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_INITIALIZER(NULL);
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 size_t count_skb_priorities(const struct xport *);
356 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
359 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
361 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
362 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
363 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
364 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
365 struct rule_dpif *miss_rule,
366 struct rule_dpif *no_packet_in_rule,
367 const struct mac_learning *, struct stp *,
368 struct rstp *, const struct mcast_snooping *,
369 const struct mbridge *,
370 const struct dpif_sflow *,
371 const struct dpif_ipfix *,
372 const struct netflow *, enum ofp_config_flags,
373 bool forward_bpdu, bool has_in_band,
375 bool variable_length_userdata,
376 size_t max_mpls_depth,
377 bool masked_set_action);
378 static void xlate_xbundle_set(struct xbundle *xbundle,
379 enum port_vlan_mode vlan_mode, int vlan,
380 unsigned long *trunks, bool use_priority_tags,
381 const struct bond *bond, const struct lacp *lacp,
383 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
384 const struct netdev *netdev, const struct cfm *cfm,
385 const struct bfd *bfd, int stp_port_no,
386 const struct rstp_port *rstp_port,
387 enum ofputil_port_config config,
388 enum ofputil_port_state state, bool is_tunnel,
390 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
391 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
392 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
393 static void xlate_xbridge_copy(struct xbridge *);
394 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
395 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
397 static void xlate_xcfg_free(struct xlate_cfg *);
401 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
403 list_init(&xbridge->xbundles);
404 hmap_init(&xbridge->xports);
405 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
406 hash_pointer(xbridge->ofproto, 0));
410 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
412 list_init(&xbundle->xports);
413 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
414 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
415 hash_pointer(xbundle->ofbundle, 0));
419 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
421 hmap_init(&xport->skb_priorities);
422 hmap_insert(&xcfg->xports, &xport->hmap_node,
423 hash_pointer(xport->ofport, 0));
424 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
425 hash_ofp_port(xport->ofp_port));
429 xlate_xbridge_set(struct xbridge *xbridge,
431 struct rule_dpif *miss_rule,
432 struct rule_dpif *no_packet_in_rule,
433 const struct mac_learning *ml, struct stp *stp,
434 struct rstp *rstp, const struct mcast_snooping *ms,
435 const struct mbridge *mbridge,
436 const struct dpif_sflow *sflow,
437 const struct dpif_ipfix *ipfix,
438 const struct netflow *netflow, enum ofp_config_flags frag,
439 bool forward_bpdu, bool has_in_band,
441 bool variable_length_userdata,
442 size_t max_mpls_depth,
443 bool masked_set_action)
445 if (xbridge->ml != ml) {
446 mac_learning_unref(xbridge->ml);
447 xbridge->ml = mac_learning_ref(ml);
450 if (xbridge->ms != ms) {
451 mcast_snooping_unref(xbridge->ms);
452 xbridge->ms = mcast_snooping_ref(ms);
455 if (xbridge->mbridge != mbridge) {
456 mbridge_unref(xbridge->mbridge);
457 xbridge->mbridge = mbridge_ref(mbridge);
460 if (xbridge->sflow != sflow) {
461 dpif_sflow_unref(xbridge->sflow);
462 xbridge->sflow = dpif_sflow_ref(sflow);
465 if (xbridge->ipfix != ipfix) {
466 dpif_ipfix_unref(xbridge->ipfix);
467 xbridge->ipfix = dpif_ipfix_ref(ipfix);
470 if (xbridge->stp != stp) {
471 stp_unref(xbridge->stp);
472 xbridge->stp = stp_ref(stp);
475 if (xbridge->rstp != rstp) {
476 rstp_unref(xbridge->rstp);
477 xbridge->rstp = rstp_ref(rstp);
480 if (xbridge->netflow != netflow) {
481 netflow_unref(xbridge->netflow);
482 xbridge->netflow = netflow_ref(netflow);
485 xbridge->dpif = dpif;
486 xbridge->forward_bpdu = forward_bpdu;
487 xbridge->has_in_band = has_in_band;
488 xbridge->frag = frag;
489 xbridge->miss_rule = miss_rule;
490 xbridge->no_packet_in_rule = no_packet_in_rule;
491 xbridge->enable_recirc = enable_recirc;
492 xbridge->variable_length_userdata = variable_length_userdata;
493 xbridge->max_mpls_depth = max_mpls_depth;
494 xbridge->masked_set_action = masked_set_action;
498 xlate_xbundle_set(struct xbundle *xbundle,
499 enum port_vlan_mode vlan_mode, int vlan,
500 unsigned long *trunks, bool use_priority_tags,
501 const struct bond *bond, const struct lacp *lacp,
504 ovs_assert(xbundle->xbridge);
506 xbundle->vlan_mode = vlan_mode;
507 xbundle->vlan = vlan;
508 xbundle->trunks = trunks;
509 xbundle->use_priority_tags = use_priority_tags;
510 xbundle->floodable = floodable;
512 if (xbundle->bond != bond) {
513 bond_unref(xbundle->bond);
514 xbundle->bond = bond_ref(bond);
517 if (xbundle->lacp != lacp) {
518 lacp_unref(xbundle->lacp);
519 xbundle->lacp = lacp_ref(lacp);
524 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
525 const struct netdev *netdev, const struct cfm *cfm,
526 const struct bfd *bfd, int stp_port_no,
527 const struct rstp_port* rstp_port,
528 enum ofputil_port_config config, enum ofputil_port_state state,
529 bool is_tunnel, bool may_enable)
531 xport->config = config;
532 xport->state = state;
533 xport->stp_port_no = stp_port_no;
534 xport->is_tunnel = is_tunnel;
535 xport->may_enable = may_enable;
536 xport->odp_port = odp_port;
538 if (xport->rstp_port != rstp_port) {
539 rstp_port_unref(xport->rstp_port);
540 xport->rstp_port = rstp_port_ref(rstp_port);
543 if (xport->cfm != cfm) {
544 cfm_unref(xport->cfm);
545 xport->cfm = cfm_ref(cfm);
548 if (xport->bfd != bfd) {
549 bfd_unref(xport->bfd);
550 xport->bfd = bfd_ref(bfd);
553 if (xport->netdev != netdev) {
554 netdev_close(xport->netdev);
555 xport->netdev = netdev_ref(netdev);
560 xlate_xbridge_copy(struct xbridge *xbridge)
562 struct xbundle *xbundle;
564 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
565 new_xbridge->ofproto = xbridge->ofproto;
566 new_xbridge->name = xstrdup(xbridge->name);
567 xlate_xbridge_init(new_xcfg, new_xbridge);
569 xlate_xbridge_set(new_xbridge,
570 xbridge->dpif, xbridge->miss_rule,
571 xbridge->no_packet_in_rule, xbridge->ml, xbridge->stp,
572 xbridge->rstp, xbridge->ms, xbridge->mbridge,
573 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
574 xbridge->frag, xbridge->forward_bpdu,
575 xbridge->has_in_band, xbridge->enable_recirc,
576 xbridge->variable_length_userdata,
577 xbridge->max_mpls_depth, xbridge->masked_set_action);
578 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
579 xlate_xbundle_copy(new_xbridge, xbundle);
582 /* Copy xports which are not part of a xbundle */
583 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
584 if (!xport->xbundle) {
585 xlate_xport_copy(new_xbridge, NULL, xport);
591 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
594 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
595 new_xbundle->ofbundle = xbundle->ofbundle;
596 new_xbundle->xbridge = xbridge;
597 new_xbundle->name = xstrdup(xbundle->name);
598 xlate_xbundle_init(new_xcfg, new_xbundle);
600 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
601 xbundle->vlan, xbundle->trunks,
602 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
604 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
605 xlate_xport_copy(xbridge, new_xbundle, xport);
610 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
613 struct skb_priority_to_dscp *pdscp, *new_pdscp;
614 struct xport *new_xport = xzalloc(sizeof *xport);
615 new_xport->ofport = xport->ofport;
616 new_xport->ofp_port = xport->ofp_port;
617 new_xport->xbridge = xbridge;
618 xlate_xport_init(new_xcfg, new_xport);
620 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
621 xport->bfd, xport->stp_port_no, xport->rstp_port,
622 xport->config, xport->state, xport->is_tunnel,
626 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
628 new_xport->peer = peer;
629 new_xport->peer->peer = new_xport;
634 new_xport->xbundle = xbundle;
635 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
638 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
639 new_pdscp = xmalloc(sizeof *pdscp);
640 new_pdscp->skb_priority = pdscp->skb_priority;
641 new_pdscp->dscp = pdscp->dscp;
642 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
643 hash_int(new_pdscp->skb_priority, 0));
647 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
648 * configuration in xcfgp.
650 * This needs to be called after editing the xlate configuration.
652 * Functions that edit the new xlate configuration are
653 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
659 * edit_xlate_configuration();
661 * xlate_txn_commit(); */
663 xlate_txn_commit(void)
665 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
667 ovsrcu_set(&xcfgp, new_xcfg);
668 ovsrcu_postpone(xlate_xcfg_free, xcfg);
673 /* Copies the current xlate configuration in xcfgp to new_xcfg.
675 * This needs to be called prior to editing the xlate configuration. */
677 xlate_txn_start(void)
679 struct xbridge *xbridge;
680 struct xlate_cfg *xcfg;
682 ovs_assert(!new_xcfg);
684 new_xcfg = xmalloc(sizeof *new_xcfg);
685 hmap_init(&new_xcfg->xbridges);
686 hmap_init(&new_xcfg->xbundles);
687 hmap_init(&new_xcfg->xports);
689 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
694 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
695 xlate_xbridge_copy(xbridge);
701 xlate_xcfg_free(struct xlate_cfg *xcfg)
703 struct xbridge *xbridge, *next_xbridge;
709 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
710 xlate_xbridge_remove(xcfg, xbridge);
713 hmap_destroy(&xcfg->xbridges);
714 hmap_destroy(&xcfg->xbundles);
715 hmap_destroy(&xcfg->xports);
720 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
721 struct dpif *dpif, struct rule_dpif *miss_rule,
722 struct rule_dpif *no_packet_in_rule,
723 const struct mac_learning *ml, struct stp *stp,
724 struct rstp *rstp, const struct mcast_snooping *ms,
725 const struct mbridge *mbridge,
726 const struct dpif_sflow *sflow,
727 const struct dpif_ipfix *ipfix,
728 const struct netflow *netflow, enum ofp_config_flags frag,
729 bool forward_bpdu, bool has_in_band, bool enable_recirc,
730 bool variable_length_userdata, size_t max_mpls_depth,
731 bool masked_set_action)
733 struct xbridge *xbridge;
735 ovs_assert(new_xcfg);
737 xbridge = xbridge_lookup(new_xcfg, ofproto);
739 xbridge = xzalloc(sizeof *xbridge);
740 xbridge->ofproto = ofproto;
742 xlate_xbridge_init(new_xcfg, xbridge);
746 xbridge->name = xstrdup(name);
748 xlate_xbridge_set(xbridge, dpif, miss_rule, no_packet_in_rule, ml, stp,
749 rstp, ms, mbridge, sflow, ipfix, netflow, frag,
750 forward_bpdu, has_in_band, enable_recirc,
751 variable_length_userdata, max_mpls_depth,
756 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
758 struct xbundle *xbundle, *next_xbundle;
759 struct xport *xport, *next_xport;
765 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
766 xlate_xport_remove(xcfg, xport);
769 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
770 xlate_xbundle_remove(xcfg, xbundle);
773 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
774 mac_learning_unref(xbridge->ml);
775 mcast_snooping_unref(xbridge->ms);
776 mbridge_unref(xbridge->mbridge);
777 dpif_sflow_unref(xbridge->sflow);
778 dpif_ipfix_unref(xbridge->ipfix);
779 stp_unref(xbridge->stp);
780 rstp_unref(xbridge->rstp);
781 hmap_destroy(&xbridge->xports);
787 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
789 struct xbridge *xbridge;
791 ovs_assert(new_xcfg);
793 xbridge = xbridge_lookup(new_xcfg, ofproto);
794 xlate_xbridge_remove(new_xcfg, xbridge);
798 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
799 const char *name, enum port_vlan_mode vlan_mode, int vlan,
800 unsigned long *trunks, bool use_priority_tags,
801 const struct bond *bond, const struct lacp *lacp,
804 struct xbundle *xbundle;
806 ovs_assert(new_xcfg);
808 xbundle = xbundle_lookup(new_xcfg, ofbundle);
810 xbundle = xzalloc(sizeof *xbundle);
811 xbundle->ofbundle = ofbundle;
812 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
814 xlate_xbundle_init(new_xcfg, xbundle);
818 xbundle->name = xstrdup(name);
820 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
821 use_priority_tags, bond, lacp, floodable);
825 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
827 struct xport *xport, *next;
833 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
834 list_remove(&xport->bundle_node);
835 xport->xbundle = NULL;
838 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
839 list_remove(&xbundle->list_node);
840 bond_unref(xbundle->bond);
841 lacp_unref(xbundle->lacp);
847 xlate_bundle_remove(struct ofbundle *ofbundle)
849 struct xbundle *xbundle;
851 ovs_assert(new_xcfg);
853 xbundle = xbundle_lookup(new_xcfg, ofbundle);
854 xlate_xbundle_remove(new_xcfg, xbundle);
858 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
859 struct ofport_dpif *ofport, ofp_port_t ofp_port,
860 odp_port_t odp_port, const struct netdev *netdev,
861 const struct cfm *cfm, const struct bfd *bfd,
862 struct ofport_dpif *peer, int stp_port_no,
863 const struct rstp_port *rstp_port,
864 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
865 enum ofputil_port_config config,
866 enum ofputil_port_state state, bool is_tunnel,
872 ovs_assert(new_xcfg);
874 xport = xport_lookup(new_xcfg, ofport);
876 xport = xzalloc(sizeof *xport);
877 xport->ofport = ofport;
878 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
879 xport->ofp_port = ofp_port;
881 xlate_xport_init(new_xcfg, xport);
884 ovs_assert(xport->ofp_port == ofp_port);
886 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, stp_port_no,
887 rstp_port, config, state, is_tunnel, may_enable);
890 xport->peer->peer = NULL;
892 xport->peer = xport_lookup(new_xcfg, peer);
894 xport->peer->peer = xport;
897 if (xport->xbundle) {
898 list_remove(&xport->bundle_node);
900 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
901 if (xport->xbundle) {
902 list_insert(&xport->xbundle->xports, &xport->bundle_node);
905 clear_skb_priorities(xport);
906 for (i = 0; i < n_qdscp; i++) {
907 struct skb_priority_to_dscp *pdscp;
908 uint32_t skb_priority;
910 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
915 pdscp = xmalloc(sizeof *pdscp);
916 pdscp->skb_priority = skb_priority;
917 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
918 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
919 hash_int(pdscp->skb_priority, 0));
924 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
931 xport->peer->peer = NULL;
935 if (xport->xbundle) {
936 list_remove(&xport->bundle_node);
939 clear_skb_priorities(xport);
940 hmap_destroy(&xport->skb_priorities);
942 hmap_remove(&xcfg->xports, &xport->hmap_node);
943 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
945 netdev_close(xport->netdev);
946 rstp_port_unref(xport->rstp_port);
947 cfm_unref(xport->cfm);
948 bfd_unref(xport->bfd);
953 xlate_ofport_remove(struct ofport_dpif *ofport)
957 ovs_assert(new_xcfg);
959 xport = xport_lookup(new_xcfg, ofport);
960 xlate_xport_remove(new_xcfg, xport);
963 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
964 * returns the corresponding struct xport, or NULL if none is found. */
965 static struct xport *
966 xlate_lookup_xport(const struct dpif_backer *backer, const struct flow *flow)
968 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
970 return xport_lookup(xcfg, tnl_port_should_receive(flow)
971 ? tnl_port_receive(flow)
972 : odp_port_to_ofport(backer, flow->in_port.odp_port));
975 static struct ofproto_dpif *
976 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
977 ofp_port_t *ofp_in_port, const struct xport **xportp)
979 const struct xport *xport;
981 *xportp = xport = xlate_lookup_xport(backer, flow);
985 *ofp_in_port = xport->ofp_port;
987 return xport->xbridge->ofproto;
993 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
994 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
995 struct ofproto_dpif *
996 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
997 ofp_port_t *ofp_in_port)
999 const struct xport *xport;
1001 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1004 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1005 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1006 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1007 * handles for those protocols if they're enabled. Caller may use the returned
1008 * pointers until quiescing, for longer term use additional references must
1011 * '*ofp_in_port' is set to OFPP_NONE if 'flow''s in_port does not exist.
1013 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport.
1016 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1017 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1018 struct dpif_sflow **sflow, struct netflow **netflow,
1019 ofp_port_t *ofp_in_port)
1021 struct ofproto_dpif *ofproto;
1022 const struct xport *xport;
1024 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1026 if (ofp_in_port && !xport) {
1027 *ofp_in_port = OFPP_NONE;
1035 *ofprotop = ofproto;
1039 *ipfix = xport->xbridge->ipfix;
1043 *sflow = xport->xbridge->sflow;
1047 *netflow = xport->xbridge->netflow;
1052 static struct xbridge *
1053 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1055 struct hmap *xbridges;
1056 struct xbridge *xbridge;
1058 if (!ofproto || !xcfg) {
1062 xbridges = &xcfg->xbridges;
1064 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1066 if (xbridge->ofproto == ofproto) {
1073 static struct xbundle *
1074 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1076 struct hmap *xbundles;
1077 struct xbundle *xbundle;
1079 if (!ofbundle || !xcfg) {
1083 xbundles = &xcfg->xbundles;
1085 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1087 if (xbundle->ofbundle == ofbundle) {
1094 static struct xport *
1095 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1097 struct hmap *xports;
1098 struct xport *xport;
1100 if (!ofport || !xcfg) {
1104 xports = &xcfg->xports;
1106 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1108 if (xport->ofport == ofport) {
1115 static struct stp_port *
1116 xport_get_stp_port(const struct xport *xport)
1118 return xport->xbridge->stp && xport->stp_port_no != -1
1119 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1124 xport_stp_learn_state(const struct xport *xport)
1126 struct stp_port *sp = xport_get_stp_port(xport);
1127 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
1131 xport_stp_forward_state(const struct xport *xport)
1133 struct stp_port *sp = xport_get_stp_port(xport);
1134 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
1138 xport_stp_should_forward_bpdu(const struct xport *xport)
1140 struct stp_port *sp = xport_get_stp_port(xport);
1141 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1144 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1145 * were used to make the determination.*/
1147 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1149 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1150 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1151 return is_stp(flow);
1155 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1157 struct stp_port *sp = xport_get_stp_port(xport);
1158 struct ofpbuf payload = *packet;
1159 struct eth_header *eth = ofpbuf_data(&payload);
1161 /* Sink packets on ports that have STP disabled when the bridge has
1163 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1167 /* Trim off padding on payload. */
1168 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1169 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1172 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1173 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
1177 static enum rstp_state
1178 xport_get_rstp_port_state(const struct xport *xport)
1180 return xport->rstp_port
1181 ? rstp_port_get_state(xport->rstp_port)
1186 xport_rstp_learn_state(const struct xport *xport)
1188 return rstp_learn_in_state(xport_get_rstp_port_state(xport));
1192 xport_rstp_forward_state(const struct xport *xport)
1194 return rstp_forward_in_state(xport_get_rstp_port_state(xport));
1198 xport_rstp_should_manage_bpdu(const struct xport *xport)
1200 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1204 rstp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
1206 struct ofpbuf payload = *packet;
1207 struct eth_header *eth = ofpbuf_data(&payload);
1209 /* Sink packets on ports that have no RSTP. */
1210 if (!xport->rstp_port) {
1214 /* Trim off padding on payload. */
1215 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1216 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1219 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1220 rstp_port_received_bpdu(xport->rstp_port, ofpbuf_data(&payload),
1221 ofpbuf_size(&payload));
1225 static struct xport *
1226 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1228 struct xport *xport;
1230 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1232 if (xport->ofp_port == ofp_port) {
1240 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1242 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1243 return xport ? xport->odp_port : ODPP_NONE;
1247 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1249 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1250 return xport && xport->may_enable;
1253 static struct ofputil_bucket *
1254 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1258 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1260 struct group_dpif *group;
1262 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1263 struct ofputil_bucket *bucket;
1265 bucket = group_first_live_bucket(ctx, group, depth);
1266 group_dpif_unref(group);
1267 return bucket == NULL;
1273 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1276 bucket_is_alive(const struct xlate_ctx *ctx,
1277 struct ofputil_bucket *bucket, int depth)
1279 if (depth >= MAX_LIVENESS_RECURSION) {
1280 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1282 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1283 MAX_LIVENESS_RECURSION);
1287 return (!ofputil_bucket_has_liveness(bucket)
1288 || (bucket->watch_port != OFPP_ANY
1289 && odp_port_is_alive(ctx, bucket->watch_port))
1290 || (bucket->watch_group != OFPG_ANY
1291 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1294 static struct ofputil_bucket *
1295 group_first_live_bucket(const struct xlate_ctx *ctx,
1296 const struct group_dpif *group, int depth)
1298 struct ofputil_bucket *bucket;
1299 const struct list *buckets;
1301 group_dpif_get_buckets(group, &buckets);
1302 LIST_FOR_EACH (bucket, list_node, buckets) {
1303 if (bucket_is_alive(ctx, bucket, depth)) {
1311 static struct ofputil_bucket *
1312 group_best_live_bucket(const struct xlate_ctx *ctx,
1313 const struct group_dpif *group,
1316 struct ofputil_bucket *best_bucket = NULL;
1317 uint32_t best_score = 0;
1320 struct ofputil_bucket *bucket;
1321 const struct list *buckets;
1323 group_dpif_get_buckets(group, &buckets);
1324 LIST_FOR_EACH (bucket, list_node, buckets) {
1325 if (bucket_is_alive(ctx, bucket, 0)) {
1326 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1327 if (score >= best_score) {
1328 best_bucket = bucket;
1339 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1341 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1342 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1346 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1348 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1351 static mirror_mask_t
1352 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1354 return xbundle != &ofpp_none_bundle
1355 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1359 static mirror_mask_t
1360 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1362 return xbundle != &ofpp_none_bundle
1363 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1367 static mirror_mask_t
1368 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1370 return xbundle != &ofpp_none_bundle
1371 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1375 static struct xbundle *
1376 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1377 bool warn, struct xport **in_xportp)
1379 struct xport *xport;
1381 /* Find the port and bundle for the received packet. */
1382 xport = get_ofp_port(xbridge, in_port);
1386 if (xport && xport->xbundle) {
1387 return xport->xbundle;
1390 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1391 * which a controller may use as the ingress port for traffic that
1392 * it is sourcing. */
1393 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1394 return &ofpp_none_bundle;
1397 /* Odd. A few possible reasons here:
1399 * - We deleted a port but there are still a few packets queued up
1402 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1403 * we don't know about.
1405 * - The ofproto client didn't configure the port as part of a bundle.
1406 * This is particularly likely to happen if a packet was received on the
1407 * port after it was created, but before the client had a chance to
1408 * configure its bundle.
1411 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1413 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1414 "port %"PRIu16, xbridge->name, in_port);
1420 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1422 const struct xbridge *xbridge = ctx->xbridge;
1423 mirror_mask_t mirrors;
1424 struct xbundle *in_xbundle;
1428 mirrors = ctx->xout->mirrors;
1429 ctx->xout->mirrors = 0;
1431 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1432 ctx->xin->packet != NULL, NULL);
1436 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1438 /* Drop frames on bundles reserved for mirroring. */
1439 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1440 if (ctx->xin->packet != NULL) {
1441 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1442 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1443 "%s, which is reserved exclusively for mirroring",
1444 ctx->xbridge->name, in_xbundle->name);
1446 ofpbuf_clear(ctx->xout->odp_actions);
1451 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1452 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1455 vlan = input_vid_to_vlan(in_xbundle, vid);
1461 /* Restore the original packet before adding the mirror actions. */
1462 ctx->xin->flow = *orig_flow;
1465 mirror_mask_t dup_mirrors;
1466 struct ofbundle *out;
1467 unsigned long *vlans;
1472 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1473 &vlans, &dup_mirrors, &out, &out_vlan);
1474 ovs_assert(has_mirror);
1477 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1479 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1482 if (!vlan_mirrored) {
1483 mirrors = zero_rightmost_1bit(mirrors);
1487 mirrors &= ~dup_mirrors;
1488 ctx->xout->mirrors |= dup_mirrors;
1490 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1491 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1493 output_normal(ctx, out_xbundle, vlan);
1495 } else if (vlan != out_vlan
1496 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1497 struct xbundle *xbundle;
1499 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1500 if (xbundle_includes_vlan(xbundle, out_vlan)
1501 && !xbundle_mirror_out(xbridge, xbundle)) {
1502 output_normal(ctx, xbundle, out_vlan);
1509 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1510 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1511 * the bundle on which the packet was received, returns the VLAN to which the
1514 * Both 'vid' and the return value are in the range 0...4095. */
1516 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1518 switch (in_xbundle->vlan_mode) {
1519 case PORT_VLAN_ACCESS:
1520 return in_xbundle->vlan;
1523 case PORT_VLAN_TRUNK:
1526 case PORT_VLAN_NATIVE_UNTAGGED:
1527 case PORT_VLAN_NATIVE_TAGGED:
1528 return vid ? vid : in_xbundle->vlan;
1535 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1536 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1539 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1540 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1543 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1545 /* Allow any VID on the OFPP_NONE port. */
1546 if (in_xbundle == &ofpp_none_bundle) {
1550 switch (in_xbundle->vlan_mode) {
1551 case PORT_VLAN_ACCESS:
1554 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1555 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1556 "packet received on port %s configured as VLAN "
1557 "%"PRIu16" access port", vid, in_xbundle->name,
1564 case PORT_VLAN_NATIVE_UNTAGGED:
1565 case PORT_VLAN_NATIVE_TAGGED:
1567 /* Port must always carry its native VLAN. */
1571 case PORT_VLAN_TRUNK:
1572 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1574 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1575 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1576 "received on port %s not configured for trunking "
1577 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1589 /* Given 'vlan', the VLAN that a packet belongs to, and
1590 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1591 * that should be included in the 802.1Q header. (If the return value is 0,
1592 * then the 802.1Q header should only be included in the packet if there is a
1595 * Both 'vlan' and the return value are in the range 0...4095. */
1597 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1599 switch (out_xbundle->vlan_mode) {
1600 case PORT_VLAN_ACCESS:
1603 case PORT_VLAN_TRUNK:
1604 case PORT_VLAN_NATIVE_TAGGED:
1607 case PORT_VLAN_NATIVE_UNTAGGED:
1608 return vlan == out_xbundle->vlan ? 0 : vlan;
1616 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1619 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1621 ovs_be16 tci, old_tci;
1622 struct xport *xport;
1624 vid = output_vlan_to_vid(out_xbundle, vlan);
1625 if (list_is_empty(&out_xbundle->xports)) {
1626 /* Partially configured bundle with no slaves. Drop the packet. */
1628 } else if (!out_xbundle->bond) {
1629 ctx->use_recirc = false;
1630 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1633 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1634 struct flow_wildcards *wc = &ctx->xout->wc;
1635 struct xlate_recirc *xr = &ctx->recirc;
1636 struct ofport_dpif *ofport;
1638 if (ctx->xbridge->enable_recirc) {
1639 ctx->use_recirc = bond_may_recirc(
1640 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1642 if (ctx->use_recirc) {
1643 /* Only TCP mode uses recirculation. */
1644 xr->hash_alg = OVS_HASH_ALG_L4;
1645 bond_update_post_recirc_rules(out_xbundle->bond, false);
1647 /* Recirculation does not require unmasking hash fields. */
1652 ofport = bond_choose_output_slave(out_xbundle->bond,
1653 &ctx->xin->flow, wc, vid);
1654 xport = xport_lookup(xcfg, ofport);
1657 /* No slaves enabled, so drop packet. */
1661 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1662 * accounting for this bond. */
1663 if (!ctx->use_recirc) {
1664 if (ctx->xin->resubmit_stats) {
1665 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1666 ctx->xin->resubmit_stats->n_bytes);
1668 if (ctx->xin->xcache) {
1669 struct xc_entry *entry;
1672 flow = &ctx->xin->flow;
1673 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1674 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1675 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1676 entry->u.bond.vid = vid;
1681 old_tci = *flow_tci;
1683 if (tci || out_xbundle->use_priority_tags) {
1684 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1686 tci |= htons(VLAN_CFI);
1691 compose_output_action(ctx, xport->ofp_port);
1692 *flow_tci = old_tci;
1695 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1696 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1697 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1699 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1701 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1705 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1706 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1710 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1711 if (flow->nw_proto == ARP_OP_REPLY) {
1713 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1714 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1715 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1717 return flow->nw_src == flow->nw_dst;
1723 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1724 * dropped. Returns true if they may be forwarded, false if they should be
1727 * 'in_port' must be the xport that corresponds to flow->in_port.
1728 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1730 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1731 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1732 * checked by input_vid_is_valid().
1734 * May also add tags to '*tags', although the current implementation only does
1735 * so in one special case.
1738 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1741 struct xbundle *in_xbundle = in_port->xbundle;
1742 const struct xbridge *xbridge = ctx->xbridge;
1743 struct flow *flow = &ctx->xin->flow;
1745 /* Drop frames for reserved multicast addresses
1746 * only if forward_bpdu option is absent. */
1747 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1748 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1752 if (in_xbundle->bond) {
1753 struct mac_entry *mac;
1755 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1761 xlate_report(ctx, "bonding refused admissibility, dropping");
1764 case BV_DROP_IF_MOVED:
1765 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1766 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1767 if (mac && mac->port.p != in_xbundle->ofbundle &&
1768 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1769 || mac_entry_is_grat_arp_locked(mac))) {
1770 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1771 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1775 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1783 /* Checks whether a MAC learning update is necessary for MAC learning table
1784 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1787 * Most packets processed through the MAC learning table do not actually
1788 * change it in any way. This function requires only a read lock on the MAC
1789 * learning table, so it is much cheaper in this common case.
1791 * Keep the code here synchronized with that in update_learning_table__()
1794 is_mac_learning_update_needed(const struct mac_learning *ml,
1795 const struct flow *flow,
1796 struct flow_wildcards *wc,
1797 int vlan, struct xbundle *in_xbundle)
1798 OVS_REQ_RDLOCK(ml->rwlock)
1800 struct mac_entry *mac;
1802 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1806 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1807 if (!mac || mac_entry_age(ml, mac)) {
1811 if (is_gratuitous_arp(flow, wc)) {
1812 /* We don't want to learn from gratuitous ARP packets that are
1813 * reflected back over bond slaves so we lock the learning table. */
1814 if (!in_xbundle->bond) {
1816 } else if (mac_entry_is_grat_arp_locked(mac)) {
1821 return mac->port.p != in_xbundle->ofbundle;
1825 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1826 * received on 'in_xbundle' in 'vlan'.
1828 * This code repeats all the checks in is_mac_learning_update_needed() because
1829 * the lock was released between there and here and thus the MAC learning state
1830 * could have changed.
1832 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1835 update_learning_table__(const struct xbridge *xbridge,
1836 const struct flow *flow, struct flow_wildcards *wc,
1837 int vlan, struct xbundle *in_xbundle)
1838 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1840 struct mac_entry *mac;
1842 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1846 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1847 if (is_gratuitous_arp(flow, wc)) {
1848 /* We don't want to learn from gratuitous ARP packets that are
1849 * reflected back over bond slaves so we lock the learning table. */
1850 if (!in_xbundle->bond) {
1851 mac_entry_set_grat_arp_lock(mac);
1852 } else if (mac_entry_is_grat_arp_locked(mac)) {
1857 if (mac->port.p != in_xbundle->ofbundle) {
1858 /* The log messages here could actually be useful in debugging,
1859 * so keep the rate limit relatively high. */
1860 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1862 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1863 "on port %s in VLAN %d",
1864 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1865 in_xbundle->name, vlan);
1867 mac->port.p = in_xbundle->ofbundle;
1868 mac_learning_changed(xbridge->ml);
1873 update_learning_table(const struct xbridge *xbridge,
1874 const struct flow *flow, struct flow_wildcards *wc,
1875 int vlan, struct xbundle *in_xbundle)
1879 /* Don't learn the OFPP_NONE port. */
1880 if (in_xbundle == &ofpp_none_bundle) {
1884 /* First try the common case: no change to MAC learning table. */
1885 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1886 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1888 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1891 /* Slow path: MAC learning table might need an update. */
1892 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1893 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1894 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1898 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1899 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
1901 update_mcast_snooping_table__(const struct xbridge *xbridge,
1902 const struct flow *flow,
1903 struct mcast_snooping *ms,
1904 ovs_be32 ip4, int vlan,
1905 struct xbundle *in_xbundle)
1906 OVS_REQ_WRLOCK(ms->rwlock)
1908 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
1910 switch (ntohs(flow->tp_src)) {
1911 case IGMP_HOST_MEMBERSHIP_REPORT:
1912 case IGMPV2_HOST_MEMBERSHIP_REPORT:
1913 if (mcast_snooping_add_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1914 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
1915 IP_FMT" is on port %s in VLAN %d",
1916 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1919 case IGMP_HOST_LEAVE_MESSAGE:
1920 if (mcast_snooping_leave_group(ms, ip4, vlan, in_xbundle->ofbundle)) {
1921 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
1922 IP_FMT" is on port %s in VLAN %d",
1923 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
1926 case IGMP_HOST_MEMBERSHIP_QUERY:
1927 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
1928 in_xbundle->ofbundle)) {
1929 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
1930 IP_FMT" is on port %s in VLAN %d",
1931 xbridge->name, IP_ARGS(flow->nw_src),
1932 in_xbundle->name, vlan);
1938 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
1939 * was received on 'in_xbundle' in 'vlan'. */
1941 update_mcast_snooping_table(const struct xbridge *xbridge,
1942 const struct flow *flow, int vlan,
1943 struct xbundle *in_xbundle)
1945 struct mcast_snooping *ms = xbridge->ms;
1946 struct xlate_cfg *xcfg;
1947 struct xbundle *mcast_xbundle;
1948 struct mcast_fport_bundle *fport;
1950 /* Don't learn the OFPP_NONE port. */
1951 if (in_xbundle == &ofpp_none_bundle) {
1955 /* Don't learn from flood ports */
1956 mcast_xbundle = NULL;
1957 ovs_rwlock_wrlock(&ms->rwlock);
1958 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1959 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
1960 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
1961 if (mcast_xbundle == in_xbundle) {
1966 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
1967 update_mcast_snooping_table__(xbridge, flow, ms, flow->igmp_group_ip4,
1970 ovs_rwlock_unlock(&ms->rwlock);
1973 /* send the packet to ports having the multicast group learned */
1975 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
1976 struct mcast_snooping *ms OVS_UNUSED,
1977 struct mcast_group *grp,
1978 struct xbundle *in_xbundle, uint16_t vlan)
1979 OVS_REQ_RDLOCK(ms->rwlock)
1981 struct xlate_cfg *xcfg;
1982 struct mcast_group_bundle *b;
1983 struct xbundle *mcast_xbundle;
1985 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1986 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
1987 mcast_xbundle = xbundle_lookup(xcfg, b->port);
1988 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
1989 xlate_report(ctx, "forwarding to mcast group port");
1990 output_normal(ctx, mcast_xbundle, vlan);
1991 } else if (!mcast_xbundle) {
1992 xlate_report(ctx, "mcast group port is unknown, dropping");
1994 xlate_report(ctx, "mcast group port is input port, dropping");
1999 /* send the packet to ports connected to multicast routers */
2001 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2002 struct mcast_snooping *ms,
2003 struct xbundle *in_xbundle, uint16_t vlan)
2004 OVS_REQ_RDLOCK(ms->rwlock)
2006 struct xlate_cfg *xcfg;
2007 struct mcast_mrouter_bundle *mrouter;
2008 struct xbundle *mcast_xbundle;
2010 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2011 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2012 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2013 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2014 xlate_report(ctx, "forwarding to mcast router port");
2015 output_normal(ctx, mcast_xbundle, vlan);
2016 } else if (!mcast_xbundle) {
2017 xlate_report(ctx, "mcast router port is unknown, dropping");
2019 xlate_report(ctx, "mcast router port is input port, dropping");
2024 /* send the packet to ports flagged to be flooded */
2026 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2027 struct mcast_snooping *ms,
2028 struct xbundle *in_xbundle, uint16_t vlan)
2029 OVS_REQ_RDLOCK(ms->rwlock)
2031 struct xlate_cfg *xcfg;
2032 struct mcast_fport_bundle *fport;
2033 struct xbundle *mcast_xbundle;
2035 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2036 LIST_FOR_EACH(fport, fport_node, &ms->fport_list) {
2037 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2038 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2039 xlate_report(ctx, "forwarding to mcast flood port");
2040 output_normal(ctx, mcast_xbundle, vlan);
2041 } else if (!mcast_xbundle) {
2042 xlate_report(ctx, "mcast flood port is unknown, dropping");
2044 xlate_report(ctx, "mcast flood port is input port, dropping");
2050 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2053 struct xbundle *xbundle;
2055 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2056 if (xbundle != in_xbundle
2057 && xbundle_includes_vlan(xbundle, vlan)
2058 && xbundle->floodable
2059 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2060 output_normal(ctx, xbundle, vlan);
2063 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2067 xlate_normal(struct xlate_ctx *ctx)
2069 struct flow_wildcards *wc = &ctx->xout->wc;
2070 struct flow *flow = &ctx->xin->flow;
2071 struct xbundle *in_xbundle;
2072 struct xport *in_port;
2073 struct mac_entry *mac;
2078 ctx->xout->has_normal = true;
2080 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2081 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2082 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2084 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2085 ctx->xin->packet != NULL, &in_port);
2087 xlate_report(ctx, "no input bundle, dropping");
2091 /* Drop malformed frames. */
2092 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2093 !(flow->vlan_tci & htons(VLAN_CFI))) {
2094 if (ctx->xin->packet != NULL) {
2095 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2096 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2097 "VLAN tag received on port %s",
2098 ctx->xbridge->name, in_xbundle->name);
2100 xlate_report(ctx, "partial VLAN tag, dropping");
2104 /* Drop frames on bundles reserved for mirroring. */
2105 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2106 if (ctx->xin->packet != NULL) {
2107 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2108 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2109 "%s, which is reserved exclusively for mirroring",
2110 ctx->xbridge->name, in_xbundle->name);
2112 xlate_report(ctx, "input port is mirror output port, dropping");
2117 vid = vlan_tci_to_vid(flow->vlan_tci);
2118 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2119 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2122 vlan = input_vid_to_vlan(in_xbundle, vid);
2124 /* Check other admissibility requirements. */
2125 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2129 /* Learn source MAC. */
2130 if (ctx->xin->may_learn) {
2131 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2133 if (ctx->xin->xcache) {
2134 struct xc_entry *entry;
2136 /* Save enough info to update mac learning table later. */
2137 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2138 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2139 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2140 entry->u.normal.vlan = vlan;
2143 /* Determine output bundle. */
2144 if (mcast_snooping_enabled(ctx->xbridge->ms)
2145 && !eth_addr_is_broadcast(flow->dl_dst)
2146 && eth_addr_is_multicast(flow->dl_dst)
2147 && flow->dl_type == htons(ETH_TYPE_IP)) {
2148 struct mcast_snooping *ms = ctx->xbridge->ms;
2149 struct mcast_group *grp;
2151 if (flow->nw_proto == IPPROTO_IGMP) {
2152 if (ctx->xin->may_learn) {
2153 if (mcast_snooping_is_membership(flow->tp_src) ||
2154 mcast_snooping_is_query(flow->tp_src)) {
2155 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2160 if (mcast_snooping_is_membership(flow->tp_src)) {
2161 ovs_rwlock_rdlock(&ms->rwlock);
2162 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2163 ovs_rwlock_unlock(&ms->rwlock);
2165 xlate_report(ctx, "multicast traffic, flooding");
2166 xlate_normal_flood(ctx, in_xbundle, vlan);
2170 if (ip_is_local_multicast(flow->nw_dst)) {
2171 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2172 * address in the 224.0.0.x range which are not IGMP must
2173 * be forwarded on all ports */
2174 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2175 xlate_normal_flood(ctx, in_xbundle, vlan);
2180 /* forwarding to group base ports */
2181 ovs_rwlock_rdlock(&ms->rwlock);
2182 grp = mcast_snooping_lookup(ms, flow->nw_dst, vlan);
2184 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2185 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2186 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2188 if (mcast_snooping_flood_unreg(ms)) {
2189 xlate_report(ctx, "unregistered multicast, flooding");
2190 xlate_normal_flood(ctx, in_xbundle, vlan);
2192 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2193 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2196 ovs_rwlock_unlock(&ms->rwlock);
2198 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2199 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2200 mac_port = mac ? mac->port.p : NULL;
2201 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2204 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2205 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2206 if (mac_xbundle && mac_xbundle != in_xbundle) {
2207 xlate_report(ctx, "forwarding to learned port");
2208 output_normal(ctx, mac_xbundle, vlan);
2209 } else if (!mac_xbundle) {
2210 xlate_report(ctx, "learned port is unknown, dropping");
2212 xlate_report(ctx, "learned port is input port, dropping");
2215 xlate_report(ctx, "no learned MAC for destination, flooding");
2216 xlate_normal_flood(ctx, in_xbundle, vlan);
2221 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
2222 * the number of packets out of UINT32_MAX to sample. The given
2223 * cookie is passed back in the callback for each sampled packet.
2226 compose_sample_action(const struct xbridge *xbridge,
2227 struct ofpbuf *odp_actions,
2228 const struct flow *flow,
2229 const uint32_t probability,
2230 const union user_action_cookie *cookie,
2231 const size_t cookie_size,
2232 const odp_port_t tunnel_out_port)
2234 size_t sample_offset, actions_offset;
2235 odp_port_t odp_port;
2239 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
2241 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2243 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
2245 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
2246 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
2247 flow_hash_5tuple(flow, 0));
2248 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2249 tunnel_out_port, odp_actions);
2251 nl_msg_end_nested(odp_actions, actions_offset);
2252 nl_msg_end_nested(odp_actions, sample_offset);
2253 return cookie_offset;
2257 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
2258 odp_port_t odp_port, unsigned int n_outputs,
2259 union user_action_cookie *cookie)
2263 cookie->type = USER_ACTION_COOKIE_SFLOW;
2264 cookie->sflow.vlan_tci = vlan_tci;
2266 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2267 * port information") for the interpretation of cookie->output. */
2268 switch (n_outputs) {
2270 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2271 cookie->sflow.output = 0x40000000 | 256;
2275 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
2277 cookie->sflow.output = ifindex;
2282 /* 0x80000000 means "multiple output ports. */
2283 cookie->sflow.output = 0x80000000 | n_outputs;
2288 /* Compose SAMPLE action for sFlow bridge sampling. */
2290 compose_sflow_action(const struct xbridge *xbridge,
2291 struct ofpbuf *odp_actions,
2292 const struct flow *flow,
2293 odp_port_t odp_port)
2295 uint32_t probability;
2296 union user_action_cookie cookie;
2298 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
2302 probability = dpif_sflow_get_probability(xbridge->sflow);
2303 compose_sflow_cookie(xbridge, htons(0), odp_port,
2304 odp_port == ODPP_NONE ? 0 : 1, &cookie);
2306 return compose_sample_action(xbridge, odp_actions, flow, probability,
2307 &cookie, sizeof cookie.sflow, ODPP_NONE);
2311 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
2312 uint32_t obs_domain_id, uint32_t obs_point_id,
2313 union user_action_cookie *cookie)
2315 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
2316 cookie->flow_sample.probability = probability;
2317 cookie->flow_sample.collector_set_id = collector_set_id;
2318 cookie->flow_sample.obs_domain_id = obs_domain_id;
2319 cookie->flow_sample.obs_point_id = obs_point_id;
2323 compose_ipfix_cookie(union user_action_cookie *cookie,
2324 odp_port_t output_odp_port)
2326 cookie->type = USER_ACTION_COOKIE_IPFIX;
2327 cookie->ipfix.output_odp_port = output_odp_port;
2330 /* Compose SAMPLE action for IPFIX bridge sampling. */
2332 compose_ipfix_action(const struct xbridge *xbridge,
2333 struct ofpbuf *odp_actions,
2334 const struct flow *flow,
2335 odp_port_t output_odp_port)
2337 uint32_t probability;
2338 union user_action_cookie cookie;
2339 odp_port_t tunnel_out_port = ODPP_NONE;
2341 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
2345 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2347 if (output_odp_port == ODPP_NONE &&
2348 !dpif_ipfix_get_bridge_exporter_input_sampling(xbridge->ipfix)) {
2352 /* For output case, output_odp_port is valid*/
2353 if (output_odp_port != ODPP_NONE) {
2354 if (!dpif_ipfix_get_bridge_exporter_output_sampling(xbridge->ipfix)) {
2357 /* If tunnel sampling is enabled, put an additional option attribute:
2358 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2360 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(xbridge->ipfix) &&
2361 dpif_ipfix_get_tunnel_port(xbridge->ipfix, output_odp_port) ) {
2362 tunnel_out_port = output_odp_port;
2366 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
2367 compose_ipfix_cookie(&cookie, output_odp_port);
2369 compose_sample_action(xbridge, odp_actions, flow, probability,
2370 &cookie, sizeof cookie.ipfix, tunnel_out_port);
2373 /* SAMPLE action for sFlow must be first action in any given list of
2374 * actions. At this point we do not have all information required to
2375 * build it. So try to build sample action as complete as possible. */
2377 add_sflow_action(struct xlate_ctx *ctx)
2379 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
2380 ctx->xout->odp_actions,
2381 &ctx->xin->flow, ODPP_NONE);
2382 ctx->sflow_odp_port = 0;
2383 ctx->sflow_n_outputs = 0;
2386 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
2387 * of actions, eventually after the SAMPLE action for sFlow. */
2389 add_ipfix_action(struct xlate_ctx *ctx)
2391 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2392 &ctx->xin->flow, ODPP_NONE);
2396 add_ipfix_output_action(struct xlate_ctx *ctx, odp_port_t port)
2398 compose_ipfix_action(ctx->xbridge, ctx->xout->odp_actions,
2399 &ctx->xin->flow, port);
2402 /* Fix SAMPLE action according to data collected while composing ODP actions.
2403 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
2404 * USERSPACE action's user-cookie which is required for sflow. */
2406 fix_sflow_action(struct xlate_ctx *ctx)
2408 const struct flow *base = &ctx->base_flow;
2409 union user_action_cookie *cookie;
2411 if (!ctx->user_cookie_offset) {
2415 cookie = ofpbuf_at(ctx->xout->odp_actions, ctx->user_cookie_offset,
2416 sizeof cookie->sflow);
2417 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2419 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
2420 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
2423 static enum slow_path_reason
2424 process_special(struct xlate_ctx *ctx, const struct flow *flow,
2425 const struct xport *xport, const struct ofpbuf *packet)
2427 struct flow_wildcards *wc = &ctx->xout->wc;
2428 const struct xbridge *xbridge = ctx->xbridge;
2432 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2434 cfm_process_heartbeat(xport->cfm, packet);
2437 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2439 bfd_process_packet(xport->bfd, flow, packet);
2440 /* If POLL received, immediately sends FINAL back. */
2441 if (bfd_should_send_packet(xport->bfd)) {
2442 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2446 } else if (xport->xbundle && xport->xbundle->lacp
2447 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2449 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2452 } else if ((xbridge->stp || xbridge->rstp) &&
2453 stp_should_process_flow(flow, wc)) {
2456 ? stp_process_packet(xport, packet)
2457 : rstp_process_packet(xport, packet);
2466 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2469 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2470 struct flow_wildcards *wc = &ctx->xout->wc;
2471 struct flow *flow = &ctx->xin->flow;
2472 ovs_be16 flow_vlan_tci;
2473 uint32_t flow_pkt_mark;
2474 uint8_t flow_nw_tos;
2475 odp_port_t out_port, odp_port;
2478 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2479 * before traversing a patch port. */
2480 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 27);
2483 xlate_report(ctx, "Nonexistent output port");
2485 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2486 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2488 } else if (check_stp) {
2489 if (is_stp(&ctx->base_flow)) {
2490 if (!xport_stp_should_forward_bpdu(xport) &&
2491 !xport_rstp_should_manage_bpdu(xport)) {
2492 if (ctx->xbridge->stp != NULL) {
2493 xlate_report(ctx, "STP not in listening state, "
2494 "skipping bpdu output");
2495 } else if (ctx->xbridge->rstp != NULL) {
2496 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2497 "skipping bpdu output");
2501 } else if (!xport_stp_forward_state(xport) ||
2502 !xport_rstp_forward_state(xport)) {
2503 if (ctx->xbridge->stp != NULL) {
2504 xlate_report(ctx, "STP not in forwarding state, "
2506 } else if (ctx->xbridge->rstp != NULL) {
2507 xlate_report(ctx, "RSTP not in forwarding state, "
2514 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2515 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2520 const struct xport *peer = xport->peer;
2521 struct flow old_flow = ctx->xin->flow;
2522 enum slow_path_reason special;
2524 ctx->xbridge = peer->xbridge;
2525 flow->in_port.ofp_port = peer->ofp_port;
2526 flow->metadata = htonll(0);
2527 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2528 memset(flow->regs, 0, sizeof flow->regs);
2530 special = process_special(ctx, &ctx->xin->flow, peer,
2533 ctx->xout->slow |= special;
2534 } else if (may_receive(peer, ctx)) {
2535 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2536 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2538 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2539 * the learning action look at the packet, then drop it. */
2540 struct flow old_base_flow = ctx->base_flow;
2541 size_t old_size = ofpbuf_size(ctx->xout->odp_actions);
2542 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2543 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2544 ctx->xout->mirrors = old_mirrors;
2545 ctx->base_flow = old_base_flow;
2546 ofpbuf_set_size(ctx->xout->odp_actions, old_size);
2550 ctx->xin->flow = old_flow;
2551 ctx->xbridge = xport->xbridge;
2553 if (ctx->xin->resubmit_stats) {
2554 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2555 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2557 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2560 if (ctx->xin->xcache) {
2561 struct xc_entry *entry;
2563 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2564 entry->u.dev.tx = netdev_ref(xport->netdev);
2565 entry->u.dev.rx = netdev_ref(peer->netdev);
2566 entry->u.dev.bfd = bfd_ref(peer->bfd);
2571 flow_vlan_tci = flow->vlan_tci;
2572 flow_pkt_mark = flow->pkt_mark;
2573 flow_nw_tos = flow->nw_tos;
2575 if (count_skb_priorities(xport)) {
2576 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2577 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2578 wc->masks.nw_tos |= IP_DSCP_MASK;
2579 flow->nw_tos &= ~IP_DSCP_MASK;
2580 flow->nw_tos |= dscp;
2584 if (xport->is_tunnel) {
2585 /* Save tunnel metadata so that changes made due to
2586 * the Logical (tunnel) Port are not visible for any further
2587 * matches, while explicit set actions on tunnel metadata are.
2589 struct flow_tnl flow_tnl = flow->tunnel;
2590 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
2591 if (odp_port == ODPP_NONE) {
2592 xlate_report(ctx, "Tunneling decided against output");
2593 goto out; /* restore flow_nw_tos */
2595 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2596 xlate_report(ctx, "Not tunneling to our own address");
2597 goto out; /* restore flow_nw_tos */
2599 if (ctx->xin->resubmit_stats) {
2600 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2602 if (ctx->xin->xcache) {
2603 struct xc_entry *entry;
2605 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2606 entry->u.dev.tx = netdev_ref(xport->netdev);
2608 out_port = odp_port;
2609 commit_odp_tunnel_action(flow, &ctx->base_flow,
2610 ctx->xout->odp_actions);
2611 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2613 odp_port = xport->odp_port;
2614 out_port = odp_port;
2615 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2616 ofp_port_t vlandev_port;
2618 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2619 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2620 ofp_port, flow->vlan_tci);
2621 if (vlandev_port != ofp_port) {
2622 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2623 flow->vlan_tci = htons(0);
2628 if (out_port != ODPP_NONE) {
2629 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2630 ctx->xout->odp_actions,
2632 ctx->xbridge->masked_set_action);
2634 if (ctx->use_recirc) {
2635 struct ovs_action_hash *act_hash;
2636 struct xlate_recirc *xr = &ctx->recirc;
2639 act_hash = nl_msg_put_unspec_uninit(ctx->xout->odp_actions,
2640 OVS_ACTION_ATTR_HASH,
2642 act_hash->hash_alg = xr->hash_alg;
2643 act_hash->hash_basis = xr->hash_basis;
2645 /* Recirc action. */
2646 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
2649 add_ipfix_output_action(ctx, out_port);
2650 nl_msg_put_odp_port(ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
2654 ctx->sflow_odp_port = odp_port;
2655 ctx->sflow_n_outputs++;
2656 ctx->xout->nf_output_iface = ofp_port;
2661 flow->vlan_tci = flow_vlan_tci;
2662 flow->pkt_mark = flow_pkt_mark;
2663 flow->nw_tos = flow_nw_tos;
2667 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
2669 compose_output_action__(ctx, ofp_port, true);
2673 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
2675 struct rule_dpif *old_rule = ctx->rule;
2676 const struct rule_actions *actions;
2678 if (ctx->xin->resubmit_stats) {
2679 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2685 actions = rule_dpif_get_actions(rule);
2686 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2687 ctx->rule = old_rule;
2692 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2694 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2696 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2697 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2698 MAX_RESUBMIT_RECURSION);
2699 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2700 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2701 } else if (ofpbuf_size(ctx->xout->odp_actions) > UINT16_MAX) {
2702 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2703 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2704 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2713 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2714 bool may_packet_in, bool honor_table_miss)
2716 if (xlate_resubmit_resource_check(ctx)) {
2717 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
2718 bool skip_wildcards = ctx->xin->skip_wildcards;
2719 uint8_t old_table_id = ctx->table_id;
2720 struct rule_dpif *rule;
2721 enum rule_dpif_lookup_verdict verdict;
2722 enum ofputil_port_config config = 0;
2724 ctx->table_id = table_id;
2726 /* Look up a flow with 'in_port' as the input port. Then restore the
2727 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2728 * have surprising behavior). */
2729 ctx->xin->flow.in_port.ofp_port = in_port;
2730 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2733 ? &ctx->xout->wc : NULL,
2735 &ctx->table_id, &rule,
2736 ctx->xin->xcache != NULL,
2737 ctx->xin->resubmit_stats);
2738 ctx->xin->flow.in_port.ofp_port = old_in_port;
2740 if (ctx->xin->resubmit_hook) {
2741 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
2745 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
2747 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
2748 if (may_packet_in) {
2749 struct xport *xport;
2751 xport = get_ofp_port(ctx->xbridge,
2752 ctx->xin->flow.in_port.ofp_port);
2753 config = xport ? xport->config : 0;
2756 /* Fall through to drop */
2757 case RULE_DPIF_LOOKUP_VERDICT_DROP:
2758 config = OFPUTIL_PC_NO_PACKET_IN;
2760 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
2761 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
2762 config = OFPUTIL_PC_NO_PACKET_IN;
2769 choose_miss_rule(config, ctx->xbridge->miss_rule,
2770 ctx->xbridge->no_packet_in_rule, &rule,
2771 ctx->xin->xcache != NULL);
2775 /* Fill in the cache entry here instead of xlate_recursively
2776 * to make the reference counting more explicit. We take a
2777 * reference in the lookups above if we are going to cache the
2779 if (ctx->xin->xcache) {
2780 struct xc_entry *entry;
2782 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2783 entry->u.rule = rule;
2785 xlate_recursively(ctx, rule);
2788 ctx->table_id = old_table_id;
2796 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
2797 struct ofputil_bucket *bucket)
2799 if (ctx->xin->resubmit_stats) {
2800 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
2802 if (ctx->xin->xcache) {
2803 struct xc_entry *entry;
2805 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
2806 entry->u.group.group = group_dpif_ref(group);
2807 entry->u.group.bucket = bucket;
2812 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
2814 uint64_t action_list_stub[1024 / 8];
2815 struct ofpbuf action_list, action_set;
2817 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2818 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2820 ofpacts_execute_action_set(&action_list, &action_set);
2822 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2825 ofpbuf_uninit(&action_set);
2826 ofpbuf_uninit(&action_list);
2830 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2832 struct ofputil_bucket *bucket;
2833 const struct list *buckets;
2834 struct flow old_flow = ctx->xin->flow;
2836 group_dpif_get_buckets(group, &buckets);
2838 LIST_FOR_EACH (bucket, list_node, buckets) {
2839 xlate_group_bucket(ctx, bucket);
2840 /* Roll back flow to previous state.
2841 * This is equivalent to cloning the packet for each bucket.
2843 * As a side effect any subsequently applied actions will
2844 * also effectively be applied to a clone of the packet taken
2845 * just before applying the all or indirect group. */
2846 ctx->xin->flow = old_flow;
2848 xlate_group_stats(ctx, group, NULL);
2852 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2854 struct ofputil_bucket *bucket;
2856 bucket = group_first_live_bucket(ctx, group, 0);
2858 xlate_group_bucket(ctx, bucket);
2859 xlate_group_stats(ctx, group, bucket);
2864 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2866 struct flow_wildcards *wc = &ctx->xout->wc;
2867 struct ofputil_bucket *bucket;
2870 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
2871 bucket = group_best_live_bucket(ctx, group, basis);
2873 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2874 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2875 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2876 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2877 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2878 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2879 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2880 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2881 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2883 xlate_group_bucket(ctx, bucket);
2884 xlate_group_stats(ctx, group, bucket);
2889 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2891 ctx->in_group = true;
2893 switch (group_dpif_get_type(group)) {
2895 case OFPGT11_INDIRECT:
2896 xlate_all_group(ctx, group);
2898 case OFPGT11_SELECT:
2899 xlate_select_group(ctx, group);
2902 xlate_ff_group(ctx, group);
2907 group_dpif_unref(group);
2909 ctx->in_group = false;
2913 xlate_group_resource_check(struct xlate_ctx *ctx)
2915 if (!xlate_resubmit_resource_check(ctx)) {
2917 } else if (ctx->in_group) {
2918 /* Prevent nested translation of OpenFlow groups.
2920 * OpenFlow allows this restriction. We enforce this restriction only
2921 * because, with the current architecture, we would otherwise have to
2922 * take a possibly recursive read lock on the ofgroup rwlock, which is
2923 * unsafe given that POSIX allows taking a read lock to block if there
2924 * is a thread blocked on taking the write lock. Other solutions
2925 * without this restriction are also possible, but seem unwarranted
2926 * given the current limited use of groups. */
2927 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2929 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2937 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2939 if (xlate_group_resource_check(ctx)) {
2940 struct group_dpif *group;
2943 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2945 xlate_group_action__(ctx, group);
2955 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2956 const struct ofpact_resubmit *resubmit)
2960 bool may_packet_in = false;
2961 bool honor_table_miss = false;
2963 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2964 /* Still allow missed packets to be sent to the controller
2965 * if resubmitting from an internal table. */
2966 may_packet_in = true;
2967 honor_table_miss = true;
2970 in_port = resubmit->in_port;
2971 if (in_port == OFPP_IN_PORT) {
2972 in_port = ctx->xin->flow.in_port.ofp_port;
2975 table_id = resubmit->table_id;
2976 if (table_id == 255) {
2977 table_id = ctx->table_id;
2980 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2985 flood_packets(struct xlate_ctx *ctx, bool all)
2987 const struct xport *xport;
2989 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2990 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2995 compose_output_action__(ctx, xport->ofp_port, false);
2996 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2997 compose_output_action(ctx, xport->ofp_port);
3001 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3005 execute_controller_action(struct xlate_ctx *ctx, int len,
3006 enum ofp_packet_in_reason reason,
3007 uint16_t controller_id)
3009 struct ofproto_packet_in *pin;
3010 struct dpif_packet *packet;
3012 ctx->xout->slow |= SLOW_CONTROLLER;
3013 if (!ctx->xin->packet) {
3017 packet = dpif_packet_clone_from_ofpbuf(ctx->xin->packet);
3019 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3020 ctx->xout->odp_actions,
3022 ctx->xbridge->masked_set_action);
3024 odp_execute_actions(NULL, &packet, 1, false,
3025 ofpbuf_data(ctx->xout->odp_actions),
3026 ofpbuf_size(ctx->xout->odp_actions), NULL);
3028 pin = xmalloc(sizeof *pin);
3029 pin->up.packet_len = ofpbuf_size(&packet->ofpbuf);
3030 pin->up.packet = ofpbuf_steal_data(&packet->ofpbuf);
3031 pin->up.reason = reason;
3032 pin->up.table_id = ctx->table_id;
3033 pin->up.cookie = (ctx->rule
3034 ? rule_dpif_get_flow_cookie(ctx->rule)
3037 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
3039 pin->controller_id = controller_id;
3040 pin->send_len = len;
3041 /* If a rule is a table-miss rule then this is
3042 * a table-miss handled by a table-miss rule.
3044 * Else, if rule is internal and has a controller action,
3045 * the later being implied by the rule being processed here,
3046 * then this is a table-miss handled without a table-miss rule.
3048 * Otherwise this is not a table-miss. */
3049 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3051 if (rule_dpif_is_table_miss(ctx->rule)) {
3052 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3053 } else if (rule_dpif_is_internal(ctx->rule)) {
3054 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3057 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3058 dpif_packet_delete(packet);
3062 compose_recirculate_action(struct xlate_ctx *ctx,
3063 const struct ofpact *ofpacts_base,
3064 const struct ofpact *ofpact_current,
3065 size_t ofpacts_base_len)
3069 unsigned ofpacts_len;
3072 struct ofpbuf ofpacts;
3076 ofpacts_len = ofpacts_base_len -
3077 ((uint8_t *)ofpact_current - (uint8_t *)ofpacts_base);
3080 id = rule_dpif_get_recirc_id(ctx->rule);
3082 /* In the case where ctx has no rule then allocate a recirc id.
3083 * The life-cycle of this recirc id is managed by associating it
3084 * with the internal rule that is created to to handle
3085 * recirculation below.
3087 * The known use-case of this is packet_out which
3088 * translates actions without a rule */
3089 id = ofproto_dpif_alloc_recirc_id(ctx->xbridge->ofproto);
3092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3093 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3097 match_init_catchall(&match);
3098 match_set_recirc_id(&match, id);
3099 ofpbuf_use_const(&ofpacts, ofpact_current, ofpacts_len);
3100 error = ofproto_dpif_add_internal_flow(ctx->xbridge->ofproto, &match,
3101 RECIRC_RULE_PRIORITY,
3102 RECIRC_TIMEOUT, &ofpacts, &rule);
3104 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3105 VLOG_ERR_RL(&rl, "Failed to add post recirculation flow %s",
3106 match_to_string(&match, 0));
3109 /* If ctx has no rule then associate the recirc id, which
3110 * was allocated above, with the internal rule. This allows
3111 * the recirc id to be released when the internal rule times out. */
3113 rule_set_recirc_id(rule, id);
3116 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3117 ctx->xout->odp_actions,
3119 ctx->xbridge->masked_set_action);
3120 nl_msg_put_u32(ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3124 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3126 struct flow_wildcards *wc = &ctx->xout->wc;
3127 struct flow *flow = &ctx->xin->flow;
3130 ovs_assert(eth_type_mpls(mpls->ethertype));
3132 n = flow_count_mpls_labels(flow, wc);
3134 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3135 ctx->xout->odp_actions,
3137 ctx->xbridge->masked_set_action);
3138 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3139 if (ctx->xin->packet != NULL) {
3140 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3141 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3142 "MPLS push action can't be performed as it would "
3143 "have more MPLS LSEs than the %d supported.",
3144 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3148 } else if (n >= ctx->xbridge->max_mpls_depth) {
3149 COVERAGE_INC(xlate_actions_mpls_overflow);
3150 ctx->xout->slow |= SLOW_ACTION;
3153 flow_push_mpls(flow, n, mpls->ethertype, wc);
3157 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3159 struct flow_wildcards *wc = &ctx->xout->wc;
3160 struct flow *flow = &ctx->xin->flow;
3161 int n = flow_count_mpls_labels(flow, wc);
3163 if (flow_pop_mpls(flow, n, eth_type, wc)) {
3164 if (ctx->xbridge->enable_recirc && !eth_type_mpls(eth_type)) {
3165 ctx->was_mpls = true;
3167 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3168 if (ctx->xin->packet != NULL) {
3169 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3170 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3171 "MPLS pop action can't be performed as it has "
3172 "more MPLS LSEs than the %d supported.",
3173 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3176 ofpbuf_clear(ctx->xout->odp_actions);
3181 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3183 struct flow *flow = &ctx->xin->flow;
3185 if (!is_ip_any(flow)) {
3189 ctx->xout->wc.masks.nw_ttl = 0xff;
3190 if (flow->nw_ttl > 1) {
3196 for (i = 0; i < ids->n_controllers; i++) {
3197 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3201 /* Stop processing for current table. */
3207 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3209 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3210 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3211 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3216 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3218 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3219 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3220 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3225 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3227 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3228 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3229 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3234 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3236 struct flow *flow = &ctx->xin->flow;
3237 struct flow_wildcards *wc = &ctx->xout->wc;
3239 if (eth_type_mpls(flow->dl_type)) {
3240 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3242 wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3245 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3248 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3252 /* Stop processing for current table. */
3257 xlate_output_action(struct xlate_ctx *ctx,
3258 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3260 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3262 ctx->xout->nf_output_iface = NF_OUT_DROP;
3266 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
3269 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3270 0, may_packet_in, true);
3276 flood_packets(ctx, false);
3279 flood_packets(ctx, true);
3281 case OFPP_CONTROLLER:
3282 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
3288 if (port != ctx->xin->flow.in_port.ofp_port) {
3289 compose_output_action(ctx, port);
3291 xlate_report(ctx, "skipping output to input port");
3296 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3297 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3298 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3299 ctx->xout->nf_output_iface = prev_nf_output_iface;
3300 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3301 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3302 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3307 xlate_output_reg_action(struct xlate_ctx *ctx,
3308 const struct ofpact_output_reg *or)
3310 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3311 if (port <= UINT16_MAX) {
3312 union mf_subvalue value;
3314 memset(&value, 0xff, sizeof value);
3315 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
3316 xlate_output_action(ctx, u16_to_ofp(port),
3317 or->max_len, false);
3322 xlate_enqueue_action(struct xlate_ctx *ctx,
3323 const struct ofpact_enqueue *enqueue)
3325 ofp_port_t ofp_port = enqueue->port;
3326 uint32_t queue_id = enqueue->queue;
3327 uint32_t flow_priority, priority;
3330 /* Translate queue to priority. */
3331 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3333 /* Fall back to ordinary output action. */
3334 xlate_output_action(ctx, enqueue->port, 0, false);
3338 /* Check output port. */
3339 if (ofp_port == OFPP_IN_PORT) {
3340 ofp_port = ctx->xin->flow.in_port.ofp_port;
3341 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3345 /* Add datapath actions. */
3346 flow_priority = ctx->xin->flow.skb_priority;
3347 ctx->xin->flow.skb_priority = priority;
3348 compose_output_action(ctx, ofp_port);
3349 ctx->xin->flow.skb_priority = flow_priority;
3351 /* Update NetFlow output port. */
3352 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3353 ctx->xout->nf_output_iface = ofp_port;
3354 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3355 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3360 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3362 uint32_t skb_priority;
3364 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3365 ctx->xin->flow.skb_priority = skb_priority;
3367 /* Couldn't translate queue to a priority. Nothing to do. A warning
3368 * has already been logged. */
3373 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3375 const struct xbridge *xbridge = xbridge_;
3386 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3389 port = get_ofp_port(xbridge, ofp_port);
3390 return port ? port->may_enable : false;
3395 xlate_bundle_action(struct xlate_ctx *ctx,
3396 const struct ofpact_bundle *bundle)
3400 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
3402 CONST_CAST(struct xbridge *, ctx->xbridge));
3403 if (bundle->dst.field) {
3404 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
3407 xlate_output_action(ctx, port, 0, false);
3412 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3413 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3415 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3416 if (ctx->xin->may_learn) {
3417 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3422 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3424 ctx->xout->has_learn = true;
3425 learn_mask(learn, &ctx->xout->wc);
3427 if (ctx->xin->xcache) {
3428 struct xc_entry *entry;
3430 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3431 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3432 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3433 entry->u.learn.ofpacts = ofpbuf_new(64);
3434 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3435 entry->u.learn.ofpacts);
3436 } else if (ctx->xin->may_learn) {
3437 uint64_t ofpacts_stub[1024 / 8];
3438 struct ofputil_flow_mod fm;
3439 struct ofpbuf ofpacts;
3441 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3442 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3443 ofpbuf_uninit(&ofpacts);
3448 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3449 uint16_t idle_timeout, uint16_t hard_timeout)
3451 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3452 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3457 xlate_fin_timeout(struct xlate_ctx *ctx,
3458 const struct ofpact_fin_timeout *oft)
3461 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3462 oft->fin_idle_timeout, oft->fin_hard_timeout);
3463 if (ctx->xin->xcache) {
3464 struct xc_entry *entry;
3466 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3467 /* XC_RULE already holds a reference on the rule, none is taken
3469 entry->u.fin.rule = ctx->rule;
3470 entry->u.fin.idle = oft->fin_idle_timeout;
3471 entry->u.fin.hard = oft->fin_hard_timeout;
3477 xlate_sample_action(struct xlate_ctx *ctx,
3478 const struct ofpact_sample *os)
3480 union user_action_cookie cookie;
3481 /* Scale the probability from 16-bit to 32-bit while representing
3482 * the same percentage. */
3483 uint32_t probability = (os->probability << 16) | os->probability;
3485 if (!ctx->xbridge->variable_length_userdata) {
3486 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3488 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3489 "lacks support (needs Linux 3.10+ or kernel module from "
3494 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3495 ctx->xout->odp_actions,
3497 ctx->xbridge->masked_set_action);
3499 compose_flow_sample_cookie(os->probability, os->collector_set_id,
3500 os->obs_domain_id, os->obs_point_id, &cookie);
3501 compose_sample_action(ctx->xbridge, ctx->xout->odp_actions, &ctx->xin->flow,
3502 probability, &cookie, sizeof cookie.flow_sample,
3507 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3509 if (xport->config & (is_stp(&ctx->xin->flow)
3510 ? OFPUTIL_PC_NO_RECV_STP
3511 : OFPUTIL_PC_NO_RECV)) {
3515 /* Only drop packets here if both forwarding and learning are
3516 * disabled. If just learning is enabled, we need to have
3517 * OFPP_NORMAL and the learning action have a look at the packet
3518 * before we can drop it. */
3519 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3520 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3528 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3530 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3531 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
3532 ofpact_pad(&ctx->action_set);
3536 xlate_action_set(struct xlate_ctx *ctx)
3538 uint64_t action_list_stub[1024 / 64];
3539 struct ofpbuf action_list;
3541 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3542 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3543 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
3544 ofpbuf_uninit(&action_list);
3548 ofpact_needs_recirculation_after_mpls(const struct xlate_ctx *ctx,
3549 const struct ofpact *a)
3551 struct flow_wildcards *wc = &ctx->xout->wc;
3552 struct flow *flow = &ctx->xin->flow;
3557 case OFPACT_CONTROLLER:
3558 case OFPACT_STRIP_VLAN:
3559 case OFPACT_SET_VLAN_PCP:
3560 case OFPACT_SET_VLAN_VID:
3561 case OFPACT_ENQUEUE:
3562 case OFPACT_PUSH_VLAN:
3563 case OFPACT_SET_ETH_SRC:
3564 case OFPACT_SET_ETH_DST:
3565 case OFPACT_SET_TUNNEL:
3566 case OFPACT_SET_QUEUE:
3567 case OFPACT_POP_QUEUE:
3568 case OFPACT_POP_MPLS:
3569 case OFPACT_DEC_MPLS_TTL:
3570 case OFPACT_SET_MPLS_TTL:
3571 case OFPACT_SET_MPLS_TC:
3572 case OFPACT_SET_MPLS_LABEL:
3574 case OFPACT_OUTPUT_REG:
3577 case OFPACT_WRITE_METADATA:
3578 case OFPACT_WRITE_ACTIONS:
3579 case OFPACT_CLEAR_ACTIONS:
3583 case OFPACT_SET_IPV4_SRC:
3584 case OFPACT_SET_IPV4_DST:
3585 case OFPACT_SET_IP_DSCP:
3586 case OFPACT_SET_IP_ECN:
3587 case OFPACT_SET_IP_TTL:
3588 case OFPACT_SET_L4_SRC_PORT:
3589 case OFPACT_SET_L4_DST_PORT:
3590 case OFPACT_RESUBMIT:
3591 case OFPACT_STACK_PUSH:
3592 case OFPACT_STACK_POP:
3593 case OFPACT_DEC_TTL:
3594 case OFPACT_MULTIPATH:
3597 case OFPACT_FIN_TIMEOUT:
3598 case OFPACT_GOTO_TABLE:
3601 case OFPACT_REG_MOVE:
3602 return (mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
3603 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
3605 case OFPACT_SET_FIELD:
3606 return mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field);
3608 case OFPACT_PUSH_MPLS:
3609 /* Recirculate if it is an IP packet with a zero ttl. This may
3610 * indicate that the packet was previously MPLS and an MPLS pop action
3611 * converted it to IP. In this case recirculating should reveal the IP
3612 * TTL which is used as the basis for a new MPLS LSE. */
3613 return (!flow_count_mpls_labels(flow, wc)
3614 && flow->nw_ttl == 0
3615 && is_ip_any(flow));
3622 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
3623 struct xlate_ctx *ctx)
3625 struct flow_wildcards *wc = &ctx->xout->wc;
3626 struct flow *flow = &ctx->xin->flow;
3627 const struct ofpact *a;
3629 /* dl_type already in the mask, not set below. */
3631 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3632 struct ofpact_controller *controller;
3633 const struct ofpact_metadata *metadata;
3634 const struct ofpact_set_field *set_field;
3635 const struct mf_field *mf;
3641 if (ctx->was_mpls && ofpact_needs_recirculation_after_mpls(ctx, a)) {
3642 compose_recirculate_action(ctx, ofpacts, a, ofpacts_len);
3648 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
3649 ofpact_get_OUTPUT(a)->max_len, true);
3653 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
3658 case OFPACT_CONTROLLER:
3659 controller = ofpact_get_CONTROLLER(a);
3660 execute_controller_action(ctx, controller->max_len,
3662 controller->controller_id);
3665 case OFPACT_ENQUEUE:
3666 memset(&wc->masks.skb_priority, 0xff,
3667 sizeof wc->masks.skb_priority);
3668 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
3671 case OFPACT_SET_VLAN_VID:
3672 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3673 if (flow->vlan_tci & htons(VLAN_CFI) ||
3674 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
3675 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
3676 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
3681 case OFPACT_SET_VLAN_PCP:
3682 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
3683 if (flow->vlan_tci & htons(VLAN_CFI) ||
3684 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
3685 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
3686 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
3687 << VLAN_PCP_SHIFT) | VLAN_CFI);
3691 case OFPACT_STRIP_VLAN:
3692 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3693 flow->vlan_tci = htons(0);
3696 case OFPACT_PUSH_VLAN:
3697 /* XXX 802.1AD(QinQ) */
3698 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
3699 flow->vlan_tci = htons(VLAN_CFI);
3702 case OFPACT_SET_ETH_SRC:
3703 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
3704 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
3707 case OFPACT_SET_ETH_DST:
3708 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
3709 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
3712 case OFPACT_SET_IPV4_SRC:
3713 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3714 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
3715 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
3719 case OFPACT_SET_IPV4_DST:
3720 if (flow->dl_type == htons(ETH_TYPE_IP)) {
3721 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
3722 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
3726 case OFPACT_SET_IP_DSCP:
3727 if (is_ip_any(flow)) {
3728 wc->masks.nw_tos |= IP_DSCP_MASK;
3729 flow->nw_tos &= ~IP_DSCP_MASK;
3730 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
3734 case OFPACT_SET_IP_ECN:
3735 if (is_ip_any(flow)) {
3736 wc->masks.nw_tos |= IP_ECN_MASK;
3737 flow->nw_tos &= ~IP_ECN_MASK;
3738 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
3742 case OFPACT_SET_IP_TTL:
3743 if (is_ip_any(flow)) {
3744 wc->masks.nw_ttl = 0xff;
3745 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
3749 case OFPACT_SET_L4_SRC_PORT:
3750 if (is_ip_any(flow)) {
3751 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3752 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
3753 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
3757 case OFPACT_SET_L4_DST_PORT:
3758 if (is_ip_any(flow)) {
3759 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3760 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
3761 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
3765 case OFPACT_RESUBMIT:
3766 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
3769 case OFPACT_SET_TUNNEL:
3770 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
3773 case OFPACT_SET_QUEUE:
3774 memset(&wc->masks.skb_priority, 0xff,
3775 sizeof wc->masks.skb_priority);
3776 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
3779 case OFPACT_POP_QUEUE:
3780 memset(&wc->masks.skb_priority, 0xff,
3781 sizeof wc->masks.skb_priority);
3782 flow->skb_priority = ctx->orig_skb_priority;
3785 case OFPACT_REG_MOVE:
3786 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
3789 case OFPACT_SET_FIELD:
3790 set_field = ofpact_get_SET_FIELD(a);
3791 mf = set_field->field;
3793 /* Set field action only ever overwrites packet's outermost
3794 * applicable header fields. Do nothing if no header exists. */
3795 if (mf->id == MFF_VLAN_VID) {
3796 wc->masks.vlan_tci |= htons(VLAN_CFI);
3797 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
3800 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
3801 /* 'dl_type' is already unwildcarded. */
3802 && !eth_type_mpls(flow->dl_type)) {
3806 mf_mask_field_and_prereqs(mf, &wc->masks);
3807 mf_set_flow_value_masked(mf, &set_field->value, &set_field->mask,
3811 case OFPACT_STACK_PUSH:
3812 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
3816 case OFPACT_STACK_POP:
3817 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
3821 case OFPACT_PUSH_MPLS:
3822 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
3825 case OFPACT_POP_MPLS:
3826 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
3829 case OFPACT_SET_MPLS_LABEL:
3830 compose_set_mpls_label_action(
3831 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
3834 case OFPACT_SET_MPLS_TC:
3835 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
3838 case OFPACT_SET_MPLS_TTL:
3839 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
3842 case OFPACT_DEC_MPLS_TTL:
3843 if (compose_dec_mpls_ttl_action(ctx)) {
3848 case OFPACT_DEC_TTL:
3849 wc->masks.nw_ttl = 0xff;
3850 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
3856 /* Nothing to do. */
3859 case OFPACT_MULTIPATH:
3860 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
3864 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
3867 case OFPACT_OUTPUT_REG:
3868 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
3872 xlate_learn_action(ctx, ofpact_get_LEARN(a));
3879 case OFPACT_FIN_TIMEOUT:
3880 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3881 ctx->xout->has_fin_timeout = true;
3882 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
3885 case OFPACT_CLEAR_ACTIONS:
3886 ofpbuf_clear(&ctx->action_set);
3889 case OFPACT_WRITE_ACTIONS:
3890 xlate_write_actions(ctx, a);
3893 case OFPACT_WRITE_METADATA:
3894 metadata = ofpact_get_WRITE_METADATA(a);
3895 flow->metadata &= ~metadata->mask;
3896 flow->metadata |= metadata->metadata & metadata->mask;
3900 /* Not implemented yet. */
3903 case OFPACT_GOTO_TABLE: {
3904 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
3906 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
3907 * than ogt->table_id. This is to allow goto_table actions that
3908 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
3909 * after recirculation. */
3910 ovs_assert(ctx->table_id == TBL_INTERNAL
3911 || ctx->table_id < ogt->table_id);
3912 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3913 ogt->table_id, true, true);
3918 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
3925 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
3926 const struct flow *flow, ofp_port_t in_port,
3927 struct rule_dpif *rule, uint16_t tcp_flags,
3928 const struct ofpbuf *packet)
3930 xin->ofproto = ofproto;
3932 xin->flow.in_port.ofp_port = in_port;
3933 xin->packet = packet;
3934 xin->may_learn = packet != NULL;
3937 xin->ofpacts = NULL;
3938 xin->ofpacts_len = 0;
3939 xin->tcp_flags = tcp_flags;
3940 xin->resubmit_hook = NULL;
3941 xin->report_hook = NULL;
3942 xin->resubmit_stats = NULL;
3943 xin->skip_wildcards = false;
3944 xin->odp_actions = NULL;
3948 xlate_out_uninit(struct xlate_out *xout)
3950 if (xout && xout->odp_actions == &xout->odp_actions_buf) {
3951 ofpbuf_uninit(xout->odp_actions);
3955 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
3956 * into datapath actions, using 'ctx', and discards the datapath actions. */
3958 xlate_actions_for_side_effects(struct xlate_in *xin)
3960 struct xlate_out xout;
3962 xlate_actions(xin, &xout);
3963 xlate_out_uninit(&xout);
3967 xlate_report(struct xlate_ctx *ctx, const char *s)
3969 if (ctx->xin->report_hook) {
3970 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
3975 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
3978 dst->slow = src->slow;
3979 dst->has_learn = src->has_learn;
3980 dst->has_normal = src->has_normal;
3981 dst->has_fin_timeout = src->has_fin_timeout;
3982 dst->nf_output_iface = src->nf_output_iface;
3983 dst->mirrors = src->mirrors;
3985 dst->odp_actions = &dst->odp_actions_buf;
3986 ofpbuf_use_stub(dst->odp_actions, dst->odp_actions_stub,
3987 sizeof dst->odp_actions_stub);
3988 ofpbuf_put(dst->odp_actions, ofpbuf_data(src->odp_actions),
3989 ofpbuf_size(src->odp_actions));
3992 static struct skb_priority_to_dscp *
3993 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
3995 struct skb_priority_to_dscp *pdscp;
3998 hash = hash_int(skb_priority, 0);
3999 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4000 if (pdscp->skb_priority == skb_priority) {
4008 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4011 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4012 *dscp = pdscp ? pdscp->dscp : 0;
4013 return pdscp != NULL;
4017 count_skb_priorities(const struct xport *xport)
4019 return hmap_count(&xport->skb_priorities);
4023 clear_skb_priorities(struct xport *xport)
4025 struct skb_priority_to_dscp *pdscp, *next;
4027 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4028 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4034 actions_output_to_local_port(const struct xlate_ctx *ctx)
4036 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4037 const struct nlattr *a;
4040 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(ctx->xout->odp_actions),
4041 ofpbuf_size(ctx->xout->odp_actions)) {
4042 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4043 && nl_attr_get_odp_port(a) == local_odp_port) {
4050 #if defined(__linux__)
4051 /* Returns the maximum number of packets that the Linux kernel is willing to
4052 * queue up internally to certain kinds of software-implemented ports, or the
4053 * default (and rarely modified) value if it cannot be determined. */
4055 netdev_max_backlog(void)
4057 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4058 static int max_backlog = 1000; /* The normal default value. */
4060 if (ovsthread_once_start(&once)) {
4061 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4065 stream = fopen(filename, "r");
4067 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4069 if (fscanf(stream, "%d", &n) != 1) {
4070 VLOG_WARN("%s: read error", filename);
4071 } else if (n <= 100) {
4072 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4078 ovsthread_once_done(&once);
4080 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4086 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4089 count_output_actions(const struct ofpbuf *odp_actions)
4091 const struct nlattr *a;
4095 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(odp_actions),
4096 ofpbuf_size(odp_actions)) {
4097 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4103 #endif /* defined(__linux__) */
4105 /* Returns true if 'odp_actions' contains more output actions than the datapath
4106 * can reliably handle in one go. On Linux, this is the value of the
4107 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4108 * packets that the kernel is willing to queue up for processing while the
4109 * datapath is processing a set of actions. */
4111 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4114 return (ofpbuf_size(odp_actions) / NL_A_U32_SIZE > netdev_max_backlog()
4115 && count_output_actions(odp_actions) > netdev_max_backlog());
4117 /* OSes other than Linux might have similar limits, but we don't know how
4118 * to determine them.*/
4123 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
4124 * into datapath actions in 'odp_actions', using 'ctx'.
4126 * The caller must take responsibility for eventually freeing 'xout', with
4127 * xlate_out_uninit(). */
4129 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4131 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4132 struct flow_wildcards *wc = &xout->wc;
4133 struct flow *flow = &xin->flow;
4134 struct rule_dpif *rule = NULL;
4136 const struct rule_actions *actions = NULL;
4137 enum slow_path_reason special;
4138 const struct ofpact *ofpacts;
4139 struct xport *in_port;
4140 struct flow orig_flow;
4141 struct xlate_ctx ctx;
4146 COVERAGE_INC(xlate_actions);
4148 /* Flow initialization rules:
4149 * - 'base_flow' must match the kernel's view of the packet at the
4150 * time that action processing starts. 'flow' represents any
4151 * transformations we wish to make through actions.
4152 * - By default 'base_flow' and 'flow' are the same since the input
4153 * packet matches the output before any actions are applied.
4154 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
4155 * of the received packet as seen by the kernel. If we later output
4156 * to another device without any modifications this will cause us to
4157 * insert a new tag since the original one was stripped off by the
4159 * - Tunnel metadata as received is retained in 'flow'. This allows
4160 * tunnel metadata matching also in later tables.
4161 * Since a kernel action for setting the tunnel metadata will only be
4162 * generated with actual tunnel output, changing the tunnel metadata
4163 * values in 'flow' (such as tun_id) will only have effect with a later
4164 * tunnel output action.
4165 * - Tunnel 'base_flow' is completely cleared since that is what the
4166 * kernel does. If we wish to maintain the original values an action
4167 * needs to be generated. */
4172 ctx.xout->has_learn = false;
4173 ctx.xout->has_normal = false;
4174 ctx.xout->has_fin_timeout = false;
4175 ctx.xout->nf_output_iface = NF_OUT_DROP;
4176 ctx.xout->mirrors = 0;
4178 xout->odp_actions = xin->odp_actions;
4179 if (!xout->odp_actions) {
4180 xout->odp_actions = &xout->odp_actions_buf;
4181 ofpbuf_use_stub(xout->odp_actions, xout->odp_actions_stub,
4182 sizeof xout->odp_actions_stub);
4184 ofpbuf_reserve(xout->odp_actions, NL_A_U32_SIZE);
4186 ctx.xbridge = xbridge_lookup(xcfg, xin->ofproto);
4191 ctx.rule = xin->rule;
4193 ctx.base_flow = *flow;
4194 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4195 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
4197 flow_wildcards_init_catchall(wc);
4198 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
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 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
4259 if (in_port && in_port->is_tunnel) {
4260 if (ctx.xin->resubmit_stats) {
4261 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4263 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4266 if (ctx.xin->xcache) {
4267 struct xc_entry *entry;
4269 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4270 entry->u.dev.rx = netdev_ref(in_port->netdev);
4271 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4275 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
4277 ctx.xout->slow |= special;
4279 size_t sample_actions_len;
4281 if (flow->in_port.ofp_port
4282 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
4283 flow->in_port.ofp_port,
4285 ctx.base_flow.vlan_tci = 0;
4288 add_sflow_action(&ctx);
4289 add_ipfix_action(&ctx);
4290 sample_actions_len = ofpbuf_size(ctx.xout->odp_actions);
4292 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
4293 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4295 /* We've let OFPP_NORMAL and the learning action look at the
4296 * packet, so drop it now if forwarding is disabled. */
4297 if (in_port && (!xport_stp_forward_state(in_port) ||
4298 !xport_rstp_forward_state(in_port))) {
4299 ofpbuf_set_size(ctx.xout->odp_actions, sample_actions_len);
4303 if (ofpbuf_size(&ctx.action_set)) {
4304 xlate_action_set(&ctx);
4307 if (ctx.xbridge->has_in_band
4308 && in_band_must_output_to_local_port(flow)
4309 && !actions_output_to_local_port(&ctx)) {
4310 compose_output_action(&ctx, OFPP_LOCAL);
4313 fix_sflow_action(&ctx);
4315 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4316 add_mirror_actions(&ctx, &orig_flow);
4320 if (nl_attr_oversized(ofpbuf_size(ctx.xout->odp_actions))) {
4321 /* These datapath actions are too big for a Netlink attribute, so we
4322 * can't hand them to the kernel directly. dpif_execute() can execute
4323 * them one by one with help, so just mark the result as SLOW_ACTION to
4324 * prevent the flow from being installed. */
4325 COVERAGE_INC(xlate_actions_oversize);
4326 ctx.xout->slow |= SLOW_ACTION;
4327 } else if (too_many_output_actions(ctx.xout->odp_actions)) {
4328 COVERAGE_INC(xlate_actions_too_many_output);
4329 ctx.xout->slow |= SLOW_ACTION;
4332 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
4333 if (ctx.xin->resubmit_stats) {
4334 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
4335 ctx.xin->resubmit_stats->n_packets,
4336 ctx.xin->resubmit_stats->n_bytes);
4338 if (ctx.xin->xcache) {
4339 struct xc_entry *entry;
4341 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
4342 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
4343 entry->u.mirror.mirrors = xout->mirrors;
4347 if (ctx.xbridge->netflow) {
4348 /* Only update netflow if we don't have controller flow. We don't
4349 * report NetFlow expiration messages for such facets because they
4350 * are just part of the control logic for the network, not real
4352 if (ofpacts_len == 0
4353 || ofpacts->type != OFPACT_CONTROLLER
4354 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
4355 if (ctx.xin->resubmit_stats) {
4356 netflow_flow_update(ctx.xbridge->netflow, flow,
4357 xout->nf_output_iface,
4358 ctx.xin->resubmit_stats);
4360 if (ctx.xin->xcache) {
4361 struct xc_entry *entry;
4363 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
4364 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
4365 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
4366 entry->u.nf.iface = xout->nf_output_iface;
4371 ofpbuf_uninit(&ctx.stack);
4372 ofpbuf_uninit(&ctx.action_set);
4374 /* Clear the metadata and register wildcard masks, because we won't
4375 * use non-header fields as part of the cache. */
4376 flow_wildcards_clear_non_packet_fields(wc);
4378 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
4379 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
4380 * these fields. The datapath interface, on the other hand, represents
4381 * them with just 8 bits each. This means that if the high 8 bits of the
4382 * masks for these fields somehow become set, then they will get chopped
4383 * off by a round trip through the datapath, and revalidation will spot
4384 * that as an inconsistency and delete the flow. Avoid the problem here by
4385 * making sure that only the low 8 bits of either field can be unwildcarded
4389 wc->masks.tp_src &= htons(UINT8_MAX);
4390 wc->masks.tp_dst &= htons(UINT8_MAX);
4394 /* Sends 'packet' out 'ofport'.
4395 * May modify 'packet'.
4396 * Returns 0 if successful, otherwise a positive errno value. */
4398 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
4400 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4401 struct xport *xport;
4402 struct ofpact_output output;
4405 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
4406 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
4407 flow_extract(packet, NULL, &flow);
4408 flow.in_port.ofp_port = OFPP_NONE;
4410 xport = xport_lookup(xcfg, ofport);
4414 output.port = xport->ofp_port;
4417 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
4418 &output.ofpact, sizeof output,
4422 struct xlate_cache *
4423 xlate_cache_new(void)
4425 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
4427 ofpbuf_init(&xcache->entries, 512);
4431 static struct xc_entry *
4432 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
4434 struct xc_entry *entry;
4436 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
4443 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
4445 if (entry->u.dev.tx) {
4446 netdev_vport_inc_tx(entry->u.dev.tx, stats);
4448 if (entry->u.dev.rx) {
4449 netdev_vport_inc_rx(entry->u.dev.rx, stats);
4451 if (entry->u.dev.bfd) {
4452 bfd_account_rx(entry->u.dev.bfd, stats);
4457 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
4459 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4460 struct xbridge *xbridge;
4461 struct xbundle *xbundle;
4462 struct flow_wildcards wc;
4464 xbridge = xbridge_lookup(xcfg, ofproto);
4469 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
4475 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
4478 /* Push stats and perform side effects of flow translation. */
4480 xlate_push_stats(struct xlate_cache *xcache,
4481 const struct dpif_flow_stats *stats)
4483 struct xc_entry *entry;
4484 struct ofpbuf entries = xcache->entries;
4486 if (!stats->n_packets) {
4490 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4491 switch (entry->type) {
4493 rule_dpif_credit_stats(entry->u.rule, stats);
4496 bond_account(entry->u.bond.bond, entry->u.bond.flow,
4497 entry->u.bond.vid, stats->n_bytes);
4500 xlate_cache_netdev(entry, stats);
4503 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
4504 entry->u.nf.iface, stats);
4507 mirror_update_stats(entry->u.mirror.mbridge,
4508 entry->u.mirror.mirrors,
4509 stats->n_packets, stats->n_bytes);
4512 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
4515 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
4516 entry->u.normal.vlan);
4518 case XC_FIN_TIMEOUT:
4519 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
4520 entry->u.fin.idle, entry->u.fin.hard);
4523 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
4533 xlate_dev_unref(struct xc_entry *entry)
4535 if (entry->u.dev.tx) {
4536 netdev_close(entry->u.dev.tx);
4538 if (entry->u.dev.rx) {
4539 netdev_close(entry->u.dev.rx);
4541 if (entry->u.dev.bfd) {
4542 bfd_unref(entry->u.dev.bfd);
4547 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
4549 netflow_flow_clear(netflow, flow);
4550 netflow_unref(netflow);
4555 xlate_cache_clear(struct xlate_cache *xcache)
4557 struct xc_entry *entry;
4558 struct ofpbuf entries;
4564 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
4565 switch (entry->type) {
4567 rule_dpif_unref(entry->u.rule);
4570 free(entry->u.bond.flow);
4571 bond_unref(entry->u.bond.bond);
4574 xlate_dev_unref(entry);
4577 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
4580 mbridge_unref(entry->u.mirror.mbridge);
4583 free(entry->u.learn.fm);
4584 ofpbuf_delete(entry->u.learn.ofpacts);
4587 free(entry->u.normal.flow);
4589 case XC_FIN_TIMEOUT:
4590 /* 'u.fin.rule' is always already held as a XC_RULE, which
4591 * has already released it's reference above. */
4594 group_dpif_unref(entry->u.group.group);
4601 ofpbuf_clear(&xcache->entries);
4605 xlate_cache_delete(struct xlate_cache *xcache)
4607 xlate_cache_clear(xcache);
4608 ofpbuf_uninit(&xcache->entries);