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"
35 #include "dynamic-string.h"
40 #include "mac-learning.h"
41 #include "mcast-snooping.h"
42 #include "meta-flow.h"
43 #include "multipath.h"
44 #include "netdev-vport.h"
47 #include "odp-execute.h"
48 #include "ofp-actions.h"
49 #include "ofproto/ofproto-dpif-ipfix.h"
50 #include "ofproto/ofproto-dpif-mirror.h"
51 #include "ofproto/ofproto-dpif-monitor.h"
52 #include "ofproto/ofproto-dpif-sflow.h"
53 #include "ofproto/ofproto-dpif.h"
54 #include "ofproto/ofproto-provider.h"
55 #include "packet-dpif.h"
56 #include "ovs-router.h"
57 #include "tnl-ports.h"
59 #include "openvswitch/vlog.h"
61 COVERAGE_DEFINE(xlate_actions);
62 COVERAGE_DEFINE(xlate_actions_oversize);
63 COVERAGE_DEFINE(xlate_actions_too_many_output);
65 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
67 /* Maximum depth of flow table recursion (due to resubmit actions) in a
68 * flow translation. */
69 #define MAX_RESUBMIT_RECURSION 64
70 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
73 /* Timeout for internal rules created to handle recirculation */
74 #define RECIRC_TIMEOUT 60
76 /* Maximum number of resubmit actions in a flow translation, whether they are
77 * recursive or not. */
78 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
81 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
82 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
84 struct ovs_list xbundles; /* Owned xbundles. */
85 struct hmap xports; /* Indexed by ofp_port. */
87 char *name; /* Name used in log messages. */
88 struct dpif *dpif; /* Datapath interface. */
89 struct mac_learning *ml; /* Mac learning handle. */
90 struct mcast_snooping *ms; /* Multicast Snooping handle. */
91 struct mbridge *mbridge; /* Mirroring. */
92 struct dpif_sflow *sflow; /* SFlow handle, or null. */
93 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
94 struct netflow *netflow; /* Netflow handle, or null. */
95 struct stp *stp; /* STP or null if disabled. */
96 struct rstp *rstp; /* RSTP or null if disabled. */
98 bool has_in_band; /* Bridge has in band control? */
99 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
101 /* True if the datapath supports recirculation. */
104 /* True if the datapath supports variable-length
105 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
106 * False if the datapath supports only 8-byte (or shorter) userdata. */
107 bool variable_length_userdata;
109 /* Number of MPLS label stack entries that the datapath supports
111 size_t max_mpls_depth;
113 /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
115 bool masked_set_action;
119 struct hmap_node hmap_node; /* In global 'xbundles' map. */
120 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
122 struct ovs_list list_node; /* In parent 'xbridges' list. */
123 struct xbridge *xbridge; /* Parent xbridge. */
125 struct ovs_list xports; /* Contains "struct xport"s. */
127 char *name; /* Name used in log messages. */
128 struct bond *bond; /* Nonnull iff more than one port. */
129 struct lacp *lacp; /* LACP handle or null. */
131 enum port_vlan_mode vlan_mode; /* VLAN mode. */
132 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
133 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
134 * NULL if all VLANs are trunked. */
135 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
136 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
140 struct hmap_node hmap_node; /* Node in global 'xports' map. */
141 struct ofport_dpif *ofport; /* Key in global 'xports map. */
143 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
144 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
146 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
148 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
149 struct xbundle *xbundle; /* Parent xbundle or null. */
151 struct netdev *netdev; /* 'ofport''s netdev. */
153 struct xbridge *xbridge; /* Parent bridge. */
154 struct xport *peer; /* Patch port peer or null. */
156 enum ofputil_port_config config; /* OpenFlow port configuration. */
157 enum ofputil_port_state state; /* OpenFlow port state. */
158 int stp_port_no; /* STP port number or -1 if not in use. */
159 struct rstp_port *rstp_port; /* RSTP port or null. */
161 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
163 bool may_enable; /* May be enabled in bonds. */
164 bool is_tunnel; /* Is a tunnel port. */
166 struct cfm *cfm; /* CFM handle or null. */
167 struct bfd *bfd; /* BFD handle or null. */
171 struct xlate_in *xin;
172 struct xlate_out *xout;
174 const struct xbridge *xbridge;
176 /* Flow at the last commit. */
177 struct flow base_flow;
179 /* Tunnel IP destination address as received. This is stored separately
180 * as the base_flow.tunnel is cleared on init to reflect the datapath
181 * behavior. Used to make sure not to send tunneled output to ourselves,
182 * which might lead to an infinite loop. This could happen easily
183 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
184 * actually set the tun_dst field. */
185 ovs_be32 orig_tunnel_ip_dst;
187 /* Stack for the push and pop actions. Each stack element is of type
188 * "union mf_subvalue". */
189 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
192 /* The rule that we are currently translating, or NULL. */
193 struct rule_dpif *rule;
195 /* Resubmit statistics, via xlate_table_action(). */
196 int recurse; /* Current resubmit nesting depth. */
197 int resubmits; /* Total number of resubmits. */
198 bool in_group; /* Currently translating ofgroup, if true. */
200 uint32_t orig_skb_priority; /* Priority when packet arrived. */
201 uint8_t table_id; /* OpenFlow table ID where flow was found. */
202 uint32_t sflow_n_outputs; /* Number of output ports. */
203 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
204 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
205 bool exit; /* No further actions should be processed. */
207 bool use_recirc; /* Should generate recirc? */
208 struct xlate_recirc recirc; /* Information used for generating
209 * recirculation actions */
211 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
212 * This is a trigger for recirculation in cases where translating an action
213 * or looking up a flow requires access to the fields of the packet after
214 * the MPLS label stack that was originally present. */
217 /* OpenFlow 1.1+ action set.
219 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
220 * When translation is otherwise complete, ofpacts_execute_action_set()
221 * converts it to a set of "struct ofpact"s that can be translated into
222 * datapath actions. */
223 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
224 struct ofpbuf action_set; /* Action set. */
225 uint64_t action_set_stub[1024 / 8];
228 /* A controller may use OFPP_NONE as the ingress port to indicate that
229 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
230 * when an input bundle is needed for validation (e.g., mirroring or
231 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
232 * any 'port' structs, so care must be taken when dealing with it. */
233 static struct xbundle ofpp_none_bundle = {
235 .vlan_mode = PORT_VLAN_TRUNK
238 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
239 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
240 * traffic egressing the 'ofport' with that priority should be marked with. */
241 struct skb_priority_to_dscp {
242 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
243 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
245 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
261 /* xlate_cache entries hold enough information to perform the side effects of
262 * xlate_actions() for a rule, without needing to perform rule translation
263 * from scratch. The primary usage of these is to submit statistics to objects
264 * that a flow relates to, although they may be used for other effects as well
265 * (for instance, refreshing hard timeouts for learned flows). */
269 struct rule_dpif *rule;
276 struct netflow *netflow;
281 struct mbridge *mbridge;
282 mirror_mask_t mirrors;
290 struct ofproto_dpif *ofproto;
291 struct ofputil_flow_mod *fm;
292 struct ofpbuf *ofpacts;
295 struct ofproto_dpif *ofproto;
300 struct rule_dpif *rule;
305 struct group_dpif *group;
306 struct ofputil_bucket *bucket;
309 char br_name[IFNAMSIZ];
315 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
316 entries = xcache->entries; \
317 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
319 entry = ofpbuf_try_pull(&entries, sizeof *entry))
322 struct ofpbuf entries;
325 /* Xlate config contains hash maps of all bridges, bundles and ports.
326 * Xcfgp contains the pointer to the current xlate configuration.
327 * When the main thread needs to change the configuration, it copies xcfgp to
328 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
329 * does not block handler and revalidator threads. */
331 struct hmap xbridges;
332 struct hmap xbundles;
335 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
336 static struct xlate_cfg *new_xcfg = NULL;
338 static bool may_receive(const struct xport *, struct xlate_ctx *);
339 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
341 static void xlate_normal(struct xlate_ctx *);
342 static inline void xlate_report(struct xlate_ctx *, const char *);
343 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
344 uint8_t table_id, bool may_packet_in,
345 bool honor_table_miss);
346 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
347 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
348 static void output_normal(struct xlate_ctx *, const struct xbundle *,
350 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
352 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
353 const struct ofproto_dpif *);
354 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
355 const struct ofbundle *);
356 static struct xport *xport_lookup(struct xlate_cfg *,
357 const struct ofport_dpif *);
358 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
359 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
360 uint32_t skb_priority);
361 static void clear_skb_priorities(struct xport *);
362 static size_t count_skb_priorities(const struct xport *);
363 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
366 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
368 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
369 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
370 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
371 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
372 const struct mac_learning *, struct stp *,
373 struct rstp *, const struct mcast_snooping *,
374 const struct mbridge *,
375 const struct dpif_sflow *,
376 const struct dpif_ipfix *,
377 const struct netflow *,
378 bool forward_bpdu, bool has_in_band,
380 bool variable_length_userdata,
381 size_t max_mpls_depth,
382 bool masked_set_action);
383 static void xlate_xbundle_set(struct xbundle *xbundle,
384 enum port_vlan_mode vlan_mode, int vlan,
385 unsigned long *trunks, bool use_priority_tags,
386 const struct bond *bond, const struct lacp *lacp,
388 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
389 const struct netdev *netdev, const struct cfm *cfm,
390 const struct bfd *bfd, int stp_port_no,
391 const struct rstp_port *rstp_port,
392 enum ofputil_port_config config,
393 enum ofputil_port_state state, bool is_tunnel,
395 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
396 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
397 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
398 static void xlate_xbridge_copy(struct xbridge *);
399 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
400 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
402 static void xlate_xcfg_free(struct xlate_cfg *);
405 xlate_report(struct xlate_ctx *ctx, const char *s)
407 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
408 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
413 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
415 list_init(&xbridge->xbundles);
416 hmap_init(&xbridge->xports);
417 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
418 hash_pointer(xbridge->ofproto, 0));
422 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
424 list_init(&xbundle->xports);
425 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
426 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
427 hash_pointer(xbundle->ofbundle, 0));
431 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
433 hmap_init(&xport->skb_priorities);
434 hmap_insert(&xcfg->xports, &xport->hmap_node,
435 hash_pointer(xport->ofport, 0));
436 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
437 hash_ofp_port(xport->ofp_port));
441 xlate_xbridge_set(struct xbridge *xbridge,
443 const struct mac_learning *ml, struct stp *stp,
444 struct rstp *rstp, const struct mcast_snooping *ms,
445 const struct mbridge *mbridge,
446 const struct dpif_sflow *sflow,
447 const struct dpif_ipfix *ipfix,
448 const struct netflow *netflow,
449 bool forward_bpdu, bool has_in_band,
451 bool variable_length_userdata,
452 size_t max_mpls_depth,
453 bool masked_set_action)
455 if (xbridge->ml != ml) {
456 mac_learning_unref(xbridge->ml);
457 xbridge->ml = mac_learning_ref(ml);
460 if (xbridge->ms != ms) {
461 mcast_snooping_unref(xbridge->ms);
462 xbridge->ms = mcast_snooping_ref(ms);
465 if (xbridge->mbridge != mbridge) {
466 mbridge_unref(xbridge->mbridge);
467 xbridge->mbridge = mbridge_ref(mbridge);
470 if (xbridge->sflow != sflow) {
471 dpif_sflow_unref(xbridge->sflow);
472 xbridge->sflow = dpif_sflow_ref(sflow);
475 if (xbridge->ipfix != ipfix) {
476 dpif_ipfix_unref(xbridge->ipfix);
477 xbridge->ipfix = dpif_ipfix_ref(ipfix);
480 if (xbridge->stp != stp) {
481 stp_unref(xbridge->stp);
482 xbridge->stp = stp_ref(stp);
485 if (xbridge->rstp != rstp) {
486 rstp_unref(xbridge->rstp);
487 xbridge->rstp = rstp_ref(rstp);
490 if (xbridge->netflow != netflow) {
491 netflow_unref(xbridge->netflow);
492 xbridge->netflow = netflow_ref(netflow);
495 xbridge->dpif = dpif;
496 xbridge->forward_bpdu = forward_bpdu;
497 xbridge->has_in_band = has_in_band;
498 xbridge->enable_recirc = enable_recirc;
499 xbridge->variable_length_userdata = variable_length_userdata;
500 xbridge->max_mpls_depth = max_mpls_depth;
501 xbridge->masked_set_action = masked_set_action;
505 xlate_xbundle_set(struct xbundle *xbundle,
506 enum port_vlan_mode vlan_mode, int vlan,
507 unsigned long *trunks, bool use_priority_tags,
508 const struct bond *bond, const struct lacp *lacp,
511 ovs_assert(xbundle->xbridge);
513 xbundle->vlan_mode = vlan_mode;
514 xbundle->vlan = vlan;
515 xbundle->trunks = trunks;
516 xbundle->use_priority_tags = use_priority_tags;
517 xbundle->floodable = floodable;
519 if (xbundle->bond != bond) {
520 bond_unref(xbundle->bond);
521 xbundle->bond = bond_ref(bond);
524 if (xbundle->lacp != lacp) {
525 lacp_unref(xbundle->lacp);
526 xbundle->lacp = lacp_ref(lacp);
531 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
532 const struct netdev *netdev, const struct cfm *cfm,
533 const struct bfd *bfd, int stp_port_no,
534 const struct rstp_port* rstp_port,
535 enum ofputil_port_config config, enum ofputil_port_state state,
536 bool is_tunnel, bool may_enable)
538 xport->config = config;
539 xport->state = state;
540 xport->stp_port_no = stp_port_no;
541 xport->is_tunnel = is_tunnel;
542 xport->may_enable = may_enable;
543 xport->odp_port = odp_port;
545 if (xport->rstp_port != rstp_port) {
546 rstp_port_unref(xport->rstp_port);
547 xport->rstp_port = rstp_port_ref(rstp_port);
550 if (xport->cfm != cfm) {
551 cfm_unref(xport->cfm);
552 xport->cfm = cfm_ref(cfm);
555 if (xport->bfd != bfd) {
556 bfd_unref(xport->bfd);
557 xport->bfd = bfd_ref(bfd);
560 if (xport->netdev != netdev) {
561 netdev_close(xport->netdev);
562 xport->netdev = netdev_ref(netdev);
567 xlate_xbridge_copy(struct xbridge *xbridge)
569 struct xbundle *xbundle;
571 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
572 new_xbridge->ofproto = xbridge->ofproto;
573 new_xbridge->name = xstrdup(xbridge->name);
574 xlate_xbridge_init(new_xcfg, new_xbridge);
576 xlate_xbridge_set(new_xbridge,
577 xbridge->dpif, xbridge->ml, xbridge->stp,
578 xbridge->rstp, xbridge->ms, xbridge->mbridge,
579 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
580 xbridge->forward_bpdu,
581 xbridge->has_in_band, xbridge->enable_recirc,
582 xbridge->variable_length_userdata,
583 xbridge->max_mpls_depth, xbridge->masked_set_action);
584 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
585 xlate_xbundle_copy(new_xbridge, xbundle);
588 /* Copy xports which are not part of a xbundle */
589 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
590 if (!xport->xbundle) {
591 xlate_xport_copy(new_xbridge, NULL, xport);
597 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
600 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
601 new_xbundle->ofbundle = xbundle->ofbundle;
602 new_xbundle->xbridge = xbridge;
603 new_xbundle->name = xstrdup(xbundle->name);
604 xlate_xbundle_init(new_xcfg, new_xbundle);
606 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
607 xbundle->vlan, xbundle->trunks,
608 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
610 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
611 xlate_xport_copy(xbridge, new_xbundle, xport);
616 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
619 struct skb_priority_to_dscp *pdscp, *new_pdscp;
620 struct xport *new_xport = xzalloc(sizeof *xport);
621 new_xport->ofport = xport->ofport;
622 new_xport->ofp_port = xport->ofp_port;
623 new_xport->xbridge = xbridge;
624 xlate_xport_init(new_xcfg, new_xport);
626 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
627 xport->bfd, xport->stp_port_no, xport->rstp_port,
628 xport->config, xport->state, xport->is_tunnel,
632 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
634 new_xport->peer = peer;
635 new_xport->peer->peer = new_xport;
640 new_xport->xbundle = xbundle;
641 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
644 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
645 new_pdscp = xmalloc(sizeof *pdscp);
646 new_pdscp->skb_priority = pdscp->skb_priority;
647 new_pdscp->dscp = pdscp->dscp;
648 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
649 hash_int(new_pdscp->skb_priority, 0));
653 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
654 * configuration in xcfgp.
656 * This needs to be called after editing the xlate configuration.
658 * Functions that edit the new xlate configuration are
659 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
665 * edit_xlate_configuration();
667 * xlate_txn_commit(); */
669 xlate_txn_commit(void)
671 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
673 ovsrcu_set(&xcfgp, new_xcfg);
674 ovsrcu_synchronize();
675 xlate_xcfg_free(xcfg);
679 /* Copies the current xlate configuration in xcfgp to new_xcfg.
681 * This needs to be called prior to editing the xlate configuration. */
683 xlate_txn_start(void)
685 struct xbridge *xbridge;
686 struct xlate_cfg *xcfg;
688 ovs_assert(!new_xcfg);
690 new_xcfg = xmalloc(sizeof *new_xcfg);
691 hmap_init(&new_xcfg->xbridges);
692 hmap_init(&new_xcfg->xbundles);
693 hmap_init(&new_xcfg->xports);
695 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
700 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
701 xlate_xbridge_copy(xbridge);
707 xlate_xcfg_free(struct xlate_cfg *xcfg)
709 struct xbridge *xbridge, *next_xbridge;
715 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
716 xlate_xbridge_remove(xcfg, xbridge);
719 hmap_destroy(&xcfg->xbridges);
720 hmap_destroy(&xcfg->xbundles);
721 hmap_destroy(&xcfg->xports);
726 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
728 const struct mac_learning *ml, struct stp *stp,
729 struct rstp *rstp, const struct mcast_snooping *ms,
730 const struct mbridge *mbridge,
731 const struct dpif_sflow *sflow,
732 const struct dpif_ipfix *ipfix,
733 const struct netflow *netflow,
734 bool forward_bpdu, bool has_in_band, bool enable_recirc,
735 bool variable_length_userdata, size_t max_mpls_depth,
736 bool masked_set_action)
738 struct xbridge *xbridge;
740 ovs_assert(new_xcfg);
742 xbridge = xbridge_lookup(new_xcfg, ofproto);
744 xbridge = xzalloc(sizeof *xbridge);
745 xbridge->ofproto = ofproto;
747 xlate_xbridge_init(new_xcfg, xbridge);
751 xbridge->name = xstrdup(name);
753 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
754 netflow, forward_bpdu, has_in_band, enable_recirc,
755 variable_length_userdata, max_mpls_depth,
760 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
762 struct xbundle *xbundle, *next_xbundle;
763 struct xport *xport, *next_xport;
769 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
770 xlate_xport_remove(xcfg, xport);
773 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
774 xlate_xbundle_remove(xcfg, xbundle);
777 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
778 mac_learning_unref(xbridge->ml);
779 mcast_snooping_unref(xbridge->ms);
780 mbridge_unref(xbridge->mbridge);
781 dpif_sflow_unref(xbridge->sflow);
782 dpif_ipfix_unref(xbridge->ipfix);
783 stp_unref(xbridge->stp);
784 rstp_unref(xbridge->rstp);
785 hmap_destroy(&xbridge->xports);
791 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
793 struct xbridge *xbridge;
795 ovs_assert(new_xcfg);
797 xbridge = xbridge_lookup(new_xcfg, ofproto);
798 xlate_xbridge_remove(new_xcfg, xbridge);
802 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
803 const char *name, enum port_vlan_mode vlan_mode, int vlan,
804 unsigned long *trunks, bool use_priority_tags,
805 const struct bond *bond, const struct lacp *lacp,
808 struct xbundle *xbundle;
810 ovs_assert(new_xcfg);
812 xbundle = xbundle_lookup(new_xcfg, ofbundle);
814 xbundle = xzalloc(sizeof *xbundle);
815 xbundle->ofbundle = ofbundle;
816 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
818 xlate_xbundle_init(new_xcfg, xbundle);
822 xbundle->name = xstrdup(name);
824 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
825 use_priority_tags, bond, lacp, floodable);
829 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
831 struct xport *xport, *next;
837 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
838 list_remove(&xport->bundle_node);
839 xport->xbundle = NULL;
842 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
843 list_remove(&xbundle->list_node);
844 bond_unref(xbundle->bond);
845 lacp_unref(xbundle->lacp);
851 xlate_bundle_remove(struct ofbundle *ofbundle)
853 struct xbundle *xbundle;
855 ovs_assert(new_xcfg);
857 xbundle = xbundle_lookup(new_xcfg, ofbundle);
858 xlate_xbundle_remove(new_xcfg, xbundle);
862 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
863 struct ofport_dpif *ofport, ofp_port_t ofp_port,
864 odp_port_t odp_port, const struct netdev *netdev,
865 const struct cfm *cfm, const struct bfd *bfd,
866 struct ofport_dpif *peer, int stp_port_no,
867 const struct rstp_port *rstp_port,
868 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
869 enum ofputil_port_config config,
870 enum ofputil_port_state state, bool is_tunnel,
876 ovs_assert(new_xcfg);
878 xport = xport_lookup(new_xcfg, ofport);
880 xport = xzalloc(sizeof *xport);
881 xport->ofport = ofport;
882 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
883 xport->ofp_port = ofp_port;
885 xlate_xport_init(new_xcfg, xport);
888 ovs_assert(xport->ofp_port == ofp_port);
890 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, stp_port_no,
891 rstp_port, config, state, is_tunnel, may_enable);
894 xport->peer->peer = NULL;
896 xport->peer = xport_lookup(new_xcfg, peer);
898 xport->peer->peer = xport;
901 if (xport->xbundle) {
902 list_remove(&xport->bundle_node);
904 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
905 if (xport->xbundle) {
906 list_insert(&xport->xbundle->xports, &xport->bundle_node);
909 clear_skb_priorities(xport);
910 for (i = 0; i < n_qdscp; i++) {
911 struct skb_priority_to_dscp *pdscp;
912 uint32_t skb_priority;
914 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
919 pdscp = xmalloc(sizeof *pdscp);
920 pdscp->skb_priority = skb_priority;
921 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
922 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
923 hash_int(pdscp->skb_priority, 0));
928 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
935 xport->peer->peer = NULL;
939 if (xport->xbundle) {
940 list_remove(&xport->bundle_node);
943 clear_skb_priorities(xport);
944 hmap_destroy(&xport->skb_priorities);
946 hmap_remove(&xcfg->xports, &xport->hmap_node);
947 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
949 netdev_close(xport->netdev);
950 rstp_port_unref(xport->rstp_port);
951 cfm_unref(xport->cfm);
952 bfd_unref(xport->bfd);
957 xlate_ofport_remove(struct ofport_dpif *ofport)
961 ovs_assert(new_xcfg);
963 xport = xport_lookup(new_xcfg, ofport);
964 xlate_xport_remove(new_xcfg, xport);
967 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
968 * returns the corresponding struct xport, or NULL if none is found. */
969 static struct xport *
970 xlate_lookup_xport(const struct dpif_backer *backer, const struct flow *flow)
972 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
974 return xport_lookup(xcfg, tnl_port_should_receive(flow)
975 ? tnl_port_receive(flow)
976 : odp_port_to_ofport(backer, flow->in_port.odp_port));
979 static struct ofproto_dpif *
980 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
981 ofp_port_t *ofp_in_port, const struct xport **xportp)
983 struct ofproto_dpif *recv_ofproto = NULL;
984 struct ofproto_dpif *recirc_ofproto = NULL;
985 const struct xport *xport;
986 ofp_port_t in_port = OFPP_NONE;
988 *xportp = xport = xlate_lookup_xport(backer, flow);
991 recv_ofproto = xport->xbridge->ofproto;
992 in_port = xport->ofp_port;
995 /* When recirc_id is set in 'flow', checks whether the ofproto_dpif that
996 * corresponds to the recirc_id is same as the receiving bridge. If they
997 * are the same, uses the 'recv_ofproto' and keeps the 'ofp_in_port' as
998 * assigned. Otherwise, uses the 'recirc_ofproto' that owns recirc_id and
999 * assigns OFPP_NONE to 'ofp_in_port'. Doing this is in that, the
1000 * recirculated flow must be processced by the ofproto which originates
1001 * the recirculation, and as bridges can only see their own ports, the
1002 * in_port of the 'recv_ofproto' should not be passed to the
1005 * Admittedly, setting the 'ofp_in_port' to OFPP_NONE limits the
1006 * 'recirc_ofproto' from meaningfully matching on in_port of recirculated
1007 * flow, and should be fixed in the near future.
1009 * TODO: Restore the original patch port.
1011 if (recv_ofproto && flow->recirc_id) {
1012 recirc_ofproto = ofproto_dpif_recirc_get_ofproto(backer,
1014 if (recv_ofproto != recirc_ofproto) {
1015 *xportp = xport = NULL;
1016 in_port = OFPP_NONE;
1021 *ofp_in_port = in_port;
1024 return xport ? recv_ofproto : recirc_ofproto;
1027 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1028 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1029 struct ofproto_dpif *
1030 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1031 ofp_port_t *ofp_in_port)
1033 const struct xport *xport;
1035 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1038 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1039 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1040 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1041 * handles for those protocols if they're enabled. Caller may use the returned
1042 * pointers until quiescing, for longer term use additional references must
1045 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1048 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1049 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1050 struct dpif_sflow **sflow, struct netflow **netflow,
1051 ofp_port_t *ofp_in_port)
1053 struct ofproto_dpif *ofproto;
1054 const struct xport *xport;
1056 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1063 *ofprotop = ofproto;
1067 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1071 *sflow = xport ? xport->xbridge->sflow : NULL;
1075 *netflow = xport ? xport->xbridge->netflow : NULL;
1081 static struct xbridge *
1082 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1084 struct hmap *xbridges;
1085 struct xbridge *xbridge;
1087 if (!ofproto || !xcfg) {
1091 xbridges = &xcfg->xbridges;
1093 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1095 if (xbridge->ofproto == ofproto) {
1102 static struct xbundle *
1103 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1105 struct hmap *xbundles;
1106 struct xbundle *xbundle;
1108 if (!ofbundle || !xcfg) {
1112 xbundles = &xcfg->xbundles;
1114 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1116 if (xbundle->ofbundle == ofbundle) {
1123 static struct xport *
1124 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1126 struct hmap *xports;
1127 struct xport *xport;
1129 if (!ofport || !xcfg) {
1133 xports = &xcfg->xports;
1135 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1137 if (xport->ofport == ofport) {
1144 static struct stp_port *
1145 xport_get_stp_port(const struct xport *xport)
1147 return xport->xbridge->stp && xport->stp_port_no != -1
1148 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1153 xport_stp_learn_state(const struct xport *xport)
1155 struct stp_port *sp = xport_get_stp_port(xport);
1157 ? stp_learn_in_state(stp_port_get_state(sp))
1162 xport_stp_forward_state(const struct xport *xport)
1164 struct stp_port *sp = xport_get_stp_port(xport);
1166 ? stp_forward_in_state(stp_port_get_state(sp))
1171 xport_stp_should_forward_bpdu(const struct xport *xport)
1173 struct stp_port *sp = xport_get_stp_port(xport);
1174 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1177 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1178 * were used to make the determination.*/
1180 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1182 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1183 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1184 return is_stp(flow);
1188 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1190 struct stp_port *sp = xport_get_stp_port(xport);
1191 struct ofpbuf payload = *packet;
1192 struct eth_header *eth = ofpbuf_data(&payload);
1194 /* Sink packets on ports that have STP disabled when the bridge has
1196 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1200 /* Trim off padding on payload. */
1201 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1202 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1205 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1206 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
1210 static enum rstp_state
1211 xport_get_rstp_port_state(const struct xport *xport)
1213 return xport->rstp_port
1214 ? rstp_port_get_state(xport->rstp_port)
1219 xport_rstp_learn_state(const struct xport *xport)
1221 return xport->xbridge->rstp && xport->rstp_port
1222 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1227 xport_rstp_forward_state(const struct xport *xport)
1229 return xport->xbridge->rstp && xport->rstp_port
1230 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1235 xport_rstp_should_manage_bpdu(const struct xport *xport)
1237 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1241 rstp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1243 struct ofpbuf payload = *packet;
1244 struct eth_header *eth = ofpbuf_data(&payload);
1246 /* Sink packets on ports that have no RSTP. */
1247 if (!xport->rstp_port) {
1251 /* Trim off padding on payload. */
1252 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1253 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1256 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1257 rstp_port_received_bpdu(xport->rstp_port, ofpbuf_data(&payload),
1258 ofpbuf_size(&payload));
1262 static struct xport *
1263 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1265 struct xport *xport;
1267 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1269 if (xport->ofp_port == ofp_port) {
1277 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1279 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1280 return xport ? xport->odp_port : ODPP_NONE;
1284 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1286 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1287 return xport && xport->may_enable;
1290 static struct ofputil_bucket *
1291 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1295 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1297 struct group_dpif *group;
1299 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1300 struct ofputil_bucket *bucket;
1302 bucket = group_first_live_bucket(ctx, group, depth);
1303 group_dpif_unref(group);
1304 return bucket == NULL;
1310 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1313 bucket_is_alive(const struct xlate_ctx *ctx,
1314 struct ofputil_bucket *bucket, int depth)
1316 if (depth >= MAX_LIVENESS_RECURSION) {
1317 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1319 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1320 MAX_LIVENESS_RECURSION);
1324 return (!ofputil_bucket_has_liveness(bucket)
1325 || (bucket->watch_port != OFPP_ANY
1326 && odp_port_is_alive(ctx, bucket->watch_port))
1327 || (bucket->watch_group != OFPG_ANY
1328 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1331 static struct ofputil_bucket *
1332 group_first_live_bucket(const struct xlate_ctx *ctx,
1333 const struct group_dpif *group, int depth)
1335 struct ofputil_bucket *bucket;
1336 const struct ovs_list *buckets;
1338 group_dpif_get_buckets(group, &buckets);
1339 LIST_FOR_EACH (bucket, list_node, buckets) {
1340 if (bucket_is_alive(ctx, bucket, depth)) {
1348 static struct ofputil_bucket *
1349 group_best_live_bucket(const struct xlate_ctx *ctx,
1350 const struct group_dpif *group,
1353 struct ofputil_bucket *best_bucket = NULL;
1354 uint32_t best_score = 0;
1357 struct ofputil_bucket *bucket;
1358 const struct ovs_list *buckets;
1360 group_dpif_get_buckets(group, &buckets);
1361 LIST_FOR_EACH (bucket, list_node, buckets) {
1362 if (bucket_is_alive(ctx, bucket, 0)) {
1363 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1364 if (score >= best_score) {
1365 best_bucket = bucket;
1376 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1378 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1379 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1383 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1385 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1388 static mirror_mask_t
1389 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1391 return xbundle != &ofpp_none_bundle
1392 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1396 static mirror_mask_t
1397 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1399 return xbundle != &ofpp_none_bundle
1400 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1404 static mirror_mask_t
1405 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1407 return xbundle != &ofpp_none_bundle
1408 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1412 static struct xbundle *
1413 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1414 bool warn, struct xport **in_xportp)
1416 struct xport *xport;
1418 /* Find the port and bundle for the received packet. */
1419 xport = get_ofp_port(xbridge, in_port);
1423 if (xport && xport->xbundle) {
1424 return xport->xbundle;
1427 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1428 * which a controller may use as the ingress port for traffic that
1429 * it is sourcing. */
1430 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1431 return &ofpp_none_bundle;
1434 /* Odd. A few possible reasons here:
1436 * - We deleted a port but there are still a few packets queued up
1439 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1440 * we don't know about.
1442 * - The ofproto client didn't configure the port as part of a bundle.
1443 * This is particularly likely to happen if a packet was received on the
1444 * port after it was created, but before the client had a chance to
1445 * configure its bundle.
1448 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1450 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1451 "port %"PRIu16, xbridge->name, in_port);
1457 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1459 const struct xbridge *xbridge = ctx->xbridge;
1460 mirror_mask_t mirrors;
1461 struct xbundle *in_xbundle;
1465 mirrors = ctx->xout->mirrors;
1466 ctx->xout->mirrors = 0;
1468 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1469 ctx->xin->packet != NULL, NULL);
1473 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1475 /* Drop frames on bundles reserved for mirroring. */
1476 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1477 if (ctx->xin->packet != NULL) {
1478 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1479 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1480 "%s, which is reserved exclusively for mirroring",
1481 ctx->xbridge->name, in_xbundle->name);
1483 ofpbuf_clear(ctx->xout->odp_actions);
1488 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1489 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1492 vlan = input_vid_to_vlan(in_xbundle, vid);
1498 /* Restore the original packet before adding the mirror actions. */
1499 ctx->xin->flow = *orig_flow;
1502 mirror_mask_t dup_mirrors;
1503 struct ofbundle *out;
1504 unsigned long *vlans;
1509 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1510 &vlans, &dup_mirrors, &out, &out_vlan);
1511 ovs_assert(has_mirror);
1514 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1516 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1519 if (!vlan_mirrored) {
1520 mirrors = zero_rightmost_1bit(mirrors);
1524 mirrors &= ~dup_mirrors;
1525 ctx->xout->mirrors |= dup_mirrors;
1527 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1528 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1530 output_normal(ctx, out_xbundle, vlan);
1532 } else if (vlan != out_vlan
1533 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1534 struct xbundle *xbundle;
1536 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1537 if (xbundle_includes_vlan(xbundle, out_vlan)
1538 && !xbundle_mirror_out(xbridge, xbundle)) {
1539 output_normal(ctx, xbundle, out_vlan);
1546 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1547 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1548 * the bundle on which the packet was received, returns the VLAN to which the
1551 * Both 'vid' and the return value are in the range 0...4095. */
1553 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1555 switch (in_xbundle->vlan_mode) {
1556 case PORT_VLAN_ACCESS:
1557 return in_xbundle->vlan;
1560 case PORT_VLAN_TRUNK:
1563 case PORT_VLAN_NATIVE_UNTAGGED:
1564 case PORT_VLAN_NATIVE_TAGGED:
1565 return vid ? vid : in_xbundle->vlan;
1572 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1573 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1576 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1577 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1580 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1582 /* Allow any VID on the OFPP_NONE port. */
1583 if (in_xbundle == &ofpp_none_bundle) {
1587 switch (in_xbundle->vlan_mode) {
1588 case PORT_VLAN_ACCESS:
1591 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1592 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1593 "packet received on port %s configured as VLAN "
1594 "%"PRIu16" access port", vid, in_xbundle->name,
1601 case PORT_VLAN_NATIVE_UNTAGGED:
1602 case PORT_VLAN_NATIVE_TAGGED:
1604 /* Port must always carry its native VLAN. */
1608 case PORT_VLAN_TRUNK:
1609 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1611 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1612 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1613 "received on port %s not configured for trunking "
1614 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1626 /* Given 'vlan', the VLAN that a packet belongs to, and
1627 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1628 * that should be included in the 802.1Q header. (If the return value is 0,
1629 * then the 802.1Q header should only be included in the packet if there is a
1632 * Both 'vlan' and the return value are in the range 0...4095. */
1634 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1636 switch (out_xbundle->vlan_mode) {
1637 case PORT_VLAN_ACCESS:
1640 case PORT_VLAN_TRUNK:
1641 case PORT_VLAN_NATIVE_TAGGED:
1644 case PORT_VLAN_NATIVE_UNTAGGED:
1645 return vlan == out_xbundle->vlan ? 0 : vlan;
1653 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1656 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1658 ovs_be16 tci, old_tci;
1659 struct xport *xport;
1661 vid = output_vlan_to_vid(out_xbundle, vlan);
1662 if (list_is_empty(&out_xbundle->xports)) {
1663 /* Partially configured bundle with no slaves. Drop the packet. */
1665 } else if (!out_xbundle->bond) {
1666 ctx->use_recirc = false;
1667 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1670 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1671 struct flow_wildcards *wc = &ctx->xout->wc;
1672 struct xlate_recirc *xr = &ctx->recirc;
1673 struct ofport_dpif *ofport;
1675 if (ctx->xbridge->enable_recirc) {
1676 ctx->use_recirc = bond_may_recirc(
1677 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1679 if (ctx->use_recirc) {
1680 /* Only TCP mode uses recirculation. */
1681 xr->hash_alg = OVS_HASH_ALG_L4;
1682 bond_update_post_recirc_rules(out_xbundle->bond, false);
1684 /* Recirculation does not require unmasking hash fields. */
1689 ofport = bond_choose_output_slave(out_xbundle->bond,
1690 &ctx->xin->flow, wc, vid);
1691 xport = xport_lookup(xcfg, ofport);
1694 /* No slaves enabled, so drop packet. */
1698 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1699 * accounting for this bond. */
1700 if (!ctx->use_recirc) {
1701 if (ctx->xin->resubmit_stats) {
1702 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1703 ctx->xin->resubmit_stats->n_bytes);
1705 if (ctx->xin->xcache) {
1706 struct xc_entry *entry;
1709 flow = &ctx->xin->flow;
1710 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1711 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1712 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1713 entry->u.bond.vid = vid;
1718 old_tci = *flow_tci;
1720 if (tci || out_xbundle->use_priority_tags) {
1721 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1723 tci |= htons(VLAN_CFI);
1728 compose_output_action(ctx, xport->ofp_port);
1729 *flow_tci = old_tci;
1732 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1733 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1734 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1736 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1738 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1742 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1743 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1747 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1748 if (flow->nw_proto == ARP_OP_REPLY) {
1750 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1751 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1752 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1754 return flow->nw_src == flow->nw_dst;
1760 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1761 * dropped. Returns true if they may be forwarded, false if they should be
1764 * 'in_port' must be the xport that corresponds to flow->in_port.
1765 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1767 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1768 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1769 * checked by input_vid_is_valid().
1771 * May also add tags to '*tags', although the current implementation only does
1772 * so in one special case.
1775 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1778 struct xbundle *in_xbundle = in_port->xbundle;
1779 const struct xbridge *xbridge = ctx->xbridge;
1780 struct flow *flow = &ctx->xin->flow;
1782 /* Drop frames for reserved multicast addresses
1783 * only if forward_bpdu option is absent. */
1784 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1785 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1789 if (in_xbundle->bond) {
1790 struct mac_entry *mac;
1792 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1798 xlate_report(ctx, "bonding refused admissibility, dropping");
1801 case BV_DROP_IF_MOVED:
1802 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1803 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1804 if (mac && mac->port.p != in_xbundle->ofbundle &&
1805 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1806 || mac_entry_is_grat_arp_locked(mac))) {
1807 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1808 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1812 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1820 /* Checks whether a MAC learning update is necessary for MAC learning table
1821 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1824 * Most packets processed through the MAC learning table do not actually
1825 * change it in any way. This function requires only a read lock on the MAC
1826 * learning table, so it is much cheaper in this common case.
1828 * Keep the code here synchronized with that in update_learning_table__()
1831 is_mac_learning_update_needed(const struct mac_learning *ml,
1832 const struct flow *flow,
1833 struct flow_wildcards *wc,
1834 int vlan, struct xbundle *in_xbundle)
1835 OVS_REQ_RDLOCK(ml->rwlock)
1837 struct mac_entry *mac;
1839 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1843 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1844 if (!mac || mac_entry_age(ml, mac)) {
1848 if (is_gratuitous_arp(flow, wc)) {
1849 /* We don't want to learn from gratuitous ARP packets that are
1850 * reflected back over bond slaves so we lock the learning table. */
1851 if (!in_xbundle->bond) {
1853 } else if (mac_entry_is_grat_arp_locked(mac)) {
1858 return mac->port.p != in_xbundle->ofbundle;
1862 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1863 * received on 'in_xbundle' in 'vlan'.
1865 * This code repeats all the checks in is_mac_learning_update_needed() because
1866 * the lock was released between there and here and thus the MAC learning state
1867 * could have changed.
1869 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1872 update_learning_table__(const struct xbridge *xbridge,
1873 const struct flow *flow, struct flow_wildcards *wc,
1874 int vlan, struct xbundle *in_xbundle)
1875 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1877 struct mac_entry *mac;
1879 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1883 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1884 if (is_gratuitous_arp(flow, wc)) {
1885 /* We don't want to learn from gratuitous ARP packets that are
1886 * reflected back over bond slaves so we lock the learning table. */
1887 if (!in_xbundle->bond) {
1888 mac_entry_set_grat_arp_lock(mac);
1889 } else if (mac_entry_is_grat_arp_locked(mac)) {
1894 if (mac->port.p != in_xbundle->ofbundle) {
1895 /* The log messages here could actually be useful in debugging,
1896 * so keep the rate limit relatively high. */
1897 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1899 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1900 "on port %s in VLAN %d",
1901 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1902 in_xbundle->name, vlan);
1904 mac->port.p = in_xbundle->ofbundle;
1905 mac_learning_changed(xbridge->ml);
1910 update_learning_table(const struct xbridge *xbridge,
1911 const struct flow *flow, struct flow_wildcards *wc,
1912 int vlan, struct xbundle *in_xbundle)
1916 /* Don't learn the OFPP_NONE port. */
1917 if (in_xbundle == &ofpp_none_bundle) {
1921 /* First try the common case: no change to MAC learning table. */
1922 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1923 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1925 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1928 /* Slow path: MAC learning table might need an update. */
1929 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1930 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1931 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1935 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1936 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
1938 update_mcast_snooping_table__(const struct xbridge *xbridge,
1939 const struct flow *flow,
1940 struct mcast_snooping *ms,
1941 ovs_be32 ip4, int vlan,
1942 struct xbundle *in_xbundle)
1943 OVS_REQ_WRLOCK(ms->rwlock)
1945 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
1947 switch (ntohs(flow->tp_src)) {
1948 case IGMP_HOST_MEMBERSHIP_REPORT:
1949 case IGMPV2_HOST_MEMBERSHIP_REPORT:
1950 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1951 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
1952 IP_FMT" is on port %s in VLAN %d",
1953 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1956 case IGMP_HOST_LEAVE_MESSAGE:
1957 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1958 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
1959 IP_FMT" is on port %s in VLAN %d",
1960 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1963 case IGMP_HOST_MEMBERSHIP_QUERY:
1964 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
1965 in_xbundle->ofbundle)) {
1966 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
1967 IP_FMT" is on port %s in VLAN %d",
1968 xbridge->name, IP_ARGS(flow->nw_src),
1969 in_xbundle->name, vlan);
1975 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1976 * was received on 'in_xbundle' in 'vlan'. */
1978 update_mcast_snooping_table(const struct xbridge *xbridge,
1979 const struct flow *flow, int vlan,
1980 struct xbundle *in_xbundle)
1982 struct mcast_snooping *ms = xbridge->ms;
1983 struct xlate_cfg *xcfg;
1984 struct xbundle *mcast_xbundle;
1985 struct mcast_port_bundle *fport;
1987 /* Don't learn the OFPP_NONE port. */
1988 if (in_xbundle == &ofpp_none_bundle) {
1992 /* Don't learn from flood ports */
1993 mcast_xbundle = NULL;
1994 ovs_rwlock_wrlock(&ms->rwlock);
1995 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1996 LIST_FOR_EACH(fport, node, &ms->fport_list) {
1997 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
1998 if (mcast_xbundle == in_xbundle) {
2003 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2004 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
2007 ovs_rwlock_unlock(&ms->rwlock);
2010 /* send the packet to ports having the multicast group learned */
2012 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2013 struct mcast_snooping *ms OVS_UNUSED,
2014 struct mcast_group *grp,
2015 struct xbundle *in_xbundle, uint16_t vlan)
2016 OVS_REQ_RDLOCK(ms->rwlock)
2018 struct xlate_cfg *xcfg;
2019 struct mcast_group_bundle *b;
2020 struct xbundle *mcast_xbundle;
2022 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2023 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2024 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2025 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2026 xlate_report(ctx, "forwarding to mcast group port");
2027 output_normal(ctx, mcast_xbundle, vlan);
2028 } else if (!mcast_xbundle) {
2029 xlate_report(ctx, "mcast group port is unknown, dropping");
2031 xlate_report(ctx, "mcast group port is input port, dropping");
2036 /* send the packet to ports connected to multicast routers */
2038 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2039 struct mcast_snooping *ms,
2040 struct xbundle *in_xbundle, uint16_t vlan)
2041 OVS_REQ_RDLOCK(ms->rwlock)
2043 struct xlate_cfg *xcfg;
2044 struct mcast_mrouter_bundle *mrouter;
2045 struct xbundle *mcast_xbundle;
2047 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2048 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2049 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2050 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2051 xlate_report(ctx, "forwarding to mcast router port");
2052 output_normal(ctx, mcast_xbundle, vlan);
2053 } else if (!mcast_xbundle) {
2054 xlate_report(ctx, "mcast router port is unknown, dropping");
2056 xlate_report(ctx, "mcast router port is input port, dropping");
2061 /* send the packet to ports flagged to be flooded */
2063 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2064 struct mcast_snooping *ms,
2065 struct xbundle *in_xbundle, uint16_t vlan)
2066 OVS_REQ_RDLOCK(ms->rwlock)
2068 struct xlate_cfg *xcfg;
2069 struct mcast_port_bundle *fport;
2070 struct xbundle *mcast_xbundle;
2072 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2073 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2074 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2075 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2076 xlate_report(ctx, "forwarding to mcast flood port");
2077 output_normal(ctx, mcast_xbundle, vlan);
2078 } else if (!mcast_xbundle) {
2079 xlate_report(ctx, "mcast flood port is unknown, dropping");
2081 xlate_report(ctx, "mcast flood port is input port, dropping");
2086 /* forward the Reports to configured ports */
2088 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2089 struct mcast_snooping *ms,
2090 struct xbundle *in_xbundle, uint16_t vlan)
2091 OVS_REQ_RDLOCK(ms->rwlock)
2093 struct xlate_cfg *xcfg;
2094 struct mcast_port_bundle *rport;
2095 struct xbundle *mcast_xbundle;
2097 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2098 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2099 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2100 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2101 xlate_report(ctx, "forwarding Report to mcast flagged port");
2102 output_normal(ctx, mcast_xbundle, vlan);
2103 } else if (!mcast_xbundle) {
2104 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2106 xlate_report(ctx, "mcast port is input port, dropping the Report");
2112 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2115 struct xbundle *xbundle;
2117 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2118 if (xbundle != in_xbundle
2119 && xbundle_includes_vlan(xbundle, vlan)
2120 && xbundle->floodable
2121 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2122 output_normal(ctx, xbundle, vlan);
2125 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2129 xlate_normal(struct xlate_ctx *ctx)
2131 struct flow_wildcards *wc = &ctx->xout->wc;
2132 struct flow *flow = &ctx->xin->flow;
2133 struct xbundle *in_xbundle;
2134 struct xport *in_port;
2135 struct mac_entry *mac;
2140 ctx->xout->has_normal = true;
2142 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2143 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2144 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2146 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2147 ctx->xin->packet != NULL, &in_port);
2149 xlate_report(ctx, "no input bundle, dropping");
2153 /* Drop malformed frames. */
2154 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2155 !(flow->vlan_tci & htons(VLAN_CFI))) {
2156 if (ctx->xin->packet != NULL) {
2157 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2158 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2159 "VLAN tag received on port %s",
2160 ctx->xbridge->name, in_xbundle->name);
2162 xlate_report(ctx, "partial VLAN tag, dropping");
2166 /* Drop frames on bundles reserved for mirroring. */
2167 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2168 if (ctx->xin->packet != NULL) {
2169 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2170 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2171 "%s, which is reserved exclusively for mirroring",
2172 ctx->xbridge->name, in_xbundle->name);
2174 xlate_report(ctx, "input port is mirror output port, dropping");
2179 vid = vlan_tci_to_vid(flow->vlan_tci);
2180 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2181 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2184 vlan = input_vid_to_vlan(in_xbundle, vid);
2186 /* Check other admissibility requirements. */
2187 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2191 /* Learn source MAC. */
2192 if (ctx->xin->may_learn) {
2193 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2195 if (ctx->xin->xcache) {
2196 struct xc_entry *entry;
2198 /* Save enough info to update mac learning table later. */
2199 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2200 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2201 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2202 entry->u.normal.vlan = vlan;
2205 /* Determine output bundle. */
2206 if (mcast_snooping_enabled(ctx->xbridge->ms)
2207 && !eth_addr_is_broadcast(flow->dl_dst)
2208 && eth_addr_is_multicast(flow->dl_dst)
2209 && flow->dl_type == htons(ETH_TYPE_IP)) {
2210 struct mcast_snooping *ms = ctx->xbridge->ms;
2211 struct mcast_group *grp;
2213 if (flow->nw_proto == IPPROTO_IGMP) {
2214 if (ctx->xin->may_learn) {
2215 if (mcast_snooping_is_membership(flow->tp_src) ||
2216 mcast_snooping_is_query(flow->tp_src)) {
2217 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2222 if (mcast_snooping_is_membership(flow->tp_src)) {
2223 ovs_rwlock_rdlock(&ms->rwlock);
2224 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2225 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2226 * forward IGMP Membership Reports only to those ports where
2227 * multicast routers are attached. Alternatively stated: a
2228 * snooping switch should not forward IGMP Membership Reports
2229 * to ports on which only hosts are attached.
2230 * An administrative control may be provided to override this
2231 * restriction, allowing the report messages to be flooded to
2233 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2234 ovs_rwlock_unlock(&ms->rwlock);
2236 xlate_report(ctx, "multicast traffic, flooding");
2237 xlate_normal_flood(ctx, in_xbundle, vlan);
2241 if (ip_is_local_multicast(flow->nw_dst)) {
2242 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2243 * address in the 224.0.0.x range which are not IGMP must
2244 * be forwarded on all ports */
2245 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2246 xlate_normal_flood(ctx, in_xbundle, vlan);
2251 /* forwarding to group base ports */
2252 ovs_rwlock_rdlock(&ms->rwlock);
2253 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2255 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2256 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2257 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2259 if (mcast_snooping_flood_unreg(ms)) {
2260 xlate_report(ctx, "unregistered multicast, flooding");
2261 xlate_normal_flood(ctx, in_xbundle, vlan);
2263 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2264 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2267 ovs_rwlock_unlock(&ms->rwlock);
2269 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2270 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2271 mac_port = mac ? mac->port.p : NULL;
2272 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2275 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2276 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2277 if (mac_xbundle && mac_xbundle != in_xbundle) {
2278 xlate_report(ctx, "forwarding to learned port");
2279 output_normal(ctx, mac_xbundle, vlan);
2280 } else if (!mac_xbundle) {
2281 xlate_report(ctx, "learned port is unknown, dropping");
2283 xlate_report(ctx, "learned port is input port, dropping");
2286 xlate_report(ctx, "no learned MAC for destination, flooding");
2287 xlate_normal_flood(ctx, in_xbundle, vlan);
2292 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2293 * the number of packets out of UINT32_MAX to sample. The given
2294 * cookie is passed back in the callback for each sampled packet.
2297 compose_sample_action(const struct xbridge *xbridge,
2298 struct ofpbuf *odp_actions,
2299 const struct flow *flow,
2300 const uint32_t probability,
2301 const union user_action_cookie *cookie,
2302 const size_t cookie_size,
2303 const odp_port_t tunnel_out_port)
2305 size_t sample_offset, actions_offset;
2306 odp_port_t odp_port;
2310 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2312 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2314 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2316 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2317 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2318 flow_hash_5tuple(flow, 0));
2319 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2320 tunnel_out_port, odp_actions);
2322 nl_msg_end_nested(odp_actions, actions_offset);
2323 nl_msg_end_nested(odp_actions, sample_offset);
2324 return cookie_offset;
2328 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2329 odp_port_t odp_port, unsigned int n_outputs,
2330 union user_action_cookie *cookie)
2334 cookie->type = USER_ACTION_COOKIE_SFLOW;
2335 cookie->sflow.vlan_tci = vlan_tci;
2337 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2338 * port information") for the interpretation of cookie->output. */
2339 switch (n_outputs) {
2341 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2342 cookie->sflow.output = 0x40000000 | 256;
2346 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2348 cookie->sflow.output = ifindex;
2353 /* 0x80000000 means "multiple output ports. */
2354 cookie->sflow.output = 0x80000000 | n_outputs;
2359 /* Compose SAMPLE action for sFlow bridge sampling. */
2361 compose_sflow_action(const struct xbridge *xbridge,
2362 struct ofpbuf *odp_actions,
2363 const struct flow *flow,
2364 odp_port_t odp_port)
2366 uint32_t probability;
2367 union user_action_cookie cookie;
2369 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2373 probability = dpif_sflow_get_probability(xbridge->sflow);
2374 compose_sflow_cookie(xbridge, htons(0), odp_port,
2375 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2377 return compose_sample_action(xbridge, odp_actions, flow, probability,
2378 &cookie, sizeof cookie.sflow, ODPP_NONE);
2382 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2383 uint32_t obs_domain_id, uint32_t obs_point_id,
2384 union user_action_cookie *cookie)
2386 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2387 cookie->flow_sample.probability = probability;
2388 cookie->flow_sample.collector_set_id = collector_set_id;
2389 cookie->flow_sample.obs_domain_id = obs_domain_id;
2390 cookie->flow_sample.obs_point_id = obs_point_id;
2394 compose_ipfix_cookie(union user_action_cookie *cookie,
2395 odp_port_t output_odp_port)
2397 cookie->type = USER_ACTION_COOKIE_IPFIX;
2398 cookie->ipfix.output_odp_port = output_odp_port;
2401 /* Compose SAMPLE action for IPFIX bridge sampling. */
2403 compose_ipfix_action(const struct xbridge *xbridge,
2404 struct ofpbuf *odp_actions,
2405 const struct flow *flow,
2406 odp_port_t output_odp_port)
2408 uint32_t probability;
2409 union user_action_cookie cookie;
2410 odp_port_t tunnel_out_port = ODPP_NONE;
2412 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2416 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2418 if (output_odp_port == ODPP_NONE &&
2419 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2423 /* For output case, output_odp_port is valid*/
2424 if (output_odp_port != ODPP_NONE) {
2425 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2428 /* If tunnel sampling is enabled, put an additional option attribute:
2429 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2431 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2432 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2433 tunnel_out_port = output_odp_port;
2437 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2438 compose_ipfix_cookie(&cookie, output_odp_port);
2440 compose_sample_action(xbridge, odp_actions, flow, probability,
2441 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2444 /* SAMPLE action for sFlow must be first action in any given list of
2445 * actions. At this point we do not have all information required to
2446 * build it. So try to build sample action as complete as possible. */
2448 add_sflow_action(struct xlate_ctx *ctx)
2450 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2451 ctx->xout->odp_actions,
2452 &ctx->xin->flow, ODPP_NONE);
2453 ctx->sflow_odp_port = 0;
2454 ctx->sflow_n_outputs = 0;
2457 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2458 * of actions, eventually after the SAMPLE action for sFlow. */
2460 add_ipfix_action(struct xlate_ctx *ctx)
2462 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2463 &ctx->xin->flow, ODPP_NONE);
2467 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2469 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2470 &ctx->xin->flow, port);
2473 /* Fix SAMPLE action according to data collected while composing ODP actions.
2474 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2475 * USERSPACE action's user-cookie which is required for sflow. */
2477 fix_sflow_action(struct xlate_ctx *ctx)
2479 const struct flow *base = &ctx->base_flow;
2480 union user_action_cookie *cookie;
2482 if (!ctx->user_cookie_offset) {
2486 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2487 sizeof cookie->sflow);
2488 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2490 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2491 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2494 static enum slow_path_reason
2495 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2496 const struct xport *xport, const struct ofpbuf *packet)
2498 struct flow_wildcards *wc = &ctx->xout->wc;
2499 const struct xbridge *xbridge = ctx->xbridge;
2503 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2505 cfm_process_heartbeat(xport->cfm, packet);
2508 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2510 bfd_process_packet(xport->bfd, flow, packet);
2511 /* If POLL received, immediately sends FINAL back. */
2512 if (bfd_should_send_packet(xport->bfd)) {
2513 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2517 } else if (xport->xbundle && xport->xbundle->lacp
2518 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2520 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2523 } else if ((xbridge->stp || xbridge->rstp) &&
2524 stp_should_process_flow(flow, wc)) {
2527 ? stp_process_packet(xport, packet)
2528 : rstp_process_packet(xport, packet);
2537 tnl_route_lookup_flow(const struct flow *oflow,
2538 ovs_be32 *ip, struct xport **out_port)
2540 char out_dev[IFNAMSIZ];
2541 struct xbridge *xbridge;
2542 struct xlate_cfg *xcfg;
2545 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2552 *ip = oflow->tunnel.ip_dst;
2555 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2558 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2559 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2562 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2563 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2574 xlate_flood_packet(struct xbridge *xbridge, struct ofpbuf *packet)
2576 struct ofpact_output output;
2579 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2580 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2581 flow_extract(packet, NULL, &flow);
2582 flow.in_port.ofp_port = OFPP_NONE;
2583 output.port = OFPP_FLOOD;
2586 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2587 &output.ofpact, sizeof output,
2592 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2593 ovs_be32 ip_src, ovs_be32 ip_dst)
2595 struct xbridge *xbridge = out_dev->xbridge;
2596 struct ofpbuf packet;
2598 ofpbuf_init(&packet, 0);
2599 compose_arp(&packet, eth_src, ip_src, ip_dst);
2601 xlate_flood_packet(xbridge, &packet);
2602 ofpbuf_uninit(&packet);
2606 build_tunnel_send(const struct xlate_ctx *ctx, const struct xport *xport,
2607 const struct flow *flow, odp_port_t tunnel_odp_port)
2609 struct ovs_action_push_tnl tnl_push_data;
2610 struct xport *out_dev = NULL;
2611 ovs_be32 s_ip, d_ip = 0;
2612 uint8_t smac[ETH_ADDR_LEN];
2613 uint8_t dmac[ETH_ADDR_LEN];
2616 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2621 /* Use mac addr of bridge port of the peer. */
2622 err = netdev_get_etheraddr(out_dev->netdev, smac);
2627 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2632 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2634 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2637 if (ctx->xin->xcache) {
2638 struct xc_entry *entry;
2640 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2641 strncpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name, IFNAMSIZ);
2642 entry->u.tnl_arp_cache.d_ip = d_ip;
2644 err = tnl_port_build_header(xport->ofport, flow,
2645 dmac, smac, s_ip, &tnl_push_data);
2649 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2650 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2651 odp_put_tnl_push_action(ctx->xout->odp_actions, &tnl_push_data);
2656 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2659 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2660 struct flow_wildcards *wc = &ctx->xout->wc;
2661 struct flow *flow = &ctx->xin->flow;
2662 struct flow_tnl flow_tnl;
2663 ovs_be16 flow_vlan_tci;
2664 uint32_t flow_pkt_mark;
2665 uint8_t flow_nw_tos;
2666 odp_port_t out_port, odp_port;
2667 bool tnl_push_pop_send = false;
2670 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2671 * before traversing a patch port. */
2672 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 30);
2673 memset(&flow_tnl, 0, sizeof flow_tnl);
2676 xlate_report(ctx, "Nonexistent output port");
2678 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2679 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2681 } else if (check_stp) {
2682 if (is_stp(&ctx->base_flow)) {
2683 if (!xport_stp_should_forward_bpdu(xport) &&
2684 !xport_rstp_should_manage_bpdu(xport)) {
2685 if (ctx->xbridge->stp != NULL) {
2686 xlate_report(ctx, "STP not in listening state, "
2687 "skipping bpdu output");
2688 } else if (ctx->xbridge->rstp != NULL) {
2689 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2690 "skipping bpdu output");
2694 } else if (!xport_stp_forward_state(xport) ||
2695 !xport_rstp_forward_state(xport)) {
2696 if (ctx->xbridge->stp != NULL) {
2697 xlate_report(ctx, "STP not in forwarding state, "
2699 } else if (ctx->xbridge->rstp != NULL) {
2700 xlate_report(ctx, "RSTP not in forwarding state, "
2707 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2708 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2713 const struct xport *peer = xport->peer;
2714 struct flow old_flow = ctx->xin->flow;
2715 enum slow_path_reason special;
2716 uint8_t table_id = rule_dpif_lookup_get_init_table_id(&ctx->xin->flow);
2718 ctx->xbridge = peer->xbridge;
2719 flow->in_port.ofp_port = peer->ofp_port;
2720 flow->metadata = htonll(0);
2721 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2722 memset(flow->regs, 0, sizeof flow->regs);
2723 flow->actset_output = OFPP_UNSET;
2725 special = process_special(ctx, &ctx->xin->flow, peer,
2728 ctx->xout->slow |= special;
2729 } else if (may_receive(peer, ctx)) {
2730 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2731 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2734 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2735 * the learning action look at the packet, then drop it. */
2736 struct flow old_base_flow = ctx->base_flow;
2737 size_t old_size = ofpbuf_size(ctx->xout->odp_actions);
2738 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2739 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2741 ctx->xout->mirrors = old_mirrors;
2742 ctx->base_flow = old_base_flow;
2743 ofpbuf_set_size(ctx->xout->odp_actions, old_size);
2747 ctx->xin->flow = old_flow;
2748 ctx->xbridge = xport->xbridge;
2750 if (ctx->xin->resubmit_stats) {
2751 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2752 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2754 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2757 if (ctx->xin->xcache) {
2758 struct xc_entry *entry;
2760 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2761 entry->u.dev.tx = netdev_ref(xport->netdev);
2762 entry->u.dev.rx = netdev_ref(peer->netdev);
2763 entry->u.dev.bfd = bfd_ref(peer->bfd);
2768 flow_vlan_tci = flow->vlan_tci;
2769 flow_pkt_mark = flow->pkt_mark;
2770 flow_nw_tos = flow->nw_tos;
2772 if (count_skb_priorities(xport)) {
2773 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2774 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2775 wc->masks.nw_tos |= IP_DSCP_MASK;
2776 flow->nw_tos &= ~IP_DSCP_MASK;
2777 flow->nw_tos |= dscp;
2781 if (xport->is_tunnel) {
2782 /* Save tunnel metadata so that changes made due to
2783 * the Logical (tunnel) Port are not visible for any further
2784 * matches, while explicit set actions on tunnel metadata are.
2786 flow_tnl = flow->tunnel;
2787 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2788 if (odp_port == ODPP_NONE) {
2789 xlate_report(ctx, "Tunneling decided against output");
2790 goto out; /* restore flow_nw_tos */
2792 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2793 xlate_report(ctx, "Not tunneling to our own address");
2794 goto out; /* restore flow_nw_tos */
2796 if (ctx->xin->resubmit_stats) {
2797 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2799 if (ctx->xin->xcache) {
2800 struct xc_entry *entry;
2802 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2803 entry->u.dev.tx = netdev_ref(xport->netdev);
2805 out_port = odp_port;
2806 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2807 tnl_push_pop_send = true;
2809 commit_odp_tunnel_action(flow, &ctx->base_flow,
2810 ctx->xout->odp_actions);
2811 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2814 odp_port = xport->odp_port;
2815 out_port = odp_port;
2816 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2817 ofp_port_t vlandev_port;
2819 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2820 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2821 ofp_port, flow->vlan_tci);
2822 if (vlandev_port != ofp_port) {
2823 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2824 flow->vlan_tci = htons(0);
2829 if (out_port != ODPP_NONE) {
2830 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2831 ctx->xout->odp_actions,
2833 ctx->xbridge->masked_set_action);
2835 if (ctx->use_recirc) {
2836 struct ovs_action_hash *act_hash;
2837 struct xlate_recirc *xr = &ctx->recirc;
2840 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2841 OVS_ACTION_ATTR_HASH,
2843 act_hash->hash_alg = xr->hash_alg;
2844 act_hash->hash_basis = xr->hash_basis;
2846 /* Recirc action. */
2847 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2851 if (tnl_push_pop_send) {
2852 build_tunnel_send(ctx, xport, flow, odp_port);
2853 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2855 odp_port_t odp_tnl_port = ODPP_NONE;
2857 /* XXX: Write better Filter for tunnel port. We can use inport
2858 * int tunnel-port flow to avoid these checks completely. */
2859 if (ofp_port == OFPP_LOCAL &&
2860 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2862 odp_tnl_port = tnl_port_map_lookup(flow, wc);
2865 if (odp_tnl_port != ODPP_NONE) {
2866 nl_msg_put_odp_port(ctx->xout->odp_actions,
2867 OVS_ACTION_ATTR_TUNNEL_POP,
2870 /* Tunnel push-pop action is not compatible with
2872 add_ipfix_output_action(ctx, out_port);
2873 nl_msg_put_odp_port(ctx->xout->odp_actions,
2874 OVS_ACTION_ATTR_OUTPUT,
2880 ctx->sflow_odp_port = odp_port;
2881 ctx->sflow_n_outputs++;
2882 ctx->xout->nf_output_iface = ofp_port;
2887 flow->vlan_tci = flow_vlan_tci;
2888 flow->pkt_mark = flow_pkt_mark;
2889 flow->nw_tos = flow_nw_tos;
2893 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
2895 compose_output_action__(ctx, ofp_port, true);
2899 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
2901 struct rule_dpif *old_rule = ctx->rule;
2902 const struct rule_actions *actions;
2904 if (ctx->xin->resubmit_stats) {
2905 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2911 actions = rule_dpif_get_actions(rule);
2912 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2913 ctx->rule = old_rule;
2918 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2922 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2923 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2924 MAX_RESUBMIT_RECURSION);
2925 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2926 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2927 } else if (ofpbuf_size(ctx->xout->odp_actions) > UINT16_MAX) {
2928 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2929 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2930 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2939 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2940 bool may_packet_in, bool honor_table_miss)
2942 if (xlate_resubmit_resource_check(ctx)) {
2943 struct flow_wildcards *wc;
2944 uint8_t old_table_id = ctx->table_id;
2945 struct rule_dpif *rule;
2947 ctx->table_id = table_id;
2948 wc = (ctx->xin->skip_wildcards) ? NULL : &ctx->xout->wc;
2950 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2951 &ctx->xin->flow, wc,
2952 ctx->xin->xcache != NULL,
2953 ctx->xin->resubmit_stats,
2954 &ctx->table_id, in_port,
2955 may_packet_in, honor_table_miss);
2957 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
2958 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
2962 /* Fill in the cache entry here instead of xlate_recursively
2963 * to make the reference counting more explicit. We take a
2964 * reference in the lookups above if we are going to cache the
2966 if (ctx->xin->xcache) {
2967 struct xc_entry *entry;
2969 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2970 entry->u.rule = rule;
2972 xlate_recursively(ctx, rule);
2975 ctx->table_id = old_table_id;
2983 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
2984 struct ofputil_bucket *bucket)
2986 if (ctx->xin->resubmit_stats) {
2987 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
2989 if (ctx->xin->xcache) {
2990 struct xc_entry *entry;
2992 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
2993 entry->u.group.group = group_dpif_ref(group);
2994 entry->u.group.bucket = bucket;
2999 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3001 uint64_t action_list_stub[1024 / 8];
3002 struct ofpbuf action_list, action_set;
3004 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3005 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3007 ofpacts_execute_action_set(&action_list, &action_set);
3009 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
3012 ofpbuf_uninit(&action_set);
3013 ofpbuf_uninit(&action_list);
3017 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3019 struct ofputil_bucket *bucket;
3020 const struct ovs_list *buckets;
3021 struct flow old_flow = ctx->xin->flow;
3023 group_dpif_get_buckets(group, &buckets);
3025 LIST_FOR_EACH (bucket, list_node, buckets) {
3026 xlate_group_bucket(ctx, bucket);
3027 /* Roll back flow to previous state.
3028 * This is equivalent to cloning the packet for each bucket.
3030 * As a side effect any subsequently applied actions will
3031 * also effectively be applied to a clone of the packet taken
3032 * just before applying the all or indirect group. */
3033 ctx->xin->flow = old_flow;
3035 xlate_group_stats(ctx, group, NULL);
3039 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3041 struct ofputil_bucket *bucket;
3043 bucket = group_first_live_bucket(ctx, group, 0);
3045 xlate_group_bucket(ctx, bucket);
3046 xlate_group_stats(ctx, group, bucket);
3051 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3053 struct flow_wildcards *wc = &ctx->xout->wc;
3054 struct ofputil_bucket *bucket;
3057 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3058 bucket = group_best_live_bucket(ctx, group, basis);
3060 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3061 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3062 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3063 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3064 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3065 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3066 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3067 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3068 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3070 xlate_group_bucket(ctx, bucket);
3071 xlate_group_stats(ctx, group, bucket);
3076 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3078 ctx->in_group = true;
3080 switch (group_dpif_get_type(group)) {
3082 case OFPGT11_INDIRECT:
3083 xlate_all_group(ctx, group);
3085 case OFPGT11_SELECT:
3086 xlate_select_group(ctx, group);
3089 xlate_ff_group(ctx, group);
3094 group_dpif_unref(group);
3096 ctx->in_group = false;
3100 xlate_group_resource_check(struct xlate_ctx *ctx)
3102 if (!xlate_resubmit_resource_check(ctx)) {
3104 } else if (ctx->in_group) {
3105 /* Prevent nested translation of OpenFlow groups.
3107 * OpenFlow allows this restriction. We enforce this restriction only
3108 * because, with the current architecture, we would otherwise have to
3109 * take a possibly recursive read lock on the ofgroup rwlock, which is
3110 * unsafe given that POSIX allows taking a read lock to block if there
3111 * is a thread blocked on taking the write lock. Other solutions
3112 * without this restriction are also possible, but seem unwarranted
3113 * given the current limited use of groups. */
3114 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3116 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
3124 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3126 if (xlate_group_resource_check(ctx)) {
3127 struct group_dpif *group;
3130 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3132 xlate_group_action__(ctx, group);
3142 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3143 const struct ofpact_resubmit *resubmit)
3147 bool may_packet_in = false;
3148 bool honor_table_miss = false;
3150 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3151 /* Still allow missed packets to be sent to the controller
3152 * if resubmitting from an internal table. */
3153 may_packet_in = true;
3154 honor_table_miss = true;
3157 in_port = resubmit->in_port;
3158 if (in_port == OFPP_IN_PORT) {
3159 in_port = ctx->xin->flow.in_port.ofp_port;
3162 table_id = resubmit->table_id;
3163 if (table_id == 255) {
3164 table_id = ctx->table_id;
3167 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3172 flood_packets(struct xlate_ctx *ctx, bool all)
3174 const struct xport *xport;
3176 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3177 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3182 compose_output_action__(ctx, xport->ofp_port, false);
3183 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3184 compose_output_action(ctx, xport->ofp_port);
3188 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3192 execute_controller_action(struct xlate_ctx *ctx, int len,
3193 enum ofp_packet_in_reason reason,
3194 uint16_t controller_id)
3196 struct ofproto_packet_in *pin;
3197 struct dpif_packet *packet;
3199 ctx->xout->slow |= SLOW_CONTROLLER;
3200 if (!ctx->xin->packet) {
3204 packet = dpif_packet_clone_from_ofpbuf(ctx->xin->packet);
3206 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3207 ctx->xout->odp_actions,
3209 ctx->xbridge->masked_set_action);
3211 odp_execute_actions(NULL, &packet, 1, false,
3212 ofpbuf_data(ctx->xout->odp_actions),
3213 ofpbuf_size(ctx->xout->odp_actions), NULL);
3215 pin = xmalloc(sizeof *pin);
3216 pin->up.packet_len = ofpbuf_size(&packet->ofpbuf);
3217 pin->up.packet = ofpbuf_steal_data(&packet->ofpbuf);
3218 pin->up.reason = reason;
3219 pin->up.table_id = ctx->table_id;
3220 pin->up.cookie = (ctx->rule
3221 ? rule_dpif_get_flow_cookie(ctx->rule)
3224 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3226 pin->controller_id = controller_id;
3227 pin->send_len = len;
3228 /* If a rule is a table-miss rule then this is
3229 * a table-miss handled by a table-miss rule.
3231 * Else, if rule is internal and has a controller action,
3232 * the later being implied by the rule being processed here,
3233 * then this is a table-miss handled without a table-miss rule.
3235 * Otherwise this is not a table-miss. */
3236 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3238 if (rule_dpif_is_table_miss(ctx->rule)) {
3239 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3240 } else if (rule_dpif_is_internal(ctx->rule)) {
3241 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3244 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3245 dpif_packet_delete(packet);
3249 compose_recirculate_action(struct xlate_ctx *ctx,
3250 const struct ofpact *ofpacts_base,
3251 const struct ofpact *ofpact_current,
3252 size_t ofpacts_base_len)
3256 unsigned ofpacts_len;
3259 struct ofpbuf ofpacts;
3263 ofpacts_len = ofpacts_base_len -
3264 ((uint8_t *)ofpact_current - (uint8_t *)ofpacts_base);
3267 id = rule_dpif_get_recirc_id(ctx->rule);
3269 /* In the case where ctx has no rule then allocate a recirc id.
3270 * The life-cycle of this recirc id is managed by associating it
3271 * with the internal rule that is created to to handle
3272 * recirculation below.
3274 * The known use-case of this is packet_out which
3275 * translates actions without a rule */
3276 id = ofproto_dpif_alloc_recirc_id(ctx->xbridge->ofproto);
3279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3280 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3284 match_init_catchall(&match);
3285 match_set_recirc_id(&match, id);
3286 ofpbuf_use_const(&ofpacts, ofpact_current, ofpacts_len);
3287 error = ofproto_dpif_add_internal_flow(ctx->xbridge->ofproto, &match,
3288 RECIRC_RULE_PRIORITY,
3289 RECIRC_TIMEOUT, &ofpacts, &rule);
3291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3292 VLOG_ERR_RL(&rl, "Failed to add post recirculation flow %s",
3293 match_to_string(&match, 0));
3296 /* If ctx has no rule then associate the recirc id, which
3297 * was allocated above, with the internal rule. This allows
3298 * the recirc id to be released when the internal rule times out. */
3300 rule_set_recirc_id(rule, id);
3303 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3304 ctx->xout->odp_actions,
3306 ctx->xbridge->masked_set_action);
3307 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3311 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3313 struct flow_wildcards *wc = &ctx->xout->wc;
3314 struct flow *flow = &ctx->xin->flow;
3317 ovs_assert(eth_type_mpls(mpls->ethertype));
3319 n = flow_count_mpls_labels(flow, wc);
3321 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3322 ctx->xout->odp_actions,
3324 ctx->xbridge->masked_set_action);
3325 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3326 if (ctx->xin->packet != NULL) {
3327 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3328 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3329 "MPLS push action can't be performed as it would "
3330 "have more MPLS LSEs than the %d supported.",
3331 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3337 flow_push_mpls(flow, n, mpls->ethertype, wc);
3341 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3343 struct flow_wildcards *wc = &ctx->xout->wc;
3344 struct flow *flow = &ctx->xin->flow;
3345 int n = flow_count_mpls_labels(flow, wc);
3347 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3348 if (ctx->xbridge->enable_recirc) {
3349 ctx->was_mpls = true;
3351 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3352 if (ctx->xin->packet != NULL) {
3353 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3354 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3355 "MPLS pop action can't be performed as it has "
3356 "more MPLS LSEs than the %d supported.",
3357 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3360 ofpbuf_clear(ctx->xout->odp_actions);
3365 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3367 struct flow *flow = &ctx->xin->flow;
3369 if (!is_ip_any(flow)) {
3373 ctx->xout->wc.masks.nw_ttl = 0xff;
3374 if (flow->nw_ttl > 1) {
3380 for (i = 0; i < ids->n_controllers; i++) {
3381 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3385 /* Stop processing for current table. */
3391 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3393 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3394 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3395 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3400 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3402 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3403 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3404 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3409 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3411 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3412 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3413 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3418 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3420 struct flow *flow = &ctx->xin->flow;
3421 struct flow_wildcards *wc = &ctx->xout->wc;
3423 if (eth_type_mpls(flow->dl_type)) {
3424 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3426 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3429 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3432 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3436 /* Stop processing for current table. */
3441 xlate_output_action(struct xlate_ctx *ctx,
3442 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3444 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3446 ctx->xout->nf_output_iface = NF_OUT_DROP;
3450 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
3453 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3454 0, may_packet_in, true);
3460 flood_packets(ctx, false);
3463 flood_packets(ctx, true);
3465 case OFPP_CONTROLLER:
3466 execute_controller_action(ctx, max_len,
3467 ctx->in_group ? OFPR_GROUP : OFPR_ACTION, 0);
3473 if (port != ctx->xin->flow.in_port.ofp_port) {
3474 compose_output_action(ctx, port);
3476 xlate_report(ctx, "skipping output to input port");
3481 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3482 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3483 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3484 ctx->xout->nf_output_iface = prev_nf_output_iface;
3485 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3486 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3487 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3492 xlate_output_reg_action(struct xlate_ctx *ctx,
3493 const struct ofpact_output_reg *or)
3495 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3496 if (port <= UINT16_MAX) {
3497 union mf_subvalue value;
3499 memset(&value, 0xff, sizeof value);
3500 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3501 xlate_output_action(ctx, u16_to_ofp(port),
3502 or->max_len, false);
3507 xlate_enqueue_action(struct xlate_ctx *ctx,
3508 const struct ofpact_enqueue *enqueue)
3510 ofp_port_t ofp_port = enqueue->port;
3511 uint32_t queue_id = enqueue->queue;
3512 uint32_t flow_priority, priority;
3515 /* Translate queue to priority. */
3516 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3518 /* Fall back to ordinary output action. */
3519 xlate_output_action(ctx, enqueue->port, 0, false);
3523 /* Check output port. */
3524 if (ofp_port == OFPP_IN_PORT) {
3525 ofp_port = ctx->xin->flow.in_port.ofp_port;
3526 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3530 /* Add datapath actions. */
3531 flow_priority = ctx->xin->flow.skb_priority;
3532 ctx->xin->flow.skb_priority = priority;
3533 compose_output_action(ctx, ofp_port);
3534 ctx->xin->flow.skb_priority = flow_priority;
3536 /* Update NetFlow output port. */
3537 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3538 ctx->xout->nf_output_iface = ofp_port;
3539 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3540 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3545 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3547 uint32_t skb_priority;
3549 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3550 ctx->xin->flow.skb_priority = skb_priority;
3552 /* Couldn't translate queue to a priority. Nothing to do. A warning
3553 * has already been logged. */
3558 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3560 const struct xbridge *xbridge = xbridge_;
3571 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3574 port = get_ofp_port(xbridge, ofp_port);
3575 return port ? port->may_enable : false;
3580 xlate_bundle_action(struct xlate_ctx *ctx,
3581 const struct ofpact_bundle *bundle)
3585 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3587 CONST_CAST(struct xbridge *, ctx->xbridge));
3588 if (bundle->dst.field) {
3589 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3592 xlate_output_action(ctx, port, 0, false);
3597 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3598 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3600 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3601 if (ctx->xin->may_learn) {
3602 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3607 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3609 ctx->xout->has_learn = true;
3610 learn_mask(learn, &ctx->xout->wc);
3612 if (ctx->xin->xcache) {
3613 struct xc_entry *entry;
3615 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3616 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3617 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3618 entry->u.learn.ofpacts = ofpbuf_new(64);
3619 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3620 entry->u.learn.ofpacts);
3621 } else if (ctx->xin->may_learn) {
3622 uint64_t ofpacts_stub[1024 / 8];
3623 struct ofputil_flow_mod fm;
3624 struct ofpbuf ofpacts;
3626 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3627 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3628 ofpbuf_uninit(&ofpacts);
3633 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3634 uint16_t idle_timeout, uint16_t hard_timeout)
3636 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3637 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3642 xlate_fin_timeout(struct xlate_ctx *ctx,
3643 const struct ofpact_fin_timeout *oft)
3646 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3647 oft->fin_idle_timeout, oft->fin_hard_timeout);
3648 if (ctx->xin->xcache) {
3649 struct xc_entry *entry;
3651 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3652 /* XC_RULE already holds a reference on the rule, none is taken
3654 entry->u.fin.rule = ctx->rule;
3655 entry->u.fin.idle = oft->fin_idle_timeout;
3656 entry->u.fin.hard = oft->fin_hard_timeout;
3662 xlate_sample_action(struct xlate_ctx *ctx,
3663 const struct ofpact_sample *os)
3665 union user_action_cookie cookie;
3666 /* Scale the probability from 16-bit to 32-bit while representing
3667 * the same percentage. */
3668 uint32_t probability = (os->probability << 16) | os->probability;
3670 if (!ctx->xbridge->variable_length_userdata) {
3671 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3673 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3674 "lacks support (needs Linux 3.10+ or kernel module from "
3679 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3680 ctx->xout->odp_actions,
3682 ctx->xbridge->masked_set_action);
3684 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3685 os->obs_domain_id, os->obs_point_id, &cookie);
3686 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions, &ctx->xin->flow,
3687 probability, &cookie, sizeof cookie.flow_sample,
3692 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3694 if (xport->config & (is_stp(&ctx->xin->flow)
3695 ? OFPUTIL_PC_NO_RECV_STP
3696 : OFPUTIL_PC_NO_RECV)) {
3700 /* Only drop packets here if both forwarding and learning are
3701 * disabled. If just learning is enabled, we need to have
3702 * OFPP_NORMAL and the learning action have a look at the packet
3703 * before we can drop it. */
3704 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3705 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3713 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3715 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3716 size_t on_len = ofpact_nest_get_action_len(on);
3717 const struct ofpact *inner;
3719 /* Maintain actset_output depending on the contents of the action set:
3721 * - OFPP_UNSET, if there is no "output" action.
3723 * - The output port, if there is an "output" action and no "group"
3726 * - OFPP_UNSET, if there is a "group" action.
3728 if (!ctx->action_set_has_group) {
3729 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3730 if (inner->type == OFPACT_OUTPUT) {
3731 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3732 } else if (inner->type == OFPACT_GROUP) {
3733 ctx->xin->flow.actset_output = OFPP_UNSET;
3734 ctx->action_set_has_group = true;
3739 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3740 ofpact_pad(&ctx->action_set);
3744 xlate_action_set(struct xlate_ctx *ctx)
3746 uint64_t action_list_stub[1024 / 64];
3747 struct ofpbuf action_list;
3749 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3750 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3751 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
3752 ofpbuf_uninit(&action_list);
3756 ofpact_needs_recirculation_after_mpls(const struct ofpact *a, struct xlate_ctx *ctx)
3758 struct flow_wildcards *wc = &ctx->xout->wc;
3759 struct flow *flow = &ctx->xin->flow;
3761 if (!ctx->was_mpls) {
3768 case OFPACT_CONTROLLER:
3769 case OFPACT_STRIP_VLAN:
3770 case OFPACT_SET_VLAN_PCP:
3771 case OFPACT_SET_VLAN_VID:
3772 case OFPACT_ENQUEUE:
3773 case OFPACT_PUSH_VLAN:
3774 case OFPACT_SET_ETH_SRC:
3775 case OFPACT_SET_ETH_DST:
3776 case OFPACT_SET_TUNNEL:
3777 case OFPACT_SET_QUEUE:
3778 case OFPACT_POP_QUEUE:
3779 case OFPACT_CONJUNCTION:
3781 case OFPACT_OUTPUT_REG:
3784 case OFPACT_WRITE_METADATA:
3785 case OFPACT_WRITE_ACTIONS:
3786 case OFPACT_CLEAR_ACTIONS:
3790 case OFPACT_POP_MPLS:
3791 case OFPACT_DEC_MPLS_TTL:
3792 case OFPACT_SET_MPLS_TTL:
3793 case OFPACT_SET_MPLS_TC:
3794 case OFPACT_SET_MPLS_LABEL:
3795 case OFPACT_SET_IPV4_SRC:
3796 case OFPACT_SET_IPV4_DST:
3797 case OFPACT_SET_IP_DSCP:
3798 case OFPACT_SET_IP_ECN:
3799 case OFPACT_SET_IP_TTL:
3800 case OFPACT_SET_L4_SRC_PORT:
3801 case OFPACT_SET_L4_DST_PORT:
3802 case OFPACT_RESUBMIT:
3803 case OFPACT_STACK_PUSH:
3804 case OFPACT_STACK_POP:
3805 case OFPACT_DEC_TTL:
3806 case OFPACT_MULTIPATH:
3809 case OFPACT_FIN_TIMEOUT:
3810 case OFPACT_GOTO_TABLE:
3813 case OFPACT_REG_MOVE:
3814 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
3815 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
3817 case OFPACT_SET_FIELD:
3818 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
3820 case OFPACT_PUSH_MPLS:
3821 /* Recirculate if it is an IP packet with a zero ttl. This may
3822 * indicate that the packet was previously MPLS and an MPLS pop action
3823 * converted it to IP. In this case recirculating should reveal the IP
3824 * TTL which is used as the basis for a new MPLS LSE. */
3825 return (!flow_count_mpls_labels(flow, wc)
3826 && flow->nw_ttl == 0
3827 && is_ip_any(flow));
3834 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
3835 struct xlate_ctx *ctx)
3837 struct flow_wildcards *wc = &ctx->xout->wc;
3838 struct flow *flow = &ctx->xin->flow;
3839 const struct ofpact *a;
3841 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3842 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
3844 /* dl_type already in the mask, not set below. */
3846 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3847 struct ofpact_controller *controller;
3848 const struct ofpact_metadata *metadata;
3849 const struct ofpact_set_field *set_field;
3850 const struct mf_field *mf;
3856 if (ofpact_needs_recirculation_after_mpls(a, ctx)) {
3857 compose_recirculate_action(ctx, ofpacts, a, ofpacts_len);
3863 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
3864 ofpact_get_OUTPUT(a)->max_len, true);
3868 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
3873 case OFPACT_CONTROLLER:
3874 controller = ofpact_get_CONTROLLER(a);
3875 execute_controller_action(ctx, controller->max_len,
3877 controller->controller_id);
3880 case OFPACT_ENQUEUE:
3881 memset(&wc->masks.skb_priority, 0xff,
3882 sizeof wc->masks.skb_priority);
3883 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
3886 case OFPACT_SET_VLAN_VID:
3887 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3888 if (flow->vlan_tci & htons(VLAN_CFI) ||
3889 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
3890 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
3891 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
3896 case OFPACT_SET_VLAN_PCP:
3897 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
3898 if (flow->vlan_tci & htons(VLAN_CFI) ||
3899 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
3900 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
3901 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
3902 << VLAN_PCP_SHIFT) | VLAN_CFI);
3906 case OFPACT_STRIP_VLAN:
3907 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3908 flow->vlan_tci = htons(0);
3911 case OFPACT_PUSH_VLAN:
3912 /* XXX 802.1AD(QinQ) */
3913 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3914 flow->vlan_tci = htons(VLAN_CFI);
3917 case OFPACT_SET_ETH_SRC:
3918 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3919 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
3922 case OFPACT_SET_ETH_DST:
3923 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3924 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
3927 case OFPACT_SET_IPV4_SRC:
3928 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3929 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3930 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
3934 case OFPACT_SET_IPV4_DST:
3935 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3936 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3937 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
3941 case OFPACT_SET_IP_DSCP:
3942 if (is_ip_any(flow)) {
3943 wc->masks.nw_tos |= IP_DSCP_MASK;
3944 flow->nw_tos &= ~IP_DSCP_MASK;
3945 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
3949 case OFPACT_SET_IP_ECN:
3950 if (is_ip_any(flow)) {
3951 wc->masks.nw_tos |= IP_ECN_MASK;
3952 flow->nw_tos &= ~IP_ECN_MASK;
3953 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
3957 case OFPACT_SET_IP_TTL:
3958 if (is_ip_any(flow)) {
3959 wc->masks.nw_ttl = 0xff;
3960 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
3964 case OFPACT_SET_L4_SRC_PORT:
3965 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3966 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3967 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3968 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
3972 case OFPACT_SET_L4_DST_PORT:
3973 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3974 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3975 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3976 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
3980 case OFPACT_RESUBMIT:
3981 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
3984 case OFPACT_SET_TUNNEL:
3985 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
3988 case OFPACT_SET_QUEUE:
3989 memset(&wc->masks.skb_priority, 0xff,
3990 sizeof wc->masks.skb_priority);
3991 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
3994 case OFPACT_POP_QUEUE:
3995 memset(&wc->masks.skb_priority, 0xff,
3996 sizeof wc->masks.skb_priority);
3997 flow->skb_priority = ctx->orig_skb_priority;
4000 case OFPACT_REG_MOVE:
4001 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4004 case OFPACT_SET_FIELD:
4005 set_field = ofpact_get_SET_FIELD(a);
4006 mf = set_field->field;
4008 /* Set field action only ever overwrites packet's outermost
4009 * applicable header fields. Do nothing if no header exists. */
4010 if (mf->id == MFF_VLAN_VID) {
4011 wc->masks.vlan_tci |= htons(VLAN_CFI);
4012 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4015 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4016 /* 'dl_type' is already unwildcarded. */
4017 && !eth_type_mpls(flow->dl_type)) {
4020 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4021 * header field on a packet that does not have them. */
4022 mf_mask_field_and_prereqs(mf, &wc->masks);
4023 if (mf_are_prereqs_ok(mf, flow)) {
4024 mf_set_flow_value_masked(mf, &set_field->value,
4025 &set_field->mask, flow);
4029 case OFPACT_STACK_PUSH:
4030 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4034 case OFPACT_STACK_POP:
4035 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4039 case OFPACT_PUSH_MPLS:
4040 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4043 case OFPACT_POP_MPLS:
4044 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4047 case OFPACT_SET_MPLS_LABEL:
4048 compose_set_mpls_label_action(
4049 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4052 case OFPACT_SET_MPLS_TC:
4053 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4056 case OFPACT_SET_MPLS_TTL:
4057 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4060 case OFPACT_DEC_MPLS_TTL:
4061 if (compose_dec_mpls_ttl_action(ctx)) {
4066 case OFPACT_DEC_TTL:
4067 wc->masks.nw_ttl = 0xff;
4068 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4074 /* Nothing to do. */
4077 case OFPACT_MULTIPATH:
4078 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4082 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4085 case OFPACT_OUTPUT_REG:
4086 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4090 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4093 case OFPACT_CONJUNCTION: {
4094 /* A flow with a "conjunction" action represents part of a special
4095 * kind of "set membership match". Such a flow should not actually
4096 * get executed, but it could via, say, a "packet-out", even though
4097 * that wouldn't be useful. Log it to help debugging. */
4098 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4099 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4107 case OFPACT_FIN_TIMEOUT:
4108 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4109 ctx->xout->has_fin_timeout = true;
4110 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4113 case OFPACT_CLEAR_ACTIONS:
4114 ofpbuf_clear(&ctx->action_set);
4115 ctx->xin->flow.actset_output = OFPP_UNSET;
4116 ctx->action_set_has_group = false;
4119 case OFPACT_WRITE_ACTIONS:
4120 xlate_write_actions(ctx, a);
4123 case OFPACT_WRITE_METADATA:
4124 metadata = ofpact_get_WRITE_METADATA(a);
4125 flow->metadata &= ~metadata->mask;
4126 flow->metadata |= metadata->metadata & metadata->mask;
4130 /* Not implemented yet. */
4133 case OFPACT_GOTO_TABLE: {
4134 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4136 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4137 * than ogt->table_id. This is to allow goto_table actions that
4138 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4139 * after recirculation. */
4140 ovs_assert(ctx->table_id == TBL_INTERNAL
4141 || ctx->table_id < ogt->table_id);
4142 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4143 ogt->table_id, true, true);
4148 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4155 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4156 const struct flow *flow, ofp_port_t in_port,
4157 struct rule_dpif *rule, uint16_t tcp_flags,
4158 const struct ofpbuf *packet)
4160 xin->ofproto = ofproto;
4162 xin->flow.in_port.ofp_port = in_port;
4163 xin->flow.actset_output = OFPP_UNSET;
4164 xin->packet = packet;
4165 xin->may_learn = packet != NULL;
4168 xin->ofpacts = NULL;
4169 xin->ofpacts_len = 0;
4170 xin->tcp_flags = tcp_flags;
4171 xin->resubmit_hook = NULL;
4172 xin->report_hook = NULL;
4173 xin->resubmit_stats = NULL;
4174 xin->skip_wildcards = false;
4175 xin->odp_actions = NULL;
4179 xlate_out_uninit(struct xlate_out *xout)
4181 if (xout && xout->odp_actions == &xout->odp_actions_buf) {
4182 ofpbuf_uninit(xout->odp_actions);
4186 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4187 * into datapath actions, using 'ctx', and discards the datapath actions. */
4189 xlate_actions_for_side_effects(struct xlate_in *xin)
4191 struct xlate_out xout;
4193 xlate_actions(xin, &xout);
4194 xlate_out_uninit(&xout);
4198 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
4201 dst->slow = src->slow;
4202 dst->has_learn = src->has_learn;
4203 dst->has_normal = src->has_normal;
4204 dst->has_fin_timeout = src->has_fin_timeout;
4205 dst->nf_output_iface = src->nf_output_iface;
4206 dst->mirrors = src->mirrors;
4208 dst->odp_actions = &dst->odp_actions_buf;
4209 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
4210 sizeof dst->odp_actions_stub);
4211 ofpbuf_put(dst->odp_actions, ofpbuf_data(src->odp_actions),
4212 ofpbuf_size(src->odp_actions));
4215 static struct skb_priority_to_dscp *
4216 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4218 struct skb_priority_to_dscp *pdscp;
4221 hash = hash_int(skb_priority, 0);
4222 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4223 if (pdscp->skb_priority == skb_priority) {
4231 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4234 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4235 *dscp = pdscp ? pdscp->dscp : 0;
4236 return pdscp != NULL;
4240 count_skb_priorities(const struct xport *xport)
4242 return hmap_count(&xport->skb_priorities);
4246 clear_skb_priorities(struct xport *xport)
4248 struct skb_priority_to_dscp *pdscp, *next;
4250 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4251 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4257 actions_output_to_local_port(const struct xlate_ctx *ctx)
4259 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4260 const struct nlattr *a;
4263 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(ctx->xout->odp_actions),
4264 ofpbuf_size(ctx->xout->odp_actions)) {
4265 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4266 && nl_attr_get_odp_port(a) == local_odp_port) {
4273 #if defined(__linux__)
4274 /* Returns the maximum number of packets that the Linux kernel is willing to
4275 * queue up internally to certain kinds of software-implemented ports, or the
4276 * default (and rarely modified) value if it cannot be determined. */
4278 netdev_max_backlog(void)
4280 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4281 static int max_backlog = 1000; /* The normal default value. */
4283 if (ovsthread_once_start(&once)) {
4284 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4288 stream = fopen(filename, "r");
4290 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4292 if (fscanf(stream, "%d", &n) != 1) {
4293 VLOG_WARN("%s: read error", filename);
4294 } else if (n <= 100) {
4295 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4301 ovsthread_once_done(&once);
4303 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4309 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4312 count_output_actions(const struct ofpbuf *odp_actions)
4314 const struct nlattr *a;
4318 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(odp_actions),
4319 ofpbuf_size(odp_actions)) {
4320 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4326 #endif /* defined(__linux__) */
4328 /* Returns true if 'odp_actions' contains more output actions than the datapath
4329 * can reliably handle in one go. On Linux, this is the value of the
4330 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4331 * packets that the kernel is willing to queue up for processing while the
4332 * datapath is processing a set of actions. */
4334 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4337 return (ofpbuf_size(odp_actions) / NL_A_U32_SIZE > netdev_max_backlog()
4338 && count_output_actions(odp_actions) > netdev_max_backlog());
4340 /* OSes other than Linux might have similar limits, but we don't know how
4341 * to determine them.*/
4346 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
4347 * into datapath actions in 'odp_actions', using 'ctx'.
4349 * The caller must take responsibility for eventually freeing 'xout', with
4350 * xlate_out_uninit(). */
4352 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4354 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4355 struct flow_wildcards *wc = &xout->wc;
4356 struct flow *flow = &xin->flow;
4357 struct rule_dpif *rule = NULL;
4359 enum slow_path_reason special;
4360 const struct ofpact *ofpacts;
4361 struct xport *in_port;
4362 struct flow orig_flow;
4363 struct xlate_ctx ctx;
4368 COVERAGE_INC(xlate_actions);
4370 /* Flow initialization rules:
4371 * - 'base_flow' must match the kernel's view of the packet at the
4372 * time that action processing starts. 'flow' represents any
4373 * transformations we wish to make through actions.
4374 * - By default 'base_flow' and 'flow' are the same since the input
4375 * packet matches the output before any actions are applied.
4376 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4377 * of the received packet as seen by the kernel. If we later output
4378 * to another device without any modifications this will cause us to
4379 * insert a new tag since the original one was stripped off by the
4381 * - Tunnel metadata as received is retained in 'flow'. This allows
4382 * tunnel metadata matching also in later tables.
4383 * Since a kernel action for setting the tunnel metadata will only be
4384 * generated with actual tunnel output, changing the tunnel metadata
4385 * values in 'flow' (such as tun_id) will only have effect with a later
4386 * tunnel output action.
4387 * - Tunnel 'base_flow' is completely cleared since that is what the
4388 * kernel does. If we wish to maintain the original values an action
4389 * needs to be generated. */
4394 ctx.xout->has_learn = false;
4395 ctx.xout->has_normal = false;
4396 ctx.xout->has_fin_timeout = false;
4397 ctx.xout->nf_output_iface = NF_OUT_DROP;
4398 ctx.xout->mirrors = 0;
4400 xout->odp_actions = xin->odp_actions;
4401 if (!xout->odp_actions) {
4402 xout->odp_actions = &xout->odp_actions_buf;
4403 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4404 sizeof xout->odp_actions_stub);
4406 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4408 ctx.xbridge = xbridge_lookup(xcfg, xin->ofproto);
4413 ctx.rule = xin->rule;
4415 ctx.base_flow = *flow;
4416 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4417 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4419 flow_wildcards_init_catchall(wc);
4420 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4421 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4422 if (is_ip_any(flow)) {
4423 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4425 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4427 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4428 if (ctx.xbridge->netflow) {
4429 netflow_mask_wc(flow, wc);
4434 ctx.in_group = false;
4435 ctx.orig_skb_priority = flow->skb_priority;
4438 ctx.use_recirc = false;
4439 ctx.was_mpls = false;
4441 if (!xin->ofpacts && !ctx.rule) {
4442 rule = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
4443 xin->skip_wildcards ? NULL : wc,
4444 ctx.xin->xcache != NULL,
4445 ctx.xin->resubmit_stats, &ctx.table_id);
4446 if (ctx.xin->resubmit_stats) {
4447 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
4449 if (ctx.xin->xcache) {
4450 struct xc_entry *entry;
4452 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4453 entry->u.rule = rule;
4457 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
4458 ctx.xin->resubmit_hook(ctx.xin, rule, 0);
4461 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4464 ofpacts = xin->ofpacts;
4465 ofpacts_len = xin->ofpacts_len;
4466 } else if (ctx.rule) {
4467 const struct rule_actions *actions = rule_dpif_get_actions(ctx.rule);
4469 ofpacts = actions->ofpacts;
4470 ofpacts_len = actions->ofpacts_len;
4475 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4477 ctx.action_set_has_group = false;
4478 ofpbuf_use_stub(&ctx.action_set,
4479 ctx.action_set_stub, sizeof ctx.action_set_stub);
4481 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4482 /* Do this conditionally because the copy is expensive enough that it
4483 * shows up in profiles. */
4487 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
4488 if (in_port && in_port->is_tunnel) {
4489 if (ctx.xin->resubmit_stats) {
4490 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4492 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4495 if (ctx.xin->xcache) {
4496 struct xc_entry *entry;
4498 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4499 entry->u.dev.rx = netdev_ref(in_port->netdev);
4500 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4504 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
4506 ctx.xout->slow |= special;
4508 size_t sample_actions_len;
4510 if (flow->in_port.ofp_port
4511 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
4512 flow->in_port.ofp_port,
4514 ctx.base_flow.vlan_tci = 0;
4517 add_sflow_action(&ctx);
4518 add_ipfix_action(&ctx);
4519 sample_actions_len = ofpbuf_size(ctx.xout->odp_actions);
4521 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4522 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4524 /* We've let OFPP_NORMAL and the learning action look at the
4525 * packet, so drop it now if forwarding is disabled. */
4526 if (in_port && (!xport_stp_forward_state(in_port) ||
4527 !xport_rstp_forward_state(in_port))) {
4528 ofpbuf_set_size(ctx.xout->odp_actions, sample_actions_len);
4532 if (ofpbuf_size(&ctx.action_set)) {
4533 xlate_action_set(&ctx);
4536 if (ctx.xbridge->has_in_band
4537 && in_band_must_output_to_local_port(flow)
4538 && !actions_output_to_local_port(&ctx)) {
4539 compose_output_action(&ctx, OFPP_LOCAL);
4542 fix_sflow_action(&ctx);
4544 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4545 add_mirror_actions(&ctx, &orig_flow);
4549 if (nl_attr_oversized(ofpbuf_size(ctx.xout->odp_actions))) {
4550 /* These datapath actions are too big for a Netlink attribute, so we
4551 * can't hand them to the kernel directly. dpif_execute() can execute
4552 * them one by one with help, so just mark the result as SLOW_ACTION to
4553 * prevent the flow from being installed. */
4554 COVERAGE_INC(xlate_actions_oversize);
4555 ctx.xout->slow |= SLOW_ACTION;
4556 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
4557 COVERAGE_INC(xlate_actions_too_many_output);
4558 ctx.xout->slow |= SLOW_ACTION;
4561 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4562 if (ctx.xin->resubmit_stats) {
4563 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
4564 ctx.xin->resubmit_stats->n_packets,
4565 ctx.xin->resubmit_stats->n_bytes);
4567 if (ctx.xin->xcache) {
4568 struct xc_entry *entry;
4570 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
4571 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
4572 entry->u.mirror.mirrors = xout->mirrors;
4576 if (ctx.xbridge->netflow) {
4577 /* Only update netflow if we don't have controller flow. We don't
4578 * report NetFlow expiration messages for such facets because they
4579 * are just part of the control logic for the network, not real
4581 if (ofpacts_len == 0
4582 || ofpacts->type != OFPACT_CONTROLLER
4583 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
4584 if (ctx.xin->resubmit_stats) {
4585 netflow_flow_update(ctx.xbridge->netflow, flow,
4586 xout->nf_output_iface,
4587 ctx.xin->resubmit_stats);
4589 if (ctx.xin->xcache) {
4590 struct xc_entry *entry;
4592 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
4593 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
4594 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
4595 entry->u.nf.iface = xout->nf_output_iface;
4600 ofpbuf_uninit(&ctx.stack);
4601 ofpbuf_uninit(&ctx.action_set);
4603 /* Clear the metadata and register wildcard masks, because we won't
4604 * use non-header fields as part of the cache. */
4605 flow_wildcards_clear_non_packet_fields(wc);
4607 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
4608 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
4609 * these fields. The datapath interface, on the other hand, represents
4610 * them with just 8 bits each. This means that if the high 8 bits of the
4611 * masks for these fields somehow become set, then they will get chopped
4612 * off by a round trip through the datapath, and revalidation will spot
4613 * that as an inconsistency and delete the flow. Avoid the problem here by
4614 * making sure that only the low 8 bits of either field can be unwildcarded
4618 wc->masks.tp_src &= htons(UINT8_MAX);
4619 wc->masks.tp_dst &= htons(UINT8_MAX);
4623 /* Sends 'packet' out 'ofport'.
4624 * May modify 'packet'.
4625 * Returns 0 if successful, otherwise a positive errno value. */
4627 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4629 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4630 struct xport *xport;
4631 struct ofpact_output output;
4634 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4635 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4636 flow_extract(packet, NULL, &flow);
4637 flow.in_port.ofp_port = OFPP_NONE;
4639 xport = xport_lookup(xcfg, ofport);
4643 output.port = xport->ofp_port;
4646 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
4647 &output.ofpact, sizeof output,
4651 struct xlate_cache *
4652 xlate_cache_new(void)
4654 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
4656 ofpbuf_init(&xcache->entries, 512);
4660 static struct xc_entry *
4661 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
4663 struct xc_entry *entry;
4665 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
4672 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
4674 if (entry->u.dev.tx) {
4675 netdev_vport_inc_tx(entry->u.dev.tx, stats);
4677 if (entry->u.dev.rx) {
4678 netdev_vport_inc_rx(entry->u.dev.rx, stats);
4680 if (entry->u.dev.bfd) {
4681 bfd_account_rx(entry->u.dev.bfd, stats);
4686 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
4688 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4689 struct xbridge *xbridge;
4690 struct xbundle *xbundle;
4691 struct flow_wildcards wc;
4693 xbridge = xbridge_lookup(xcfg, ofproto);
4698 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
4704 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
4707 /* Push stats and perform side effects of flow translation. */
4709 xlate_push_stats(struct xlate_cache *xcache,
4710 const struct dpif_flow_stats *stats)
4712 struct xc_entry *entry;
4713 struct ofpbuf entries = xcache->entries;
4714 uint8_t dmac[ETH_ADDR_LEN];
4716 if (!stats->n_packets) {
4720 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4721 switch (entry->type) {
4723 rule_dpif_credit_stats(entry->u.rule, stats);
4726 bond_account(entry->u.bond.bond, entry->u.bond.flow,
4727 entry->u.bond.vid, stats->n_bytes);
4730 xlate_cache_netdev(entry, stats);
4733 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
4734 entry->u.nf.iface, stats);
4737 mirror_update_stats(entry->u.mirror.mbridge,
4738 entry->u.mirror.mirrors,
4739 stats->n_packets, stats->n_bytes);
4742 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
4745 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
4746 entry->u.normal.vlan);
4748 case XC_FIN_TIMEOUT:
4749 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
4750 entry->u.fin.idle, entry->u.fin.hard);
4753 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
4757 /* Lookup arp to avoid arp timeout. */
4758 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
4767 xlate_dev_unref(struct xc_entry *entry)
4769 if (entry->u.dev.tx) {
4770 netdev_close(entry->u.dev.tx);
4772 if (entry->u.dev.rx) {
4773 netdev_close(entry->u.dev.rx);
4775 if (entry->u.dev.bfd) {
4776 bfd_unref(entry->u.dev.bfd);
4781 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
4783 netflow_flow_clear(netflow, flow);
4784 netflow_unref(netflow);
4789 xlate_cache_clear(struct xlate_cache *xcache)
4791 struct xc_entry *entry;
4792 struct ofpbuf entries;
4798 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4799 switch (entry->type) {
4801 rule_dpif_unref(entry->u.rule);
4804 free(entry->u.bond.flow);
4805 bond_unref(entry->u.bond.bond);
4808 xlate_dev_unref(entry);
4811 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
4814 mbridge_unref(entry->u.mirror.mbridge);
4817 free(entry->u.learn.fm);
4818 ofpbuf_delete(entry->u.learn.ofpacts);
4821 free(entry->u.normal.flow);
4823 case XC_FIN_TIMEOUT:
4824 /* 'u.fin.rule' is always already held as a XC_RULE, which
4825 * has already released it's reference above. */
4828 group_dpif_unref(entry->u.group.group);
4837 ofpbuf_clear(&xcache->entries);
4841 xlate_cache_delete(struct xlate_cache *xcache)
4843 xlate_cache_clear(xcache);
4844 ofpbuf_uninit(&xcache->entries);