofpact_pad(&ctx->action_set);
4024 xlate_action_set(struct xlate_ctx *ctx)
4026 uint64_t action_list_stub[1024 / 64];
4027 struct ofpbuf action_list;
4029 ctx->in_action_set = true;
4030 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
4031 ofpacts_execute_action_set(&action_list, &ctx->action_set);
4032 /* Clear the action set, as it is not needed any more. */
4033 ofpbuf_clear(&ctx->action_set);
4034 do_xlate_actions(action_list.data, action_list.size, ctx);
4035 ctx->in_action_set = false;
4036 ofpbuf_uninit(&action_list);
4040 recirc_put_unroll_xlate(struct xlate_ctx *ctx)
4042 struct ofpact_unroll_xlate *unroll;
4044 unroll = ctx->last_unroll_offset < 0
4046 : ALIGNED_CAST(struct ofpact_unroll_xlate *,
4047 (char *)ctx->action_set.data + ctx->last_unroll_offset);
4049 /* Restore the table_id and rule cookie for a potential PACKET
4052 (ctx->table_id != unroll->rule_table_id
4053 || ctx->rule_cookie != unroll->rule_cookie)) {
4055 ctx->last_unroll_offset = ctx->action_set.size;
4056 unroll = ofpact_put_UNROLL_XLATE(&ctx->action_set);
4057 unroll->rule_table_id = ctx->table_id;
4058 unroll->rule_cookie = ctx->rule_cookie;
4063 /* Copy remaining actions to the action_set to be executed after recirculation.
4064 * UNROLL_XLATE action is inserted, if not already done so, before actions that
4065 * may generate PACKET_INs from the current table and without matching another
4068 recirc_unroll_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4069 struct xlate_ctx *ctx)
4071 const struct ofpact *a;
4073 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4075 /* May generate PACKET INs. */
4076 case OFPACT_OUTPUT_REG:
4079 case OFPACT_CONTROLLER:
4080 case OFPACT_DEC_MPLS_TTL:
4081 case OFPACT_DEC_TTL:
4082 recirc_put_unroll_xlate(ctx);
4085 /* These may not generate PACKET INs. */
4086 case OFPACT_SET_TUNNEL:
4087 case OFPACT_REG_MOVE:
4088 case OFPACT_SET_FIELD:
4089 case OFPACT_STACK_PUSH:
4090 case OFPACT_STACK_POP:
4092 case OFPACT_WRITE_METADATA:
4093 case OFPACT_RESUBMIT: /* May indirectly generate PACKET INs, */
4094 case OFPACT_GOTO_TABLE: /* but from a different table and rule. */
4095 case OFPACT_ENQUEUE:
4096 case OFPACT_SET_VLAN_VID:
4097 case OFPACT_SET_VLAN_PCP:
4098 case OFPACT_STRIP_VLAN:
4099 case OFPACT_PUSH_VLAN:
4100 case OFPACT_SET_ETH_SRC:
4101 case OFPACT_SET_ETH_DST:
4102 case OFPACT_SET_IPV4_SRC:
4103 case OFPACT_SET_IPV4_DST:
4104 case OFPACT_SET_IP_DSCP:
4105 case OFPACT_SET_IP_ECN:
4106 case OFPACT_SET_IP_TTL:
4107 case OFPACT_SET_L4_SRC_PORT:
4108 case OFPACT_SET_L4_DST_PORT:
4109 case OFPACT_SET_QUEUE:
4110 case OFPACT_POP_QUEUE:
4111 case OFPACT_PUSH_MPLS:
4112 case OFPACT_POP_MPLS:
4113 case OFPACT_SET_MPLS_LABEL:
4114 case OFPACT_SET_MPLS_TC:
4115 case OFPACT_SET_MPLS_TTL:
4116 case OFPACT_MULTIPATH:
4119 case OFPACT_UNROLL_XLATE:
4120 case OFPACT_FIN_TIMEOUT:
4121 case OFPACT_CLEAR_ACTIONS:
4122 case OFPACT_WRITE_ACTIONS:
4125 case OFPACT_DEBUG_RECIRC:
4129 /* These need not be copied for restoration. */
4131 case OFPACT_CONJUNCTION:
4134 /* Copy the action over. */
4135 ofpbuf_put(&ctx->action_set, a, OFPACT_ALIGN(a->len));
4139 #define CHECK_MPLS_RECIRCULATION() \
4140 if (ctx->was_mpls) { \
4141 ctx_trigger_recirculation(ctx); \
4144 #define CHECK_MPLS_RECIRCULATION_IF(COND) \
4146 CHECK_MPLS_RECIRCULATION(); \
4150 put_ct_mark(const struct flow *flow, struct flow *base_flow,
4151 struct ofpbuf *odp_actions, struct flow_wildcards *wc)
4158 odp_attr.key = flow->ct_mark;
4159 odp_attr.mask = wc->masks.ct_mark;
4161 if (odp_attr.mask && odp_attr.key != base_flow->ct_mark) {
4162 nl_msg_put_unspec(odp_actions, OVS_CT_ATTR_MARK, &odp_attr,
4168 compose_conntrack_action(struct xlate_ctx *ctx, struct ofpact_conntrack *ofc)
4170 uint32_t old_ct_mark = ctx->base_flow.ct_mark;
4174 /* Ensure that any prior actions are applied before composing the new
4175 * conntrack action. */
4176 xlate_commit_actions(ctx);
4178 /* Process nested actions first, to populate the key. */
4179 do_xlate_actions(ofc->actions, ofpact_ct_get_action_len(ofc), ctx);
4181 if (ofc->zone_src.field) {
4182 zone = mf_get_subfield(&ofc->zone_src, &ctx->xin->flow);
4184 zone = ofc->zone_imm;
4187 ct_offset = nl_msg_start_nested(ctx->odp_actions, OVS_ACTION_ATTR_CT);
4188 if (ofc->flags & NX_CT_F_COMMIT) {
4189 nl_msg_put_flag(ctx->odp_actions, OVS_CT_ATTR_COMMIT);
4191 nl_msg_put_u16(ctx->odp_actions, OVS_CT_ATTR_ZONE, zone);
4192 put_ct_mark(&ctx->xin->flow, &ctx->base_flow, ctx->odp_actions, ctx->wc);
4193 nl_msg_end_nested(ctx->odp_actions, ct_offset);
4195 /* Restore the original ct fields in the key. These should only be exposed
4196 * after recirculation to another table. */
4197 ctx->base_flow.ct_mark = old_ct_mark;
4199 if (ofc->recirc_table == NX_CT_RECIRC_NONE) {
4200 /* If we do not recirculate as part of this action, hide the results of
4201 * connection tracking from subsequent recirculations. */
4202 ctx->conntracked = false;
4204 /* Use ct_* fields from datapath during recirculation upcall. */
4205 ctx->conntracked = true;
4206 ctx_trigger_recirculation(ctx);
4207 compose_recirculate_action__(ctx, ofc->recirc_table);
4212 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4213 struct xlate_ctx *ctx)
4215 struct flow_wildcards *wc = ctx->wc;
4216 struct flow *flow = &ctx->xin->flow;
4217 const struct ofpact *a;
4219 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4220 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
4221 tnl_nd_snoop(flow, wc, ctx->xbridge->name);
4223 /* dl_type already in the mask, not set below. */
4225 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4226 struct ofpact_controller *controller;
4227 const struct ofpact_metadata *metadata;
4228 const struct ofpact_set_field *set_field;
4229 const struct mf_field *mf;
4232 /* Check if need to store the remaining actions for later
4234 if (exit_recirculates(ctx)) {
4235 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4237 (uint8_t *)ofpacts)),
4245 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4246 ofpact_get_OUTPUT(a)->max_len, true);
4250 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4251 /* Group could not be found. */
4256 case OFPACT_CONTROLLER:
4257 controller = ofpact_get_CONTROLLER(a);
4258 execute_controller_action(ctx, controller->max_len,
4260 controller->controller_id);
4263 case OFPACT_ENQUEUE:
4264 memset(&wc->masks.skb_priority, 0xff,
4265 sizeof wc->masks.skb_priority);
4266 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4269 case OFPACT_SET_VLAN_VID:
4270 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4271 if (flow->vlan_tci & htons(VLAN_CFI) ||
4272 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4273 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4274 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4279 case OFPACT_SET_VLAN_PCP:
4280 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4281 if (flow->vlan_tci & htons(VLAN_CFI) ||
4282 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4283 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4284 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4285 << VLAN_PCP_SHIFT) | VLAN_CFI);
4289 case OFPACT_STRIP_VLAN:
4290 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4291 flow->vlan_tci = htons(0);
4294 case OFPACT_PUSH_VLAN:
4295 /* XXX 802.1AD(QinQ) */
4296 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4297 flow->vlan_tci = htons(VLAN_CFI);
4300 case OFPACT_SET_ETH_SRC:
4301 WC_MASK_FIELD(wc, dl_src);
4302 flow->dl_src = ofpact_get_SET_ETH_SRC(a)->mac;
4305 case OFPACT_SET_ETH_DST:
4306 WC_MASK_FIELD(wc, dl_dst);
4307 flow->dl_dst = ofpact_get_SET_ETH_DST(a)->mac;
4310 case OFPACT_SET_IPV4_SRC:
4311 CHECK_MPLS_RECIRCULATION();
4312 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4313 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4314 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4318 case OFPACT_SET_IPV4_DST:
4319 CHECK_MPLS_RECIRCULATION();
4320 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4321 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4322 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4326 case OFPACT_SET_IP_DSCP:
4327 CHECK_MPLS_RECIRCULATION();
4328 if (is_ip_any(flow)) {
4329 wc->masks.nw_tos |= IP_DSCP_MASK;
4330 flow->nw_tos &= ~IP_DSCP_MASK;
4331 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4335 case OFPACT_SET_IP_ECN:
4336 CHECK_MPLS_RECIRCULATION();
4337 if (is_ip_any(flow)) {
4338 wc->masks.nw_tos |= IP_ECN_MASK;
4339 flow->nw_tos &= ~IP_ECN_MASK;
4340 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4344 case OFPACT_SET_IP_TTL:
4345 CHECK_MPLS_RECIRCULATION();
4346 if (is_ip_any(flow)) {
4347 wc->masks.nw_ttl = 0xff;
4348 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4352 case OFPACT_SET_L4_SRC_PORT:
4353 CHECK_MPLS_RECIRCULATION();
4354 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4355 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4356 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4357 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4361 case OFPACT_SET_L4_DST_PORT:
4362 CHECK_MPLS_RECIRCULATION();
4363 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4364 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4365 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4366 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4370 case OFPACT_RESUBMIT:
4371 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4374 case OFPACT_SET_TUNNEL:
4375 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4378 case OFPACT_SET_QUEUE:
4379 memset(&wc->masks.skb_priority, 0xff,
4380 sizeof wc->masks.skb_priority);
4381 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4384 case OFPACT_POP_QUEUE:
4385 memset(&wc->masks.skb_priority, 0xff,
4386 sizeof wc->masks.skb_priority);
4387 flow->skb_priority = ctx->orig_skb_priority;
4390 case OFPACT_REG_MOVE:
4391 CHECK_MPLS_RECIRCULATION_IF(
4392 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4393 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4394 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4397 case OFPACT_SET_FIELD:
4398 CHECK_MPLS_RECIRCULATION_IF(
4399 mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field));
4400 set_field = ofpact_get_SET_FIELD(a);
4401 mf = set_field->field;
4403 /* Set field action only ever overwrites packet's outermost
4404 * applicable header fields. Do nothing if no header exists. */
4405 if (mf->id == MFF_VLAN_VID) {
4406 wc->masks.vlan_tci |= htons(VLAN_CFI);
4407 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4410 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4411 /* 'dl_type' is already unwildcarded. */
4412 && !eth_type_mpls(flow->dl_type)) {
4415 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4416 * header field on a packet that does not have them. */
4417 mf_mask_field_and_prereqs(mf, wc);
4418 if (mf_are_prereqs_ok(mf, flow)) {
4419 mf_set_flow_value_masked(mf, &set_field->value,
4420 &set_field->mask, flow);
4424 case OFPACT_STACK_PUSH:
4425 CHECK_MPLS_RECIRCULATION_IF(
4426 mf_is_l3_or_higher(ofpact_get_STACK_PUSH(a)->subfield.field));
4427 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4431 case OFPACT_STACK_POP:
4432 CHECK_MPLS_RECIRCULATION_IF(
4433 mf_is_l3_or_higher(ofpact_get_STACK_POP(a)->subfield.field));
4434 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4438 case OFPACT_PUSH_MPLS:
4439 /* Recirculate if it is an IP packet with a zero ttl. This may
4440 * indicate that the packet was previously MPLS and an MPLS pop
4441 * action converted it to IP. In this case recirculating should
4442 * reveal the IP TTL which is used as the basis for a new MPLS
4444 CHECK_MPLS_RECIRCULATION_IF(
4445 !flow_count_mpls_labels(flow, wc)
4446 && flow->nw_ttl == 0
4447 && is_ip_any(flow));
4448 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4451 case OFPACT_POP_MPLS:
4452 CHECK_MPLS_RECIRCULATION();
4453 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4456 case OFPACT_SET_MPLS_LABEL:
4457 CHECK_MPLS_RECIRCULATION();
4458 compose_set_mpls_label_action(
4459 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4462 case OFPACT_SET_MPLS_TC:
4463 CHECK_MPLS_RECIRCULATION();
4464 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4467 case OFPACT_SET_MPLS_TTL:
4468 CHECK_MPLS_RECIRCULATION();
4469 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4472 case OFPACT_DEC_MPLS_TTL:
4473 CHECK_MPLS_RECIRCULATION();
4474 if (compose_dec_mpls_ttl_action(ctx)) {
4479 case OFPACT_DEC_TTL:
4480 CHECK_MPLS_RECIRCULATION();
4481 wc->masks.nw_ttl = 0xff;
4482 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4488 /* Nothing to do. */
4491 case OFPACT_MULTIPATH:
4492 CHECK_MPLS_RECIRCULATION();
4493 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4497 CHECK_MPLS_RECIRCULATION();
4498 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4501 case OFPACT_OUTPUT_REG:
4502 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4506 CHECK_MPLS_RECIRCULATION();
4507 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4510 case OFPACT_CONJUNCTION: {
4511 /* A flow with a "conjunction" action represents part of a special
4512 * kind of "set membership match". Such a flow should not actually
4513 * get executed, but it could via, say, a "packet-out", even though
4514 * that wouldn't be useful. Log it to help debugging. */
4515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4516 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4524 case OFPACT_UNROLL_XLATE: {
4525 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4527 /* Restore translation context data that was stored earlier. */
4528 ctx->table_id = unroll->rule_table_id;
4529 ctx->rule_cookie = unroll->rule_cookie;
4532 case OFPACT_FIN_TIMEOUT:
4533 CHECK_MPLS_RECIRCULATION();
4534 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4535 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4538 case OFPACT_CLEAR_ACTIONS:
4539 ofpbuf_clear(&ctx->action_set);
4540 ctx->xin->flow.actset_output = OFPP_UNSET;
4541 ctx->action_set_has_group = false;
4544 case OFPACT_WRITE_ACTIONS:
4545 xlate_write_actions(ctx, a);
4548 case OFPACT_WRITE_METADATA:
4549 metadata = ofpact_get_WRITE_METADATA(a);
4550 flow->metadata &= ~metadata->mask;
4551 flow->metadata |= metadata->metadata & metadata->mask;
4555 /* Not implemented yet. */
4558 case OFPACT_GOTO_TABLE: {
4559 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4561 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4562 * than ogt->table_id. This is to allow goto_table actions that
4563 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4564 * after recirculation. */
4565 ovs_assert(ctx->table_id == TBL_INTERNAL
4566 || ctx->table_id < ogt->table_id);
4567 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4568 ogt->table_id, true, true);
4573 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4577 CHECK_MPLS_RECIRCULATION();
4578 compose_conntrack_action(ctx, ofpact_get_CT(a));
4581 case OFPACT_DEBUG_RECIRC:
4582 ctx_trigger_recirculation(ctx);
4587 /* Check if need to store this and the remaining actions for later
4589 if (ctx->exit && ctx_first_recirculation_action(ctx)) {
4590 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4592 (uint8_t *)ofpacts)),
4600 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4601 const struct flow *flow, ofp_port_t in_port,
4602 struct rule_dpif *rule, uint16_t tcp_flags,
4603 const struct dp_packet *packet, struct flow_wildcards *wc,
4604 struct ofpbuf *odp_actions)
4606 xin->ofproto = ofproto;
4608 xin->flow.in_port.ofp_port = in_port;
4609 xin->flow.actset_output = OFPP_UNSET;
4610 xin->packet = packet;
4611 xin->may_learn = packet != NULL;
4614 xin->ofpacts = NULL;
4615 xin->ofpacts_len = 0;
4616 xin->tcp_flags = tcp_flags;
4617 xin->resubmit_hook = NULL;
4618 xin->report_hook = NULL;
4619 xin->resubmit_stats = NULL;
4623 xin->odp_actions = odp_actions;
4625 /* Do recirc lookup. */
4626 xin->recirc = flow->recirc_id
4627 ? recirc_id_node_find(flow->recirc_id)
4632 xlate_out_uninit(struct xlate_out *xout)
4635 xlate_out_free_recircs(xout);
4639 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4640 * into datapath actions, using 'ctx', and discards the datapath actions. */
4642 xlate_actions_for_side_effects(struct xlate_in *xin)
4644 struct xlate_out xout;
4646 xlate_actions(xin, &xout);
4647 xlate_out_uninit(&xout);
4650 static struct skb_priority_to_dscp *
4651 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4653 struct skb_priority_to_dscp *pdscp;
4656 hash = hash_int(skb_priority, 0);
4657 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4658 if (pdscp->skb_priority == skb_priority) {
4666 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4669 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4670 *dscp = pdscp ? pdscp->dscp : 0;
4671 return pdscp != NULL;
4675 count_skb_priorities(const struct xport *xport)
4677 return hmap_count(&xport->skb_priorities);
4681 clear_skb_priorities(struct xport *xport)
4683 struct skb_priority_to_dscp *pdscp, *next;
4685 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4686 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4692 actions_output_to_local_port(const struct xlate_ctx *ctx)
4694 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4695 const struct nlattr *a;
4698 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->odp_actions->data,
4699 ctx->odp_actions->size) {
4700 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4701 && nl_attr_get_odp_port(a) == local_odp_port) {
4708 #if defined(__linux__)
4709 /* Returns the maximum number of packets that the Linux kernel is willing to
4710 * queue up internally to certain kinds of software-implemented ports, or the
4711 * default (and rarely modified) value if it cannot be determined. */
4713 netdev_max_backlog(void)
4715 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4716 static int max_backlog = 1000; /* The normal default value. */
4718 if (ovsthread_once_start(&once)) {
4719 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4723 stream = fopen(filename, "r");
4725 VLOG_INFO("%s: open failed (%s)", filename, ovs_strerror(errno));
4727 if (fscanf(stream, "%d", &n) != 1) {
4728 VLOG_WARN("%s: read error", filename);
4729 } else if (n <= 100) {
4730 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4736 ovsthread_once_done(&once);
4738 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4744 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4747 count_output_actions(const struct ofpbuf *odp_actions)
4749 const struct nlattr *a;
4753 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4754 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4760 #endif /* defined(__linux__) */
4762 /* Returns true if 'odp_actions' contains more output actions than the datapath
4763 * can reliably handle in one go. On Linux, this is the value of the
4764 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4765 * packets that the kernel is willing to queue up for processing while the
4766 * datapath is processing a set of actions. */
4768 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4771 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4772 && count_output_actions(odp_actions) > netdev_max_backlog());
4774 /* OSes other than Linux might have similar limits, but we don't know how
4775 * to determine them.*/
4781 xlate_wc_init(struct xlate_ctx *ctx)
4783 flow_wildcards_init_catchall(ctx->wc);
4785 /* Some fields we consider to always be examined. */
4786 WC_MASK_FIELD(ctx->wc, in_port);
4787 WC_MASK_FIELD(ctx->wc, dl_type);
4788 if (is_ip_any(&ctx->xin->flow)) {
4789 WC_MASK_FIELD_MASK(ctx->wc, nw_frag, FLOW_NW_FRAG_MASK);
4792 if (ctx->xbridge->support.odp.recirc) {
4793 /* Always exactly match recirc_id when datapath supports
4795 WC_MASK_FIELD(ctx->wc, recirc_id);
4798 if (ctx->xbridge->netflow) {
4799 netflow_mask_wc(&ctx->xin->flow, ctx->wc);
4802 tnl_wc_init(&ctx->xin->flow, ctx->wc);
4806 xlate_wc_finish(struct xlate_ctx *ctx)
4808 /* Clear the metadata and register wildcard masks, because we won't
4809 * use non-header fields as part of the cache. */
4810 flow_wildcards_clear_non_packet_fields(ctx->wc);
4812 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
4813 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
4814 * represent these fields. The datapath interface, on the other hand,
4815 * represents them with just 8 bits each. This means that if the high
4816 * 8 bits of the masks for these fields somehow become set, then they
4817 * will get chopped off by a round trip through the datapath, and
4818 * revalidation will spot that as an inconsistency and delete the flow.
4819 * Avoid the problem here by making sure that only the low 8 bits of
4820 * either field can be unwildcarded for ICMP.
4822 if (is_icmpv4(&ctx->xin->flow) || is_icmpv6(&ctx->xin->flow)) {
4823 ctx->wc->masks.tp_src &= htons(UINT8_MAX);
4824 ctx->wc->masks.tp_dst &= htons(UINT8_MAX);
4826 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
4827 if (ctx->wc->masks.vlan_tci) {
4828 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI);
4832 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
4834 * The caller must take responsibility for eventually freeing 'xout', with
4835 * xlate_out_uninit(). */
4837 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4839 *xout = (struct xlate_out) {
4845 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4846 struct xbridge *xbridge = xbridge_lookup(xcfg, xin->ofproto);
4851 struct flow *flow = &xin->flow;
4853 union mf_subvalue stack_stub[1024 / sizeof(union mf_subvalue)];
4854 uint64_t action_set_stub[1024 / 8];
4855 struct flow_wildcards scratch_wc;
4856 uint64_t actions_stub[256 / 8];
4857 struct ofpbuf scratch_actions = OFPBUF_STUB_INITIALIZER(actions_stub);
4858 struct xlate_ctx ctx = {
4862 .orig_tunnel_ip_dst = flow->tunnel.ip_dst,
4864 .stack = OFPBUF_STUB_INITIALIZER(stack_stub),
4866 .wc = xin->wc ? xin->wc : &scratch_wc,
4867 .odp_actions = xin->odp_actions ? xin->odp_actions : &scratch_actions,
4869 .recurse = xin->recurse,
4870 .resubmits = xin->resubmits,
4872 .in_action_set = false,
4875 .rule_cookie = OVS_BE64_MAX,
4876 .orig_skb_priority = flow->skb_priority,
4877 .sflow_n_outputs = 0,
4878 .sflow_odp_port = 0,
4879 .nf_output_iface = NF_OUT_DROP,
4883 .recirc_action_offset = -1,
4884 .last_unroll_offset = -1,
4887 .conntracked = false,
4889 .action_set_has_group = false,
4890 .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
4893 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
4894 * the packet as the datapath will treat it for output actions:
4896 * - Our datapath doesn't retain tunneling information without us
4897 * re-setting it, so clear the tunnel data.
4899 * - For VLAN splinters, a higher layer may pretend that the packet
4900 * came in on 'flow->in_port.ofp_port' with 'flow->vlan_tci'
4901 * attached, because that's how we want to treat it from an OpenFlow
4902 * perspective. But from the datapath's perspective it actually came
4903 * in on a VLAN device without any VLAN attached. So here we put the
4904 * datapath's view of the VLAN information in 'base_flow' to ensure
4905 * correct treatment.
4907 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4908 if (flow->in_port.ofp_port
4909 != vsp_realdev_to_vlandev(xbridge->ofproto,
4910 flow->in_port.ofp_port,
4912 ctx.base_flow.vlan_tci = 0;
4915 ofpbuf_reserve(ctx.odp_actions, NL_A_U32_SIZE);
4917 xlate_wc_init(&ctx);
4920 COVERAGE_INC(xlate_actions);
4923 const struct recirc_state *state = &xin->recirc->state;
4925 xlate_report(&ctx, "Restoring state post-recirculation:");
4927 if (xin->ofpacts_len > 0 || ctx.rule) {
4928 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4929 const char *conflict = xin->ofpacts_len ? "actions" : "rule";
4931 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!", conflict);
4932 xlate_report(&ctx, "- Recirculation conflict (%s)!", conflict);
4936 /* Set the bridge for post-recirculation processing if needed. */
4937 if (ctx.xbridge->ofproto != state->ofproto) {
4938 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4939 const struct xbridge *new_bridge
4940 = xbridge_lookup(xcfg, state->ofproto);
4942 if (OVS_UNLIKELY(!new_bridge)) {
4943 /* Drop the packet if the bridge cannot be found. */
4944 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4945 VLOG_WARN_RL(&rl, "Recirculation bridge no longer exists.");
4946 xlate_report(&ctx, "- Recirculation bridge no longer exists.");
4949 ctx.xbridge = new_bridge;
4952 /* Set the post-recirculation table id. Note: A table lookup is done
4953 * only if there are no post-recirculation actions. */
4954 ctx.table_id = state->table_id;
4955 xlate_report(&ctx, "- Resuming from table %"PRIu8, ctx.table_id);
4957 if (!state->conntracked) {
4958 clear_conntrack(flow);
4961 /* Restore pipeline metadata. May change flow's in_port and other
4962 * metadata to the values that existed when recirculation was
4964 recirc_metadata_to_flow(&state->metadata, flow);
4966 /* Restore stack, if any. */
4968 ofpbuf_put(&ctx.stack, state->stack->data, state->stack->size);
4971 /* Restore mirror state. */
4972 ctx.mirrors = state->mirrors;
4974 /* Restore action set, if any. */
4975 if (state->action_set_len) {
4976 const struct ofpact *a;
4978 xlate_report_actions(&ctx, "- Restoring action set",
4979 state->ofpacts, state->action_set_len);
4981 ofpbuf_put(&ctx.action_set, state->ofpacts, state->action_set_len);
4983 OFPACT_FOR_EACH(a, state->ofpacts, state->action_set_len) {
4984 if (a->type == OFPACT_GROUP) {
4985 ctx.action_set_has_group = true;
4991 /* Restore recirculation actions. If there are no actions, processing
4992 * will start with a lookup in the table set above. */
4993 if (state->ofpacts_len > state->action_set_len) {
4994 xin->ofpacts_len = state->ofpacts_len - state->action_set_len;
4995 xin->ofpacts = state->ofpacts +
4996 state->action_set_len / sizeof *state->ofpacts;
4998 xlate_report_actions(&ctx, "- Restoring actions",
4999 xin->ofpacts, xin->ofpacts_len);
5001 } else if (OVS_UNLIKELY(flow->recirc_id)) {
5002 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
5004 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
5008 /* The bridge is now known so obtain its table version. */
5009 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
5011 if (!xin->ofpacts && !ctx.rule) {
5012 ctx.rule = rule_dpif_lookup_from_table(
5013 ctx.xbridge->ofproto, ctx.tables_version, flow, xin->wc,
5014 ctx.xin->resubmit_stats, &ctx.table_id,
5015 flow->in_port.ofp_port, true, true);
5016 if (ctx.xin->resubmit_stats) {
5017 rule_dpif_credit_stats(ctx.rule, ctx.xin->resubmit_stats);
5019 if (ctx.xin->xcache) {
5020 struct xc_entry *entry;
5022 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
5023 entry->u.rule = ctx.rule;
5024 rule_dpif_ref(ctx.rule);
5027 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
5028 ctx.xin->resubmit_hook(ctx.xin, ctx.rule, 0);
5031 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
5033 /* Get the proximate input port of the packet. (If xin->recirc,
5034 * flow->in_port is the ultimate input port of the packet.) */
5035 struct xport *in_port = get_ofp_port(xbridge,
5036 ctx.base_flow.in_port.ofp_port);
5038 /* Tunnel stats only for non-recirculated packets. */
5039 if (!xin->recirc && in_port && in_port->is_tunnel) {
5040 if (ctx.xin->resubmit_stats) {
5041 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
5043 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
5046 if (ctx.xin->xcache) {
5047 struct xc_entry *entry;
5049 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
5050 entry->u.dev.rx = netdev_ref(in_port->netdev);
5051 entry->u.dev.bfd = bfd_ref(in_port->bfd);
5055 if (!xin->recirc && process_special(&ctx, in_port)) {
5056 /* process_special() did all the processing for this packet.
5058 * We do not perform special processing on recirculated packets, as
5059 * recirculated packets are not really received by the bridge.*/
5060 } else if (in_port && in_port->xbundle
5061 && xbundle_mirror_out(xbridge, in_port->xbundle)) {
5062 if (ctx.xin->packet != NULL) {
5063 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5064 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5065 "%s, which is reserved exclusively for mirroring",
5066 ctx.xbridge->name, in_port->xbundle->name);
5069 /* Sampling is done only for packets really received by the bridge. */
5070 unsigned int user_cookie_offset = 0;
5072 user_cookie_offset = compose_sflow_action(&ctx);
5073 compose_ipfix_action(&ctx, ODPP_NONE);
5075 size_t sample_actions_len = ctx.odp_actions->size;
5077 if (tnl_process_ecn(flow)
5078 && (!in_port || may_receive(in_port, &ctx))) {
5079 const struct ofpact *ofpacts;
5083 ofpacts = xin->ofpacts;
5084 ofpacts_len = xin->ofpacts_len;
5085 } else if (ctx.rule) {
5086 const struct rule_actions *actions
5087 = rule_dpif_get_actions(ctx.rule);
5088 ofpacts = actions->ofpacts;
5089 ofpacts_len = actions->ofpacts_len;
5090 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
5095 mirror_ingress_packet(&ctx);
5096 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
5098 /* We've let OFPP_NORMAL and the learning action look at the
5099 * packet, so drop it now if forwarding is disabled. */
5100 if (in_port && (!xport_stp_forward_state(in_port) ||
5101 !xport_rstp_forward_state(in_port))) {
5102 /* Drop all actions added by do_xlate_actions() above. */
5103 ctx.odp_actions->size = sample_actions_len;
5105 /* Undo changes that may have been done for recirculation. */
5106 if (exit_recirculates(&ctx)) {
5107 ctx.action_set.size = ctx.recirc_action_offset;
5108 ctx.recirc_action_offset = -1;
5109 ctx.last_unroll_offset = -1;
5111 } else if (ctx.action_set.size) {
5112 /* Translate action set only if not dropping the packet and
5113 * not recirculating. */
5114 if (!exit_recirculates(&ctx)) {
5115 xlate_action_set(&ctx);
5118 /* Check if need to recirculate. */
5119 if (exit_recirculates(&ctx)) {
5120 compose_recirculate_action(&ctx);
5124 /* Output only fully processed packets. */
5125 if (!exit_recirculates(&ctx)
5126 && xbridge->has_in_band
5127 && in_band_must_output_to_local_port(flow)
5128 && !actions_output_to_local_port(&ctx)) {
5129 compose_output_action(&ctx, OFPP_LOCAL, NULL);
5132 if (user_cookie_offset) {
5133 fix_sflow_action(&ctx, user_cookie_offset);
5137 if (nl_attr_oversized(ctx.odp_actions->size)) {
5138 /* These datapath actions are too big for a Netlink attribute, so we
5139 * can't hand them to the kernel directly. dpif_execute() can execute
5140 * them one by one with help, so just mark the result as SLOW_ACTION to
5141 * prevent the flow from being installed. */
5142 COVERAGE_INC(xlate_actions_oversize);
5143 ctx.xout->slow |= SLOW_ACTION;
5144 } else if (too_many_output_actions(ctx.odp_actions)) {
5145 COVERAGE_INC(xlate_actions_too_many_output);
5146 ctx.xout->slow |= SLOW_ACTION;
5149 /* Do netflow only for packets really received by the bridge and not sent
5150 * to the controller. We consider packets sent to the controller to be
5151 * part of the control plane rather than the data plane. */
5152 if (!xin->recirc && xbridge->netflow && !(xout->slow & SLOW_CONTROLLER)) {
5153 if (ctx.xin->resubmit_stats) {
5154 netflow_flow_update(xbridge->netflow, flow,
5155 ctx.nf_output_iface,
5156 ctx.xin->resubmit_stats);
5158 if (ctx.xin->xcache) {
5159 struct xc_entry *entry;
5161 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5162 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5163 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5164 entry->u.nf.iface = ctx.nf_output_iface;
5169 xlate_wc_finish(&ctx);
5173 ofpbuf_uninit(&ctx.stack);
5174 ofpbuf_uninit(&ctx.action_set);
5175 ofpbuf_uninit(&scratch_actions);
5178 /* Sends 'packet' out 'ofport'.
5179 * May modify 'packet'.
5180 * Returns 0 if successful, otherwise a positive errno value. */
5182 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5184 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5185 struct xport *xport;
5186 struct ofpact_output output;
5189 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5190 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5191 flow_extract(packet, &flow);
5192 flow.in_port.ofp_port = OFPP_NONE;
5194 xport = xport_lookup(xcfg, ofport);
5198 output.port = xport->ofp_port;
5201 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5202 &output.ofpact, sizeof output,
5206 struct xlate_cache *
5207 xlate_cache_new(void)
5209 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5211 ofpbuf_init(&xcache->entries, 512);
5215 static struct xc_entry *
5216 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5218 struct xc_entry *entry;
5220 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5227 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5229 if (entry->u.dev.tx) {
5230 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5232 if (entry->u.dev.rx) {
5233 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5235 if (entry->u.dev.bfd) {
5236 bfd_account_rx(entry->u.dev.bfd, stats);
5241 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5243 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5244 struct xbridge *xbridge;
5245 struct xbundle *xbundle;
5246 struct flow_wildcards wc;
5248 xbridge = xbridge_lookup(xcfg, ofproto);
5253 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5259 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5262 /* Push stats and perform side effects of flow translation. */
5264 xlate_push_stats(struct xlate_cache *xcache,
5265 const struct dpif_flow_stats *stats)
5267 struct xc_entry *entry;
5268 struct ofpbuf entries = xcache->entries;
5269 struct eth_addr dmac;
5271 if (!stats->n_packets) {
5275 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5276 switch (entry->type) {
5278 rule_dpif_credit_stats(entry->u.rule, stats);
5281 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5282 entry->u.bond.vid, stats->n_bytes);
5285 xlate_cache_netdev(entry, stats);
5288 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5289 entry->u.nf.iface, stats);
5292 mirror_update_stats(entry->u.mirror.mbridge,
5293 entry->u.mirror.mirrors,
5294 stats->n_packets, stats->n_bytes);
5297 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5300 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5301 entry->u.normal.vlan);
5303 case XC_FIN_TIMEOUT:
5304 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5305 entry->u.fin.idle, entry->u.fin.hard);
5308 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5312 /* Lookup arp to avoid arp timeout. */
5313 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name,
5314 entry->u.tnl_arp_cache.d_ip, &dmac);
5323 xlate_dev_unref(struct xc_entry *entry)
5325 if (entry->u.dev.tx) {
5326 netdev_close(entry->u.dev.tx);
5328 if (entry->u.dev.rx) {
5329 netdev_close(entry->u.dev.rx);
5331 if (entry->u.dev.bfd) {
5332 bfd_unref(entry->u.dev.bfd);
5337 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5339 netflow_flow_clear(netflow, flow);
5340 netflow_unref(netflow);
5345 xlate_cache_clear(struct xlate_cache *xcache)
5347 struct xc_entry *entry;
5348 struct ofpbuf entries;
5354 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5355 switch (entry->type) {
5357 rule_dpif_unref(entry->u.rule);
5360 free(entry->u.bond.flow);
5361 bond_unref(entry->u.bond.bond);
5364 xlate_dev_unref(entry);
5367 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5370 mbridge_unref(entry->u.mirror.mbridge);
5373 free(entry->u.learn.fm);
5374 ofpbuf_delete(entry->u.learn.ofpacts);
5377 free(entry->u.normal.flow);
5379 case XC_FIN_TIMEOUT:
5380 /* 'u.fin.rule' is always already held as a XC_RULE, which
5381 * has already released it's reference above. */
5384 group_dpif_unref(entry->u.group.group);
5393 ofpbuf_clear(&xcache->entries);
5397 xlate_cache_delete(struct xlate_cache *xcache)
5399 xlate_cache_clear(xcache);
5400 ofpbuf_uninit(&xcache->entries);