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 /* Timeout for internal rules created to handle recirculation */
75 #define RECIRC_TIMEOUT 60
77 /* Maximum number of resubmit actions in a flow translation, whether they are
78 * recursive or not. */
79 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
82 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
83 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
85 struct ovs_list xbundles; /* Owned xbundles. */
86 struct hmap xports; /* Indexed by ofp_port. */
88 char *name; /* Name used in log messages. */
89 struct dpif *dpif; /* Datapath interface. */
90 struct mac_learning *ml; /* Mac learning handle. */
91 struct mcast_snooping *ms; /* Multicast Snooping handle. */
92 struct mbridge *mbridge; /* Mirroring. */
93 struct dpif_sflow *sflow; /* SFlow handle, or null. */
94 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
95 struct netflow *netflow; /* Netflow handle, or null. */
96 struct stp *stp; /* STP or null if disabled. */
97 struct rstp *rstp; /* RSTP or null if disabled. */
99 bool has_in_band; /* Bridge has in band control? */
100 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
102 /* True if the datapath supports recirculation. */
105 /* True if the datapath supports variable-length
106 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
107 * False if the datapath supports only 8-byte (or shorter) userdata. */
108 bool variable_length_userdata;
110 /* Number of MPLS label stack entries that the datapath supports
112 size_t max_mpls_depth;
114 /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
116 bool masked_set_action;
120 struct hmap_node hmap_node; /* In global 'xbundles' map. */
121 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
123 struct ovs_list list_node; /* In parent 'xbridges' list. */
124 struct xbridge *xbridge; /* Parent xbridge. */
126 struct ovs_list xports; /* Contains "struct xport"s. */
128 char *name; /* Name used in log messages. */
129 struct bond *bond; /* Nonnull iff more than one port. */
130 struct lacp *lacp; /* LACP handle or null. */
132 enum port_vlan_mode vlan_mode; /* VLAN mode. */
133 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
134 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
135 * NULL if all VLANs are trunked. */
136 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
137 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
141 struct hmap_node hmap_node; /* Node in global 'xports' map. */
142 struct ofport_dpif *ofport; /* Key in global 'xports map. */
144 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
145 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
147 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
149 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
150 struct xbundle *xbundle; /* Parent xbundle or null. */
152 struct netdev *netdev; /* 'ofport''s netdev. */
154 struct xbridge *xbridge; /* Parent bridge. */
155 struct xport *peer; /* Patch port peer or null. */
157 enum ofputil_port_config config; /* OpenFlow port configuration. */
158 enum ofputil_port_state state; /* OpenFlow port state. */
159 int stp_port_no; /* STP port number or -1 if not in use. */
160 struct rstp_port *rstp_port; /* RSTP port or null. */
162 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
164 bool may_enable; /* May be enabled in bonds. */
165 bool is_tunnel; /* Is a tunnel port. */
167 struct cfm *cfm; /* CFM handle or null. */
168 struct bfd *bfd; /* BFD handle or null. */
169 struct lldp *lldp; /* LLDP handle or null. */
173 struct xlate_in *xin;
174 struct xlate_out *xout;
176 const struct xbridge *xbridge;
178 /* Flow at the last commit. */
179 struct flow base_flow;
181 /* Tunnel IP destination address as received. This is stored separately
182 * as the base_flow.tunnel is cleared on init to reflect the datapath
183 * behavior. Used to make sure not to send tunneled output to ourselves,
184 * which might lead to an infinite loop. This could happen easily
185 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
186 * actually set the tun_dst field. */
187 ovs_be32 orig_tunnel_ip_dst;
189 /* Stack for the push and pop actions. Each stack element is of type
190 * "union mf_subvalue". */
191 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
194 /* The rule that we are currently translating, or NULL. */
195 struct rule_dpif *rule;
197 /* Resubmit statistics, via xlate_table_action(). */
198 int recurse; /* Current resubmit nesting depth. */
199 int resubmits; /* Total number of resubmits. */
200 bool in_group; /* Currently translating ofgroup, if true. */
201 bool in_action_set; /* Currently translating action_set, if true. */
203 uint32_t orig_skb_priority; /* Priority when packet arrived. */
204 uint8_t table_id; /* OpenFlow table ID where flow was found. */
205 uint32_t sflow_n_outputs; /* Number of output ports. */
206 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
207 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
208 bool exit; /* No further actions should be processed. */
210 bool use_recirc; /* Should generate recirc? */
211 struct xlate_recirc recirc; /* Information used for generating
212 * recirculation actions */
214 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
215 * This is a trigger for recirculation in cases where translating an action
216 * or looking up a flow requires access to the fields of the packet after
217 * the MPLS label stack that was originally present. */
220 /* OpenFlow 1.1+ action set.
222 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
223 * When translation is otherwise complete, ofpacts_execute_action_set()
224 * converts it to a set of "struct ofpact"s that can be translated into
225 * datapath actions. */
226 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
227 struct ofpbuf action_set; /* Action set. */
228 uint64_t action_set_stub[1024 / 8];
231 /* A controller may use OFPP_NONE as the ingress port to indicate that
232 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
233 * when an input bundle is needed for validation (e.g., mirroring or
234 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
235 * any 'port' structs, so care must be taken when dealing with it. */
236 static struct xbundle ofpp_none_bundle = {
238 .vlan_mode = PORT_VLAN_TRUNK
241 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
242 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
243 * traffic egressing the 'ofport' with that priority should be marked with. */
244 struct skb_priority_to_dscp {
245 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
246 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
248 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
264 /* xlate_cache entries hold enough information to perform the side effects of
265 * xlate_actions() for a rule, without needing to perform rule translation
266 * from scratch. The primary usage of these is to submit statistics to objects
267 * that a flow relates to, although they may be used for other effects as well
268 * (for instance, refreshing hard timeouts for learned flows). */
272 struct rule_dpif *rule;
279 struct netflow *netflow;
284 struct mbridge *mbridge;
285 mirror_mask_t mirrors;
293 struct ofproto_dpif *ofproto;
294 struct ofputil_flow_mod *fm;
295 struct ofpbuf *ofpacts;
298 struct ofproto_dpif *ofproto;
303 struct rule_dpif *rule;
308 struct group_dpif *group;
309 struct ofputil_bucket *bucket;
312 char br_name[IFNAMSIZ];
318 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
319 entries = xcache->entries; \
320 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
322 entry = ofpbuf_try_pull(&entries, sizeof *entry))
325 struct ofpbuf entries;
328 /* Xlate config contains hash maps of all bridges, bundles and ports.
329 * Xcfgp contains the pointer to the current xlate configuration.
330 * When the main thread needs to change the configuration, it copies xcfgp to
331 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
332 * does not block handler and revalidator threads. */
334 struct hmap xbridges;
335 struct hmap xbundles;
338 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
339 static struct xlate_cfg *new_xcfg = NULL;
341 static bool may_receive(const struct xport *, struct xlate_ctx *);
342 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
344 static void xlate_normal(struct xlate_ctx *);
345 static inline void xlate_report(struct xlate_ctx *, const char *);
346 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
347 uint8_t table_id, bool may_packet_in,
348 bool honor_table_miss);
349 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
350 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
351 static void output_normal(struct xlate_ctx *, const struct xbundle *,
353 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
355 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
356 const struct ofproto_dpif *);
357 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
358 const struct ofbundle *);
359 static struct xport *xport_lookup(struct xlate_cfg *,
360 const struct ofport_dpif *);
361 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
362 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
363 uint32_t skb_priority);
364 static void clear_skb_priorities(struct xport *);
365 static size_t count_skb_priorities(const struct xport *);
366 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
369 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
371 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
372 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
373 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
374 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
375 const struct mac_learning *, struct stp *,
376 struct rstp *, const struct mcast_snooping *,
377 const struct mbridge *,
378 const struct dpif_sflow *,
379 const struct dpif_ipfix *,
380 const struct netflow *,
381 bool forward_bpdu, bool has_in_band,
383 bool variable_length_userdata,
384 size_t max_mpls_depth,
385 bool masked_set_action);
386 static void xlate_xbundle_set(struct xbundle *xbundle,
387 enum port_vlan_mode vlan_mode, int vlan,
388 unsigned long *trunks, bool use_priority_tags,
389 const struct bond *bond, const struct lacp *lacp,
391 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
392 const struct netdev *netdev, const struct cfm *cfm,
393 const struct bfd *bfd, const struct lldp *lldp,
394 int stp_port_no, const struct rstp_port *rstp_port,
395 enum ofputil_port_config config,
396 enum ofputil_port_state state, bool is_tunnel,
398 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
399 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
400 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
401 static void xlate_xbridge_copy(struct xbridge *);
402 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
403 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
405 static void xlate_xcfg_free(struct xlate_cfg *);
408 xlate_report(struct xlate_ctx *ctx, const char *s)
410 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
411 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
416 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
418 list_init(&xbridge->xbundles);
419 hmap_init(&xbridge->xports);
420 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
421 hash_pointer(xbridge->ofproto, 0));
425 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
427 list_init(&xbundle->xports);
428 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
429 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
430 hash_pointer(xbundle->ofbundle, 0));
434 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
436 hmap_init(&xport->skb_priorities);
437 hmap_insert(&xcfg->xports, &xport->hmap_node,
438 hash_pointer(xport->ofport, 0));
439 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
440 hash_ofp_port(xport->ofp_port));
444 xlate_xbridge_set(struct xbridge *xbridge,
446 const struct mac_learning *ml, struct stp *stp,
447 struct rstp *rstp, const struct mcast_snooping *ms,
448 const struct mbridge *mbridge,
449 const struct dpif_sflow *sflow,
450 const struct dpif_ipfix *ipfix,
451 const struct netflow *netflow,
452 bool forward_bpdu, bool has_in_band,
454 bool variable_length_userdata,
455 size_t max_mpls_depth,
456 bool masked_set_action)
458 if (xbridge->ml != ml) {
459 mac_learning_unref(xbridge->ml);
460 xbridge->ml = mac_learning_ref(ml);
463 if (xbridge->ms != ms) {
464 mcast_snooping_unref(xbridge->ms);
465 xbridge->ms = mcast_snooping_ref(ms);
468 if (xbridge->mbridge != mbridge) {
469 mbridge_unref(xbridge->mbridge);
470 xbridge->mbridge = mbridge_ref(mbridge);
473 if (xbridge->sflow != sflow) {
474 dpif_sflow_unref(xbridge->sflow);
475 xbridge->sflow = dpif_sflow_ref(sflow);
478 if (xbridge->ipfix != ipfix) {
479 dpif_ipfix_unref(xbridge->ipfix);
480 xbridge->ipfix = dpif_ipfix_ref(ipfix);
483 if (xbridge->stp != stp) {
484 stp_unref(xbridge->stp);
485 xbridge->stp = stp_ref(stp);
488 if (xbridge->rstp != rstp) {
489 rstp_unref(xbridge->rstp);
490 xbridge->rstp = rstp_ref(rstp);
493 if (xbridge->netflow != netflow) {
494 netflow_unref(xbridge->netflow);
495 xbridge->netflow = netflow_ref(netflow);
498 xbridge->dpif = dpif;
499 xbridge->forward_bpdu = forward_bpdu;
500 xbridge->has_in_band = has_in_band;
501 xbridge->enable_recirc = enable_recirc;
502 xbridge->variable_length_userdata = variable_length_userdata;
503 xbridge->max_mpls_depth = max_mpls_depth;
504 xbridge->masked_set_action = masked_set_action;
508 xlate_xbundle_set(struct xbundle *xbundle,
509 enum port_vlan_mode vlan_mode, int vlan,
510 unsigned long *trunks, bool use_priority_tags,
511 const struct bond *bond, const struct lacp *lacp,
514 ovs_assert(xbundle->xbridge);
516 xbundle->vlan_mode = vlan_mode;
517 xbundle->vlan = vlan;
518 xbundle->trunks = trunks;
519 xbundle->use_priority_tags = use_priority_tags;
520 xbundle->floodable = floodable;
522 if (xbundle->bond != bond) {
523 bond_unref(xbundle->bond);
524 xbundle->bond = bond_ref(bond);
527 if (xbundle->lacp != lacp) {
528 lacp_unref(xbundle->lacp);
529 xbundle->lacp = lacp_ref(lacp);
534 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
535 const struct netdev *netdev, const struct cfm *cfm,
536 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
537 const struct rstp_port* rstp_port,
538 enum ofputil_port_config config, enum ofputil_port_state state,
539 bool is_tunnel, bool may_enable)
541 xport->config = config;
542 xport->state = state;
543 xport->stp_port_no = stp_port_no;
544 xport->is_tunnel = is_tunnel;
545 xport->may_enable = may_enable;
546 xport->odp_port = odp_port;
548 if (xport->rstp_port != rstp_port) {
549 rstp_port_unref(xport->rstp_port);
550 xport->rstp_port = rstp_port_ref(rstp_port);
553 if (xport->cfm != cfm) {
554 cfm_unref(xport->cfm);
555 xport->cfm = cfm_ref(cfm);
558 if (xport->bfd != bfd) {
559 bfd_unref(xport->bfd);
560 xport->bfd = bfd_ref(bfd);
563 if (xport->lldp != lldp) {
564 lldp_unref(xport->lldp);
565 xport->lldp = lldp_ref(lldp);
568 if (xport->netdev != netdev) {
569 netdev_close(xport->netdev);
570 xport->netdev = netdev_ref(netdev);
575 xlate_xbridge_copy(struct xbridge *xbridge)
577 struct xbundle *xbundle;
579 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
580 new_xbridge->ofproto = xbridge->ofproto;
581 new_xbridge->name = xstrdup(xbridge->name);
582 xlate_xbridge_init(new_xcfg, new_xbridge);
584 xlate_xbridge_set(new_xbridge,
585 xbridge->dpif, xbridge->ml, xbridge->stp,
586 xbridge->rstp, xbridge->ms, xbridge->mbridge,
587 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
588 xbridge->forward_bpdu,
589 xbridge->has_in_band, xbridge->enable_recirc,
590 xbridge->variable_length_userdata,
591 xbridge->max_mpls_depth, xbridge->masked_set_action);
592 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
593 xlate_xbundle_copy(new_xbridge, xbundle);
596 /* Copy xports which are not part of a xbundle */
597 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
598 if (!xport->xbundle) {
599 xlate_xport_copy(new_xbridge, NULL, xport);
605 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
608 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
609 new_xbundle->ofbundle = xbundle->ofbundle;
610 new_xbundle->xbridge = xbridge;
611 new_xbundle->name = xstrdup(xbundle->name);
612 xlate_xbundle_init(new_xcfg, new_xbundle);
614 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
615 xbundle->vlan, xbundle->trunks,
616 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
618 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
619 xlate_xport_copy(xbridge, new_xbundle, xport);
624 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
627 struct skb_priority_to_dscp *pdscp, *new_pdscp;
628 struct xport *new_xport = xzalloc(sizeof *xport);
629 new_xport->ofport = xport->ofport;
630 new_xport->ofp_port = xport->ofp_port;
631 new_xport->xbridge = xbridge;
632 xlate_xport_init(new_xcfg, new_xport);
634 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
635 xport->bfd, xport->lldp, xport->stp_port_no,
636 xport->rstp_port, xport->config, xport->state,
637 xport->is_tunnel, xport->may_enable);
640 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
642 new_xport->peer = peer;
643 new_xport->peer->peer = new_xport;
648 new_xport->xbundle = xbundle;
649 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
652 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
653 new_pdscp = xmalloc(sizeof *pdscp);
654 new_pdscp->skb_priority = pdscp->skb_priority;
655 new_pdscp->dscp = pdscp->dscp;
656 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
657 hash_int(new_pdscp->skb_priority, 0));
661 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
662 * configuration in xcfgp.
664 * This needs to be called after editing the xlate configuration.
666 * Functions that edit the new xlate configuration are
667 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
673 * edit_xlate_configuration();
675 * xlate_txn_commit(); */
677 xlate_txn_commit(void)
679 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
681 ovsrcu_set(&xcfgp, new_xcfg);
682 ovsrcu_synchronize();
683 xlate_xcfg_free(xcfg);
687 /* Copies the current xlate configuration in xcfgp to new_xcfg.
689 * This needs to be called prior to editing the xlate configuration. */
691 xlate_txn_start(void)
693 struct xbridge *xbridge;
694 struct xlate_cfg *xcfg;
696 ovs_assert(!new_xcfg);
698 new_xcfg = xmalloc(sizeof *new_xcfg);
699 hmap_init(&new_xcfg->xbridges);
700 hmap_init(&new_xcfg->xbundles);
701 hmap_init(&new_xcfg->xports);
703 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
708 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
709 xlate_xbridge_copy(xbridge);
715 xlate_xcfg_free(struct xlate_cfg *xcfg)
717 struct xbridge *xbridge, *next_xbridge;
723 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
724 xlate_xbridge_remove(xcfg, xbridge);
727 hmap_destroy(&xcfg->xbridges);
728 hmap_destroy(&xcfg->xbundles);
729 hmap_destroy(&xcfg->xports);
734 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
736 const struct mac_learning *ml, struct stp *stp,
737 struct rstp *rstp, const struct mcast_snooping *ms,
738 const struct mbridge *mbridge,
739 const struct dpif_sflow *sflow,
740 const struct dpif_ipfix *ipfix,
741 const struct netflow *netflow,
742 bool forward_bpdu, bool has_in_band, bool enable_recirc,
743 bool variable_length_userdata, size_t max_mpls_depth,
744 bool masked_set_action)
746 struct xbridge *xbridge;
748 ovs_assert(new_xcfg);
750 xbridge = xbridge_lookup(new_xcfg, ofproto);
752 xbridge = xzalloc(sizeof *xbridge);
753 xbridge->ofproto = ofproto;
755 xlate_xbridge_init(new_xcfg, xbridge);
759 xbridge->name = xstrdup(name);
761 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
762 netflow, forward_bpdu, has_in_band, enable_recirc,
763 variable_length_userdata, max_mpls_depth,
768 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
770 struct xbundle *xbundle, *next_xbundle;
771 struct xport *xport, *next_xport;
777 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
778 xlate_xport_remove(xcfg, xport);
781 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
782 xlate_xbundle_remove(xcfg, xbundle);
785 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
786 mac_learning_unref(xbridge->ml);
787 mcast_snooping_unref(xbridge->ms);
788 mbridge_unref(xbridge->mbridge);
789 dpif_sflow_unref(xbridge->sflow);
790 dpif_ipfix_unref(xbridge->ipfix);
791 stp_unref(xbridge->stp);
792 rstp_unref(xbridge->rstp);
793 hmap_destroy(&xbridge->xports);
799 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
801 struct xbridge *xbridge;
803 ovs_assert(new_xcfg);
805 xbridge = xbridge_lookup(new_xcfg, ofproto);
806 xlate_xbridge_remove(new_xcfg, xbridge);
810 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
811 const char *name, enum port_vlan_mode vlan_mode, int vlan,
812 unsigned long *trunks, bool use_priority_tags,
813 const struct bond *bond, const struct lacp *lacp,
816 struct xbundle *xbundle;
818 ovs_assert(new_xcfg);
820 xbundle = xbundle_lookup(new_xcfg, ofbundle);
822 xbundle = xzalloc(sizeof *xbundle);
823 xbundle->ofbundle = ofbundle;
824 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
826 xlate_xbundle_init(new_xcfg, xbundle);
830 xbundle->name = xstrdup(name);
832 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
833 use_priority_tags, bond, lacp, floodable);
837 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
839 struct xport *xport, *next;
845 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
846 list_remove(&xport->bundle_node);
847 xport->xbundle = NULL;
850 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
851 list_remove(&xbundle->list_node);
852 bond_unref(xbundle->bond);
853 lacp_unref(xbundle->lacp);
859 xlate_bundle_remove(struct ofbundle *ofbundle)
861 struct xbundle *xbundle;
863 ovs_assert(new_xcfg);
865 xbundle = xbundle_lookup(new_xcfg, ofbundle);
866 xlate_xbundle_remove(new_xcfg, xbundle);
870 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
871 struct ofport_dpif *ofport, ofp_port_t ofp_port,
872 odp_port_t odp_port, const struct netdev *netdev,
873 const struct cfm *cfm, const struct bfd *bfd,
874 const struct lldp *lldp, struct ofport_dpif *peer,
875 int stp_port_no, const struct rstp_port *rstp_port,
876 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
877 enum ofputil_port_config config,
878 enum ofputil_port_state state, bool is_tunnel,
884 ovs_assert(new_xcfg);
886 xport = xport_lookup(new_xcfg, ofport);
888 xport = xzalloc(sizeof *xport);
889 xport->ofport = ofport;
890 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
891 xport->ofp_port = ofp_port;
893 xlate_xport_init(new_xcfg, xport);
896 ovs_assert(xport->ofp_port == ofp_port);
898 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
899 stp_port_no, rstp_port, config, state, is_tunnel,
903 xport->peer->peer = NULL;
905 xport->peer = xport_lookup(new_xcfg, peer);
907 xport->peer->peer = xport;
910 if (xport->xbundle) {
911 list_remove(&xport->bundle_node);
913 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
914 if (xport->xbundle) {
915 list_insert(&xport->xbundle->xports, &xport->bundle_node);
918 clear_skb_priorities(xport);
919 for (i = 0; i < n_qdscp; i++) {
920 struct skb_priority_to_dscp *pdscp;
921 uint32_t skb_priority;
923 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
928 pdscp = xmalloc(sizeof *pdscp);
929 pdscp->skb_priority = skb_priority;
930 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
931 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
932 hash_int(pdscp->skb_priority, 0));
937 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
944 xport->peer->peer = NULL;
948 if (xport->xbundle) {
949 list_remove(&xport->bundle_node);
952 clear_skb_priorities(xport);
953 hmap_destroy(&xport->skb_priorities);
955 hmap_remove(&xcfg->xports, &xport->hmap_node);
956 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
958 netdev_close(xport->netdev);
959 rstp_port_unref(xport->rstp_port);
960 cfm_unref(xport->cfm);
961 bfd_unref(xport->bfd);
962 lldp_unref(xport->lldp);
967 xlate_ofport_remove(struct ofport_dpif *ofport)
971 ovs_assert(new_xcfg);
973 xport = xport_lookup(new_xcfg, ofport);
974 xlate_xport_remove(new_xcfg, xport);
977 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
978 * returns the corresponding struct xport, or NULL if none is found. */
979 static struct xport *
980 xlate_lookup_xport(const struct dpif_backer *backer, const struct flow *flow)
982 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
984 return xport_lookup(xcfg, tnl_port_should_receive(flow)
985 ? tnl_port_receive(flow)
986 : odp_port_to_ofport(backer, flow->in_port.odp_port));
989 static struct ofproto_dpif *
990 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
991 ofp_port_t *ofp_in_port, const struct xport **xportp)
993 struct ofproto_dpif *recv_ofproto = NULL;
994 struct ofproto_dpif *recirc_ofproto = NULL;
995 const struct xport *xport;
996 ofp_port_t in_port = OFPP_NONE;
998 *xportp = xport = xlate_lookup_xport(backer, flow);
1001 recv_ofproto = xport->xbridge->ofproto;
1002 in_port = xport->ofp_port;
1005 /* When recirc_id is set in 'flow', checks whether the ofproto_dpif that
1006 * corresponds to the recirc_id is same as the receiving bridge. If they
1007 * are the same, uses the 'recv_ofproto' and keeps the 'ofp_in_port' as
1008 * assigned. Otherwise, uses the 'recirc_ofproto' that owns recirc_id and
1009 * assigns OFPP_NONE to 'ofp_in_port'. Doing this is in that, the
1010 * recirculated flow must be processced by the ofproto which originates
1011 * the recirculation, and as bridges can only see their own ports, the
1012 * in_port of the 'recv_ofproto' should not be passed to the
1015 * Admittedly, setting the 'ofp_in_port' to OFPP_NONE limits the
1016 * 'recirc_ofproto' from meaningfully matching on in_port of recirculated
1017 * flow, and should be fixed in the near future.
1019 * TODO: Restore the original patch port.
1021 if (recv_ofproto && flow->recirc_id) {
1022 recirc_ofproto = ofproto_dpif_recirc_get_ofproto(backer,
1024 if (recv_ofproto != recirc_ofproto) {
1025 *xportp = xport = NULL;
1026 in_port = OFPP_NONE;
1031 *ofp_in_port = in_port;
1034 return xport ? recv_ofproto : recirc_ofproto;
1037 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1038 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1039 struct ofproto_dpif *
1040 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1041 ofp_port_t *ofp_in_port)
1043 const struct xport *xport;
1045 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1048 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1049 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1050 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1051 * handles for those protocols if they're enabled. Caller may use the returned
1052 * pointers until quiescing, for longer term use additional references must
1055 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1058 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1059 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1060 struct dpif_sflow **sflow, struct netflow **netflow,
1061 ofp_port_t *ofp_in_port)
1063 struct ofproto_dpif *ofproto;
1064 const struct xport *xport;
1066 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1073 *ofprotop = ofproto;
1077 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1081 *sflow = xport ? xport->xbridge->sflow : NULL;
1085 *netflow = xport ? xport->xbridge->netflow : NULL;
1091 static struct xbridge *
1092 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1094 struct hmap *xbridges;
1095 struct xbridge *xbridge;
1097 if (!ofproto || !xcfg) {
1101 xbridges = &xcfg->xbridges;
1103 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1105 if (xbridge->ofproto == ofproto) {
1112 static struct xbundle *
1113 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1115 struct hmap *xbundles;
1116 struct xbundle *xbundle;
1118 if (!ofbundle || !xcfg) {
1122 xbundles = &xcfg->xbundles;
1124 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1126 if (xbundle->ofbundle == ofbundle) {
1133 static struct xport *
1134 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1136 struct hmap *xports;
1137 struct xport *xport;
1139 if (!ofport || !xcfg) {
1143 xports = &xcfg->xports;
1145 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1147 if (xport->ofport == ofport) {
1154 static struct stp_port *
1155 xport_get_stp_port(const struct xport *xport)
1157 return xport->xbridge->stp && xport->stp_port_no != -1
1158 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1163 xport_stp_learn_state(const struct xport *xport)
1165 struct stp_port *sp = xport_get_stp_port(xport);
1167 ? stp_learn_in_state(stp_port_get_state(sp))
1172 xport_stp_forward_state(const struct xport *xport)
1174 struct stp_port *sp = xport_get_stp_port(xport);
1176 ? stp_forward_in_state(stp_port_get_state(sp))
1181 xport_stp_should_forward_bpdu(const struct xport *xport)
1183 struct stp_port *sp = xport_get_stp_port(xport);
1184 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1187 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1188 * were used to make the determination.*/
1190 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1192 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1193 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1194 return is_stp(flow);
1198 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1200 struct stp_port *sp = xport_get_stp_port(xport);
1201 struct dp_packet payload = *packet;
1202 struct eth_header *eth = dp_packet_data(&payload);
1204 /* Sink packets on ports that have STP disabled when the bridge has
1206 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1210 /* Trim off padding on payload. */
1211 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1212 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1215 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1216 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1220 static enum rstp_state
1221 xport_get_rstp_port_state(const struct xport *xport)
1223 return xport->rstp_port
1224 ? rstp_port_get_state(xport->rstp_port)
1229 xport_rstp_learn_state(const struct xport *xport)
1231 return xport->xbridge->rstp && xport->rstp_port
1232 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1237 xport_rstp_forward_state(const struct xport *xport)
1239 return xport->xbridge->rstp && xport->rstp_port
1240 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1245 xport_rstp_should_manage_bpdu(const struct xport *xport)
1247 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1251 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1253 struct dp_packet payload = *packet;
1254 struct eth_header *eth = dp_packet_data(&payload);
1256 /* Sink packets on ports that have no RSTP. */
1257 if (!xport->rstp_port) {
1261 /* Trim off padding on payload. */
1262 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1263 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1266 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1267 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1268 dp_packet_size(&payload));
1272 static struct xport *
1273 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1275 struct xport *xport;
1277 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1279 if (xport->ofp_port == ofp_port) {
1287 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1289 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1290 return xport ? xport->odp_port : ODPP_NONE;
1294 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1296 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1297 return xport && xport->may_enable;
1300 static struct ofputil_bucket *
1301 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1305 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1307 struct group_dpif *group;
1309 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1310 struct ofputil_bucket *bucket;
1312 bucket = group_first_live_bucket(ctx, group, depth);
1313 group_dpif_unref(group);
1314 return bucket == NULL;
1320 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1323 bucket_is_alive(const struct xlate_ctx *ctx,
1324 struct ofputil_bucket *bucket, int depth)
1326 if (depth >= MAX_LIVENESS_RECURSION) {
1327 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1329 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1330 MAX_LIVENESS_RECURSION);
1334 return (!ofputil_bucket_has_liveness(bucket)
1335 || (bucket->watch_port != OFPP_ANY
1336 && odp_port_is_alive(ctx, bucket->watch_port))
1337 || (bucket->watch_group != OFPG_ANY
1338 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1341 static struct ofputil_bucket *
1342 group_first_live_bucket(const struct xlate_ctx *ctx,
1343 const struct group_dpif *group, int depth)
1345 struct ofputil_bucket *bucket;
1346 const struct ovs_list *buckets;
1348 group_dpif_get_buckets(group, &buckets);
1349 LIST_FOR_EACH (bucket, list_node, buckets) {
1350 if (bucket_is_alive(ctx, bucket, depth)) {
1358 static struct ofputil_bucket *
1359 group_best_live_bucket(const struct xlate_ctx *ctx,
1360 const struct group_dpif *group,
1363 struct ofputil_bucket *best_bucket = NULL;
1364 uint32_t best_score = 0;
1367 struct ofputil_bucket *bucket;
1368 const struct ovs_list *buckets;
1370 group_dpif_get_buckets(group, &buckets);
1371 LIST_FOR_EACH (bucket, list_node, buckets) {
1372 if (bucket_is_alive(ctx, bucket, 0)) {
1373 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1374 if (score >= best_score) {
1375 best_bucket = bucket;
1386 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1388 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1389 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1393 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1395 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1398 static mirror_mask_t
1399 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1401 return xbundle != &ofpp_none_bundle
1402 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1406 static mirror_mask_t
1407 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1409 return xbundle != &ofpp_none_bundle
1410 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1414 static mirror_mask_t
1415 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1417 return xbundle != &ofpp_none_bundle
1418 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1422 static struct xbundle *
1423 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1424 bool warn, struct xport **in_xportp)
1426 struct xport *xport;
1428 /* Find the port and bundle for the received packet. */
1429 xport = get_ofp_port(xbridge, in_port);
1433 if (xport && xport->xbundle) {
1434 return xport->xbundle;
1437 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1438 * which a controller may use as the ingress port for traffic that
1439 * it is sourcing. */
1440 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1441 return &ofpp_none_bundle;
1444 /* Odd. A few possible reasons here:
1446 * - We deleted a port but there are still a few packets queued up
1449 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1450 * we don't know about.
1452 * - The ofproto client didn't configure the port as part of a bundle.
1453 * This is particularly likely to happen if a packet was received on the
1454 * port after it was created, but before the client had a chance to
1455 * configure its bundle.
1458 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1460 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1461 "port %"PRIu16, xbridge->name, in_port);
1467 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1469 const struct xbridge *xbridge = ctx->xbridge;
1470 mirror_mask_t mirrors;
1471 struct xbundle *in_xbundle;
1475 mirrors = ctx->xout->mirrors;
1476 ctx->xout->mirrors = 0;
1478 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1479 ctx->xin->packet != NULL, NULL);
1483 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1485 /* Drop frames on bundles reserved for mirroring. */
1486 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1487 if (ctx->xin->packet != NULL) {
1488 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1489 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1490 "%s, which is reserved exclusively for mirroring",
1491 ctx->xbridge->name, in_xbundle->name);
1493 ofpbuf_clear(ctx->xout->odp_actions);
1498 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1499 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1502 vlan = input_vid_to_vlan(in_xbundle, vid);
1508 /* Restore the original packet before adding the mirror actions. */
1509 ctx->xin->flow = *orig_flow;
1512 mirror_mask_t dup_mirrors;
1513 struct ofbundle *out;
1514 unsigned long *vlans;
1519 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1520 &vlans, &dup_mirrors, &out, &out_vlan);
1521 ovs_assert(has_mirror);
1524 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1526 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1529 if (!vlan_mirrored) {
1530 mirrors = zero_rightmost_1bit(mirrors);
1534 mirrors &= ~dup_mirrors;
1535 ctx->xout->mirrors |= dup_mirrors;
1537 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1538 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1540 output_normal(ctx, out_xbundle, vlan);
1542 } else if (vlan != out_vlan
1543 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1544 struct xbundle *xbundle;
1546 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1547 if (xbundle_includes_vlan(xbundle, out_vlan)
1548 && !xbundle_mirror_out(xbridge, xbundle)) {
1549 output_normal(ctx, xbundle, out_vlan);
1556 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1557 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1558 * the bundle on which the packet was received, returns the VLAN to which the
1561 * Both 'vid' and the return value are in the range 0...4095. */
1563 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1565 switch (in_xbundle->vlan_mode) {
1566 case PORT_VLAN_ACCESS:
1567 return in_xbundle->vlan;
1570 case PORT_VLAN_TRUNK:
1573 case PORT_VLAN_NATIVE_UNTAGGED:
1574 case PORT_VLAN_NATIVE_TAGGED:
1575 return vid ? vid : in_xbundle->vlan;
1582 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1583 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1586 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1587 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1590 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1592 /* Allow any VID on the OFPP_NONE port. */
1593 if (in_xbundle == &ofpp_none_bundle) {
1597 switch (in_xbundle->vlan_mode) {
1598 case PORT_VLAN_ACCESS:
1601 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1602 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1603 "packet received on port %s configured as VLAN "
1604 "%"PRIu16" access port", vid, in_xbundle->name,
1611 case PORT_VLAN_NATIVE_UNTAGGED:
1612 case PORT_VLAN_NATIVE_TAGGED:
1614 /* Port must always carry its native VLAN. */
1618 case PORT_VLAN_TRUNK:
1619 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1621 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1622 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1623 "received on port %s not configured for trunking "
1624 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1636 /* Given 'vlan', the VLAN that a packet belongs to, and
1637 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1638 * that should be included in the 802.1Q header. (If the return value is 0,
1639 * then the 802.1Q header should only be included in the packet if there is a
1642 * Both 'vlan' and the return value are in the range 0...4095. */
1644 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1646 switch (out_xbundle->vlan_mode) {
1647 case PORT_VLAN_ACCESS:
1650 case PORT_VLAN_TRUNK:
1651 case PORT_VLAN_NATIVE_TAGGED:
1654 case PORT_VLAN_NATIVE_UNTAGGED:
1655 return vlan == out_xbundle->vlan ? 0 : vlan;
1663 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1666 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1668 ovs_be16 tci, old_tci;
1669 struct xport *xport;
1671 vid = output_vlan_to_vid(out_xbundle, vlan);
1672 if (list_is_empty(&out_xbundle->xports)) {
1673 /* Partially configured bundle with no slaves. Drop the packet. */
1675 } else if (!out_xbundle->bond) {
1676 ctx->use_recirc = false;
1677 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1680 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1681 struct flow_wildcards *wc = &ctx->xout->wc;
1682 struct xlate_recirc *xr = &ctx->recirc;
1683 struct ofport_dpif *ofport;
1685 if (ctx->xbridge->enable_recirc) {
1686 ctx->use_recirc = bond_may_recirc(
1687 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1689 if (ctx->use_recirc) {
1690 /* Only TCP mode uses recirculation. */
1691 xr->hash_alg = OVS_HASH_ALG_L4;
1692 bond_update_post_recirc_rules(out_xbundle->bond, false);
1694 /* Recirculation does not require unmasking hash fields. */
1699 ofport = bond_choose_output_slave(out_xbundle->bond,
1700 &ctx->xin->flow, wc, vid);
1701 xport = xport_lookup(xcfg, ofport);
1704 /* No slaves enabled, so drop packet. */
1708 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1709 * accounting for this bond. */
1710 if (!ctx->use_recirc) {
1711 if (ctx->xin->resubmit_stats) {
1712 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1713 ctx->xin->resubmit_stats->n_bytes);
1715 if (ctx->xin->xcache) {
1716 struct xc_entry *entry;
1719 flow = &ctx->xin->flow;
1720 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1721 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1722 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1723 entry->u.bond.vid = vid;
1728 old_tci = *flow_tci;
1730 if (tci || out_xbundle->use_priority_tags) {
1731 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1733 tci |= htons(VLAN_CFI);
1738 compose_output_action(ctx, xport->ofp_port);
1739 *flow_tci = old_tci;
1742 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1743 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1744 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1746 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1748 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1752 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1753 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1757 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1758 if (flow->nw_proto == ARP_OP_REPLY) {
1760 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1761 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1762 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1764 return flow->nw_src == flow->nw_dst;
1770 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1771 * dropped. Returns true if they may be forwarded, false if they should be
1774 * 'in_port' must be the xport that corresponds to flow->in_port.
1775 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1777 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1778 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1779 * checked by input_vid_is_valid().
1781 * May also add tags to '*tags', although the current implementation only does
1782 * so in one special case.
1785 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1788 struct xbundle *in_xbundle = in_port->xbundle;
1789 const struct xbridge *xbridge = ctx->xbridge;
1790 struct flow *flow = &ctx->xin->flow;
1792 /* Drop frames for reserved multicast addresses
1793 * only if forward_bpdu option is absent. */
1794 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1795 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1799 if (in_xbundle->bond) {
1800 struct mac_entry *mac;
1802 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1808 xlate_report(ctx, "bonding refused admissibility, dropping");
1811 case BV_DROP_IF_MOVED:
1812 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1813 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1815 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1816 && (!is_gratuitous_arp(flow, &ctx->xout->wc)
1817 || mac_entry_is_grat_arp_locked(mac))) {
1818 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1819 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1823 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1831 /* Checks whether a MAC learning update is necessary for MAC learning table
1832 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1835 * Most packets processed through the MAC learning table do not actually
1836 * change it in any way. This function requires only a read lock on the MAC
1837 * learning table, so it is much cheaper in this common case.
1839 * Keep the code here synchronized with that in update_learning_table__()
1842 is_mac_learning_update_needed(const struct mac_learning *ml,
1843 const struct flow *flow,
1844 struct flow_wildcards *wc,
1845 int vlan, struct xbundle *in_xbundle)
1846 OVS_REQ_RDLOCK(ml->rwlock)
1848 struct mac_entry *mac;
1850 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1854 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1855 if (!mac || mac_entry_age(ml, mac)) {
1859 if (is_gratuitous_arp(flow, wc)) {
1860 /* We don't want to learn from gratuitous ARP packets that are
1861 * reflected back over bond slaves so we lock the learning table. */
1862 if (!in_xbundle->bond) {
1864 } else if (mac_entry_is_grat_arp_locked(mac)) {
1869 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1873 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1874 * received on 'in_xbundle' in 'vlan'.
1876 * This code repeats all the checks in is_mac_learning_update_needed() because
1877 * the lock was released between there and here and thus the MAC learning state
1878 * could have changed.
1880 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1883 update_learning_table__(const struct xbridge *xbridge,
1884 const struct flow *flow, struct flow_wildcards *wc,
1885 int vlan, struct xbundle *in_xbundle)
1886 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1888 struct mac_entry *mac;
1890 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1894 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1895 if (is_gratuitous_arp(flow, wc)) {
1896 /* We don't want to learn from gratuitous ARP packets that are
1897 * reflected back over bond slaves so we lock the learning table. */
1898 if (!in_xbundle->bond) {
1899 mac_entry_set_grat_arp_lock(mac);
1900 } else if (mac_entry_is_grat_arp_locked(mac)) {
1905 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1906 /* The log messages here could actually be useful in debugging,
1907 * so keep the rate limit relatively high. */
1908 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1910 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1911 "on port %s in VLAN %d",
1912 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1913 in_xbundle->name, vlan);
1915 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
1920 update_learning_table(const struct xbridge *xbridge,
1921 const struct flow *flow, struct flow_wildcards *wc,
1922 int vlan, struct xbundle *in_xbundle)
1926 /* Don't learn the OFPP_NONE port. */
1927 if (in_xbundle == &ofpp_none_bundle) {
1931 /* First try the common case: no change to MAC learning table. */
1932 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1933 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1935 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1938 /* Slow path: MAC learning table might need an update. */
1939 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1940 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1941 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1945 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1946 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
1948 update_mcast_snooping_table__(const struct xbridge *xbridge,
1949 const struct flow *flow,
1950 struct mcast_snooping *ms,
1951 ovs_be32 ip4, int vlan,
1952 struct xbundle *in_xbundle)
1953 OVS_REQ_WRLOCK(ms->rwlock)
1955 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
1957 switch (ntohs(flow->tp_src)) {
1958 case IGMP_HOST_MEMBERSHIP_REPORT:
1959 case IGMPV2_HOST_MEMBERSHIP_REPORT:
1960 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1961 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
1962 IP_FMT" is on port %s in VLAN %d",
1963 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1966 case IGMP_HOST_LEAVE_MESSAGE:
1967 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1968 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
1969 IP_FMT" is on port %s in VLAN %d",
1970 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1973 case IGMP_HOST_MEMBERSHIP_QUERY:
1974 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
1975 in_xbundle->ofbundle)) {
1976 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
1977 IP_FMT" is on port %s in VLAN %d",
1978 xbridge->name, IP_ARGS(flow->nw_src),
1979 in_xbundle->name, vlan);
1985 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1986 * was received on 'in_xbundle' in 'vlan'. */
1988 update_mcast_snooping_table(const struct xbridge *xbridge,
1989 const struct flow *flow, int vlan,
1990 struct xbundle *in_xbundle)
1992 struct mcast_snooping *ms = xbridge->ms;
1993 struct xlate_cfg *xcfg;
1994 struct xbundle *mcast_xbundle;
1995 struct mcast_port_bundle *fport;
1997 /* Don't learn the OFPP_NONE port. */
1998 if (in_xbundle == &ofpp_none_bundle) {
2002 /* Don't learn from flood ports */
2003 mcast_xbundle = NULL;
2004 ovs_rwlock_wrlock(&ms->rwlock);
2005 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2006 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2007 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2008 if (mcast_xbundle == in_xbundle) {
2013 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2014 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
2017 ovs_rwlock_unlock(&ms->rwlock);
2020 /* send the packet to ports having the multicast group learned */
2022 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2023 struct mcast_snooping *ms OVS_UNUSED,
2024 struct mcast_group *grp,
2025 struct xbundle *in_xbundle, uint16_t vlan)
2026 OVS_REQ_RDLOCK(ms->rwlock)
2028 struct xlate_cfg *xcfg;
2029 struct mcast_group_bundle *b;
2030 struct xbundle *mcast_xbundle;
2032 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2033 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2034 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2035 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2036 xlate_report(ctx, "forwarding to mcast group port");
2037 output_normal(ctx, mcast_xbundle, vlan);
2038 } else if (!mcast_xbundle) {
2039 xlate_report(ctx, "mcast group port is unknown, dropping");
2041 xlate_report(ctx, "mcast group port is input port, dropping");
2046 /* send the packet to ports connected to multicast routers */
2048 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2049 struct mcast_snooping *ms,
2050 struct xbundle *in_xbundle, uint16_t vlan)
2051 OVS_REQ_RDLOCK(ms->rwlock)
2053 struct xlate_cfg *xcfg;
2054 struct mcast_mrouter_bundle *mrouter;
2055 struct xbundle *mcast_xbundle;
2057 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2058 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2059 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2060 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2061 xlate_report(ctx, "forwarding to mcast router port");
2062 output_normal(ctx, mcast_xbundle, vlan);
2063 } else if (!mcast_xbundle) {
2064 xlate_report(ctx, "mcast router port is unknown, dropping");
2066 xlate_report(ctx, "mcast router port is input port, dropping");
2071 /* send the packet to ports flagged to be flooded */
2073 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2074 struct mcast_snooping *ms,
2075 struct xbundle *in_xbundle, uint16_t vlan)
2076 OVS_REQ_RDLOCK(ms->rwlock)
2078 struct xlate_cfg *xcfg;
2079 struct mcast_port_bundle *fport;
2080 struct xbundle *mcast_xbundle;
2082 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2083 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2084 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2085 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2086 xlate_report(ctx, "forwarding to mcast flood port");
2087 output_normal(ctx, mcast_xbundle, vlan);
2088 } else if (!mcast_xbundle) {
2089 xlate_report(ctx, "mcast flood port is unknown, dropping");
2091 xlate_report(ctx, "mcast flood port is input port, dropping");
2096 /* forward the Reports to configured ports */
2098 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2099 struct mcast_snooping *ms,
2100 struct xbundle *in_xbundle, uint16_t vlan)
2101 OVS_REQ_RDLOCK(ms->rwlock)
2103 struct xlate_cfg *xcfg;
2104 struct mcast_port_bundle *rport;
2105 struct xbundle *mcast_xbundle;
2107 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2108 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2109 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2110 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2111 xlate_report(ctx, "forwarding Report to mcast flagged port");
2112 output_normal(ctx, mcast_xbundle, vlan);
2113 } else if (!mcast_xbundle) {
2114 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2116 xlate_report(ctx, "mcast port is input port, dropping the Report");
2122 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2125 struct xbundle *xbundle;
2127 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2128 if (xbundle != in_xbundle
2129 && xbundle_includes_vlan(xbundle, vlan)
2130 && xbundle->floodable
2131 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2132 output_normal(ctx, xbundle, vlan);
2135 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2139 xlate_normal(struct xlate_ctx *ctx)
2141 struct flow_wildcards *wc = &ctx->xout->wc;
2142 struct flow *flow = &ctx->xin->flow;
2143 struct xbundle *in_xbundle;
2144 struct xport *in_port;
2145 struct mac_entry *mac;
2150 ctx->xout->has_normal = true;
2152 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2153 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2154 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2156 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2157 ctx->xin->packet != NULL, &in_port);
2159 xlate_report(ctx, "no input bundle, dropping");
2163 /* Drop malformed frames. */
2164 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2165 !(flow->vlan_tci & htons(VLAN_CFI))) {
2166 if (ctx->xin->packet != NULL) {
2167 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2168 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2169 "VLAN tag received on port %s",
2170 ctx->xbridge->name, in_xbundle->name);
2172 xlate_report(ctx, "partial VLAN tag, dropping");
2176 /* Drop frames on bundles reserved for mirroring. */
2177 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2178 if (ctx->xin->packet != NULL) {
2179 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2180 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2181 "%s, which is reserved exclusively for mirroring",
2182 ctx->xbridge->name, in_xbundle->name);
2184 xlate_report(ctx, "input port is mirror output port, dropping");
2189 vid = vlan_tci_to_vid(flow->vlan_tci);
2190 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2191 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2194 vlan = input_vid_to_vlan(in_xbundle, vid);
2196 /* Check other admissibility requirements. */
2197 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2201 /* Learn source MAC. */
2202 if (ctx->xin->may_learn) {
2203 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2205 if (ctx->xin->xcache) {
2206 struct xc_entry *entry;
2208 /* Save enough info to update mac learning table later. */
2209 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2210 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2211 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2212 entry->u.normal.vlan = vlan;
2215 /* Determine output bundle. */
2216 if (mcast_snooping_enabled(ctx->xbridge->ms)
2217 && !eth_addr_is_broadcast(flow->dl_dst)
2218 && eth_addr_is_multicast(flow->dl_dst)
2219 && flow->dl_type == htons(ETH_TYPE_IP)) {
2220 struct mcast_snooping *ms = ctx->xbridge->ms;
2221 struct mcast_group *grp;
2223 if (flow->nw_proto == IPPROTO_IGMP) {
2224 if (ctx->xin->may_learn) {
2225 if (mcast_snooping_is_membership(flow->tp_src) ||
2226 mcast_snooping_is_query(flow->tp_src)) {
2227 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2232 if (mcast_snooping_is_membership(flow->tp_src)) {
2233 ovs_rwlock_rdlock(&ms->rwlock);
2234 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2235 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2236 * forward IGMP Membership Reports only to those ports where
2237 * multicast routers are attached. Alternatively stated: a
2238 * snooping switch should not forward IGMP Membership Reports
2239 * to ports on which only hosts are attached.
2240 * An administrative control may be provided to override this
2241 * restriction, allowing the report messages to be flooded to
2243 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2244 ovs_rwlock_unlock(&ms->rwlock);
2246 xlate_report(ctx, "multicast traffic, flooding");
2247 xlate_normal_flood(ctx, in_xbundle, vlan);
2251 if (ip_is_local_multicast(flow->nw_dst)) {
2252 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2253 * address in the 224.0.0.x range which are not IGMP must
2254 * be forwarded on all ports */
2255 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2256 xlate_normal_flood(ctx, in_xbundle, vlan);
2261 /* forwarding to group base ports */
2262 ovs_rwlock_rdlock(&ms->rwlock);
2263 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2265 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2266 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2267 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2269 if (mcast_snooping_flood_unreg(ms)) {
2270 xlate_report(ctx, "unregistered multicast, flooding");
2271 xlate_normal_flood(ctx, in_xbundle, vlan);
2273 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2274 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2277 ovs_rwlock_unlock(&ms->rwlock);
2279 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2280 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2281 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2282 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2285 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2286 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2287 if (mac_xbundle && mac_xbundle != in_xbundle) {
2288 xlate_report(ctx, "forwarding to learned port");
2289 output_normal(ctx, mac_xbundle, vlan);
2290 } else if (!mac_xbundle) {
2291 xlate_report(ctx, "learned port is unknown, dropping");
2293 xlate_report(ctx, "learned port is input port, dropping");
2296 xlate_report(ctx, "no learned MAC for destination, flooding");
2297 xlate_normal_flood(ctx, in_xbundle, vlan);
2302 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2303 * the number of packets out of UINT32_MAX to sample. The given
2304 * cookie is passed back in the callback for each sampled packet.
2307 compose_sample_action(const struct xbridge *xbridge,
2308 struct ofpbuf *odp_actions,
2309 const struct flow *flow,
2310 const uint32_t probability,
2311 const union user_action_cookie *cookie,
2312 const size_t cookie_size,
2313 const odp_port_t tunnel_out_port)
2315 size_t sample_offset, actions_offset;
2316 odp_port_t odp_port;
2320 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2322 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2324 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2326 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2327 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2328 flow_hash_5tuple(flow, 0));
2329 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2330 tunnel_out_port, odp_actions);
2332 nl_msg_end_nested(odp_actions, actions_offset);
2333 nl_msg_end_nested(odp_actions, sample_offset);
2334 return cookie_offset;
2338 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2339 odp_port_t odp_port, unsigned int n_outputs,
2340 union user_action_cookie *cookie)
2344 cookie->type = USER_ACTION_COOKIE_SFLOW;
2345 cookie->sflow.vlan_tci = vlan_tci;
2347 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2348 * port information") for the interpretation of cookie->output. */
2349 switch (n_outputs) {
2351 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2352 cookie->sflow.output = 0x40000000 | 256;
2356 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2358 cookie->sflow.output = ifindex;
2363 /* 0x80000000 means "multiple output ports. */
2364 cookie->sflow.output = 0x80000000 | n_outputs;
2369 /* Compose SAMPLE action for sFlow bridge sampling. */
2371 compose_sflow_action(const struct xbridge *xbridge,
2372 struct ofpbuf *odp_actions,
2373 const struct flow *flow,
2374 odp_port_t odp_port)
2376 uint32_t probability;
2377 union user_action_cookie cookie;
2379 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2383 probability = dpif_sflow_get_probability(xbridge->sflow);
2384 compose_sflow_cookie(xbridge, htons(0), odp_port,
2385 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2387 return compose_sample_action(xbridge, odp_actions, flow, probability,
2388 &cookie, sizeof cookie.sflow, ODPP_NONE);
2392 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2393 uint32_t obs_domain_id, uint32_t obs_point_id,
2394 union user_action_cookie *cookie)
2396 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2397 cookie->flow_sample.probability = probability;
2398 cookie->flow_sample.collector_set_id = collector_set_id;
2399 cookie->flow_sample.obs_domain_id = obs_domain_id;
2400 cookie->flow_sample.obs_point_id = obs_point_id;
2404 compose_ipfix_cookie(union user_action_cookie *cookie,
2405 odp_port_t output_odp_port)
2407 cookie->type = USER_ACTION_COOKIE_IPFIX;
2408 cookie->ipfix.output_odp_port = output_odp_port;
2411 /* Compose SAMPLE action for IPFIX bridge sampling. */
2413 compose_ipfix_action(const struct xbridge *xbridge,
2414 struct ofpbuf *odp_actions,
2415 const struct flow *flow,
2416 odp_port_t output_odp_port)
2418 uint32_t probability;
2419 union user_action_cookie cookie;
2420 odp_port_t tunnel_out_port = ODPP_NONE;
2422 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2426 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2428 if (output_odp_port == ODPP_NONE &&
2429 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2433 /* For output case, output_odp_port is valid*/
2434 if (output_odp_port != ODPP_NONE) {
2435 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2438 /* If tunnel sampling is enabled, put an additional option attribute:
2439 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2441 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2442 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2443 tunnel_out_port = output_odp_port;
2447 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2448 compose_ipfix_cookie(&cookie, output_odp_port);
2450 compose_sample_action(xbridge, odp_actions, flow, probability,
2451 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2454 /* SAMPLE action for sFlow must be first action in any given list of
2455 * actions. At this point we do not have all information required to
2456 * build it. So try to build sample action as complete as possible. */
2458 add_sflow_action(struct xlate_ctx *ctx)
2460 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2461 ctx->xout->odp_actions,
2462 &ctx->xin->flow, ODPP_NONE);
2463 ctx->sflow_odp_port = 0;
2464 ctx->sflow_n_outputs = 0;
2467 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2468 * of actions, eventually after the SAMPLE action for sFlow. */
2470 add_ipfix_action(struct xlate_ctx *ctx)
2472 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2473 &ctx->xin->flow, ODPP_NONE);
2477 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2479 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2480 &ctx->xin->flow, port);
2483 /* Fix SAMPLE action according to data collected while composing ODP actions.
2484 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2485 * USERSPACE action's user-cookie which is required for sflow. */
2487 fix_sflow_action(struct xlate_ctx *ctx)
2489 const struct flow *base = &ctx->base_flow;
2490 union user_action_cookie *cookie;
2492 if (!ctx->user_cookie_offset) {
2496 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2497 sizeof cookie->sflow);
2498 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2500 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2501 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2504 static enum slow_path_reason
2505 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2506 const struct xport *xport, const struct dp_packet *packet)
2508 struct flow_wildcards *wc = &ctx->xout->wc;
2509 const struct xbridge *xbridge = ctx->xbridge;
2513 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2515 cfm_process_heartbeat(xport->cfm, packet);
2518 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2520 bfd_process_packet(xport->bfd, flow, packet);
2521 /* If POLL received, immediately sends FINAL back. */
2522 if (bfd_should_send_packet(xport->bfd)) {
2523 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2527 } else if (xport->xbundle && xport->xbundle->lacp
2528 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2530 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2533 } else if ((xbridge->stp || xbridge->rstp) &&
2534 stp_should_process_flow(flow, wc)) {
2537 ? stp_process_packet(xport, packet)
2538 : rstp_process_packet(xport, packet);
2541 } else if (xport->lldp && lldp_should_process_flow(flow)) {
2543 lldp_process_packet(xport->lldp, packet);
2552 tnl_route_lookup_flow(const struct flow *oflow,
2553 ovs_be32 *ip, struct xport **out_port)
2555 char out_dev[IFNAMSIZ];
2556 struct xbridge *xbridge;
2557 struct xlate_cfg *xcfg;
2560 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2567 *ip = oflow->tunnel.ip_dst;
2570 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2573 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2574 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2577 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2578 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2589 xlate_flood_packet(struct xbridge *xbridge, struct dp_packet *packet)
2591 struct ofpact_output output;
2594 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2595 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2596 flow_extract(packet, &flow);
2597 flow.in_port.ofp_port = OFPP_NONE;
2598 output.port = OFPP_FLOOD;
2601 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2602 &output.ofpact, sizeof output,
2607 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2608 ovs_be32 ip_src, ovs_be32 ip_dst)
2610 struct xbridge *xbridge = out_dev->xbridge;
2611 struct dp_packet packet;
2613 dp_packet_init(&packet, 0);
2614 compose_arp(&packet, eth_src, ip_src, ip_dst);
2616 xlate_flood_packet(xbridge, &packet);
2617 dp_packet_uninit(&packet);
2621 build_tunnel_send(const struct xlate_ctx *ctx, const struct xport *xport,
2622 const struct flow *flow, odp_port_t tunnel_odp_port)
2624 struct ovs_action_push_tnl tnl_push_data;
2625 struct xport *out_dev = NULL;
2626 ovs_be32 s_ip, d_ip = 0;
2627 uint8_t smac[ETH_ADDR_LEN];
2628 uint8_t dmac[ETH_ADDR_LEN];
2631 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2636 /* Use mac addr of bridge port of the peer. */
2637 err = netdev_get_etheraddr(out_dev->netdev, smac);
2642 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2647 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2649 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2652 if (ctx->xin->xcache) {
2653 struct xc_entry *entry;
2655 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2656 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2657 sizeof entry->u.tnl_arp_cache.br_name);
2658 entry->u.tnl_arp_cache.d_ip = d_ip;
2660 err = tnl_port_build_header(xport->ofport, flow,
2661 dmac, smac, s_ip, &tnl_push_data);
2665 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2666 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2667 odp_put_tnl_push_action(ctx->xout->odp_actions, &tnl_push_data);
2672 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2675 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2676 struct flow_wildcards *wc = &ctx->xout->wc;
2677 struct flow *flow = &ctx->xin->flow;
2678 struct flow_tnl flow_tnl;
2679 ovs_be16 flow_vlan_tci;
2680 uint32_t flow_pkt_mark;
2681 uint8_t flow_nw_tos;
2682 odp_port_t out_port, odp_port;
2683 bool tnl_push_pop_send = false;
2686 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2687 * before traversing a patch port. */
2688 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 31);
2689 memset(&flow_tnl, 0, sizeof flow_tnl);
2692 xlate_report(ctx, "Nonexistent output port");
2694 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2695 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2697 } else if (check_stp) {
2698 if (is_stp(&ctx->base_flow)) {
2699 if (!xport_stp_should_forward_bpdu(xport) &&
2700 !xport_rstp_should_manage_bpdu(xport)) {
2701 if (ctx->xbridge->stp != NULL) {
2702 xlate_report(ctx, "STP not in listening state, "
2703 "skipping bpdu output");
2704 } else if (ctx->xbridge->rstp != NULL) {
2705 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2706 "skipping bpdu output");
2710 } else if (!xport_stp_forward_state(xport) ||
2711 !xport_rstp_forward_state(xport)) {
2712 if (ctx->xbridge->stp != NULL) {
2713 xlate_report(ctx, "STP not in forwarding state, "
2715 } else if (ctx->xbridge->rstp != NULL) {
2716 xlate_report(ctx, "RSTP not in forwarding state, "
2723 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2724 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2729 const struct xport *peer = xport->peer;
2730 struct flow old_flow = ctx->xin->flow;
2731 enum slow_path_reason special;
2732 uint8_t table_id = rule_dpif_lookup_get_init_table_id(&ctx->xin->flow);
2734 ctx->xbridge = peer->xbridge;
2735 flow->in_port.ofp_port = peer->ofp_port;
2736 flow->metadata = htonll(0);
2737 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2738 memset(flow->regs, 0, sizeof flow->regs);
2739 flow->actset_output = OFPP_UNSET;
2741 special = process_special(ctx, &ctx->xin->flow, peer,
2744 ctx->xout->slow |= special;
2745 } else if (may_receive(peer, ctx)) {
2746 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2747 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2750 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2751 * the learning action look at the packet, then drop it. */
2752 struct flow old_base_flow = ctx->base_flow;
2753 size_t old_size = ctx->xout->odp_actions->size;
2754 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2755 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2757 ctx->xout->mirrors = old_mirrors;
2758 ctx->base_flow = old_base_flow;
2759 ctx->xout->odp_actions->size = old_size;
2763 ctx->xin->flow = old_flow;
2764 ctx->xbridge = xport->xbridge;
2766 if (ctx->xin->resubmit_stats) {
2767 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2768 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2770 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2773 if (ctx->xin->xcache) {
2774 struct xc_entry *entry;
2776 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2777 entry->u.dev.tx = netdev_ref(xport->netdev);
2778 entry->u.dev.rx = netdev_ref(peer->netdev);
2779 entry->u.dev.bfd = bfd_ref(peer->bfd);
2784 flow_vlan_tci = flow->vlan_tci;
2785 flow_pkt_mark = flow->pkt_mark;
2786 flow_nw_tos = flow->nw_tos;
2788 if (count_skb_priorities(xport)) {
2789 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2790 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2791 wc->masks.nw_tos |= IP_DSCP_MASK;
2792 flow->nw_tos &= ~IP_DSCP_MASK;
2793 flow->nw_tos |= dscp;
2797 if (xport->is_tunnel) {
2798 /* Save tunnel metadata so that changes made due to
2799 * the Logical (tunnel) Port are not visible for any further
2800 * matches, while explicit set actions on tunnel metadata are.
2802 flow_tnl = flow->tunnel;
2803 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2804 if (odp_port == ODPP_NONE) {
2805 xlate_report(ctx, "Tunneling decided against output");
2806 goto out; /* restore flow_nw_tos */
2808 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2809 xlate_report(ctx, "Not tunneling to our own address");
2810 goto out; /* restore flow_nw_tos */
2812 if (ctx->xin->resubmit_stats) {
2813 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2815 if (ctx->xin->xcache) {
2816 struct xc_entry *entry;
2818 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2819 entry->u.dev.tx = netdev_ref(xport->netdev);
2821 out_port = odp_port;
2822 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2823 tnl_push_pop_send = true;
2825 commit_odp_tunnel_action(flow, &ctx->base_flow,
2826 ctx->xout->odp_actions);
2827 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2830 odp_port = xport->odp_port;
2831 out_port = odp_port;
2832 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2833 ofp_port_t vlandev_port;
2835 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2836 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2837 ofp_port, flow->vlan_tci);
2838 if (vlandev_port != ofp_port) {
2839 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2840 flow->vlan_tci = htons(0);
2845 if (out_port != ODPP_NONE) {
2846 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2847 ctx->xout->odp_actions,
2849 ctx->xbridge->masked_set_action);
2851 if (ctx->use_recirc) {
2852 struct ovs_action_hash *act_hash;
2853 struct xlate_recirc *xr = &ctx->recirc;
2856 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2857 OVS_ACTION_ATTR_HASH,
2859 act_hash->hash_alg = xr->hash_alg;
2860 act_hash->hash_basis = xr->hash_basis;
2862 /* Recirc action. */
2863 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2867 if (tnl_push_pop_send) {
2868 build_tunnel_send(ctx, xport, flow, odp_port);
2869 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2871 odp_port_t odp_tnl_port = ODPP_NONE;
2873 /* XXX: Write better Filter for tunnel port. We can use inport
2874 * int tunnel-port flow to avoid these checks completely. */
2875 if (ofp_port == OFPP_LOCAL &&
2876 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2878 odp_tnl_port = tnl_port_map_lookup(flow, wc);
2881 if (odp_tnl_port != ODPP_NONE) {
2882 nl_msg_put_odp_port(ctx->xout->odp_actions,
2883 OVS_ACTION_ATTR_TUNNEL_POP,
2886 /* Tunnel push-pop action is not compatible with
2888 add_ipfix_output_action(ctx, out_port);
2889 nl_msg_put_odp_port(ctx->xout->odp_actions,
2890 OVS_ACTION_ATTR_OUTPUT,
2896 ctx->sflow_odp_port = odp_port;
2897 ctx->sflow_n_outputs++;
2898 ctx->xout->nf_output_iface = ofp_port;
2903 flow->vlan_tci = flow_vlan_tci;
2904 flow->pkt_mark = flow_pkt_mark;
2905 flow->nw_tos = flow_nw_tos;
2909 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
2911 compose_output_action__(ctx, ofp_port, true);
2915 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
2917 struct rule_dpif *old_rule = ctx->rule;
2918 const struct rule_actions *actions;
2920 if (ctx->xin->resubmit_stats) {
2921 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2927 actions = rule_dpif_get_actions(rule);
2928 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2929 ctx->rule = old_rule;
2934 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2936 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2938 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2939 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2940 MAX_RESUBMIT_RECURSION);
2941 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2942 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2943 } else if (ctx->xout->odp_actions->size > UINT16_MAX) {
2944 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2945 } else if (ctx->stack.size >= 65536) {
2946 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2955 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2956 bool may_packet_in, bool honor_table_miss)
2958 if (xlate_resubmit_resource_check(ctx)) {
2959 struct flow_wildcards *wc;
2960 uint8_t old_table_id = ctx->table_id;
2961 struct rule_dpif *rule;
2963 ctx->table_id = table_id;
2964 wc = (ctx->xin->skip_wildcards) ? NULL : &ctx->xout->wc;
2966 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2967 &ctx->xin->flow, wc,
2968 ctx->xin->xcache != NULL,
2969 ctx->xin->resubmit_stats,
2970 &ctx->table_id, in_port,
2971 may_packet_in, honor_table_miss);
2973 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
2974 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
2978 /* Fill in the cache entry here instead of xlate_recursively
2979 * to make the reference counting more explicit. We take a
2980 * reference in the lookups above if we are going to cache the
2982 if (ctx->xin->xcache) {
2983 struct xc_entry *entry;
2985 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2986 entry->u.rule = rule;
2988 xlate_recursively(ctx, rule);
2991 ctx->table_id = old_table_id;
2999 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3000 struct ofputil_bucket *bucket)
3002 if (ctx->xin->resubmit_stats) {
3003 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3005 if (ctx->xin->xcache) {
3006 struct xc_entry *entry;
3008 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3009 entry->u.group.group = group_dpif_ref(group);
3010 entry->u.group.bucket = bucket;
3015 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3017 uint64_t action_list_stub[1024 / 8];
3018 struct ofpbuf action_list, action_set;
3020 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3021 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3023 ofpacts_execute_action_set(&action_list, &action_set);
3025 do_xlate_actions(action_list.data, action_list.size, ctx);
3028 ofpbuf_uninit(&action_set);
3029 ofpbuf_uninit(&action_list);
3033 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3035 struct ofputil_bucket *bucket;
3036 const struct ovs_list *buckets;
3037 struct flow old_flow = ctx->xin->flow;
3039 group_dpif_get_buckets(group, &buckets);
3041 LIST_FOR_EACH (bucket, list_node, buckets) {
3042 xlate_group_bucket(ctx, bucket);
3043 /* Roll back flow to previous state.
3044 * This is equivalent to cloning the packet for each bucket.
3046 * As a side effect any subsequently applied actions will
3047 * also effectively be applied to a clone of the packet taken
3048 * just before applying the all or indirect group. */
3049 ctx->xin->flow = old_flow;
3051 xlate_group_stats(ctx, group, NULL);
3055 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3057 struct ofputil_bucket *bucket;
3059 bucket = group_first_live_bucket(ctx, group, 0);
3061 xlate_group_bucket(ctx, bucket);
3062 xlate_group_stats(ctx, group, bucket);
3067 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3069 struct flow_wildcards *wc = &ctx->xout->wc;
3070 struct ofputil_bucket *bucket;
3073 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3074 bucket = group_best_live_bucket(ctx, group, basis);
3076 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3077 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3078 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3079 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3080 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3081 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3082 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3083 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3084 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3086 xlate_group_bucket(ctx, bucket);
3087 xlate_group_stats(ctx, group, bucket);
3092 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3094 ctx->in_group = true;
3096 switch (group_dpif_get_type(group)) {
3098 case OFPGT11_INDIRECT:
3099 xlate_all_group(ctx, group);
3101 case OFPGT11_SELECT:
3102 xlate_select_group(ctx, group);
3105 xlate_ff_group(ctx, group);
3110 group_dpif_unref(group);
3112 ctx->in_group = false;
3116 xlate_group_resource_check(struct xlate_ctx *ctx)
3118 if (!xlate_resubmit_resource_check(ctx)) {
3120 } else if (ctx->in_group) {
3121 /* Prevent nested translation of OpenFlow groups.
3123 * OpenFlow allows this restriction. We enforce this restriction only
3124 * because, with the current architecture, we would otherwise have to
3125 * take a possibly recursive read lock on the ofgroup rwlock, which is
3126 * unsafe given that POSIX allows taking a read lock to block if there
3127 * is a thread blocked on taking the write lock. Other solutions
3128 * without this restriction are also possible, but seem unwarranted
3129 * given the current limited use of groups. */
3130 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3132 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
3140 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3142 if (xlate_group_resource_check(ctx)) {
3143 struct group_dpif *group;
3146 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3148 xlate_group_action__(ctx, group);
3158 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3159 const struct ofpact_resubmit *resubmit)
3163 bool may_packet_in = false;
3164 bool honor_table_miss = false;
3166 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3167 /* Still allow missed packets to be sent to the controller
3168 * if resubmitting from an internal table. */
3169 may_packet_in = true;
3170 honor_table_miss = true;
3173 in_port = resubmit->in_port;
3174 if (in_port == OFPP_IN_PORT) {
3175 in_port = ctx->xin->flow.in_port.ofp_port;
3178 table_id = resubmit->table_id;
3179 if (table_id == 255) {
3180 table_id = ctx->table_id;
3183 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3188 flood_packets(struct xlate_ctx *ctx, bool all)
3190 const struct xport *xport;
3192 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3193 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3198 compose_output_action__(ctx, xport->ofp_port, false);
3199 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3200 compose_output_action(ctx, xport->ofp_port);
3204 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3208 execute_controller_action(struct xlate_ctx *ctx, int len,
3209 enum ofp_packet_in_reason reason,
3210 uint16_t controller_id)
3212 struct ofproto_packet_in *pin;
3213 struct dp_packet *packet;
3215 ctx->xout->slow |= SLOW_CONTROLLER;
3216 if (!ctx->xin->packet) {
3220 packet = dp_packet_clone(ctx->xin->packet);
3222 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3223 ctx->xout->odp_actions,
3225 ctx->xbridge->masked_set_action);
3227 odp_execute_actions(NULL, &packet, 1, false,
3228 ctx->xout->odp_actions->data,
3229 ctx->xout->odp_actions->size, NULL);
3231 pin = xmalloc(sizeof *pin);
3232 pin->up.packet_len = dp_packet_size(packet);
3233 pin->up.packet = dp_packet_steal_data(packet);
3234 pin->up.reason = reason;
3235 pin->up.table_id = ctx->table_id;
3236 pin->up.cookie = (ctx->rule
3237 ? rule_dpif_get_flow_cookie(ctx->rule)
3240 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3242 pin->controller_id = controller_id;
3243 pin->send_len = len;
3244 /* If a rule is a table-miss rule then this is
3245 * a table-miss handled by a table-miss rule.
3247 * Else, if rule is internal and has a controller action,
3248 * the later being implied by the rule being processed here,
3249 * then this is a table-miss handled without a table-miss rule.
3251 * Otherwise this is not a table-miss. */
3252 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3254 if (rule_dpif_is_table_miss(ctx->rule)) {
3255 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3256 } else if (rule_dpif_is_internal(ctx->rule)) {
3257 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3260 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3261 dp_packet_delete(packet);
3265 compose_recirculate_action(struct xlate_ctx *ctx,
3266 const struct ofpact *ofpacts_base,
3267 const struct ofpact *ofpact_current,
3268 size_t ofpacts_base_len)
3272 unsigned ofpacts_len;
3275 struct ofpbuf ofpacts;
3279 ofpacts_len = ofpacts_base_len -
3280 ((uint8_t *)ofpact_current - (uint8_t *)ofpacts_base);
3283 id = rule_dpif_get_recirc_id(ctx->rule);
3285 /* In the case where ctx has no rule then allocate a recirc id.
3286 * The life-cycle of this recirc id is managed by associating it
3287 * with the internal rule that is created to to handle
3288 * recirculation below.
3290 * The known use-case of this is packet_out which
3291 * translates actions without a rule */
3292 id = ofproto_dpif_alloc_recirc_id(ctx->xbridge->ofproto);
3295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3296 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3300 match_init_catchall(&match);
3301 match_set_recirc_id(&match, id);
3302 ofpbuf_use_const(&ofpacts, ofpact_current, ofpacts_len);
3303 error = ofproto_dpif_add_internal_flow(ctx->xbridge->ofproto, &match,
3304 RECIRC_RULE_PRIORITY,
3305 RECIRC_TIMEOUT, &ofpacts, &rule);
3307 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3308 VLOG_ERR_RL(&rl, "Failed to add post recirculation flow %s",
3309 match_to_string(&match, 0));
3311 ofproto_dpif_free_recirc_id(ctx->xbridge->ofproto, id);
3315 /* If ctx has no rule then associate the recirc id, which
3316 * was allocated above, with the internal rule. This allows
3317 * the recirc id to be released when the internal rule times out. */
3319 rule_set_recirc_id(rule, id);
3322 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3323 ctx->xout->odp_actions,
3325 ctx->xbridge->masked_set_action);
3326 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3330 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3332 struct flow_wildcards *wc = &ctx->xout->wc;
3333 struct flow *flow = &ctx->xin->flow;
3336 ovs_assert(eth_type_mpls(mpls->ethertype));
3338 n = flow_count_mpls_labels(flow, wc);
3340 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3341 ctx->xout->odp_actions,
3343 ctx->xbridge->masked_set_action);
3344 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3345 if (ctx->xin->packet != NULL) {
3346 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3347 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3348 "MPLS push action can't be performed as it would "
3349 "have more MPLS LSEs than the %d supported.",
3350 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3356 flow_push_mpls(flow, n, mpls->ethertype, wc);
3360 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3362 struct flow_wildcards *wc = &ctx->xout->wc;
3363 struct flow *flow = &ctx->xin->flow;
3364 int n = flow_count_mpls_labels(flow, wc);
3366 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3367 if (ctx->xbridge->enable_recirc) {
3368 ctx->was_mpls = true;
3370 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3371 if (ctx->xin->packet != NULL) {
3372 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3373 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3374 "MPLS pop action can't be performed as it has "
3375 "more MPLS LSEs than the %d supported.",
3376 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3379 ofpbuf_clear(ctx->xout->odp_actions);
3384 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3386 struct flow *flow = &ctx->xin->flow;
3388 if (!is_ip_any(flow)) {
3392 ctx->xout->wc.masks.nw_ttl = 0xff;
3393 if (flow->nw_ttl > 1) {
3399 for (i = 0; i < ids->n_controllers; i++) {
3400 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3404 /* Stop processing for current table. */
3410 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3412 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3413 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3414 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3419 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3421 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3422 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3423 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3428 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3430 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3431 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3432 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3437 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3439 struct flow *flow = &ctx->xin->flow;
3440 struct flow_wildcards *wc = &ctx->xout->wc;
3442 if (eth_type_mpls(flow->dl_type)) {
3443 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3445 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3448 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3451 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3455 /* Stop processing for current table. */
3460 xlate_output_action(struct xlate_ctx *ctx,
3461 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3463 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3465 ctx->xout->nf_output_iface = NF_OUT_DROP;
3469 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
3472 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3473 0, may_packet_in, true);
3479 flood_packets(ctx, false);
3482 flood_packets(ctx, true);
3484 case OFPP_CONTROLLER:
3485 execute_controller_action(ctx, max_len,
3486 (ctx->in_group ? OFPR_GROUP
3487 : ctx->in_action_set ? OFPR_ACTION_SET
3495 if (port != ctx->xin->flow.in_port.ofp_port) {
3496 compose_output_action(ctx, port);
3498 xlate_report(ctx, "skipping output to input port");
3503 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3504 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3505 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3506 ctx->xout->nf_output_iface = prev_nf_output_iface;
3507 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3508 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3509 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3514 xlate_output_reg_action(struct xlate_ctx *ctx,
3515 const struct ofpact_output_reg *or)
3517 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3518 if (port <= UINT16_MAX) {
3519 union mf_subvalue value;
3521 memset(&value, 0xff, sizeof value);
3522 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3523 xlate_output_action(ctx, u16_to_ofp(port),
3524 or->max_len, false);
3529 xlate_enqueue_action(struct xlate_ctx *ctx,
3530 const struct ofpact_enqueue *enqueue)
3532 ofp_port_t ofp_port = enqueue->port;
3533 uint32_t queue_id = enqueue->queue;
3534 uint32_t flow_priority, priority;
3537 /* Translate queue to priority. */
3538 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3540 /* Fall back to ordinary output action. */
3541 xlate_output_action(ctx, enqueue->port, 0, false);
3545 /* Check output port. */
3546 if (ofp_port == OFPP_IN_PORT) {
3547 ofp_port = ctx->xin->flow.in_port.ofp_port;
3548 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3552 /* Add datapath actions. */
3553 flow_priority = ctx->xin->flow.skb_priority;
3554 ctx->xin->flow.skb_priority = priority;
3555 compose_output_action(ctx, ofp_port);
3556 ctx->xin->flow.skb_priority = flow_priority;
3558 /* Update NetFlow output port. */
3559 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3560 ctx->xout->nf_output_iface = ofp_port;
3561 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3562 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3567 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3569 uint32_t skb_priority;
3571 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3572 ctx->xin->flow.skb_priority = skb_priority;
3574 /* Couldn't translate queue to a priority. Nothing to do. A warning
3575 * has already been logged. */
3580 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3582 const struct xbridge *xbridge = xbridge_;
3593 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3596 port = get_ofp_port(xbridge, ofp_port);
3597 return port ? port->may_enable : false;
3602 xlate_bundle_action(struct xlate_ctx *ctx,
3603 const struct ofpact_bundle *bundle)
3607 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3609 CONST_CAST(struct xbridge *, ctx->xbridge));
3610 if (bundle->dst.field) {
3611 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3614 xlate_output_action(ctx, port, 0, false);
3619 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3620 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3622 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3623 if (ctx->xin->may_learn) {
3624 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3629 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3631 ctx->xout->has_learn = true;
3632 learn_mask(learn, &ctx->xout->wc);
3634 if (ctx->xin->xcache) {
3635 struct xc_entry *entry;
3637 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3638 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3639 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3640 entry->u.learn.ofpacts = ofpbuf_new(64);
3641 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3642 entry->u.learn.ofpacts);
3643 } else if (ctx->xin->may_learn) {
3644 uint64_t ofpacts_stub[1024 / 8];
3645 struct ofputil_flow_mod fm;
3646 struct ofpbuf ofpacts;
3648 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3649 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3650 ofpbuf_uninit(&ofpacts);
3655 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3656 uint16_t idle_timeout, uint16_t hard_timeout)
3658 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3659 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3664 xlate_fin_timeout(struct xlate_ctx *ctx,
3665 const struct ofpact_fin_timeout *oft)
3668 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3669 oft->fin_idle_timeout, oft->fin_hard_timeout);
3670 if (ctx->xin->xcache) {
3671 struct xc_entry *entry;
3673 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3674 /* XC_RULE already holds a reference on the rule, none is taken
3676 entry->u.fin.rule = ctx->rule;
3677 entry->u.fin.idle = oft->fin_idle_timeout;
3678 entry->u.fin.hard = oft->fin_hard_timeout;
3684 xlate_sample_action(struct xlate_ctx *ctx,
3685 const struct ofpact_sample *os)
3687 union user_action_cookie cookie;
3688 /* Scale the probability from 16-bit to 32-bit while representing
3689 * the same percentage. */
3690 uint32_t probability = (os->probability << 16) | os->probability;
3692 if (!ctx->xbridge->variable_length_userdata) {
3693 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3695 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3696 "lacks support (needs Linux 3.10+ or kernel module from "
3701 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3702 ctx->xout->odp_actions,
3704 ctx->xbridge->masked_set_action);
3706 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3707 os->obs_domain_id, os->obs_point_id, &cookie);
3708 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions,
3709 &ctx->xin->flow, probability, &cookie,
3710 sizeof cookie.flow_sample, ODPP_NONE);
3714 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3716 if (xport->config & (is_stp(&ctx->xin->flow)
3717 ? OFPUTIL_PC_NO_RECV_STP
3718 : OFPUTIL_PC_NO_RECV)) {
3722 /* Only drop packets here if both forwarding and learning are
3723 * disabled. If just learning is enabled, we need to have
3724 * OFPP_NORMAL and the learning action have a look at the packet
3725 * before we can drop it. */
3726 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3727 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3735 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3737 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3738 size_t on_len = ofpact_nest_get_action_len(on);
3739 const struct ofpact *inner;
3741 /* Maintain actset_output depending on the contents of the action set:
3743 * - OFPP_UNSET, if there is no "output" action.
3745 * - The output port, if there is an "output" action and no "group"
3748 * - OFPP_UNSET, if there is a "group" action.
3750 if (!ctx->action_set_has_group) {
3751 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3752 if (inner->type == OFPACT_OUTPUT) {
3753 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3754 } else if (inner->type == OFPACT_GROUP) {
3755 ctx->xin->flow.actset_output = OFPP_UNSET;
3756 ctx->action_set_has_group = true;
3761 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3762 ofpact_pad(&ctx->action_set);
3766 xlate_action_set(struct xlate_ctx *ctx)
3768 uint64_t action_list_stub[1024 / 64];
3769 struct ofpbuf action_list;
3771 ctx->in_action_set = true;
3772 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3773 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3774 do_xlate_actions(action_list.data, action_list.size, ctx);
3775 ctx->in_action_set = false;
3776 ofpbuf_uninit(&action_list);
3780 ofpact_needs_recirculation_after_mpls(const struct ofpact *a, struct xlate_ctx *ctx)
3782 struct flow_wildcards *wc = &ctx->xout->wc;
3783 struct flow *flow = &ctx->xin->flow;
3785 if (!ctx->was_mpls) {
3792 case OFPACT_CONTROLLER:
3793 case OFPACT_STRIP_VLAN:
3794 case OFPACT_SET_VLAN_PCP:
3795 case OFPACT_SET_VLAN_VID:
3796 case OFPACT_ENQUEUE:
3797 case OFPACT_PUSH_VLAN:
3798 case OFPACT_SET_ETH_SRC:
3799 case OFPACT_SET_ETH_DST:
3800 case OFPACT_SET_TUNNEL:
3801 case OFPACT_SET_QUEUE:
3802 case OFPACT_POP_QUEUE:
3803 case OFPACT_CONJUNCTION:
3805 case OFPACT_OUTPUT_REG:
3808 case OFPACT_WRITE_METADATA:
3809 case OFPACT_WRITE_ACTIONS:
3810 case OFPACT_CLEAR_ACTIONS:
3814 case OFPACT_POP_MPLS:
3815 case OFPACT_DEC_MPLS_TTL:
3816 case OFPACT_SET_MPLS_TTL:
3817 case OFPACT_SET_MPLS_TC:
3818 case OFPACT_SET_MPLS_LABEL:
3819 case OFPACT_SET_IPV4_SRC:
3820 case OFPACT_SET_IPV4_DST:
3821 case OFPACT_SET_IP_DSCP:
3822 case OFPACT_SET_IP_ECN:
3823 case OFPACT_SET_IP_TTL:
3824 case OFPACT_SET_L4_SRC_PORT:
3825 case OFPACT_SET_L4_DST_PORT:
3826 case OFPACT_RESUBMIT:
3827 case OFPACT_STACK_PUSH:
3828 case OFPACT_STACK_POP:
3829 case OFPACT_DEC_TTL:
3830 case OFPACT_MULTIPATH:
3833 case OFPACT_FIN_TIMEOUT:
3834 case OFPACT_GOTO_TABLE:
3837 case OFPACT_REG_MOVE:
3838 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
3839 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
3841 case OFPACT_SET_FIELD:
3842 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
3844 case OFPACT_PUSH_MPLS:
3845 /* Recirculate if it is an IP packet with a zero ttl. This may
3846 * indicate that the packet was previously MPLS and an MPLS pop action
3847 * converted it to IP. In this case recirculating should reveal the IP
3848 * TTL which is used as the basis for a new MPLS LSE. */
3849 return (!flow_count_mpls_labels(flow, wc)
3850 && flow->nw_ttl == 0
3851 && is_ip_any(flow));
3858 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
3859 struct xlate_ctx *ctx)
3861 struct flow_wildcards *wc = &ctx->xout->wc;
3862 struct flow *flow = &ctx->xin->flow;
3863 const struct ofpact *a;
3865 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3866 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
3868 /* dl_type already in the mask, not set below. */
3870 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3871 struct ofpact_controller *controller;
3872 const struct ofpact_metadata *metadata;
3873 const struct ofpact_set_field *set_field;
3874 const struct mf_field *mf;
3880 if (ofpact_needs_recirculation_after_mpls(a, ctx)) {
3881 compose_recirculate_action(ctx, ofpacts, a, ofpacts_len);
3887 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
3888 ofpact_get_OUTPUT(a)->max_len, true);
3892 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
3897 case OFPACT_CONTROLLER:
3898 controller = ofpact_get_CONTROLLER(a);
3899 execute_controller_action(ctx, controller->max_len,
3901 controller->controller_id);
3904 case OFPACT_ENQUEUE:
3905 memset(&wc->masks.skb_priority, 0xff,
3906 sizeof wc->masks.skb_priority);
3907 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
3910 case OFPACT_SET_VLAN_VID:
3911 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3912 if (flow->vlan_tci & htons(VLAN_CFI) ||
3913 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
3914 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
3915 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
3920 case OFPACT_SET_VLAN_PCP:
3921 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
3922 if (flow->vlan_tci & htons(VLAN_CFI) ||
3923 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
3924 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
3925 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
3926 << VLAN_PCP_SHIFT) | VLAN_CFI);
3930 case OFPACT_STRIP_VLAN:
3931 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3932 flow->vlan_tci = htons(0);
3935 case OFPACT_PUSH_VLAN:
3936 /* XXX 802.1AD(QinQ) */
3937 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3938 flow->vlan_tci = htons(VLAN_CFI);
3941 case OFPACT_SET_ETH_SRC:
3942 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3943 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
3946 case OFPACT_SET_ETH_DST:
3947 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3948 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
3951 case OFPACT_SET_IPV4_SRC:
3952 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3953 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3954 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
3958 case OFPACT_SET_IPV4_DST:
3959 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3960 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3961 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
3965 case OFPACT_SET_IP_DSCP:
3966 if (is_ip_any(flow)) {
3967 wc->masks.nw_tos |= IP_DSCP_MASK;
3968 flow->nw_tos &= ~IP_DSCP_MASK;
3969 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
3973 case OFPACT_SET_IP_ECN:
3974 if (is_ip_any(flow)) {
3975 wc->masks.nw_tos |= IP_ECN_MASK;
3976 flow->nw_tos &= ~IP_ECN_MASK;
3977 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
3981 case OFPACT_SET_IP_TTL:
3982 if (is_ip_any(flow)) {
3983 wc->masks.nw_ttl = 0xff;
3984 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
3988 case OFPACT_SET_L4_SRC_PORT:
3989 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3990 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3991 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3992 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
3996 case OFPACT_SET_L4_DST_PORT:
3997 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3998 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3999 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4000 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4004 case OFPACT_RESUBMIT:
4005 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4008 case OFPACT_SET_TUNNEL:
4009 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4012 case OFPACT_SET_QUEUE:
4013 memset(&wc->masks.skb_priority, 0xff,
4014 sizeof wc->masks.skb_priority);
4015 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4018 case OFPACT_POP_QUEUE:
4019 memset(&wc->masks.skb_priority, 0xff,
4020 sizeof wc->masks.skb_priority);
4021 flow->skb_priority = ctx->orig_skb_priority;
4024 case OFPACT_REG_MOVE:
4025 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4028 case OFPACT_SET_FIELD:
4029 set_field = ofpact_get_SET_FIELD(a);
4030 mf = set_field->field;
4032 /* Set field action only ever overwrites packet's outermost
4033 * applicable header fields. Do nothing if no header exists. */
4034 if (mf->id == MFF_VLAN_VID) {
4035 wc->masks.vlan_tci |= htons(VLAN_CFI);
4036 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4039 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4040 /* 'dl_type' is already unwildcarded. */
4041 && !eth_type_mpls(flow->dl_type)) {
4044 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4045 * header field on a packet that does not have them. */
4046 mf_mask_field_and_prereqs(mf, &wc->masks);
4047 if (mf_are_prereqs_ok(mf, flow)) {
4048 mf_set_flow_value_masked(mf, &set_field->value,
4049 &set_field->mask, flow);
4053 case OFPACT_STACK_PUSH:
4054 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4058 case OFPACT_STACK_POP:
4059 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4063 case OFPACT_PUSH_MPLS:
4064 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4067 case OFPACT_POP_MPLS:
4068 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4071 case OFPACT_SET_MPLS_LABEL:
4072 compose_set_mpls_label_action(
4073 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4076 case OFPACT_SET_MPLS_TC:
4077 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4080 case OFPACT_SET_MPLS_TTL:
4081 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4084 case OFPACT_DEC_MPLS_TTL:
4085 if (compose_dec_mpls_ttl_action(ctx)) {
4090 case OFPACT_DEC_TTL:
4091 wc->masks.nw_ttl = 0xff;
4092 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4098 /* Nothing to do. */
4101 case OFPACT_MULTIPATH:
4102 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4106 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4109 case OFPACT_OUTPUT_REG:
4110 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4114 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4117 case OFPACT_CONJUNCTION: {
4118 /* A flow with a "conjunction" action represents part of a special
4119 * kind of "set membership match". Such a flow should not actually
4120 * get executed, but it could via, say, a "packet-out", even though
4121 * that wouldn't be useful. Log it to help debugging. */
4122 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4123 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4131 case OFPACT_FIN_TIMEOUT:
4132 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4133 ctx->xout->has_fin_timeout = true;
4134 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4137 case OFPACT_CLEAR_ACTIONS:
4138 ofpbuf_clear(&ctx->action_set);
4139 ctx->xin->flow.actset_output = OFPP_UNSET;
4140 ctx->action_set_has_group = false;
4143 case OFPACT_WRITE_ACTIONS:
4144 xlate_write_actions(ctx, a);
4147 case OFPACT_WRITE_METADATA:
4148 metadata = ofpact_get_WRITE_METADATA(a);
4149 flow->metadata &= ~metadata->mask;
4150 flow->metadata |= metadata->metadata & metadata->mask;
4154 /* Not implemented yet. */
4157 case OFPACT_GOTO_TABLE: {
4158 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4160 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4161 * than ogt->table_id. This is to allow goto_table actions that
4162 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4163 * after recirculation. */
4164 ovs_assert(ctx->table_id == TBL_INTERNAL
4165 || ctx->table_id < ogt->table_id);
4166 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4167 ogt->table_id, true, true);
4172 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4179 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4180 const struct flow *flow, ofp_port_t in_port,
4181 struct rule_dpif *rule, uint16_t tcp_flags,
4182 const struct dp_packet *packet)
4184 xin->ofproto = ofproto;
4186 xin->flow.in_port.ofp_port = in_port;
4187 xin->flow.actset_output = OFPP_UNSET;
4188 xin->packet = packet;
4189 xin->may_learn = packet != NULL;
4192 xin->ofpacts = NULL;
4193 xin->ofpacts_len = 0;
4194 xin->tcp_flags = tcp_flags;
4195 xin->resubmit_hook = NULL;
4196 xin->report_hook = NULL;
4197 xin->resubmit_stats = NULL;
4198 xin->skip_wildcards = false;
4199 xin->odp_actions = NULL;
4203 xlate_out_uninit(struct xlate_out *xout)
4205 if (xout && xout->odp_actions == &xout->odp_actions_buf) {
4206 ofpbuf_uninit(xout->odp_actions);
4210 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4211 * into datapath actions, using 'ctx', and discards the datapath actions. */
4213 xlate_actions_for_side_effects(struct xlate_in *xin)
4215 struct xlate_out xout;
4217 xlate_actions(xin, &xout);
4218 xlate_out_uninit(&xout);
4222 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
4225 dst->slow = src->slow;
4226 dst->has_learn = src->has_learn;
4227 dst->has_normal = src->has_normal;
4228 dst->has_fin_timeout = src->has_fin_timeout;
4229 dst->nf_output_iface = src->nf_output_iface;
4230 dst->mirrors = src->mirrors;
4232 dst->odp_actions = &dst->odp_actions_buf;
4233 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
4234 sizeof dst->odp_actions_stub);
4235 ofpbuf_put(dst->odp_actions, src->odp_actions->data, src->odp_actions->size);
4238 static struct skb_priority_to_dscp *
4239 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4241 struct skb_priority_to_dscp *pdscp;
4244 hash = hash_int(skb_priority, 0);
4245 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4246 if (pdscp->skb_priority == skb_priority) {
4254 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4257 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4258 *dscp = pdscp ? pdscp->dscp : 0;
4259 return pdscp != NULL;
4263 count_skb_priorities(const struct xport *xport)
4265 return hmap_count(&xport->skb_priorities);
4269 clear_skb_priorities(struct xport *xport)
4271 struct skb_priority_to_dscp *pdscp, *next;
4273 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4274 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4280 actions_output_to_local_port(const struct xlate_ctx *ctx)
4282 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4283 const struct nlattr *a;
4286 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions->data,
4287 ctx->xout->odp_actions->size) {
4288 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4289 && nl_attr_get_odp_port(a) == local_odp_port) {
4296 #if defined(__linux__)
4297 /* Returns the maximum number of packets that the Linux kernel is willing to
4298 * queue up internally to certain kinds of software-implemented ports, or the
4299 * default (and rarely modified) value if it cannot be determined. */
4301 netdev_max_backlog(void)
4303 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4304 static int max_backlog = 1000; /* The normal default value. */
4306 if (ovsthread_once_start(&once)) {
4307 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4311 stream = fopen(filename, "r");
4313 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4315 if (fscanf(stream, "%d", &n) != 1) {
4316 VLOG_WARN("%s: read error", filename);
4317 } else if (n <= 100) {
4318 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4324 ovsthread_once_done(&once);
4326 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4332 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4335 count_output_actions(const struct ofpbuf *odp_actions)
4337 const struct nlattr *a;
4341 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4342 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4348 #endif /* defined(__linux__) */
4350 /* Returns true if 'odp_actions' contains more output actions than the datapath
4351 * can reliably handle in one go. On Linux, this is the value of the
4352 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4353 * packets that the kernel is willing to queue up for processing while the
4354 * datapath is processing a set of actions. */
4356 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4359 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4360 && count_output_actions(odp_actions) > netdev_max_backlog());
4362 /* OSes other than Linux might have similar limits, but we don't know how
4363 * to determine them.*/
4368 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
4369 * into datapath actions in 'odp_actions', using 'ctx'.
4371 * The caller must take responsibility for eventually freeing 'xout', with
4372 * xlate_out_uninit(). */
4374 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4376 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4377 struct flow_wildcards *wc = &xout->wc;
4378 struct flow *flow = &xin->flow;
4379 struct rule_dpif *rule = NULL;
4381 enum slow_path_reason special;
4382 const struct ofpact *ofpacts;
4383 struct xport *in_port;
4384 struct flow orig_flow;
4385 struct xlate_ctx ctx;
4390 COVERAGE_INC(xlate_actions);
4392 /* Flow initialization rules:
4393 * - 'base_flow' must match the kernel's view of the packet at the
4394 * time that action processing starts. 'flow' represents any
4395 * transformations we wish to make through actions.
4396 * - By default 'base_flow' and 'flow' are the same since the input
4397 * packet matches the output before any actions are applied.
4398 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4399 * of the received packet as seen by the kernel. If we later output
4400 * to another device without any modifications this will cause us to
4401 * insert a new tag since the original one was stripped off by the
4403 * - Tunnel metadata as received is retained in 'flow'. This allows
4404 * tunnel metadata matching also in later tables.
4405 * Since a kernel action for setting the tunnel metadata will only be
4406 * generated with actual tunnel output, changing the tunnel metadata
4407 * values in 'flow' (such as tun_id) will only have effect with a later
4408 * tunnel output action.
4409 * - Tunnel 'base_flow' is completely cleared since that is what the
4410 * kernel does. If we wish to maintain the original values an action
4411 * needs to be generated. */
4416 ctx.xout->has_learn = false;
4417 ctx.xout->has_normal = false;
4418 ctx.xout->has_fin_timeout = false;
4419 ctx.xout->nf_output_iface = NF_OUT_DROP;
4420 ctx.xout->mirrors = 0;
4422 xout->odp_actions = xin->odp_actions;
4423 if (!xout->odp_actions) {
4424 xout->odp_actions = &xout->odp_actions_buf;
4425 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4426 sizeof xout->odp_actions_stub);
4428 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4430 ctx.xbridge = xbridge_lookup(xcfg, xin->ofproto);
4435 ctx.rule = xin->rule;
4437 ctx.base_flow = *flow;
4438 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4439 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4441 flow_wildcards_init_catchall(wc);
4442 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4443 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4444 if (is_ip_any(flow)) {
4445 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4447 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4449 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4450 if (ctx.xbridge->netflow) {
4451 netflow_mask_wc(flow, wc);
4456 ctx.in_group = false;
4457 ctx.in_action_set = false;
4458 ctx.orig_skb_priority = flow->skb_priority;
4461 ctx.use_recirc = false;
4462 ctx.was_mpls = false;
4464 if (!xin->ofpacts && !ctx.rule) {
4465 rule = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
4466 xin->skip_wildcards ? NULL : wc,
4467 ctx.xin->xcache != NULL,
4468 ctx.xin->resubmit_stats, &ctx.table_id);
4469 if (ctx.xin->resubmit_stats) {
4470 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
4472 if (ctx.xin->xcache) {
4473 struct xc_entry *entry;
4475 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4476 entry->u.rule = rule;
4480 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
4481 ctx.xin->resubmit_hook(ctx.xin, rule, 0);
4484 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4487 ofpacts = xin->ofpacts;
4488 ofpacts_len = xin->ofpacts_len;
4489 } else if (ctx.rule) {
4490 const struct rule_actions *actions = rule_dpif_get_actions(ctx.rule);
4492 ofpacts = actions->ofpacts;
4493 ofpacts_len = actions->ofpacts_len;
4498 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4500 ctx.action_set_has_group = false;
4501 ofpbuf_use_stub(&ctx.action_set,
4502 ctx.action_set_stub, sizeof ctx.action_set_stub);
4504 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4505 /* Do this conditionally because the copy is expensive enough that it
4506 * shows up in profiles. */
4510 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
4511 if (in_port && in_port->is_tunnel) {
4512 if (ctx.xin->resubmit_stats) {
4513 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4515 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4518 if (ctx.xin->xcache) {
4519 struct xc_entry *entry;
4521 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4522 entry->u.dev.rx = netdev_ref(in_port->netdev);
4523 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4527 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
4529 ctx.xout->slow |= special;
4531 size_t sample_actions_len;
4533 if (flow->in_port.ofp_port
4534 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
4535 flow->in_port.ofp_port,
4537 ctx.base_flow.vlan_tci = 0;
4540 add_sflow_action(&ctx);
4541 add_ipfix_action(&ctx);
4542 sample_actions_len = ctx.xout->odp_actions->size;
4544 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4545 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4547 /* We've let OFPP_NORMAL and the learning action look at the
4548 * packet, so drop it now if forwarding is disabled. */
4549 if (in_port && (!xport_stp_forward_state(in_port) ||
4550 !xport_rstp_forward_state(in_port))) {
4551 /* Drop all actions added by do_xlate_actions() above. */
4552 ctx.xout->odp_actions->size = sample_actions_len;
4553 } else if (ctx.action_set.size) {
4554 /* Translate action set only if not dropping the packet. */
4555 xlate_action_set(&ctx);
4559 if (ctx.xbridge->has_in_band
4560 && in_band_must_output_to_local_port(flow)
4561 && !actions_output_to_local_port(&ctx)) {
4562 compose_output_action(&ctx, OFPP_LOCAL);
4565 fix_sflow_action(&ctx);
4567 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4568 add_mirror_actions(&ctx, &orig_flow);
4572 if (nl_attr_oversized(ctx.xout->odp_actions->size)) {
4573 /* These datapath actions are too big for a Netlink attribute, so we
4574 * can't hand them to the kernel directly. dpif_execute() can execute
4575 * them one by one with help, so just mark the result as SLOW_ACTION to
4576 * prevent the flow from being installed. */
4577 COVERAGE_INC(xlate_actions_oversize);
4578 ctx.xout->slow |= SLOW_ACTION;
4579 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
4580 COVERAGE_INC(xlate_actions_too_many_output);
4581 ctx.xout->slow |= SLOW_ACTION;
4584 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4585 if (ctx.xin->resubmit_stats) {
4586 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
4587 ctx.xin->resubmit_stats->n_packets,
4588 ctx.xin->resubmit_stats->n_bytes);
4590 if (ctx.xin->xcache) {
4591 struct xc_entry *entry;
4593 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
4594 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
4595 entry->u.mirror.mirrors = xout->mirrors;
4599 if (ctx.xbridge->netflow) {
4600 /* Only update netflow if we don't have controller flow. We don't
4601 * report NetFlow expiration messages for such facets because they
4602 * are just part of the control logic for the network, not real
4604 if (ofpacts_len == 0
4605 || ofpacts->type != OFPACT_CONTROLLER
4606 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
4607 if (ctx.xin->resubmit_stats) {
4608 netflow_flow_update(ctx.xbridge->netflow, flow,
4609 xout->nf_output_iface,
4610 ctx.xin->resubmit_stats);
4612 if (ctx.xin->xcache) {
4613 struct xc_entry *entry;
4615 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
4616 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
4617 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
4618 entry->u.nf.iface = xout->nf_output_iface;
4623 ofpbuf_uninit(&ctx.stack);
4624 ofpbuf_uninit(&ctx.action_set);
4626 /* Clear the metadata and register wildcard masks, because we won't
4627 * use non-header fields as part of the cache. */
4628 flow_wildcards_clear_non_packet_fields(wc);
4630 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
4631 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
4632 * these fields. The datapath interface, on the other hand, represents
4633 * them with just 8 bits each. This means that if the high 8 bits of the
4634 * masks for these fields somehow become set, then they will get chopped
4635 * off by a round trip through the datapath, and revalidation will spot
4636 * that as an inconsistency and delete the flow. Avoid the problem here by
4637 * making sure that only the low 8 bits of either field can be unwildcarded
4641 wc->masks.tp_src &= htons(UINT8_MAX);
4642 wc->masks.tp_dst &= htons(UINT8_MAX);
4646 /* Sends 'packet' out 'ofport'.
4647 * May modify 'packet'.
4648 * Returns 0 if successful, otherwise a positive errno value. */
4650 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
4652 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4653 struct xport *xport;
4654 struct ofpact_output output;
4657 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4658 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4659 flow_extract(packet, &flow);
4660 flow.in_port.ofp_port = OFPP_NONE;
4662 xport = xport_lookup(xcfg, ofport);
4666 output.port = xport->ofp_port;
4669 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
4670 &output.ofpact, sizeof output,
4674 struct xlate_cache *
4675 xlate_cache_new(void)
4677 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
4679 ofpbuf_init(&xcache->entries, 512);
4683 static struct xc_entry *
4684 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
4686 struct xc_entry *entry;
4688 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
4695 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
4697 if (entry->u.dev.tx) {
4698 netdev_vport_inc_tx(entry->u.dev.tx, stats);
4700 if (entry->u.dev.rx) {
4701 netdev_vport_inc_rx(entry->u.dev.rx, stats);
4703 if (entry->u.dev.bfd) {
4704 bfd_account_rx(entry->u.dev.bfd, stats);
4709 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
4711 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4712 struct xbridge *xbridge;
4713 struct xbundle *xbundle;
4714 struct flow_wildcards wc;
4716 xbridge = xbridge_lookup(xcfg, ofproto);
4721 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
4727 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
4730 /* Push stats and perform side effects of flow translation. */
4732 xlate_push_stats(struct xlate_cache *xcache,
4733 const struct dpif_flow_stats *stats)
4735 struct xc_entry *entry;
4736 struct ofpbuf entries = xcache->entries;
4737 uint8_t dmac[ETH_ADDR_LEN];
4739 if (!stats->n_packets) {
4743 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4744 switch (entry->type) {
4746 rule_dpif_credit_stats(entry->u.rule, stats);
4749 bond_account(entry->u.bond.bond, entry->u.bond.flow,
4750 entry->u.bond.vid, stats->n_bytes);
4753 xlate_cache_netdev(entry, stats);
4756 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
4757 entry->u.nf.iface, stats);
4760 mirror_update_stats(entry->u.mirror.mbridge,
4761 entry->u.mirror.mirrors,
4762 stats->n_packets, stats->n_bytes);
4765 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
4768 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
4769 entry->u.normal.vlan);
4771 case XC_FIN_TIMEOUT:
4772 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
4773 entry->u.fin.idle, entry->u.fin.hard);
4776 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
4780 /* Lookup arp to avoid arp timeout. */
4781 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
4790 xlate_dev_unref(struct xc_entry *entry)
4792 if (entry->u.dev.tx) {
4793 netdev_close(entry->u.dev.tx);
4795 if (entry->u.dev.rx) {
4796 netdev_close(entry->u.dev.rx);
4798 if (entry->u.dev.bfd) {
4799 bfd_unref(entry->u.dev.bfd);
4804 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
4806 netflow_flow_clear(netflow, flow);
4807 netflow_unref(netflow);
4812 xlate_cache_clear(struct xlate_cache *xcache)
4814 struct xc_entry *entry;
4815 struct ofpbuf entries;
4821 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4822 switch (entry->type) {
4824 rule_dpif_unref(entry->u.rule);
4827 free(entry->u.bond.flow);
4828 bond_unref(entry->u.bond.bond);
4831 xlate_dev_unref(entry);
4834 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
4837 mbridge_unref(entry->u.mirror.mbridge);
4840 free(entry->u.learn.fm);
4841 ofpbuf_delete(entry->u.learn.ofpacts);
4844 free(entry->u.normal.flow);
4846 case XC_FIN_TIMEOUT:
4847 /* 'u.fin.rule' is always already held as a XC_RULE, which
4848 * has already released it's reference above. */
4851 group_dpif_unref(entry->u.group.group);
4860 ofpbuf_clear(&xcache->entries);
4864 xlate_cache_delete(struct xlate_cache *xcache)
4866 xlate_cache_clear(xcache);
4867 ofpbuf_uninit(&xcache->entries);