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 /* True if the datapath supports recirculation. */
102 /* True if the datapath supports variable-length
103 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
104 * False if the datapath supports only 8-byte (or shorter) userdata. */
105 bool variable_length_userdata;
107 /* Number of MPLS label stack entries that the datapath supports
109 size_t max_mpls_depth;
111 /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
113 bool masked_set_action;
117 struct hmap_node hmap_node; /* In global 'xbundles' map. */
118 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
120 struct ovs_list list_node; /* In parent 'xbridges' list. */
121 struct xbridge *xbridge; /* Parent xbridge. */
123 struct ovs_list xports; /* Contains "struct xport"s. */
125 char *name; /* Name used in log messages. */
126 struct bond *bond; /* Nonnull iff more than one port. */
127 struct lacp *lacp; /* LACP handle or null. */
129 enum port_vlan_mode vlan_mode; /* VLAN mode. */
130 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
131 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
132 * NULL if all VLANs are trunked. */
133 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
134 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
138 struct hmap_node hmap_node; /* Node in global 'xports' map. */
139 struct ofport_dpif *ofport; /* Key in global 'xports map. */
141 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
142 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
144 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
146 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
147 struct xbundle *xbundle; /* Parent xbundle or null. */
149 struct netdev *netdev; /* 'ofport''s netdev. */
151 struct xbridge *xbridge; /* Parent bridge. */
152 struct xport *peer; /* Patch port peer or null. */
154 enum ofputil_port_config config; /* OpenFlow port configuration. */
155 enum ofputil_port_state state; /* OpenFlow port state. */
156 int stp_port_no; /* STP port number or -1 if not in use. */
157 struct rstp_port *rstp_port; /* RSTP port or null. */
159 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
161 bool may_enable; /* May be enabled in bonds. */
162 bool is_tunnel; /* Is a tunnel port. */
164 struct cfm *cfm; /* CFM handle or null. */
165 struct bfd *bfd; /* BFD handle or null. */
166 struct lldp *lldp; /* LLDP handle or null. */
170 struct xlate_in *xin;
171 struct xlate_out *xout;
173 const struct xbridge *xbridge;
175 /* Flow at the last commit. */
176 struct flow base_flow;
178 /* Tunnel IP destination address as received. This is stored separately
179 * as the base_flow.tunnel is cleared on init to reflect the datapath
180 * behavior. Used to make sure not to send tunneled output to ourselves,
181 * which might lead to an infinite loop. This could happen easily
182 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
183 * actually set the tun_dst field. */
184 ovs_be32 orig_tunnel_ip_dst;
186 /* Stack for the push and pop actions. Each stack element is of type
187 * "union mf_subvalue". */
188 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
191 /* The rule that we are currently translating, or NULL. */
192 struct rule_dpif *rule;
194 /* Resubmit statistics, via xlate_table_action(). */
195 int recurse; /* Current resubmit nesting depth. */
196 int resubmits; /* Total number of resubmits. */
197 bool in_group; /* Currently translating ofgroup, if true. */
198 bool in_action_set; /* Currently translating action_set, if true. */
200 uint8_t table_id; /* OpenFlow table ID where flow was found. */
201 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
202 uint32_t orig_skb_priority; /* Priority when packet arrived. */
203 uint32_t sflow_n_outputs; /* Number of output ports. */
204 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
205 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
206 bool exit; /* No further actions should be processed. */
208 /* These are used for non-bond recirculation. The recirculation IDs are
209 * stored in xout and must be associated with a datapath flow (ukey),
210 * otherwise they will be freed when the xout is uninitialized.
213 * Steps in Recirculation Translation
214 * ==================================
216 * At some point during translation, the code recognizes the need for
217 * recirculation. For example, recirculation is necessary when, after
218 * popping the last MPLS label, an action or a match tries to examine or
219 * modify a field that has been newly revealed following the MPLS label.
221 * The simplest part of the work to be done is to commit existing changes to
222 * the packet, which produces datapath actions corresponding to the changes,
223 * and after this, add an OVS_ACTION_ATTR_RECIRC datapath action.
225 * The main problem here is preserving state. When the datapath executes
226 * OVS_ACTION_ATTR_RECIRC, it will upcall to userspace to get a translation
227 * for the post-recirculation actions. At this point userspace has to
228 * resume the translation where it left off, which means that it has to
229 * execute the following:
231 * - The action that prompted recirculation, and any actions following
232 * it within the same flow.
234 * - If the action that prompted recirculation was invoked within a
235 * NXAST_RESUBMIT, then any actions following the resubmit. These
236 * "resubmit"s can be nested, so this has to go all the way up the
239 * - The OpenFlow 1.1+ action set.
241 * State that actions and flow table lookups can depend on, such as the
242 * following, must also be preserved:
244 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
246 * - Action set, stack
248 * - The table ID and cookie of the flow being translated at each level
249 * of the control stack (since OFPAT_CONTROLLER actions send these to
252 * Translation allows for the control of this state preservation via these
253 * members. When a need for recirculation is identified, the translation
256 * 1. Sets 'recirc_action_offset' to the current size of 'action_set'. The
257 * action set is part of what needs to be preserved, so this allows the
258 * action set and the additional state to share the 'action_set' buffer.
259 * Later steps can tell that setup for recirculation is in progress from
260 * the nonnegative value of 'recirc_action_offset'.
262 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
263 * translation process.
265 * 3. Adds an OFPACT_UNROLL_XLATE action to 'action_set'. This action
266 * holds the current table ID and cookie so that they can be restored
267 * during a post-recirculation upcall translation.
269 * 4. Adds the action that prompted recirculation and any actions following
270 * it within the same flow to 'action_set', so that they can be executed
271 * during a post-recirculation upcall translation.
275 * 6. The action that prompted recirculation might be nested in a stack of
276 * nested "resubmit"s that have actions remaining. Each of these notices
277 * that we're exiting (from 'exit') and that recirculation setup is in
278 * progress (from 'recirc_action_offset') and responds by adding more
279 * OFPACT_UNROLL_XLATE actions to 'action_set', as necessary, and any
280 * actions that were yet unprocessed.
282 * The caller stores all the state produced by this process associated with
283 * the recirculation ID. For post-recirculation upcall translation, the
284 * caller passes it back in for the new translation to execute. The
285 * process yielded a set of ofpacts that can be translated directly, so it
286 * is not much of a special case at that point.
288 int recirc_action_offset; /* Offset in 'action_set' to actions to be
289 * executed after recirculation, or -1. */
290 int last_unroll_offset; /* Offset in 'action_set' to the latest unroll
293 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
294 * This is a trigger for recirculation in cases where translating an action
295 * or looking up a flow requires access to the fields of the packet after
296 * the MPLS label stack that was originally present.
298 * XXX: output to a table and patch port do not currently recirculate even
299 * if this is true. */
302 /* OpenFlow 1.1+ action set.
304 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
305 * When translation is otherwise complete, ofpacts_execute_action_set()
306 * converts it to a set of "struct ofpact"s that can be translated into
307 * datapath actions. */
308 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
309 struct ofpbuf action_set; /* Action set. */
310 uint64_t action_set_stub[1024 / 8];
313 static void xlate_action_set(struct xlate_ctx *ctx);
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 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
441 uint8_t table_id, bool may_packet_in,
442 bool honor_table_miss);
443 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
444 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
445 static void output_normal(struct xlate_ctx *, const struct xbundle *,
448 /* Optional bond recirculation parameter to compose_output_action(). */
449 struct xlate_bond_recirc {
450 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
451 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
452 uint32_t hash_basis; /* Compute hash for recirc before. */
455 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
456 const struct xlate_bond_recirc *xr);
458 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
459 const struct ofproto_dpif *);
460 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
461 const struct ofbundle *);
462 static struct xport *xport_lookup(struct xlate_cfg *,
463 const struct ofport_dpif *);
464 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
465 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
466 uint32_t skb_priority);
467 static void clear_skb_priorities(struct xport *);
468 static size_t count_skb_priorities(const struct xport *);
469 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
472 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
474 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
475 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
476 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
477 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
478 const struct mac_learning *, struct stp *,
479 struct rstp *, const struct mcast_snooping *,
480 const struct mbridge *,
481 const struct dpif_sflow *,
482 const struct dpif_ipfix *,
483 const struct netflow *,
484 bool forward_bpdu, bool has_in_band,
486 bool variable_length_userdata,
487 size_t max_mpls_depth,
488 bool masked_set_action);
489 static void xlate_xbundle_set(struct xbundle *xbundle,
490 enum port_vlan_mode vlan_mode, int vlan,
491 unsigned long *trunks, bool use_priority_tags,
492 const struct bond *bond, const struct lacp *lacp,
494 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
495 const struct netdev *netdev, const struct cfm *cfm,
496 const struct bfd *bfd, const struct lldp *lldp,
497 int stp_port_no, const struct rstp_port *rstp_port,
498 enum ofputil_port_config config,
499 enum ofputil_port_state state, bool is_tunnel,
501 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
502 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
503 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
504 static void xlate_xbridge_copy(struct xbridge *);
505 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
506 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
508 static void xlate_xcfg_free(struct xlate_cfg *);
511 xlate_report(struct xlate_ctx *ctx, const char *s)
513 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
514 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
519 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
521 list_init(&xbridge->xbundles);
522 hmap_init(&xbridge->xports);
523 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
524 hash_pointer(xbridge->ofproto, 0));
528 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
530 list_init(&xbundle->xports);
531 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
532 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
533 hash_pointer(xbundle->ofbundle, 0));
537 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
539 hmap_init(&xport->skb_priorities);
540 hmap_insert(&xcfg->xports, &xport->hmap_node,
541 hash_pointer(xport->ofport, 0));
542 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
543 hash_ofp_port(xport->ofp_port));
547 xlate_xbridge_set(struct xbridge *xbridge,
549 const struct mac_learning *ml, struct stp *stp,
550 struct rstp *rstp, const struct mcast_snooping *ms,
551 const struct mbridge *mbridge,
552 const struct dpif_sflow *sflow,
553 const struct dpif_ipfix *ipfix,
554 const struct netflow *netflow,
555 bool forward_bpdu, bool has_in_band,
557 bool variable_length_userdata,
558 size_t max_mpls_depth,
559 bool masked_set_action)
561 if (xbridge->ml != ml) {
562 mac_learning_unref(xbridge->ml);
563 xbridge->ml = mac_learning_ref(ml);
566 if (xbridge->ms != ms) {
567 mcast_snooping_unref(xbridge->ms);
568 xbridge->ms = mcast_snooping_ref(ms);
571 if (xbridge->mbridge != mbridge) {
572 mbridge_unref(xbridge->mbridge);
573 xbridge->mbridge = mbridge_ref(mbridge);
576 if (xbridge->sflow != sflow) {
577 dpif_sflow_unref(xbridge->sflow);
578 xbridge->sflow = dpif_sflow_ref(sflow);
581 if (xbridge->ipfix != ipfix) {
582 dpif_ipfix_unref(xbridge->ipfix);
583 xbridge->ipfix = dpif_ipfix_ref(ipfix);
586 if (xbridge->stp != stp) {
587 stp_unref(xbridge->stp);
588 xbridge->stp = stp_ref(stp);
591 if (xbridge->rstp != rstp) {
592 rstp_unref(xbridge->rstp);
593 xbridge->rstp = rstp_ref(rstp);
596 if (xbridge->netflow != netflow) {
597 netflow_unref(xbridge->netflow);
598 xbridge->netflow = netflow_ref(netflow);
601 xbridge->dpif = dpif;
602 xbridge->forward_bpdu = forward_bpdu;
603 xbridge->has_in_band = has_in_band;
604 xbridge->enable_recirc = enable_recirc;
605 xbridge->variable_length_userdata = variable_length_userdata;
606 xbridge->max_mpls_depth = max_mpls_depth;
607 xbridge->masked_set_action = masked_set_action;
611 xlate_xbundle_set(struct xbundle *xbundle,
612 enum port_vlan_mode vlan_mode, int vlan,
613 unsigned long *trunks, bool use_priority_tags,
614 const struct bond *bond, const struct lacp *lacp,
617 ovs_assert(xbundle->xbridge);
619 xbundle->vlan_mode = vlan_mode;
620 xbundle->vlan = vlan;
621 xbundle->trunks = trunks;
622 xbundle->use_priority_tags = use_priority_tags;
623 xbundle->floodable = floodable;
625 if (xbundle->bond != bond) {
626 bond_unref(xbundle->bond);
627 xbundle->bond = bond_ref(bond);
630 if (xbundle->lacp != lacp) {
631 lacp_unref(xbundle->lacp);
632 xbundle->lacp = lacp_ref(lacp);
637 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
638 const struct netdev *netdev, const struct cfm *cfm,
639 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
640 const struct rstp_port* rstp_port,
641 enum ofputil_port_config config, enum ofputil_port_state state,
642 bool is_tunnel, bool may_enable)
644 xport->config = config;
645 xport->state = state;
646 xport->stp_port_no = stp_port_no;
647 xport->is_tunnel = is_tunnel;
648 xport->may_enable = may_enable;
649 xport->odp_port = odp_port;
651 if (xport->rstp_port != rstp_port) {
652 rstp_port_unref(xport->rstp_port);
653 xport->rstp_port = rstp_port_ref(rstp_port);
656 if (xport->cfm != cfm) {
657 cfm_unref(xport->cfm);
658 xport->cfm = cfm_ref(cfm);
661 if (xport->bfd != bfd) {
662 bfd_unref(xport->bfd);
663 xport->bfd = bfd_ref(bfd);
666 if (xport->lldp != lldp) {
667 lldp_unref(xport->lldp);
668 xport->lldp = lldp_ref(lldp);
671 if (xport->netdev != netdev) {
672 netdev_close(xport->netdev);
673 xport->netdev = netdev_ref(netdev);
678 xlate_xbridge_copy(struct xbridge *xbridge)
680 struct xbundle *xbundle;
682 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
683 new_xbridge->ofproto = xbridge->ofproto;
684 new_xbridge->name = xstrdup(xbridge->name);
685 xlate_xbridge_init(new_xcfg, new_xbridge);
687 xlate_xbridge_set(new_xbridge,
688 xbridge->dpif, xbridge->ml, xbridge->stp,
689 xbridge->rstp, xbridge->ms, xbridge->mbridge,
690 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
691 xbridge->forward_bpdu,
692 xbridge->has_in_band, xbridge->enable_recirc,
693 xbridge->variable_length_userdata,
694 xbridge->max_mpls_depth, xbridge->masked_set_action);
695 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
696 xlate_xbundle_copy(new_xbridge, xbundle);
699 /* Copy xports which are not part of a xbundle */
700 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
701 if (!xport->xbundle) {
702 xlate_xport_copy(new_xbridge, NULL, xport);
708 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
711 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
712 new_xbundle->ofbundle = xbundle->ofbundle;
713 new_xbundle->xbridge = xbridge;
714 new_xbundle->name = xstrdup(xbundle->name);
715 xlate_xbundle_init(new_xcfg, new_xbundle);
717 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
718 xbundle->vlan, xbundle->trunks,
719 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
721 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
722 xlate_xport_copy(xbridge, new_xbundle, xport);
727 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
730 struct skb_priority_to_dscp *pdscp, *new_pdscp;
731 struct xport *new_xport = xzalloc(sizeof *xport);
732 new_xport->ofport = xport->ofport;
733 new_xport->ofp_port = xport->ofp_port;
734 new_xport->xbridge = xbridge;
735 xlate_xport_init(new_xcfg, new_xport);
737 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
738 xport->bfd, xport->lldp, xport->stp_port_no,
739 xport->rstp_port, xport->config, xport->state,
740 xport->is_tunnel, xport->may_enable);
743 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
745 new_xport->peer = peer;
746 new_xport->peer->peer = new_xport;
751 new_xport->xbundle = xbundle;
752 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
755 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
756 new_pdscp = xmalloc(sizeof *pdscp);
757 new_pdscp->skb_priority = pdscp->skb_priority;
758 new_pdscp->dscp = pdscp->dscp;
759 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
760 hash_int(new_pdscp->skb_priority, 0));
764 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
765 * configuration in xcfgp.
767 * This needs to be called after editing the xlate configuration.
769 * Functions that edit the new xlate configuration are
770 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
776 * edit_xlate_configuration();
778 * xlate_txn_commit(); */
780 xlate_txn_commit(void)
782 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
784 ovsrcu_set(&xcfgp, new_xcfg);
785 ovsrcu_synchronize();
786 xlate_xcfg_free(xcfg);
790 /* Copies the current xlate configuration in xcfgp to new_xcfg.
792 * This needs to be called prior to editing the xlate configuration. */
794 xlate_txn_start(void)
796 struct xbridge *xbridge;
797 struct xlate_cfg *xcfg;
799 ovs_assert(!new_xcfg);
801 new_xcfg = xmalloc(sizeof *new_xcfg);
802 hmap_init(&new_xcfg->xbridges);
803 hmap_init(&new_xcfg->xbundles);
804 hmap_init(&new_xcfg->xports);
806 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
811 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
812 xlate_xbridge_copy(xbridge);
818 xlate_xcfg_free(struct xlate_cfg *xcfg)
820 struct xbridge *xbridge, *next_xbridge;
826 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
827 xlate_xbridge_remove(xcfg, xbridge);
830 hmap_destroy(&xcfg->xbridges);
831 hmap_destroy(&xcfg->xbundles);
832 hmap_destroy(&xcfg->xports);
837 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
839 const struct mac_learning *ml, struct stp *stp,
840 struct rstp *rstp, const struct mcast_snooping *ms,
841 const struct mbridge *mbridge,
842 const struct dpif_sflow *sflow,
843 const struct dpif_ipfix *ipfix,
844 const struct netflow *netflow,
845 bool forward_bpdu, bool has_in_band, bool enable_recirc,
846 bool variable_length_userdata, size_t max_mpls_depth,
847 bool masked_set_action)
849 struct xbridge *xbridge;
851 ovs_assert(new_xcfg);
853 xbridge = xbridge_lookup(new_xcfg, ofproto);
855 xbridge = xzalloc(sizeof *xbridge);
856 xbridge->ofproto = ofproto;
858 xlate_xbridge_init(new_xcfg, xbridge);
862 xbridge->name = xstrdup(name);
864 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
865 netflow, forward_bpdu, has_in_band, enable_recirc,
866 variable_length_userdata, max_mpls_depth,
871 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
873 struct xbundle *xbundle, *next_xbundle;
874 struct xport *xport, *next_xport;
880 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
881 xlate_xport_remove(xcfg, xport);
884 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
885 xlate_xbundle_remove(xcfg, xbundle);
888 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
889 mac_learning_unref(xbridge->ml);
890 mcast_snooping_unref(xbridge->ms);
891 mbridge_unref(xbridge->mbridge);
892 dpif_sflow_unref(xbridge->sflow);
893 dpif_ipfix_unref(xbridge->ipfix);
894 stp_unref(xbridge->stp);
895 rstp_unref(xbridge->rstp);
896 hmap_destroy(&xbridge->xports);
902 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
904 struct xbridge *xbridge;
906 ovs_assert(new_xcfg);
908 xbridge = xbridge_lookup(new_xcfg, ofproto);
909 xlate_xbridge_remove(new_xcfg, xbridge);
913 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
914 const char *name, enum port_vlan_mode vlan_mode, int vlan,
915 unsigned long *trunks, bool use_priority_tags,
916 const struct bond *bond, const struct lacp *lacp,
919 struct xbundle *xbundle;
921 ovs_assert(new_xcfg);
923 xbundle = xbundle_lookup(new_xcfg, ofbundle);
925 xbundle = xzalloc(sizeof *xbundle);
926 xbundle->ofbundle = ofbundle;
927 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
929 xlate_xbundle_init(new_xcfg, xbundle);
933 xbundle->name = xstrdup(name);
935 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
936 use_priority_tags, bond, lacp, floodable);
940 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
942 struct xport *xport, *next;
948 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
949 list_remove(&xport->bundle_node);
950 xport->xbundle = NULL;
953 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
954 list_remove(&xbundle->list_node);
955 bond_unref(xbundle->bond);
956 lacp_unref(xbundle->lacp);
962 xlate_bundle_remove(struct ofbundle *ofbundle)
964 struct xbundle *xbundle;
966 ovs_assert(new_xcfg);
968 xbundle = xbundle_lookup(new_xcfg, ofbundle);
969 xlate_xbundle_remove(new_xcfg, xbundle);
973 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
974 struct ofport_dpif *ofport, ofp_port_t ofp_port,
975 odp_port_t odp_port, const struct netdev *netdev,
976 const struct cfm *cfm, const struct bfd *bfd,
977 const struct lldp *lldp, struct ofport_dpif *peer,
978 int stp_port_no, const struct rstp_port *rstp_port,
979 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
980 enum ofputil_port_config config,
981 enum ofputil_port_state state, bool is_tunnel,
987 ovs_assert(new_xcfg);
989 xport = xport_lookup(new_xcfg, ofport);
991 xport = xzalloc(sizeof *xport);
992 xport->ofport = ofport;
993 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
994 xport->ofp_port = ofp_port;
996 xlate_xport_init(new_xcfg, xport);
999 ovs_assert(xport->ofp_port == ofp_port);
1001 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1002 stp_port_no, rstp_port, config, state, is_tunnel,
1006 xport->peer->peer = NULL;
1008 xport->peer = xport_lookup(new_xcfg, peer);
1010 xport->peer->peer = xport;
1013 if (xport->xbundle) {
1014 list_remove(&xport->bundle_node);
1016 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1017 if (xport->xbundle) {
1018 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1021 clear_skb_priorities(xport);
1022 for (i = 0; i < n_qdscp; i++) {
1023 struct skb_priority_to_dscp *pdscp;
1024 uint32_t skb_priority;
1026 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1031 pdscp = xmalloc(sizeof *pdscp);
1032 pdscp->skb_priority = skb_priority;
1033 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1034 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1035 hash_int(pdscp->skb_priority, 0));
1040 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1047 xport->peer->peer = NULL;
1051 if (xport->xbundle) {
1052 list_remove(&xport->bundle_node);
1055 clear_skb_priorities(xport);
1056 hmap_destroy(&xport->skb_priorities);
1058 hmap_remove(&xcfg->xports, &xport->hmap_node);
1059 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1061 netdev_close(xport->netdev);
1062 rstp_port_unref(xport->rstp_port);
1063 cfm_unref(xport->cfm);
1064 bfd_unref(xport->bfd);
1065 lldp_unref(xport->lldp);
1070 xlate_ofport_remove(struct ofport_dpif *ofport)
1072 struct xport *xport;
1074 ovs_assert(new_xcfg);
1076 xport = xport_lookup(new_xcfg, ofport);
1077 xlate_xport_remove(new_xcfg, xport);
1080 static struct ofproto_dpif *
1081 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1082 ofp_port_t *ofp_in_port, const struct xport **xportp)
1084 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1085 const struct xport *xport;
1087 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1088 ? tnl_port_receive(flow)
1089 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1090 if (OVS_UNLIKELY(!xport)) {
1095 *ofp_in_port = xport->ofp_port;
1097 return xport->xbridge->ofproto;
1100 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1101 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1102 struct ofproto_dpif *
1103 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1104 ofp_port_t *ofp_in_port)
1106 const struct xport *xport;
1108 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1111 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1112 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1113 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1114 * handles for those protocols if they're enabled. Caller may use the returned
1115 * pointers until quiescing, for longer term use additional references must
1118 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1121 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1122 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1123 struct dpif_sflow **sflow, struct netflow **netflow,
1124 ofp_port_t *ofp_in_port)
1126 struct ofproto_dpif *ofproto;
1127 const struct xport *xport;
1129 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1136 *ofprotop = ofproto;
1140 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1144 *sflow = xport ? xport->xbridge->sflow : NULL;
1148 *netflow = xport ? xport->xbridge->netflow : NULL;
1154 static struct xbridge *
1155 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1157 struct hmap *xbridges;
1158 struct xbridge *xbridge;
1160 if (!ofproto || !xcfg) {
1164 xbridges = &xcfg->xbridges;
1166 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1168 if (xbridge->ofproto == ofproto) {
1175 static struct xbundle *
1176 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1178 struct hmap *xbundles;
1179 struct xbundle *xbundle;
1181 if (!ofbundle || !xcfg) {
1185 xbundles = &xcfg->xbundles;
1187 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1189 if (xbundle->ofbundle == ofbundle) {
1196 static struct xport *
1197 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1199 struct hmap *xports;
1200 struct xport *xport;
1202 if (!ofport || !xcfg) {
1206 xports = &xcfg->xports;
1208 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1210 if (xport->ofport == ofport) {
1217 static struct stp_port *
1218 xport_get_stp_port(const struct xport *xport)
1220 return xport->xbridge->stp && xport->stp_port_no != -1
1221 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1226 xport_stp_learn_state(const struct xport *xport)
1228 struct stp_port *sp = xport_get_stp_port(xport);
1230 ? stp_learn_in_state(stp_port_get_state(sp))
1235 xport_stp_forward_state(const struct xport *xport)
1237 struct stp_port *sp = xport_get_stp_port(xport);
1239 ? stp_forward_in_state(stp_port_get_state(sp))
1244 xport_stp_should_forward_bpdu(const struct xport *xport)
1246 struct stp_port *sp = xport_get_stp_port(xport);
1247 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1250 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1251 * were used to make the determination.*/
1253 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1255 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1256 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1257 return is_stp(flow);
1261 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1263 struct stp_port *sp = xport_get_stp_port(xport);
1264 struct dp_packet payload = *packet;
1265 struct eth_header *eth = dp_packet_data(&payload);
1267 /* Sink packets on ports that have STP disabled when the bridge has
1269 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1273 /* Trim off padding on payload. */
1274 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1275 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1278 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1279 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1283 static enum rstp_state
1284 xport_get_rstp_port_state(const struct xport *xport)
1286 return xport->rstp_port
1287 ? rstp_port_get_state(xport->rstp_port)
1292 xport_rstp_learn_state(const struct xport *xport)
1294 return xport->xbridge->rstp && xport->rstp_port
1295 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1300 xport_rstp_forward_state(const struct xport *xport)
1302 return xport->xbridge->rstp && xport->rstp_port
1303 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1308 xport_rstp_should_manage_bpdu(const struct xport *xport)
1310 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1314 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1316 struct dp_packet payload = *packet;
1317 struct eth_header *eth = dp_packet_data(&payload);
1319 /* Sink packets on ports that have no RSTP. */
1320 if (!xport->rstp_port) {
1324 /* Trim off padding on payload. */
1325 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1326 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1329 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1330 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1331 dp_packet_size(&payload));
1335 static struct xport *
1336 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1338 struct xport *xport;
1340 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1342 if (xport->ofp_port == ofp_port) {
1350 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1352 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1353 return xport ? xport->odp_port : ODPP_NONE;
1357 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1359 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1360 return xport && xport->may_enable;
1363 static struct ofputil_bucket *
1364 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1368 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1370 struct group_dpif *group;
1372 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1373 struct ofputil_bucket *bucket;
1375 bucket = group_first_live_bucket(ctx, group, depth);
1376 group_dpif_unref(group);
1377 return bucket == NULL;
1383 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1386 bucket_is_alive(const struct xlate_ctx *ctx,
1387 struct ofputil_bucket *bucket, int depth)
1389 if (depth >= MAX_LIVENESS_RECURSION) {
1390 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1392 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1393 MAX_LIVENESS_RECURSION);
1397 return (!ofputil_bucket_has_liveness(bucket)
1398 || (bucket->watch_port != OFPP_ANY
1399 && odp_port_is_alive(ctx, bucket->watch_port))
1400 || (bucket->watch_group != OFPG_ANY
1401 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1404 static struct ofputil_bucket *
1405 group_first_live_bucket(const struct xlate_ctx *ctx,
1406 const struct group_dpif *group, int depth)
1408 struct ofputil_bucket *bucket;
1409 const struct ovs_list *buckets;
1411 group_dpif_get_buckets(group, &buckets);
1412 LIST_FOR_EACH (bucket, list_node, buckets) {
1413 if (bucket_is_alive(ctx, bucket, depth)) {
1421 static struct ofputil_bucket *
1422 group_best_live_bucket(const struct xlate_ctx *ctx,
1423 const struct group_dpif *group,
1426 struct ofputil_bucket *best_bucket = NULL;
1427 uint32_t best_score = 0;
1430 struct ofputil_bucket *bucket;
1431 const struct ovs_list *buckets;
1433 group_dpif_get_buckets(group, &buckets);
1434 LIST_FOR_EACH (bucket, list_node, buckets) {
1435 if (bucket_is_alive(ctx, bucket, 0)) {
1436 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1437 if (score >= best_score) {
1438 best_bucket = bucket;
1449 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1451 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1452 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1456 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1458 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1461 static mirror_mask_t
1462 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1464 return xbundle != &ofpp_none_bundle
1465 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1469 static mirror_mask_t
1470 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1472 return xbundle != &ofpp_none_bundle
1473 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1477 static mirror_mask_t
1478 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1480 return xbundle != &ofpp_none_bundle
1481 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1485 static struct xbundle *
1486 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1487 bool warn, struct xport **in_xportp)
1489 struct xport *xport;
1491 /* Find the port and bundle for the received packet. */
1492 xport = get_ofp_port(xbridge, in_port);
1496 if (xport && xport->xbundle) {
1497 return xport->xbundle;
1500 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1501 * which a controller may use as the ingress port for traffic that
1502 * it is sourcing. */
1503 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1504 return &ofpp_none_bundle;
1507 /* Odd. A few possible reasons here:
1509 * - We deleted a port but there are still a few packets queued up
1512 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1513 * we don't know about.
1515 * - The ofproto client didn't configure the port as part of a bundle.
1516 * This is particularly likely to happen if a packet was received on the
1517 * port after it was created, but before the client had a chance to
1518 * configure its bundle.
1521 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1523 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1524 "port %"PRIu16, xbridge->name, in_port);
1530 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1532 const struct xbridge *xbridge = ctx->xbridge;
1533 mirror_mask_t mirrors;
1534 struct xbundle *in_xbundle;
1538 mirrors = ctx->xout->mirrors;
1539 ctx->xout->mirrors = 0;
1541 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1542 ctx->xin->packet != NULL, NULL);
1546 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1548 /* Drop frames on bundles reserved for mirroring. */
1549 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1550 if (ctx->xin->packet != NULL) {
1551 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1552 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1553 "%s, which is reserved exclusively for mirroring",
1554 ctx->xbridge->name, in_xbundle->name);
1556 ofpbuf_clear(ctx->xout->odp_actions);
1561 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1562 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1565 vlan = input_vid_to_vlan(in_xbundle, vid);
1571 /* Restore the original packet before adding the mirror actions. */
1572 ctx->xin->flow = *orig_flow;
1575 mirror_mask_t dup_mirrors;
1576 struct ofbundle *out;
1577 unsigned long *vlans;
1582 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1583 &vlans, &dup_mirrors, &out, &out_vlan);
1584 ovs_assert(has_mirror);
1587 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1589 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1592 if (!vlan_mirrored) {
1593 mirrors = zero_rightmost_1bit(mirrors);
1597 mirrors &= ~dup_mirrors;
1598 ctx->xout->mirrors |= dup_mirrors;
1600 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1601 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1603 output_normal(ctx, out_xbundle, vlan);
1605 } else if (vlan != out_vlan
1606 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1607 struct xbundle *xbundle;
1609 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1610 if (xbundle_includes_vlan(xbundle, out_vlan)
1611 && !xbundle_mirror_out(xbridge, xbundle)) {
1612 output_normal(ctx, xbundle, out_vlan);
1619 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1620 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1621 * the bundle on which the packet was received, returns the VLAN to which the
1624 * Both 'vid' and the return value are in the range 0...4095. */
1626 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1628 switch (in_xbundle->vlan_mode) {
1629 case PORT_VLAN_ACCESS:
1630 return in_xbundle->vlan;
1633 case PORT_VLAN_TRUNK:
1636 case PORT_VLAN_NATIVE_UNTAGGED:
1637 case PORT_VLAN_NATIVE_TAGGED:
1638 return vid ? vid : in_xbundle->vlan;
1645 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1646 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1649 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1650 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1653 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1655 /* Allow any VID on the OFPP_NONE port. */
1656 if (in_xbundle == &ofpp_none_bundle) {
1660 switch (in_xbundle->vlan_mode) {
1661 case PORT_VLAN_ACCESS:
1664 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1665 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1666 "packet received on port %s configured as VLAN "
1667 "%"PRIu16" access port", vid, in_xbundle->name,
1674 case PORT_VLAN_NATIVE_UNTAGGED:
1675 case PORT_VLAN_NATIVE_TAGGED:
1677 /* Port must always carry its native VLAN. */
1681 case PORT_VLAN_TRUNK:
1682 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1685 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1686 "received on port %s not configured for trunking "
1687 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1699 /* Given 'vlan', the VLAN that a packet belongs to, and
1700 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1701 * that should be included in the 802.1Q header. (If the return value is 0,
1702 * then the 802.1Q header should only be included in the packet if there is a
1705 * Both 'vlan' and the return value are in the range 0...4095. */
1707 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1709 switch (out_xbundle->vlan_mode) {
1710 case PORT_VLAN_ACCESS:
1713 case PORT_VLAN_TRUNK:
1714 case PORT_VLAN_NATIVE_TAGGED:
1717 case PORT_VLAN_NATIVE_UNTAGGED:
1718 return vlan == out_xbundle->vlan ? 0 : vlan;
1726 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1729 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1731 ovs_be16 tci, old_tci;
1732 struct xport *xport;
1733 struct xlate_bond_recirc xr;
1734 bool use_recirc = false;
1736 vid = output_vlan_to_vid(out_xbundle, vlan);
1737 if (list_is_empty(&out_xbundle->xports)) {
1738 /* Partially configured bundle with no slaves. Drop the packet. */
1740 } else if (!out_xbundle->bond) {
1741 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1744 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1745 struct flow_wildcards *wc = &ctx->xout->wc;
1746 struct ofport_dpif *ofport;
1748 if (ctx->xbridge->enable_recirc) {
1749 use_recirc = bond_may_recirc(
1750 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1753 /* Only TCP mode uses recirculation. */
1754 xr.hash_alg = OVS_HASH_ALG_L4;
1755 bond_update_post_recirc_rules(out_xbundle->bond, false);
1757 /* Recirculation does not require unmasking hash fields. */
1762 ofport = bond_choose_output_slave(out_xbundle->bond,
1763 &ctx->xin->flow, wc, vid);
1764 xport = xport_lookup(xcfg, ofport);
1767 /* No slaves enabled, so drop packet. */
1771 /* If use_recirc is set, the main thread will handle stats
1772 * accounting for this bond. */
1774 if (ctx->xin->resubmit_stats) {
1775 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1776 ctx->xin->resubmit_stats->n_bytes);
1778 if (ctx->xin->xcache) {
1779 struct xc_entry *entry;
1782 flow = &ctx->xin->flow;
1783 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1784 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1785 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1786 entry->u.bond.vid = vid;
1791 old_tci = *flow_tci;
1793 if (tci || out_xbundle->use_priority_tags) {
1794 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1796 tci |= htons(VLAN_CFI);
1801 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1802 *flow_tci = old_tci;
1805 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1806 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1807 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1809 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1811 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1815 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1816 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1820 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1821 if (flow->nw_proto == ARP_OP_REPLY) {
1823 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1824 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1825 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1827 return flow->nw_src == flow->nw_dst;
1833 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1834 * dropped. Returns true if they may be forwarded, false if they should be
1837 * 'in_port' must be the xport that corresponds to flow->in_port.
1838 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1840 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1841 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1842 * checked by input_vid_is_valid().
1844 * May also add tags to '*tags', although the current implementation only does
1845 * so in one special case.
1848 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1851 struct xbundle *in_xbundle = in_port->xbundle;
1852 const struct xbridge *xbridge = ctx->xbridge;
1853 struct flow *flow = &ctx->xin->flow;
1855 /* Drop frames for reserved multicast addresses
1856 * only if forward_bpdu option is absent. */
1857 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1858 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1862 if (in_xbundle->bond) {
1863 struct mac_entry *mac;
1865 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1871 xlate_report(ctx, "bonding refused admissibility, dropping");
1874 case BV_DROP_IF_MOVED:
1875 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1876 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1878 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1879 && (!is_gratuitous_arp(flow, &ctx->xout->wc)
1880 || mac_entry_is_grat_arp_locked(mac))) {
1881 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1882 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1886 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1894 /* Checks whether a MAC learning update is necessary for MAC learning table
1895 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1898 * Most packets processed through the MAC learning table do not actually
1899 * change it in any way. This function requires only a read lock on the MAC
1900 * learning table, so it is much cheaper in this common case.
1902 * Keep the code here synchronized with that in update_learning_table__()
1905 is_mac_learning_update_needed(const struct mac_learning *ml,
1906 const struct flow *flow,
1907 struct flow_wildcards *wc,
1908 int vlan, struct xbundle *in_xbundle)
1909 OVS_REQ_RDLOCK(ml->rwlock)
1911 struct mac_entry *mac;
1913 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1917 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1918 if (!mac || mac_entry_age(ml, mac)) {
1922 if (is_gratuitous_arp(flow, wc)) {
1923 /* We don't want to learn from gratuitous ARP packets that are
1924 * reflected back over bond slaves so we lock the learning table. */
1925 if (!in_xbundle->bond) {
1927 } else if (mac_entry_is_grat_arp_locked(mac)) {
1932 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1936 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1937 * received on 'in_xbundle' in 'vlan'.
1939 * This code repeats all the checks in is_mac_learning_update_needed() because
1940 * the lock was released between there and here and thus the MAC learning state
1941 * could have changed.
1943 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1946 update_learning_table__(const struct xbridge *xbridge,
1947 const struct flow *flow, struct flow_wildcards *wc,
1948 int vlan, struct xbundle *in_xbundle)
1949 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1951 struct mac_entry *mac;
1953 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1957 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1958 if (is_gratuitous_arp(flow, wc)) {
1959 /* We don't want to learn from gratuitous ARP packets that are
1960 * reflected back over bond slaves so we lock the learning table. */
1961 if (!in_xbundle->bond) {
1962 mac_entry_set_grat_arp_lock(mac);
1963 } else if (mac_entry_is_grat_arp_locked(mac)) {
1968 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1969 /* The log messages here could actually be useful in debugging,
1970 * so keep the rate limit relatively high. */
1971 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1973 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1974 "on port %s in VLAN %d",
1975 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1976 in_xbundle->name, vlan);
1978 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
1983 update_learning_table(const struct xbridge *xbridge,
1984 const struct flow *flow, struct flow_wildcards *wc,
1985 int vlan, struct xbundle *in_xbundle)
1989 /* Don't learn the OFPP_NONE port. */
1990 if (in_xbundle == &ofpp_none_bundle) {
1994 /* First try the common case: no change to MAC learning table. */
1995 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1996 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1998 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2001 /* Slow path: MAC learning table might need an update. */
2002 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2003 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2004 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2008 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2009 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2011 update_mcast_snooping_table__(const struct xbridge *xbridge,
2012 const struct flow *flow,
2013 struct mcast_snooping *ms,
2014 ovs_be32 ip4, int vlan,
2015 struct xbundle *in_xbundle)
2016 OVS_REQ_WRLOCK(ms->rwlock)
2018 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2020 switch (ntohs(flow->tp_src)) {
2021 case IGMP_HOST_MEMBERSHIP_REPORT:
2022 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2023 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
2024 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2025 IP_FMT" is on port %s in VLAN %d",
2026 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2029 case IGMP_HOST_LEAVE_MESSAGE:
2030 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
2031 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2032 IP_FMT" is on port %s in VLAN %d",
2033 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2036 case IGMP_HOST_MEMBERSHIP_QUERY:
2037 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2038 in_xbundle->ofbundle)) {
2039 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2040 IP_FMT" is on port %s in VLAN %d",
2041 xbridge->name, IP_ARGS(flow->nw_src),
2042 in_xbundle->name, vlan);
2048 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2049 * was received on 'in_xbundle' in 'vlan'. */
2051 update_mcast_snooping_table(const struct xbridge *xbridge,
2052 const struct flow *flow, int vlan,
2053 struct xbundle *in_xbundle)
2055 struct mcast_snooping *ms = xbridge->ms;
2056 struct xlate_cfg *xcfg;
2057 struct xbundle *mcast_xbundle;
2058 struct mcast_port_bundle *fport;
2060 /* Don't learn the OFPP_NONE port. */
2061 if (in_xbundle == &ofpp_none_bundle) {
2065 /* Don't learn from flood ports */
2066 mcast_xbundle = NULL;
2067 ovs_rwlock_wrlock(&ms->rwlock);
2068 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2069 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2070 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2071 if (mcast_xbundle == in_xbundle) {
2076 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2077 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
2080 ovs_rwlock_unlock(&ms->rwlock);
2083 /* send the packet to ports having the multicast group learned */
2085 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2086 struct mcast_snooping *ms OVS_UNUSED,
2087 struct mcast_group *grp,
2088 struct xbundle *in_xbundle, uint16_t vlan)
2089 OVS_REQ_RDLOCK(ms->rwlock)
2091 struct xlate_cfg *xcfg;
2092 struct mcast_group_bundle *b;
2093 struct xbundle *mcast_xbundle;
2095 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2096 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2097 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2098 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2099 xlate_report(ctx, "forwarding to mcast group port");
2100 output_normal(ctx, mcast_xbundle, vlan);
2101 } else if (!mcast_xbundle) {
2102 xlate_report(ctx, "mcast group port is unknown, dropping");
2104 xlate_report(ctx, "mcast group port is input port, dropping");
2109 /* send the packet to ports connected to multicast routers */
2111 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2112 struct mcast_snooping *ms,
2113 struct xbundle *in_xbundle, uint16_t vlan)
2114 OVS_REQ_RDLOCK(ms->rwlock)
2116 struct xlate_cfg *xcfg;
2117 struct mcast_mrouter_bundle *mrouter;
2118 struct xbundle *mcast_xbundle;
2120 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2121 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2122 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2123 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2124 xlate_report(ctx, "forwarding to mcast router port");
2125 output_normal(ctx, mcast_xbundle, vlan);
2126 } else if (!mcast_xbundle) {
2127 xlate_report(ctx, "mcast router port is unknown, dropping");
2129 xlate_report(ctx, "mcast router port is input port, dropping");
2134 /* send the packet to ports flagged to be flooded */
2136 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2137 struct mcast_snooping *ms,
2138 struct xbundle *in_xbundle, uint16_t vlan)
2139 OVS_REQ_RDLOCK(ms->rwlock)
2141 struct xlate_cfg *xcfg;
2142 struct mcast_port_bundle *fport;
2143 struct xbundle *mcast_xbundle;
2145 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2146 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2147 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2148 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2149 xlate_report(ctx, "forwarding to mcast flood port");
2150 output_normal(ctx, mcast_xbundle, vlan);
2151 } else if (!mcast_xbundle) {
2152 xlate_report(ctx, "mcast flood port is unknown, dropping");
2154 xlate_report(ctx, "mcast flood port is input port, dropping");
2159 /* forward the Reports to configured ports */
2161 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2162 struct mcast_snooping *ms,
2163 struct xbundle *in_xbundle, uint16_t vlan)
2164 OVS_REQ_RDLOCK(ms->rwlock)
2166 struct xlate_cfg *xcfg;
2167 struct mcast_port_bundle *rport;
2168 struct xbundle *mcast_xbundle;
2170 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2171 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2172 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2173 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2174 xlate_report(ctx, "forwarding Report to mcast flagged port");
2175 output_normal(ctx, mcast_xbundle, vlan);
2176 } else if (!mcast_xbundle) {
2177 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2179 xlate_report(ctx, "mcast port is input port, dropping the Report");
2185 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2188 struct xbundle *xbundle;
2190 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2191 if (xbundle != in_xbundle
2192 && xbundle_includes_vlan(xbundle, vlan)
2193 && xbundle->floodable
2194 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2195 output_normal(ctx, xbundle, vlan);
2198 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2202 xlate_normal(struct xlate_ctx *ctx)
2204 struct flow_wildcards *wc = &ctx->xout->wc;
2205 struct flow *flow = &ctx->xin->flow;
2206 struct xbundle *in_xbundle;
2207 struct xport *in_port;
2208 struct mac_entry *mac;
2213 ctx->xout->has_normal = true;
2215 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2216 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2217 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2219 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2220 ctx->xin->packet != NULL, &in_port);
2222 xlate_report(ctx, "no input bundle, dropping");
2226 /* Drop malformed frames. */
2227 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2228 !(flow->vlan_tci & htons(VLAN_CFI))) {
2229 if (ctx->xin->packet != NULL) {
2230 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2231 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2232 "VLAN tag received on port %s",
2233 ctx->xbridge->name, in_xbundle->name);
2235 xlate_report(ctx, "partial VLAN tag, dropping");
2239 /* Drop frames on bundles reserved for mirroring. */
2240 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2241 if (ctx->xin->packet != NULL) {
2242 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2243 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2244 "%s, which is reserved exclusively for mirroring",
2245 ctx->xbridge->name, in_xbundle->name);
2247 xlate_report(ctx, "input port is mirror output port, dropping");
2252 vid = vlan_tci_to_vid(flow->vlan_tci);
2253 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2254 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2257 vlan = input_vid_to_vlan(in_xbundle, vid);
2259 /* Check other admissibility requirements. */
2260 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2264 /* Learn source MAC. */
2265 if (ctx->xin->may_learn) {
2266 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2268 if (ctx->xin->xcache) {
2269 struct xc_entry *entry;
2271 /* Save enough info to update mac learning table later. */
2272 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2273 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2274 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2275 entry->u.normal.vlan = vlan;
2278 /* Determine output bundle. */
2279 if (mcast_snooping_enabled(ctx->xbridge->ms)
2280 && !eth_addr_is_broadcast(flow->dl_dst)
2281 && eth_addr_is_multicast(flow->dl_dst)
2282 && flow->dl_type == htons(ETH_TYPE_IP)) {
2283 struct mcast_snooping *ms = ctx->xbridge->ms;
2284 struct mcast_group *grp;
2286 if (flow->nw_proto == IPPROTO_IGMP) {
2287 if (ctx->xin->may_learn) {
2288 if (mcast_snooping_is_membership(flow->tp_src) ||
2289 mcast_snooping_is_query(flow->tp_src)) {
2290 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2295 if (mcast_snooping_is_membership(flow->tp_src)) {
2296 ovs_rwlock_rdlock(&ms->rwlock);
2297 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2298 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2299 * forward IGMP Membership Reports only to those ports where
2300 * multicast routers are attached. Alternatively stated: a
2301 * snooping switch should not forward IGMP Membership Reports
2302 * to ports on which only hosts are attached.
2303 * An administrative control may be provided to override this
2304 * restriction, allowing the report messages to be flooded to
2306 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2307 ovs_rwlock_unlock(&ms->rwlock);
2309 xlate_report(ctx, "multicast traffic, flooding");
2310 xlate_normal_flood(ctx, in_xbundle, vlan);
2314 if (ip_is_local_multicast(flow->nw_dst)) {
2315 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2316 * address in the 224.0.0.x range which are not IGMP must
2317 * be forwarded on all ports */
2318 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2319 xlate_normal_flood(ctx, in_xbundle, vlan);
2324 /* forwarding to group base ports */
2325 ovs_rwlock_rdlock(&ms->rwlock);
2326 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2328 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2329 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2330 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2332 if (mcast_snooping_flood_unreg(ms)) {
2333 xlate_report(ctx, "unregistered multicast, flooding");
2334 xlate_normal_flood(ctx, in_xbundle, vlan);
2336 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2337 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2340 ovs_rwlock_unlock(&ms->rwlock);
2342 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2343 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2344 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2345 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2348 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2349 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2350 if (mac_xbundle && mac_xbundle != in_xbundle) {
2351 xlate_report(ctx, "forwarding to learned port");
2352 output_normal(ctx, mac_xbundle, vlan);
2353 } else if (!mac_xbundle) {
2354 xlate_report(ctx, "learned port is unknown, dropping");
2356 xlate_report(ctx, "learned port is input port, dropping");
2359 xlate_report(ctx, "no learned MAC for destination, flooding");
2360 xlate_normal_flood(ctx, in_xbundle, vlan);
2365 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2366 * the number of packets out of UINT32_MAX to sample. The given
2367 * cookie is passed back in the callback for each sampled packet.
2370 compose_sample_action(const struct xbridge *xbridge,
2371 struct ofpbuf *odp_actions,
2372 const struct flow *flow,
2373 const uint32_t probability,
2374 const union user_action_cookie *cookie,
2375 const size_t cookie_size,
2376 const odp_port_t tunnel_out_port)
2378 size_t sample_offset, actions_offset;
2379 odp_port_t odp_port;
2383 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2385 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2387 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2389 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2390 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2391 flow_hash_5tuple(flow, 0));
2392 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2393 tunnel_out_port, odp_actions);
2395 nl_msg_end_nested(odp_actions, actions_offset);
2396 nl_msg_end_nested(odp_actions, sample_offset);
2397 return cookie_offset;
2401 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2402 odp_port_t odp_port, unsigned int n_outputs,
2403 union user_action_cookie *cookie)
2407 cookie->type = USER_ACTION_COOKIE_SFLOW;
2408 cookie->sflow.vlan_tci = vlan_tci;
2410 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2411 * port information") for the interpretation of cookie->output. */
2412 switch (n_outputs) {
2414 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2415 cookie->sflow.output = 0x40000000 | 256;
2419 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2421 cookie->sflow.output = ifindex;
2426 /* 0x80000000 means "multiple output ports. */
2427 cookie->sflow.output = 0x80000000 | n_outputs;
2432 /* Compose SAMPLE action for sFlow bridge sampling. */
2434 compose_sflow_action(const struct xbridge *xbridge,
2435 struct ofpbuf *odp_actions,
2436 const struct flow *flow,
2437 odp_port_t odp_port)
2439 uint32_t probability;
2440 union user_action_cookie cookie;
2442 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2446 probability = dpif_sflow_get_probability(xbridge->sflow);
2447 compose_sflow_cookie(xbridge, htons(0), odp_port,
2448 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2450 return compose_sample_action(xbridge, odp_actions, flow, probability,
2451 &cookie, sizeof cookie.sflow, ODPP_NONE);
2455 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2456 uint32_t obs_domain_id, uint32_t obs_point_id,
2457 union user_action_cookie *cookie)
2459 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2460 cookie->flow_sample.probability = probability;
2461 cookie->flow_sample.collector_set_id = collector_set_id;
2462 cookie->flow_sample.obs_domain_id = obs_domain_id;
2463 cookie->flow_sample.obs_point_id = obs_point_id;
2467 compose_ipfix_cookie(union user_action_cookie *cookie,
2468 odp_port_t output_odp_port)
2470 cookie->type = USER_ACTION_COOKIE_IPFIX;
2471 cookie->ipfix.output_odp_port = output_odp_port;
2474 /* Compose SAMPLE action for IPFIX bridge sampling. */
2476 compose_ipfix_action(const struct xbridge *xbridge,
2477 struct ofpbuf *odp_actions,
2478 const struct flow *flow,
2479 odp_port_t output_odp_port)
2481 uint32_t probability;
2482 union user_action_cookie cookie;
2483 odp_port_t tunnel_out_port = ODPP_NONE;
2485 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2489 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2491 if (output_odp_port == ODPP_NONE &&
2492 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2496 /* For output case, output_odp_port is valid*/
2497 if (output_odp_port != ODPP_NONE) {
2498 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2501 /* If tunnel sampling is enabled, put an additional option attribute:
2502 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2504 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2505 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2506 tunnel_out_port = output_odp_port;
2510 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2511 compose_ipfix_cookie(&cookie, output_odp_port);
2513 compose_sample_action(xbridge, odp_actions, flow, probability,
2514 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2517 /* SAMPLE action for sFlow must be first action in any given list of
2518 * actions. At this point we do not have all information required to
2519 * build it. So try to build sample action as complete as possible. */
2521 add_sflow_action(struct xlate_ctx *ctx)
2523 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2524 ctx->xout->odp_actions,
2525 &ctx->xin->flow, ODPP_NONE);
2526 ctx->sflow_odp_port = 0;
2527 ctx->sflow_n_outputs = 0;
2530 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2531 * of actions, eventually after the SAMPLE action for sFlow. */
2533 add_ipfix_action(struct xlate_ctx *ctx)
2535 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2536 &ctx->xin->flow, ODPP_NONE);
2540 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2542 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2543 &ctx->xin->flow, port);
2546 /* Fix SAMPLE action according to data collected while composing ODP actions.
2547 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2548 * USERSPACE action's user-cookie which is required for sflow. */
2550 fix_sflow_action(struct xlate_ctx *ctx)
2552 const struct flow *base = &ctx->base_flow;
2553 union user_action_cookie *cookie;
2555 if (!ctx->user_cookie_offset) {
2559 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2560 sizeof cookie->sflow);
2561 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2563 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2564 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2567 static enum slow_path_reason
2568 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2569 const struct xport *xport, const struct dp_packet *packet)
2571 struct flow_wildcards *wc = &ctx->xout->wc;
2572 const struct xbridge *xbridge = ctx->xbridge;
2576 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2578 cfm_process_heartbeat(xport->cfm, packet);
2581 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2583 bfd_process_packet(xport->bfd, flow, packet);
2584 /* If POLL received, immediately sends FINAL back. */
2585 if (bfd_should_send_packet(xport->bfd)) {
2586 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2590 } else if (xport->xbundle && xport->xbundle->lacp
2591 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2593 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2596 } else if ((xbridge->stp || xbridge->rstp) &&
2597 stp_should_process_flow(flow, wc)) {
2600 ? stp_process_packet(xport, packet)
2601 : rstp_process_packet(xport, packet);
2604 } else if (xport->lldp && lldp_should_process_flow(flow)) {
2606 lldp_process_packet(xport->lldp, packet);
2615 tnl_route_lookup_flow(const struct flow *oflow,
2616 ovs_be32 *ip, struct xport **out_port)
2618 char out_dev[IFNAMSIZ];
2619 struct xbridge *xbridge;
2620 struct xlate_cfg *xcfg;
2623 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2630 *ip = oflow->tunnel.ip_dst;
2633 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2636 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2637 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2640 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2641 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2652 xlate_flood_packet(struct xbridge *xbridge, struct dp_packet *packet)
2654 struct ofpact_output output;
2657 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2658 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2659 flow_extract(packet, &flow);
2660 flow.in_port.ofp_port = OFPP_NONE;
2661 output.port = OFPP_FLOOD;
2664 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2665 &output.ofpact, sizeof output,
2670 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2671 ovs_be32 ip_src, ovs_be32 ip_dst)
2673 struct xbridge *xbridge = out_dev->xbridge;
2674 struct dp_packet packet;
2676 dp_packet_init(&packet, 0);
2677 compose_arp(&packet, eth_src, ip_src, ip_dst);
2679 xlate_flood_packet(xbridge, &packet);
2680 dp_packet_uninit(&packet);
2684 build_tunnel_send(const struct xlate_ctx *ctx, const struct xport *xport,
2685 const struct flow *flow, odp_port_t tunnel_odp_port)
2687 struct ovs_action_push_tnl tnl_push_data;
2688 struct xport *out_dev = NULL;
2689 ovs_be32 s_ip, d_ip = 0;
2690 uint8_t smac[ETH_ADDR_LEN];
2691 uint8_t dmac[ETH_ADDR_LEN];
2694 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2699 /* Use mac addr of bridge port of the peer. */
2700 err = netdev_get_etheraddr(out_dev->netdev, smac);
2705 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2710 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2712 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2715 if (ctx->xin->xcache) {
2716 struct xc_entry *entry;
2718 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2719 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2720 sizeof entry->u.tnl_arp_cache.br_name);
2721 entry->u.tnl_arp_cache.d_ip = d_ip;
2723 err = tnl_port_build_header(xport->ofport, flow,
2724 dmac, smac, s_ip, &tnl_push_data);
2728 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2729 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2730 odp_put_tnl_push_action(ctx->xout->odp_actions, &tnl_push_data);
2735 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2736 const struct xlate_bond_recirc *xr, bool check_stp)
2738 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2739 struct flow_wildcards *wc = &ctx->xout->wc;
2740 struct flow *flow = &ctx->xin->flow;
2741 struct flow_tnl flow_tnl;
2742 ovs_be16 flow_vlan_tci;
2743 uint32_t flow_pkt_mark;
2744 uint8_t flow_nw_tos;
2745 odp_port_t out_port, odp_port;
2746 bool tnl_push_pop_send = false;
2749 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2750 * before traversing a patch port. */
2751 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 31);
2752 memset(&flow_tnl, 0, sizeof flow_tnl);
2755 xlate_report(ctx, "Nonexistent output port");
2757 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2758 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2760 } else if (check_stp) {
2761 if (is_stp(&ctx->base_flow)) {
2762 if (!xport_stp_should_forward_bpdu(xport) &&
2763 !xport_rstp_should_manage_bpdu(xport)) {
2764 if (ctx->xbridge->stp != NULL) {
2765 xlate_report(ctx, "STP not in listening state, "
2766 "skipping bpdu output");
2767 } else if (ctx->xbridge->rstp != NULL) {
2768 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2769 "skipping bpdu output");
2773 } else if (!xport_stp_forward_state(xport) ||
2774 !xport_rstp_forward_state(xport)) {
2775 if (ctx->xbridge->stp != NULL) {
2776 xlate_report(ctx, "STP not in forwarding state, "
2778 } else if (ctx->xbridge->rstp != NULL) {
2779 xlate_report(ctx, "RSTP not in forwarding state, "
2786 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2787 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2792 const struct xport *peer = xport->peer;
2793 struct flow old_flow = ctx->xin->flow;
2794 bool old_was_mpls = ctx->was_mpls;
2795 enum slow_path_reason special;
2796 struct ofpbuf old_stack = ctx->stack;
2797 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2798 struct ofpbuf old_action_set = ctx->action_set;
2799 uint64_t actset_stub[1024 / 8];
2801 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2802 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2803 ctx->xbridge = peer->xbridge;
2804 flow->in_port.ofp_port = peer->ofp_port;
2805 flow->metadata = htonll(0);
2806 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2807 memset(flow->regs, 0, sizeof flow->regs);
2808 flow->actset_output = OFPP_UNSET;
2810 special = process_special(ctx, &ctx->xin->flow, peer,
2813 ctx->xout->slow |= special;
2814 } else if (may_receive(peer, ctx)) {
2815 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2816 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2817 if (ctx->action_set.size) {
2818 /* Translate action set only if not dropping the packet and
2819 * not recirculating. */
2820 if (!exit_recirculates(ctx)) {
2821 xlate_action_set(ctx);
2824 /* Check if need to recirculate. */
2825 if (exit_recirculates(ctx)) {
2826 compose_recirculate_action(ctx);
2829 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2830 * the learning action look at the packet, then drop it. */
2831 struct flow old_base_flow = ctx->base_flow;
2832 size_t old_size = ctx->xout->odp_actions->size;
2833 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2835 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2836 ctx->xout->mirrors = old_mirrors;
2837 ctx->base_flow = old_base_flow;
2838 ctx->xout->odp_actions->size = old_size;
2840 /* Undo changes that may have been done for recirculation. */
2841 if (exit_recirculates(ctx)) {
2842 ctx->action_set.size = ctx->recirc_action_offset;
2843 ctx->recirc_action_offset = -1;
2844 ctx->last_unroll_offset = -1;
2849 ctx->xin->flow = old_flow;
2850 ctx->xbridge = xport->xbridge;
2851 ofpbuf_uninit(&ctx->action_set);
2852 ctx->action_set = old_action_set;
2853 ofpbuf_uninit(&ctx->stack);
2854 ctx->stack = old_stack;
2856 /* The peer bridge popping MPLS should have no effect on the original
2858 ctx->was_mpls = old_was_mpls;
2860 /* The fact that the peer bridge exits (for any reason) does not mean
2861 * that the original bridge should exit. Specifically, if the peer
2862 * bridge recirculates (which typically modifies the packet), the
2863 * original bridge must continue processing with the original, not the
2864 * recirculated packet! */
2867 if (ctx->xin->resubmit_stats) {
2868 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2869 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2871 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2874 if (ctx->xin->xcache) {
2875 struct xc_entry *entry;
2877 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2878 entry->u.dev.tx = netdev_ref(xport->netdev);
2879 entry->u.dev.rx = netdev_ref(peer->netdev);
2880 entry->u.dev.bfd = bfd_ref(peer->bfd);
2885 flow_vlan_tci = flow->vlan_tci;
2886 flow_pkt_mark = flow->pkt_mark;
2887 flow_nw_tos = flow->nw_tos;
2889 if (count_skb_priorities(xport)) {
2890 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2891 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2892 wc->masks.nw_tos |= IP_DSCP_MASK;
2893 flow->nw_tos &= ~IP_DSCP_MASK;
2894 flow->nw_tos |= dscp;
2898 if (xport->is_tunnel) {
2899 /* Save tunnel metadata so that changes made due to
2900 * the Logical (tunnel) Port are not visible for any further
2901 * matches, while explicit set actions on tunnel metadata are.
2903 flow_tnl = flow->tunnel;
2904 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2905 if (odp_port == ODPP_NONE) {
2906 xlate_report(ctx, "Tunneling decided against output");
2907 goto out; /* restore flow_nw_tos */
2909 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2910 xlate_report(ctx, "Not tunneling to our own address");
2911 goto out; /* restore flow_nw_tos */
2913 if (ctx->xin->resubmit_stats) {
2914 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2916 if (ctx->xin->xcache) {
2917 struct xc_entry *entry;
2919 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2920 entry->u.dev.tx = netdev_ref(xport->netdev);
2922 out_port = odp_port;
2923 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2924 tnl_push_pop_send = true;
2926 commit_odp_tunnel_action(flow, &ctx->base_flow,
2927 ctx->xout->odp_actions);
2928 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2931 odp_port = xport->odp_port;
2932 out_port = odp_port;
2933 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2934 ofp_port_t vlandev_port;
2936 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2937 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2938 ofp_port, flow->vlan_tci);
2939 if (vlandev_port != ofp_port) {
2940 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2941 flow->vlan_tci = htons(0);
2946 if (out_port != ODPP_NONE) {
2947 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2948 ctx->xout->odp_actions,
2950 ctx->xbridge->masked_set_action);
2953 struct ovs_action_hash *act_hash;
2956 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2957 OVS_ACTION_ATTR_HASH,
2959 act_hash->hash_alg = xr->hash_alg;
2960 act_hash->hash_basis = xr->hash_basis;
2962 /* Recirc action. */
2963 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2967 if (tnl_push_pop_send) {
2968 build_tunnel_send(ctx, xport, flow, odp_port);
2969 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2971 odp_port_t odp_tnl_port = ODPP_NONE;
2973 /* XXX: Write better Filter for tunnel port. We can use inport
2974 * int tunnel-port flow to avoid these checks completely. */
2975 if (ofp_port == OFPP_LOCAL &&
2976 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2978 odp_tnl_port = tnl_port_map_lookup(flow, wc);
2981 if (odp_tnl_port != ODPP_NONE) {
2982 nl_msg_put_odp_port(ctx->xout->odp_actions,
2983 OVS_ACTION_ATTR_TUNNEL_POP,
2986 /* Tunnel push-pop action is not compatible with
2988 add_ipfix_output_action(ctx, out_port);
2989 nl_msg_put_odp_port(ctx->xout->odp_actions,
2990 OVS_ACTION_ATTR_OUTPUT,
2996 ctx->sflow_odp_port = odp_port;
2997 ctx->sflow_n_outputs++;
2998 ctx->xout->nf_output_iface = ofp_port;
3003 flow->vlan_tci = flow_vlan_tci;
3004 flow->pkt_mark = flow_pkt_mark;
3005 flow->nw_tos = flow_nw_tos;
3009 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3010 const struct xlate_bond_recirc *xr)
3012 compose_output_action__(ctx, ofp_port, xr, true);
3016 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3018 struct rule_dpif *old_rule = ctx->rule;
3019 ovs_be64 old_cookie = ctx->rule_cookie;
3020 const struct rule_actions *actions;
3022 if (ctx->xin->resubmit_stats) {
3023 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3029 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3030 actions = rule_dpif_get_actions(rule);
3031 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3032 ctx->rule_cookie = old_cookie;
3033 ctx->rule = old_rule;
3038 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3040 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3042 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3043 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
3044 MAX_RESUBMIT_RECURSION);
3045 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3046 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
3047 } else if (ctx->xout->odp_actions->size > UINT16_MAX) {
3048 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
3049 } else if (ctx->stack.size >= 65536) {
3050 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
3059 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3060 bool may_packet_in, bool honor_table_miss)
3062 if (xlate_resubmit_resource_check(ctx)) {
3063 struct flow_wildcards *wc;
3064 uint8_t old_table_id = ctx->table_id;
3065 struct rule_dpif *rule;
3067 ctx->table_id = table_id;
3068 wc = (ctx->xin->skip_wildcards) ? NULL : &ctx->xout->wc;
3070 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3071 &ctx->xin->flow, wc,
3072 ctx->xin->xcache != NULL,
3073 ctx->xin->resubmit_stats,
3074 &ctx->table_id, in_port,
3075 may_packet_in, honor_table_miss);
3077 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3078 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3082 /* Fill in the cache entry here instead of xlate_recursively
3083 * to make the reference counting more explicit. We take a
3084 * reference in the lookups above if we are going to cache the
3086 if (ctx->xin->xcache) {
3087 struct xc_entry *entry;
3089 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3090 entry->u.rule = rule;
3092 xlate_recursively(ctx, rule);
3095 ctx->table_id = old_table_id;
3103 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3104 struct ofputil_bucket *bucket)
3106 if (ctx->xin->resubmit_stats) {
3107 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3109 if (ctx->xin->xcache) {
3110 struct xc_entry *entry;
3112 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3113 entry->u.group.group = group_dpif_ref(group);
3114 entry->u.group.bucket = bucket;
3119 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3121 uint64_t action_list_stub[1024 / 8];
3122 struct ofpbuf action_list, action_set;
3123 struct flow old_flow = ctx->xin->flow;
3124 bool old_was_mpls = ctx->was_mpls;
3126 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3127 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3129 ofpacts_execute_action_set(&action_list, &action_set);
3131 do_xlate_actions(action_list.data, action_list.size, ctx);
3134 ofpbuf_uninit(&action_set);
3135 ofpbuf_uninit(&action_list);
3137 /* Check if need to recirculate. */
3138 if (exit_recirculates(ctx)) {
3139 compose_recirculate_action(ctx);
3142 /* Roll back flow to previous state.
3143 * This is equivalent to cloning the packet for each bucket.
3145 * As a side effect any subsequently applied actions will
3146 * also effectively be applied to a clone of the packet taken
3147 * just before applying the all or indirect group.
3149 * Note that group buckets are action sets, hence they cannot modify the
3150 * main action set. Also any stack actions are ignored when executing an
3151 * action set, so group buckets cannot change the stack either.
3152 * However, we do allow resubmit actions in group buckets, which could
3153 * break the above assumptions. It is up to the controller to not mess up
3154 * with the action_set and stack in the tables resubmitted to from
3156 ctx->xin->flow = old_flow;
3158 /* The group bucket popping MPLS should have no effect after bucket
3160 ctx->was_mpls = old_was_mpls;
3162 /* The fact that the group bucket exits (for any reason) does not mean that
3163 * the translation after the group action should exit. Specifically, if
3164 * the group bucket recirculates (which typically modifies the packet), the
3165 * actions after the group action must continue processing with the
3166 * original, not the recirculated packet! */
3171 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3173 struct ofputil_bucket *bucket;
3174 const struct ovs_list *buckets;
3176 group_dpif_get_buckets(group, &buckets);
3178 LIST_FOR_EACH (bucket, list_node, buckets) {
3179 xlate_group_bucket(ctx, bucket);
3181 xlate_group_stats(ctx, group, NULL);
3185 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3187 struct ofputil_bucket *bucket;
3189 bucket = group_first_live_bucket(ctx, group, 0);
3191 xlate_group_bucket(ctx, bucket);
3192 xlate_group_stats(ctx, group, bucket);
3197 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3199 struct flow_wildcards *wc = &ctx->xout->wc;
3200 struct ofputil_bucket *bucket;
3203 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3204 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3205 bucket = group_best_live_bucket(ctx, group, basis);
3207 xlate_group_bucket(ctx, bucket);
3208 xlate_group_stats(ctx, group, bucket);
3213 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3215 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3216 struct flow_wildcards *wc = &ctx->xout->wc;
3217 const struct field_array *fields;
3218 struct ofputil_bucket *bucket;
3222 fields = group_dpif_get_fields(group);
3223 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3225 /* Determine which fields to hash */
3226 for (i = 0; i < MFF_N_IDS; i++) {
3227 if (bitmap_is_set(fields->used.bm, i)) {
3228 const struct mf_field *mf;
3230 /* If the field is already present in 'hash_fields' then
3231 * this loop has already checked that it and its pre-requisites
3232 * are present in the flow and its pre-requisites have
3233 * already been added to 'hash_fields'. There is nothing more
3234 * to do here and as an optimisation the loop can continue. */
3235 if (bitmap_is_set(hash_fields.bm, i)) {
3241 /* Only hash a field if it and its pre-requisites are present
3243 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3247 /* Hash both the field and its pre-requisites */
3248 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3252 /* Hash the fields */
3253 for (i = 0; i < MFF_N_IDS; i++) {
3254 if (bitmap_is_set(hash_fields.bm, i)) {
3255 const struct mf_field *mf = mf_from_id(i);
3256 union mf_value value;
3259 mf_get_value(mf, &ctx->xin->flow, &value);
3260 /* This seems inefficient but so does apply_mask() */
3261 for (j = 0; j < mf->n_bytes; j++) {
3262 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3264 basis = hash_bytes(&value, mf->n_bytes, basis);
3266 mf_mask_field(mf, &wc->masks);
3270 bucket = group_best_live_bucket(ctx, group, basis);
3272 xlate_group_bucket(ctx, bucket);
3273 xlate_group_stats(ctx, group, bucket);
3278 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3280 const char *selection_method = group_dpif_get_selection_method(group);
3282 if (selection_method[0] == '\0') {
3283 xlate_default_select_group(ctx, group);
3284 } else if (!strcasecmp("hash", selection_method)) {
3285 xlate_hash_fields_select_group(ctx, group);
3287 /* Parsing of groups should ensure this never happens */
3293 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3295 ctx->in_group = true;
3297 switch (group_dpif_get_type(group)) {
3299 case OFPGT11_INDIRECT:
3300 xlate_all_group(ctx, group);
3302 case OFPGT11_SELECT:
3303 xlate_select_group(ctx, group);
3306 xlate_ff_group(ctx, group);
3311 group_dpif_unref(group);
3313 ctx->in_group = false;
3317 xlate_group_resource_check(struct xlate_ctx *ctx)
3319 if (!xlate_resubmit_resource_check(ctx)) {
3321 } else if (ctx->in_group) {
3322 /* Prevent nested translation of OpenFlow groups.
3324 * OpenFlow allows this restriction. We enforce this restriction only
3325 * because, with the current architecture, we would otherwise have to
3326 * take a possibly recursive read lock on the ofgroup rwlock, which is
3327 * unsafe given that POSIX allows taking a read lock to block if there
3328 * is a thread blocked on taking the write lock. Other solutions
3329 * without this restriction are also possible, but seem unwarranted
3330 * given the current limited use of groups. */
3331 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3333 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
3341 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3343 if (xlate_group_resource_check(ctx)) {
3344 struct group_dpif *group;
3347 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3349 xlate_group_action__(ctx, group);
3359 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3360 const struct ofpact_resubmit *resubmit)
3364 bool may_packet_in = false;
3365 bool honor_table_miss = false;
3367 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3368 /* Still allow missed packets to be sent to the controller
3369 * if resubmitting from an internal table. */
3370 may_packet_in = true;
3371 honor_table_miss = true;
3374 in_port = resubmit->in_port;
3375 if (in_port == OFPP_IN_PORT) {
3376 in_port = ctx->xin->flow.in_port.ofp_port;
3379 table_id = resubmit->table_id;
3380 if (table_id == 255) {
3381 table_id = ctx->table_id;
3384 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3389 flood_packets(struct xlate_ctx *ctx, bool all)
3391 const struct xport *xport;
3393 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3394 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3399 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3400 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3401 compose_output_action(ctx, xport->ofp_port, NULL);
3405 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3409 execute_controller_action(struct xlate_ctx *ctx, int len,
3410 enum ofp_packet_in_reason reason,
3411 uint16_t controller_id)
3413 struct ofproto_packet_in *pin;
3414 struct dp_packet *packet;
3416 ctx->xout->slow |= SLOW_CONTROLLER;
3417 if (!ctx->xin->packet) {
3421 packet = dp_packet_clone(ctx->xin->packet);
3423 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3424 ctx->xout->odp_actions,
3426 ctx->xbridge->masked_set_action);
3428 odp_execute_actions(NULL, &packet, 1, false,
3429 ctx->xout->odp_actions->data,
3430 ctx->xout->odp_actions->size, NULL);
3432 pin = xmalloc(sizeof *pin);
3433 pin->up.packet_len = dp_packet_size(packet);
3434 pin->up.packet = dp_packet_steal_data(packet);
3435 pin->up.reason = reason;
3436 pin->up.table_id = ctx->table_id;
3437 pin->up.cookie = ctx->rule_cookie;
3439 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3441 pin->controller_id = controller_id;
3442 pin->send_len = len;
3443 /* If a rule is a table-miss rule then this is
3444 * a table-miss handled by a table-miss rule.
3446 * Else, if rule is internal and has a controller action,
3447 * the later being implied by the rule being processed here,
3448 * then this is a table-miss handled without a table-miss rule.
3450 * Otherwise this is not a table-miss. */
3451 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3453 if (rule_dpif_is_table_miss(ctx->rule)) {
3454 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3455 } else if (rule_dpif_is_internal(ctx->rule)) {
3456 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3459 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3460 dp_packet_delete(packet);
3463 /* Called only when ctx->recirc_action_offset is set. */
3465 compose_recirculate_action(struct xlate_ctx *ctx)
3467 struct recirc_metadata md;
3470 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3471 ctx->xout->odp_actions,
3473 ctx->xbridge->masked_set_action);
3475 recirc_metadata_from_flow(&md, &ctx->xin->flow);
3477 ovs_assert(ctx->recirc_action_offset >= 0);
3479 /* Only allocate recirculation ID if we have a packet. */
3480 if (ctx->xin->packet) {
3481 /* Allocate a unique recirc id for the given metadata state in the
3482 * flow. The life-cycle of this recirc id is managed by associating it
3483 * with the udpif key ('ukey') created for each new datapath flow. */
3484 id = recirc_alloc_id_ctx(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3485 ctx->recirc_action_offset,
3486 ctx->action_set.size, ctx->action_set.data);
3488 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3489 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3492 xlate_out_add_recirc(ctx->xout, id);
3494 /* Look up an existing recirc id for the given metadata state in the
3495 * flow. No new reference is taken, as the ID is RCU protected and is
3496 * only required temporarily for verification. */
3497 id = recirc_find_id(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3498 ctx->recirc_action_offset,
3499 ctx->action_set.size, ctx->action_set.data);
3500 /* We let zero 'id' to be used in the RECIRC action below, which will
3501 * fail all revalidations as zero is not a valid recirculation ID. */
3504 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3506 /* Undo changes done by recirculation. */
3507 ctx->action_set.size = ctx->recirc_action_offset;
3508 ctx->recirc_action_offset = -1;
3509 ctx->last_unroll_offset = -1;
3513 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3515 struct flow_wildcards *wc = &ctx->xout->wc;
3516 struct flow *flow = &ctx->xin->flow;
3519 ovs_assert(eth_type_mpls(mpls->ethertype));
3521 n = flow_count_mpls_labels(flow, wc);
3523 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3524 ctx->xout->odp_actions,
3526 ctx->xbridge->masked_set_action);
3527 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3528 if (ctx->xin->packet != NULL) {
3529 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3530 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3531 "MPLS push action can't be performed as it would "
3532 "have more MPLS LSEs than the %d supported.",
3533 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3539 flow_push_mpls(flow, n, mpls->ethertype, wc);
3543 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3545 struct flow_wildcards *wc = &ctx->xout->wc;
3546 struct flow *flow = &ctx->xin->flow;
3547 int n = flow_count_mpls_labels(flow, wc);
3549 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3550 if (ctx->xbridge->enable_recirc) {
3551 ctx->was_mpls = true;
3553 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3554 if (ctx->xin->packet != NULL) {
3555 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3556 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3557 "MPLS pop action can't be performed as it has "
3558 "more MPLS LSEs than the %d supported.",
3559 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3562 ofpbuf_clear(ctx->xout->odp_actions);
3567 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3569 struct flow *flow = &ctx->xin->flow;
3571 if (!is_ip_any(flow)) {
3575 ctx->xout->wc.masks.nw_ttl = 0xff;
3576 if (flow->nw_ttl > 1) {
3582 for (i = 0; i < ids->n_controllers; i++) {
3583 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3587 /* Stop processing for current table. */
3593 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3595 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3596 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3597 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3602 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3604 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3605 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3606 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3611 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3613 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3614 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3615 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3620 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3622 struct flow *flow = &ctx->xin->flow;
3623 struct flow_wildcards *wc = &ctx->xout->wc;
3625 if (eth_type_mpls(flow->dl_type)) {
3626 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3628 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3631 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3634 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3638 /* Stop processing for current table. */
3643 xlate_output_action(struct xlate_ctx *ctx,
3644 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3646 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3648 ctx->xout->nf_output_iface = NF_OUT_DROP;
3652 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3655 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3656 0, may_packet_in, true);
3662 flood_packets(ctx, false);
3665 flood_packets(ctx, true);
3667 case OFPP_CONTROLLER:
3668 execute_controller_action(ctx, max_len,
3669 (ctx->in_group ? OFPR_GROUP
3670 : ctx->in_action_set ? OFPR_ACTION_SET
3678 if (port != ctx->xin->flow.in_port.ofp_port) {
3679 compose_output_action(ctx, port, NULL);
3681 xlate_report(ctx, "skipping output to input port");
3686 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3687 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3688 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3689 ctx->xout->nf_output_iface = prev_nf_output_iface;
3690 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3691 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3692 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3697 xlate_output_reg_action(struct xlate_ctx *ctx,
3698 const struct ofpact_output_reg *or)
3700 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3701 if (port <= UINT16_MAX) {
3702 union mf_subvalue value;
3704 memset(&value, 0xff, sizeof value);
3705 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3706 xlate_output_action(ctx, u16_to_ofp(port),
3707 or->max_len, false);
3712 xlate_enqueue_action(struct xlate_ctx *ctx,
3713 const struct ofpact_enqueue *enqueue)
3715 ofp_port_t ofp_port = enqueue->port;
3716 uint32_t queue_id = enqueue->queue;
3717 uint32_t flow_priority, priority;
3720 /* Translate queue to priority. */
3721 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3723 /* Fall back to ordinary output action. */
3724 xlate_output_action(ctx, enqueue->port, 0, false);
3728 /* Check output port. */
3729 if (ofp_port == OFPP_IN_PORT) {
3730 ofp_port = ctx->xin->flow.in_port.ofp_port;
3731 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3735 /* Add datapath actions. */
3736 flow_priority = ctx->xin->flow.skb_priority;
3737 ctx->xin->flow.skb_priority = priority;
3738 compose_output_action(ctx, ofp_port, NULL);
3739 ctx->xin->flow.skb_priority = flow_priority;
3741 /* Update NetFlow output port. */
3742 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3743 ctx->xout->nf_output_iface = ofp_port;
3744 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3745 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3750 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3752 uint32_t skb_priority;
3754 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3755 ctx->xin->flow.skb_priority = skb_priority;
3757 /* Couldn't translate queue to a priority. Nothing to do. A warning
3758 * has already been logged. */
3763 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3765 const struct xbridge *xbridge = xbridge_;
3776 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3779 port = get_ofp_port(xbridge, ofp_port);
3780 return port ? port->may_enable : false;
3785 xlate_bundle_action(struct xlate_ctx *ctx,
3786 const struct ofpact_bundle *bundle)
3790 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3792 CONST_CAST(struct xbridge *, ctx->xbridge));
3793 if (bundle->dst.field) {
3794 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3797 xlate_output_action(ctx, port, 0, false);
3802 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3803 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3805 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3806 if (ctx->xin->may_learn) {
3807 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3812 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3814 ctx->xout->has_learn = true;
3815 learn_mask(learn, &ctx->xout->wc);
3817 if (ctx->xin->xcache) {
3818 struct xc_entry *entry;
3820 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3821 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3822 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3823 entry->u.learn.ofpacts = ofpbuf_new(64);
3824 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3825 entry->u.learn.ofpacts);
3826 } else if (ctx->xin->may_learn) {
3827 uint64_t ofpacts_stub[1024 / 8];
3828 struct ofputil_flow_mod fm;
3829 struct ofpbuf ofpacts;
3831 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3832 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3833 ofpbuf_uninit(&ofpacts);
3838 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3839 uint16_t idle_timeout, uint16_t hard_timeout)
3841 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3842 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3847 xlate_fin_timeout(struct xlate_ctx *ctx,
3848 const struct ofpact_fin_timeout *oft)
3851 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3852 oft->fin_idle_timeout, oft->fin_hard_timeout);
3853 if (ctx->xin->xcache) {
3854 struct xc_entry *entry;
3856 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3857 /* XC_RULE already holds a reference on the rule, none is taken
3859 entry->u.fin.rule = ctx->rule;
3860 entry->u.fin.idle = oft->fin_idle_timeout;
3861 entry->u.fin.hard = oft->fin_hard_timeout;
3867 xlate_sample_action(struct xlate_ctx *ctx,
3868 const struct ofpact_sample *os)
3870 union user_action_cookie cookie;
3871 /* Scale the probability from 16-bit to 32-bit while representing
3872 * the same percentage. */
3873 uint32_t probability = (os->probability << 16) | os->probability;
3875 if (!ctx->xbridge->variable_length_userdata) {
3876 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3878 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3879 "lacks support (needs Linux 3.10+ or kernel module from "
3884 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3885 ctx->xout->odp_actions,
3887 ctx->xbridge->masked_set_action);
3889 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3890 os->obs_domain_id, os->obs_point_id, &cookie);
3891 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions,
3892 &ctx->xin->flow, probability, &cookie,
3893 sizeof cookie.flow_sample, ODPP_NONE);
3897 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3899 if (xport->config & (is_stp(&ctx->xin->flow)
3900 ? OFPUTIL_PC_NO_RECV_STP
3901 : OFPUTIL_PC_NO_RECV)) {
3905 /* Only drop packets here if both forwarding and learning are
3906 * disabled. If just learning is enabled, we need to have
3907 * OFPP_NORMAL and the learning action have a look at the packet
3908 * before we can drop it. */
3909 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3910 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3918 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3920 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3921 size_t on_len = ofpact_nest_get_action_len(on);
3922 const struct ofpact *inner;
3924 /* Maintain actset_output depending on the contents of the action set:
3926 * - OFPP_UNSET, if there is no "output" action.
3928 * - The output port, if there is an "output" action and no "group"
3931 * - OFPP_UNSET, if there is a "group" action.
3933 if (!ctx->action_set_has_group) {
3934 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3935 if (inner->type == OFPACT_OUTPUT) {
3936 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3937 } else if (inner->type == OFPACT_GROUP) {
3938 ctx->xin->flow.actset_output = OFPP_UNSET;
3939 ctx->action_set_has_group = true;
3944 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3945 ofpact_pad(&ctx->action_set);
3949 xlate_action_set(struct xlate_ctx *ctx)
3951 uint64_t action_list_stub[1024 / 64];
3952 struct ofpbuf action_list;
3954 ctx->in_action_set = true;
3955 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3956 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3957 /* Clear the action set, as it is not needed any more. */
3958 ofpbuf_clear(&ctx->action_set);
3959 do_xlate_actions(action_list.data, action_list.size, ctx);
3960 ctx->in_action_set = false;
3961 ofpbuf_uninit(&action_list);
3965 ofpact_needs_recirculation_after_mpls(const struct ofpact *a, struct xlate_ctx *ctx)
3967 struct flow_wildcards *wc = &ctx->xout->wc;
3968 struct flow *flow = &ctx->xin->flow;
3970 if (!ctx->was_mpls) {
3977 case OFPACT_CONTROLLER:
3978 case OFPACT_STRIP_VLAN:
3979 case OFPACT_SET_VLAN_PCP:
3980 case OFPACT_SET_VLAN_VID:
3981 case OFPACT_ENQUEUE:
3982 case OFPACT_PUSH_VLAN:
3983 case OFPACT_SET_ETH_SRC:
3984 case OFPACT_SET_ETH_DST:
3985 case OFPACT_SET_TUNNEL:
3986 case OFPACT_SET_QUEUE:
3987 case OFPACT_POP_QUEUE:
3988 case OFPACT_CONJUNCTION:
3990 case OFPACT_UNROLL_XLATE:
3991 case OFPACT_OUTPUT_REG:
3994 case OFPACT_WRITE_METADATA:
3995 case OFPACT_WRITE_ACTIONS:
3996 case OFPACT_CLEAR_ACTIONS:
4000 case OFPACT_POP_MPLS:
4001 case OFPACT_DEC_MPLS_TTL:
4002 case OFPACT_SET_MPLS_TTL:
4003 case OFPACT_SET_MPLS_TC:
4004 case OFPACT_SET_MPLS_LABEL:
4005 case OFPACT_SET_IPV4_SRC:
4006 case OFPACT_SET_IPV4_DST:
4007 case OFPACT_SET_IP_DSCP:
4008 case OFPACT_SET_IP_ECN:
4009 case OFPACT_SET_IP_TTL:
4010 case OFPACT_SET_L4_SRC_PORT:
4011 case OFPACT_SET_L4_DST_PORT:
4012 case OFPACT_RESUBMIT:
4013 case OFPACT_STACK_PUSH:
4014 case OFPACT_STACK_POP:
4015 case OFPACT_DEC_TTL:
4016 case OFPACT_MULTIPATH:
4019 case OFPACT_FIN_TIMEOUT:
4020 case OFPACT_GOTO_TABLE:
4023 case OFPACT_REG_MOVE:
4024 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4025 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4027 case OFPACT_SET_FIELD:
4028 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
4030 case OFPACT_PUSH_MPLS:
4031 /* Recirculate if it is an IP packet with a zero ttl. This may
4032 * indicate that the packet was previously MPLS and an MPLS pop action
4033 * converted it to IP. In this case recirculating should reveal the IP
4034 * TTL which is used as the basis for a new MPLS LSE. */
4035 return (!flow_count_mpls_labels(flow, wc)
4036 && flow->nw_ttl == 0
4037 && is_ip_any(flow));
4044 recirc_put_unroll_xlate(struct xlate_ctx *ctx)
4046 struct ofpact_unroll_xlate *unroll;
4048 unroll = ctx->last_unroll_offset < 0
4050 : ALIGNED_CAST(struct ofpact_unroll_xlate *,
4051 (char *)ctx->action_set.data + ctx->last_unroll_offset);
4053 /* Restore the table_id and rule cookie for a potential PACKET
4056 (ctx->table_id != unroll->rule_table_id
4057 || ctx->rule_cookie != unroll->rule_cookie)) {
4059 ctx->last_unroll_offset = ctx->action_set.size;
4060 unroll = ofpact_put_UNROLL_XLATE(&ctx->action_set);
4061 unroll->rule_table_id = ctx->table_id;
4062 unroll->rule_cookie = ctx->rule_cookie;
4067 /* Copy remaining actions to the action_set to be executed after recirculation.
4068 * UNROLL_XLATE action is inserted, if not already done so, before actions that
4069 * may generate PACKET_INs from the current table and without matching another
4072 recirc_unroll_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4073 struct xlate_ctx *ctx)
4075 const struct ofpact *a;
4077 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4079 /* May generate PACKET INs. */
4080 case OFPACT_OUTPUT_REG:
4083 case OFPACT_CONTROLLER:
4084 case OFPACT_DEC_MPLS_TTL:
4085 case OFPACT_DEC_TTL:
4086 recirc_put_unroll_xlate(ctx);
4089 /* These may not generate PACKET INs. */
4090 case OFPACT_SET_TUNNEL:
4091 case OFPACT_REG_MOVE:
4092 case OFPACT_SET_FIELD:
4093 case OFPACT_STACK_PUSH:
4094 case OFPACT_STACK_POP:
4096 case OFPACT_WRITE_METADATA:
4097 case OFPACT_RESUBMIT: /* May indirectly generate PACKET INs, */
4098 case OFPACT_GOTO_TABLE: /* but from a different table and rule. */
4099 case OFPACT_ENQUEUE:
4100 case OFPACT_SET_VLAN_VID:
4101 case OFPACT_SET_VLAN_PCP:
4102 case OFPACT_STRIP_VLAN:
4103 case OFPACT_PUSH_VLAN:
4104 case OFPACT_SET_ETH_SRC:
4105 case OFPACT_SET_ETH_DST:
4106 case OFPACT_SET_IPV4_SRC:
4107 case OFPACT_SET_IPV4_DST:
4108 case OFPACT_SET_IP_DSCP:
4109 case OFPACT_SET_IP_ECN:
4110 case OFPACT_SET_IP_TTL:
4111 case OFPACT_SET_L4_SRC_PORT:
4112 case OFPACT_SET_L4_DST_PORT:
4113 case OFPACT_SET_QUEUE:
4114 case OFPACT_POP_QUEUE:
4115 case OFPACT_PUSH_MPLS:
4116 case OFPACT_POP_MPLS:
4117 case OFPACT_SET_MPLS_LABEL:
4118 case OFPACT_SET_MPLS_TC:
4119 case OFPACT_SET_MPLS_TTL:
4120 case OFPACT_MULTIPATH:
4123 case OFPACT_UNROLL_XLATE:
4124 case OFPACT_FIN_TIMEOUT:
4125 case OFPACT_CLEAR_ACTIONS:
4126 case OFPACT_WRITE_ACTIONS:
4131 /* These need not be copied for restoration. */
4133 case OFPACT_CONJUNCTION:
4136 /* Copy the action over. */
4137 ofpbuf_put(&ctx->action_set, a, OFPACT_ALIGN(a->len));
4142 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4143 struct xlate_ctx *ctx)
4145 struct flow_wildcards *wc = &ctx->xout->wc;
4146 struct flow *flow = &ctx->xin->flow;
4147 const struct ofpact *a;
4149 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4150 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
4152 /* dl_type already in the mask, not set below. */
4154 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4155 struct ofpact_controller *controller;
4156 const struct ofpact_metadata *metadata;
4157 const struct ofpact_set_field *set_field;
4158 const struct mf_field *mf;
4160 if (!ctx->exit && ofpact_needs_recirculation_after_mpls(a, ctx)) {
4162 ctx->recirc_action_offset = ctx->action_set.size;
4166 /* Check if need to store the remaining actions for later
4168 if (exit_recirculates(ctx)) {
4169 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4171 (uint8_t *)ofpacts)),
4179 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4180 ofpact_get_OUTPUT(a)->max_len, true);
4184 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4189 case OFPACT_CONTROLLER:
4190 controller = ofpact_get_CONTROLLER(a);
4191 execute_controller_action(ctx, controller->max_len,
4193 controller->controller_id);
4196 case OFPACT_ENQUEUE:
4197 memset(&wc->masks.skb_priority, 0xff,
4198 sizeof wc->masks.skb_priority);
4199 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4202 case OFPACT_SET_VLAN_VID:
4203 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4204 if (flow->vlan_tci & htons(VLAN_CFI) ||
4205 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4206 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4207 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4212 case OFPACT_SET_VLAN_PCP:
4213 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4214 if (flow->vlan_tci & htons(VLAN_CFI) ||
4215 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4216 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4217 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4218 << VLAN_PCP_SHIFT) | VLAN_CFI);
4222 case OFPACT_STRIP_VLAN:
4223 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4224 flow->vlan_tci = htons(0);
4227 case OFPACT_PUSH_VLAN:
4228 /* XXX 802.1AD(QinQ) */
4229 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4230 flow->vlan_tci = htons(VLAN_CFI);
4233 case OFPACT_SET_ETH_SRC:
4234 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
4235 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
4238 case OFPACT_SET_ETH_DST:
4239 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
4240 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
4243 case OFPACT_SET_IPV4_SRC:
4244 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4245 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4246 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4250 case OFPACT_SET_IPV4_DST:
4251 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4252 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4253 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4257 case OFPACT_SET_IP_DSCP:
4258 if (is_ip_any(flow)) {
4259 wc->masks.nw_tos |= IP_DSCP_MASK;
4260 flow->nw_tos &= ~IP_DSCP_MASK;
4261 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4265 case OFPACT_SET_IP_ECN:
4266 if (is_ip_any(flow)) {
4267 wc->masks.nw_tos |= IP_ECN_MASK;
4268 flow->nw_tos &= ~IP_ECN_MASK;
4269 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4273 case OFPACT_SET_IP_TTL:
4274 if (is_ip_any(flow)) {
4275 wc->masks.nw_ttl = 0xff;
4276 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4280 case OFPACT_SET_L4_SRC_PORT:
4281 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4282 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4283 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4284 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4288 case OFPACT_SET_L4_DST_PORT:
4289 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4290 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4291 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4292 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4296 case OFPACT_RESUBMIT:
4297 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4300 case OFPACT_SET_TUNNEL:
4301 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4304 case OFPACT_SET_QUEUE:
4305 memset(&wc->masks.skb_priority, 0xff,
4306 sizeof wc->masks.skb_priority);
4307 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4310 case OFPACT_POP_QUEUE:
4311 memset(&wc->masks.skb_priority, 0xff,
4312 sizeof wc->masks.skb_priority);
4313 flow->skb_priority = ctx->orig_skb_priority;
4316 case OFPACT_REG_MOVE:
4317 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4320 case OFPACT_SET_FIELD:
4321 set_field = ofpact_get_SET_FIELD(a);
4322 mf = set_field->field;
4324 /* Set field action only ever overwrites packet's outermost
4325 * applicable header fields. Do nothing if no header exists. */
4326 if (mf->id == MFF_VLAN_VID) {
4327 wc->masks.vlan_tci |= htons(VLAN_CFI);
4328 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4331 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4332 /* 'dl_type' is already unwildcarded. */
4333 && !eth_type_mpls(flow->dl_type)) {
4336 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4337 * header field on a packet that does not have them. */
4338 mf_mask_field_and_prereqs(mf, &wc->masks);
4339 if (mf_are_prereqs_ok(mf, flow)) {
4340 mf_set_flow_value_masked(mf, &set_field->value,
4341 &set_field->mask, flow);
4345 case OFPACT_STACK_PUSH:
4346 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4350 case OFPACT_STACK_POP:
4351 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4355 case OFPACT_PUSH_MPLS:
4356 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4359 case OFPACT_POP_MPLS:
4360 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4363 case OFPACT_SET_MPLS_LABEL:
4364 compose_set_mpls_label_action(
4365 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4368 case OFPACT_SET_MPLS_TC:
4369 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4372 case OFPACT_SET_MPLS_TTL:
4373 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4376 case OFPACT_DEC_MPLS_TTL:
4377 if (compose_dec_mpls_ttl_action(ctx)) {
4382 case OFPACT_DEC_TTL:
4383 wc->masks.nw_ttl = 0xff;
4384 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4390 /* Nothing to do. */
4393 case OFPACT_MULTIPATH:
4394 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4398 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4401 case OFPACT_OUTPUT_REG:
4402 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4406 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4409 case OFPACT_CONJUNCTION: {
4410 /* A flow with a "conjunction" action represents part of a special
4411 * kind of "set membership match". Such a flow should not actually
4412 * get executed, but it could via, say, a "packet-out", even though
4413 * that wouldn't be useful. Log it to help debugging. */
4414 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4415 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4423 case OFPACT_UNROLL_XLATE: {
4424 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4426 /* Restore translation context data that was stored earlier. */
4427 ctx->table_id = unroll->rule_table_id;
4428 ctx->rule_cookie = unroll->rule_cookie;
4431 case OFPACT_FIN_TIMEOUT:
4432 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4433 ctx->xout->has_fin_timeout = true;
4434 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4437 case OFPACT_CLEAR_ACTIONS:
4438 ofpbuf_clear(&ctx->action_set);
4439 ctx->xin->flow.actset_output = OFPP_UNSET;
4440 ctx->action_set_has_group = false;
4443 case OFPACT_WRITE_ACTIONS:
4444 xlate_write_actions(ctx, a);
4447 case OFPACT_WRITE_METADATA:
4448 metadata = ofpact_get_WRITE_METADATA(a);
4449 flow->metadata &= ~metadata->mask;
4450 flow->metadata |= metadata->metadata & metadata->mask;
4454 /* Not implemented yet. */
4457 case OFPACT_GOTO_TABLE: {
4458 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4460 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4461 * than ogt->table_id. This is to allow goto_table actions that
4462 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4463 * after recirculation. */
4464 ovs_assert(ctx->table_id == TBL_INTERNAL
4465 || ctx->table_id < ogt->table_id);
4466 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4467 ogt->table_id, true, true);
4472 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4479 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4480 const struct flow *flow, ofp_port_t in_port,
4481 struct rule_dpif *rule, uint16_t tcp_flags,
4482 const struct dp_packet *packet)
4484 xin->ofproto = ofproto;
4486 xin->flow.in_port.ofp_port = in_port;
4487 xin->flow.actset_output = OFPP_UNSET;
4488 xin->packet = packet;
4489 xin->may_learn = packet != NULL;
4492 xin->ofpacts = NULL;
4493 xin->ofpacts_len = 0;
4494 xin->tcp_flags = tcp_flags;
4495 xin->resubmit_hook = NULL;
4496 xin->report_hook = NULL;
4497 xin->resubmit_stats = NULL;
4498 xin->skip_wildcards = false;
4499 xin->odp_actions = NULL;
4501 /* Do recirc lookup. */
4502 xin->recirc = flow->recirc_id
4503 ? recirc_id_node_find(flow->recirc_id)
4508 xlate_out_uninit(struct xlate_out *xout)
4511 if (xout->odp_actions == &xout->odp_actions_buf) {
4512 ofpbuf_uninit(xout->odp_actions);
4514 xlate_out_free_recircs(xout);
4518 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4519 * into datapath actions, using 'ctx', and discards the datapath actions. */
4521 xlate_actions_for_side_effects(struct xlate_in *xin)
4523 struct xlate_out xout;
4525 xlate_actions(xin, &xout);
4526 xlate_out_uninit(&xout);
4530 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
4533 dst->slow = src->slow;
4534 dst->has_learn = src->has_learn;
4535 dst->has_normal = src->has_normal;
4536 dst->has_fin_timeout = src->has_fin_timeout;
4537 dst->nf_output_iface = src->nf_output_iface;
4538 dst->mirrors = src->mirrors;
4540 dst->odp_actions = &dst->odp_actions_buf;
4541 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
4542 sizeof dst->odp_actions_stub);
4543 ofpbuf_put(dst->odp_actions, src->odp_actions->data, src->odp_actions->size);
4546 static struct skb_priority_to_dscp *
4547 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4549 struct skb_priority_to_dscp *pdscp;
4552 hash = hash_int(skb_priority, 0);
4553 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4554 if (pdscp->skb_priority == skb_priority) {
4562 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4565 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4566 *dscp = pdscp ? pdscp->dscp : 0;
4567 return pdscp != NULL;
4571 count_skb_priorities(const struct xport *xport)
4573 return hmap_count(&xport->skb_priorities);
4577 clear_skb_priorities(struct xport *xport)
4579 struct skb_priority_to_dscp *pdscp, *next;
4581 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4582 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4588 actions_output_to_local_port(const struct xlate_ctx *ctx)
4590 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4591 const struct nlattr *a;
4594 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions->data,
4595 ctx->xout->odp_actions->size) {
4596 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4597 && nl_attr_get_odp_port(a) == local_odp_port) {
4604 #if defined(__linux__)
4605 /* Returns the maximum number of packets that the Linux kernel is willing to
4606 * queue up internally to certain kinds of software-implemented ports, or the
4607 * default (and rarely modified) value if it cannot be determined. */
4609 netdev_max_backlog(void)
4611 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4612 static int max_backlog = 1000; /* The normal default value. */
4614 if (ovsthread_once_start(&once)) {
4615 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4619 stream = fopen(filename, "r");
4621 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4623 if (fscanf(stream, "%d", &n) != 1) {
4624 VLOG_WARN("%s: read error", filename);
4625 } else if (n <= 100) {
4626 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4632 ovsthread_once_done(&once);
4634 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4640 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4643 count_output_actions(const struct ofpbuf *odp_actions)
4645 const struct nlattr *a;
4649 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4650 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4656 #endif /* defined(__linux__) */
4658 /* Returns true if 'odp_actions' contains more output actions than the datapath
4659 * can reliably handle in one go. On Linux, this is the value of the
4660 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4661 * packets that the kernel is willing to queue up for processing while the
4662 * datapath is processing a set of actions. */
4664 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4667 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4668 && count_output_actions(odp_actions) > netdev_max_backlog());
4670 /* OSes other than Linux might have similar limits, but we don't know how
4671 * to determine them.*/
4676 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
4678 * The caller must take responsibility for eventually freeing 'xout', with
4679 * xlate_out_uninit(). */
4681 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4683 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4684 struct flow_wildcards *wc = NULL;
4685 struct flow *flow = &xin->flow;
4686 struct rule_dpif *rule = NULL;
4688 enum slow_path_reason special;
4689 const struct ofpact *ofpacts;
4690 struct xbridge *xbridge;
4691 struct xport *in_port;
4692 struct flow orig_flow;
4693 struct xlate_ctx ctx;
4698 COVERAGE_INC(xlate_actions);
4700 /* Flow initialization rules:
4701 * - 'base_flow' must match the kernel's view of the packet at the
4702 * time that action processing starts. 'flow' represents any
4703 * transformations we wish to make through actions.
4704 * - By default 'base_flow' and 'flow' are the same since the input
4705 * packet matches the output before any actions are applied.
4706 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4707 * of the received packet as seen by the kernel. If we later output
4708 * to another device without any modifications this will cause us to
4709 * insert a new tag since the original one was stripped off by the
4711 * - Tunnel metadata as received is retained in 'flow'. This allows
4712 * tunnel metadata matching also in later tables.
4713 * Since a kernel action for setting the tunnel metadata will only be
4714 * generated with actual tunnel output, changing the tunnel metadata
4715 * values in 'flow' (such as tun_id) will only have effect with a later
4716 * tunnel output action.
4717 * - Tunnel 'base_flow' is completely cleared since that is what the
4718 * kernel does. If we wish to maintain the original values an action
4719 * needs to be generated. */
4724 ctx.xout->has_learn = false;
4725 ctx.xout->has_normal = false;
4726 ctx.xout->has_fin_timeout = false;
4727 ctx.xout->nf_output_iface = NF_OUT_DROP;
4728 ctx.xout->mirrors = 0;
4729 ctx.xout->n_recircs = 0;
4731 xout->odp_actions = xin->odp_actions;
4732 if (!xout->odp_actions) {
4733 xout->odp_actions = &xout->odp_actions_buf;
4734 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4735 sizeof xout->odp_actions_stub);
4737 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4739 xbridge = xbridge_lookup(xcfg, xin->ofproto);
4743 /* 'ctx.xbridge' may be changed by action processing, whereas 'xbridge'
4744 * will remain set on the original input bridge. */
4745 ctx.xbridge = xbridge;
4746 ctx.rule = xin->rule;
4748 ctx.base_flow = *flow;
4749 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4750 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4752 if (!xin->skip_wildcards) {
4754 flow_wildcards_init_catchall(wc);
4755 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4756 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4757 if (is_ip_any(flow)) {
4758 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4760 if (xbridge->enable_recirc) {
4761 /* Always exactly match recirc_id when datapath supports
4763 wc->masks.recirc_id = UINT32_MAX;
4765 if (xbridge->netflow) {
4766 netflow_mask_wc(flow, wc);
4769 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4771 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4775 ctx.in_group = false;
4776 ctx.in_action_set = false;
4777 ctx.orig_skb_priority = flow->skb_priority;
4779 ctx.rule_cookie = OVS_BE64_MAX;
4781 ctx.was_mpls = false;
4782 ctx.recirc_action_offset = -1;
4783 ctx.last_unroll_offset = -1;
4785 ctx.action_set_has_group = false;
4786 ofpbuf_use_stub(&ctx.action_set,
4787 ctx.action_set_stub, sizeof ctx.action_set_stub);
4789 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4791 /* The in_port of the original packet before recirculation. */
4792 in_port = get_ofp_port(xbridge, flow->in_port.ofp_port);
4795 const struct recirc_id_node *recirc = xin->recirc;
4797 if (xin->ofpacts_len > 0 || ctx.rule) {
4798 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4800 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!",
4801 xin->ofpacts_len > 0
4807 /* Set the bridge for post-recirculation processing if needed. */
4808 if (ctx.xbridge->ofproto != recirc->ofproto) {
4809 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4810 const struct xbridge *new_bridge = xbridge_lookup(xcfg,
4813 if (OVS_UNLIKELY(!new_bridge)) {
4814 /* Drop the packet if the bridge cannot be found. */
4815 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4816 VLOG_WARN_RL(&rl, "Recirculation bridge no longer exists.");
4819 ctx.xbridge = new_bridge;
4822 /* Set the post-recirculation table id. Note: A table lookup is done
4823 * only if there are no post-recirculation actions. */
4824 ctx.table_id = recirc->table_id;
4826 /* Restore pipeline metadata. May change flow's in_port and other
4827 * metadata to the values that existed when recirculation was
4829 recirc_metadata_to_flow(&recirc->metadata, flow);
4831 /* Restore stack, if any. */
4832 if (recirc->stack) {
4833 ofpbuf_put(&ctx.stack, recirc->stack->data, recirc->stack->size);
4836 /* Restore action set, if any. */
4837 if (recirc->action_set_len) {
4838 const struct ofpact *a;
4840 ofpbuf_put(&ctx.action_set, recirc->ofpacts,
4841 recirc->action_set_len);
4843 OFPACT_FOR_EACH(a, recirc->ofpacts, recirc->action_set_len) {
4844 if (a->type == OFPACT_GROUP) {
4845 ctx.action_set_has_group = true;
4851 /* Restore recirculation actions. If there are no actions, processing
4852 * will start with a lookup in the table set above. */
4853 if (recirc->ofpacts_len > recirc->action_set_len) {
4854 xin->ofpacts_len = recirc->ofpacts_len - recirc->action_set_len;
4855 xin->ofpacts = recirc->ofpacts +
4856 recirc->action_set_len / sizeof *recirc->ofpacts;
4858 } else if (OVS_UNLIKELY(flow->recirc_id)) {
4859 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4861 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
4866 if (!xin->ofpacts && !ctx.rule) {
4867 rule = rule_dpif_lookup_from_table(ctx.xbridge->ofproto, 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);
5075 /* Sends 'packet' out 'ofport'.
5076 * May modify 'packet'.
5077 * Returns 0 if successful, otherwise a positive errno value. */
5079 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5081 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5082 struct xport *xport;
5083 struct ofpact_output output;
5086 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5087 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5088 flow_extract(packet, &flow);
5089 flow.in_port.ofp_port = OFPP_NONE;
5091 xport = xport_lookup(xcfg, ofport);
5095 output.port = xport->ofp_port;
5098 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5099 &output.ofpact, sizeof output,
5103 struct xlate_cache *
5104 xlate_cache_new(void)
5106 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5108 ofpbuf_init(&xcache->entries, 512);
5112 static struct xc_entry *
5113 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5115 struct xc_entry *entry;
5117 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5124 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5126 if (entry->u.dev.tx) {
5127 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5129 if (entry->u.dev.rx) {
5130 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5132 if (entry->u.dev.bfd) {
5133 bfd_account_rx(entry->u.dev.bfd, stats);
5138 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5140 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5141 struct xbridge *xbridge;
5142 struct xbundle *xbundle;
5143 struct flow_wildcards wc;
5145 xbridge = xbridge_lookup(xcfg, ofproto);
5150 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5156 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5159 /* Push stats and perform side effects of flow translation. */
5161 xlate_push_stats(struct xlate_cache *xcache,
5162 const struct dpif_flow_stats *stats)
5164 struct xc_entry *entry;
5165 struct ofpbuf entries = xcache->entries;
5166 uint8_t dmac[ETH_ADDR_LEN];
5168 if (!stats->n_packets) {
5172 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5173 switch (entry->type) {
5175 rule_dpif_credit_stats(entry->u.rule, stats);
5178 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5179 entry->u.bond.vid, stats->n_bytes);
5182 xlate_cache_netdev(entry, stats);
5185 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5186 entry->u.nf.iface, stats);
5189 mirror_update_stats(entry->u.mirror.mbridge,
5190 entry->u.mirror.mirrors,
5191 stats->n_packets, stats->n_bytes);
5194 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5197 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5198 entry->u.normal.vlan);
5200 case XC_FIN_TIMEOUT:
5201 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5202 entry->u.fin.idle, entry->u.fin.hard);
5205 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5209 /* Lookup arp to avoid arp timeout. */
5210 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
5219 xlate_dev_unref(struct xc_entry *entry)
5221 if (entry->u.dev.tx) {
5222 netdev_close(entry->u.dev.tx);
5224 if (entry->u.dev.rx) {
5225 netdev_close(entry->u.dev.rx);
5227 if (entry->u.dev.bfd) {
5228 bfd_unref(entry->u.dev.bfd);
5233 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5235 netflow_flow_clear(netflow, flow);
5236 netflow_unref(netflow);
5241 xlate_cache_clear(struct xlate_cache *xcache)
5243 struct xc_entry *entry;
5244 struct ofpbuf entries;
5250 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5251 switch (entry->type) {
5253 rule_dpif_unref(entry->u.rule);
5256 free(entry->u.bond.flow);
5257 bond_unref(entry->u.bond.bond);
5260 xlate_dev_unref(entry);
5263 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5266 mbridge_unref(entry->u.mirror.mbridge);
5269 free(entry->u.learn.fm);
5270 ofpbuf_delete(entry->u.learn.ofpacts);
5273 free(entry->u.normal.flow);
5275 case XC_FIN_TIMEOUT:
5276 /* 'u.fin.rule' is always already held as a XC_RULE, which
5277 * has already released it's reference above. */
5280 group_dpif_unref(entry->u.group.group);
5289 ofpbuf_clear(&xcache->entries);
5293 xlate_cache_delete(struct xlate_cache *xcache)
5295 xlate_cache_clear(xcache);
5296 ofpbuf_uninit(&xcache->entries);