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 long long 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". */
178 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
181 /* The rule that we are currently translating, or NULL. */
182 struct rule_dpif *rule;
184 /* Resubmit statistics, via xlate_table_action(). */
185 int recurse; /* Current resubmit nesting depth. */
186 int resubmits; /* Total number of resubmits. */
187 bool in_group; /* Currently translating ofgroup, if true. */
188 bool in_action_set; /* Currently translating action_set, if true. */
190 uint8_t table_id; /* OpenFlow table ID where flow was found. */
191 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
192 uint32_t orig_skb_priority; /* Priority when packet arrived. */
193 uint32_t sflow_n_outputs; /* Number of output ports. */
194 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
195 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
196 bool exit; /* No further actions should be processed. */
198 /* These are used for non-bond recirculation. The recirculation IDs are
199 * stored in xout and must be associated with a datapath flow (ukey),
200 * otherwise they will be freed when the xout is uninitialized.
203 * Steps in Recirculation Translation
204 * ==================================
206 * At some point during translation, the code recognizes the need for
207 * recirculation. For example, recirculation is necessary when, after
208 * popping the last MPLS label, an action or a match tries to examine or
209 * modify a field that has been newly revealed following the MPLS label.
211 * The simplest part of the work to be done is to commit existing changes to
212 * the packet, which produces datapath actions corresponding to the changes,
213 * and after this, add an OVS_ACTION_ATTR_RECIRC datapath action.
215 * The main problem here is preserving state. When the datapath executes
216 * OVS_ACTION_ATTR_RECIRC, it will upcall to userspace to get a translation
217 * for the post-recirculation actions. At this point userspace has to
218 * resume the translation where it left off, which means that it has to
219 * execute the following:
221 * - The action that prompted recirculation, and any actions following
222 * it within the same flow.
224 * - If the action that prompted recirculation was invoked within a
225 * NXAST_RESUBMIT, then any actions following the resubmit. These
226 * "resubmit"s can be nested, so this has to go all the way up the
229 * - The OpenFlow 1.1+ action set.
231 * State that actions and flow table lookups can depend on, such as the
232 * following, must also be preserved:
234 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
236 * - Action set, stack
238 * - The table ID and cookie of the flow being translated at each level
239 * of the control stack (since OFPAT_CONTROLLER actions send these to
242 * Translation allows for the control of this state preservation via these
243 * members. When a need for recirculation is identified, the translation
246 * 1. Sets 'recirc_action_offset' to the current size of 'action_set'. The
247 * action set is part of what needs to be preserved, so this allows the
248 * action set and the additional state to share the 'action_set' buffer.
249 * Later steps can tell that setup for recirculation is in progress from
250 * the nonnegative value of 'recirc_action_offset'.
252 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
253 * translation process.
255 * 3. Adds an OFPACT_UNROLL_XLATE action to 'action_set'. This action
256 * holds the current table ID and cookie so that they can be restored
257 * during a post-recirculation upcall translation.
259 * 4. Adds the action that prompted recirculation and any actions following
260 * it within the same flow to 'action_set', so that they can be executed
261 * during a post-recirculation upcall translation.
265 * 6. The action that prompted recirculation might be nested in a stack of
266 * nested "resubmit"s that have actions remaining. Each of these notices
267 * that we're exiting (from 'exit') and that recirculation setup is in
268 * progress (from 'recirc_action_offset') and responds by adding more
269 * OFPACT_UNROLL_XLATE actions to 'action_set', as necessary, and any
270 * actions that were yet unprocessed.
272 * The caller stores all the state produced by this process associated with
273 * the recirculation ID. For post-recirculation upcall translation, the
274 * caller passes it back in for the new translation to execute. The
275 * process yielded a set of ofpacts that can be translated directly, so it
276 * is not much of a special case at that point.
278 int recirc_action_offset; /* Offset in 'action_set' to actions to be
279 * executed after recirculation, or -1. */
280 int last_unroll_offset; /* Offset in 'action_set' to the latest unroll
283 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
284 * This is a trigger for recirculation in cases where translating an action
285 * or looking up a flow requires access to the fields of the packet after
286 * the MPLS label stack that was originally present. */
289 /* OpenFlow 1.1+ action set.
291 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
292 * When translation is otherwise complete, ofpacts_execute_action_set()
293 * converts it to a set of "struct ofpact"s that can be translated into
294 * datapath actions. */
295 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
296 struct ofpbuf action_set; /* Action set. */
297 uint64_t action_set_stub[1024 / 8];
300 static void xlate_action_set(struct xlate_ctx *ctx);
303 ctx_trigger_recirculation(struct xlate_ctx *ctx)
306 ctx->recirc_action_offset = ctx->action_set.size;
310 ctx_first_recirculation_action(const struct xlate_ctx *ctx)
312 return ctx->recirc_action_offset == ctx->action_set.size;
316 exit_recirculates(const struct xlate_ctx *ctx)
318 /* When recirculating the 'recirc_action_offset' has a non-negative value.
320 return ctx->recirc_action_offset >= 0;
323 static void compose_recirculate_action(struct xlate_ctx *ctx);
325 /* A controller may use OFPP_NONE as the ingress port to indicate that
326 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
327 * when an input bundle is needed for validation (e.g., mirroring or
328 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
329 * any 'port' structs, so care must be taken when dealing with it. */
330 static struct xbundle ofpp_none_bundle = {
332 .vlan_mode = PORT_VLAN_TRUNK
335 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
336 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
337 * traffic egressing the 'ofport' with that priority should be marked with. */
338 struct skb_priority_to_dscp {
339 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
340 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
342 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
358 /* xlate_cache entries hold enough information to perform the side effects of
359 * xlate_actions() for a rule, without needing to perform rule translation
360 * from scratch. The primary usage of these is to submit statistics to objects
361 * that a flow relates to, although they may be used for other effects as well
362 * (for instance, refreshing hard timeouts for learned flows). */
366 struct rule_dpif *rule;
373 struct netflow *netflow;
378 struct mbridge *mbridge;
379 mirror_mask_t mirrors;
387 struct ofproto_dpif *ofproto;
388 struct ofputil_flow_mod *fm;
389 struct ofpbuf *ofpacts;
392 struct ofproto_dpif *ofproto;
397 struct rule_dpif *rule;
402 struct group_dpif *group;
403 struct ofputil_bucket *bucket;
406 char br_name[IFNAMSIZ];
412 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
413 entries = xcache->entries; \
414 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
416 entry = ofpbuf_try_pull(&entries, sizeof *entry))
419 struct ofpbuf entries;
422 /* Xlate config contains hash maps of all bridges, bundles and ports.
423 * Xcfgp contains the pointer to the current xlate configuration.
424 * When the main thread needs to change the configuration, it copies xcfgp to
425 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
426 * does not block handler and revalidator threads. */
428 struct hmap xbridges;
429 struct hmap xbundles;
432 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
433 static struct xlate_cfg *new_xcfg = NULL;
435 static bool may_receive(const struct xport *, struct xlate_ctx *);
436 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
438 static void xlate_normal(struct xlate_ctx *);
439 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
440 OVS_PRINTF_FORMAT(2, 3);
441 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
442 uint8_t table_id, bool may_packet_in,
443 bool honor_table_miss);
444 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
445 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
446 static void output_normal(struct xlate_ctx *, const struct xbundle *,
449 /* Optional bond recirculation parameter to compose_output_action(). */
450 struct xlate_bond_recirc {
451 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
452 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
453 uint32_t hash_basis; /* Compute hash for recirc before. */
456 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
457 const struct xlate_bond_recirc *xr);
459 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
460 const struct ofproto_dpif *);
461 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
462 const struct ofbundle *);
463 static struct xport *xport_lookup(struct xlate_cfg *,
464 const struct ofport_dpif *);
465 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
466 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
467 uint32_t skb_priority);
468 static void clear_skb_priorities(struct xport *);
469 static size_t count_skb_priorities(const struct xport *);
470 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
473 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
475 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
476 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
477 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
478 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
479 const struct mac_learning *, struct stp *,
480 struct rstp *, const struct mcast_snooping *,
481 const struct mbridge *,
482 const struct dpif_sflow *,
483 const struct dpif_ipfix *,
484 const struct netflow *,
485 bool forward_bpdu, bool has_in_band,
486 const struct dpif_backer_support *);
487 static void xlate_xbundle_set(struct xbundle *xbundle,
488 enum port_vlan_mode vlan_mode, int vlan,
489 unsigned long *trunks, bool use_priority_tags,
490 const struct bond *bond, const struct lacp *lacp,
492 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
493 const struct netdev *netdev, const struct cfm *cfm,
494 const struct bfd *bfd, const struct lldp *lldp,
495 int stp_port_no, const struct rstp_port *rstp_port,
496 enum ofputil_port_config config,
497 enum ofputil_port_state state, bool is_tunnel,
499 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
500 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
501 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
502 static void xlate_xbridge_copy(struct xbridge *);
503 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
504 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
506 static void xlate_xcfg_free(struct xlate_cfg *);
509 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
511 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
514 va_start(args, format);
515 ctx->xin->report_hook(ctx->xin, ctx->recurse, format, args);
521 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
523 list_init(&xbridge->xbundles);
524 hmap_init(&xbridge->xports);
525 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
526 hash_pointer(xbridge->ofproto, 0));
530 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
532 list_init(&xbundle->xports);
533 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
534 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
535 hash_pointer(xbundle->ofbundle, 0));
539 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
541 hmap_init(&xport->skb_priorities);
542 hmap_insert(&xcfg->xports, &xport->hmap_node,
543 hash_pointer(xport->ofport, 0));
544 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
545 hash_ofp_port(xport->ofp_port));
549 xlate_xbridge_set(struct xbridge *xbridge,
551 const struct mac_learning *ml, struct stp *stp,
552 struct rstp *rstp, const struct mcast_snooping *ms,
553 const struct mbridge *mbridge,
554 const struct dpif_sflow *sflow,
555 const struct dpif_ipfix *ipfix,
556 const struct netflow *netflow,
557 bool forward_bpdu, bool has_in_band,
558 const struct dpif_backer_support *support)
560 if (xbridge->ml != ml) {
561 mac_learning_unref(xbridge->ml);
562 xbridge->ml = mac_learning_ref(ml);
565 if (xbridge->ms != ms) {
566 mcast_snooping_unref(xbridge->ms);
567 xbridge->ms = mcast_snooping_ref(ms);
570 if (xbridge->mbridge != mbridge) {
571 mbridge_unref(xbridge->mbridge);
572 xbridge->mbridge = mbridge_ref(mbridge);
575 if (xbridge->sflow != sflow) {
576 dpif_sflow_unref(xbridge->sflow);
577 xbridge->sflow = dpif_sflow_ref(sflow);
580 if (xbridge->ipfix != ipfix) {
581 dpif_ipfix_unref(xbridge->ipfix);
582 xbridge->ipfix = dpif_ipfix_ref(ipfix);
585 if (xbridge->stp != stp) {
586 stp_unref(xbridge->stp);
587 xbridge->stp = stp_ref(stp);
590 if (xbridge->rstp != rstp) {
591 rstp_unref(xbridge->rstp);
592 xbridge->rstp = rstp_ref(rstp);
595 if (xbridge->netflow != netflow) {
596 netflow_unref(xbridge->netflow);
597 xbridge->netflow = netflow_ref(netflow);
600 xbridge->dpif = dpif;
601 xbridge->forward_bpdu = forward_bpdu;
602 xbridge->has_in_band = has_in_band;
603 xbridge->support = *support;
607 xlate_xbundle_set(struct xbundle *xbundle,
608 enum port_vlan_mode vlan_mode, int vlan,
609 unsigned long *trunks, bool use_priority_tags,
610 const struct bond *bond, const struct lacp *lacp,
613 ovs_assert(xbundle->xbridge);
615 xbundle->vlan_mode = vlan_mode;
616 xbundle->vlan = vlan;
617 xbundle->trunks = trunks;
618 xbundle->use_priority_tags = use_priority_tags;
619 xbundle->floodable = floodable;
621 if (xbundle->bond != bond) {
622 bond_unref(xbundle->bond);
623 xbundle->bond = bond_ref(bond);
626 if (xbundle->lacp != lacp) {
627 lacp_unref(xbundle->lacp);
628 xbundle->lacp = lacp_ref(lacp);
633 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
634 const struct netdev *netdev, const struct cfm *cfm,
635 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
636 const struct rstp_port* rstp_port,
637 enum ofputil_port_config config, enum ofputil_port_state state,
638 bool is_tunnel, bool may_enable)
640 xport->config = config;
641 xport->state = state;
642 xport->stp_port_no = stp_port_no;
643 xport->is_tunnel = is_tunnel;
644 xport->may_enable = may_enable;
645 xport->odp_port = odp_port;
647 if (xport->rstp_port != rstp_port) {
648 rstp_port_unref(xport->rstp_port);
649 xport->rstp_port = rstp_port_ref(rstp_port);
652 if (xport->cfm != cfm) {
653 cfm_unref(xport->cfm);
654 xport->cfm = cfm_ref(cfm);
657 if (xport->bfd != bfd) {
658 bfd_unref(xport->bfd);
659 xport->bfd = bfd_ref(bfd);
662 if (xport->lldp != lldp) {
663 lldp_unref(xport->lldp);
664 xport->lldp = lldp_ref(lldp);
667 if (xport->netdev != netdev) {
668 netdev_close(xport->netdev);
669 xport->netdev = netdev_ref(netdev);
674 xlate_xbridge_copy(struct xbridge *xbridge)
676 struct xbundle *xbundle;
678 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
679 new_xbridge->ofproto = xbridge->ofproto;
680 new_xbridge->name = xstrdup(xbridge->name);
681 xlate_xbridge_init(new_xcfg, new_xbridge);
683 xlate_xbridge_set(new_xbridge,
684 xbridge->dpif, xbridge->ml, xbridge->stp,
685 xbridge->rstp, xbridge->ms, xbridge->mbridge,
686 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
687 xbridge->forward_bpdu, xbridge->has_in_band,
689 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
690 xlate_xbundle_copy(new_xbridge, xbundle);
693 /* Copy xports which are not part of a xbundle */
694 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
695 if (!xport->xbundle) {
696 xlate_xport_copy(new_xbridge, NULL, xport);
702 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
705 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
706 new_xbundle->ofbundle = xbundle->ofbundle;
707 new_xbundle->xbridge = xbridge;
708 new_xbundle->name = xstrdup(xbundle->name);
709 xlate_xbundle_init(new_xcfg, new_xbundle);
711 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
712 xbundle->vlan, xbundle->trunks,
713 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
715 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
716 xlate_xport_copy(xbridge, new_xbundle, xport);
721 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
724 struct skb_priority_to_dscp *pdscp, *new_pdscp;
725 struct xport *new_xport = xzalloc(sizeof *xport);
726 new_xport->ofport = xport->ofport;
727 new_xport->ofp_port = xport->ofp_port;
728 new_xport->xbridge = xbridge;
729 xlate_xport_init(new_xcfg, new_xport);
731 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
732 xport->bfd, xport->lldp, xport->stp_port_no,
733 xport->rstp_port, xport->config, xport->state,
734 xport->is_tunnel, xport->may_enable);
737 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
739 new_xport->peer = peer;
740 new_xport->peer->peer = new_xport;
745 new_xport->xbundle = xbundle;
746 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
749 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
750 new_pdscp = xmalloc(sizeof *pdscp);
751 new_pdscp->skb_priority = pdscp->skb_priority;
752 new_pdscp->dscp = pdscp->dscp;
753 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
754 hash_int(new_pdscp->skb_priority, 0));
758 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
759 * configuration in xcfgp.
761 * This needs to be called after editing the xlate configuration.
763 * Functions that edit the new xlate configuration are
764 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
770 * edit_xlate_configuration();
772 * xlate_txn_commit(); */
774 xlate_txn_commit(void)
776 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
778 ovsrcu_set(&xcfgp, new_xcfg);
779 ovsrcu_synchronize();
780 xlate_xcfg_free(xcfg);
784 /* Copies the current xlate configuration in xcfgp to new_xcfg.
786 * This needs to be called prior to editing the xlate configuration. */
788 xlate_txn_start(void)
790 struct xbridge *xbridge;
791 struct xlate_cfg *xcfg;
793 ovs_assert(!new_xcfg);
795 new_xcfg = xmalloc(sizeof *new_xcfg);
796 hmap_init(&new_xcfg->xbridges);
797 hmap_init(&new_xcfg->xbundles);
798 hmap_init(&new_xcfg->xports);
800 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
805 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
806 xlate_xbridge_copy(xbridge);
812 xlate_xcfg_free(struct xlate_cfg *xcfg)
814 struct xbridge *xbridge, *next_xbridge;
820 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
821 xlate_xbridge_remove(xcfg, xbridge);
824 hmap_destroy(&xcfg->xbridges);
825 hmap_destroy(&xcfg->xbundles);
826 hmap_destroy(&xcfg->xports);
831 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
833 const struct mac_learning *ml, struct stp *stp,
834 struct rstp *rstp, const struct mcast_snooping *ms,
835 const struct mbridge *mbridge,
836 const struct dpif_sflow *sflow,
837 const struct dpif_ipfix *ipfix,
838 const struct netflow *netflow,
839 bool forward_bpdu, bool has_in_band,
840 const struct dpif_backer_support *support)
842 struct xbridge *xbridge;
844 ovs_assert(new_xcfg);
846 xbridge = xbridge_lookup(new_xcfg, ofproto);
848 xbridge = xzalloc(sizeof *xbridge);
849 xbridge->ofproto = ofproto;
851 xlate_xbridge_init(new_xcfg, xbridge);
855 xbridge->name = xstrdup(name);
857 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
858 netflow, forward_bpdu, has_in_band, support);
862 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
864 struct xbundle *xbundle, *next_xbundle;
865 struct xport *xport, *next_xport;
871 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
872 xlate_xport_remove(xcfg, xport);
875 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
876 xlate_xbundle_remove(xcfg, xbundle);
879 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
880 mac_learning_unref(xbridge->ml);
881 mcast_snooping_unref(xbridge->ms);
882 mbridge_unref(xbridge->mbridge);
883 dpif_sflow_unref(xbridge->sflow);
884 dpif_ipfix_unref(xbridge->ipfix);
885 stp_unref(xbridge->stp);
886 rstp_unref(xbridge->rstp);
887 hmap_destroy(&xbridge->xports);
893 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
895 struct xbridge *xbridge;
897 ovs_assert(new_xcfg);
899 xbridge = xbridge_lookup(new_xcfg, ofproto);
900 xlate_xbridge_remove(new_xcfg, xbridge);
904 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
905 const char *name, enum port_vlan_mode vlan_mode, int vlan,
906 unsigned long *trunks, bool use_priority_tags,
907 const struct bond *bond, const struct lacp *lacp,
910 struct xbundle *xbundle;
912 ovs_assert(new_xcfg);
914 xbundle = xbundle_lookup(new_xcfg, ofbundle);
916 xbundle = xzalloc(sizeof *xbundle);
917 xbundle->ofbundle = ofbundle;
918 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
920 xlate_xbundle_init(new_xcfg, xbundle);
924 xbundle->name = xstrdup(name);
926 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
927 use_priority_tags, bond, lacp, floodable);
931 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
939 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
940 xport->xbundle = NULL;
943 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
944 list_remove(&xbundle->list_node);
945 bond_unref(xbundle->bond);
946 lacp_unref(xbundle->lacp);
952 xlate_bundle_remove(struct ofbundle *ofbundle)
954 struct xbundle *xbundle;
956 ovs_assert(new_xcfg);
958 xbundle = xbundle_lookup(new_xcfg, ofbundle);
959 xlate_xbundle_remove(new_xcfg, xbundle);
963 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
964 struct ofport_dpif *ofport, ofp_port_t ofp_port,
965 odp_port_t odp_port, const struct netdev *netdev,
966 const struct cfm *cfm, const struct bfd *bfd,
967 const struct lldp *lldp, struct ofport_dpif *peer,
968 int stp_port_no, const struct rstp_port *rstp_port,
969 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
970 enum ofputil_port_config config,
971 enum ofputil_port_state state, bool is_tunnel,
977 ovs_assert(new_xcfg);
979 xport = xport_lookup(new_xcfg, ofport);
981 xport = xzalloc(sizeof *xport);
982 xport->ofport = ofport;
983 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
984 xport->ofp_port = ofp_port;
986 xlate_xport_init(new_xcfg, xport);
989 ovs_assert(xport->ofp_port == ofp_port);
991 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
992 stp_port_no, rstp_port, config, state, is_tunnel,
996 xport->peer->peer = NULL;
998 xport->peer = xport_lookup(new_xcfg, peer);
1000 xport->peer->peer = xport;
1003 if (xport->xbundle) {
1004 list_remove(&xport->bundle_node);
1006 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1007 if (xport->xbundle) {
1008 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1011 clear_skb_priorities(xport);
1012 for (i = 0; i < n_qdscp; i++) {
1013 struct skb_priority_to_dscp *pdscp;
1014 uint32_t skb_priority;
1016 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1021 pdscp = xmalloc(sizeof *pdscp);
1022 pdscp->skb_priority = skb_priority;
1023 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1024 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1025 hash_int(pdscp->skb_priority, 0));
1030 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1037 xport->peer->peer = NULL;
1041 if (xport->xbundle) {
1042 list_remove(&xport->bundle_node);
1045 clear_skb_priorities(xport);
1046 hmap_destroy(&xport->skb_priorities);
1048 hmap_remove(&xcfg->xports, &xport->hmap_node);
1049 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1051 netdev_close(xport->netdev);
1052 rstp_port_unref(xport->rstp_port);
1053 cfm_unref(xport->cfm);
1054 bfd_unref(xport->bfd);
1055 lldp_unref(xport->lldp);
1060 xlate_ofport_remove(struct ofport_dpif *ofport)
1062 struct xport *xport;
1064 ovs_assert(new_xcfg);
1066 xport = xport_lookup(new_xcfg, ofport);
1067 xlate_xport_remove(new_xcfg, xport);
1070 static struct ofproto_dpif *
1071 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1072 ofp_port_t *ofp_in_port, const struct xport **xportp)
1074 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1075 const struct xport *xport;
1077 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1078 ? tnl_port_receive(flow)
1079 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1080 if (OVS_UNLIKELY(!xport)) {
1085 *ofp_in_port = xport->ofp_port;
1087 return xport->xbridge->ofproto;
1090 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1091 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1092 struct ofproto_dpif *
1093 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1094 ofp_port_t *ofp_in_port)
1096 const struct xport *xport;
1098 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1101 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1102 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1103 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1104 * handles for those protocols if they're enabled. Caller may use the returned
1105 * pointers until quiescing, for longer term use additional references must
1108 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1111 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1112 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1113 struct dpif_sflow **sflow, struct netflow **netflow,
1114 ofp_port_t *ofp_in_port)
1116 struct ofproto_dpif *ofproto;
1117 const struct xport *xport;
1119 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1126 *ofprotop = ofproto;
1130 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1134 *sflow = xport ? xport->xbridge->sflow : NULL;
1138 *netflow = xport ? xport->xbridge->netflow : NULL;
1144 static struct xbridge *
1145 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1147 struct hmap *xbridges;
1148 struct xbridge *xbridge;
1150 if (!ofproto || !xcfg) {
1154 xbridges = &xcfg->xbridges;
1156 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1158 if (xbridge->ofproto == ofproto) {
1165 static struct xbundle *
1166 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1168 struct hmap *xbundles;
1169 struct xbundle *xbundle;
1171 if (!ofbundle || !xcfg) {
1175 xbundles = &xcfg->xbundles;
1177 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1179 if (xbundle->ofbundle == ofbundle) {
1186 static struct xport *
1187 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1189 struct hmap *xports;
1190 struct xport *xport;
1192 if (!ofport || !xcfg) {
1196 xports = &xcfg->xports;
1198 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1200 if (xport->ofport == ofport) {
1207 static struct stp_port *
1208 xport_get_stp_port(const struct xport *xport)
1210 return xport->xbridge->stp && xport->stp_port_no != -1
1211 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1216 xport_stp_learn_state(const struct xport *xport)
1218 struct stp_port *sp = xport_get_stp_port(xport);
1220 ? stp_learn_in_state(stp_port_get_state(sp))
1225 xport_stp_forward_state(const struct xport *xport)
1227 struct stp_port *sp = xport_get_stp_port(xport);
1229 ? stp_forward_in_state(stp_port_get_state(sp))
1234 xport_stp_should_forward_bpdu(const struct xport *xport)
1236 struct stp_port *sp = xport_get_stp_port(xport);
1237 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1240 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1241 * were used to make the determination.*/
1243 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1245 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1246 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1247 return is_stp(flow);
1251 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1253 struct stp_port *sp = xport_get_stp_port(xport);
1254 struct dp_packet payload = *packet;
1255 struct eth_header *eth = dp_packet_data(&payload);
1257 /* Sink packets on ports that have STP disabled when the bridge has
1259 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1263 /* Trim off padding on payload. */
1264 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1265 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1268 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1269 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1273 static enum rstp_state
1274 xport_get_rstp_port_state(const struct xport *xport)
1276 return xport->rstp_port
1277 ? rstp_port_get_state(xport->rstp_port)
1282 xport_rstp_learn_state(const struct xport *xport)
1284 return xport->xbridge->rstp && xport->rstp_port
1285 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1290 xport_rstp_forward_state(const struct xport *xport)
1292 return xport->xbridge->rstp && xport->rstp_port
1293 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1298 xport_rstp_should_manage_bpdu(const struct xport *xport)
1300 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1304 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1306 struct dp_packet payload = *packet;
1307 struct eth_header *eth = dp_packet_data(&payload);
1309 /* Sink packets on ports that have no RSTP. */
1310 if (!xport->rstp_port) {
1314 /* Trim off padding on payload. */
1315 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1316 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1319 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1320 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1321 dp_packet_size(&payload));
1325 static struct xport *
1326 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1328 struct xport *xport;
1330 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1332 if (xport->ofp_port == ofp_port) {
1340 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1342 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1343 return xport ? xport->odp_port : ODPP_NONE;
1347 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1349 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1350 return xport && xport->may_enable;
1353 static struct ofputil_bucket *
1354 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1358 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1360 struct group_dpif *group;
1362 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1363 struct ofputil_bucket *bucket;
1365 bucket = group_first_live_bucket(ctx, group, depth);
1366 group_dpif_unref(group);
1367 return bucket == NULL;
1373 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1376 bucket_is_alive(const struct xlate_ctx *ctx,
1377 struct ofputil_bucket *bucket, int depth)
1379 if (depth >= MAX_LIVENESS_RECURSION) {
1380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1382 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1383 MAX_LIVENESS_RECURSION);
1387 return (!ofputil_bucket_has_liveness(bucket)
1388 || (bucket->watch_port != OFPP_ANY
1389 && odp_port_is_alive(ctx, bucket->watch_port))
1390 || (bucket->watch_group != OFPG_ANY
1391 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1394 static struct ofputil_bucket *
1395 group_first_live_bucket(const struct xlate_ctx *ctx,
1396 const struct group_dpif *group, int depth)
1398 struct ofputil_bucket *bucket;
1399 const struct ovs_list *buckets;
1401 group_dpif_get_buckets(group, &buckets);
1402 LIST_FOR_EACH (bucket, list_node, buckets) {
1403 if (bucket_is_alive(ctx, bucket, depth)) {
1411 static struct ofputil_bucket *
1412 group_best_live_bucket(const struct xlate_ctx *ctx,
1413 const struct group_dpif *group,
1416 struct ofputil_bucket *best_bucket = NULL;
1417 uint32_t best_score = 0;
1420 struct ofputil_bucket *bucket;
1421 const struct ovs_list *buckets;
1423 group_dpif_get_buckets(group, &buckets);
1424 LIST_FOR_EACH (bucket, list_node, buckets) {
1425 if (bucket_is_alive(ctx, bucket, 0)) {
1426 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1427 if (score >= best_score) {
1428 best_bucket = bucket;
1439 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1441 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1442 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1446 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1448 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1451 static mirror_mask_t
1452 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1454 return xbundle != &ofpp_none_bundle
1455 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1459 static mirror_mask_t
1460 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1462 return xbundle != &ofpp_none_bundle
1463 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1467 static mirror_mask_t
1468 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1470 return xbundle != &ofpp_none_bundle
1471 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1475 static struct xbundle *
1476 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1477 bool warn, struct xport **in_xportp)
1479 struct xport *xport;
1481 /* Find the port and bundle for the received packet. */
1482 xport = get_ofp_port(xbridge, in_port);
1486 if (xport && xport->xbundle) {
1487 return xport->xbundle;
1490 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1491 * which a controller may use as the ingress port for traffic that
1492 * it is sourcing. */
1493 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1494 return &ofpp_none_bundle;
1497 /* Odd. A few possible reasons here:
1499 * - We deleted a port but there are still a few packets queued up
1502 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1503 * we don't know about.
1505 * - The ofproto client didn't configure the port as part of a bundle.
1506 * This is particularly likely to happen if a packet was received on the
1507 * port after it was created, but before the client had a chance to
1508 * configure its bundle.
1511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1513 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1514 "port %"PRIu16, xbridge->name, in_port);
1520 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1522 const struct xbridge *xbridge = ctx->xbridge;
1523 mirror_mask_t mirrors;
1524 struct xbundle *in_xbundle;
1528 mirrors = ctx->xout->mirrors;
1529 ctx->xout->mirrors = 0;
1531 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1532 ctx->xin->packet != NULL, NULL);
1536 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1538 /* Drop frames on bundles reserved for mirroring. */
1539 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1540 if (ctx->xin->packet != NULL) {
1541 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1542 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1543 "%s, which is reserved exclusively for mirroring",
1544 ctx->xbridge->name, in_xbundle->name);
1546 ofpbuf_clear(ctx->xout->odp_actions);
1551 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1552 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1555 vlan = input_vid_to_vlan(in_xbundle, vid);
1561 /* Restore the original packet before adding the mirror actions. */
1562 ctx->xin->flow = *orig_flow;
1565 mirror_mask_t dup_mirrors;
1566 struct ofbundle *out;
1567 unsigned long *vlans;
1572 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1573 &vlans, &dup_mirrors, &out, &out_vlan);
1574 ovs_assert(has_mirror);
1577 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1579 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1582 if (!vlan_mirrored) {
1583 mirrors = zero_rightmost_1bit(mirrors);
1587 mirrors &= ~dup_mirrors;
1588 ctx->xout->mirrors |= dup_mirrors;
1590 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1591 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1593 output_normal(ctx, out_xbundle, vlan);
1595 } else if (vlan != out_vlan
1596 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1597 struct xbundle *xbundle;
1599 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1600 if (xbundle_includes_vlan(xbundle, out_vlan)
1601 && !xbundle_mirror_out(xbridge, xbundle)) {
1602 output_normal(ctx, xbundle, out_vlan);
1609 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1610 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1611 * the bundle on which the packet was received, returns the VLAN to which the
1614 * Both 'vid' and the return value are in the range 0...4095. */
1616 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1618 switch (in_xbundle->vlan_mode) {
1619 case PORT_VLAN_ACCESS:
1620 return in_xbundle->vlan;
1623 case PORT_VLAN_TRUNK:
1626 case PORT_VLAN_NATIVE_UNTAGGED:
1627 case PORT_VLAN_NATIVE_TAGGED:
1628 return vid ? vid : in_xbundle->vlan;
1635 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1636 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1639 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1640 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1643 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1645 /* Allow any VID on the OFPP_NONE port. */
1646 if (in_xbundle == &ofpp_none_bundle) {
1650 switch (in_xbundle->vlan_mode) {
1651 case PORT_VLAN_ACCESS:
1654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1655 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1656 "packet received on port %s configured as VLAN "
1657 "%"PRIu16" access port", vid, in_xbundle->name,
1664 case PORT_VLAN_NATIVE_UNTAGGED:
1665 case PORT_VLAN_NATIVE_TAGGED:
1667 /* Port must always carry its native VLAN. */
1671 case PORT_VLAN_TRUNK:
1672 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1674 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1675 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1676 "received on port %s not configured for trunking "
1677 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1689 /* Given 'vlan', the VLAN that a packet belongs to, and
1690 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1691 * that should be included in the 802.1Q header. (If the return value is 0,
1692 * then the 802.1Q header should only be included in the packet if there is a
1695 * Both 'vlan' and the return value are in the range 0...4095. */
1697 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1699 switch (out_xbundle->vlan_mode) {
1700 case PORT_VLAN_ACCESS:
1703 case PORT_VLAN_TRUNK:
1704 case PORT_VLAN_NATIVE_TAGGED:
1707 case PORT_VLAN_NATIVE_UNTAGGED:
1708 return vlan == out_xbundle->vlan ? 0 : vlan;
1716 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1719 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1721 ovs_be16 tci, old_tci;
1722 struct xport *xport;
1723 struct xlate_bond_recirc xr;
1724 bool use_recirc = false;
1726 vid = output_vlan_to_vid(out_xbundle, vlan);
1727 if (list_is_empty(&out_xbundle->xports)) {
1728 /* Partially configured bundle with no slaves. Drop the packet. */
1730 } else if (!out_xbundle->bond) {
1731 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1734 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1735 struct flow_wildcards *wc = &ctx->xout->wc;
1736 struct ofport_dpif *ofport;
1738 if (ctx->xbridge->support.recirc) {
1739 use_recirc = bond_may_recirc(
1740 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1743 /* Only TCP mode uses recirculation. */
1744 xr.hash_alg = OVS_HASH_ALG_L4;
1745 bond_update_post_recirc_rules(out_xbundle->bond, false);
1747 /* Recirculation does not require unmasking hash fields. */
1752 ofport = bond_choose_output_slave(out_xbundle->bond,
1753 &ctx->xin->flow, wc, vid);
1754 xport = xport_lookup(xcfg, ofport);
1757 /* No slaves enabled, so drop packet. */
1761 /* If use_recirc is set, the main thread will handle stats
1762 * accounting for this bond. */
1764 if (ctx->xin->resubmit_stats) {
1765 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1766 ctx->xin->resubmit_stats->n_bytes);
1768 if (ctx->xin->xcache) {
1769 struct xc_entry *entry;
1772 flow = &ctx->xin->flow;
1773 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1774 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1775 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1776 entry->u.bond.vid = vid;
1781 old_tci = *flow_tci;
1783 if (tci || out_xbundle->use_priority_tags) {
1784 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1786 tci |= htons(VLAN_CFI);
1791 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1792 *flow_tci = old_tci;
1795 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1796 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1797 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1799 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1801 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1805 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1806 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1810 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1811 if (flow->nw_proto == ARP_OP_REPLY) {
1813 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1814 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1815 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1817 return flow->nw_src == flow->nw_dst;
1823 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1824 * dropped. Returns true if they may be forwarded, false if they should be
1827 * 'in_port' must be the xport that corresponds to flow->in_port.
1828 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1830 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1831 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1832 * checked by input_vid_is_valid().
1834 * May also add tags to '*tags', although the current implementation only does
1835 * so in one special case.
1838 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1841 struct xbundle *in_xbundle = in_port->xbundle;
1842 const struct xbridge *xbridge = ctx->xbridge;
1843 struct flow *flow = &ctx->xin->flow;
1845 /* Drop frames for reserved multicast addresses
1846 * only if forward_bpdu option is absent. */
1847 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1848 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1852 if (in_xbundle->bond) {
1853 struct mac_entry *mac;
1855 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1861 xlate_report(ctx, "bonding refused admissibility, dropping");
1864 case BV_DROP_IF_MOVED:
1865 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1866 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1868 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1869 && (!is_gratuitous_arp(flow, &ctx->xout->wc)
1870 || mac_entry_is_grat_arp_locked(mac))) {
1871 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1872 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1876 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1884 /* Checks whether a MAC learning update is necessary for MAC learning table
1885 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1888 * Most packets processed through the MAC learning table do not actually
1889 * change it in any way. This function requires only a read lock on the MAC
1890 * learning table, so it is much cheaper in this common case.
1892 * Keep the code here synchronized with that in update_learning_table__()
1895 is_mac_learning_update_needed(const struct mac_learning *ml,
1896 const struct flow *flow,
1897 struct flow_wildcards *wc,
1898 int vlan, struct xbundle *in_xbundle)
1899 OVS_REQ_RDLOCK(ml->rwlock)
1901 struct mac_entry *mac;
1903 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1907 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1908 if (!mac || mac_entry_age(ml, mac)) {
1912 if (is_gratuitous_arp(flow, wc)) {
1913 /* We don't want to learn from gratuitous ARP packets that are
1914 * reflected back over bond slaves so we lock the learning table. */
1915 if (!in_xbundle->bond) {
1917 } else if (mac_entry_is_grat_arp_locked(mac)) {
1922 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1926 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1927 * received on 'in_xbundle' in 'vlan'.
1929 * This code repeats all the checks in is_mac_learning_update_needed() because
1930 * the lock was released between there and here and thus the MAC learning state
1931 * could have changed.
1933 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1936 update_learning_table__(const struct xbridge *xbridge,
1937 const struct flow *flow, struct flow_wildcards *wc,
1938 int vlan, struct xbundle *in_xbundle)
1939 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1941 struct mac_entry *mac;
1943 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1947 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1948 if (is_gratuitous_arp(flow, wc)) {
1949 /* We don't want to learn from gratuitous ARP packets that are
1950 * reflected back over bond slaves so we lock the learning table. */
1951 if (!in_xbundle->bond) {
1952 mac_entry_set_grat_arp_lock(mac);
1953 } else if (mac_entry_is_grat_arp_locked(mac)) {
1958 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1959 /* The log messages here could actually be useful in debugging,
1960 * so keep the rate limit relatively high. */
1961 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1963 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1964 "on port %s in VLAN %d",
1965 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1966 in_xbundle->name, vlan);
1968 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
1973 update_learning_table(const struct xbridge *xbridge,
1974 const struct flow *flow, struct flow_wildcards *wc,
1975 int vlan, struct xbundle *in_xbundle)
1979 /* Don't learn the OFPP_NONE port. */
1980 if (in_xbundle == &ofpp_none_bundle) {
1984 /* First try the common case: no change to MAC learning table. */
1985 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1986 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1988 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1991 /* Slow path: MAC learning table might need an update. */
1992 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1993 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1994 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1998 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1999 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2001 update_mcast_snooping_table__(const struct xbridge *xbridge,
2002 const struct flow *flow,
2003 struct mcast_snooping *ms,
2004 ovs_be32 ip4, int vlan,
2005 struct xbundle *in_xbundle)
2006 OVS_REQ_WRLOCK(ms->rwlock)
2008 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2010 switch (ntohs(flow->tp_src)) {
2011 case IGMP_HOST_MEMBERSHIP_REPORT:
2012 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2013 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
2014 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2015 IP_FMT" is on port %s in VLAN %d",
2016 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2019 case IGMP_HOST_LEAVE_MESSAGE:
2020 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
2021 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2022 IP_FMT" is on port %s in VLAN %d",
2023 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2026 case IGMP_HOST_MEMBERSHIP_QUERY:
2027 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2028 in_xbundle->ofbundle)) {
2029 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2030 IP_FMT" is on port %s in VLAN %d",
2031 xbridge->name, IP_ARGS(flow->nw_src),
2032 in_xbundle->name, vlan);
2038 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2039 * was received on 'in_xbundle' in 'vlan'. */
2041 update_mcast_snooping_table(const struct xbridge *xbridge,
2042 const struct flow *flow, int vlan,
2043 struct xbundle *in_xbundle)
2045 struct mcast_snooping *ms = xbridge->ms;
2046 struct xlate_cfg *xcfg;
2047 struct xbundle *mcast_xbundle;
2048 struct mcast_port_bundle *fport;
2050 /* Don't learn the OFPP_NONE port. */
2051 if (in_xbundle == &ofpp_none_bundle) {
2055 /* Don't learn from flood ports */
2056 mcast_xbundle = NULL;
2057 ovs_rwlock_wrlock(&ms->rwlock);
2058 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2059 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2060 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2061 if (mcast_xbundle == in_xbundle) {
2066 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2067 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
2070 ovs_rwlock_unlock(&ms->rwlock);
2073 /* send the packet to ports having the multicast group learned */
2075 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2076 struct mcast_snooping *ms OVS_UNUSED,
2077 struct mcast_group *grp,
2078 struct xbundle *in_xbundle, uint16_t vlan)
2079 OVS_REQ_RDLOCK(ms->rwlock)
2081 struct xlate_cfg *xcfg;
2082 struct mcast_group_bundle *b;
2083 struct xbundle *mcast_xbundle;
2085 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2086 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2087 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2088 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2089 xlate_report(ctx, "forwarding to mcast group port");
2090 output_normal(ctx, mcast_xbundle, vlan);
2091 } else if (!mcast_xbundle) {
2092 xlate_report(ctx, "mcast group port is unknown, dropping");
2094 xlate_report(ctx, "mcast group port is input port, dropping");
2099 /* send the packet to ports connected to multicast routers */
2101 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2102 struct mcast_snooping *ms,
2103 struct xbundle *in_xbundle, uint16_t vlan)
2104 OVS_REQ_RDLOCK(ms->rwlock)
2106 struct xlate_cfg *xcfg;
2107 struct mcast_mrouter_bundle *mrouter;
2108 struct xbundle *mcast_xbundle;
2110 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2111 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2112 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2113 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2114 xlate_report(ctx, "forwarding to mcast router port");
2115 output_normal(ctx, mcast_xbundle, vlan);
2116 } else if (!mcast_xbundle) {
2117 xlate_report(ctx, "mcast router port is unknown, dropping");
2119 xlate_report(ctx, "mcast router port is input port, dropping");
2124 /* send the packet to ports flagged to be flooded */
2126 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2127 struct mcast_snooping *ms,
2128 struct xbundle *in_xbundle, uint16_t vlan)
2129 OVS_REQ_RDLOCK(ms->rwlock)
2131 struct xlate_cfg *xcfg;
2132 struct mcast_port_bundle *fport;
2133 struct xbundle *mcast_xbundle;
2135 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2136 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2137 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2138 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2139 xlate_report(ctx, "forwarding to mcast flood port");
2140 output_normal(ctx, mcast_xbundle, vlan);
2141 } else if (!mcast_xbundle) {
2142 xlate_report(ctx, "mcast flood port is unknown, dropping");
2144 xlate_report(ctx, "mcast flood port is input port, dropping");
2149 /* forward the Reports to configured ports */
2151 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2152 struct mcast_snooping *ms,
2153 struct xbundle *in_xbundle, uint16_t vlan)
2154 OVS_REQ_RDLOCK(ms->rwlock)
2156 struct xlate_cfg *xcfg;
2157 struct mcast_port_bundle *rport;
2158 struct xbundle *mcast_xbundle;
2160 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2161 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2162 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2163 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2164 xlate_report(ctx, "forwarding Report to mcast flagged port");
2165 output_normal(ctx, mcast_xbundle, vlan);
2166 } else if (!mcast_xbundle) {
2167 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2169 xlate_report(ctx, "mcast port is input port, dropping the Report");
2175 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2178 struct xbundle *xbundle;
2180 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2181 if (xbundle != in_xbundle
2182 && xbundle_includes_vlan(xbundle, vlan)
2183 && xbundle->floodable
2184 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2185 output_normal(ctx, xbundle, vlan);
2188 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2192 xlate_normal(struct xlate_ctx *ctx)
2194 struct flow_wildcards *wc = &ctx->xout->wc;
2195 struct flow *flow = &ctx->xin->flow;
2196 struct xbundle *in_xbundle;
2197 struct xport *in_port;
2198 struct mac_entry *mac;
2203 ctx->xout->has_normal = true;
2205 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2206 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2207 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2209 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2210 ctx->xin->packet != NULL, &in_port);
2212 xlate_report(ctx, "no input bundle, dropping");
2216 /* Drop malformed frames. */
2217 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2218 !(flow->vlan_tci & htons(VLAN_CFI))) {
2219 if (ctx->xin->packet != NULL) {
2220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2221 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2222 "VLAN tag received on port %s",
2223 ctx->xbridge->name, in_xbundle->name);
2225 xlate_report(ctx, "partial VLAN tag, dropping");
2229 /* Drop frames on bundles reserved for mirroring. */
2230 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2231 if (ctx->xin->packet != NULL) {
2232 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2233 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2234 "%s, which is reserved exclusively for mirroring",
2235 ctx->xbridge->name, in_xbundle->name);
2237 xlate_report(ctx, "input port is mirror output port, dropping");
2242 vid = vlan_tci_to_vid(flow->vlan_tci);
2243 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2244 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2247 vlan = input_vid_to_vlan(in_xbundle, vid);
2249 /* Check other admissibility requirements. */
2250 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2254 /* Learn source MAC. */
2255 if (ctx->xin->may_learn) {
2256 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2258 if (ctx->xin->xcache) {
2259 struct xc_entry *entry;
2261 /* Save enough info to update mac learning table later. */
2262 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2263 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2264 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2265 entry->u.normal.vlan = vlan;
2268 /* Determine output bundle. */
2269 if (mcast_snooping_enabled(ctx->xbridge->ms)
2270 && !eth_addr_is_broadcast(flow->dl_dst)
2271 && eth_addr_is_multicast(flow->dl_dst)
2272 && flow->dl_type == htons(ETH_TYPE_IP)) {
2273 struct mcast_snooping *ms = ctx->xbridge->ms;
2274 struct mcast_group *grp;
2276 if (flow->nw_proto == IPPROTO_IGMP) {
2277 if (ctx->xin->may_learn) {
2278 if (mcast_snooping_is_membership(flow->tp_src) ||
2279 mcast_snooping_is_query(flow->tp_src)) {
2280 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2285 if (mcast_snooping_is_membership(flow->tp_src)) {
2286 ovs_rwlock_rdlock(&ms->rwlock);
2287 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2288 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2289 * forward IGMP Membership Reports only to those ports where
2290 * multicast routers are attached. Alternatively stated: a
2291 * snooping switch should not forward IGMP Membership Reports
2292 * to ports on which only hosts are attached.
2293 * An administrative control may be provided to override this
2294 * restriction, allowing the report messages to be flooded to
2296 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2297 ovs_rwlock_unlock(&ms->rwlock);
2299 xlate_report(ctx, "multicast traffic, flooding");
2300 xlate_normal_flood(ctx, in_xbundle, vlan);
2304 if (ip_is_local_multicast(flow->nw_dst)) {
2305 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2306 * address in the 224.0.0.x range which are not IGMP must
2307 * be forwarded on all ports */
2308 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2309 xlate_normal_flood(ctx, in_xbundle, vlan);
2314 /* forwarding to group base ports */
2315 ovs_rwlock_rdlock(&ms->rwlock);
2316 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2318 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2319 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2320 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2322 if (mcast_snooping_flood_unreg(ms)) {
2323 xlate_report(ctx, "unregistered multicast, flooding");
2324 xlate_normal_flood(ctx, in_xbundle, vlan);
2326 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2327 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2330 ovs_rwlock_unlock(&ms->rwlock);
2332 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2333 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2334 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2335 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2338 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2339 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2340 if (mac_xbundle && mac_xbundle != in_xbundle) {
2341 xlate_report(ctx, "forwarding to learned port");
2342 output_normal(ctx, mac_xbundle, vlan);
2343 } else if (!mac_xbundle) {
2344 xlate_report(ctx, "learned port is unknown, dropping");
2346 xlate_report(ctx, "learned port is input port, dropping");
2349 xlate_report(ctx, "no learned MAC for destination, flooding");
2350 xlate_normal_flood(ctx, in_xbundle, vlan);
2355 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2356 * the number of packets out of UINT32_MAX to sample. The given
2357 * cookie is passed back in the callback for each sampled packet.
2360 compose_sample_action(const struct xbridge *xbridge,
2361 struct ofpbuf *odp_actions,
2362 const struct flow *flow,
2363 const uint32_t probability,
2364 const union user_action_cookie *cookie,
2365 const size_t cookie_size,
2366 const odp_port_t tunnel_out_port)
2368 size_t sample_offset, actions_offset;
2369 odp_port_t odp_port;
2373 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2375 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2377 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2379 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2380 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2381 flow_hash_5tuple(flow, 0));
2382 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2383 tunnel_out_port, odp_actions);
2385 nl_msg_end_nested(odp_actions, actions_offset);
2386 nl_msg_end_nested(odp_actions, sample_offset);
2387 return cookie_offset;
2391 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2392 odp_port_t odp_port, unsigned int n_outputs,
2393 union user_action_cookie *cookie)
2397 cookie->type = USER_ACTION_COOKIE_SFLOW;
2398 cookie->sflow.vlan_tci = vlan_tci;
2400 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2401 * port information") for the interpretation of cookie->output. */
2402 switch (n_outputs) {
2404 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2405 cookie->sflow.output = 0x40000000 | 256;
2409 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2411 cookie->sflow.output = ifindex;
2416 /* 0x80000000 means "multiple output ports. */
2417 cookie->sflow.output = 0x80000000 | n_outputs;
2422 /* Compose SAMPLE action for sFlow bridge sampling. */
2424 compose_sflow_action(const struct xbridge *xbridge,
2425 struct ofpbuf *odp_actions,
2426 const struct flow *flow,
2427 odp_port_t odp_port)
2429 uint32_t probability;
2430 union user_action_cookie cookie;
2432 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2436 probability = dpif_sflow_get_probability(xbridge->sflow);
2437 compose_sflow_cookie(xbridge, htons(0), odp_port,
2438 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2440 return compose_sample_action(xbridge, odp_actions, flow, probability,
2441 &cookie, sizeof cookie.sflow, ODPP_NONE);
2445 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2446 uint32_t obs_domain_id, uint32_t obs_point_id,
2447 union user_action_cookie *cookie)
2449 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2450 cookie->flow_sample.probability = probability;
2451 cookie->flow_sample.collector_set_id = collector_set_id;
2452 cookie->flow_sample.obs_domain_id = obs_domain_id;
2453 cookie->flow_sample.obs_point_id = obs_point_id;
2457 compose_ipfix_cookie(union user_action_cookie *cookie,
2458 odp_port_t output_odp_port)
2460 cookie->type = USER_ACTION_COOKIE_IPFIX;
2461 cookie->ipfix.output_odp_port = output_odp_port;
2464 /* Compose SAMPLE action for IPFIX bridge sampling. */
2466 compose_ipfix_action(const struct xbridge *xbridge,
2467 struct ofpbuf *odp_actions,
2468 const struct flow *flow,
2469 odp_port_t output_odp_port)
2471 uint32_t probability;
2472 union user_action_cookie cookie;
2473 odp_port_t tunnel_out_port = ODPP_NONE;
2475 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2479 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2481 if (output_odp_port == ODPP_NONE &&
2482 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2486 /* For output case, output_odp_port is valid*/
2487 if (output_odp_port != ODPP_NONE) {
2488 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2491 /* If tunnel sampling is enabled, put an additional option attribute:
2492 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2494 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2495 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2496 tunnel_out_port = output_odp_port;
2500 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2501 compose_ipfix_cookie(&cookie, output_odp_port);
2503 compose_sample_action(xbridge, odp_actions, flow, probability,
2504 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2507 /* SAMPLE action for sFlow must be first action in any given list of
2508 * actions. At this point we do not have all information required to
2509 * build it. So try to build sample action as complete as possible. */
2511 add_sflow_action(struct xlate_ctx *ctx)
2513 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2514 ctx->xout->odp_actions,
2515 &ctx->xin->flow, ODPP_NONE);
2516 ctx->sflow_odp_port = 0;
2517 ctx->sflow_n_outputs = 0;
2520 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2521 * of actions, eventually after the SAMPLE action for sFlow. */
2523 add_ipfix_action(struct xlate_ctx *ctx)
2525 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2526 &ctx->xin->flow, ODPP_NONE);
2530 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2532 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2533 &ctx->xin->flow, port);
2536 /* Fix SAMPLE action according to data collected while composing ODP actions.
2537 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2538 * USERSPACE action's user-cookie which is required for sflow. */
2540 fix_sflow_action(struct xlate_ctx *ctx)
2542 const struct flow *base = &ctx->base_flow;
2543 union user_action_cookie *cookie;
2545 if (!ctx->user_cookie_offset) {
2549 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2550 sizeof cookie->sflow);
2551 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2553 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2554 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2557 static enum slow_path_reason
2558 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2559 const struct xport *xport, const struct dp_packet *packet)
2561 struct flow_wildcards *wc = &ctx->xout->wc;
2562 const struct xbridge *xbridge = ctx->xbridge;
2566 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2568 cfm_process_heartbeat(xport->cfm, packet);
2571 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2573 bfd_process_packet(xport->bfd, flow, packet);
2574 /* If POLL received, immediately sends FINAL back. */
2575 if (bfd_should_send_packet(xport->bfd)) {
2576 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2580 } else if (xport->xbundle && xport->xbundle->lacp
2581 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2583 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2586 } else if ((xbridge->stp || xbridge->rstp) &&
2587 stp_should_process_flow(flow, wc)) {
2590 ? stp_process_packet(xport, packet)
2591 : rstp_process_packet(xport, packet);
2594 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2596 lldp_process_packet(xport->lldp, packet);
2605 tnl_route_lookup_flow(const struct flow *oflow,
2606 ovs_be32 *ip, struct xport **out_port)
2608 char out_dev[IFNAMSIZ];
2609 struct xbridge *xbridge;
2610 struct xlate_cfg *xcfg;
2613 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2620 *ip = oflow->tunnel.ip_dst;
2623 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2626 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2627 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2630 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2631 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2642 xlate_flood_packet(struct xbridge *xbridge, struct dp_packet *packet)
2644 struct ofpact_output output;
2647 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2648 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2649 flow_extract(packet, &flow);
2650 flow.in_port.ofp_port = OFPP_NONE;
2651 output.port = OFPP_FLOOD;
2654 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2655 &output.ofpact, sizeof output,
2660 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2661 ovs_be32 ip_src, ovs_be32 ip_dst)
2663 struct xbridge *xbridge = out_dev->xbridge;
2664 struct dp_packet packet;
2666 dp_packet_init(&packet, 0);
2667 compose_arp(&packet, ARP_OP_REQUEST,
2668 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2670 xlate_flood_packet(xbridge, &packet);
2671 dp_packet_uninit(&packet);
2675 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2676 const struct flow *flow, odp_port_t tunnel_odp_port)
2678 struct ovs_action_push_tnl tnl_push_data;
2679 struct xport *out_dev = NULL;
2680 ovs_be32 s_ip, d_ip = 0;
2681 uint8_t smac[ETH_ADDR_LEN];
2682 uint8_t dmac[ETH_ADDR_LEN];
2685 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2687 xlate_report(ctx, "native tunnel routing failed");
2690 xlate_report(ctx, "tunneling to "IP_FMT" via %s",
2691 IP_ARGS(d_ip), netdev_get_name(out_dev->netdev));
2693 /* Use mac addr of bridge port of the peer. */
2694 err = netdev_get_etheraddr(out_dev->netdev, smac);
2696 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2700 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2702 xlate_report(ctx, "tunnel output device lacks IPv4 address");
2706 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2708 xlate_report(ctx, "ARP cache miss for "IP_FMT" on bridge %s, "
2709 "sending ARP request",
2710 IP_ARGS(d_ip), out_dev->xbridge->name);
2711 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2714 if (ctx->xin->xcache) {
2715 struct xc_entry *entry;
2717 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2718 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2719 sizeof entry->u.tnl_arp_cache.br_name);
2720 entry->u.tnl_arp_cache.d_ip = d_ip;
2723 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" "IP_FMT
2724 " to "ETH_ADDR_FMT" "IP_FMT,
2725 ETH_ADDR_ARGS(smac), IP_ARGS(s_ip),
2726 ETH_ADDR_ARGS(dmac), IP_ARGS(d_ip));
2727 err = tnl_port_build_header(xport->ofport, flow,
2728 dmac, smac, s_ip, &tnl_push_data);
2732 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2733 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2734 odp_put_tnl_push_action(ctx->xout->odp_actions, &tnl_push_data);
2739 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2740 const struct xlate_bond_recirc *xr, bool check_stp)
2742 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2743 struct flow_wildcards *wc = &ctx->xout->wc;
2744 struct flow *flow = &ctx->xin->flow;
2745 struct flow_tnl flow_tnl;
2746 ovs_be16 flow_vlan_tci;
2747 uint32_t flow_pkt_mark;
2748 uint8_t flow_nw_tos;
2749 odp_port_t out_port, odp_port;
2750 bool tnl_push_pop_send = false;
2753 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2754 * before traversing a patch port. */
2755 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 31);
2756 memset(&flow_tnl, 0, sizeof flow_tnl);
2759 xlate_report(ctx, "Nonexistent output port");
2761 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2762 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2764 } else if (check_stp) {
2765 if (is_stp(&ctx->base_flow)) {
2766 if (!xport_stp_should_forward_bpdu(xport) &&
2767 !xport_rstp_should_manage_bpdu(xport)) {
2768 if (ctx->xbridge->stp != NULL) {
2769 xlate_report(ctx, "STP not in listening state, "
2770 "skipping bpdu output");
2771 } else if (ctx->xbridge->rstp != NULL) {
2772 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2773 "skipping bpdu output");
2777 } else if (!xport_stp_forward_state(xport) ||
2778 !xport_rstp_forward_state(xport)) {
2779 if (ctx->xbridge->stp != NULL) {
2780 xlate_report(ctx, "STP not in forwarding state, "
2782 } else if (ctx->xbridge->rstp != NULL) {
2783 xlate_report(ctx, "RSTP not in forwarding state, "
2790 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2791 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2796 const struct xport *peer = xport->peer;
2797 struct flow old_flow = ctx->xin->flow;
2798 bool old_was_mpls = ctx->was_mpls;
2799 long long old_version = ctx->tables_version;
2800 enum slow_path_reason special;
2801 struct ofpbuf old_stack = ctx->stack;
2802 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2803 struct ofpbuf old_action_set = ctx->action_set;
2804 uint64_t actset_stub[1024 / 8];
2806 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2807 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2808 ctx->xbridge = peer->xbridge;
2809 flow->in_port.ofp_port = peer->ofp_port;
2810 flow->metadata = htonll(0);
2811 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2812 memset(flow->regs, 0, sizeof flow->regs);
2813 flow->actset_output = OFPP_UNSET;
2815 /* The bridge is now known so obtain its table version. */
2817 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
2819 special = process_special(ctx, &ctx->xin->flow, peer,
2822 ctx->xout->slow |= special;
2823 } else if (may_receive(peer, ctx)) {
2824 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2825 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2826 if (ctx->action_set.size) {
2827 /* Translate action set only if not dropping the packet and
2828 * not recirculating. */
2829 if (!exit_recirculates(ctx)) {
2830 xlate_action_set(ctx);
2833 /* Check if need to recirculate. */
2834 if (exit_recirculates(ctx)) {
2835 compose_recirculate_action(ctx);
2838 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2839 * the learning action look at the packet, then drop it. */
2840 struct flow old_base_flow = ctx->base_flow;
2841 size_t old_size = ctx->xout->odp_actions->size;
2842 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2844 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2845 ctx->xout->mirrors = old_mirrors;
2846 ctx->base_flow = old_base_flow;
2847 ctx->xout->odp_actions->size = old_size;
2849 /* Undo changes that may have been done for recirculation. */
2850 if (exit_recirculates(ctx)) {
2851 ctx->action_set.size = ctx->recirc_action_offset;
2852 ctx->recirc_action_offset = -1;
2853 ctx->last_unroll_offset = -1;
2858 ctx->xin->flow = old_flow;
2859 ctx->xbridge = xport->xbridge;
2860 ofpbuf_uninit(&ctx->action_set);
2861 ctx->action_set = old_action_set;
2862 ofpbuf_uninit(&ctx->stack);
2863 ctx->stack = old_stack;
2865 /* Restore calling bridge's lookup version. */
2866 ctx->tables_version = old_version;
2868 /* The peer bridge popping MPLS should have no effect on the original
2870 ctx->was_mpls = old_was_mpls;
2872 /* The fact that the peer bridge exits (for any reason) does not mean
2873 * that the original bridge should exit. Specifically, if the peer
2874 * bridge recirculates (which typically modifies the packet), the
2875 * original bridge must continue processing with the original, not the
2876 * recirculated packet! */
2879 if (ctx->xin->resubmit_stats) {
2880 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2881 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2883 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2886 if (ctx->xin->xcache) {
2887 struct xc_entry *entry;
2889 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2890 entry->u.dev.tx = netdev_ref(xport->netdev);
2891 entry->u.dev.rx = netdev_ref(peer->netdev);
2892 entry->u.dev.bfd = bfd_ref(peer->bfd);
2897 flow_vlan_tci = flow->vlan_tci;
2898 flow_pkt_mark = flow->pkt_mark;
2899 flow_nw_tos = flow->nw_tos;
2901 if (count_skb_priorities(xport)) {
2902 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2903 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2904 wc->masks.nw_tos |= IP_DSCP_MASK;
2905 flow->nw_tos &= ~IP_DSCP_MASK;
2906 flow->nw_tos |= dscp;
2910 if (xport->is_tunnel) {
2911 /* Save tunnel metadata so that changes made due to
2912 * the Logical (tunnel) Port are not visible for any further
2913 * matches, while explicit set actions on tunnel metadata are.
2915 flow_tnl = flow->tunnel;
2916 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2917 if (odp_port == ODPP_NONE) {
2918 xlate_report(ctx, "Tunneling decided against output");
2919 goto out; /* restore flow_nw_tos */
2921 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2922 xlate_report(ctx, "Not tunneling to our own address");
2923 goto out; /* restore flow_nw_tos */
2925 if (ctx->xin->resubmit_stats) {
2926 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2928 if (ctx->xin->xcache) {
2929 struct xc_entry *entry;
2931 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2932 entry->u.dev.tx = netdev_ref(xport->netdev);
2934 out_port = odp_port;
2935 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2936 xlate_report(ctx, "output to native tunnel");
2937 tnl_push_pop_send = true;
2939 xlate_report(ctx, "output to kernel tunnel");
2940 commit_odp_tunnel_action(flow, &ctx->base_flow,
2941 ctx->xout->odp_actions);
2942 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2945 odp_port = xport->odp_port;
2946 out_port = odp_port;
2947 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2948 ofp_port_t vlandev_port;
2950 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2951 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2952 ofp_port, flow->vlan_tci);
2953 if (vlandev_port != ofp_port) {
2954 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2955 flow->vlan_tci = htons(0);
2960 if (out_port != ODPP_NONE) {
2961 bool use_masked = ctx->xbridge->support.masked_set_action;
2963 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2964 ctx->xout->odp_actions,
2968 struct ovs_action_hash *act_hash;
2971 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2972 OVS_ACTION_ATTR_HASH,
2974 act_hash->hash_alg = xr->hash_alg;
2975 act_hash->hash_basis = xr->hash_basis;
2977 /* Recirc action. */
2978 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2982 if (tnl_push_pop_send) {
2983 build_tunnel_send(ctx, xport, flow, odp_port);
2984 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2986 odp_port_t odp_tnl_port = ODPP_NONE;
2988 /* XXX: Write better Filter for tunnel port. We can use inport
2989 * int tunnel-port flow to avoid these checks completely. */
2990 if (ofp_port == OFPP_LOCAL &&
2991 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2993 odp_tnl_port = tnl_port_map_lookup(flow, wc);
2996 if (odp_tnl_port != ODPP_NONE) {
2997 nl_msg_put_odp_port(ctx->xout->odp_actions,
2998 OVS_ACTION_ATTR_TUNNEL_POP,
3001 /* Tunnel push-pop action is not compatible with
3003 add_ipfix_output_action(ctx, out_port);
3004 nl_msg_put_odp_port(ctx->xout->odp_actions,
3005 OVS_ACTION_ATTR_OUTPUT,
3011 ctx->sflow_odp_port = odp_port;
3012 ctx->sflow_n_outputs++;
3013 ctx->xout->nf_output_iface = ofp_port;
3018 flow->vlan_tci = flow_vlan_tci;
3019 flow->pkt_mark = flow_pkt_mark;
3020 flow->nw_tos = flow_nw_tos;
3024 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3025 const struct xlate_bond_recirc *xr)
3027 compose_output_action__(ctx, ofp_port, xr, true);
3031 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3033 struct rule_dpif *old_rule = ctx->rule;
3034 ovs_be64 old_cookie = ctx->rule_cookie;
3035 const struct rule_actions *actions;
3037 if (ctx->xin->resubmit_stats) {
3038 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3044 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3045 actions = rule_dpif_get_actions(rule);
3046 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3047 ctx->rule_cookie = old_cookie;
3048 ctx->rule = old_rule;
3053 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3055 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3057 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3058 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
3059 MAX_RESUBMIT_RECURSION);
3060 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3061 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
3062 } else if (ctx->xout->odp_actions->size > UINT16_MAX) {
3063 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
3064 } else if (ctx->stack.size >= 65536) {
3065 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
3074 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3075 bool may_packet_in, bool honor_table_miss)
3077 /* Check if we need to recirculate before matching in a table. */
3078 if (ctx->was_mpls) {
3079 ctx_trigger_recirculation(ctx);
3082 if (xlate_resubmit_resource_check(ctx)) {
3083 struct flow_wildcards *wc;
3084 uint8_t old_table_id = ctx->table_id;
3085 struct rule_dpif *rule;
3087 ctx->table_id = table_id;
3088 wc = (ctx->xin->skip_wildcards) ? NULL : &ctx->xout->wc;
3090 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3091 ctx->tables_version,
3092 &ctx->xin->flow, wc,
3093 ctx->xin->xcache != NULL,
3094 ctx->xin->resubmit_stats,
3095 &ctx->table_id, in_port,
3096 may_packet_in, honor_table_miss);
3098 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3099 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3103 /* Fill in the cache entry here instead of xlate_recursively
3104 * to make the reference counting more explicit. We take a
3105 * reference in the lookups above if we are going to cache the
3107 if (ctx->xin->xcache) {
3108 struct xc_entry *entry;
3110 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3111 entry->u.rule = rule;
3113 xlate_recursively(ctx, rule);
3116 ctx->table_id = old_table_id;
3124 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3125 struct ofputil_bucket *bucket)
3127 if (ctx->xin->resubmit_stats) {
3128 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3130 if (ctx->xin->xcache) {
3131 struct xc_entry *entry;
3133 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3134 entry->u.group.group = group_dpif_ref(group);
3135 entry->u.group.bucket = bucket;
3140 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3142 uint64_t action_list_stub[1024 / 8];
3143 struct ofpbuf action_list, action_set;
3144 struct flow old_flow = ctx->xin->flow;
3145 bool old_was_mpls = ctx->was_mpls;
3147 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3148 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3150 ofpacts_execute_action_set(&action_list, &action_set);
3152 do_xlate_actions(action_list.data, action_list.size, ctx);
3155 ofpbuf_uninit(&action_set);
3156 ofpbuf_uninit(&action_list);
3158 /* Check if need to recirculate. */
3159 if (exit_recirculates(ctx)) {
3160 compose_recirculate_action(ctx);
3163 /* Roll back flow to previous state.
3164 * This is equivalent to cloning the packet for each bucket.
3166 * As a side effect any subsequently applied actions will
3167 * also effectively be applied to a clone of the packet taken
3168 * just before applying the all or indirect group.
3170 * Note that group buckets are action sets, hence they cannot modify the
3171 * main action set. Also any stack actions are ignored when executing an
3172 * action set, so group buckets cannot change the stack either.
3173 * However, we do allow resubmit actions in group buckets, which could
3174 * break the above assumptions. It is up to the controller to not mess up
3175 * with the action_set and stack in the tables resubmitted to from
3177 ctx->xin->flow = old_flow;
3179 /* The group bucket popping MPLS should have no effect after bucket
3181 ctx->was_mpls = old_was_mpls;
3183 /* The fact that the group bucket exits (for any reason) does not mean that
3184 * the translation after the group action should exit. Specifically, if
3185 * the group bucket recirculates (which typically modifies the packet), the
3186 * actions after the group action must continue processing with the
3187 * original, not the recirculated packet! */
3192 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3194 struct ofputil_bucket *bucket;
3195 const struct ovs_list *buckets;
3197 group_dpif_get_buckets(group, &buckets);
3199 LIST_FOR_EACH (bucket, list_node, buckets) {
3200 xlate_group_bucket(ctx, bucket);
3202 xlate_group_stats(ctx, group, NULL);
3206 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3208 struct ofputil_bucket *bucket;
3210 bucket = group_first_live_bucket(ctx, group, 0);
3212 xlate_group_bucket(ctx, bucket);
3213 xlate_group_stats(ctx, group, bucket);
3218 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3220 struct flow_wildcards *wc = &ctx->xout->wc;
3221 struct ofputil_bucket *bucket;
3224 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3225 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3226 bucket = group_best_live_bucket(ctx, group, basis);
3228 xlate_group_bucket(ctx, bucket);
3229 xlate_group_stats(ctx, group, bucket);
3234 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3236 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3237 struct flow_wildcards *wc = &ctx->xout->wc;
3238 const struct field_array *fields;
3239 struct ofputil_bucket *bucket;
3243 fields = group_dpif_get_fields(group);
3244 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3246 /* Determine which fields to hash */
3247 for (i = 0; i < MFF_N_IDS; i++) {
3248 if (bitmap_is_set(fields->used.bm, i)) {
3249 const struct mf_field *mf;
3251 /* If the field is already present in 'hash_fields' then
3252 * this loop has already checked that it and its pre-requisites
3253 * are present in the flow and its pre-requisites have
3254 * already been added to 'hash_fields'. There is nothing more
3255 * to do here and as an optimisation the loop can continue. */
3256 if (bitmap_is_set(hash_fields.bm, i)) {
3262 /* Only hash a field if it and its pre-requisites are present
3264 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3268 /* Hash both the field and its pre-requisites */
3269 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3273 /* Hash the fields */
3274 for (i = 0; i < MFF_N_IDS; i++) {
3275 if (bitmap_is_set(hash_fields.bm, i)) {
3276 const struct mf_field *mf = mf_from_id(i);
3277 union mf_value value;
3280 mf_get_value(mf, &ctx->xin->flow, &value);
3281 /* This seems inefficient but so does apply_mask() */
3282 for (j = 0; j < mf->n_bytes; j++) {
3283 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3285 basis = hash_bytes(&value, mf->n_bytes, basis);
3287 mf_mask_field(mf, &wc->masks);
3291 bucket = group_best_live_bucket(ctx, group, basis);
3293 xlate_group_bucket(ctx, bucket);
3294 xlate_group_stats(ctx, group, bucket);
3299 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3301 const char *selection_method = group_dpif_get_selection_method(group);
3303 if (selection_method[0] == '\0') {
3304 xlate_default_select_group(ctx, group);
3305 } else if (!strcasecmp("hash", selection_method)) {
3306 xlate_hash_fields_select_group(ctx, group);
3308 /* Parsing of groups should ensure this never happens */
3314 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3316 ctx->in_group = true;
3318 switch (group_dpif_get_type(group)) {
3320 case OFPGT11_INDIRECT:
3321 xlate_all_group(ctx, group);
3323 case OFPGT11_SELECT:
3324 xlate_select_group(ctx, group);
3327 xlate_ff_group(ctx, group);
3332 group_dpif_unref(group);
3334 ctx->in_group = false;
3338 xlate_group_resource_check(struct xlate_ctx *ctx)
3340 if (!xlate_resubmit_resource_check(ctx)) {
3342 } else if (ctx->in_group) {
3343 /* Prevent nested translation of OpenFlow groups.
3345 * OpenFlow allows this restriction. We enforce this restriction only
3346 * because, with the current architecture, we would otherwise have to
3347 * take a possibly recursive read lock on the ofgroup rwlock, which is
3348 * unsafe given that POSIX allows taking a read lock to block if there
3349 * is a thread blocked on taking the write lock. Other solutions
3350 * without this restriction are also possible, but seem unwarranted
3351 * given the current limited use of groups. */
3352 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3354 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
3362 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3364 if (xlate_group_resource_check(ctx)) {
3365 struct group_dpif *group;
3368 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3370 xlate_group_action__(ctx, group);
3380 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3381 const struct ofpact_resubmit *resubmit)
3385 bool may_packet_in = false;
3386 bool honor_table_miss = false;
3388 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3389 /* Still allow missed packets to be sent to the controller
3390 * if resubmitting from an internal table. */
3391 may_packet_in = true;
3392 honor_table_miss = true;
3395 in_port = resubmit->in_port;
3396 if (in_port == OFPP_IN_PORT) {
3397 in_port = ctx->xin->flow.in_port.ofp_port;
3400 table_id = resubmit->table_id;
3401 if (table_id == 255) {
3402 table_id = ctx->table_id;
3405 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3410 flood_packets(struct xlate_ctx *ctx, bool all)
3412 const struct xport *xport;
3414 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3415 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3420 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3421 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3422 compose_output_action(ctx, xport->ofp_port, NULL);
3426 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3430 execute_controller_action(struct xlate_ctx *ctx, int len,
3431 enum ofp_packet_in_reason reason,
3432 uint16_t controller_id)
3434 struct ofproto_packet_in *pin;
3435 struct dp_packet *packet;
3438 ctx->xout->slow |= SLOW_CONTROLLER;
3439 if (!ctx->xin->packet) {
3443 packet = dp_packet_clone(ctx->xin->packet);
3445 use_masked = ctx->xbridge->support.masked_set_action;
3446 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3447 ctx->xout->odp_actions,
3448 &ctx->xout->wc, use_masked);
3450 odp_execute_actions(NULL, &packet, 1, false,
3451 ctx->xout->odp_actions->data,
3452 ctx->xout->odp_actions->size, NULL);
3454 pin = xmalloc(sizeof *pin);
3455 pin->up.packet_len = dp_packet_size(packet);
3456 pin->up.packet = dp_packet_steal_data(packet);
3457 pin->up.reason = reason;
3458 pin->up.table_id = ctx->table_id;
3459 pin->up.cookie = ctx->rule_cookie;
3461 flow_get_metadata(&ctx->xin->flow, &pin->up.flow_metadata);
3463 pin->controller_id = controller_id;
3464 pin->send_len = len;
3465 /* If a rule is a table-miss rule then this is
3466 * a table-miss handled by a table-miss rule.
3468 * Else, if rule is internal and has a controller action,
3469 * the later being implied by the rule being processed here,
3470 * then this is a table-miss handled without a table-miss rule.
3472 * Otherwise this is not a table-miss. */
3473 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3475 if (rule_dpif_is_table_miss(ctx->rule)) {
3476 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3477 } else if (rule_dpif_is_internal(ctx->rule)) {
3478 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3481 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3482 dp_packet_delete(packet);
3485 /* Called only when ctx->recirc_action_offset is set. */
3487 compose_recirculate_action(struct xlate_ctx *ctx)
3489 struct recirc_metadata md;
3493 use_masked = ctx->xbridge->support.masked_set_action;
3494 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3495 ctx->xout->odp_actions,
3496 &ctx->xout->wc, use_masked);
3498 recirc_metadata_from_flow(&md, &ctx->xin->flow);
3500 ovs_assert(ctx->recirc_action_offset >= 0);
3502 /* Only allocate recirculation ID if we have a packet. */
3503 if (ctx->xin->packet) {
3504 /* Allocate a unique recirc id for the given metadata state in the
3505 * flow. The life-cycle of this recirc id is managed by associating it
3506 * with the udpif key ('ukey') created for each new datapath flow. */
3507 id = recirc_alloc_id_ctx(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3508 ctx->recirc_action_offset,
3509 ctx->action_set.size, ctx->action_set.data);
3511 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3512 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3515 xlate_out_add_recirc(ctx->xout, id);
3517 /* Look up an existing recirc id for the given metadata state in the
3518 * flow. No new reference is taken, as the ID is RCU protected and is
3519 * only required temporarily for verification. */
3520 id = recirc_find_id(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3521 ctx->recirc_action_offset,
3522 ctx->action_set.size, ctx->action_set.data);
3523 /* We let zero 'id' to be used in the RECIRC action below, which will
3524 * fail all revalidations as zero is not a valid recirculation ID. */
3527 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3529 /* Undo changes done by recirculation. */
3530 ctx->action_set.size = ctx->recirc_action_offset;
3531 ctx->recirc_action_offset = -1;
3532 ctx->last_unroll_offset = -1;
3536 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3538 struct flow_wildcards *wc = &ctx->xout->wc;
3539 struct flow *flow = &ctx->xin->flow;
3542 ovs_assert(eth_type_mpls(mpls->ethertype));
3544 n = flow_count_mpls_labels(flow, wc);
3546 bool use_masked = ctx->xbridge->support.masked_set_action;
3548 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3549 ctx->xout->odp_actions,
3550 &ctx->xout->wc, use_masked);
3551 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3552 if (ctx->xin->packet != NULL) {
3553 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3554 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3555 "MPLS push action can't be performed as it would "
3556 "have more MPLS LSEs than the %d supported.",
3557 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3563 flow_push_mpls(flow, n, mpls->ethertype, wc);
3567 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3569 struct flow_wildcards *wc = &ctx->xout->wc;
3570 struct flow *flow = &ctx->xin->flow;
3571 int n = flow_count_mpls_labels(flow, wc);
3573 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3574 if (ctx->xbridge->support.recirc) {
3575 ctx->was_mpls = true;
3577 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3578 if (ctx->xin->packet != NULL) {
3579 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3580 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3581 "MPLS pop action can't be performed as it has "
3582 "more MPLS LSEs than the %d supported.",
3583 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3586 ofpbuf_clear(ctx->xout->odp_actions);
3591 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3593 struct flow *flow = &ctx->xin->flow;
3595 if (!is_ip_any(flow)) {
3599 ctx->xout->wc.masks.nw_ttl = 0xff;
3600 if (flow->nw_ttl > 1) {
3606 for (i = 0; i < ids->n_controllers; i++) {
3607 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3611 /* Stop processing for current table. */
3617 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3619 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3620 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3621 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3626 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3628 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3629 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3630 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3635 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3637 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3638 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3639 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3644 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3646 struct flow *flow = &ctx->xin->flow;
3647 struct flow_wildcards *wc = &ctx->xout->wc;
3649 if (eth_type_mpls(flow->dl_type)) {
3650 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3652 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3655 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3658 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3662 /* Stop processing for current table. */
3667 xlate_output_action(struct xlate_ctx *ctx,
3668 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3670 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3672 ctx->xout->nf_output_iface = NF_OUT_DROP;
3676 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3679 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3680 0, may_packet_in, true);
3686 flood_packets(ctx, false);
3689 flood_packets(ctx, true);
3691 case OFPP_CONTROLLER:
3692 execute_controller_action(ctx, max_len,
3693 (ctx->in_group ? OFPR_GROUP
3694 : ctx->in_action_set ? OFPR_ACTION_SET
3702 if (port != ctx->xin->flow.in_port.ofp_port) {
3703 compose_output_action(ctx, port, NULL);
3705 xlate_report(ctx, "skipping output to input port");
3710 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3711 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3712 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3713 ctx->xout->nf_output_iface = prev_nf_output_iface;
3714 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3715 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3716 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3721 xlate_output_reg_action(struct xlate_ctx *ctx,
3722 const struct ofpact_output_reg *or)
3724 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3725 if (port <= UINT16_MAX) {
3726 union mf_subvalue value;
3728 memset(&value, 0xff, sizeof value);
3729 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3730 xlate_output_action(ctx, u16_to_ofp(port),
3731 or->max_len, false);
3736 xlate_enqueue_action(struct xlate_ctx *ctx,
3737 const struct ofpact_enqueue *enqueue)
3739 ofp_port_t ofp_port = enqueue->port;
3740 uint32_t queue_id = enqueue->queue;
3741 uint32_t flow_priority, priority;
3744 /* Translate queue to priority. */
3745 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3747 /* Fall back to ordinary output action. */
3748 xlate_output_action(ctx, enqueue->port, 0, false);
3752 /* Check output port. */
3753 if (ofp_port == OFPP_IN_PORT) {
3754 ofp_port = ctx->xin->flow.in_port.ofp_port;
3755 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3759 /* Add datapath actions. */
3760 flow_priority = ctx->xin->flow.skb_priority;
3761 ctx->xin->flow.skb_priority = priority;
3762 compose_output_action(ctx, ofp_port, NULL);
3763 ctx->xin->flow.skb_priority = flow_priority;
3765 /* Update NetFlow output port. */
3766 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3767 ctx->xout->nf_output_iface = ofp_port;
3768 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3769 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3774 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3776 uint32_t skb_priority;
3778 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3779 ctx->xin->flow.skb_priority = skb_priority;
3781 /* Couldn't translate queue to a priority. Nothing to do. A warning
3782 * has already been logged. */
3787 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3789 const struct xbridge *xbridge = xbridge_;
3800 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3803 port = get_ofp_port(xbridge, ofp_port);
3804 return port ? port->may_enable : false;
3809 xlate_bundle_action(struct xlate_ctx *ctx,
3810 const struct ofpact_bundle *bundle)
3814 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3816 CONST_CAST(struct xbridge *, ctx->xbridge));
3817 if (bundle->dst.field) {
3818 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3821 xlate_output_action(ctx, port, 0, false);
3826 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3827 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3829 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3830 if (ctx->xin->may_learn) {
3831 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3836 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3838 ctx->xout->has_learn = true;
3839 learn_mask(learn, &ctx->xout->wc);
3841 if (ctx->xin->xcache) {
3842 struct xc_entry *entry;
3844 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3845 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3846 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3847 entry->u.learn.ofpacts = ofpbuf_new(64);
3848 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3849 entry->u.learn.ofpacts);
3850 } else if (ctx->xin->may_learn) {
3851 uint64_t ofpacts_stub[1024 / 8];
3852 struct ofputil_flow_mod fm;
3853 struct ofpbuf ofpacts;
3855 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3856 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3857 ofpbuf_uninit(&ofpacts);
3862 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3863 uint16_t idle_timeout, uint16_t hard_timeout)
3865 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3866 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3871 xlate_fin_timeout(struct xlate_ctx *ctx,
3872 const struct ofpact_fin_timeout *oft)
3875 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3876 oft->fin_idle_timeout, oft->fin_hard_timeout);
3877 if (ctx->xin->xcache) {
3878 struct xc_entry *entry;
3880 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3881 /* XC_RULE already holds a reference on the rule, none is taken
3883 entry->u.fin.rule = ctx->rule;
3884 entry->u.fin.idle = oft->fin_idle_timeout;
3885 entry->u.fin.hard = oft->fin_hard_timeout;
3891 xlate_sample_action(struct xlate_ctx *ctx,
3892 const struct ofpact_sample *os)
3894 union user_action_cookie cookie;
3895 /* Scale the probability from 16-bit to 32-bit while representing
3896 * the same percentage. */
3897 uint32_t probability = (os->probability << 16) | os->probability;
3900 if (!ctx->xbridge->support.variable_length_userdata) {
3901 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3903 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3904 "lacks support (needs Linux 3.10+ or kernel module from "
3909 use_masked = ctx->xbridge->support.masked_set_action;
3910 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3911 ctx->xout->odp_actions,
3912 &ctx->xout->wc, use_masked);
3914 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3915 os->obs_domain_id, os->obs_point_id, &cookie);
3916 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions,
3917 &ctx->xin->flow, probability, &cookie,
3918 sizeof cookie.flow_sample, ODPP_NONE);
3922 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3924 if (xport->config & (is_stp(&ctx->xin->flow)
3925 ? OFPUTIL_PC_NO_RECV_STP
3926 : OFPUTIL_PC_NO_RECV)) {
3930 /* Only drop packets here if both forwarding and learning are
3931 * disabled. If just learning is enabled, we need to have
3932 * OFPP_NORMAL and the learning action have a look at the packet
3933 * before we can drop it. */
3934 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3935 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3943 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3945 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3946 size_t on_len = ofpact_nest_get_action_len(on);
3947 const struct ofpact *inner;
3949 /* Maintain actset_output depending on the contents of the action set:
3951 * - OFPP_UNSET, if there is no "output" action.
3953 * - The output port, if there is an "output" action and no "group"
3956 * - OFPP_UNSET, if there is a "group" action.
3958 if (!ctx->action_set_has_group) {
3959 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3960 if (inner->type == OFPACT_OUTPUT) {
3961 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3962 } else if (inner->type == OFPACT_GROUP) {
3963 ctx->xin->flow.actset_output = OFPP_UNSET;
3964 ctx->action_set_has_group = true;
3969 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3970 ofpact_pad(&ctx->action_set);
3974 xlate_action_set(struct xlate_ctx *ctx)
3976 uint64_t action_list_stub[1024 / 64];
3977 struct ofpbuf action_list;
3979 ctx->in_action_set = true;
3980 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3981 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3982 /* Clear the action set, as it is not needed any more. */
3983 ofpbuf_clear(&ctx->action_set);
3984 do_xlate_actions(action_list.data, action_list.size, ctx);
3985 ctx->in_action_set = false;
3986 ofpbuf_uninit(&action_list);
3990 recirc_put_unroll_xlate(struct xlate_ctx *ctx)
3992 struct ofpact_unroll_xlate *unroll;
3994 unroll = ctx->last_unroll_offset < 0
3996 : ALIGNED_CAST(struct ofpact_unroll_xlate *,
3997 (char *)ctx->action_set.data + ctx->last_unroll_offset);
3999 /* Restore the table_id and rule cookie for a potential PACKET
4002 (ctx->table_id != unroll->rule_table_id
4003 || ctx->rule_cookie != unroll->rule_cookie)) {
4005 ctx->last_unroll_offset = ctx->action_set.size;
4006 unroll = ofpact_put_UNROLL_XLATE(&ctx->action_set);
4007 unroll->rule_table_id = ctx->table_id;
4008 unroll->rule_cookie = ctx->rule_cookie;
4013 /* Copy remaining actions to the action_set to be executed after recirculation.
4014 * UNROLL_XLATE action is inserted, if not already done so, before actions that
4015 * may generate PACKET_INs from the current table and without matching another
4018 recirc_unroll_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4019 struct xlate_ctx *ctx)
4021 const struct ofpact *a;
4023 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4025 /* May generate PACKET INs. */
4026 case OFPACT_OUTPUT_REG:
4029 case OFPACT_CONTROLLER:
4030 case OFPACT_DEC_MPLS_TTL:
4031 case OFPACT_DEC_TTL:
4032 recirc_put_unroll_xlate(ctx);
4035 /* These may not generate PACKET INs. */
4036 case OFPACT_SET_TUNNEL:
4037 case OFPACT_REG_MOVE:
4038 case OFPACT_SET_FIELD:
4039 case OFPACT_STACK_PUSH:
4040 case OFPACT_STACK_POP:
4042 case OFPACT_WRITE_METADATA:
4043 case OFPACT_RESUBMIT: /* May indirectly generate PACKET INs, */
4044 case OFPACT_GOTO_TABLE: /* but from a different table and rule. */
4045 case OFPACT_ENQUEUE:
4046 case OFPACT_SET_VLAN_VID:
4047 case OFPACT_SET_VLAN_PCP:
4048 case OFPACT_STRIP_VLAN:
4049 case OFPACT_PUSH_VLAN:
4050 case OFPACT_SET_ETH_SRC:
4051 case OFPACT_SET_ETH_DST:
4052 case OFPACT_SET_IPV4_SRC:
4053 case OFPACT_SET_IPV4_DST:
4054 case OFPACT_SET_IP_DSCP:
4055 case OFPACT_SET_IP_ECN:
4056 case OFPACT_SET_IP_TTL:
4057 case OFPACT_SET_L4_SRC_PORT:
4058 case OFPACT_SET_L4_DST_PORT:
4059 case OFPACT_SET_QUEUE:
4060 case OFPACT_POP_QUEUE:
4061 case OFPACT_PUSH_MPLS:
4062 case OFPACT_POP_MPLS:
4063 case OFPACT_SET_MPLS_LABEL:
4064 case OFPACT_SET_MPLS_TC:
4065 case OFPACT_SET_MPLS_TTL:
4066 case OFPACT_MULTIPATH:
4069 case OFPACT_UNROLL_XLATE:
4070 case OFPACT_FIN_TIMEOUT:
4071 case OFPACT_CLEAR_ACTIONS:
4072 case OFPACT_WRITE_ACTIONS:
4077 /* These need not be copied for restoration. */
4079 case OFPACT_CONJUNCTION:
4082 /* Copy the action over. */
4083 ofpbuf_put(&ctx->action_set, a, OFPACT_ALIGN(a->len));
4087 #define CHECK_MPLS_RECIRCULATION() \
4088 if (ctx->was_mpls) { \
4089 ctx_trigger_recirculation(ctx); \
4092 #define CHECK_MPLS_RECIRCULATION_IF(COND) \
4094 CHECK_MPLS_RECIRCULATION(); \
4098 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4099 struct xlate_ctx *ctx)
4101 struct flow_wildcards *wc = &ctx->xout->wc;
4102 struct flow *flow = &ctx->xin->flow;
4103 const struct ofpact *a;
4105 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4106 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
4108 /* dl_type already in the mask, not set below. */
4110 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4111 struct ofpact_controller *controller;
4112 const struct ofpact_metadata *metadata;
4113 const struct ofpact_set_field *set_field;
4114 const struct mf_field *mf;
4117 /* Check if need to store the remaining actions for later
4119 if (exit_recirculates(ctx)) {
4120 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4122 (uint8_t *)ofpacts)),
4130 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4131 ofpact_get_OUTPUT(a)->max_len, true);
4135 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4136 /* Group could not be found. */
4141 case OFPACT_CONTROLLER:
4142 controller = ofpact_get_CONTROLLER(a);
4143 execute_controller_action(ctx, controller->max_len,
4145 controller->controller_id);
4148 case OFPACT_ENQUEUE:
4149 memset(&wc->masks.skb_priority, 0xff,
4150 sizeof wc->masks.skb_priority);
4151 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4154 case OFPACT_SET_VLAN_VID:
4155 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4156 if (flow->vlan_tci & htons(VLAN_CFI) ||
4157 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4158 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4159 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4164 case OFPACT_SET_VLAN_PCP:
4165 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4166 if (flow->vlan_tci & htons(VLAN_CFI) ||
4167 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4168 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4169 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4170 << VLAN_PCP_SHIFT) | VLAN_CFI);
4174 case OFPACT_STRIP_VLAN:
4175 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4176 flow->vlan_tci = htons(0);
4179 case OFPACT_PUSH_VLAN:
4180 /* XXX 802.1AD(QinQ) */
4181 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4182 flow->vlan_tci = htons(VLAN_CFI);
4185 case OFPACT_SET_ETH_SRC:
4186 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
4187 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
4190 case OFPACT_SET_ETH_DST:
4191 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
4192 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
4195 case OFPACT_SET_IPV4_SRC:
4196 CHECK_MPLS_RECIRCULATION();
4197 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4198 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4199 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4203 case OFPACT_SET_IPV4_DST:
4204 CHECK_MPLS_RECIRCULATION();
4205 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4206 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4207 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4211 case OFPACT_SET_IP_DSCP:
4212 CHECK_MPLS_RECIRCULATION();
4213 if (is_ip_any(flow)) {
4214 wc->masks.nw_tos |= IP_DSCP_MASK;
4215 flow->nw_tos &= ~IP_DSCP_MASK;
4216 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4220 case OFPACT_SET_IP_ECN:
4221 CHECK_MPLS_RECIRCULATION();
4222 if (is_ip_any(flow)) {
4223 wc->masks.nw_tos |= IP_ECN_MASK;
4224 flow->nw_tos &= ~IP_ECN_MASK;
4225 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4229 case OFPACT_SET_IP_TTL:
4230 CHECK_MPLS_RECIRCULATION();
4231 if (is_ip_any(flow)) {
4232 wc->masks.nw_ttl = 0xff;
4233 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4237 case OFPACT_SET_L4_SRC_PORT:
4238 CHECK_MPLS_RECIRCULATION();
4239 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4240 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4241 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4242 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4246 case OFPACT_SET_L4_DST_PORT:
4247 CHECK_MPLS_RECIRCULATION();
4248 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4249 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4250 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4251 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4255 case OFPACT_RESUBMIT:
4256 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4259 case OFPACT_SET_TUNNEL:
4260 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4263 case OFPACT_SET_QUEUE:
4264 memset(&wc->masks.skb_priority, 0xff,
4265 sizeof wc->masks.skb_priority);
4266 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4269 case OFPACT_POP_QUEUE:
4270 memset(&wc->masks.skb_priority, 0xff,
4271 sizeof wc->masks.skb_priority);
4272 flow->skb_priority = ctx->orig_skb_priority;
4275 case OFPACT_REG_MOVE:
4276 CHECK_MPLS_RECIRCULATION_IF(
4277 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4278 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4279 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4282 case OFPACT_SET_FIELD:
4283 CHECK_MPLS_RECIRCULATION_IF(
4284 mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field));
4285 set_field = ofpact_get_SET_FIELD(a);
4286 mf = set_field->field;
4288 /* Set field action only ever overwrites packet's outermost
4289 * applicable header fields. Do nothing if no header exists. */
4290 if (mf->id == MFF_VLAN_VID) {
4291 wc->masks.vlan_tci |= htons(VLAN_CFI);
4292 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4295 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4296 /* 'dl_type' is already unwildcarded. */
4297 && !eth_type_mpls(flow->dl_type)) {
4300 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4301 * header field on a packet that does not have them. */
4302 mf_mask_field_and_prereqs(mf, &wc->masks);
4303 if (mf_are_prereqs_ok(mf, flow)) {
4304 mf_set_flow_value_masked(mf, &set_field->value,
4305 &set_field->mask, flow);
4309 case OFPACT_STACK_PUSH:
4310 CHECK_MPLS_RECIRCULATION_IF(
4311 mf_is_l3_or_higher(ofpact_get_STACK_PUSH(a)->subfield.field));
4312 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4316 case OFPACT_STACK_POP:
4317 CHECK_MPLS_RECIRCULATION_IF(
4318 mf_is_l3_or_higher(ofpact_get_STACK_POP(a)->subfield.field));
4319 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4323 case OFPACT_PUSH_MPLS:
4324 /* Recirculate if it is an IP packet with a zero ttl. This may
4325 * indicate that the packet was previously MPLS and an MPLS pop
4326 * action converted it to IP. In this case recirculating should
4327 * reveal the IP TTL which is used as the basis for a new MPLS
4329 CHECK_MPLS_RECIRCULATION_IF(
4330 !flow_count_mpls_labels(flow, wc)
4331 && flow->nw_ttl == 0
4332 && is_ip_any(flow));
4333 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4336 case OFPACT_POP_MPLS:
4337 CHECK_MPLS_RECIRCULATION();
4338 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4341 case OFPACT_SET_MPLS_LABEL:
4342 CHECK_MPLS_RECIRCULATION();
4343 compose_set_mpls_label_action(
4344 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4347 case OFPACT_SET_MPLS_TC:
4348 CHECK_MPLS_RECIRCULATION();
4349 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4352 case OFPACT_SET_MPLS_TTL:
4353 CHECK_MPLS_RECIRCULATION();
4354 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4357 case OFPACT_DEC_MPLS_TTL:
4358 CHECK_MPLS_RECIRCULATION();
4359 if (compose_dec_mpls_ttl_action(ctx)) {
4364 case OFPACT_DEC_TTL:
4365 CHECK_MPLS_RECIRCULATION();
4366 wc->masks.nw_ttl = 0xff;
4367 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4373 /* Nothing to do. */
4376 case OFPACT_MULTIPATH:
4377 CHECK_MPLS_RECIRCULATION();
4378 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4382 CHECK_MPLS_RECIRCULATION();
4383 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4386 case OFPACT_OUTPUT_REG:
4387 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4391 CHECK_MPLS_RECIRCULATION();
4392 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4395 case OFPACT_CONJUNCTION: {
4396 /* A flow with a "conjunction" action represents part of a special
4397 * kind of "set membership match". Such a flow should not actually
4398 * get executed, but it could via, say, a "packet-out", even though
4399 * that wouldn't be useful. Log it to help debugging. */
4400 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4401 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4409 case OFPACT_UNROLL_XLATE: {
4410 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4412 /* Restore translation context data that was stored earlier. */
4413 ctx->table_id = unroll->rule_table_id;
4414 ctx->rule_cookie = unroll->rule_cookie;
4417 case OFPACT_FIN_TIMEOUT:
4418 CHECK_MPLS_RECIRCULATION();
4419 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4420 ctx->xout->has_fin_timeout = true;
4421 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4424 case OFPACT_CLEAR_ACTIONS:
4425 ofpbuf_clear(&ctx->action_set);
4426 ctx->xin->flow.actset_output = OFPP_UNSET;
4427 ctx->action_set_has_group = false;
4430 case OFPACT_WRITE_ACTIONS:
4431 xlate_write_actions(ctx, a);
4434 case OFPACT_WRITE_METADATA:
4435 metadata = ofpact_get_WRITE_METADATA(a);
4436 flow->metadata &= ~metadata->mask;
4437 flow->metadata |= metadata->metadata & metadata->mask;
4441 /* Not implemented yet. */
4444 case OFPACT_GOTO_TABLE: {
4445 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4447 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4448 * than ogt->table_id. This is to allow goto_table actions that
4449 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4450 * after recirculation. */
4451 ovs_assert(ctx->table_id == TBL_INTERNAL
4452 || ctx->table_id < ogt->table_id);
4453 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4454 ogt->table_id, true, true);
4459 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4463 /* Check if need to store this and the remaining actions for later
4465 if (ctx->exit && ctx_first_recirculation_action(ctx)) {
4466 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4468 (uint8_t *)ofpacts)),
4476 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4477 const struct flow *flow, ofp_port_t in_port,
4478 struct rule_dpif *rule, uint16_t tcp_flags,
4479 const struct dp_packet *packet)
4481 xin->ofproto = ofproto;
4483 xin->flow.in_port.ofp_port = in_port;
4484 xin->flow.actset_output = OFPP_UNSET;
4485 xin->packet = packet;
4486 xin->may_learn = packet != NULL;
4489 xin->ofpacts = NULL;
4490 xin->ofpacts_len = 0;
4491 xin->tcp_flags = tcp_flags;
4492 xin->resubmit_hook = NULL;
4493 xin->report_hook = NULL;
4494 xin->resubmit_stats = NULL;
4495 xin->skip_wildcards = false;
4496 xin->odp_actions = NULL;
4498 /* Do recirc lookup. */
4499 xin->recirc = flow->recirc_id
4500 ? recirc_id_node_find(flow->recirc_id)
4505 xlate_out_uninit(struct xlate_out *xout)
4508 if (xout->odp_actions == &xout->odp_actions_buf) {
4509 ofpbuf_uninit(xout->odp_actions);
4511 xlate_out_free_recircs(xout);
4515 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4516 * into datapath actions, using 'ctx', and discards the datapath actions. */
4518 xlate_actions_for_side_effects(struct xlate_in *xin)
4520 struct xlate_out xout;
4522 xlate_actions(xin, &xout);
4523 xlate_out_uninit(&xout);
4527 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
4530 dst->slow = src->slow;
4531 dst->has_learn = src->has_learn;
4532 dst->has_normal = src->has_normal;
4533 dst->has_fin_timeout = src->has_fin_timeout;
4534 dst->nf_output_iface = src->nf_output_iface;
4535 dst->mirrors = src->mirrors;
4537 dst->odp_actions = &dst->odp_actions_buf;
4538 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
4539 sizeof dst->odp_actions_stub);
4540 ofpbuf_put(dst->odp_actions, src->odp_actions->data, src->odp_actions->size);
4543 static struct skb_priority_to_dscp *
4544 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4546 struct skb_priority_to_dscp *pdscp;
4549 hash = hash_int(skb_priority, 0);
4550 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4551 if (pdscp->skb_priority == skb_priority) {
4559 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4562 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4563 *dscp = pdscp ? pdscp->dscp : 0;
4564 return pdscp != NULL;
4568 count_skb_priorities(const struct xport *xport)
4570 return hmap_count(&xport->skb_priorities);
4574 clear_skb_priorities(struct xport *xport)
4576 struct skb_priority_to_dscp *pdscp, *next;
4578 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4579 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4585 actions_output_to_local_port(const struct xlate_ctx *ctx)
4587 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4588 const struct nlattr *a;
4591 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions->data,
4592 ctx->xout->odp_actions->size) {
4593 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4594 && nl_attr_get_odp_port(a) == local_odp_port) {
4601 #if defined(__linux__)
4602 /* Returns the maximum number of packets that the Linux kernel is willing to
4603 * queue up internally to certain kinds of software-implemented ports, or the
4604 * default (and rarely modified) value if it cannot be determined. */
4606 netdev_max_backlog(void)
4608 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4609 static int max_backlog = 1000; /* The normal default value. */
4611 if (ovsthread_once_start(&once)) {
4612 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4616 stream = fopen(filename, "r");
4618 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4620 if (fscanf(stream, "%d", &n) != 1) {
4621 VLOG_WARN("%s: read error", filename);
4622 } else if (n <= 100) {
4623 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4629 ovsthread_once_done(&once);
4631 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4637 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4640 count_output_actions(const struct ofpbuf *odp_actions)
4642 const struct nlattr *a;
4646 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4647 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4653 #endif /* defined(__linux__) */
4655 /* Returns true if 'odp_actions' contains more output actions than the datapath
4656 * can reliably handle in one go. On Linux, this is the value of the
4657 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4658 * packets that the kernel is willing to queue up for processing while the
4659 * datapath is processing a set of actions. */
4661 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4664 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4665 && count_output_actions(odp_actions) > netdev_max_backlog());
4667 /* OSes other than Linux might have similar limits, but we don't know how
4668 * to determine them.*/
4673 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
4675 * The caller must take responsibility for eventually freeing 'xout', with
4676 * xlate_out_uninit(). */
4678 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4680 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4681 struct flow_wildcards *wc = NULL;
4682 struct flow *flow = &xin->flow;
4683 struct rule_dpif *rule = NULL;
4685 enum slow_path_reason special;
4686 const struct ofpact *ofpacts;
4687 struct xbridge *xbridge;
4688 struct xport *in_port;
4689 struct flow orig_flow;
4690 struct xlate_ctx ctx;
4695 COVERAGE_INC(xlate_actions);
4697 /* Flow initialization rules:
4698 * - 'base_flow' must match the kernel's view of the packet at the
4699 * time that action processing starts. 'flow' represents any
4700 * transformations we wish to make through actions.
4701 * - By default 'base_flow' and 'flow' are the same since the input
4702 * packet matches the output before any actions are applied.
4703 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4704 * of the received packet as seen by the kernel. If we later output
4705 * to another device without any modifications this will cause us to
4706 * insert a new tag since the original one was stripped off by the
4708 * - Tunnel metadata as received is retained in 'flow'. This allows
4709 * tunnel metadata matching also in later tables.
4710 * Since a kernel action for setting the tunnel metadata will only be
4711 * generated with actual tunnel output, changing the tunnel metadata
4712 * values in 'flow' (such as tun_id) will only have effect with a later
4713 * tunnel output action.
4714 * - Tunnel 'base_flow' is completely cleared since that is what the
4715 * kernel does. If we wish to maintain the original values an action
4716 * needs to be generated. */
4721 ctx.xout->has_learn = false;
4722 ctx.xout->has_normal = false;
4723 ctx.xout->has_fin_timeout = false;
4724 ctx.xout->nf_output_iface = NF_OUT_DROP;
4725 ctx.xout->mirrors = 0;
4726 ctx.xout->n_recircs = 0;
4728 xout->odp_actions = xin->odp_actions;
4729 if (!xout->odp_actions) {
4730 xout->odp_actions = &xout->odp_actions_buf;
4731 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4732 sizeof xout->odp_actions_stub);
4734 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4736 xbridge = xbridge_lookup(xcfg, xin->ofproto);
4740 /* 'ctx.xbridge' may be changed by action processing, whereas 'xbridge'
4741 * will remain set on the original input bridge. */
4742 ctx.xbridge = xbridge;
4743 ctx.rule = xin->rule;
4745 ctx.base_flow = *flow;
4746 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4747 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4749 if (!xin->skip_wildcards) {
4751 flow_wildcards_init_catchall(wc);
4752 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4753 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4754 if (is_ip_any(flow)) {
4755 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4757 if (xbridge->support.recirc) {
4758 /* Always exactly match recirc_id when datapath supports
4760 wc->masks.recirc_id = UINT32_MAX;
4762 if (xbridge->netflow) {
4763 netflow_mask_wc(flow, wc);
4766 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4768 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4772 ctx.in_group = false;
4773 ctx.in_action_set = false;
4774 ctx.orig_skb_priority = flow->skb_priority;
4776 ctx.rule_cookie = OVS_BE64_MAX;
4778 ctx.was_mpls = false;
4779 ctx.recirc_action_offset = -1;
4780 ctx.last_unroll_offset = -1;
4782 ctx.action_set_has_group = false;
4783 ofpbuf_use_stub(&ctx.action_set,
4784 ctx.action_set_stub, sizeof ctx.action_set_stub);
4786 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4788 /* The in_port of the original packet before recirculation. */
4789 in_port = get_ofp_port(xbridge, flow->in_port.ofp_port);
4792 const struct recirc_id_node *recirc = xin->recirc;
4794 if (xin->ofpacts_len > 0 || ctx.rule) {
4795 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4797 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!",
4798 xin->ofpacts_len > 0
4804 /* Set the bridge for post-recirculation processing if needed. */
4805 if (ctx.xbridge->ofproto != recirc->ofproto) {
4806 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4807 const struct xbridge *new_bridge = xbridge_lookup(xcfg,
4810 if (OVS_UNLIKELY(!new_bridge)) {
4811 /* Drop the packet if the bridge cannot be found. */
4812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4813 VLOG_WARN_RL(&rl, "Recirculation bridge no longer exists.");
4816 ctx.xbridge = new_bridge;
4819 /* Set the post-recirculation table id. Note: A table lookup is done
4820 * only if there are no post-recirculation actions. */
4821 ctx.table_id = recirc->table_id;
4823 /* Restore pipeline metadata. May change flow's in_port and other
4824 * metadata to the values that existed when recirculation was
4826 recirc_metadata_to_flow(&recirc->metadata, flow);
4828 /* Restore stack, if any. */
4829 if (recirc->stack) {
4830 ofpbuf_put(&ctx.stack, recirc->stack->data, recirc->stack->size);
4833 /* Restore action set, if any. */
4834 if (recirc->action_set_len) {
4835 const struct ofpact *a;
4837 ofpbuf_put(&ctx.action_set, recirc->ofpacts,
4838 recirc->action_set_len);
4840 OFPACT_FOR_EACH(a, recirc->ofpacts, recirc->action_set_len) {
4841 if (a->type == OFPACT_GROUP) {
4842 ctx.action_set_has_group = true;
4848 /* Restore recirculation actions. If there are no actions, processing
4849 * will start with a lookup in the table set above. */
4850 if (recirc->ofpacts_len > recirc->action_set_len) {
4851 xin->ofpacts_len = recirc->ofpacts_len - recirc->action_set_len;
4852 xin->ofpacts = recirc->ofpacts +
4853 recirc->action_set_len / sizeof *recirc->ofpacts;
4855 } else if (OVS_UNLIKELY(flow->recirc_id)) {
4856 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4858 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
4862 /* The bridge is now known so obtain its table version. */
4863 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
4865 if (!xin->ofpacts && !ctx.rule) {
4866 rule = rule_dpif_lookup_from_table(ctx.xbridge->ofproto,
4867 ctx.tables_version, flow, wc,
4868 ctx.xin->xcache != NULL,
4869 ctx.xin->resubmit_stats,
4871 flow->in_port.ofp_port, true, true);
4872 if (ctx.xin->resubmit_stats) {
4873 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
4875 if (ctx.xin->xcache) {
4876 struct xc_entry *entry;
4878 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4879 entry->u.rule = rule;
4883 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
4884 ctx.xin->resubmit_hook(ctx.xin, rule, 0);
4887 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4890 ofpacts = xin->ofpacts;
4891 ofpacts_len = xin->ofpacts_len;
4892 } else if (ctx.rule) {
4893 const struct rule_actions *actions = rule_dpif_get_actions(ctx.rule);
4895 ofpacts = actions->ofpacts;
4896 ofpacts_len = actions->ofpacts_len;
4898 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
4903 if (mbridge_has_mirrors(xbridge->mbridge)) {
4904 /* Do this conditionally because the copy is expensive enough that it
4905 * shows up in profiles. */
4909 /* Tunnel stats only for non-recirculated packets. */
4910 if (!xin->recirc && in_port && in_port->is_tunnel) {
4911 if (ctx.xin->resubmit_stats) {
4912 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4914 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4917 if (ctx.xin->xcache) {
4918 struct xc_entry *entry;
4920 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4921 entry->u.dev.rx = netdev_ref(in_port->netdev);
4922 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4926 /* Do not perform special processing on recirculated packets,
4927 * as recirculated packets are not really received by the bridge. */
4929 (special = process_special(&ctx, flow, in_port, ctx.xin->packet))) {
4930 ctx.xout->slow |= special;
4932 size_t sample_actions_len;
4934 if (flow->in_port.ofp_port
4935 != vsp_realdev_to_vlandev(xbridge->ofproto,
4936 flow->in_port.ofp_port,
4938 ctx.base_flow.vlan_tci = 0;
4941 /* Sampling is done only for packets really received by the bridge. */
4943 add_sflow_action(&ctx);
4944 add_ipfix_action(&ctx);
4945 sample_actions_len = ctx.xout->odp_actions->size;
4947 sample_actions_len = 0;
4950 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4951 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4953 /* We've let OFPP_NORMAL and the learning action look at the
4954 * packet, so drop it now if forwarding is disabled. */
4955 if (in_port && (!xport_stp_forward_state(in_port) ||
4956 !xport_rstp_forward_state(in_port))) {
4957 /* Drop all actions added by do_xlate_actions() above. */
4958 ctx.xout->odp_actions->size = sample_actions_len;
4960 /* Undo changes that may have been done for recirculation. */
4961 if (exit_recirculates(&ctx)) {
4962 ctx.action_set.size = ctx.recirc_action_offset;
4963 ctx.recirc_action_offset = -1;
4964 ctx.last_unroll_offset = -1;
4966 } else if (ctx.action_set.size) {
4967 /* Translate action set only if not dropping the packet and
4968 * not recirculating. */
4969 if (!exit_recirculates(&ctx)) {
4970 xlate_action_set(&ctx);
4973 /* Check if need to recirculate. */
4974 if (exit_recirculates(&ctx)) {
4975 compose_recirculate_action(&ctx);
4979 /* Output only fully processed packets. */
4980 if (!exit_recirculates(&ctx)
4981 && xbridge->has_in_band
4982 && in_band_must_output_to_local_port(flow)
4983 && !actions_output_to_local_port(&ctx)) {
4984 compose_output_action(&ctx, OFPP_LOCAL, NULL);
4988 fix_sflow_action(&ctx);
4990 /* Only mirror fully processed packets. */
4991 if (!exit_recirculates(&ctx)
4992 && mbridge_has_mirrors(xbridge->mbridge)) {
4993 add_mirror_actions(&ctx, &orig_flow);
4997 if (nl_attr_oversized(ctx.xout->odp_actions->size)) {
4998 /* These datapath actions are too big for a Netlink attribute, so we
4999 * can't hand them to the kernel directly. dpif_execute() can execute
5000 * them one by one with help, so just mark the result as SLOW_ACTION to
5001 * prevent the flow from being installed. */
5002 COVERAGE_INC(xlate_actions_oversize);
5003 ctx.xout->slow |= SLOW_ACTION;
5004 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
5005 COVERAGE_INC(xlate_actions_too_many_output);
5006 ctx.xout->slow |= SLOW_ACTION;
5009 /* Update mirror stats only for packets really received by the bridge. */
5010 if (!xin->recirc && mbridge_has_mirrors(xbridge->mbridge)) {
5011 if (ctx.xin->resubmit_stats) {
5012 mirror_update_stats(xbridge->mbridge, xout->mirrors,
5013 ctx.xin->resubmit_stats->n_packets,
5014 ctx.xin->resubmit_stats->n_bytes);
5016 if (ctx.xin->xcache) {
5017 struct xc_entry *entry;
5019 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
5020 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
5021 entry->u.mirror.mirrors = xout->mirrors;
5025 /* Do netflow only for packets really received by the bridge. */
5026 if (!xin->recirc && xbridge->netflow) {
5027 /* Only update netflow if we don't have controller flow. We don't
5028 * report NetFlow expiration messages for such facets because they
5029 * are just part of the control logic for the network, not real
5031 if (ofpacts_len == 0
5032 || ofpacts->type != OFPACT_CONTROLLER
5033 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
5034 if (ctx.xin->resubmit_stats) {
5035 netflow_flow_update(xbridge->netflow, flow,
5036 xout->nf_output_iface,
5037 ctx.xin->resubmit_stats);
5039 if (ctx.xin->xcache) {
5040 struct xc_entry *entry;
5042 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5043 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5044 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5045 entry->u.nf.iface = xout->nf_output_iface;
5050 ofpbuf_uninit(&ctx.stack);
5051 ofpbuf_uninit(&ctx.action_set);
5054 /* Clear the metadata and register wildcard masks, because we won't
5055 * use non-header fields as part of the cache. */
5056 flow_wildcards_clear_non_packet_fields(wc);
5058 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
5059 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
5060 * represent these fields. The datapath interface, on the other hand,
5061 * represents them with just 8 bits each. This means that if the high
5062 * 8 bits of the masks for these fields somehow become set, then they
5063 * will get chopped off by a round trip through the datapath, and
5064 * revalidation will spot that as an inconsistency and delete the flow.
5065 * Avoid the problem here by making sure that only the low 8 bits of
5066 * either field can be unwildcarded for ICMP.
5069 wc->masks.tp_src &= htons(UINT8_MAX);
5070 wc->masks.tp_dst &= htons(UINT8_MAX);
5072 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
5073 if (wc->masks.vlan_tci) {
5074 wc->masks.vlan_tci |= htons(VLAN_CFI);
5079 /* Sends 'packet' out 'ofport'.
5080 * May modify 'packet'.
5081 * Returns 0 if successful, otherwise a positive errno value. */
5083 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5085 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5086 struct xport *xport;
5087 struct ofpact_output output;
5090 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5091 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5092 flow_extract(packet, &flow);
5093 flow.in_port.ofp_port = OFPP_NONE;
5095 xport = xport_lookup(xcfg, ofport);
5099 output.port = xport->ofp_port;
5102 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5103 &output.ofpact, sizeof output,
5107 struct xlate_cache *
5108 xlate_cache_new(void)
5110 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5112 ofpbuf_init(&xcache->entries, 512);
5116 static struct xc_entry *
5117 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5119 struct xc_entry *entry;
5121 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5128 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5130 if (entry->u.dev.tx) {
5131 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5133 if (entry->u.dev.rx) {
5134 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5136 if (entry->u.dev.bfd) {
5137 bfd_account_rx(entry->u.dev.bfd, stats);
5142 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5144 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5145 struct xbridge *xbridge;
5146 struct xbundle *xbundle;
5147 struct flow_wildcards wc;
5149 xbridge = xbridge_lookup(xcfg, ofproto);
5154 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5160 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5163 /* Push stats and perform side effects of flow translation. */
5165 xlate_push_stats(struct xlate_cache *xcache,
5166 const struct dpif_flow_stats *stats)
5168 struct xc_entry *entry;
5169 struct ofpbuf entries = xcache->entries;
5170 uint8_t dmac[ETH_ADDR_LEN];
5172 if (!stats->n_packets) {
5176 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5177 switch (entry->type) {
5179 rule_dpif_credit_stats(entry->u.rule, stats);
5182 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5183 entry->u.bond.vid, stats->n_bytes);
5186 xlate_cache_netdev(entry, stats);
5189 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5190 entry->u.nf.iface, stats);
5193 mirror_update_stats(entry->u.mirror.mbridge,
5194 entry->u.mirror.mirrors,
5195 stats->n_packets, stats->n_bytes);
5198 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5201 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5202 entry->u.normal.vlan);
5204 case XC_FIN_TIMEOUT:
5205 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5206 entry->u.fin.idle, entry->u.fin.hard);
5209 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5213 /* Lookup arp to avoid arp timeout. */
5214 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
5223 xlate_dev_unref(struct xc_entry *entry)
5225 if (entry->u.dev.tx) {
5226 netdev_close(entry->u.dev.tx);
5228 if (entry->u.dev.rx) {
5229 netdev_close(entry->u.dev.rx);
5231 if (entry->u.dev.bfd) {
5232 bfd_unref(entry->u.dev.bfd);
5237 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5239 netflow_flow_clear(netflow, flow);
5240 netflow_unref(netflow);
5245 xlate_cache_clear(struct xlate_cache *xcache)
5247 struct xc_entry *entry;
5248 struct ofpbuf entries;
5254 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5255 switch (entry->type) {
5257 rule_dpif_unref(entry->u.rule);
5260 free(entry->u.bond.flow);
5261 bond_unref(entry->u.bond.bond);
5264 xlate_dev_unref(entry);
5267 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5270 mbridge_unref(entry->u.mirror.mbridge);
5273 free(entry->u.learn.fm);
5274 ofpbuf_delete(entry->u.learn.ofpacts);
5277 free(entry->u.normal.flow);
5279 case XC_FIN_TIMEOUT:
5280 /* 'u.fin.rule' is always already held as a XC_RULE, which
5281 * has already released it's reference above. */
5284 group_dpif_unref(entry->u.group.group);
5293 ofpbuf_clear(&xcache->entries);
5297 xlate_cache_delete(struct xlate_cache *xcache)
5299 xlate_cache_clear(xcache);
5300 ofpbuf_uninit(&xcache->entries);