1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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 const struct xport *xport;
985 *xportp = xport = xlate_lookup_xport(backer, flow);
989 *ofp_in_port = xport->ofp_port;
991 return xport->xbridge->ofproto;
997 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
998 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
999 struct ofproto_dpif *
1000 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1001 ofp_port_t *ofp_in_port)
1003 const struct xport *xport;
1005 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1008 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1009 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1010 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1011 * handles for those protocols if they're enabled. Caller may use the returned
1012 * pointers until quiescing, for longer term use additional references must
1015 * '*ofp_in_port' is set to OFPP_NONE if 'flow''s in_port does not exist.
1017 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport.
1020 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1021 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1022 struct dpif_sflow **sflow, struct netflow **netflow,
1023 ofp_port_t *ofp_in_port)
1025 struct ofproto_dpif *ofproto;
1026 const struct xport *xport;
1028 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1030 if (ofp_in_port && !xport) {
1031 *ofp_in_port = OFPP_NONE;
1039 *ofprotop = ofproto;
1043 *ipfix = xport->xbridge->ipfix;
1047 *sflow = xport->xbridge->sflow;
1051 *netflow = xport->xbridge->netflow;
1056 static struct xbridge *
1057 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1059 struct hmap *xbridges;
1060 struct xbridge *xbridge;
1062 if (!ofproto || !xcfg) {
1066 xbridges = &xcfg->xbridges;
1068 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1070 if (xbridge->ofproto == ofproto) {
1077 static struct xbundle *
1078 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1080 struct hmap *xbundles;
1081 struct xbundle *xbundle;
1083 if (!ofbundle || !xcfg) {
1087 xbundles = &xcfg->xbundles;
1089 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1091 if (xbundle->ofbundle == ofbundle) {
1098 static struct xport *
1099 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1101 struct hmap *xports;
1102 struct xport *xport;
1104 if (!ofport || !xcfg) {
1108 xports = &xcfg->xports;
1110 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1112 if (xport->ofport == ofport) {
1119 static struct stp_port *
1120 xport_get_stp_port(const struct xport *xport)
1122 return xport->xbridge->stp && xport->stp_port_no != -1
1123 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1128 xport_stp_learn_state(const struct xport *xport)
1130 struct stp_port *sp = xport_get_stp_port(xport);
1132 ? stp_learn_in_state(stp_port_get_state(sp))
1137 xport_stp_forward_state(const struct xport *xport)
1139 struct stp_port *sp = xport_get_stp_port(xport);
1141 ? stp_forward_in_state(stp_port_get_state(sp))
1146 xport_stp_should_forward_bpdu(const struct xport *xport)
1148 struct stp_port *sp = xport_get_stp_port(xport);
1149 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1152 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1153 * were used to make the determination.*/
1155 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1157 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1158 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1159 return is_stp(flow);
1163 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1165 struct stp_port *sp = xport_get_stp_port(xport);
1166 struct ofpbuf payload = *packet;
1167 struct eth_header *eth = ofpbuf_data(&payload);
1169 /* Sink packets on ports that have STP disabled when the bridge has
1171 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1175 /* Trim off padding on payload. */
1176 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1177 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1180 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1181 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
1185 static enum rstp_state
1186 xport_get_rstp_port_state(const struct xport *xport)
1188 return xport->rstp_port
1189 ? rstp_port_get_state(xport->rstp_port)
1194 xport_rstp_learn_state(const struct xport *xport)
1196 return xport->xbridge->rstp && xport->rstp_port
1197 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1202 xport_rstp_forward_state(const struct xport *xport)
1204 return xport->xbridge->rstp && xport->rstp_port
1205 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1210 xport_rstp_should_manage_bpdu(const struct xport *xport)
1212 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1216 rstp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1218 struct ofpbuf payload = *packet;
1219 struct eth_header *eth = ofpbuf_data(&payload);
1221 /* Sink packets on ports that have no RSTP. */
1222 if (!xport->rstp_port) {
1226 /* Trim off padding on payload. */
1227 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1228 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1231 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1232 rstp_port_received_bpdu(xport->rstp_port, ofpbuf_data(&payload),
1233 ofpbuf_size(&payload));
1237 static struct xport *
1238 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1240 struct xport *xport;
1242 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1244 if (xport->ofp_port == ofp_port) {
1252 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1254 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1255 return xport ? xport->odp_port : ODPP_NONE;
1259 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1261 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1262 return xport && xport->may_enable;
1265 static struct ofputil_bucket *
1266 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1270 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1272 struct group_dpif *group;
1274 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1275 struct ofputil_bucket *bucket;
1277 bucket = group_first_live_bucket(ctx, group, depth);
1278 group_dpif_unref(group);
1279 return bucket == NULL;
1285 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1288 bucket_is_alive(const struct xlate_ctx *ctx,
1289 struct ofputil_bucket *bucket, int depth)
1291 if (depth >= MAX_LIVENESS_RECURSION) {
1292 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1294 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1295 MAX_LIVENESS_RECURSION);
1299 return (!ofputil_bucket_has_liveness(bucket)
1300 || (bucket->watch_port != OFPP_ANY
1301 && odp_port_is_alive(ctx, bucket->watch_port))
1302 || (bucket->watch_group != OFPG_ANY
1303 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1306 static struct ofputil_bucket *
1307 group_first_live_bucket(const struct xlate_ctx *ctx,
1308 const struct group_dpif *group, int depth)
1310 struct ofputil_bucket *bucket;
1311 const struct ovs_list *buckets;
1313 group_dpif_get_buckets(group, &buckets);
1314 LIST_FOR_EACH (bucket, list_node, buckets) {
1315 if (bucket_is_alive(ctx, bucket, depth)) {
1323 static struct ofputil_bucket *
1324 group_best_live_bucket(const struct xlate_ctx *ctx,
1325 const struct group_dpif *group,
1328 struct ofputil_bucket *best_bucket = NULL;
1329 uint32_t best_score = 0;
1332 struct ofputil_bucket *bucket;
1333 const struct ovs_list *buckets;
1335 group_dpif_get_buckets(group, &buckets);
1336 LIST_FOR_EACH (bucket, list_node, buckets) {
1337 if (bucket_is_alive(ctx, bucket, 0)) {
1338 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1339 if (score >= best_score) {
1340 best_bucket = bucket;
1351 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1353 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1354 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1358 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1360 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1363 static mirror_mask_t
1364 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1366 return xbundle != &ofpp_none_bundle
1367 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1371 static mirror_mask_t
1372 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1374 return xbundle != &ofpp_none_bundle
1375 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1379 static mirror_mask_t
1380 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1382 return xbundle != &ofpp_none_bundle
1383 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1387 static struct xbundle *
1388 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1389 bool warn, struct xport **in_xportp)
1391 struct xport *xport;
1393 /* Find the port and bundle for the received packet. */
1394 xport = get_ofp_port(xbridge, in_port);
1398 if (xport && xport->xbundle) {
1399 return xport->xbundle;
1402 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1403 * which a controller may use as the ingress port for traffic that
1404 * it is sourcing. */
1405 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1406 return &ofpp_none_bundle;
1409 /* Odd. A few possible reasons here:
1411 * - We deleted a port but there are still a few packets queued up
1414 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1415 * we don't know about.
1417 * - The ofproto client didn't configure the port as part of a bundle.
1418 * This is particularly likely to happen if a packet was received on the
1419 * port after it was created, but before the client had a chance to
1420 * configure its bundle.
1423 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1425 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1426 "port %"PRIu16, xbridge->name, in_port);
1432 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1434 const struct xbridge *xbridge = ctx->xbridge;
1435 mirror_mask_t mirrors;
1436 struct xbundle *in_xbundle;
1440 mirrors = ctx->xout->mirrors;
1441 ctx->xout->mirrors = 0;
1443 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1444 ctx->xin->packet != NULL, NULL);
1448 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1450 /* Drop frames on bundles reserved for mirroring. */
1451 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1452 if (ctx->xin->packet != NULL) {
1453 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1454 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1455 "%s, which is reserved exclusively for mirroring",
1456 ctx->xbridge->name, in_xbundle->name);
1458 ofpbuf_clear(ctx->xout->odp_actions);
1463 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1464 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1467 vlan = input_vid_to_vlan(in_xbundle, vid);
1473 /* Restore the original packet before adding the mirror actions. */
1474 ctx->xin->flow = *orig_flow;
1477 mirror_mask_t dup_mirrors;
1478 struct ofbundle *out;
1479 unsigned long *vlans;
1484 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1485 &vlans, &dup_mirrors, &out, &out_vlan);
1486 ovs_assert(has_mirror);
1489 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1491 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1494 if (!vlan_mirrored) {
1495 mirrors = zero_rightmost_1bit(mirrors);
1499 mirrors &= ~dup_mirrors;
1500 ctx->xout->mirrors |= dup_mirrors;
1502 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1503 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1505 output_normal(ctx, out_xbundle, vlan);
1507 } else if (vlan != out_vlan
1508 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1509 struct xbundle *xbundle;
1511 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1512 if (xbundle_includes_vlan(xbundle, out_vlan)
1513 && !xbundle_mirror_out(xbridge, xbundle)) {
1514 output_normal(ctx, xbundle, out_vlan);
1521 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1522 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1523 * the bundle on which the packet was received, returns the VLAN to which the
1526 * Both 'vid' and the return value are in the range 0...4095. */
1528 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1530 switch (in_xbundle->vlan_mode) {
1531 case PORT_VLAN_ACCESS:
1532 return in_xbundle->vlan;
1535 case PORT_VLAN_TRUNK:
1538 case PORT_VLAN_NATIVE_UNTAGGED:
1539 case PORT_VLAN_NATIVE_TAGGED:
1540 return vid ? vid : in_xbundle->vlan;
1547 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1548 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1551 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1552 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1555 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1557 /* Allow any VID on the OFPP_NONE port. */
1558 if (in_xbundle == &ofpp_none_bundle) {
1562 switch (in_xbundle->vlan_mode) {
1563 case PORT_VLAN_ACCESS:
1566 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1567 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1568 "packet received on port %s configured as VLAN "
1569 "%"PRIu16" access port", vid, in_xbundle->name,
1576 case PORT_VLAN_NATIVE_UNTAGGED:
1577 case PORT_VLAN_NATIVE_TAGGED:
1579 /* Port must always carry its native VLAN. */
1583 case PORT_VLAN_TRUNK:
1584 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1587 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1588 "received on port %s not configured for trunking "
1589 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1601 /* Given 'vlan', the VLAN that a packet belongs to, and
1602 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1603 * that should be included in the 802.1Q header. (If the return value is 0,
1604 * then the 802.1Q header should only be included in the packet if there is a
1607 * Both 'vlan' and the return value are in the range 0...4095. */
1609 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1611 switch (out_xbundle->vlan_mode) {
1612 case PORT_VLAN_ACCESS:
1615 case PORT_VLAN_TRUNK:
1616 case PORT_VLAN_NATIVE_TAGGED:
1619 case PORT_VLAN_NATIVE_UNTAGGED:
1620 return vlan == out_xbundle->vlan ? 0 : vlan;
1628 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1631 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1633 ovs_be16 tci, old_tci;
1634 struct xport *xport;
1636 vid = output_vlan_to_vid(out_xbundle, vlan);
1637 if (list_is_empty(&out_xbundle->xports)) {
1638 /* Partially configured bundle with no slaves. Drop the packet. */
1640 } else if (!out_xbundle->bond) {
1641 ctx->use_recirc = false;
1642 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1645 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1646 struct flow_wildcards *wc = &ctx->xout->wc;
1647 struct xlate_recirc *xr = &ctx->recirc;
1648 struct ofport_dpif *ofport;
1650 if (ctx->xbridge->enable_recirc) {
1651 ctx->use_recirc = bond_may_recirc(
1652 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1654 if (ctx->use_recirc) {
1655 /* Only TCP mode uses recirculation. */
1656 xr->hash_alg = OVS_HASH_ALG_L4;
1657 bond_update_post_recirc_rules(out_xbundle->bond, false);
1659 /* Recirculation does not require unmasking hash fields. */
1664 ofport = bond_choose_output_slave(out_xbundle->bond,
1665 &ctx->xin->flow, wc, vid);
1666 xport = xport_lookup(xcfg, ofport);
1669 /* No slaves enabled, so drop packet. */
1673 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1674 * accounting for this bond. */
1675 if (!ctx->use_recirc) {
1676 if (ctx->xin->resubmit_stats) {
1677 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1678 ctx->xin->resubmit_stats->n_bytes);
1680 if (ctx->xin->xcache) {
1681 struct xc_entry *entry;
1684 flow = &ctx->xin->flow;
1685 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1686 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1687 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1688 entry->u.bond.vid = vid;
1693 old_tci = *flow_tci;
1695 if (tci || out_xbundle->use_priority_tags) {
1696 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1698 tci |= htons(VLAN_CFI);
1703 compose_output_action(ctx, xport->ofp_port);
1704 *flow_tci = old_tci;
1707 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1708 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1709 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1711 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1713 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1717 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1718 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1722 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1723 if (flow->nw_proto == ARP_OP_REPLY) {
1725 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1726 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1727 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1729 return flow->nw_src == flow->nw_dst;
1735 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1736 * dropped. Returns true if they may be forwarded, false if they should be
1739 * 'in_port' must be the xport that corresponds to flow->in_port.
1740 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1742 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1743 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1744 * checked by input_vid_is_valid().
1746 * May also add tags to '*tags', although the current implementation only does
1747 * so in one special case.
1750 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1753 struct xbundle *in_xbundle = in_port->xbundle;
1754 const struct xbridge *xbridge = ctx->xbridge;
1755 struct flow *flow = &ctx->xin->flow;
1757 /* Drop frames for reserved multicast addresses
1758 * only if forward_bpdu option is absent. */
1759 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1760 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1764 if (in_xbundle->bond) {
1765 struct mac_entry *mac;
1767 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1773 xlate_report(ctx, "bonding refused admissibility, dropping");
1776 case BV_DROP_IF_MOVED:
1777 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1778 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1779 if (mac && mac->port.p != in_xbundle->ofbundle &&
1780 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1781 || mac_entry_is_grat_arp_locked(mac))) {
1782 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1783 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1787 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1795 /* Checks whether a MAC learning update is necessary for MAC learning table
1796 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1799 * Most packets processed through the MAC learning table do not actually
1800 * change it in any way. This function requires only a read lock on the MAC
1801 * learning table, so it is much cheaper in this common case.
1803 * Keep the code here synchronized with that in update_learning_table__()
1806 is_mac_learning_update_needed(const struct mac_learning *ml,
1807 const struct flow *flow,
1808 struct flow_wildcards *wc,
1809 int vlan, struct xbundle *in_xbundle)
1810 OVS_REQ_RDLOCK(ml->rwlock)
1812 struct mac_entry *mac;
1814 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1818 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1819 if (!mac || mac_entry_age(ml, mac)) {
1823 if (is_gratuitous_arp(flow, wc)) {
1824 /* We don't want to learn from gratuitous ARP packets that are
1825 * reflected back over bond slaves so we lock the learning table. */
1826 if (!in_xbundle->bond) {
1828 } else if (mac_entry_is_grat_arp_locked(mac)) {
1833 return mac->port.p != in_xbundle->ofbundle;
1837 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1838 * received on 'in_xbundle' in 'vlan'.
1840 * This code repeats all the checks in is_mac_learning_update_needed() because
1841 * the lock was released between there and here and thus the MAC learning state
1842 * could have changed.
1844 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1847 update_learning_table__(const struct xbridge *xbridge,
1848 const struct flow *flow, struct flow_wildcards *wc,
1849 int vlan, struct xbundle *in_xbundle)
1850 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1852 struct mac_entry *mac;
1854 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1858 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1859 if (is_gratuitous_arp(flow, wc)) {
1860 /* We don't want to learn from gratuitous ARP packets that are
1861 * reflected back over bond slaves so we lock the learning table. */
1862 if (!in_xbundle->bond) {
1863 mac_entry_set_grat_arp_lock(mac);
1864 } else if (mac_entry_is_grat_arp_locked(mac)) {
1869 if (mac->port.p != in_xbundle->ofbundle) {
1870 /* The log messages here could actually be useful in debugging,
1871 * so keep the rate limit relatively high. */
1872 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1874 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1875 "on port %s in VLAN %d",
1876 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1877 in_xbundle->name, vlan);
1879 mac->port.p = in_xbundle->ofbundle;
1880 mac_learning_changed(xbridge->ml);
1885 update_learning_table(const struct xbridge *xbridge,
1886 const struct flow *flow, struct flow_wildcards *wc,
1887 int vlan, struct xbundle *in_xbundle)
1891 /* Don't learn the OFPP_NONE port. */
1892 if (in_xbundle == &ofpp_none_bundle) {
1896 /* First try the common case: no change to MAC learning table. */
1897 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1898 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1900 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1903 /* Slow path: MAC learning table might need an update. */
1904 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1905 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1906 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1910 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1911 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
1913 update_mcast_snooping_table__(const struct xbridge *xbridge,
1914 const struct flow *flow,
1915 struct mcast_snooping *ms,
1916 ovs_be32 ip4, int vlan,
1917 struct xbundle *in_xbundle)
1918 OVS_REQ_WRLOCK(ms->rwlock)
1920 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
1922 switch (ntohs(flow->tp_src)) {
1923 case IGMP_HOST_MEMBERSHIP_REPORT:
1924 case IGMPV2_HOST_MEMBERSHIP_REPORT:
1925 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1926 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
1927 IP_FMT" is on port %s in VLAN %d",
1928 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1931 case IGMP_HOST_LEAVE_MESSAGE:
1932 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1933 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
1934 IP_FMT" is on port %s in VLAN %d",
1935 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1938 case IGMP_HOST_MEMBERSHIP_QUERY:
1939 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
1940 in_xbundle->ofbundle)) {
1941 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
1942 IP_FMT" is on port %s in VLAN %d",
1943 xbridge->name, IP_ARGS(flow->nw_src),
1944 in_xbundle->name, vlan);
1950 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1951 * was received on 'in_xbundle' in 'vlan'. */
1953 update_mcast_snooping_table(const struct xbridge *xbridge,
1954 const struct flow *flow, int vlan,
1955 struct xbundle *in_xbundle)
1957 struct mcast_snooping *ms = xbridge->ms;
1958 struct xlate_cfg *xcfg;
1959 struct xbundle *mcast_xbundle;
1960 struct mcast_fport_bundle *fport;
1962 /* Don't learn the OFPP_NONE port. */
1963 if (in_xbundle == &ofpp_none_bundle) {
1967 /* Don't learn from flood ports */
1968 mcast_xbundle = NULL;
1969 ovs_rwlock_wrlock(&ms->rwlock);
1970 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1971 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
1972 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
1973 if (mcast_xbundle == in_xbundle) {
1978 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
1979 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
1982 ovs_rwlock_unlock(&ms->rwlock);
1985 /* send the packet to ports having the multicast group learned */
1987 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
1988 struct mcast_snooping *ms OVS_UNUSED,
1989 struct mcast_group *grp,
1990 struct xbundle *in_xbundle, uint16_t vlan)
1991 OVS_REQ_RDLOCK(ms->rwlock)
1993 struct xlate_cfg *xcfg;
1994 struct mcast_group_bundle *b;
1995 struct xbundle *mcast_xbundle;
1997 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1998 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
1999 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2000 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2001 xlate_report(ctx, "forwarding to mcast group port");
2002 output_normal(ctx, mcast_xbundle, vlan);
2003 } else if (!mcast_xbundle) {
2004 xlate_report(ctx, "mcast group port is unknown, dropping");
2006 xlate_report(ctx, "mcast group port is input port, dropping");
2011 /* send the packet to ports connected to multicast routers */
2013 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2014 struct mcast_snooping *ms,
2015 struct xbundle *in_xbundle, uint16_t vlan)
2016 OVS_REQ_RDLOCK(ms->rwlock)
2018 struct xlate_cfg *xcfg;
2019 struct mcast_mrouter_bundle *mrouter;
2020 struct xbundle *mcast_xbundle;
2022 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2023 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2024 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2025 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2026 xlate_report(ctx, "forwarding to mcast router port");
2027 output_normal(ctx, mcast_xbundle, vlan);
2028 } else if (!mcast_xbundle) {
2029 xlate_report(ctx, "mcast router port is unknown, dropping");
2031 xlate_report(ctx, "mcast router port is input port, dropping");
2036 /* send the packet to ports flagged to be flooded */
2038 xlate_normal_mcast_send_fports(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_fport_bundle *fport;
2045 struct xbundle *mcast_xbundle;
2047 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2048 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
2049 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2050 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2051 xlate_report(ctx, "forwarding to mcast flood port");
2052 output_normal(ctx, mcast_xbundle, vlan);
2053 } else if (!mcast_xbundle) {
2054 xlate_report(ctx, "mcast flood port is unknown, dropping");
2056 xlate_report(ctx, "mcast flood port is input port, dropping");
2062 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2065 struct xbundle *xbundle;
2067 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2068 if (xbundle != in_xbundle
2069 && xbundle_includes_vlan(xbundle, vlan)
2070 && xbundle->floodable
2071 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2072 output_normal(ctx, xbundle, vlan);
2075 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2079 xlate_normal(struct xlate_ctx *ctx)
2081 struct flow_wildcards *wc = &ctx->xout->wc;
2082 struct flow *flow = &ctx->xin->flow;
2083 struct xbundle *in_xbundle;
2084 struct xport *in_port;
2085 struct mac_entry *mac;
2090 ctx->xout->has_normal = true;
2092 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2093 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2094 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2096 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2097 ctx->xin->packet != NULL, &in_port);
2099 xlate_report(ctx, "no input bundle, dropping");
2103 /* Drop malformed frames. */
2104 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2105 !(flow->vlan_tci & htons(VLAN_CFI))) {
2106 if (ctx->xin->packet != NULL) {
2107 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2108 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2109 "VLAN tag received on port %s",
2110 ctx->xbridge->name, in_xbundle->name);
2112 xlate_report(ctx, "partial VLAN tag, dropping");
2116 /* Drop frames on bundles reserved for mirroring. */
2117 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2118 if (ctx->xin->packet != NULL) {
2119 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2120 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2121 "%s, which is reserved exclusively for mirroring",
2122 ctx->xbridge->name, in_xbundle->name);
2124 xlate_report(ctx, "input port is mirror output port, dropping");
2129 vid = vlan_tci_to_vid(flow->vlan_tci);
2130 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2131 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2134 vlan = input_vid_to_vlan(in_xbundle, vid);
2136 /* Check other admissibility requirements. */
2137 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2141 /* Learn source MAC. */
2142 if (ctx->xin->may_learn) {
2143 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2145 if (ctx->xin->xcache) {
2146 struct xc_entry *entry;
2148 /* Save enough info to update mac learning table later. */
2149 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2150 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2151 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2152 entry->u.normal.vlan = vlan;
2155 /* Determine output bundle. */
2156 if (mcast_snooping_enabled(ctx->xbridge->ms)
2157 && !eth_addr_is_broadcast(flow->dl_dst)
2158 && eth_addr_is_multicast(flow->dl_dst)
2159 && flow->dl_type == htons(ETH_TYPE_IP)) {
2160 struct mcast_snooping *ms = ctx->xbridge->ms;
2161 struct mcast_group *grp;
2163 if (flow->nw_proto == IPPROTO_IGMP) {
2164 if (ctx->xin->may_learn) {
2165 if (mcast_snooping_is_membership(flow->tp_src) ||
2166 mcast_snooping_is_query(flow->tp_src)) {
2167 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2172 if (mcast_snooping_is_membership(flow->tp_src)) {
2173 ovs_rwlock_rdlock(&ms->rwlock);
2174 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2175 ovs_rwlock_unlock(&ms->rwlock);
2177 xlate_report(ctx, "multicast traffic, flooding");
2178 xlate_normal_flood(ctx, in_xbundle, vlan);
2182 if (ip_is_local_multicast(flow->nw_dst)) {
2183 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2184 * address in the 224.0.0.x range which are not IGMP must
2185 * be forwarded on all ports */
2186 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2187 xlate_normal_flood(ctx, in_xbundle, vlan);
2192 /* forwarding to group base ports */
2193 ovs_rwlock_rdlock(&ms->rwlock);
2194 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2196 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2197 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2198 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2200 if (mcast_snooping_flood_unreg(ms)) {
2201 xlate_report(ctx, "unregistered multicast, flooding");
2202 xlate_normal_flood(ctx, in_xbundle, vlan);
2204 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2205 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2208 ovs_rwlock_unlock(&ms->rwlock);
2210 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2211 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2212 mac_port = mac ? mac->port.p : NULL;
2213 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2216 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2217 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2218 if (mac_xbundle && mac_xbundle != in_xbundle) {
2219 xlate_report(ctx, "forwarding to learned port");
2220 output_normal(ctx, mac_xbundle, vlan);
2221 } else if (!mac_xbundle) {
2222 xlate_report(ctx, "learned port is unknown, dropping");
2224 xlate_report(ctx, "learned port is input port, dropping");
2227 xlate_report(ctx, "no learned MAC for destination, flooding");
2228 xlate_normal_flood(ctx, in_xbundle, vlan);
2233 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2234 * the number of packets out of UINT32_MAX to sample. The given
2235 * cookie is passed back in the callback for each sampled packet.
2238 compose_sample_action(const struct xbridge *xbridge,
2239 struct ofpbuf *odp_actions,
2240 const struct flow *flow,
2241 const uint32_t probability,
2242 const union user_action_cookie *cookie,
2243 const size_t cookie_size,
2244 const odp_port_t tunnel_out_port)
2246 size_t sample_offset, actions_offset;
2247 odp_port_t odp_port;
2251 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2253 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2255 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2257 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2258 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2259 flow_hash_5tuple(flow, 0));
2260 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2261 tunnel_out_port, odp_actions);
2263 nl_msg_end_nested(odp_actions, actions_offset);
2264 nl_msg_end_nested(odp_actions, sample_offset);
2265 return cookie_offset;
2269 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2270 odp_port_t odp_port, unsigned int n_outputs,
2271 union user_action_cookie *cookie)
2275 cookie->type = USER_ACTION_COOKIE_SFLOW;
2276 cookie->sflow.vlan_tci = vlan_tci;
2278 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2279 * port information") for the interpretation of cookie->output. */
2280 switch (n_outputs) {
2282 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2283 cookie->sflow.output = 0x40000000 | 256;
2287 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2289 cookie->sflow.output = ifindex;
2294 /* 0x80000000 means "multiple output ports. */
2295 cookie->sflow.output = 0x80000000 | n_outputs;
2300 /* Compose SAMPLE action for sFlow bridge sampling. */
2302 compose_sflow_action(const struct xbridge *xbridge,
2303 struct ofpbuf *odp_actions,
2304 const struct flow *flow,
2305 odp_port_t odp_port)
2307 uint32_t probability;
2308 union user_action_cookie cookie;
2310 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2314 probability = dpif_sflow_get_probability(xbridge->sflow);
2315 compose_sflow_cookie(xbridge, htons(0), odp_port,
2316 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2318 return compose_sample_action(xbridge, odp_actions, flow, probability,
2319 &cookie, sizeof cookie.sflow, ODPP_NONE);
2323 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2324 uint32_t obs_domain_id, uint32_t obs_point_id,
2325 union user_action_cookie *cookie)
2327 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2328 cookie->flow_sample.probability = probability;
2329 cookie->flow_sample.collector_set_id = collector_set_id;
2330 cookie->flow_sample.obs_domain_id = obs_domain_id;
2331 cookie->flow_sample.obs_point_id = obs_point_id;
2335 compose_ipfix_cookie(union user_action_cookie *cookie,
2336 odp_port_t output_odp_port)
2338 cookie->type = USER_ACTION_COOKIE_IPFIX;
2339 cookie->ipfix.output_odp_port = output_odp_port;
2342 /* Compose SAMPLE action for IPFIX bridge sampling. */
2344 compose_ipfix_action(const struct xbridge *xbridge,
2345 struct ofpbuf *odp_actions,
2346 const struct flow *flow,
2347 odp_port_t output_odp_port)
2349 uint32_t probability;
2350 union user_action_cookie cookie;
2351 odp_port_t tunnel_out_port = ODPP_NONE;
2353 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2357 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2359 if (output_odp_port == ODPP_NONE &&
2360 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2364 /* For output case, output_odp_port is valid*/
2365 if (output_odp_port != ODPP_NONE) {
2366 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2369 /* If tunnel sampling is enabled, put an additional option attribute:
2370 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2372 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2373 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2374 tunnel_out_port = output_odp_port;
2378 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2379 compose_ipfix_cookie(&cookie, output_odp_port);
2381 compose_sample_action(xbridge, odp_actions, flow, probability,
2382 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2385 /* SAMPLE action for sFlow must be first action in any given list of
2386 * actions. At this point we do not have all information required to
2387 * build it. So try to build sample action as complete as possible. */
2389 add_sflow_action(struct xlate_ctx *ctx)
2391 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2392 ctx->xout->odp_actions,
2393 &ctx->xin->flow, ODPP_NONE);
2394 ctx->sflow_odp_port = 0;
2395 ctx->sflow_n_outputs = 0;
2398 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2399 * of actions, eventually after the SAMPLE action for sFlow. */
2401 add_ipfix_action(struct xlate_ctx *ctx)
2403 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2404 &ctx->xin->flow, ODPP_NONE);
2408 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2410 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2411 &ctx->xin->flow, port);
2414 /* Fix SAMPLE action according to data collected while composing ODP actions.
2415 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2416 * USERSPACE action's user-cookie which is required for sflow. */
2418 fix_sflow_action(struct xlate_ctx *ctx)
2420 const struct flow *base = &ctx->base_flow;
2421 union user_action_cookie *cookie;
2423 if (!ctx->user_cookie_offset) {
2427 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2428 sizeof cookie->sflow);
2429 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2431 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2432 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2435 static enum slow_path_reason
2436 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2437 const struct xport *xport, const struct ofpbuf *packet)
2439 struct flow_wildcards *wc = &ctx->xout->wc;
2440 const struct xbridge *xbridge = ctx->xbridge;
2444 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2446 cfm_process_heartbeat(xport->cfm, packet);
2449 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2451 bfd_process_packet(xport->bfd, flow, packet);
2452 /* If POLL received, immediately sends FINAL back. */
2453 if (bfd_should_send_packet(xport->bfd)) {
2454 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2458 } else if (xport->xbundle && xport->xbundle->lacp
2459 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2461 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2464 } else if ((xbridge->stp || xbridge->rstp) &&
2465 stp_should_process_flow(flow, wc)) {
2468 ? stp_process_packet(xport, packet)
2469 : rstp_process_packet(xport, packet);
2478 tnl_route_lookup_flow(const struct flow *oflow,
2479 ovs_be32 *ip, struct xport **out_port)
2481 char out_dev[IFNAMSIZ];
2482 struct xbridge *xbridge;
2483 struct xlate_cfg *xcfg;
2486 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2493 *ip = oflow->tunnel.ip_dst;
2496 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2499 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2500 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2503 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2504 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2515 xlate_flood_packet(struct xbridge *xbridge, struct ofpbuf *packet)
2517 struct ofpact_output output;
2520 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2521 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2522 flow_extract(packet, NULL, &flow);
2523 flow.in_port.ofp_port = OFPP_NONE;
2524 output.port = OFPP_FLOOD;
2527 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2528 &output.ofpact, sizeof output,
2533 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2534 ovs_be32 ip_src, ovs_be32 ip_dst)
2536 struct xbridge *xbridge = out_dev->xbridge;
2537 struct ofpbuf packet;
2539 ofpbuf_init(&packet, 0);
2540 compose_arp(&packet, eth_src, ip_src, ip_dst);
2542 xlate_flood_packet(xbridge, &packet);
2543 ofpbuf_uninit(&packet);
2547 build_tunnel_send(const struct xlate_ctx *ctx, const struct xport *xport,
2548 const struct flow *flow, odp_port_t tunnel_odp_port)
2550 struct ovs_action_push_tnl tnl_push_data;
2551 struct xport *out_dev = NULL;
2552 ovs_be32 s_ip, d_ip = 0;
2553 uint8_t smac[ETH_ADDR_LEN];
2554 uint8_t dmac[ETH_ADDR_LEN];
2557 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2562 /* Use mac addr of bridge port of the peer. */
2563 err = netdev_get_etheraddr(out_dev->netdev, smac);
2568 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2573 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2575 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2578 if (ctx->xin->xcache) {
2579 struct xc_entry *entry;
2581 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2582 strncpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name, IFNAMSIZ);
2583 entry->u.tnl_arp_cache.d_ip = d_ip;
2585 err = tnl_port_build_header(xport->ofport, flow,
2586 dmac, smac, s_ip, &tnl_push_data);
2590 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2591 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2592 odp_put_tnl_push_action(ctx->xout->odp_actions, &tnl_push_data);
2597 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2600 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2601 struct flow_wildcards *wc = &ctx->xout->wc;
2602 struct flow *flow = &ctx->xin->flow;
2603 struct flow_tnl flow_tnl;
2604 ovs_be16 flow_vlan_tci;
2605 uint32_t flow_pkt_mark;
2606 uint8_t flow_nw_tos;
2607 odp_port_t out_port, odp_port;
2608 bool tnl_push_pop_send = false;
2611 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2612 * before traversing a patch port. */
2613 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 28);
2614 memset(&flow_tnl, 0, sizeof flow_tnl);
2617 xlate_report(ctx, "Nonexistent output port");
2619 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2620 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2622 } else if (check_stp) {
2623 if (is_stp(&ctx->base_flow)) {
2624 if (!xport_stp_should_forward_bpdu(xport) &&
2625 !xport_rstp_should_manage_bpdu(xport)) {
2626 if (ctx->xbridge->stp != NULL) {
2627 xlate_report(ctx, "STP not in listening state, "
2628 "skipping bpdu output");
2629 } else if (ctx->xbridge->rstp != NULL) {
2630 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2631 "skipping bpdu output");
2635 } else if (!xport_stp_forward_state(xport) ||
2636 !xport_rstp_forward_state(xport)) {
2637 if (ctx->xbridge->stp != NULL) {
2638 xlate_report(ctx, "STP not in forwarding state, "
2640 } else if (ctx->xbridge->rstp != NULL) {
2641 xlate_report(ctx, "RSTP not in forwarding state, "
2648 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2649 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2654 const struct xport *peer = xport->peer;
2655 struct flow old_flow = ctx->xin->flow;
2656 enum slow_path_reason special;
2657 uint8_t table_id = rule_dpif_lookup_get_init_table_id(&ctx->xin->flow);
2659 ctx->xbridge = peer->xbridge;
2660 flow->in_port.ofp_port = peer->ofp_port;
2661 flow->metadata = htonll(0);
2662 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2663 memset(flow->regs, 0, sizeof flow->regs);
2664 flow->actset_output = OFPP_UNSET;
2666 special = process_special(ctx, &ctx->xin->flow, peer,
2669 ctx->xout->slow |= special;
2670 } else if (may_receive(peer, ctx)) {
2671 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2672 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2675 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2676 * the learning action look at the packet, then drop it. */
2677 struct flow old_base_flow = ctx->base_flow;
2678 size_t old_size = ofpbuf_size(ctx->xout->odp_actions);
2679 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2680 xlate_table_action(ctx, flow->in_port.ofp_port, table_id,
2682 ctx->xout->mirrors = old_mirrors;
2683 ctx->base_flow = old_base_flow;
2684 ofpbuf_set_size(ctx->xout->odp_actions, old_size);
2688 ctx->xin->flow = old_flow;
2689 ctx->xbridge = xport->xbridge;
2691 if (ctx->xin->resubmit_stats) {
2692 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2693 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2695 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2698 if (ctx->xin->xcache) {
2699 struct xc_entry *entry;
2701 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2702 entry->u.dev.tx = netdev_ref(xport->netdev);
2703 entry->u.dev.rx = netdev_ref(peer->netdev);
2704 entry->u.dev.bfd = bfd_ref(peer->bfd);
2709 flow_vlan_tci = flow->vlan_tci;
2710 flow_pkt_mark = flow->pkt_mark;
2711 flow_nw_tos = flow->nw_tos;
2713 if (count_skb_priorities(xport)) {
2714 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2715 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2716 wc->masks.nw_tos |= IP_DSCP_MASK;
2717 flow->nw_tos &= ~IP_DSCP_MASK;
2718 flow->nw_tos |= dscp;
2722 if (xport->is_tunnel) {
2723 /* Save tunnel metadata so that changes made due to
2724 * the Logical (tunnel) Port are not visible for any further
2725 * matches, while explicit set actions on tunnel metadata are.
2727 flow_tnl = flow->tunnel;
2728 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2729 if (odp_port == ODPP_NONE) {
2730 xlate_report(ctx, "Tunneling decided against output");
2731 goto out; /* restore flow_nw_tos */
2733 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2734 xlate_report(ctx, "Not tunneling to our own address");
2735 goto out; /* restore flow_nw_tos */
2737 if (ctx->xin->resubmit_stats) {
2738 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2740 if (ctx->xin->xcache) {
2741 struct xc_entry *entry;
2743 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2744 entry->u.dev.tx = netdev_ref(xport->netdev);
2746 out_port = odp_port;
2747 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2748 tnl_push_pop_send = true;
2750 commit_odp_tunnel_action(flow, &ctx->base_flow,
2751 ctx->xout->odp_actions);
2752 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2755 odp_port = xport->odp_port;
2756 out_port = odp_port;
2757 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2758 ofp_port_t vlandev_port;
2760 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2761 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2762 ofp_port, flow->vlan_tci);
2763 if (vlandev_port != ofp_port) {
2764 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2765 flow->vlan_tci = htons(0);
2770 if (out_port != ODPP_NONE) {
2771 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2772 ctx->xout->odp_actions,
2774 ctx->xbridge->masked_set_action);
2776 if (ctx->use_recirc) {
2777 struct ovs_action_hash *act_hash;
2778 struct xlate_recirc *xr = &ctx->recirc;
2781 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2782 OVS_ACTION_ATTR_HASH,
2784 act_hash->hash_alg = xr->hash_alg;
2785 act_hash->hash_basis = xr->hash_basis;
2787 /* Recirc action. */
2788 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2792 if (tnl_push_pop_send) {
2793 build_tunnel_send(ctx, xport, flow, odp_port);
2794 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2796 odp_port_t odp_tnl_port = ODPP_NONE;
2798 /* XXX: Write better Filter for tunnel port. We can use inport
2799 * int tunnel-port flow to avoid these checks completely. */
2800 if (ofp_port == OFPP_LOCAL &&
2801 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2803 odp_tnl_port = tnl_port_map_lookup(flow, wc);
2806 if (odp_tnl_port != ODPP_NONE) {
2807 nl_msg_put_odp_port(ctx->xout->odp_actions,
2808 OVS_ACTION_ATTR_TUNNEL_POP,
2811 /* Tunnel push-pop action is not compatible with
2813 add_ipfix_output_action(ctx, out_port);
2814 nl_msg_put_odp_port(ctx->xout->odp_actions,
2815 OVS_ACTION_ATTR_OUTPUT,
2821 ctx->sflow_odp_port = odp_port;
2822 ctx->sflow_n_outputs++;
2823 ctx->xout->nf_output_iface = ofp_port;
2828 flow->vlan_tci = flow_vlan_tci;
2829 flow->pkt_mark = flow_pkt_mark;
2830 flow->nw_tos = flow_nw_tos;
2834 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
2836 compose_output_action__(ctx, ofp_port, true);
2840 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
2842 struct rule_dpif *old_rule = ctx->rule;
2843 const struct rule_actions *actions;
2845 if (ctx->xin->resubmit_stats) {
2846 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2852 actions = rule_dpif_get_actions(rule);
2853 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2854 ctx->rule = old_rule;
2859 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2861 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2863 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2864 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2865 MAX_RESUBMIT_RECURSION);
2866 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2867 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2868 } else if (ofpbuf_size(ctx->xout->odp_actions) > UINT16_MAX) {
2869 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2870 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2871 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2880 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2881 bool may_packet_in, bool honor_table_miss)
2883 if (xlate_resubmit_resource_check(ctx)) {
2884 struct flow_wildcards *wc;
2885 uint8_t old_table_id = ctx->table_id;
2886 struct rule_dpif *rule;
2888 ctx->table_id = table_id;
2889 wc = (ctx->xin->skip_wildcards) ? NULL : &ctx->xout->wc;
2891 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2892 &ctx->xin->flow, wc,
2893 ctx->xin->xcache != NULL,
2894 ctx->xin->resubmit_stats,
2895 &ctx->table_id, in_port,
2896 may_packet_in, honor_table_miss);
2898 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
2899 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
2903 /* Fill in the cache entry here instead of xlate_recursively
2904 * to make the reference counting more explicit. We take a
2905 * reference in the lookups above if we are going to cache the
2907 if (ctx->xin->xcache) {
2908 struct xc_entry *entry;
2910 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2911 entry->u.rule = rule;
2913 xlate_recursively(ctx, rule);
2916 ctx->table_id = old_table_id;
2924 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
2925 struct ofputil_bucket *bucket)
2927 if (ctx->xin->resubmit_stats) {
2928 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
2930 if (ctx->xin->xcache) {
2931 struct xc_entry *entry;
2933 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
2934 entry->u.group.group = group_dpif_ref(group);
2935 entry->u.group.bucket = bucket;
2940 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
2942 uint64_t action_list_stub[1024 / 8];
2943 struct ofpbuf action_list, action_set;
2945 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2946 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2948 ofpacts_execute_action_set(&action_list, &action_set);
2950 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2953 ofpbuf_uninit(&action_set);
2954 ofpbuf_uninit(&action_list);
2958 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2960 struct ofputil_bucket *bucket;
2961 const struct ovs_list *buckets;
2962 struct flow old_flow = ctx->xin->flow;
2964 group_dpif_get_buckets(group, &buckets);
2966 LIST_FOR_EACH (bucket, list_node, buckets) {
2967 xlate_group_bucket(ctx, bucket);
2968 /* Roll back flow to previous state.
2969 * This is equivalent to cloning the packet for each bucket.
2971 * As a side effect any subsequently applied actions will
2972 * also effectively be applied to a clone of the packet taken
2973 * just before applying the all or indirect group. */
2974 ctx->xin->flow = old_flow;
2976 xlate_group_stats(ctx, group, NULL);
2980 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2982 struct ofputil_bucket *bucket;
2984 bucket = group_first_live_bucket(ctx, group, 0);
2986 xlate_group_bucket(ctx, bucket);
2987 xlate_group_stats(ctx, group, bucket);
2992 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2994 struct flow_wildcards *wc = &ctx->xout->wc;
2995 struct ofputil_bucket *bucket;
2998 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
2999 bucket = group_best_live_bucket(ctx, group, basis);
3001 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3002 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3003 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3004 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3005 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3006 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3007 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3008 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3009 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3011 xlate_group_bucket(ctx, bucket);
3012 xlate_group_stats(ctx, group, bucket);
3017 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3019 ctx->in_group = true;
3021 switch (group_dpif_get_type(group)) {
3023 case OFPGT11_INDIRECT:
3024 xlate_all_group(ctx, group);
3026 case OFPGT11_SELECT:
3027 xlate_select_group(ctx, group);
3030 xlate_ff_group(ctx, group);
3035 group_dpif_unref(group);
3037 ctx->in_group = false;
3041 xlate_group_resource_check(struct xlate_ctx *ctx)
3043 if (!xlate_resubmit_resource_check(ctx)) {
3045 } else if (ctx->in_group) {
3046 /* Prevent nested translation of OpenFlow groups.
3048 * OpenFlow allows this restriction. We enforce this restriction only
3049 * because, with the current architecture, we would otherwise have to
3050 * take a possibly recursive read lock on the ofgroup rwlock, which is
3051 * unsafe given that POSIX allows taking a read lock to block if there
3052 * is a thread blocked on taking the write lock. Other solutions
3053 * without this restriction are also possible, but seem unwarranted
3054 * given the current limited use of groups. */
3055 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3057 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
3065 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3067 if (xlate_group_resource_check(ctx)) {
3068 struct group_dpif *group;
3071 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3073 xlate_group_action__(ctx, group);
3083 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3084 const struct ofpact_resubmit *resubmit)
3088 bool may_packet_in = false;
3089 bool honor_table_miss = false;
3091 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3092 /* Still allow missed packets to be sent to the controller
3093 * if resubmitting from an internal table. */
3094 may_packet_in = true;
3095 honor_table_miss = true;
3098 in_port = resubmit->in_port;
3099 if (in_port == OFPP_IN_PORT) {
3100 in_port = ctx->xin->flow.in_port.ofp_port;
3103 table_id = resubmit->table_id;
3104 if (table_id == 255) {
3105 table_id = ctx->table_id;
3108 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3113 flood_packets(struct xlate_ctx *ctx, bool all)
3115 const struct xport *xport;
3117 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3118 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3123 compose_output_action__(ctx, xport->ofp_port, false);
3124 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3125 compose_output_action(ctx, xport->ofp_port);
3129 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3133 execute_controller_action(struct xlate_ctx *ctx, int len,
3134 enum ofp_packet_in_reason reason,
3135 uint16_t controller_id)
3137 struct ofproto_packet_in *pin;
3138 struct dpif_packet *packet;
3140 ctx->xout->slow |= SLOW_CONTROLLER;
3141 if (!ctx->xin->packet) {
3145 packet = dpif_packet_clone_from_ofpbuf(ctx->xin->packet);
3147 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3148 ctx->xout->odp_actions,
3150 ctx->xbridge->masked_set_action);
3152 odp_execute_actions(NULL, &packet, 1, false,
3153 ofpbuf_data(ctx->xout->odp_actions),
3154 ofpbuf_size(ctx->xout->odp_actions), NULL);
3156 pin = xmalloc(sizeof *pin);
3157 pin->up.packet_len = ofpbuf_size(&packet->ofpbuf);
3158 pin->up.packet = ofpbuf_steal_data(&packet->ofpbuf);
3159 pin->up.reason = reason;
3160 pin->up.table_id = ctx->table_id;
3161 pin->up.cookie = (ctx->rule
3162 ? rule_dpif_get_flow_cookie(ctx->rule)
3165 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3167 pin->controller_id = controller_id;
3168 pin->send_len = len;
3169 /* If a rule is a table-miss rule then this is
3170 * a table-miss handled by a table-miss rule.
3172 * Else, if rule is internal and has a controller action,
3173 * the later being implied by the rule being processed here,
3174 * then this is a table-miss handled without a table-miss rule.
3176 * Otherwise this is not a table-miss. */
3177 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3179 if (rule_dpif_is_table_miss(ctx->rule)) {
3180 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3181 } else if (rule_dpif_is_internal(ctx->rule)) {
3182 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3185 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3186 dpif_packet_delete(packet);
3190 compose_recirculate_action(struct xlate_ctx *ctx,
3191 const struct ofpact *ofpacts_base,
3192 const struct ofpact *ofpact_current,
3193 size_t ofpacts_base_len)
3197 unsigned ofpacts_len;
3200 struct ofpbuf ofpacts;
3204 ofpacts_len = ofpacts_base_len -
3205 ((uint8_t *)ofpact_current - (uint8_t *)ofpacts_base);
3208 id = rule_dpif_get_recirc_id(ctx->rule);
3210 /* In the case where ctx has no rule then allocate a recirc id.
3211 * The life-cycle of this recirc id is managed by associating it
3212 * with the internal rule that is created to to handle
3213 * recirculation below.
3215 * The known use-case of this is packet_out which
3216 * translates actions without a rule */
3217 id = ofproto_dpif_alloc_recirc_id(ctx->xbridge->ofproto);
3220 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3221 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3225 match_init_catchall(&match);
3226 match_set_recirc_id(&match, id);
3227 ofpbuf_use_const(&ofpacts, ofpact_current, ofpacts_len);
3228 error = ofproto_dpif_add_internal_flow(ctx->xbridge->ofproto, &match,
3229 RECIRC_RULE_PRIORITY,
3230 RECIRC_TIMEOUT, &ofpacts, &rule);
3232 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3233 VLOG_ERR_RL(&rl, "Failed to add post recirculation flow %s",
3234 match_to_string(&match, 0));
3237 /* If ctx has no rule then associate the recirc id, which
3238 * was allocated above, with the internal rule. This allows
3239 * the recirc id to be released when the internal rule times out. */
3241 rule_set_recirc_id(rule, id);
3244 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3245 ctx->xout->odp_actions,
3247 ctx->xbridge->masked_set_action);
3248 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3252 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3254 struct flow_wildcards *wc = &ctx->xout->wc;
3255 struct flow *flow = &ctx->xin->flow;
3258 ovs_assert(eth_type_mpls(mpls->ethertype));
3260 n = flow_count_mpls_labels(flow, wc);
3262 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3263 ctx->xout->odp_actions,
3265 ctx->xbridge->masked_set_action);
3266 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3267 if (ctx->xin->packet != NULL) {
3268 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3269 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3270 "MPLS push action can't be performed as it would "
3271 "have more MPLS LSEs than the %d supported.",
3272 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3278 flow_push_mpls(flow, n, mpls->ethertype, wc);
3282 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3284 struct flow_wildcards *wc = &ctx->xout->wc;
3285 struct flow *flow = &ctx->xin->flow;
3286 int n = flow_count_mpls_labels(flow, wc);
3288 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3289 if (ctx->xbridge->enable_recirc) {
3290 ctx->was_mpls = true;
3292 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3293 if (ctx->xin->packet != NULL) {
3294 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3295 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3296 "MPLS pop action can't be performed as it has "
3297 "more MPLS LSEs than the %d supported.",
3298 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3301 ofpbuf_clear(ctx->xout->odp_actions);
3306 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3308 struct flow *flow = &ctx->xin->flow;
3310 if (!is_ip_any(flow)) {
3314 ctx->xout->wc.masks.nw_ttl = 0xff;
3315 if (flow->nw_ttl > 1) {
3321 for (i = 0; i < ids->n_controllers; i++) {
3322 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3326 /* Stop processing for current table. */
3332 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3334 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3335 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3336 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3341 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3343 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3344 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3345 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3350 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3352 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3353 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3354 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3359 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3361 struct flow *flow = &ctx->xin->flow;
3362 struct flow_wildcards *wc = &ctx->xout->wc;
3364 if (eth_type_mpls(flow->dl_type)) {
3365 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3367 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3370 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3373 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3377 /* Stop processing for current table. */
3382 xlate_output_action(struct xlate_ctx *ctx,
3383 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3385 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3387 ctx->xout->nf_output_iface = NF_OUT_DROP;
3391 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
3394 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3395 0, may_packet_in, true);
3401 flood_packets(ctx, false);
3404 flood_packets(ctx, true);
3406 case OFPP_CONTROLLER:
3407 execute_controller_action(ctx, max_len,
3408 ctx->in_group ? OFPR_GROUP : OFPR_ACTION, 0);
3414 if (port != ctx->xin->flow.in_port.ofp_port) {
3415 compose_output_action(ctx, port);
3417 xlate_report(ctx, "skipping output to input port");
3422 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3423 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3424 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3425 ctx->xout->nf_output_iface = prev_nf_output_iface;
3426 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3427 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3428 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3433 xlate_output_reg_action(struct xlate_ctx *ctx,
3434 const struct ofpact_output_reg *or)
3436 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3437 if (port <= UINT16_MAX) {
3438 union mf_subvalue value;
3440 memset(&value, 0xff, sizeof value);
3441 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3442 xlate_output_action(ctx, u16_to_ofp(port),
3443 or->max_len, false);
3448 xlate_enqueue_action(struct xlate_ctx *ctx,
3449 const struct ofpact_enqueue *enqueue)
3451 ofp_port_t ofp_port = enqueue->port;
3452 uint32_t queue_id = enqueue->queue;
3453 uint32_t flow_priority, priority;
3456 /* Translate queue to priority. */
3457 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3459 /* Fall back to ordinary output action. */
3460 xlate_output_action(ctx, enqueue->port, 0, false);
3464 /* Check output port. */
3465 if (ofp_port == OFPP_IN_PORT) {
3466 ofp_port = ctx->xin->flow.in_port.ofp_port;
3467 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3471 /* Add datapath actions. */
3472 flow_priority = ctx->xin->flow.skb_priority;
3473 ctx->xin->flow.skb_priority = priority;
3474 compose_output_action(ctx, ofp_port);
3475 ctx->xin->flow.skb_priority = flow_priority;
3477 /* Update NetFlow output port. */
3478 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3479 ctx->xout->nf_output_iface = ofp_port;
3480 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3481 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3486 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3488 uint32_t skb_priority;
3490 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3491 ctx->xin->flow.skb_priority = skb_priority;
3493 /* Couldn't translate queue to a priority. Nothing to do. A warning
3494 * has already been logged. */
3499 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3501 const struct xbridge *xbridge = xbridge_;
3512 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3515 port = get_ofp_port(xbridge, ofp_port);
3516 return port ? port->may_enable : false;
3521 xlate_bundle_action(struct xlate_ctx *ctx,
3522 const struct ofpact_bundle *bundle)
3526 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3528 CONST_CAST(struct xbridge *, ctx->xbridge));
3529 if (bundle->dst.field) {
3530 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3533 xlate_output_action(ctx, port, 0, false);
3538 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3539 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3541 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3542 if (ctx->xin->may_learn) {
3543 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3548 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3550 ctx->xout->has_learn = true;
3551 learn_mask(learn, &ctx->xout->wc);
3553 if (ctx->xin->xcache) {
3554 struct xc_entry *entry;
3556 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3557 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3558 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3559 entry->u.learn.ofpacts = ofpbuf_new(64);
3560 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3561 entry->u.learn.ofpacts);
3562 } else if (ctx->xin->may_learn) {
3563 uint64_t ofpacts_stub[1024 / 8];
3564 struct ofputil_flow_mod fm;
3565 struct ofpbuf ofpacts;
3567 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3568 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3569 ofpbuf_uninit(&ofpacts);
3574 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3575 uint16_t idle_timeout, uint16_t hard_timeout)
3577 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3578 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3583 xlate_fin_timeout(struct xlate_ctx *ctx,
3584 const struct ofpact_fin_timeout *oft)
3587 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3588 oft->fin_idle_timeout, oft->fin_hard_timeout);
3589 if (ctx->xin->xcache) {
3590 struct xc_entry *entry;
3592 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3593 /* XC_RULE already holds a reference on the rule, none is taken
3595 entry->u.fin.rule = ctx->rule;
3596 entry->u.fin.idle = oft->fin_idle_timeout;
3597 entry->u.fin.hard = oft->fin_hard_timeout;
3603 xlate_sample_action(struct xlate_ctx *ctx,
3604 const struct ofpact_sample *os)
3606 union user_action_cookie cookie;
3607 /* Scale the probability from 16-bit to 32-bit while representing
3608 * the same percentage. */
3609 uint32_t probability = (os->probability << 16) | os->probability;
3611 if (!ctx->xbridge->variable_length_userdata) {
3612 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3614 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3615 "lacks support (needs Linux 3.10+ or kernel module from "
3620 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3621 ctx->xout->odp_actions,
3623 ctx->xbridge->masked_set_action);
3625 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3626 os->obs_domain_id, os->obs_point_id, &cookie);
3627 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions, &ctx->xin->flow,
3628 probability, &cookie, sizeof cookie.flow_sample,
3633 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3635 if (xport->config & (is_stp(&ctx->xin->flow)
3636 ? OFPUTIL_PC_NO_RECV_STP
3637 : OFPUTIL_PC_NO_RECV)) {
3641 /* Only drop packets here if both forwarding and learning are
3642 * disabled. If just learning is enabled, we need to have
3643 * OFPP_NORMAL and the learning action have a look at the packet
3644 * before we can drop it. */
3645 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3646 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3654 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3656 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3657 size_t on_len = ofpact_nest_get_action_len(on);
3658 const struct ofpact *inner;
3660 /* Maintain actset_output depending on the contents of the action set:
3662 * - OFPP_UNSET, if there is no "output" action.
3664 * - The output port, if there is an "output" action and no "group"
3667 * - OFPP_UNSET, if there is a "group" action.
3669 if (!ctx->action_set_has_group) {
3670 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3671 if (inner->type == OFPACT_OUTPUT) {
3672 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3673 } else if (inner->type == OFPACT_GROUP) {
3674 ctx->xin->flow.actset_output = OFPP_UNSET;
3675 ctx->action_set_has_group = true;
3680 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3681 ofpact_pad(&ctx->action_set);
3685 xlate_action_set(struct xlate_ctx *ctx)
3687 uint64_t action_list_stub[1024 / 64];
3688 struct ofpbuf action_list;
3690 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3691 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3692 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
3693 ofpbuf_uninit(&action_list);
3697 ofpact_needs_recirculation_after_mpls(const struct ofpact *a, struct xlate_ctx *ctx)
3699 struct flow_wildcards *wc = &ctx->xout->wc;
3700 struct flow *flow = &ctx->xin->flow;
3702 if (!ctx->was_mpls) {
3709 case OFPACT_CONTROLLER:
3710 case OFPACT_STRIP_VLAN:
3711 case OFPACT_SET_VLAN_PCP:
3712 case OFPACT_SET_VLAN_VID:
3713 case OFPACT_ENQUEUE:
3714 case OFPACT_PUSH_VLAN:
3715 case OFPACT_SET_ETH_SRC:
3716 case OFPACT_SET_ETH_DST:
3717 case OFPACT_SET_TUNNEL:
3718 case OFPACT_SET_QUEUE:
3719 case OFPACT_POP_QUEUE:
3721 case OFPACT_OUTPUT_REG:
3724 case OFPACT_WRITE_METADATA:
3725 case OFPACT_WRITE_ACTIONS:
3726 case OFPACT_CLEAR_ACTIONS:
3730 case OFPACT_POP_MPLS:
3731 case OFPACT_DEC_MPLS_TTL:
3732 case OFPACT_SET_MPLS_TTL:
3733 case OFPACT_SET_MPLS_TC:
3734 case OFPACT_SET_MPLS_LABEL:
3735 case OFPACT_SET_IPV4_SRC:
3736 case OFPACT_SET_IPV4_DST:
3737 case OFPACT_SET_IP_DSCP:
3738 case OFPACT_SET_IP_ECN:
3739 case OFPACT_SET_IP_TTL:
3740 case OFPACT_SET_L4_SRC_PORT:
3741 case OFPACT_SET_L4_DST_PORT:
3742 case OFPACT_RESUBMIT:
3743 case OFPACT_STACK_PUSH:
3744 case OFPACT_STACK_POP:
3745 case OFPACT_DEC_TTL:
3746 case OFPACT_MULTIPATH:
3749 case OFPACT_FIN_TIMEOUT:
3750 case OFPACT_GOTO_TABLE:
3753 case OFPACT_REG_MOVE:
3754 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
3755 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
3757 case OFPACT_SET_FIELD:
3758 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
3760 case OFPACT_PUSH_MPLS:
3761 /* Recirculate if it is an IP packet with a zero ttl. This may
3762 * indicate that the packet was previously MPLS and an MPLS pop action
3763 * converted it to IP. In this case recirculating should reveal the IP
3764 * TTL which is used as the basis for a new MPLS LSE. */
3765 return (!flow_count_mpls_labels(flow, wc)
3766 && flow->nw_ttl == 0
3767 && is_ip_any(flow));
3774 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
3775 struct xlate_ctx *ctx)
3777 struct flow_wildcards *wc = &ctx->xout->wc;
3778 struct flow *flow = &ctx->xin->flow;
3779 const struct ofpact *a;
3781 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3782 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
3784 /* dl_type already in the mask, not set below. */
3786 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3787 struct ofpact_controller *controller;
3788 const struct ofpact_metadata *metadata;
3789 const struct ofpact_set_field *set_field;
3790 const struct mf_field *mf;
3796 if (ofpact_needs_recirculation_after_mpls(a, ctx)) {
3797 compose_recirculate_action(ctx, ofpacts, a, ofpacts_len);
3803 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
3804 ofpact_get_OUTPUT(a)->max_len, true);
3808 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
3813 case OFPACT_CONTROLLER:
3814 controller = ofpact_get_CONTROLLER(a);
3815 execute_controller_action(ctx, controller->max_len,
3817 controller->controller_id);
3820 case OFPACT_ENQUEUE:
3821 memset(&wc->masks.skb_priority, 0xff,
3822 sizeof wc->masks.skb_priority);
3823 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
3826 case OFPACT_SET_VLAN_VID:
3827 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3828 if (flow->vlan_tci & htons(VLAN_CFI) ||
3829 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
3830 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
3831 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
3836 case OFPACT_SET_VLAN_PCP:
3837 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
3838 if (flow->vlan_tci & htons(VLAN_CFI) ||
3839 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
3840 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
3841 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
3842 << VLAN_PCP_SHIFT) | VLAN_CFI);
3846 case OFPACT_STRIP_VLAN:
3847 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3848 flow->vlan_tci = htons(0);
3851 case OFPACT_PUSH_VLAN:
3852 /* XXX 802.1AD(QinQ) */
3853 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3854 flow->vlan_tci = htons(VLAN_CFI);
3857 case OFPACT_SET_ETH_SRC:
3858 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3859 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
3862 case OFPACT_SET_ETH_DST:
3863 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3864 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
3867 case OFPACT_SET_IPV4_SRC:
3868 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3869 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3870 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
3874 case OFPACT_SET_IPV4_DST:
3875 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3876 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3877 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
3881 case OFPACT_SET_IP_DSCP:
3882 if (is_ip_any(flow)) {
3883 wc->masks.nw_tos |= IP_DSCP_MASK;
3884 flow->nw_tos &= ~IP_DSCP_MASK;
3885 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
3889 case OFPACT_SET_IP_ECN:
3890 if (is_ip_any(flow)) {
3891 wc->masks.nw_tos |= IP_ECN_MASK;
3892 flow->nw_tos &= ~IP_ECN_MASK;
3893 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
3897 case OFPACT_SET_IP_TTL:
3898 if (is_ip_any(flow)) {
3899 wc->masks.nw_ttl = 0xff;
3900 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
3904 case OFPACT_SET_L4_SRC_PORT:
3905 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3906 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3907 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3908 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
3912 case OFPACT_SET_L4_DST_PORT:
3913 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
3914 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3915 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3916 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
3920 case OFPACT_RESUBMIT:
3921 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
3924 case OFPACT_SET_TUNNEL:
3925 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
3928 case OFPACT_SET_QUEUE:
3929 memset(&wc->masks.skb_priority, 0xff,
3930 sizeof wc->masks.skb_priority);
3931 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
3934 case OFPACT_POP_QUEUE:
3935 memset(&wc->masks.skb_priority, 0xff,
3936 sizeof wc->masks.skb_priority);
3937 flow->skb_priority = ctx->orig_skb_priority;
3940 case OFPACT_REG_MOVE:
3941 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
3944 case OFPACT_SET_FIELD:
3945 set_field = ofpact_get_SET_FIELD(a);
3946 mf = set_field->field;
3948 /* Set field action only ever overwrites packet's outermost
3949 * applicable header fields. Do nothing if no header exists. */
3950 if (mf->id == MFF_VLAN_VID) {
3951 wc->masks.vlan_tci |= htons(VLAN_CFI);
3952 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
3955 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
3956 /* 'dl_type' is already unwildcarded. */
3957 && !eth_type_mpls(flow->dl_type)) {
3960 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
3961 * header field on a packet that does not have them. */
3962 mf_mask_field_and_prereqs(mf, &wc->masks);
3963 if (mf_are_prereqs_ok(mf, flow)) {
3964 mf_set_flow_value_masked(mf, &set_field->value,
3965 &set_field->mask, flow);
3969 case OFPACT_STACK_PUSH:
3970 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
3974 case OFPACT_STACK_POP:
3975 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
3979 case OFPACT_PUSH_MPLS:
3980 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
3983 case OFPACT_POP_MPLS:
3984 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
3987 case OFPACT_SET_MPLS_LABEL:
3988 compose_set_mpls_label_action(
3989 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
3992 case OFPACT_SET_MPLS_TC:
3993 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
3996 case OFPACT_SET_MPLS_TTL:
3997 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4000 case OFPACT_DEC_MPLS_TTL:
4001 if (compose_dec_mpls_ttl_action(ctx)) {
4006 case OFPACT_DEC_TTL:
4007 wc->masks.nw_ttl = 0xff;
4008 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4014 /* Nothing to do. */
4017 case OFPACT_MULTIPATH:
4018 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4022 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4025 case OFPACT_OUTPUT_REG:
4026 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4030 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4037 case OFPACT_FIN_TIMEOUT:
4038 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4039 ctx->xout->has_fin_timeout = true;
4040 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4043 case OFPACT_CLEAR_ACTIONS:
4044 ofpbuf_clear(&ctx->action_set);
4045 ctx->xin->flow.actset_output = OFPP_UNSET;
4046 ctx->action_set_has_group = false;
4049 case OFPACT_WRITE_ACTIONS:
4050 xlate_write_actions(ctx, a);
4053 case OFPACT_WRITE_METADATA:
4054 metadata = ofpact_get_WRITE_METADATA(a);
4055 flow->metadata &= ~metadata->mask;
4056 flow->metadata |= metadata->metadata & metadata->mask;
4060 /* Not implemented yet. */
4063 case OFPACT_GOTO_TABLE: {
4064 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4066 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4067 * than ogt->table_id. This is to allow goto_table actions that
4068 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4069 * after recirculation. */
4070 ovs_assert(ctx->table_id == TBL_INTERNAL
4071 || ctx->table_id < ogt->table_id);
4072 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4073 ogt->table_id, true, true);
4078 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4085 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4086 const struct flow *flow, ofp_port_t in_port,
4087 struct rule_dpif *rule, uint16_t tcp_flags,
4088 const struct ofpbuf *packet)
4090 xin->ofproto = ofproto;
4092 xin->flow.in_port.ofp_port = in_port;
4093 xin->flow.actset_output = OFPP_UNSET;
4094 xin->packet = packet;
4095 xin->may_learn = packet != NULL;
4098 xin->ofpacts = NULL;
4099 xin->ofpacts_len = 0;
4100 xin->tcp_flags = tcp_flags;
4101 xin->resubmit_hook = NULL;
4102 xin->report_hook = NULL;
4103 xin->resubmit_stats = NULL;
4104 xin->skip_wildcards = false;
4105 xin->odp_actions = NULL;
4109 xlate_out_uninit(struct xlate_out *xout)
4111 if (xout && xout->odp_actions == &xout->odp_actions_buf) {
4112 ofpbuf_uninit(xout->odp_actions);
4116 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4117 * into datapath actions, using 'ctx', and discards the datapath actions. */
4119 xlate_actions_for_side_effects(struct xlate_in *xin)
4121 struct xlate_out xout;
4123 xlate_actions(xin, &xout);
4124 xlate_out_uninit(&xout);
4128 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
4131 dst->slow = src->slow;
4132 dst->has_learn = src->has_learn;
4133 dst->has_normal = src->has_normal;
4134 dst->has_fin_timeout = src->has_fin_timeout;
4135 dst->nf_output_iface = src->nf_output_iface;
4136 dst->mirrors = src->mirrors;
4138 dst->odp_actions = &dst->odp_actions_buf;
4139 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
4140 sizeof dst->odp_actions_stub);
4141 ofpbuf_put(dst->odp_actions, ofpbuf_data(src->odp_actions),
4142 ofpbuf_size(src->odp_actions));
4145 static struct skb_priority_to_dscp *
4146 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4148 struct skb_priority_to_dscp *pdscp;
4151 hash = hash_int(skb_priority, 0);
4152 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4153 if (pdscp->skb_priority == skb_priority) {
4161 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4164 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4165 *dscp = pdscp ? pdscp->dscp : 0;
4166 return pdscp != NULL;
4170 count_skb_priorities(const struct xport *xport)
4172 return hmap_count(&xport->skb_priorities);
4176 clear_skb_priorities(struct xport *xport)
4178 struct skb_priority_to_dscp *pdscp, *next;
4180 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4181 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4187 actions_output_to_local_port(const struct xlate_ctx *ctx)
4189 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4190 const struct nlattr *a;
4193 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(ctx->xout->odp_actions),
4194 ofpbuf_size(ctx->xout->odp_actions)) {
4195 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4196 && nl_attr_get_odp_port(a) == local_odp_port) {
4203 #if defined(__linux__)
4204 /* Returns the maximum number of packets that the Linux kernel is willing to
4205 * queue up internally to certain kinds of software-implemented ports, or the
4206 * default (and rarely modified) value if it cannot be determined. */
4208 netdev_max_backlog(void)
4210 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4211 static int max_backlog = 1000; /* The normal default value. */
4213 if (ovsthread_once_start(&once)) {
4214 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4218 stream = fopen(filename, "r");
4220 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4222 if (fscanf(stream, "%d", &n) != 1) {
4223 VLOG_WARN("%s: read error", filename);
4224 } else if (n <= 100) {
4225 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4231 ovsthread_once_done(&once);
4233 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4239 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4242 count_output_actions(const struct ofpbuf *odp_actions)
4244 const struct nlattr *a;
4248 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(odp_actions),
4249 ofpbuf_size(odp_actions)) {
4250 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4256 #endif /* defined(__linux__) */
4258 /* Returns true if 'odp_actions' contains more output actions than the datapath
4259 * can reliably handle in one go. On Linux, this is the value of the
4260 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4261 * packets that the kernel is willing to queue up for processing while the
4262 * datapath is processing a set of actions. */
4264 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4267 return (ofpbuf_size(odp_actions) / NL_A_U32_SIZE > netdev_max_backlog()
4268 && count_output_actions(odp_actions) > netdev_max_backlog());
4270 /* OSes other than Linux might have similar limits, but we don't know how
4271 * to determine them.*/
4276 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
4277 * into datapath actions in 'odp_actions', using 'ctx'.
4279 * The caller must take responsibility for eventually freeing 'xout', with
4280 * xlate_out_uninit(). */
4282 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4284 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4285 struct flow_wildcards *wc = &xout->wc;
4286 struct flow *flow = &xin->flow;
4287 struct rule_dpif *rule = NULL;
4289 enum slow_path_reason special;
4290 const struct ofpact *ofpacts;
4291 struct xport *in_port;
4292 struct flow orig_flow;
4293 struct xlate_ctx ctx;
4298 COVERAGE_INC(xlate_actions);
4300 /* Flow initialization rules:
4301 * - 'base_flow' must match the kernel's view of the packet at the
4302 * time that action processing starts. 'flow' represents any
4303 * transformations we wish to make through actions.
4304 * - By default 'base_flow' and 'flow' are the same since the input
4305 * packet matches the output before any actions are applied.
4306 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4307 * of the received packet as seen by the kernel. If we later output
4308 * to another device without any modifications this will cause us to
4309 * insert a new tag since the original one was stripped off by the
4311 * - Tunnel metadata as received is retained in 'flow'. This allows
4312 * tunnel metadata matching also in later tables.
4313 * Since a kernel action for setting the tunnel metadata will only be
4314 * generated with actual tunnel output, changing the tunnel metadata
4315 * values in 'flow' (such as tun_id) will only have effect with a later
4316 * tunnel output action.
4317 * - Tunnel 'base_flow' is completely cleared since that is what the
4318 * kernel does. If we wish to maintain the original values an action
4319 * needs to be generated. */
4324 ctx.xout->has_learn = false;
4325 ctx.xout->has_normal = false;
4326 ctx.xout->has_fin_timeout = false;
4327 ctx.xout->nf_output_iface = NF_OUT_DROP;
4328 ctx.xout->mirrors = 0;
4330 xout->odp_actions = xin->odp_actions;
4331 if (!xout->odp_actions) {
4332 xout->odp_actions = &xout->odp_actions_buf;
4333 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4334 sizeof xout->odp_actions_stub);
4336 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4338 ctx.xbridge = xbridge_lookup(xcfg, xin->ofproto);
4343 ctx.rule = xin->rule;
4345 ctx.base_flow = *flow;
4346 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4347 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4349 flow_wildcards_init_catchall(wc);
4350 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4351 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4352 if (is_ip_any(flow)) {
4353 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4355 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4357 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4358 if (ctx.xbridge->netflow) {
4359 netflow_mask_wc(flow, wc);
4364 ctx.in_group = false;
4365 ctx.orig_skb_priority = flow->skb_priority;
4368 ctx.use_recirc = false;
4369 ctx.was_mpls = false;
4371 if (!xin->ofpacts && !ctx.rule) {
4372 rule = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
4373 xin->skip_wildcards ? NULL : wc,
4374 ctx.xin->xcache != NULL,
4375 ctx.xin->resubmit_stats, &ctx.table_id);
4376 if (ctx.xin->resubmit_stats) {
4377 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
4379 if (ctx.xin->xcache) {
4380 struct xc_entry *entry;
4382 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4383 entry->u.rule = rule;
4387 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
4388 ctx.xin->resubmit_hook(ctx.xin, rule, 0);
4391 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4394 ofpacts = xin->ofpacts;
4395 ofpacts_len = xin->ofpacts_len;
4396 } else if (ctx.rule) {
4397 const struct rule_actions *actions = rule_dpif_get_actions(ctx.rule);
4399 ofpacts = actions->ofpacts;
4400 ofpacts_len = actions->ofpacts_len;
4405 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4407 ctx.action_set_has_group = false;
4408 ofpbuf_use_stub(&ctx.action_set,
4409 ctx.action_set_stub, sizeof ctx.action_set_stub);
4411 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4412 /* Do this conditionally because the copy is expensive enough that it
4413 * shows up in profiles. */
4417 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
4418 if (in_port && in_port->is_tunnel) {
4419 if (ctx.xin->resubmit_stats) {
4420 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4422 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4425 if (ctx.xin->xcache) {
4426 struct xc_entry *entry;
4428 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4429 entry->u.dev.rx = netdev_ref(in_port->netdev);
4430 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4434 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
4436 ctx.xout->slow |= special;
4438 size_t sample_actions_len;
4440 if (flow->in_port.ofp_port
4441 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
4442 flow->in_port.ofp_port,
4444 ctx.base_flow.vlan_tci = 0;
4447 add_sflow_action(&ctx);
4448 add_ipfix_action(&ctx);
4449 sample_actions_len = ofpbuf_size(ctx.xout->odp_actions);
4451 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4452 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4454 /* We've let OFPP_NORMAL and the learning action look at the
4455 * packet, so drop it now if forwarding is disabled. */
4456 if (in_port && (!xport_stp_forward_state(in_port) ||
4457 !xport_rstp_forward_state(in_port))) {
4458 ofpbuf_set_size(ctx.xout->odp_actions, sample_actions_len);
4462 if (ofpbuf_size(&ctx.action_set)) {
4463 xlate_action_set(&ctx);
4466 if (ctx.xbridge->has_in_band
4467 && in_band_must_output_to_local_port(flow)
4468 && !actions_output_to_local_port(&ctx)) {
4469 compose_output_action(&ctx, OFPP_LOCAL);
4472 fix_sflow_action(&ctx);
4474 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4475 add_mirror_actions(&ctx, &orig_flow);
4479 if (nl_attr_oversized(ofpbuf_size(ctx.xout->odp_actions))) {
4480 /* These datapath actions are too big for a Netlink attribute, so we
4481 * can't hand them to the kernel directly. dpif_execute() can execute
4482 * them one by one with help, so just mark the result as SLOW_ACTION to
4483 * prevent the flow from being installed. */
4484 COVERAGE_INC(xlate_actions_oversize);
4485 ctx.xout->slow |= SLOW_ACTION;
4486 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
4487 COVERAGE_INC(xlate_actions_too_many_output);
4488 ctx.xout->slow |= SLOW_ACTION;
4491 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4492 if (ctx.xin->resubmit_stats) {
4493 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
4494 ctx.xin->resubmit_stats->n_packets,
4495 ctx.xin->resubmit_stats->n_bytes);
4497 if (ctx.xin->xcache) {
4498 struct xc_entry *entry;
4500 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
4501 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
4502 entry->u.mirror.mirrors = xout->mirrors;
4506 if (ctx.xbridge->netflow) {
4507 /* Only update netflow if we don't have controller flow. We don't
4508 * report NetFlow expiration messages for such facets because they
4509 * are just part of the control logic for the network, not real
4511 if (ofpacts_len == 0
4512 || ofpacts->type != OFPACT_CONTROLLER
4513 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
4514 if (ctx.xin->resubmit_stats) {
4515 netflow_flow_update(ctx.xbridge->netflow, flow,
4516 xout->nf_output_iface,
4517 ctx.xin->resubmit_stats);
4519 if (ctx.xin->xcache) {
4520 struct xc_entry *entry;
4522 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
4523 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
4524 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
4525 entry->u.nf.iface = xout->nf_output_iface;
4530 ofpbuf_uninit(&ctx.stack);
4531 ofpbuf_uninit(&ctx.action_set);
4533 /* Clear the metadata and register wildcard masks, because we won't
4534 * use non-header fields as part of the cache. */
4535 flow_wildcards_clear_non_packet_fields(wc);
4537 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
4538 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
4539 * these fields. The datapath interface, on the other hand, represents
4540 * them with just 8 bits each. This means that if the high 8 bits of the
4541 * masks for these fields somehow become set, then they will get chopped
4542 * off by a round trip through the datapath, and revalidation will spot
4543 * that as an inconsistency and delete the flow. Avoid the problem here by
4544 * making sure that only the low 8 bits of either field can be unwildcarded
4548 wc->masks.tp_src &= htons(UINT8_MAX);
4549 wc->masks.tp_dst &= htons(UINT8_MAX);
4553 /* Sends 'packet' out 'ofport'.
4554 * May modify 'packet'.
4555 * Returns 0 if successful, otherwise a positive errno value. */
4557 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4559 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4560 struct xport *xport;
4561 struct ofpact_output output;
4564 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4565 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4566 flow_extract(packet, NULL, &flow);
4567 flow.in_port.ofp_port = OFPP_NONE;
4569 xport = xport_lookup(xcfg, ofport);
4573 output.port = xport->ofp_port;
4576 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
4577 &output.ofpact, sizeof output,
4581 struct xlate_cache *
4582 xlate_cache_new(void)
4584 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
4586 ofpbuf_init(&xcache->entries, 512);
4590 static struct xc_entry *
4591 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
4593 struct xc_entry *entry;
4595 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
4602 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
4604 if (entry->u.dev.tx) {
4605 netdev_vport_inc_tx(entry->u.dev.tx, stats);
4607 if (entry->u.dev.rx) {
4608 netdev_vport_inc_rx(entry->u.dev.rx, stats);
4610 if (entry->u.dev.bfd) {
4611 bfd_account_rx(entry->u.dev.bfd, stats);
4616 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
4618 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4619 struct xbridge *xbridge;
4620 struct xbundle *xbundle;
4621 struct flow_wildcards wc;
4623 xbridge = xbridge_lookup(xcfg, ofproto);
4628 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
4634 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
4637 /* Push stats and perform side effects of flow translation. */
4639 xlate_push_stats(struct xlate_cache *xcache,
4640 const struct dpif_flow_stats *stats)
4642 struct xc_entry *entry;
4643 struct ofpbuf entries = xcache->entries;
4644 uint8_t dmac[ETH_ADDR_LEN];
4646 if (!stats->n_packets) {
4650 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4651 switch (entry->type) {
4653 rule_dpif_credit_stats(entry->u.rule, stats);
4656 bond_account(entry->u.bond.bond, entry->u.bond.flow,
4657 entry->u.bond.vid, stats->n_bytes);
4660 xlate_cache_netdev(entry, stats);
4663 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
4664 entry->u.nf.iface, stats);
4667 mirror_update_stats(entry->u.mirror.mbridge,
4668 entry->u.mirror.mirrors,
4669 stats->n_packets, stats->n_bytes);
4672 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
4675 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
4676 entry->u.normal.vlan);
4678 case XC_FIN_TIMEOUT:
4679 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
4680 entry->u.fin.idle, entry->u.fin.hard);
4683 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
4687 /* Lookup arp to avoid arp timeout. */
4688 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
4697 xlate_dev_unref(struct xc_entry *entry)
4699 if (entry->u.dev.tx) {
4700 netdev_close(entry->u.dev.tx);
4702 if (entry->u.dev.rx) {
4703 netdev_close(entry->u.dev.rx);
4705 if (entry->u.dev.bfd) {
4706 bfd_unref(entry->u.dev.bfd);
4711 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
4713 netflow_flow_clear(netflow, flow);
4714 netflow_unref(netflow);
4719 xlate_cache_clear(struct xlate_cache *xcache)
4721 struct xc_entry *entry;
4722 struct ofpbuf entries;
4728 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4729 switch (entry->type) {
4731 rule_dpif_unref(entry->u.rule);
4734 free(entry->u.bond.flow);
4735 bond_unref(entry->u.bond.bond);
4738 xlate_dev_unref(entry);
4741 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
4744 mbridge_unref(entry->u.mirror.mbridge);
4747 free(entry->u.learn.fm);
4748 ofpbuf_delete(entry->u.learn.ofpacts);
4751 free(entry->u.normal.flow);
4753 case XC_FIN_TIMEOUT:
4754 /* 'u.fin.rule' is always already held as a XC_RULE, which
4755 * has already released it's reference above. */
4758 group_dpif_unref(entry->u.group.group);
4767 ofpbuf_clear(&xcache->entries);
4771 xlate_cache_delete(struct xlate_cache *xcache)
4773 xlate_cache_clear(xcache);
4774 ofpbuf_uninit(&xcache->entries);