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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
35 #include "mac-learning.h"
36 #include "mcast-snooping.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
42 #include "odp-execute.h"
43 #include "ofp-actions.h"
44 #include "ofproto/ofproto-dpif-ipfix.h"
45 #include "ofproto/ofproto-dpif-mirror.h"
46 #include "ofproto/ofproto-dpif-monitor.h"
47 #include "ofproto/ofproto-dpif-sflow.h"
48 #include "ofproto/ofproto-dpif.h"
49 #include "ofproto/ofproto-provider.h"
50 #include "packet-dpif.h"
54 COVERAGE_DEFINE(xlate_actions);
55 COVERAGE_DEFINE(xlate_actions_oversize);
56 COVERAGE_DEFINE(xlate_actions_too_many_output);
57 COVERAGE_DEFINE(xlate_actions_mpls_overflow);
59 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
61 /* Maximum depth of flow table recursion (due to resubmit actions) in a
62 * flow translation. */
63 #define MAX_RESUBMIT_RECURSION 64
64 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
67 /* Timeout for internal rules created to handle recirculation */
68 #define RECIRC_TIMEOUT 60
70 /* Maximum number of resubmit actions in a flow translation, whether they are
71 * recursive or not. */
72 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
75 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
76 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
78 struct list xbundles; /* Owned xbundles. */
79 struct hmap xports; /* Indexed by ofp_port. */
81 char *name; /* Name used in log messages. */
82 struct dpif *dpif; /* Datapath interface. */
83 struct mac_learning *ml; /* Mac learning handle. */
84 struct mcast_snooping *ms; /* Multicast Snooping handle. */
85 struct mbridge *mbridge; /* Mirroring. */
86 struct dpif_sflow *sflow; /* SFlow handle, or null. */
87 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
88 struct netflow *netflow; /* Netflow handle, or null. */
89 struct stp *stp; /* STP or null if disabled. */
90 struct rstp *rstp; /* RSTP or null if disabled. */
92 /* Special rules installed by ofproto-dpif. */
93 struct rule_dpif *miss_rule;
94 struct rule_dpif *no_packet_in_rule;
96 enum ofp_config_flags frag; /* Fragmentation handling. */
97 bool has_in_band; /* Bridge has in band control? */
98 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
100 /* True if the datapath supports recirculation. */
103 /* True if the datapath supports variable-length
104 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
105 * False if the datapath supports only 8-byte (or shorter) userdata. */
106 bool variable_length_userdata;
108 /* Number of MPLS label stack entries that the datapath supports
110 size_t max_mpls_depth;
112 /* True if the datapath supports masked data in OVS_ACTION_ATTR_SET
114 bool masked_set_action;
118 struct hmap_node hmap_node; /* In global 'xbundles' map. */
119 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
121 struct list list_node; /* In parent 'xbridges' list. */
122 struct xbridge *xbridge; /* Parent xbridge. */
124 struct list xports; /* Contains "struct xport"s. */
126 char *name; /* Name used in log messages. */
127 struct bond *bond; /* Nonnull iff more than one port. */
128 struct lacp *lacp; /* LACP handle or null. */
130 enum port_vlan_mode vlan_mode; /* VLAN mode. */
131 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
132 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
133 * NULL if all VLANs are trunked. */
134 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
135 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
139 struct hmap_node hmap_node; /* Node in global 'xports' map. */
140 struct ofport_dpif *ofport; /* Key in global 'xports map. */
142 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
143 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
145 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
147 struct list bundle_node; /* In parent xbundle (if it exists). */
148 struct xbundle *xbundle; /* Parent xbundle or null. */
150 struct netdev *netdev; /* 'ofport''s netdev. */
152 struct xbridge *xbridge; /* Parent bridge. */
153 struct xport *peer; /* Patch port peer or null. */
155 enum ofputil_port_config config; /* OpenFlow port configuration. */
156 enum ofputil_port_state state; /* OpenFlow port state. */
157 int stp_port_no; /* STP port number or -1 if not in use. */
158 struct rstp_port *rstp_port; /* RSTP port or null. */
160 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
162 bool may_enable; /* May be enabled in bonds. */
163 bool is_tunnel; /* Is a tunnel port. */
165 struct cfm *cfm; /* CFM handle or null. */
166 struct bfd *bfd; /* BFD handle or null. */
170 struct xlate_in *xin;
171 struct xlate_out *xout;
173 const struct xbridge *xbridge;
175 /* Flow at the last commit. */
176 struct flow base_flow;
178 /* Tunnel IP destination address as received. This is stored separately
179 * as the base_flow.tunnel is cleared on init to reflect the datapath
180 * behavior. Used to make sure not to send tunneled output to ourselves,
181 * which might lead to an infinite loop. This could happen easily
182 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
183 * actually set the tun_dst field. */
184 ovs_be32 orig_tunnel_ip_dst;
186 /* Stack for the push and pop actions. Each stack element is of type
187 * "union mf_subvalue". */
188 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
191 /* The rule that we are currently translating, or NULL. */
192 struct rule_dpif *rule;
194 /* Resubmit statistics, via xlate_table_action(). */
195 int recurse; /* Current resubmit nesting depth. */
196 int resubmits; /* Total number of resubmits. */
197 bool in_group; /* Currently translating ofgroup, if true. */
199 uint32_t orig_skb_priority; /* Priority when packet arrived. */
200 uint8_t table_id; /* OpenFlow table ID where flow was found. */
201 uint32_t sflow_n_outputs; /* Number of output ports. */
202 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
203 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
204 bool exit; /* No further actions should be processed. */
206 bool use_recirc; /* Should generate recirc? */
207 struct xlate_recirc recirc; /* Information used for generating
208 * recirculation actions */
210 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
211 * This is a trigger for recirculation in cases where translating an action
212 * or looking up a flow requires access to the fields of the packet after
213 * the MPLS label stack that was originally present. */
216 /* OpenFlow 1.1+ action set.
218 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
219 * When translation is otherwise complete, ofpacts_execute_action_set()
220 * converts it to a set of "struct ofpact"s that can be translated into
221 * datapath actions. */
222 struct ofpbuf action_set; /* Action set. */
223 uint64_t action_set_stub[1024 / 8];
226 /* A controller may use OFPP_NONE as the ingress port to indicate that
227 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
228 * when an input bundle is needed for validation (e.g., mirroring or
229 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
230 * any 'port' structs, so care must be taken when dealing with it. */
231 static struct xbundle ofpp_none_bundle = {
233 .vlan_mode = PORT_VLAN_TRUNK
236 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
237 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
238 * traffic egressing the 'ofport' with that priority should be marked with. */
239 struct skb_priority_to_dscp {
240 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
241 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
243 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
258 /* xlate_cache entries hold enough information to perform the side effects of
259 * xlate_actions() for a rule, without needing to perform rule translation
260 * from scratch. The primary usage of these is to submit statistics to objects
261 * that a flow relates to, although they may be used for other effects as well
262 * (for instance, refreshing hard timeouts for learned flows). */
266 struct rule_dpif *rule;
273 struct netflow *netflow;
278 struct mbridge *mbridge;
279 mirror_mask_t mirrors;
287 struct ofproto_dpif *ofproto;
288 struct ofputil_flow_mod *fm;
289 struct ofpbuf *ofpacts;
292 struct ofproto_dpif *ofproto;
297 struct rule_dpif *rule;
302 struct group_dpif *group;
303 struct ofputil_bucket *bucket;
308 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
309 entries = xcache->entries; \
310 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
312 entry = ofpbuf_try_pull(&entries, sizeof *entry))
315 struct ofpbuf entries;
318 /* Xlate config contains hash maps of all bridges, bundles and ports.
319 * Xcfgp contains the pointer to the current xlate configuration.
320 * When the main thread needs to change the configuration, it copies xcfgp to
321 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
322 * does not block handler and revalidator threads. */
324 struct hmap xbridges;
325 struct hmap xbundles;
328 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_TYPE_INITIALIZER;
329 static struct xlate_cfg *new_xcfg = NULL;
331 static bool may_receive(const struct xport *, struct xlate_ctx *);
332 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
334 static void xlate_normal(struct xlate_ctx *);
335 static void xlate_report(struct xlate_ctx *, const char *);
336 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
337 uint8_t table_id, bool may_packet_in,
338 bool honor_table_miss);
339 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
340 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
341 static void output_normal(struct xlate_ctx *, const struct xbundle *,
343 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
345 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
346 const struct ofproto_dpif *);
347 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
348 const struct ofbundle *);
349 static struct xport *xport_lookup(struct xlate_cfg *,
350 const struct ofport_dpif *);
351 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
352 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
353 uint32_t skb_priority);
354 static void clear_skb_priorities(struct xport *);
355 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
358 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
360 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
361 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
362 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
363 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
364 struct rule_dpif *miss_rule,
365 struct rule_dpif *no_packet_in_rule,
366 const struct mac_learning *, struct stp *,
367 struct rstp *, const struct mcast_snooping *,
368 const struct mbridge *,
369 const struct dpif_sflow *,
370 const struct dpif_ipfix *,
371 const struct netflow *, enum ofp_config_flags,
372 bool forward_bpdu, bool has_in_band,
374 bool variable_length_userdata,
375 size_t max_mpls_depth,
376 bool masked_set_action);
377 static void xlate_xbundle_set(struct xbundle *xbundle,
378 enum port_vlan_mode vlan_mode, int vlan,
379 unsigned long *trunks, bool use_priority_tags,
380 const struct bond *bond, const struct lacp *lacp,
382 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
383 const struct netdev *netdev, const struct cfm *cfm,
384 const struct bfd *bfd, int stp_port_no,
385 const struct rstp_port *rstp_port,
386 enum ofputil_port_config config,
387 enum ofputil_port_state state, bool is_tunnel,
389 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
390 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
391 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
392 static void xlate_xbridge_copy(struct xbridge *);
393 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
394 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
396 static void xlate_xcfg_free(struct xlate_cfg *);
400 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
402 list_init(&xbridge->xbundles);
403 hmap_init(&xbridge->xports);
404 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
405 hash_pointer(xbridge->ofproto, 0));
409 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
411 list_init(&xbundle->xports);
412 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
413 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
414 hash_pointer(xbundle->ofbundle, 0));
418 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
420 hmap_init(&xport->skb_priorities);
421 hmap_insert(&xcfg->xports, &xport->hmap_node,
422 hash_pointer(xport->ofport, 0));
423 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
424 hash_ofp_port(xport->ofp_port));
428 xlate_xbridge_set(struct xbridge *xbridge,
430 struct rule_dpif *miss_rule,
431 struct rule_dpif *no_packet_in_rule,
432 const struct mac_learning *ml, struct stp *stp,
433 struct rstp *rstp, const struct mcast_snooping *ms,
434 const struct mbridge *mbridge,
435 const struct dpif_sflow *sflow,
436 const struct dpif_ipfix *ipfix,
437 const struct netflow *netflow, enum ofp_config_flags frag,
438 bool forward_bpdu, bool has_in_band,
440 bool variable_length_userdata,
441 size_t max_mpls_depth,
442 bool masked_set_action)
444 if (xbridge->ml != ml) {
445 mac_learning_unref(xbridge->ml);
446 xbridge->ml = mac_learning_ref(ml);
449 if (xbridge->ms != ms) {
450 mcast_snooping_unref(xbridge->ms);
451 xbridge->ms = mcast_snooping_ref(ms);
454 if (xbridge->mbridge != mbridge) {
455 mbridge_unref(xbridge->mbridge);
456 xbridge->mbridge = mbridge_ref(mbridge);
459 if (xbridge->sflow != sflow) {
460 dpif_sflow_unref(xbridge->sflow);
461 xbridge->sflow = dpif_sflow_ref(sflow);
464 if (xbridge->ipfix != ipfix) {
465 dpif_ipfix_unref(xbridge->ipfix);
466 xbridge->ipfix = dpif_ipfix_ref(ipfix);
469 if (xbridge->stp != stp) {
470 stp_unref(xbridge->stp);
471 xbridge->stp = stp_ref(stp);
474 if (xbridge->rstp != rstp) {
475 rstp_unref(xbridge->rstp);
476 xbridge->rstp = rstp_ref(rstp);
479 if (xbridge->netflow != netflow) {
480 netflow_unref(xbridge->netflow);
481 xbridge->netflow = netflow_ref(netflow);
484 xbridge->dpif = dpif;
485 xbridge->forward_bpdu = forward_bpdu;
486 xbridge->has_in_band = has_in_band;
487 xbridge->frag = frag;
488 xbridge->miss_rule = miss_rule;
489 xbridge->no_packet_in_rule = no_packet_in_rule;
490 xbridge->enable_recirc = enable_recirc;
491 xbridge->variable_length_userdata = variable_length_userdata;
492 xbridge->max_mpls_depth = max_mpls_depth;
493 xbridge->masked_set_action = masked_set_action;
497 xlate_xbundle_set(struct xbundle *xbundle,
498 enum port_vlan_mode vlan_mode, int vlan,
499 unsigned long *trunks, bool use_priority_tags,
500 const struct bond *bond, const struct lacp *lacp,
503 ovs_assert(xbundle->xbridge);
505 xbundle->vlan_mode = vlan_mode;
506 xbundle->vlan = vlan;
507 xbundle->trunks = trunks;
508 xbundle->use_priority_tags = use_priority_tags;
509 xbundle->floodable = floodable;
511 if (xbundle->bond != bond) {
512 bond_unref(xbundle->bond);
513 xbundle->bond = bond_ref(bond);
516 if (xbundle->lacp != lacp) {
517 lacp_unref(xbundle->lacp);
518 xbundle->lacp = lacp_ref(lacp);
523 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
524 const struct netdev *netdev, const struct cfm *cfm,
525 const struct bfd *bfd, int stp_port_no,
526 const struct rstp_port* rstp_port,
527 enum ofputil_port_config config, enum ofputil_port_state state,
528 bool is_tunnel, bool may_enable)
530 xport->config = config;
531 xport->state = state;
532 xport->stp_port_no = stp_port_no;
533 xport->is_tunnel = is_tunnel;
534 xport->may_enable = may_enable;
535 xport->odp_port = odp_port;
537 if (xport->rstp_port != rstp_port) {
538 rstp_port_unref(xport->rstp_port);
539 xport->rstp_port = rstp_port_ref(rstp_port);
542 if (xport->cfm != cfm) {
543 cfm_unref(xport->cfm);
544 xport->cfm = cfm_ref(cfm);
547 if (xport->bfd != bfd) {
548 bfd_unref(xport->bfd);
549 xport->bfd = bfd_ref(bfd);
552 if (xport->netdev != netdev) {
553 netdev_close(xport->netdev);
554 xport->netdev = netdev_ref(netdev);
559 xlate_xbridge_copy(struct xbridge *xbridge)
561 struct xbundle *xbundle;
563 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
564 new_xbridge->ofproto = xbridge->ofproto;
565 new_xbridge->name = xstrdup(xbridge->name);
566 xlate_xbridge_init(new_xcfg, new_xbridge);
568 xlate_xbridge_set(new_xbridge,
569 xbridge->dpif, xbridge->miss_rule,
570 xbridge->no_packet_in_rule, xbridge->ml, xbridge->stp,
571 xbridge->rstp, xbridge->ms, xbridge->mbridge,
572 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
573 xbridge->frag, xbridge->forward_bpdu,
574 xbridge->has_in_band, xbridge->enable_recirc,
575 xbridge->variable_length_userdata,
576 xbridge->max_mpls_depth, xbridge->masked_set_action);
577 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
578 xlate_xbundle_copy(new_xbridge, xbundle);
581 /* Copy xports which are not part of a xbundle */
582 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
583 if (!xport->xbundle) {
584 xlate_xport_copy(new_xbridge, NULL, xport);
590 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
593 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
594 new_xbundle->ofbundle = xbundle->ofbundle;
595 new_xbundle->xbridge = xbridge;
596 new_xbundle->name = xstrdup(xbundle->name);
597 xlate_xbundle_init(new_xcfg, new_xbundle);
599 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
600 xbundle->vlan, xbundle->trunks,
601 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
603 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
604 xlate_xport_copy(xbridge, new_xbundle, xport);
609 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
612 struct skb_priority_to_dscp *pdscp, *new_pdscp;
613 struct xport *new_xport = xzalloc(sizeof *xport);
614 new_xport->ofport = xport->ofport;
615 new_xport->ofp_port = xport->ofp_port;
616 new_xport->xbridge = xbridge;
617 xlate_xport_init(new_xcfg, new_xport);
619 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
620 xport->bfd, xport->stp_port_no, xport->rstp_port,
621 xport->config, xport->state, xport->is_tunnel,
625 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
627 new_xport->peer = peer;
628 new_xport->peer->peer = new_xport;
633 new_xport->xbundle = xbundle;
634 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
637 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
638 new_pdscp = xmalloc(sizeof *pdscp);
639 new_pdscp->skb_priority = pdscp->skb_priority;
640 new_pdscp->dscp = pdscp->dscp;
641 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
642 hash_int(new_pdscp->skb_priority, 0));
646 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
647 * configuration in xcfgp.
649 * This needs to be called after editing the xlate configuration.
651 * Functions that edit the new xlate configuration are
652 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
658 * edit_xlate_configuration();
660 * xlate_txn_commit(); */
662 xlate_txn_commit(void)
664 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
666 ovsrcu_set(&xcfgp, new_xcfg);
667 ovsrcu_postpone(xlate_xcfg_free, xcfg);
672 /* Copies the current xlate configuration in xcfgp to new_xcfg.
674 * This needs to be called prior to editing the xlate configuration. */
676 xlate_txn_start(void)
678 struct xbridge *xbridge;
679 struct xlate_cfg *xcfg;
681 ovs_assert(!new_xcfg);
683 new_xcfg = xmalloc(sizeof *new_xcfg);
684 hmap_init(&new_xcfg->xbridges);
685 hmap_init(&new_xcfg->xbundles);
686 hmap_init(&new_xcfg->xports);
688 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
693 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
694 xlate_xbridge_copy(xbridge);
700 xlate_xcfg_free(struct xlate_cfg *xcfg)
702 struct xbridge *xbridge, *next_xbridge;
708 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
709 xlate_xbridge_remove(xcfg, xbridge);
712 hmap_destroy(&xcfg->xbridges);
713 hmap_destroy(&xcfg->xbundles);
714 hmap_destroy(&xcfg->xports);
719 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
720 struct dpif *dpif, struct rule_dpif *miss_rule,
721 struct rule_dpif *no_packet_in_rule,
722 const struct mac_learning *ml, struct stp *stp,
723 struct rstp *rstp, const struct mcast_snooping *ms,
724 const struct mbridge *mbridge,
725 const struct dpif_sflow *sflow,
726 const struct dpif_ipfix *ipfix,
727 const struct netflow *netflow, enum ofp_config_flags frag,
728 bool forward_bpdu, bool has_in_band, bool enable_recirc,
729 bool variable_length_userdata, size_t max_mpls_depth,
730 bool masked_set_action)
732 struct xbridge *xbridge;
734 ovs_assert(new_xcfg);
736 xbridge = xbridge_lookup(new_xcfg, ofproto);
738 xbridge = xzalloc(sizeof *xbridge);
739 xbridge->ofproto = ofproto;
741 xlate_xbridge_init(new_xcfg, xbridge);
745 xbridge->name = xstrdup(name);
747 xlate_xbridge_set(xbridge, dpif, miss_rule, no_packet_in_rule, ml, stp,
748 rstp, ms, mbridge, sflow, ipfix, netflow, frag,
749 forward_bpdu, has_in_band, enable_recirc,
750 variable_length_userdata, max_mpls_depth,
755 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
757 struct xbundle *xbundle, *next_xbundle;
758 struct xport *xport, *next_xport;
764 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
765 xlate_xport_remove(xcfg, xport);
768 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
769 xlate_xbundle_remove(xcfg, xbundle);
772 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
773 mac_learning_unref(xbridge->ml);
774 mcast_snooping_unref(xbridge->ms);
775 mbridge_unref(xbridge->mbridge);
776 dpif_sflow_unref(xbridge->sflow);
777 dpif_ipfix_unref(xbridge->ipfix);
778 stp_unref(xbridge->stp);
779 rstp_unref(xbridge->rstp);
780 hmap_destroy(&xbridge->xports);
786 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
788 struct xbridge *xbridge;
790 ovs_assert(new_xcfg);
792 xbridge = xbridge_lookup(new_xcfg, ofproto);
793 xlate_xbridge_remove(new_xcfg, xbridge);
797 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
798 const char *name, enum port_vlan_mode vlan_mode, int vlan,
799 unsigned long *trunks, bool use_priority_tags,
800 const struct bond *bond, const struct lacp *lacp,
803 struct xbundle *xbundle;
805 ovs_assert(new_xcfg);
807 xbundle = xbundle_lookup(new_xcfg, ofbundle);
809 xbundle = xzalloc(sizeof *xbundle);
810 xbundle->ofbundle = ofbundle;
811 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
813 xlate_xbundle_init(new_xcfg, xbundle);
817 xbundle->name = xstrdup(name);
819 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
820 use_priority_tags, bond, lacp, floodable);
824 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
826 struct xport *xport, *next;
832 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
833 list_remove(&xport->bundle_node);
834 xport->xbundle = NULL;
837 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
838 list_remove(&xbundle->list_node);
839 bond_unref(xbundle->bond);
840 lacp_unref(xbundle->lacp);
846 xlate_bundle_remove(struct ofbundle *ofbundle)
848 struct xbundle *xbundle;
850 ovs_assert(new_xcfg);
852 xbundle = xbundle_lookup(new_xcfg, ofbundle);
853 xlate_xbundle_remove(new_xcfg, xbundle);
857 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
858 struct ofport_dpif *ofport, ofp_port_t ofp_port,
859 odp_port_t odp_port, const struct netdev *netdev,
860 const struct cfm *cfm, const struct bfd *bfd,
861 struct ofport_dpif *peer, int stp_port_no,
862 const struct rstp_port *rstp_port,
863 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
864 enum ofputil_port_config config,
865 enum ofputil_port_state state, bool is_tunnel,
871 ovs_assert(new_xcfg);
873 xport = xport_lookup(new_xcfg, ofport);
875 xport = xzalloc(sizeof *xport);
876 xport->ofport = ofport;
877 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
878 xport->ofp_port = ofp_port;
880 xlate_xport_init(new_xcfg, xport);
883 ovs_assert(xport->ofp_port == ofp_port);
885 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, stp_port_no,
886 rstp_port, config, state, is_tunnel, may_enable);
889 xport->peer->peer = NULL;
891 xport->peer = xport_lookup(new_xcfg, peer);
893 xport->peer->peer = xport;
896 if (xport->xbundle) {
897 list_remove(&xport->bundle_node);
899 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
900 if (xport->xbundle) {
901 list_insert(&xport->xbundle->xports, &xport->bundle_node);
904 clear_skb_priorities(xport);
905 for (i = 0; i < n_qdscp; i++) {
906 struct skb_priority_to_dscp *pdscp;
907 uint32_t skb_priority;
909 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
914 pdscp = xmalloc(sizeof *pdscp);
915 pdscp->skb_priority = skb_priority;
916 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
917 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
918 hash_int(pdscp->skb_priority, 0));
923 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
930 xport->peer->peer = NULL;
934 if (xport->xbundle) {
935 list_remove(&xport->bundle_node);
938 clear_skb_priorities(xport);
939 hmap_destroy(&xport->skb_priorities);
941 hmap_remove(&xcfg->xports, &xport->hmap_node);
942 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
944 netdev_close(xport->netdev);
945 rstp_port_unref(xport->rstp_port);
946 cfm_unref(xport->cfm);
947 bfd_unref(xport->bfd);
952 xlate_ofport_remove(struct ofport_dpif *ofport)
956 ovs_assert(new_xcfg);
958 xport = xport_lookup(new_xcfg, ofport);
959 xlate_xport_remove(new_xcfg, xport);
962 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
963 * returns the corresponding struct xport, or NULL if none is found. */
964 static struct xport *
965 xlate_lookup_xport(const struct dpif_backer *backer, const struct flow *flow)
967 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
969 return xport_lookup(xcfg, tnl_port_should_receive(flow)
970 ? tnl_port_receive(flow)
971 : odp_port_to_ofport(backer, flow->in_port.odp_port));
974 static struct ofproto_dpif *
975 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
976 ofp_port_t *ofp_in_port, const struct xport **xportp)
978 const struct xport *xport;
980 *xportp = xport = xlate_lookup_xport(backer, flow);
984 *ofp_in_port = xport->ofp_port;
986 return xport->xbridge->ofproto;
992 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
993 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
994 struct ofproto_dpif *
995 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
996 ofp_port_t *ofp_in_port)
998 const struct xport *xport;
1000 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1003 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1004 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1005 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1006 * handles for those protocols if they're enabled. Caller may use the returned
1007 * pointers until quiescing, for longer term use additional references must
1010 * '*ofp_in_port' is set to OFPP_NONE if 'flow''s in_port does not exist.
1012 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport.
1015 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1016 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1017 struct dpif_sflow **sflow, struct netflow **netflow,
1018 ofp_port_t *ofp_in_port)
1020 struct ofproto_dpif *ofproto;
1021 const struct xport *xport;
1023 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1025 if (ofp_in_port && !xport) {
1026 *ofp_in_port = OFPP_NONE;
1034 *ofprotop = ofproto;
1038 *ipfix = xport->xbridge->ipfix;
1042 *sflow = xport->xbridge->sflow;
1046 *netflow = xport->xbridge->netflow;
1051 static struct xbridge *
1052 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1054 struct hmap *xbridges;
1055 struct xbridge *xbridge;
1057 if (!ofproto || !xcfg) {
1061 xbridges = &xcfg->xbridges;
1063 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1065 if (xbridge->ofproto == ofproto) {
1072 static struct xbundle *
1073 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1075 struct hmap *xbundles;
1076 struct xbundle *xbundle;
1078 if (!ofbundle || !xcfg) {
1082 xbundles = &xcfg->xbundles;
1084 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1086 if (xbundle->ofbundle == ofbundle) {
1093 static struct xport *
1094 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1096 struct hmap *xports;
1097 struct xport *xport;
1099 if (!ofport || !xcfg) {
1103 xports = &xcfg->xports;
1105 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1107 if (xport->ofport == ofport) {
1114 static struct stp_port *
1115 xport_get_stp_port(const struct xport *xport)
1117 return xport->xbridge->stp && xport->stp_port_no != -1
1118 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1123 xport_stp_learn_state(const struct xport *xport)
1125 struct stp_port *sp = xport_get_stp_port(xport);
1126 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
1130 xport_stp_forward_state(const struct xport *xport)
1132 struct stp_port *sp = xport_get_stp_port(xport);
1133 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
1137 xport_stp_should_forward_bpdu(const struct xport *xport)
1139 struct stp_port *sp = xport_get_stp_port(xport);
1140 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1143 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1144 * were used to make the determination.*/
1146 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1148 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1149 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1150 return is_stp(flow);
1154 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1156 struct stp_port *sp = xport_get_stp_port(xport);
1157 struct ofpbuf payload = *packet;
1158 struct eth_header *eth = ofpbuf_data(&payload);
1160 /* Sink packets on ports that have STP disabled when the bridge has
1162 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1166 /* Trim off padding on payload. */
1167 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1168 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1171 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1172 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
1176 static enum rstp_state
1177 xport_get_rstp_port_state(const struct xport *xport)
1179 return xport->rstp_port
1180 ? rstp_port_get_state(xport->rstp_port)
1185 xport_rstp_learn_state(const struct xport *xport)
1187 return rstp_learn_in_state(xport_get_rstp_port_state(xport));
1191 xport_rstp_forward_state(const struct xport *xport)
1193 return rstp_forward_in_state(xport_get_rstp_port_state(xport));
1197 xport_rstp_should_manage_bpdu(const struct xport *xport)
1199 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1203 rstp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1205 struct ofpbuf payload = *packet;
1206 struct eth_header *eth = ofpbuf_data(&payload);
1208 /* Sink packets on ports that have no RSTP. */
1209 if (!xport->rstp_port) {
1213 /* Trim off padding on payload. */
1214 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1215 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1218 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1219 rstp_port_received_bpdu(xport->rstp_port, ofpbuf_data(&payload),
1220 ofpbuf_size(&payload));
1224 static struct xport *
1225 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1227 struct xport *xport;
1229 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1231 if (xport->ofp_port == ofp_port) {
1239 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1241 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1242 return xport ? xport->odp_port : ODPP_NONE;
1246 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1248 struct xport *xport;
1250 xport = get_ofp_port(ctx->xbridge, ofp_port);
1251 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
1252 xport->state & OFPUTIL_PS_LINK_DOWN) {
1259 static struct ofputil_bucket *
1260 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1264 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1266 struct group_dpif *group;
1268 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1269 struct ofputil_bucket *bucket;
1271 bucket = group_first_live_bucket(ctx, group, depth);
1272 group_dpif_unref(group);
1273 return bucket == NULL;
1279 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1282 bucket_is_alive(const struct xlate_ctx *ctx,
1283 struct ofputil_bucket *bucket, int depth)
1285 if (depth >= MAX_LIVENESS_RECURSION) {
1286 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1288 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1289 MAX_LIVENESS_RECURSION);
1293 return (!ofputil_bucket_has_liveness(bucket)
1294 || (bucket->watch_port != OFPP_ANY
1295 && odp_port_is_alive(ctx, bucket->watch_port))
1296 || (bucket->watch_group != OFPG_ANY
1297 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1300 static struct ofputil_bucket *
1301 group_first_live_bucket(const struct xlate_ctx *ctx,
1302 const struct group_dpif *group, int depth)
1304 struct ofputil_bucket *bucket;
1305 const struct list *buckets;
1307 group_dpif_get_buckets(group, &buckets);
1308 LIST_FOR_EACH (bucket, list_node, buckets) {
1309 if (bucket_is_alive(ctx, bucket, depth)) {
1317 static struct ofputil_bucket *
1318 group_best_live_bucket(const struct xlate_ctx *ctx,
1319 const struct group_dpif *group,
1322 struct ofputil_bucket *best_bucket = NULL;
1323 uint32_t best_score = 0;
1326 struct ofputil_bucket *bucket;
1327 const struct list *buckets;
1329 group_dpif_get_buckets(group, &buckets);
1330 LIST_FOR_EACH (bucket, list_node, buckets) {
1331 if (bucket_is_alive(ctx, bucket, 0)) {
1332 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1333 if (score >= best_score) {
1334 best_bucket = bucket;
1345 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1347 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1348 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1352 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1354 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1357 static mirror_mask_t
1358 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1360 return xbundle != &ofpp_none_bundle
1361 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1365 static mirror_mask_t
1366 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1368 return xbundle != &ofpp_none_bundle
1369 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1373 static mirror_mask_t
1374 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1376 return xbundle != &ofpp_none_bundle
1377 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1381 static struct xbundle *
1382 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1383 bool warn, struct xport **in_xportp)
1385 struct xport *xport;
1387 /* Find the port and bundle for the received packet. */
1388 xport = get_ofp_port(xbridge, in_port);
1392 if (xport && xport->xbundle) {
1393 return xport->xbundle;
1396 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1397 * which a controller may use as the ingress port for traffic that
1398 * it is sourcing. */
1399 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1400 return &ofpp_none_bundle;
1403 /* Odd. A few possible reasons here:
1405 * - We deleted a port but there are still a few packets queued up
1408 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1409 * we don't know about.
1411 * - The ofproto client didn't configure the port as part of a bundle.
1412 * This is particularly likely to happen if a packet was received on the
1413 * port after it was created, but before the client had a chance to
1414 * configure its bundle.
1417 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1419 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1420 "port %"PRIu16, xbridge->name, in_port);
1426 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1428 const struct xbridge *xbridge = ctx->xbridge;
1429 mirror_mask_t mirrors;
1430 struct xbundle *in_xbundle;
1434 mirrors = ctx->xout->mirrors;
1435 ctx->xout->mirrors = 0;
1437 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1438 ctx->xin->packet != NULL, NULL);
1442 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1444 /* Drop frames on bundles reserved for mirroring. */
1445 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1446 if (ctx->xin->packet != NULL) {
1447 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1448 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1449 "%s, which is reserved exclusively for mirroring",
1450 ctx->xbridge->name, in_xbundle->name);
1452 ofpbuf_clear(ctx->xout->odp_actions);
1457 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1458 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1461 vlan = input_vid_to_vlan(in_xbundle, vid);
1467 /* Restore the original packet before adding the mirror actions. */
1468 ctx->xin->flow = *orig_flow;
1471 mirror_mask_t dup_mirrors;
1472 struct ofbundle *out;
1473 unsigned long *vlans;
1478 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1479 &vlans, &dup_mirrors, &out, &out_vlan);
1480 ovs_assert(has_mirror);
1483 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1485 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1488 if (!vlan_mirrored) {
1489 mirrors = zero_rightmost_1bit(mirrors);
1493 mirrors &= ~dup_mirrors;
1494 ctx->xout->mirrors |= dup_mirrors;
1496 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1497 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1499 output_normal(ctx, out_xbundle, vlan);
1501 } else if (vlan != out_vlan
1502 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1503 struct xbundle *xbundle;
1505 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1506 if (xbundle_includes_vlan(xbundle, out_vlan)
1507 && !xbundle_mirror_out(xbridge, xbundle)) {
1508 output_normal(ctx, xbundle, out_vlan);
1515 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1516 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1517 * the bundle on which the packet was received, returns the VLAN to which the
1520 * Both 'vid' and the return value are in the range 0...4095. */
1522 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1524 switch (in_xbundle->vlan_mode) {
1525 case PORT_VLAN_ACCESS:
1526 return in_xbundle->vlan;
1529 case PORT_VLAN_TRUNK:
1532 case PORT_VLAN_NATIVE_UNTAGGED:
1533 case PORT_VLAN_NATIVE_TAGGED:
1534 return vid ? vid : in_xbundle->vlan;
1541 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1542 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1545 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1546 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1549 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1551 /* Allow any VID on the OFPP_NONE port. */
1552 if (in_xbundle == &ofpp_none_bundle) {
1556 switch (in_xbundle->vlan_mode) {
1557 case PORT_VLAN_ACCESS:
1560 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1561 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1562 "packet received on port %s configured as VLAN "
1563 "%"PRIu16" access port", vid, in_xbundle->name,
1570 case PORT_VLAN_NATIVE_UNTAGGED:
1571 case PORT_VLAN_NATIVE_TAGGED:
1573 /* Port must always carry its native VLAN. */
1577 case PORT_VLAN_TRUNK:
1578 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1580 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1581 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1582 "received on port %s not configured for trunking "
1583 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1595 /* Given 'vlan', the VLAN that a packet belongs to, and
1596 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1597 * that should be included in the 802.1Q header. (If the return value is 0,
1598 * then the 802.1Q header should only be included in the packet if there is a
1601 * Both 'vlan' and the return value are in the range 0...4095. */
1603 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1605 switch (out_xbundle->vlan_mode) {
1606 case PORT_VLAN_ACCESS:
1609 case PORT_VLAN_TRUNK:
1610 case PORT_VLAN_NATIVE_TAGGED:
1613 case PORT_VLAN_NATIVE_UNTAGGED:
1614 return vlan == out_xbundle->vlan ? 0 : vlan;
1622 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1625 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1627 ovs_be16 tci, old_tci;
1628 struct xport *xport;
1630 vid = output_vlan_to_vid(out_xbundle, vlan);
1631 if (list_is_empty(&out_xbundle->xports)) {
1632 /* Partially configured bundle with no slaves. Drop the packet. */
1634 } else if (!out_xbundle->bond) {
1635 ctx->use_recirc = false;
1636 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1639 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1640 struct flow_wildcards *wc = &ctx->xout->wc;
1641 struct xlate_recirc *xr = &ctx->recirc;
1642 struct ofport_dpif *ofport;
1644 if (ctx->xbridge->enable_recirc) {
1645 ctx->use_recirc = bond_may_recirc(
1646 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1648 if (ctx->use_recirc) {
1649 /* Only TCP mode uses recirculation. */
1650 xr->hash_alg = OVS_HASH_ALG_L4;
1651 bond_update_post_recirc_rules(out_xbundle->bond, false);
1653 /* Recirculation does not require unmasking hash fields. */
1658 ofport = bond_choose_output_slave(out_xbundle->bond,
1659 &ctx->xin->flow, wc, vid);
1660 xport = xport_lookup(xcfg, ofport);
1663 /* No slaves enabled, so drop packet. */
1667 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1668 * accounting for this bond. */
1669 if (!ctx->use_recirc) {
1670 if (ctx->xin->resubmit_stats) {
1671 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1672 ctx->xin->resubmit_stats->n_bytes);
1674 if (ctx->xin->xcache) {
1675 struct xc_entry *entry;
1678 flow = &ctx->xin->flow;
1679 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1680 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1681 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1682 entry->u.bond.vid = vid;
1687 old_tci = *flow_tci;
1689 if (tci || out_xbundle->use_priority_tags) {
1690 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1692 tci |= htons(VLAN_CFI);
1697 compose_output_action(ctx, xport->ofp_port);
1698 *flow_tci = old_tci;
1701 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1702 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1703 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1705 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1707 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1711 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1712 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1716 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1717 if (flow->nw_proto == ARP_OP_REPLY) {
1719 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1720 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1721 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1723 return flow->nw_src == flow->nw_dst;
1729 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1730 * dropped. Returns true if they may be forwarded, false if they should be
1733 * 'in_port' must be the xport that corresponds to flow->in_port.
1734 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1736 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1737 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1738 * checked by input_vid_is_valid().
1740 * May also add tags to '*tags', although the current implementation only does
1741 * so in one special case.
1744 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1747 struct xbundle *in_xbundle = in_port->xbundle;
1748 const struct xbridge *xbridge = ctx->xbridge;
1749 struct flow *flow = &ctx->xin->flow;
1751 /* Drop frames for reserved multicast addresses
1752 * only if forward_bpdu option is absent. */
1753 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1754 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1758 if (in_xbundle->bond) {
1759 struct mac_entry *mac;
1761 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1767 xlate_report(ctx, "bonding refused admissibility, dropping");
1770 case BV_DROP_IF_MOVED:
1771 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1772 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1773 if (mac && mac->port.p != in_xbundle->ofbundle &&
1774 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1775 || mac_entry_is_grat_arp_locked(mac))) {
1776 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1777 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1781 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1789 /* Checks whether a MAC learning update is necessary for MAC learning table
1790 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1793 * Most packets processed through the MAC learning table do not actually
1794 * change it in any way. This function requires only a read lock on the MAC
1795 * learning table, so it is much cheaper in this common case.
1797 * Keep the code here synchronized with that in update_learning_table__()
1800 is_mac_learning_update_needed(const struct mac_learning *ml,
1801 const struct flow *flow,
1802 struct flow_wildcards *wc,
1803 int vlan, struct xbundle *in_xbundle)
1804 OVS_REQ_RDLOCK(ml->rwlock)
1806 struct mac_entry *mac;
1808 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1812 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1813 if (!mac || mac_entry_age(ml, mac)) {
1817 if (is_gratuitous_arp(flow, wc)) {
1818 /* We don't want to learn from gratuitous ARP packets that are
1819 * reflected back over bond slaves so we lock the learning table. */
1820 if (!in_xbundle->bond) {
1822 } else if (mac_entry_is_grat_arp_locked(mac)) {
1827 return mac->port.p != in_xbundle->ofbundle;
1831 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1832 * received on 'in_xbundle' in 'vlan'.
1834 * This code repeats all the checks in is_mac_learning_update_needed() because
1835 * the lock was released between there and here and thus the MAC learning state
1836 * could have changed.
1838 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1841 update_learning_table__(const struct xbridge *xbridge,
1842 const struct flow *flow, struct flow_wildcards *wc,
1843 int vlan, struct xbundle *in_xbundle)
1844 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1846 struct mac_entry *mac;
1848 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1852 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1853 if (is_gratuitous_arp(flow, wc)) {
1854 /* We don't want to learn from gratuitous ARP packets that are
1855 * reflected back over bond slaves so we lock the learning table. */
1856 if (!in_xbundle->bond) {
1857 mac_entry_set_grat_arp_lock(mac);
1858 } else if (mac_entry_is_grat_arp_locked(mac)) {
1863 if (mac->port.p != in_xbundle->ofbundle) {
1864 /* The log messages here could actually be useful in debugging,
1865 * so keep the rate limit relatively high. */
1866 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1868 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1869 "on port %s in VLAN %d",
1870 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1871 in_xbundle->name, vlan);
1873 mac->port.p = in_xbundle->ofbundle;
1874 mac_learning_changed(xbridge->ml);
1879 update_learning_table(const struct xbridge *xbridge,
1880 const struct flow *flow, struct flow_wildcards *wc,
1881 int vlan, struct xbundle *in_xbundle)
1885 /* Don't learn the OFPP_NONE port. */
1886 if (in_xbundle == &ofpp_none_bundle) {
1890 /* First try the common case: no change to MAC learning table. */
1891 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1892 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1894 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1897 /* Slow path: MAC learning table might need an update. */
1898 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1899 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1900 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1904 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1905 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
1907 update_mcast_snooping_table__(const struct xbridge *xbridge,
1908 const struct flow *flow,
1909 struct mcast_snooping *ms,
1910 ovs_be32 ip4, int vlan,
1911 struct xbundle *in_xbundle)
1912 OVS_REQ_WRLOCK(ms->rwlock)
1914 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
1916 switch (ntohs(flow->tp_src)) {
1917 case IGMP_HOST_MEMBERSHIP_REPORT:
1918 case IGMPV2_HOST_MEMBERSHIP_REPORT:
1919 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1920 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
1921 IP_FMT" is on port %s in VLAN %d",
1922 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1925 case IGMP_HOST_LEAVE_MESSAGE:
1926 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1927 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
1928 IP_FMT" is on port %s in VLAN %d",
1929 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1932 case IGMP_HOST_MEMBERSHIP_QUERY:
1933 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
1934 in_xbundle->ofbundle)) {
1935 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
1936 IP_FMT" is on port %s in VLAN %d",
1937 xbridge->name, IP_ARGS(flow->nw_src),
1938 in_xbundle->name, vlan);
1944 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1945 * was received on 'in_xbundle' in 'vlan'. */
1947 update_mcast_snooping_table(const struct xbridge *xbridge,
1948 const struct flow *flow, int vlan,
1949 struct xbundle *in_xbundle)
1951 struct mcast_snooping *ms = xbridge->ms;
1952 struct xlate_cfg *xcfg;
1953 struct xbundle *mcast_xbundle;
1954 struct mcast_fport_bundle *fport;
1956 /* Don't learn the OFPP_NONE port. */
1957 if (in_xbundle == &ofpp_none_bundle) {
1961 /* Don't learn from flood ports */
1962 mcast_xbundle = NULL;
1963 ovs_rwlock_wrlock(&ms->rwlock);
1964 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1965 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
1966 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
1967 if (mcast_xbundle == in_xbundle) {
1972 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
1973 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
1976 ovs_rwlock_unlock(&ms->rwlock);
1979 /* send the packet to ports having the multicast group learned */
1981 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
1982 struct mcast_snooping *ms OVS_UNUSED,
1983 struct mcast_group *grp,
1984 struct xbundle *in_xbundle, uint16_t vlan)
1985 OVS_REQ_RDLOCK(ms->rwlock)
1987 struct xlate_cfg *xcfg;
1988 struct mcast_group_bundle *b;
1989 struct xbundle *mcast_xbundle;
1991 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1992 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
1993 mcast_xbundle = xbundle_lookup(xcfg, b->port);
1994 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
1995 xlate_report(ctx, "forwarding to mcast group port");
1996 output_normal(ctx, mcast_xbundle, vlan);
1997 } else if (!mcast_xbundle) {
1998 xlate_report(ctx, "mcast group port is unknown, dropping");
2000 xlate_report(ctx, "mcast group port is input port, dropping");
2005 /* send the packet to ports connected to multicast routers */
2007 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2008 struct mcast_snooping *ms,
2009 struct xbundle *in_xbundle, uint16_t vlan)
2010 OVS_REQ_RDLOCK(ms->rwlock)
2012 struct xlate_cfg *xcfg;
2013 struct mcast_mrouter_bundle *mrouter;
2014 struct xbundle *mcast_xbundle;
2016 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2017 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2018 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2019 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2020 xlate_report(ctx, "forwarding to mcast router port");
2021 output_normal(ctx, mcast_xbundle, vlan);
2022 } else if (!mcast_xbundle) {
2023 xlate_report(ctx, "mcast router port is unknown, dropping");
2025 xlate_report(ctx, "mcast router port is input port, dropping");
2030 /* send the packet to ports flagged to be flooded */
2032 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2033 struct mcast_snooping *ms,
2034 struct xbundle *in_xbundle, uint16_t vlan)
2035 OVS_REQ_RDLOCK(ms->rwlock)
2037 struct xlate_cfg *xcfg;
2038 struct mcast_fport_bundle *fport;
2039 struct xbundle *mcast_xbundle;
2041 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2042 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
2043 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2044 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2045 xlate_report(ctx, "forwarding to mcast flood port");
2046 output_normal(ctx, mcast_xbundle, vlan);
2047 } else if (!mcast_xbundle) {
2048 xlate_report(ctx, "mcast flood port is unknown, dropping");
2050 xlate_report(ctx, "mcast flood port is input port, dropping");
2056 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2059 struct xbundle *xbundle;
2061 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2062 if (xbundle != in_xbundle
2063 && xbundle_includes_vlan(xbundle, vlan)
2064 && xbundle->floodable
2065 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2066 output_normal(ctx, xbundle, vlan);
2069 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2073 xlate_normal(struct xlate_ctx *ctx)
2075 struct flow_wildcards *wc = &ctx->xout->wc;
2076 struct flow *flow = &ctx->xin->flow;
2077 struct xbundle *in_xbundle;
2078 struct xport *in_port;
2079 struct mac_entry *mac;
2084 ctx->xout->has_normal = true;
2086 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2087 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2088 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2090 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2091 ctx->xin->packet != NULL, &in_port);
2093 xlate_report(ctx, "no input bundle, dropping");
2097 /* Drop malformed frames. */
2098 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2099 !(flow->vlan_tci & htons(VLAN_CFI))) {
2100 if (ctx->xin->packet != NULL) {
2101 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2102 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2103 "VLAN tag received on port %s",
2104 ctx->xbridge->name, in_xbundle->name);
2106 xlate_report(ctx, "partial VLAN tag, dropping");
2110 /* Drop frames on bundles reserved for mirroring. */
2111 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2112 if (ctx->xin->packet != NULL) {
2113 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2114 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2115 "%s, which is reserved exclusively for mirroring",
2116 ctx->xbridge->name, in_xbundle->name);
2118 xlate_report(ctx, "input port is mirror output port, dropping");
2123 vid = vlan_tci_to_vid(flow->vlan_tci);
2124 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2125 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2128 vlan = input_vid_to_vlan(in_xbundle, vid);
2130 /* Check other admissibility requirements. */
2131 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2135 /* Learn source MAC. */
2136 if (ctx->xin->may_learn) {
2137 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2139 if (ctx->xin->xcache) {
2140 struct xc_entry *entry;
2142 /* Save enough info to update mac learning table later. */
2143 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2144 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2145 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2146 entry->u.normal.vlan = vlan;
2149 /* Determine output bundle. */
2150 if (mcast_snooping_enabled(ctx->xbridge->ms)
2151 && !eth_addr_is_broadcast(flow->dl_dst)
2152 && eth_addr_is_multicast(flow->dl_dst)
2153 && flow->dl_type == htons(ETH_TYPE_IP)) {
2154 struct mcast_snooping *ms = ctx->xbridge->ms;
2155 struct mcast_group *grp;
2157 if (flow->nw_proto == IPPROTO_IGMP) {
2158 if (ctx->xin->may_learn) {
2159 if (mcast_snooping_is_membership(flow->tp_src) ||
2160 mcast_snooping_is_query(flow->tp_src)) {
2161 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2166 if (mcast_snooping_is_membership(flow->tp_src)) {
2167 ovs_rwlock_rdlock(&ms->rwlock);
2168 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2169 ovs_rwlock_unlock(&ms->rwlock);
2171 xlate_report(ctx, "multicast traffic, flooding");
2172 xlate_normal_flood(ctx, in_xbundle, vlan);
2176 if (ip_is_local_multicast(flow->nw_dst)) {
2177 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2178 * address in the 224.0.0.x range which are not IGMP must
2179 * be forwarded on all ports */
2180 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2181 xlate_normal_flood(ctx, in_xbundle, vlan);
2186 /* forwarding to group base ports */
2187 ovs_rwlock_rdlock(&ms->rwlock);
2188 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2190 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2191 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2192 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2194 if (mcast_snooping_flood_unreg(ms)) {
2195 xlate_report(ctx, "unregistered multicast, flooding");
2196 xlate_normal_flood(ctx, in_xbundle, vlan);
2198 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2199 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2202 ovs_rwlock_unlock(&ms->rwlock);
2204 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2205 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2206 mac_port = mac ? mac->port.p : NULL;
2207 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2210 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2211 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2212 if (mac_xbundle && mac_xbundle != in_xbundle) {
2213 xlate_report(ctx, "forwarding to learned port");
2214 output_normal(ctx, mac_xbundle, vlan);
2215 } else if (!mac_xbundle) {
2216 xlate_report(ctx, "learned port is unknown, dropping");
2218 xlate_report(ctx, "learned port is input port, dropping");
2221 xlate_report(ctx, "no learned MAC for destination, flooding");
2222 xlate_normal_flood(ctx, in_xbundle, vlan);
2227 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2228 * the number of packets out of UINT32_MAX to sample. The given
2229 * cookie is passed back in the callback for each sampled packet.
2232 compose_sample_action(const struct xbridge *xbridge,
2233 struct ofpbuf *odp_actions,
2234 const struct flow *flow,
2235 const uint32_t probability,
2236 const union user_action_cookie *cookie,
2237 const size_t cookie_size,
2238 const odp_port_t tunnel_out_port)
2240 size_t sample_offset, actions_offset;
2241 odp_port_t odp_port;
2245 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2247 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2249 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2251 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2252 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2253 flow_hash_5tuple(flow, 0));
2254 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2255 tunnel_out_port, odp_actions);
2257 nl_msg_end_nested(odp_actions, actions_offset);
2258 nl_msg_end_nested(odp_actions, sample_offset);
2259 return cookie_offset;
2263 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2264 odp_port_t odp_port, unsigned int n_outputs,
2265 union user_action_cookie *cookie)
2269 cookie->type = USER_ACTION_COOKIE_SFLOW;
2270 cookie->sflow.vlan_tci = vlan_tci;
2272 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2273 * port information") for the interpretation of cookie->output. */
2274 switch (n_outputs) {
2276 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2277 cookie->sflow.output = 0x40000000 | 256;
2281 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2283 cookie->sflow.output = ifindex;
2288 /* 0x80000000 means "multiple output ports. */
2289 cookie->sflow.output = 0x80000000 | n_outputs;
2294 /* Compose SAMPLE action for sFlow bridge sampling. */
2296 compose_sflow_action(const struct xbridge *xbridge,
2297 struct ofpbuf *odp_actions,
2298 const struct flow *flow,
2299 odp_port_t odp_port)
2301 uint32_t probability;
2302 union user_action_cookie cookie;
2304 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2308 probability = dpif_sflow_get_probability(xbridge->sflow);
2309 compose_sflow_cookie(xbridge, htons(0), odp_port,
2310 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2312 return compose_sample_action(xbridge, odp_actions, flow, probability,
2313 &cookie, sizeof cookie.sflow, ODPP_NONE);
2317 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2318 uint32_t obs_domain_id, uint32_t obs_point_id,
2319 union user_action_cookie *cookie)
2321 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2322 cookie->flow_sample.probability = probability;
2323 cookie->flow_sample.collector_set_id = collector_set_id;
2324 cookie->flow_sample.obs_domain_id = obs_domain_id;
2325 cookie->flow_sample.obs_point_id = obs_point_id;
2329 compose_ipfix_cookie(union user_action_cookie *cookie,
2330 odp_port_t output_odp_port)
2332 cookie->type = USER_ACTION_COOKIE_IPFIX;
2333 cookie->ipfix.output_odp_port = output_odp_port;
2336 /* Compose SAMPLE action for IPFIX bridge sampling. */
2338 compose_ipfix_action(const struct xbridge *xbridge,
2339 struct ofpbuf *odp_actions,
2340 const struct flow *flow,
2341 odp_port_t output_odp_port)
2343 uint32_t probability;
2344 union user_action_cookie cookie;
2345 odp_port_t tunnel_out_port = ODPP_NONE;
2347 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2351 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2353 if (output_odp_port == ODPP_NONE &&
2354 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2358 /* For output case, output_odp_port is valid*/
2359 if (output_odp_port != ODPP_NONE) {
2360 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2363 /* If tunnel sampling is enabled, put an additional option attribute:
2364 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2366 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2367 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2368 tunnel_out_port = output_odp_port;
2372 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2373 compose_ipfix_cookie(&cookie, output_odp_port);
2375 compose_sample_action(xbridge, odp_actions, flow, probability,
2376 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2379 /* SAMPLE action for sFlow must be first action in any given list of
2380 * actions. At this point we do not have all information required to
2381 * build it. So try to build sample action as complete as possible. */
2383 add_sflow_action(struct xlate_ctx *ctx)
2385 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2386 ctx->xout->odp_actions,
2387 &ctx->xin->flow, ODPP_NONE);
2388 ctx->sflow_odp_port = 0;
2389 ctx->sflow_n_outputs = 0;
2392 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2393 * of actions, eventually after the SAMPLE action for sFlow. */
2395 add_ipfix_action(struct xlate_ctx *ctx)
2397 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2398 &ctx->xin->flow, ODPP_NONE);
2402 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2404 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2405 &ctx->xin->flow, port);
2408 /* Fix SAMPLE action according to data collected while composing ODP actions.
2409 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2410 * USERSPACE action's user-cookie which is required for sflow. */
2412 fix_sflow_action(struct xlate_ctx *ctx)
2414 const struct flow *base = &ctx->base_flow;
2415 union user_action_cookie *cookie;
2417 if (!ctx->user_cookie_offset) {
2421 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2422 sizeof cookie->sflow);
2423 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2425 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2426 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2429 static enum slow_path_reason
2430 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2431 const struct xport *xport, const struct ofpbuf *packet)
2433 struct flow_wildcards *wc = &ctx->xout->wc;
2434 const struct xbridge *xbridge = ctx->xbridge;
2438 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2440 cfm_process_heartbeat(xport->cfm, packet);
2443 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2445 bfd_process_packet(xport->bfd, flow, packet);
2446 /* If POLL received, immediately sends FINAL back. */
2447 if (bfd_should_send_packet(xport->bfd)) {
2448 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2452 } else if (xport->xbundle && xport->xbundle->lacp
2453 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2455 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2458 } else if ((xbridge->stp || xbridge->rstp) &&
2459 stp_should_process_flow(flow, wc)) {
2462 ? stp_process_packet(xport, packet)
2463 : rstp_process_packet(xport, packet);
2472 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2475 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2476 struct flow_wildcards *wc = &ctx->xout->wc;
2477 struct flow *flow = &ctx->xin->flow;
2478 ovs_be16 flow_vlan_tci;
2479 uint32_t flow_pkt_mark;
2480 uint8_t flow_nw_tos;
2481 odp_port_t out_port, odp_port;
2484 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2485 * before traversing a patch port. */
2486 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 27);
2489 xlate_report(ctx, "Nonexistent output port");
2491 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2492 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2494 } else if (check_stp) {
2495 if (is_stp(&ctx->base_flow)) {
2496 if (!xport_stp_should_forward_bpdu(xport) &&
2497 !xport_rstp_should_manage_bpdu(xport)) {
2498 if (ctx->xbridge->stp != NULL) {
2499 xlate_report(ctx, "STP not in listening state, "
2500 "skipping bpdu output");
2501 } else if (ctx->xbridge->rstp != NULL) {
2502 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2503 "skipping bpdu output");
2507 } else if (!xport_stp_forward_state(xport) ||
2508 !xport_rstp_forward_state(xport)) {
2509 if (ctx->xbridge->stp != NULL) {
2510 xlate_report(ctx, "STP not in forwarding state, "
2512 } else if (ctx->xbridge->rstp != NULL) {
2513 xlate_report(ctx, "RSTP not in forwarding state, "
2520 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2521 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2526 const struct xport *peer = xport->peer;
2527 struct flow old_flow = ctx->xin->flow;
2528 enum slow_path_reason special;
2530 ctx->xbridge = peer->xbridge;
2531 flow->in_port.ofp_port = peer->ofp_port;
2532 flow->metadata = htonll(0);
2533 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2534 memset(flow->regs, 0, sizeof flow->regs);
2536 special = process_special(ctx, &ctx->xin->flow, peer,
2539 ctx->xout->slow |= special;
2540 } else if (may_receive(peer, ctx)) {
2541 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2542 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2544 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2545 * the learning action look at the packet, then drop it. */
2546 struct flow old_base_flow = ctx->base_flow;
2547 size_t old_size = ofpbuf_size(ctx->xout->odp_actions);
2548 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2549 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2550 ctx->xout->mirrors = old_mirrors;
2551 ctx->base_flow = old_base_flow;
2552 ofpbuf_set_size(ctx->xout->odp_actions, old_size);
2556 ctx->xin->flow = old_flow;
2557 ctx->xbridge = xport->xbridge;
2559 if (ctx->xin->resubmit_stats) {
2560 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2561 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2563 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2566 if (ctx->xin->xcache) {
2567 struct xc_entry *entry;
2569 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2570 entry->u.dev.tx = netdev_ref(xport->netdev);
2571 entry->u.dev.rx = netdev_ref(peer->netdev);
2572 entry->u.dev.bfd = bfd_ref(peer->bfd);
2577 flow_vlan_tci = flow->vlan_tci;
2578 flow_pkt_mark = flow->pkt_mark;
2579 flow_nw_tos = flow->nw_tos;
2581 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2582 wc->masks.nw_tos |= IP_DSCP_MASK;
2583 flow->nw_tos &= ~IP_DSCP_MASK;
2584 flow->nw_tos |= dscp;
2587 if (xport->is_tunnel) {
2588 /* Save tunnel metadata so that changes made due to
2589 * the Logical (tunnel) Port are not visible for any further
2590 * matches, while explicit set actions on tunnel metadata are.
2592 struct flow_tnl flow_tnl = flow->tunnel;
2593 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2594 if (odp_port == ODPP_NONE) {
2595 xlate_report(ctx, "Tunneling decided against output");
2596 goto out; /* restore flow_nw_tos */
2598 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2599 xlate_report(ctx, "Not tunneling to our own address");
2600 goto out; /* restore flow_nw_tos */
2602 if (ctx->xin->resubmit_stats) {
2603 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2605 if (ctx->xin->xcache) {
2606 struct xc_entry *entry;
2608 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2609 entry->u.dev.tx = netdev_ref(xport->netdev);
2611 out_port = odp_port;
2612 commit_odp_tunnel_action(flow, &ctx->base_flow,
2613 ctx->xout->odp_actions);
2614 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2616 odp_port = xport->odp_port;
2617 out_port = odp_port;
2618 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2619 ofp_port_t vlandev_port;
2621 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2622 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2623 ofp_port, flow->vlan_tci);
2624 if (vlandev_port != ofp_port) {
2625 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2626 flow->vlan_tci = htons(0);
2631 if (out_port != ODPP_NONE) {
2632 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2633 ctx->xout->odp_actions,
2635 ctx->xbridge->masked_set_action);
2637 if (ctx->use_recirc) {
2638 struct ovs_action_hash *act_hash;
2639 struct xlate_recirc *xr = &ctx->recirc;
2642 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2643 OVS_ACTION_ATTR_HASH,
2645 act_hash->hash_alg = xr->hash_alg;
2646 act_hash->hash_basis = xr->hash_basis;
2648 /* Recirc action. */
2649 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2652 add_ipfix_output_action(ctx, out_port);
2653 nl_msg_put_odp_port(ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
2657 ctx->sflow_odp_port = odp_port;
2658 ctx->sflow_n_outputs++;
2659 ctx->xout->nf_output_iface = ofp_port;
2664 flow->vlan_tci = flow_vlan_tci;
2665 flow->pkt_mark = flow_pkt_mark;
2666 flow->nw_tos = flow_nw_tos;
2670 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
2672 compose_output_action__(ctx, ofp_port, true);
2676 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
2678 struct rule_dpif *old_rule = ctx->rule;
2679 const struct rule_actions *actions;
2681 if (ctx->xin->resubmit_stats) {
2682 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2688 actions = rule_dpif_get_actions(rule);
2689 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2690 ctx->rule = old_rule;
2695 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2697 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2699 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2700 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2701 MAX_RESUBMIT_RECURSION);
2702 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2703 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2704 } else if (ofpbuf_size(ctx->xout->odp_actions) > UINT16_MAX) {
2705 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2706 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2707 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2716 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2717 bool may_packet_in, bool honor_table_miss)
2719 if (xlate_resubmit_resource_check(ctx)) {
2720 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
2721 bool skip_wildcards = ctx->xin->skip_wildcards;
2722 uint8_t old_table_id = ctx->table_id;
2723 struct rule_dpif *rule;
2724 enum rule_dpif_lookup_verdict verdict;
2725 enum ofputil_port_config config = 0;
2727 ctx->table_id = table_id;
2729 /* Look up a flow with 'in_port' as the input port. Then restore the
2730 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2731 * have surprising behavior). */
2732 ctx->xin->flow.in_port.ofp_port = in_port;
2733 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2736 ? &ctx->xout->wc : NULL,
2738 &ctx->table_id, &rule,
2739 ctx->xin->xcache != NULL,
2740 ctx->xin->resubmit_stats);
2741 ctx->xin->flow.in_port.ofp_port = old_in_port;
2743 if (ctx->xin->resubmit_hook) {
2744 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
2748 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
2750 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
2751 if (may_packet_in) {
2752 struct xport *xport;
2754 xport = get_ofp_port(ctx->xbridge,
2755 ctx->xin->flow.in_port.ofp_port);
2756 config = xport ? xport->config : 0;
2759 /* Fall through to drop */
2760 case RULE_DPIF_LOOKUP_VERDICT_DROP:
2761 config = OFPUTIL_PC_NO_PACKET_IN;
2763 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
2764 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
2765 config = OFPUTIL_PC_NO_PACKET_IN;
2772 choose_miss_rule(config, ctx->xbridge->miss_rule,
2773 ctx->xbridge->no_packet_in_rule, &rule,
2774 ctx->xin->xcache != NULL);
2778 /* Fill in the cache entry here instead of xlate_recursively
2779 * to make the reference counting more explicit. We take a
2780 * reference in the lookups above if we are going to cache the
2782 if (ctx->xin->xcache) {
2783 struct xc_entry *entry;
2785 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2786 entry->u.rule = rule;
2788 xlate_recursively(ctx, rule);
2791 ctx->table_id = old_table_id;
2799 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
2800 struct ofputil_bucket *bucket)
2802 if (ctx->xin->resubmit_stats) {
2803 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
2805 if (ctx->xin->xcache) {
2806 struct xc_entry *entry;
2808 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
2809 entry->u.group.group = group_dpif_ref(group);
2810 entry->u.group.bucket = bucket;
2815 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
2817 uint64_t action_list_stub[1024 / 8];
2818 struct ofpbuf action_list, action_set;
2820 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2821 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2823 ofpacts_execute_action_set(&action_list, &action_set);
2825 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2828 ofpbuf_uninit(&action_set);
2829 ofpbuf_uninit(&action_list);
2833 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2835 struct ofputil_bucket *bucket;
2836 const struct list *buckets;
2837 struct flow old_flow = ctx->xin->flow;
2839 group_dpif_get_buckets(group, &buckets);
2841 LIST_FOR_EACH (bucket, list_node, buckets) {
2842 xlate_group_bucket(ctx, bucket);
2843 /* Roll back flow to previous state.
2844 * This is equivalent to cloning the packet for each bucket.
2846 * As a side effect any subsequently applied actions will
2847 * also effectively be applied to a clone of the packet taken
2848 * just before applying the all or indirect group. */
2849 ctx->xin->flow = old_flow;
2851 xlate_group_stats(ctx, group, NULL);
2855 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2857 struct ofputil_bucket *bucket;
2859 bucket = group_first_live_bucket(ctx, group, 0);
2861 xlate_group_bucket(ctx, bucket);
2862 xlate_group_stats(ctx, group, bucket);
2867 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2869 struct flow_wildcards *wc = &ctx->xout->wc;
2870 struct ofputil_bucket *bucket;
2873 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
2874 bucket = group_best_live_bucket(ctx, group, basis);
2876 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2877 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2878 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2879 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2880 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2881 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2882 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2883 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2884 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2886 xlate_group_bucket(ctx, bucket);
2887 xlate_group_stats(ctx, group, bucket);
2892 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2894 ctx->in_group = true;
2896 switch (group_dpif_get_type(group)) {
2898 case OFPGT11_INDIRECT:
2899 xlate_all_group(ctx, group);
2901 case OFPGT11_SELECT:
2902 xlate_select_group(ctx, group);
2905 xlate_ff_group(ctx, group);
2910 group_dpif_unref(group);
2912 ctx->in_group = false;
2916 xlate_group_resource_check(struct xlate_ctx *ctx)
2918 if (!xlate_resubmit_resource_check(ctx)) {
2920 } else if (ctx->in_group) {
2921 /* Prevent nested translation of OpenFlow groups.
2923 * OpenFlow allows this restriction. We enforce this restriction only
2924 * because, with the current architecture, we would otherwise have to
2925 * take a possibly recursive read lock on the ofgroup rwlock, which is
2926 * unsafe given that POSIX allows taking a read lock to block if there
2927 * is a thread blocked on taking the write lock. Other solutions
2928 * without this restriction are also possible, but seem unwarranted
2929 * given the current limited use of groups. */
2930 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2932 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2940 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2942 if (xlate_group_resource_check(ctx)) {
2943 struct group_dpif *group;
2946 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2948 xlate_group_action__(ctx, group);
2958 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2959 const struct ofpact_resubmit *resubmit)
2963 bool may_packet_in = false;
2964 bool honor_table_miss = false;
2966 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2967 /* Still allow missed packets to be sent to the controller
2968 * if resubmitting from an internal table. */
2969 may_packet_in = true;
2970 honor_table_miss = true;
2973 in_port = resubmit->in_port;
2974 if (in_port == OFPP_IN_PORT) {
2975 in_port = ctx->xin->flow.in_port.ofp_port;
2978 table_id = resubmit->table_id;
2979 if (table_id == 255) {
2980 table_id = ctx->table_id;
2983 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2988 flood_packets(struct xlate_ctx *ctx, bool all)
2990 const struct xport *xport;
2992 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2993 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2998 compose_output_action__(ctx, xport->ofp_port, false);
2999 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3000 compose_output_action(ctx, xport->ofp_port);
3004 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3008 execute_controller_action(struct xlate_ctx *ctx, int len,
3009 enum ofp_packet_in_reason reason,
3010 uint16_t controller_id)
3012 struct ofproto_packet_in *pin;
3013 struct dpif_packet *packet;
3014 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
3016 ctx->xout->slow |= SLOW_CONTROLLER;
3017 if (!ctx->xin->packet) {
3021 packet = dpif_packet_clone_from_ofpbuf(ctx->xin->packet);
3023 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3024 ctx->xout->odp_actions,
3026 ctx->xbridge->masked_set_action);
3028 odp_execute_actions(NULL, &packet, 1, false, &md,
3029 ofpbuf_data(ctx->xout->odp_actions),
3030 ofpbuf_size(ctx->xout->odp_actions), NULL);
3032 pin = xmalloc(sizeof *pin);
3033 pin->up.packet_len = ofpbuf_size(&packet->ofpbuf);
3034 pin->up.packet = ofpbuf_steal_data(&packet->ofpbuf);
3035 pin->up.reason = reason;
3036 pin->up.table_id = ctx->table_id;
3037 pin->up.cookie = (ctx->rule
3038 ? rule_dpif_get_flow_cookie(ctx->rule)
3041 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3043 pin->controller_id = controller_id;
3044 pin->send_len = len;
3045 /* If a rule is a table-miss rule then this is
3046 * a table-miss handled by a table-miss rule.
3048 * Else, if rule is internal and has a controller action,
3049 * the later being implied by the rule being processed here,
3050 * then this is a table-miss handled without a table-miss rule.
3052 * Otherwise this is not a table-miss. */
3053 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3055 if (rule_dpif_is_table_miss(ctx->rule)) {
3056 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3057 } else if (rule_dpif_is_internal(ctx->rule)) {
3058 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3061 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3062 dpif_packet_delete(packet);
3066 compose_recirculate_action(struct xlate_ctx *ctx,
3067 const struct ofpact *ofpacts_base,
3068 const struct ofpact *ofpact_current,
3069 size_t ofpacts_base_len)
3073 unsigned ofpacts_len;
3076 struct ofpbuf ofpacts;
3080 ofpacts_len = ofpacts_base_len -
3081 ((uint8_t *)ofpact_current - (uint8_t *)ofpacts_base);
3084 id = rule_dpif_get_recirc_id(ctx->rule);
3086 /* In the case where ctx has no rule then allocate a recirc id.
3087 * The life-cycle of this recirc id is managed by associating it
3088 * with the internal rule that is created to to handle
3089 * recirculation below.
3091 * The known use-case of this is packet_out which
3092 * translates actions without a rule */
3093 id = ofproto_dpif_alloc_recirc_id(ctx->xbridge->ofproto);
3096 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3097 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3101 match_init_catchall(&match);
3102 match_set_recirc_id(&match, id);
3103 ofpbuf_use_const(&ofpacts, ofpact_current, ofpacts_len);
3104 error = ofproto_dpif_add_internal_flow(ctx->xbridge->ofproto, &match,
3105 RECIRC_RULE_PRIORITY,
3106 RECIRC_TIMEOUT, &ofpacts, &rule);
3108 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3109 VLOG_ERR_RL(&rl, "Failed to add post recirculation flow %s",
3110 match_to_string(&match, 0));
3113 /* If ctx has no rule then associate the recirc id, which
3114 * was allocated above, with the internal rule. This allows
3115 * the recirc id to be released when the internal rule times out. */
3117 rule_set_recirc_id(rule, id);
3120 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3121 ctx->xout->odp_actions,
3123 ctx->xbridge->masked_set_action);
3124 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3128 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3130 struct flow_wildcards *wc = &ctx->xout->wc;
3131 struct flow *flow = &ctx->xin->flow;
3134 ovs_assert(eth_type_mpls(mpls->ethertype));
3136 n = flow_count_mpls_labels(flow, wc);
3138 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3139 ctx->xout->odp_actions,
3141 ctx->xbridge->masked_set_action);
3142 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3143 if (ctx->xin->packet != NULL) {
3144 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3145 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3146 "MPLS push action can't be performed as it would "
3147 "have more MPLS LSEs than the %d supported.",
3148 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3152 } else if (n >= ctx->xbridge->max_mpls_depth) {
3153 COVERAGE_INC(xlate_actions_mpls_overflow);
3154 ctx->xout->slow |= SLOW_ACTION;
3157 flow_push_mpls(flow, n, mpls->ethertype, wc);
3161 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3163 struct flow_wildcards *wc = &ctx->xout->wc;
3164 struct flow *flow = &ctx->xin->flow;
3165 int n = flow_count_mpls_labels(flow, wc);
3167 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3168 if (ctx->xbridge->enable_recirc && !eth_type_mpls(eth_type)) {
3169 ctx->was_mpls = true;
3171 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3172 if (ctx->xin->packet != NULL) {
3173 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3174 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3175 "MPLS pop action can't be performed as it has "
3176 "more MPLS LSEs than the %d supported.",
3177 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3180 ofpbuf_clear(ctx->xout->odp_actions);
3185 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3187 struct flow *flow = &ctx->xin->flow;
3189 if (!is_ip_any(flow)) {
3193 ctx->xout->wc.masks.nw_ttl = 0xff;
3194 if (flow->nw_ttl > 1) {
3200 for (i = 0; i < ids->n_controllers; i++) {
3201 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3205 /* Stop processing for current table. */
3211 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3213 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3214 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3215 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3220 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3222 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3223 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3224 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3229 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3231 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3232 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3233 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3238 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3240 struct flow *flow = &ctx->xin->flow;
3241 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3242 struct flow_wildcards *wc = &ctx->xout->wc;
3244 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
3245 if (eth_type_mpls(flow->dl_type)) {
3248 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3251 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3253 /* Stop processing for current table. */
3262 xlate_output_action(struct xlate_ctx *ctx,
3263 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3265 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3267 ctx->xout->nf_output_iface = NF_OUT_DROP;
3271 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
3274 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3275 0, may_packet_in, true);
3281 flood_packets(ctx, false);
3284 flood_packets(ctx, true);
3286 case OFPP_CONTROLLER:
3287 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
3293 if (port != ctx->xin->flow.in_port.ofp_port) {
3294 compose_output_action(ctx, port);
3296 xlate_report(ctx, "skipping output to input port");
3301 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3302 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3303 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3304 ctx->xout->nf_output_iface = prev_nf_output_iface;
3305 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3306 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3307 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3312 xlate_output_reg_action(struct xlate_ctx *ctx,
3313 const struct ofpact_output_reg *or)
3315 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3316 if (port <= UINT16_MAX) {
3317 union mf_subvalue value;
3319 memset(&value, 0xff, sizeof value);
3320 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3321 xlate_output_action(ctx, u16_to_ofp(port),
3322 or->max_len, false);
3327 xlate_enqueue_action(struct xlate_ctx *ctx,
3328 const struct ofpact_enqueue *enqueue)
3330 ofp_port_t ofp_port = enqueue->port;
3331 uint32_t queue_id = enqueue->queue;
3332 uint32_t flow_priority, priority;
3335 /* Translate queue to priority. */
3336 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3338 /* Fall back to ordinary output action. */
3339 xlate_output_action(ctx, enqueue->port, 0, false);
3343 /* Check output port. */
3344 if (ofp_port == OFPP_IN_PORT) {
3345 ofp_port = ctx->xin->flow.in_port.ofp_port;
3346 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3350 /* Add datapath actions. */
3351 flow_priority = ctx->xin->flow.skb_priority;
3352 ctx->xin->flow.skb_priority = priority;
3353 compose_output_action(ctx, ofp_port);
3354 ctx->xin->flow.skb_priority = flow_priority;
3356 /* Update NetFlow output port. */
3357 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3358 ctx->xout->nf_output_iface = ofp_port;
3359 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3360 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3365 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3367 uint32_t skb_priority;
3369 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3370 ctx->xin->flow.skb_priority = skb_priority;
3372 /* Couldn't translate queue to a priority. Nothing to do. A warning
3373 * has already been logged. */
3378 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3380 const struct xbridge *xbridge = xbridge_;
3391 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3394 port = get_ofp_port(xbridge, ofp_port);
3395 return port ? port->may_enable : false;
3400 xlate_bundle_action(struct xlate_ctx *ctx,
3401 const struct ofpact_bundle *bundle)
3405 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3407 CONST_CAST(struct xbridge *, ctx->xbridge));
3408 if (bundle->dst.field) {
3409 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3412 xlate_output_action(ctx, port, 0, false);
3417 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3418 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3420 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3421 if (ctx->xin->may_learn) {
3422 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3427 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3429 ctx->xout->has_learn = true;
3430 learn_mask(learn, &ctx->xout->wc);
3432 if (ctx->xin->xcache) {
3433 struct xc_entry *entry;
3435 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3436 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3437 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3438 entry->u.learn.ofpacts = ofpbuf_new(64);
3439 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3440 entry->u.learn.ofpacts);
3441 } else if (ctx->xin->may_learn) {
3442 uint64_t ofpacts_stub[1024 / 8];
3443 struct ofputil_flow_mod fm;
3444 struct ofpbuf ofpacts;
3446 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3447 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3448 ofpbuf_uninit(&ofpacts);
3453 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3454 uint16_t idle_timeout, uint16_t hard_timeout)
3456 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3457 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3462 xlate_fin_timeout(struct xlate_ctx *ctx,
3463 const struct ofpact_fin_timeout *oft)
3466 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3467 oft->fin_idle_timeout, oft->fin_hard_timeout);
3468 if (ctx->xin->xcache) {
3469 struct xc_entry *entry;
3471 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3472 /* XC_RULE already holds a reference on the rule, none is taken
3474 entry->u.fin.rule = ctx->rule;
3475 entry->u.fin.idle = oft->fin_idle_timeout;
3476 entry->u.fin.hard = oft->fin_hard_timeout;
3482 xlate_sample_action(struct xlate_ctx *ctx,
3483 const struct ofpact_sample *os)
3485 union user_action_cookie cookie;
3486 /* Scale the probability from 16-bit to 32-bit while representing
3487 * the same percentage. */
3488 uint32_t probability = (os->probability << 16) | os->probability;
3490 if (!ctx->xbridge->variable_length_userdata) {
3491 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3493 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3494 "lacks support (needs Linux 3.10+ or kernel module from "
3499 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3500 ctx->xout->odp_actions,
3502 ctx->xbridge->masked_set_action);
3504 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3505 os->obs_domain_id, os->obs_point_id, &cookie);
3506 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions, &ctx->xin->flow,
3507 probability, &cookie, sizeof cookie.flow_sample,
3512 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3514 if (xport->config & (is_stp(&ctx->xin->flow)
3515 ? OFPUTIL_PC_NO_RECV_STP
3516 : OFPUTIL_PC_NO_RECV)) {
3520 /* Only drop packets here if both forwarding and learning are
3521 * disabled. If just learning is enabled, we need to have
3522 * OFPP_NORMAL and the learning action have a look at the packet
3523 * before we can drop it. */
3524 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3525 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3533 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3535 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3536 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
3537 ofpact_pad(&ctx->action_set);
3541 xlate_action_set(struct xlate_ctx *ctx)
3543 uint64_t action_list_stub[1024 / 64];
3544 struct ofpbuf action_list;
3546 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3547 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3548 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
3549 ofpbuf_uninit(&action_list);
3553 ofpact_needs_recirculation_after_mpls(const struct xlate_ctx *ctx,
3554 const struct ofpact *a)
3556 struct flow_wildcards *wc = &ctx->xout->wc;
3557 struct flow *flow = &ctx->xin->flow;
3562 case OFPACT_CONTROLLER:
3563 case OFPACT_STRIP_VLAN:
3564 case OFPACT_SET_VLAN_PCP:
3565 case OFPACT_SET_VLAN_VID:
3566 case OFPACT_ENQUEUE:
3567 case OFPACT_PUSH_VLAN:
3568 case OFPACT_SET_ETH_SRC:
3569 case OFPACT_SET_ETH_DST:
3570 case OFPACT_SET_TUNNEL:
3571 case OFPACT_SET_QUEUE:
3572 case OFPACT_POP_QUEUE:
3573 case OFPACT_POP_MPLS:
3574 case OFPACT_DEC_MPLS_TTL:
3575 case OFPACT_SET_MPLS_TTL:
3576 case OFPACT_SET_MPLS_TC:
3577 case OFPACT_SET_MPLS_LABEL:
3579 case OFPACT_OUTPUT_REG:
3582 case OFPACT_WRITE_METADATA:
3583 case OFPACT_WRITE_ACTIONS:
3584 case OFPACT_CLEAR_ACTIONS:
3588 case OFPACT_SET_IPV4_SRC:
3589 case OFPACT_SET_IPV4_DST:
3590 case OFPACT_SET_IP_DSCP:
3591 case OFPACT_SET_IP_ECN:
3592 case OFPACT_SET_IP_TTL:
3593 case OFPACT_SET_L4_SRC_PORT:
3594 case OFPACT_SET_L4_DST_PORT:
3595 case OFPACT_RESUBMIT:
3596 case OFPACT_STACK_PUSH:
3597 case OFPACT_STACK_POP:
3598 case OFPACT_DEC_TTL:
3599 case OFPACT_MULTIPATH:
3602 case OFPACT_FIN_TIMEOUT:
3603 case OFPACT_GOTO_TABLE:
3606 case OFPACT_REG_MOVE:
3607 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
3608 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
3610 case OFPACT_REG_LOAD:
3611 return mf_is_l3_or_higher(ofpact_get_REG_LOAD(a)->dst.field);
3613 case OFPACT_SET_FIELD:
3614 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
3616 case OFPACT_PUSH_MPLS:
3617 /* Recirculate if it is an IP packet with a zero ttl. This may
3618 * indicate that the packet was previously MPLS and an MPLS pop action
3619 * converted it to IP. In this case recirculating should reveal the IP
3620 * TTL which is used as the basis for a new MPLS LSE. */
3621 return (!flow_count_mpls_labels(flow, wc)
3622 && flow->nw_ttl == 0
3623 && is_ip_any(flow));
3630 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
3631 struct xlate_ctx *ctx)
3633 struct flow_wildcards *wc = &ctx->xout->wc;
3634 struct flow *flow = &ctx->xin->flow;
3635 const struct ofpact *a;
3637 /* dl_type already in the mask, not set below. */
3639 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3640 struct ofpact_controller *controller;
3641 const struct ofpact_metadata *metadata;
3642 const struct ofpact_set_field *set_field;
3643 const struct mf_field *mf;
3649 if (ctx->was_mpls && ofpact_needs_recirculation_after_mpls(ctx, a)) {
3650 compose_recirculate_action(ctx, ofpacts, a, ofpacts_len);
3656 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
3657 ofpact_get_OUTPUT(a)->max_len, true);
3661 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
3666 case OFPACT_CONTROLLER:
3667 controller = ofpact_get_CONTROLLER(a);
3668 execute_controller_action(ctx, controller->max_len,
3670 controller->controller_id);
3673 case OFPACT_ENQUEUE:
3674 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
3677 case OFPACT_SET_VLAN_VID:
3678 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3679 if (flow->vlan_tci & htons(VLAN_CFI) ||
3680 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
3681 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
3682 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
3687 case OFPACT_SET_VLAN_PCP:
3688 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
3689 if (flow->vlan_tci & htons(VLAN_CFI) ||
3690 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
3691 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
3692 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
3693 << VLAN_PCP_SHIFT) | VLAN_CFI);
3697 case OFPACT_STRIP_VLAN:
3698 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3699 flow->vlan_tci = htons(0);
3702 case OFPACT_PUSH_VLAN:
3703 /* XXX 802.1AD(QinQ) */
3704 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3705 flow->vlan_tci = htons(VLAN_CFI);
3708 case OFPACT_SET_ETH_SRC:
3709 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3710 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
3713 case OFPACT_SET_ETH_DST:
3714 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3715 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
3718 case OFPACT_SET_IPV4_SRC:
3719 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3720 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3721 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
3725 case OFPACT_SET_IPV4_DST:
3726 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3727 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3728 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
3732 case OFPACT_SET_IP_DSCP:
3733 if (is_ip_any(flow)) {
3734 wc->masks.nw_tos |= IP_DSCP_MASK;
3735 flow->nw_tos &= ~IP_DSCP_MASK;
3736 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
3740 case OFPACT_SET_IP_ECN:
3741 if (is_ip_any(flow)) {
3742 wc->masks.nw_tos |= IP_ECN_MASK;
3743 flow->nw_tos &= ~IP_ECN_MASK;
3744 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
3748 case OFPACT_SET_IP_TTL:
3749 if (is_ip_any(flow)) {
3750 wc->masks.nw_ttl = 0xff;
3751 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
3755 case OFPACT_SET_L4_SRC_PORT:
3756 if (is_ip_any(flow)) {
3757 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3758 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3759 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
3763 case OFPACT_SET_L4_DST_PORT:
3764 if (is_ip_any(flow)) {
3765 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3766 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3767 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
3771 case OFPACT_RESUBMIT:
3772 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
3775 case OFPACT_SET_TUNNEL:
3776 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
3779 case OFPACT_SET_QUEUE:
3780 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
3783 case OFPACT_POP_QUEUE:
3784 flow->skb_priority = ctx->orig_skb_priority;
3787 case OFPACT_REG_MOVE:
3788 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
3791 case OFPACT_REG_LOAD:
3792 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
3795 case OFPACT_SET_FIELD:
3796 set_field = ofpact_get_SET_FIELD(a);
3797 mf = set_field->field;
3799 /* Set field action only ever overwrites packet's outermost
3800 * applicable header fields. Do nothing if no header exists. */
3801 if (mf->id == MFF_VLAN_VID) {
3802 wc->masks.vlan_tci |= htons(VLAN_CFI);
3803 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
3806 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
3807 /* 'dl_type' is already unwildcarded. */
3808 && !eth_type_mpls(flow->dl_type)) {
3812 mf_mask_field_and_prereqs(mf, &wc->masks);
3813 mf_set_flow_value(mf, &set_field->value, flow);
3816 case OFPACT_STACK_PUSH:
3817 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
3821 case OFPACT_STACK_POP:
3822 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
3826 case OFPACT_PUSH_MPLS:
3827 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
3830 case OFPACT_POP_MPLS:
3831 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
3834 case OFPACT_SET_MPLS_LABEL:
3835 compose_set_mpls_label_action(
3836 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
3839 case OFPACT_SET_MPLS_TC:
3840 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
3843 case OFPACT_SET_MPLS_TTL:
3844 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
3847 case OFPACT_DEC_MPLS_TTL:
3848 if (compose_dec_mpls_ttl_action(ctx)) {
3853 case OFPACT_DEC_TTL:
3854 wc->masks.nw_ttl = 0xff;
3855 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
3861 /* Nothing to do. */
3864 case OFPACT_MULTIPATH:
3865 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
3869 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
3872 case OFPACT_OUTPUT_REG:
3873 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
3877 xlate_learn_action(ctx, ofpact_get_LEARN(a));
3884 case OFPACT_FIN_TIMEOUT:
3885 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3886 ctx->xout->has_fin_timeout = true;
3887 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
3890 case OFPACT_CLEAR_ACTIONS:
3891 ofpbuf_clear(&ctx->action_set);
3894 case OFPACT_WRITE_ACTIONS:
3895 xlate_write_actions(ctx, a);
3898 case OFPACT_WRITE_METADATA:
3899 metadata = ofpact_get_WRITE_METADATA(a);
3900 flow->metadata &= ~metadata->mask;
3901 flow->metadata |= metadata->metadata & metadata->mask;
3905 /* Not implemented yet. */
3908 case OFPACT_GOTO_TABLE: {
3909 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
3911 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
3912 * than ogt->table_id. This is to allow goto_table actions that
3913 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
3914 * after recirculation. */
3915 ovs_assert(ctx->table_id == TBL_INTERNAL
3916 || ctx->table_id < ogt->table_id);
3917 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3918 ogt->table_id, true, true);
3923 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
3930 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
3931 const struct flow *flow, ofp_port_t in_port,
3932 struct rule_dpif *rule, uint16_t tcp_flags,
3933 const struct ofpbuf *packet)
3935 xin->ofproto = ofproto;
3937 xin->flow.in_port.ofp_port = in_port;
3938 xin->packet = packet;
3939 xin->may_learn = packet != NULL;
3942 xin->ofpacts = NULL;
3943 xin->ofpacts_len = 0;
3944 xin->tcp_flags = tcp_flags;
3945 xin->resubmit_hook = NULL;
3946 xin->report_hook = NULL;
3947 xin->resubmit_stats = NULL;
3948 xin->skip_wildcards = false;
3949 xin->odp_actions = NULL;
3953 xlate_out_uninit(struct xlate_out *xout)
3955 if (xout && xout->odp_actions == &xout->odp_actions_buf) {
3956 ofpbuf_uninit(xout->odp_actions);
3960 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
3961 * into datapath actions, using 'ctx', and discards the datapath actions. */
3963 xlate_actions_for_side_effects(struct xlate_in *xin)
3965 struct xlate_out xout;
3967 xlate_actions(xin, &xout);
3968 xlate_out_uninit(&xout);
3972 xlate_report(struct xlate_ctx *ctx, const char *s)
3974 if (ctx->xin->report_hook) {
3975 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
3980 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
3983 dst->slow = src->slow;
3984 dst->has_learn = src->has_learn;
3985 dst->has_normal = src->has_normal;
3986 dst->has_fin_timeout = src->has_fin_timeout;
3987 dst->nf_output_iface = src->nf_output_iface;
3988 dst->mirrors = src->mirrors;
3990 dst->odp_actions = &dst->odp_actions_buf;
3991 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
3992 sizeof dst->odp_actions_stub);
3993 ofpbuf_put(dst->odp_actions, ofpbuf_data(src->odp_actions),
3994 ofpbuf_size(src->odp_actions));
3997 static struct skb_priority_to_dscp *
3998 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4000 struct skb_priority_to_dscp *pdscp;
4003 hash = hash_int(skb_priority, 0);
4004 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4005 if (pdscp->skb_priority == skb_priority) {
4013 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4016 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4017 *dscp = pdscp ? pdscp->dscp : 0;
4018 return pdscp != NULL;
4022 clear_skb_priorities(struct xport *xport)
4024 struct skb_priority_to_dscp *pdscp, *next;
4026 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4027 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4033 actions_output_to_local_port(const struct xlate_ctx *ctx)
4035 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4036 const struct nlattr *a;
4039 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(ctx->xout->odp_actions),
4040 ofpbuf_size(ctx->xout->odp_actions)) {
4041 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4042 && nl_attr_get_odp_port(a) == local_odp_port) {
4049 #if defined(__linux__)
4050 /* Returns the maximum number of packets that the Linux kernel is willing to
4051 * queue up internally to certain kinds of software-implemented ports, or the
4052 * default (and rarely modified) value if it cannot be determined. */
4054 netdev_max_backlog(void)
4056 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4057 static int max_backlog = 1000; /* The normal default value. */
4059 if (ovsthread_once_start(&once)) {
4060 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4064 stream = fopen(filename, "r");
4066 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4068 if (fscanf(stream, "%d", &n) != 1) {
4069 VLOG_WARN("%s: read error", filename);
4070 } else if (n <= 100) {
4071 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4077 ovsthread_once_done(&once);
4079 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4085 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4088 count_output_actions(const struct ofpbuf *odp_actions)
4090 const struct nlattr *a;
4094 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(odp_actions),
4095 ofpbuf_size(odp_actions)) {
4096 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4102 #endif /* defined(__linux__) */
4104 /* Returns true if 'odp_actions' contains more output actions than the datapath
4105 * can reliably handle in one go. On Linux, this is the value of the
4106 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4107 * packets that the kernel is willing to queue up for processing while the
4108 * datapath is processing a set of actions. */
4110 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4113 return (ofpbuf_size(odp_actions) / NL_A_U32_SIZE > netdev_max_backlog()
4114 && count_output_actions(odp_actions) > netdev_max_backlog());
4116 /* OSes other than Linux might have similar limits, but we don't know how
4117 * to determine them.*/
4122 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
4123 * into datapath actions in 'odp_actions', using 'ctx'.
4125 * The caller must take responsibility for eventually freeing 'xout', with
4126 * xlate_out_uninit(). */
4128 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4130 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4131 struct flow_wildcards *wc = &xout->wc;
4132 struct flow *flow = &xin->flow;
4133 struct rule_dpif *rule = NULL;
4135 const struct rule_actions *actions = NULL;
4136 enum slow_path_reason special;
4137 const struct ofpact *ofpacts;
4138 struct xport *in_port;
4139 struct flow orig_flow;
4140 struct xlate_ctx ctx;
4145 COVERAGE_INC(xlate_actions);
4147 /* Flow initialization rules:
4148 * - 'base_flow' must match the kernel's view of the packet at the
4149 * time that action processing starts. 'flow' represents any
4150 * transformations we wish to make through actions.
4151 * - By default 'base_flow' and 'flow' are the same since the input
4152 * packet matches the output before any actions are applied.
4153 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4154 * of the received packet as seen by the kernel. If we later output
4155 * to another device without any modifications this will cause us to
4156 * insert a new tag since the original one was stripped off by the
4158 * - Tunnel metadata as received is retained in 'flow'. This allows
4159 * tunnel metadata matching also in later tables.
4160 * Since a kernel action for setting the tunnel metadata will only be
4161 * generated with actual tunnel output, changing the tunnel metadata
4162 * values in 'flow' (such as tun_id) will only have effect with a later
4163 * tunnel output action.
4164 * - Tunnel 'base_flow' is completely cleared since that is what the
4165 * kernel does. If we wish to maintain the original values an action
4166 * needs to be generated. */
4171 ctx.xout->has_learn = false;
4172 ctx.xout->has_normal = false;
4173 ctx.xout->has_fin_timeout = false;
4174 ctx.xout->nf_output_iface = NF_OUT_DROP;
4175 ctx.xout->mirrors = 0;
4177 xout->odp_actions = xin->odp_actions;
4178 if (!xout->odp_actions) {
4179 xout->odp_actions = &xout->odp_actions_buf;
4180 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4181 sizeof xout->odp_actions_stub);
4183 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4185 ctx.xbridge = xbridge_lookup(xcfg, xin->ofproto);
4190 ctx.rule = xin->rule;
4192 ctx.base_flow = *flow;
4193 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4194 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4196 flow_wildcards_init_catchall(wc);
4197 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
4198 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
4199 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
4200 if (is_ip_any(flow)) {
4201 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4203 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
4205 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
4206 if (ctx.xbridge->netflow) {
4207 netflow_mask_wc(flow, wc);
4212 ctx.in_group = false;
4213 ctx.orig_skb_priority = flow->skb_priority;
4216 ctx.use_recirc = false;
4217 ctx.was_mpls = false;
4219 if (!xin->ofpacts && !ctx.rule) {
4220 ctx.table_id = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
4221 !xin->skip_wildcards ? wc : NULL,
4222 &rule, ctx.xin->xcache != NULL,
4223 ctx.xin->resubmit_stats);
4224 if (ctx.xin->resubmit_stats) {
4225 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
4227 if (ctx.xin->xcache) {
4228 struct xc_entry *entry;
4230 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4231 entry->u.rule = rule;
4235 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4238 ofpacts = xin->ofpacts;
4239 ofpacts_len = xin->ofpacts_len;
4240 } else if (ctx.rule) {
4241 actions = rule_dpif_get_actions(ctx.rule);
4242 ofpacts = actions->ofpacts;
4243 ofpacts_len = actions->ofpacts_len;
4248 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
4249 ofpbuf_use_stub(&ctx.action_set,
4250 ctx.action_set_stub, sizeof ctx.action_set_stub);
4252 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4253 /* Do this conditionally because the copy is expensive enough that it
4254 * shows up in profiles. */
4258 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
4259 switch (ctx.xbridge->frag) {
4260 case OFPC_FRAG_NORMAL:
4261 /* We must pretend that transport ports are unavailable. */
4262 flow->tp_src = ctx.base_flow.tp_src = htons(0);
4263 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
4266 case OFPC_FRAG_DROP:
4269 case OFPC_FRAG_REASM:
4272 case OFPC_FRAG_NX_MATCH:
4273 /* Nothing to do. */
4276 case OFPC_INVALID_TTL_TO_CONTROLLER:
4281 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
4282 if (in_port && in_port->is_tunnel) {
4283 if (ctx.xin->resubmit_stats) {
4284 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4286 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4289 if (ctx.xin->xcache) {
4290 struct xc_entry *entry;
4292 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4293 entry->u.dev.rx = netdev_ref(in_port->netdev);
4294 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4298 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
4300 ctx.xout->slow |= special;
4302 size_t sample_actions_len;
4304 if (flow->in_port.ofp_port
4305 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
4306 flow->in_port.ofp_port,
4308 ctx.base_flow.vlan_tci = 0;
4311 add_sflow_action(&ctx);
4312 add_ipfix_action(&ctx);
4313 sample_actions_len = ofpbuf_size(ctx.xout->odp_actions);
4315 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4316 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4318 /* We've let OFPP_NORMAL and the learning action look at the
4319 * packet, so drop it now if forwarding is disabled. */
4320 if (in_port && (!xport_stp_forward_state(in_port) ||
4321 !xport_rstp_forward_state(in_port))) {
4322 ofpbuf_set_size(ctx.xout->odp_actions, sample_actions_len);
4326 if (ofpbuf_size(&ctx.action_set)) {
4327 xlate_action_set(&ctx);
4330 if (ctx.xbridge->has_in_band
4331 && in_band_must_output_to_local_port(flow)
4332 && !actions_output_to_local_port(&ctx)) {
4333 compose_output_action(&ctx, OFPP_LOCAL);
4336 fix_sflow_action(&ctx);
4338 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4339 add_mirror_actions(&ctx, &orig_flow);
4343 if (nl_attr_oversized(ofpbuf_size(ctx.xout->odp_actions))) {
4344 /* These datapath actions are too big for a Netlink attribute, so we
4345 * can't hand them to the kernel directly. dpif_execute() can execute
4346 * them one by one with help, so just mark the result as SLOW_ACTION to
4347 * prevent the flow from being installed. */
4348 COVERAGE_INC(xlate_actions_oversize);
4349 ctx.xout->slow |= SLOW_ACTION;
4350 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
4351 COVERAGE_INC(xlate_actions_too_many_output);
4352 ctx.xout->slow |= SLOW_ACTION;
4355 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4356 if (ctx.xin->resubmit_stats) {
4357 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
4358 ctx.xin->resubmit_stats->n_packets,
4359 ctx.xin->resubmit_stats->n_bytes);
4361 if (ctx.xin->xcache) {
4362 struct xc_entry *entry;
4364 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
4365 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
4366 entry->u.mirror.mirrors = xout->mirrors;
4370 if (ctx.xbridge->netflow) {
4371 /* Only update netflow if we don't have controller flow. We don't
4372 * report NetFlow expiration messages for such facets because they
4373 * are just part of the control logic for the network, not real
4375 if (ofpacts_len == 0
4376 || ofpacts->type != OFPACT_CONTROLLER
4377 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
4378 if (ctx.xin->resubmit_stats) {
4379 netflow_flow_update(ctx.xbridge->netflow, flow,
4380 xout->nf_output_iface,
4381 ctx.xin->resubmit_stats);
4383 if (ctx.xin->xcache) {
4384 struct xc_entry *entry;
4386 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
4387 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
4388 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
4389 entry->u.nf.iface = xout->nf_output_iface;
4394 ofpbuf_uninit(&ctx.stack);
4395 ofpbuf_uninit(&ctx.action_set);
4397 /* Clear the metadata and register wildcard masks, because we won't
4398 * use non-header fields as part of the cache. */
4399 flow_wildcards_clear_non_packet_fields(wc);
4401 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
4402 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
4403 * these fields. The datapath interface, on the other hand, represents
4404 * them with just 8 bits each. This means that if the high 8 bits of the
4405 * masks for these fields somehow become set, then they will get chopped
4406 * off by a round trip through the datapath, and revalidation will spot
4407 * that as an inconsistency and delete the flow. Avoid the problem here by
4408 * making sure that only the low 8 bits of either field can be unwildcarded
4412 wc->masks.tp_src &= htons(UINT8_MAX);
4413 wc->masks.tp_dst &= htons(UINT8_MAX);
4417 /* Sends 'packet' out 'ofport'.
4418 * May modify 'packet'.
4419 * Returns 0 if successful, otherwise a positive errno value. */
4421 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4423 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4424 struct xport *xport;
4425 struct ofpact_output output;
4428 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4429 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4430 flow_extract(packet, NULL, &flow);
4431 flow.in_port.ofp_port = OFPP_NONE;
4433 xport = xport_lookup(xcfg, ofport);
4437 output.port = xport->ofp_port;
4440 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
4441 &output.ofpact, sizeof output,
4445 struct xlate_cache *
4446 xlate_cache_new(void)
4448 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
4450 ofpbuf_init(&xcache->entries, 512);
4454 static struct xc_entry *
4455 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
4457 struct xc_entry *entry;
4459 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
4466 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
4468 if (entry->u.dev.tx) {
4469 netdev_vport_inc_tx(entry->u.dev.tx, stats);
4471 if (entry->u.dev.rx) {
4472 netdev_vport_inc_rx(entry->u.dev.rx, stats);
4474 if (entry->u.dev.bfd) {
4475 bfd_account_rx(entry->u.dev.bfd, stats);
4480 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
4482 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4483 struct xbridge *xbridge;
4484 struct xbundle *xbundle;
4485 struct flow_wildcards wc;
4487 xbridge = xbridge_lookup(xcfg, ofproto);
4492 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
4498 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
4501 /* Push stats and perform side effects of flow translation. */
4503 xlate_push_stats(struct xlate_cache *xcache,
4504 const struct dpif_flow_stats *stats)
4506 struct xc_entry *entry;
4507 struct ofpbuf entries = xcache->entries;
4509 if (!stats->n_packets) {
4513 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4514 switch (entry->type) {
4516 rule_dpif_credit_stats(entry->u.rule, stats);
4519 bond_account(entry->u.bond.bond, entry->u.bond.flow,
4520 entry->u.bond.vid, stats->n_bytes);
4523 xlate_cache_netdev(entry, stats);
4526 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
4527 entry->u.nf.iface, stats);
4530 mirror_update_stats(entry->u.mirror.mbridge,
4531 entry->u.mirror.mirrors,
4532 stats->n_packets, stats->n_bytes);
4535 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
4538 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
4539 entry->u.normal.vlan);
4541 case XC_FIN_TIMEOUT:
4542 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
4543 entry->u.fin.idle, entry->u.fin.hard);
4546 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
4556 xlate_dev_unref(struct xc_entry *entry)
4558 if (entry->u.dev.tx) {
4559 netdev_close(entry->u.dev.tx);
4561 if (entry->u.dev.rx) {
4562 netdev_close(entry->u.dev.rx);
4564 if (entry->u.dev.bfd) {
4565 bfd_unref(entry->u.dev.bfd);
4570 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
4572 netflow_flow_clear(netflow, flow);
4573 netflow_unref(netflow);
4578 xlate_cache_clear(struct xlate_cache *xcache)
4580 struct xc_entry *entry;
4581 struct ofpbuf entries;
4587 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4588 switch (entry->type) {
4590 rule_dpif_unref(entry->u.rule);
4593 free(entry->u.bond.flow);
4594 bond_unref(entry->u.bond.bond);
4597 xlate_dev_unref(entry);
4600 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
4603 mbridge_unref(entry->u.mirror.mbridge);
4606 free(entry->u.learn.fm);
4607 ofpbuf_delete(entry->u.learn.ofpacts);
4610 free(entry->u.normal.flow);
4612 case XC_FIN_TIMEOUT:
4613 /* 'u.fin.rule' is always already held as a XC_RULE, which
4614 * has already released it's reference above. */
4617 group_dpif_unref(entry->u.group.group);
4624 ofpbuf_clear(&xcache->entries);
4628 xlate_cache_delete(struct xlate_cache *xcache)
4630 xlate_cache_clear(xcache);
4631 ofpbuf_uninit(&xcache->entries);