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 "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
41 #include "odp-execute.h"
42 #include "ofp-actions.h"
43 #include "ofproto/ofproto-dpif-ipfix.h"
44 #include "ofproto/ofproto-dpif-mirror.h"
45 #include "ofproto/ofproto-dpif-monitor.h"
46 #include "ofproto/ofproto-dpif-sflow.h"
47 #include "ofproto/ofproto-dpif.h"
48 #include "ofproto/ofproto-provider.h"
52 COVERAGE_DEFINE(xlate_actions);
53 COVERAGE_DEFINE(xlate_actions_oversize);
54 COVERAGE_DEFINE(xlate_actions_mpls_overflow);
56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
58 /* Maximum depth of flow table recursion (due to resubmit actions) in a
59 * flow translation. */
60 #define MAX_RESUBMIT_RECURSION 64
61 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
64 /* Maximum number of resubmit actions in a flow translation, whether they are
65 * recursive or not. */
66 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
68 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
71 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
72 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
74 struct list xbundles; /* Owned xbundles. */
75 struct hmap xports; /* Indexed by ofp_port. */
77 char *name; /* Name used in log messages. */
78 struct dpif *dpif; /* Datapath interface. */
79 struct mac_learning *ml; /* Mac learning handle. */
80 struct mbridge *mbridge; /* Mirroring. */
81 struct dpif_sflow *sflow; /* SFlow handle, or null. */
82 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
83 struct netflow *netflow; /* Netflow handle, or null. */
84 struct stp *stp; /* STP or null if disabled. */
86 /* Special rules installed by ofproto-dpif. */
87 struct rule_dpif *miss_rule;
88 struct rule_dpif *no_packet_in_rule;
90 enum ofp_config_flags frag; /* Fragmentation handling. */
91 bool has_in_band; /* Bridge has in band control? */
92 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
94 /* True if the datapath supports recirculation. */
97 /* True if the datapath supports variable-length
98 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
99 * False if the datapath supports only 8-byte (or shorter) userdata. */
100 bool variable_length_userdata;
102 /* Number of MPLS label stack entries that the datapath supports
104 size_t max_mpls_depth;
108 struct hmap_node hmap_node; /* In global 'xbundles' map. */
109 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
111 struct list list_node; /* In parent 'xbridges' list. */
112 struct xbridge *xbridge; /* Parent xbridge. */
114 struct list xports; /* Contains "struct xport"s. */
116 char *name; /* Name used in log messages. */
117 struct bond *bond; /* Nonnull iff more than one port. */
118 struct lacp *lacp; /* LACP handle or null. */
120 enum port_vlan_mode vlan_mode; /* VLAN mode. */
121 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
122 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
123 * NULL if all VLANs are trunked. */
124 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
125 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
129 struct hmap_node hmap_node; /* Node in global 'xports' map. */
130 struct ofport_dpif *ofport; /* Key in global 'xports map. */
132 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
133 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
135 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
137 struct list bundle_node; /* In parent xbundle (if it exists). */
138 struct xbundle *xbundle; /* Parent xbundle or null. */
140 struct netdev *netdev; /* 'ofport''s netdev. */
142 struct xbridge *xbridge; /* Parent bridge. */
143 struct xport *peer; /* Patch port peer or null. */
145 enum ofputil_port_config config; /* OpenFlow port configuration. */
146 enum ofputil_port_state state; /* OpenFlow port state. */
147 int stp_port_no; /* STP port number or -1 if not in use. */
149 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
151 bool may_enable; /* May be enabled in bonds. */
152 bool is_tunnel; /* Is a tunnel port. */
154 struct cfm *cfm; /* CFM handle or null. */
155 struct bfd *bfd; /* BFD handle or null. */
159 struct xlate_in *xin;
160 struct xlate_out *xout;
162 const struct xbridge *xbridge;
164 /* Flow at the last commit. */
165 struct flow base_flow;
167 /* Tunnel IP destination address as received. This is stored separately
168 * as the base_flow.tunnel is cleared on init to reflect the datapath
169 * behavior. Used to make sure not to send tunneled output to ourselves,
170 * which might lead to an infinite loop. This could happen easily
171 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
172 * actually set the tun_dst field. */
173 ovs_be32 orig_tunnel_ip_dst;
175 /* Stack for the push and pop actions. Each stack element is of type
176 * "union mf_subvalue". */
177 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
180 /* The rule that we are currently translating, or NULL. */
181 struct rule_dpif *rule;
183 /* Resubmit statistics, via xlate_table_action(). */
184 int recurse; /* Current resubmit nesting depth. */
185 int resubmits; /* Total number of resubmits. */
186 bool in_group; /* Currently translating ofgroup, if true. */
188 uint32_t orig_skb_priority; /* Priority when packet arrived. */
189 uint8_t table_id; /* OpenFlow table ID where flow was found. */
190 uint32_t sflow_n_outputs; /* Number of output ports. */
191 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
192 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
193 bool exit; /* No further actions should be processed. */
195 /* OpenFlow 1.1+ action set.
197 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
198 * When translation is otherwise complete, ofpacts_execute_action_set()
199 * converts it to a set of "struct ofpact"s that can be translated into
200 * datapath actions. */
201 struct ofpbuf action_set; /* Action set. */
202 uint64_t action_set_stub[1024 / 8];
205 /* A controller may use OFPP_NONE as the ingress port to indicate that
206 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
207 * when an input bundle is needed for validation (e.g., mirroring or
208 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
209 * any 'port' structs, so care must be taken when dealing with it. */
210 static struct xbundle ofpp_none_bundle = {
212 .vlan_mode = PORT_VLAN_TRUNK
215 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
216 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
217 * traffic egressing the 'ofport' with that priority should be marked with. */
218 struct skb_priority_to_dscp {
219 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
220 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
222 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
236 /* xlate_cache entries hold enough information to perform the side effects of
237 * xlate_actions() for a rule, without needing to perform rule translation
238 * from scratch. The primary usage of these is to submit statistics to objects
239 * that a flow relates to, although they may be used for other effects as well
240 * (for instance, refreshing hard timeouts for learned flows). */
244 struct rule_dpif *rule;
251 struct netflow *netflow;
256 struct mbridge *mbridge;
257 mirror_mask_t mirrors;
265 struct ofproto_dpif *ofproto;
266 struct rule_dpif *rule;
269 struct ofproto_dpif *ofproto;
274 struct rule_dpif *rule;
281 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
282 entries = xcache->entries; \
283 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
285 entry = ofpbuf_try_pull(&entries, sizeof *entry))
288 struct ofpbuf entries;
291 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
292 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
293 static struct hmap xports = HMAP_INITIALIZER(&xports);
295 static bool may_receive(const struct xport *, struct xlate_ctx *);
296 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
298 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
299 OVS_REQ_RDLOCK(xlate_rwlock);
300 static void xlate_normal(struct xlate_ctx *);
301 static void xlate_report(struct xlate_ctx *, const char *);
302 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
303 uint8_t table_id, bool may_packet_in,
304 bool honor_table_miss);
305 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
306 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
307 static void output_normal(struct xlate_ctx *, const struct xbundle *,
309 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
311 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
312 static struct xbundle *xbundle_lookup(const struct ofbundle *);
313 static struct xport *xport_lookup(const struct ofport_dpif *);
314 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
315 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
316 uint32_t skb_priority);
317 static void clear_skb_priorities(struct xport *);
318 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
321 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
325 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
326 struct dpif *dpif, struct rule_dpif *miss_rule,
327 struct rule_dpif *no_packet_in_rule,
328 const struct mac_learning *ml, struct stp *stp,
329 const struct mbridge *mbridge,
330 const struct dpif_sflow *sflow,
331 const struct dpif_ipfix *ipfix,
332 const struct netflow *netflow, enum ofp_config_flags frag,
333 bool forward_bpdu, bool has_in_band,
335 bool variable_length_userdata,
336 size_t max_mpls_depth)
338 struct xbridge *xbridge = xbridge_lookup(ofproto);
341 xbridge = xzalloc(sizeof *xbridge);
342 xbridge->ofproto = ofproto;
344 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
345 hmap_init(&xbridge->xports);
346 list_init(&xbridge->xbundles);
349 if (xbridge->ml != ml) {
350 mac_learning_unref(xbridge->ml);
351 xbridge->ml = mac_learning_ref(ml);
354 if (xbridge->mbridge != mbridge) {
355 mbridge_unref(xbridge->mbridge);
356 xbridge->mbridge = mbridge_ref(mbridge);
359 if (xbridge->sflow != sflow) {
360 dpif_sflow_unref(xbridge->sflow);
361 xbridge->sflow = dpif_sflow_ref(sflow);
364 if (xbridge->ipfix != ipfix) {
365 dpif_ipfix_unref(xbridge->ipfix);
366 xbridge->ipfix = dpif_ipfix_ref(ipfix);
369 if (xbridge->stp != stp) {
370 stp_unref(xbridge->stp);
371 xbridge->stp = stp_ref(stp);
374 if (xbridge->netflow != netflow) {
375 netflow_unref(xbridge->netflow);
376 xbridge->netflow = netflow_ref(netflow);
380 xbridge->name = xstrdup(name);
382 xbridge->dpif = dpif;
383 xbridge->forward_bpdu = forward_bpdu;
384 xbridge->has_in_band = has_in_band;
385 xbridge->frag = frag;
386 xbridge->miss_rule = miss_rule;
387 xbridge->no_packet_in_rule = no_packet_in_rule;
388 xbridge->enable_recirc = enable_recirc;
389 xbridge->variable_length_userdata = variable_length_userdata;
390 xbridge->max_mpls_depth = max_mpls_depth;
394 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
396 struct xbridge *xbridge = xbridge_lookup(ofproto);
397 struct xbundle *xbundle, *next_xbundle;
398 struct xport *xport, *next_xport;
404 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
405 xlate_ofport_remove(xport->ofport);
408 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
409 xlate_bundle_remove(xbundle->ofbundle);
412 hmap_remove(&xbridges, &xbridge->hmap_node);
413 mac_learning_unref(xbridge->ml);
414 mbridge_unref(xbridge->mbridge);
415 dpif_sflow_unref(xbridge->sflow);
416 dpif_ipfix_unref(xbridge->ipfix);
417 stp_unref(xbridge->stp);
418 hmap_destroy(&xbridge->xports);
424 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
425 const char *name, enum port_vlan_mode vlan_mode, int vlan,
426 unsigned long *trunks, bool use_priority_tags,
427 const struct bond *bond, const struct lacp *lacp,
430 struct xbundle *xbundle = xbundle_lookup(ofbundle);
433 xbundle = xzalloc(sizeof *xbundle);
434 xbundle->ofbundle = ofbundle;
435 xbundle->xbridge = xbridge_lookup(ofproto);
437 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
438 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
439 list_init(&xbundle->xports);
442 ovs_assert(xbundle->xbridge);
445 xbundle->name = xstrdup(name);
447 xbundle->vlan_mode = vlan_mode;
448 xbundle->vlan = vlan;
449 xbundle->trunks = trunks;
450 xbundle->use_priority_tags = use_priority_tags;
451 xbundle->floodable = floodable;
453 if (xbundle->bond != bond) {
454 bond_unref(xbundle->bond);
455 xbundle->bond = bond_ref(bond);
458 if (xbundle->lacp != lacp) {
459 lacp_unref(xbundle->lacp);
460 xbundle->lacp = lacp_ref(lacp);
465 xlate_bundle_remove(struct ofbundle *ofbundle)
467 struct xbundle *xbundle = xbundle_lookup(ofbundle);
468 struct xport *xport, *next;
474 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
475 list_remove(&xport->bundle_node);
476 xport->xbundle = NULL;
479 hmap_remove(&xbundles, &xbundle->hmap_node);
480 list_remove(&xbundle->list_node);
481 bond_unref(xbundle->bond);
482 lacp_unref(xbundle->lacp);
488 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
489 struct ofport_dpif *ofport, ofp_port_t ofp_port,
490 odp_port_t odp_port, const struct netdev *netdev,
491 const struct cfm *cfm, const struct bfd *bfd,
492 struct ofport_dpif *peer, int stp_port_no,
493 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
494 enum ofputil_port_config config,
495 enum ofputil_port_state state, bool is_tunnel,
498 struct xport *xport = xport_lookup(ofport);
502 xport = xzalloc(sizeof *xport);
503 xport->ofport = ofport;
504 xport->xbridge = xbridge_lookup(ofproto);
505 xport->ofp_port = ofp_port;
507 hmap_init(&xport->skb_priorities);
508 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
509 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
510 hash_ofp_port(xport->ofp_port));
513 ovs_assert(xport->ofp_port == ofp_port);
515 xport->config = config;
516 xport->state = state;
517 xport->stp_port_no = stp_port_no;
518 xport->is_tunnel = is_tunnel;
519 xport->may_enable = may_enable;
520 xport->odp_port = odp_port;
522 if (xport->netdev != netdev) {
523 netdev_close(xport->netdev);
524 xport->netdev = netdev_ref(netdev);
527 if (xport->cfm != cfm) {
528 cfm_unref(xport->cfm);
529 xport->cfm = cfm_ref(cfm);
532 if (xport->bfd != bfd) {
533 bfd_unref(xport->bfd);
534 xport->bfd = bfd_ref(bfd);
538 xport->peer->peer = NULL;
540 xport->peer = xport_lookup(peer);
542 xport->peer->peer = xport;
545 if (xport->xbundle) {
546 list_remove(&xport->bundle_node);
548 xport->xbundle = xbundle_lookup(ofbundle);
549 if (xport->xbundle) {
550 list_insert(&xport->xbundle->xports, &xport->bundle_node);
553 clear_skb_priorities(xport);
554 for (i = 0; i < n_qdscp; i++) {
555 struct skb_priority_to_dscp *pdscp;
556 uint32_t skb_priority;
558 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
563 pdscp = xmalloc(sizeof *pdscp);
564 pdscp->skb_priority = skb_priority;
565 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
566 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
567 hash_int(pdscp->skb_priority, 0));
572 xlate_ofport_remove(struct ofport_dpif *ofport)
574 struct xport *xport = xport_lookup(ofport);
581 xport->peer->peer = NULL;
585 if (xport->xbundle) {
586 list_remove(&xport->bundle_node);
589 clear_skb_priorities(xport);
590 hmap_destroy(&xport->skb_priorities);
592 hmap_remove(&xports, &xport->hmap_node);
593 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
595 netdev_close(xport->netdev);
596 cfm_unref(xport->cfm);
597 bfd_unref(xport->bfd);
601 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
602 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
603 * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
604 * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
605 * 'netflow' with the appropriate handles for those protocols if they're
606 * enabled. Caller is responsible for unrefing them.
608 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
609 * 'flow''s in_port to OFPP_NONE.
611 * This function does post-processing on data returned from
612 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
613 * of the upcall processing logic. In particular, if the extracted in_port is
614 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
615 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
616 * a VLAN header onto 'packet' (if it is nonnull).
618 * Similarly, this function also includes some logic to help with tunnels. It
619 * may modify 'flow' as necessary to make the tunneling implementation
620 * transparent to the upcall processing logic.
622 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
623 * or some other positive errno if there are other problems. */
625 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
626 const struct nlattr *key, size_t key_len, struct flow *flow,
627 struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
628 struct dpif_sflow **sflow, struct netflow **netflow,
629 odp_port_t *odp_in_port)
631 const struct xport *xport;
634 ovs_rwlock_rdlock(&xlate_rwlock);
635 if (odp_flow_key_to_flow(key, key_len, flow) == ODP_FIT_ERROR) {
641 *odp_in_port = flow->in_port.odp_port;
644 xport = xport_lookup(tnl_port_should_receive(flow)
645 ? tnl_port_receive(flow)
646 : odp_port_to_ofport(backer, flow->in_port.odp_port));
648 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
653 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
655 /* Make the packet resemble the flow, so that it gets sent to
656 * an OpenFlow controller properly, so that it looks correct
657 * for sFlow, and so that flow_extract() will get the correct
658 * vlan_tci if it is called on 'packet'. */
659 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
665 *ofproto = xport->xbridge->ofproto;
669 *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
673 *sflow = dpif_sflow_ref(xport->xbridge->sflow);
677 *netflow = netflow_ref(xport->xbridge->netflow);
681 ovs_rwlock_unlock(&xlate_rwlock);
685 static struct xbridge *
686 xbridge_lookup(const struct ofproto_dpif *ofproto)
688 struct xbridge *xbridge;
694 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
696 if (xbridge->ofproto == ofproto) {
703 static struct xbundle *
704 xbundle_lookup(const struct ofbundle *ofbundle)
706 struct xbundle *xbundle;
712 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
714 if (xbundle->ofbundle == ofbundle) {
721 static struct xport *
722 xport_lookup(const struct ofport_dpif *ofport)
730 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
732 if (xport->ofport == ofport) {
739 static struct stp_port *
740 xport_get_stp_port(const struct xport *xport)
742 return xport->xbridge->stp && xport->stp_port_no != -1
743 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
748 xport_stp_learn_state(const struct xport *xport)
750 struct stp_port *sp = xport_get_stp_port(xport);
751 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
755 xport_stp_forward_state(const struct xport *xport)
757 struct stp_port *sp = xport_get_stp_port(xport);
758 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
762 xport_stp_listen_state(const struct xport *xport)
764 struct stp_port *sp = xport_get_stp_port(xport);
765 return stp_listen_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
768 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
769 * were used to make the determination.*/
771 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
773 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
774 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
779 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
781 struct stp_port *sp = xport_get_stp_port(xport);
782 struct ofpbuf payload = *packet;
783 struct eth_header *eth = ofpbuf_data(&payload);
785 /* Sink packets on ports that have STP disabled when the bridge has
787 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
791 /* Trim off padding on payload. */
792 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
793 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
796 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
797 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
801 static struct xport *
802 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
806 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
808 if (xport->ofp_port == ofp_port) {
816 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
818 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
819 return xport ? xport->odp_port : ODPP_NONE;
823 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
827 xport = get_ofp_port(ctx->xbridge, ofp_port);
828 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
829 xport->state & OFPUTIL_PS_LINK_DOWN) {
836 static const struct ofputil_bucket *
837 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
841 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
843 struct group_dpif *group;
846 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
851 hit = group_first_live_bucket(ctx, group, depth) != NULL;
853 group_dpif_release(group);
857 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
860 bucket_is_alive(const struct xlate_ctx *ctx,
861 const struct ofputil_bucket *bucket, int depth)
863 if (depth >= MAX_LIVENESS_RECURSION) {
864 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
866 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
867 MAX_LIVENESS_RECURSION);
871 return !ofputil_bucket_has_liveness(bucket) ||
872 (bucket->watch_port != OFPP_ANY &&
873 odp_port_is_alive(ctx, bucket->watch_port)) ||
874 (bucket->watch_group != OFPG_ANY &&
875 group_is_alive(ctx, bucket->watch_group, depth + 1));
878 static const struct ofputil_bucket *
879 group_first_live_bucket(const struct xlate_ctx *ctx,
880 const struct group_dpif *group, int depth)
882 struct ofputil_bucket *bucket;
883 const struct list *buckets;
885 group_dpif_get_buckets(group, &buckets);
886 LIST_FOR_EACH (bucket, list_node, buckets) {
887 if (bucket_is_alive(ctx, bucket, depth)) {
895 static const struct ofputil_bucket *
896 group_best_live_bucket(const struct xlate_ctx *ctx,
897 const struct group_dpif *group,
900 const struct ofputil_bucket *best_bucket = NULL;
901 uint32_t best_score = 0;
904 const struct ofputil_bucket *bucket;
905 const struct list *buckets;
907 group_dpif_get_buckets(group, &buckets);
908 LIST_FOR_EACH (bucket, list_node, buckets) {
909 if (bucket_is_alive(ctx, bucket, 0)) {
910 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
911 if (score >= best_score) {
912 best_bucket = bucket;
923 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
925 return (bundle->vlan_mode != PORT_VLAN_ACCESS
926 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
930 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
932 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
936 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
938 return xbundle != &ofpp_none_bundle
939 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
944 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
946 return xbundle != &ofpp_none_bundle
947 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
952 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
954 return xbundle != &ofpp_none_bundle
955 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
959 static struct xbundle *
960 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
961 bool warn, struct xport **in_xportp)
965 /* Find the port and bundle for the received packet. */
966 xport = get_ofp_port(xbridge, in_port);
970 if (xport && xport->xbundle) {
971 return xport->xbundle;
974 /* Special-case OFPP_NONE, which a controller may use as the ingress
975 * port for traffic that it is sourcing. */
976 if (in_port == OFPP_NONE) {
977 return &ofpp_none_bundle;
980 /* Odd. A few possible reasons here:
982 * - We deleted a port but there are still a few packets queued up
985 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
986 * we don't know about.
988 * - The ofproto client didn't configure the port as part of a bundle.
989 * This is particularly likely to happen if a packet was received on the
990 * port after it was created, but before the client had a chance to
991 * configure its bundle.
994 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
996 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
997 "port %"PRIu16, xbridge->name, in_port);
1003 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1005 const struct xbridge *xbridge = ctx->xbridge;
1006 mirror_mask_t mirrors;
1007 struct xbundle *in_xbundle;
1011 mirrors = ctx->xout->mirrors;
1012 ctx->xout->mirrors = 0;
1014 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1015 ctx->xin->packet != NULL, NULL);
1019 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1021 /* Drop frames on bundles reserved for mirroring. */
1022 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1023 if (ctx->xin->packet != NULL) {
1024 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1025 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1026 "%s, which is reserved exclusively for mirroring",
1027 ctx->xbridge->name, in_xbundle->name);
1029 ofpbuf_clear(&ctx->xout->odp_actions);
1034 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1035 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1038 vlan = input_vid_to_vlan(in_xbundle, vid);
1044 /* Restore the original packet before adding the mirror actions. */
1045 ctx->xin->flow = *orig_flow;
1048 mirror_mask_t dup_mirrors;
1049 struct ofbundle *out;
1050 unsigned long *vlans;
1055 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1056 &vlans, &dup_mirrors, &out, &out_vlan);
1057 ovs_assert(has_mirror);
1060 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1062 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1065 if (!vlan_mirrored) {
1066 mirrors = zero_rightmost_1bit(mirrors);
1070 mirrors &= ~dup_mirrors;
1071 ctx->xout->mirrors |= dup_mirrors;
1073 struct xbundle *out_xbundle = xbundle_lookup(out);
1075 output_normal(ctx, out_xbundle, vlan);
1077 } else if (vlan != out_vlan
1078 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1079 struct xbundle *xbundle;
1081 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1082 if (xbundle_includes_vlan(xbundle, out_vlan)
1083 && !xbundle_mirror_out(xbridge, xbundle)) {
1084 output_normal(ctx, xbundle, out_vlan);
1091 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1092 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1093 * the bundle on which the packet was received, returns the VLAN to which the
1096 * Both 'vid' and the return value are in the range 0...4095. */
1098 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1100 switch (in_xbundle->vlan_mode) {
1101 case PORT_VLAN_ACCESS:
1102 return in_xbundle->vlan;
1105 case PORT_VLAN_TRUNK:
1108 case PORT_VLAN_NATIVE_UNTAGGED:
1109 case PORT_VLAN_NATIVE_TAGGED:
1110 return vid ? vid : in_xbundle->vlan;
1117 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1118 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1121 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1122 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1125 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1127 /* Allow any VID on the OFPP_NONE port. */
1128 if (in_xbundle == &ofpp_none_bundle) {
1132 switch (in_xbundle->vlan_mode) {
1133 case PORT_VLAN_ACCESS:
1136 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1137 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1138 "packet received on port %s configured as VLAN "
1139 "%"PRIu16" access port", vid, in_xbundle->name,
1146 case PORT_VLAN_NATIVE_UNTAGGED:
1147 case PORT_VLAN_NATIVE_TAGGED:
1149 /* Port must always carry its native VLAN. */
1153 case PORT_VLAN_TRUNK:
1154 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1156 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1157 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1158 "received on port %s not configured for trunking "
1159 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1171 /* Given 'vlan', the VLAN that a packet belongs to, and
1172 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1173 * that should be included in the 802.1Q header. (If the return value is 0,
1174 * then the 802.1Q header should only be included in the packet if there is a
1177 * Both 'vlan' and the return value are in the range 0...4095. */
1179 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1181 switch (out_xbundle->vlan_mode) {
1182 case PORT_VLAN_ACCESS:
1185 case PORT_VLAN_TRUNK:
1186 case PORT_VLAN_NATIVE_TAGGED:
1189 case PORT_VLAN_NATIVE_UNTAGGED:
1190 return vlan == out_xbundle->vlan ? 0 : vlan;
1198 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1201 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1203 ovs_be16 tci, old_tci;
1204 struct xport *xport;
1206 vid = output_vlan_to_vid(out_xbundle, vlan);
1207 if (list_is_empty(&out_xbundle->xports)) {
1208 /* Partially configured bundle with no slaves. Drop the packet. */
1210 } else if (!out_xbundle->bond) {
1211 ctx->xout->use_recirc = false;
1212 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1215 struct ofport_dpif *ofport;
1216 struct xlate_recirc *xr = &ctx->xout->recirc;
1217 struct flow_wildcards *wc = &ctx->xout->wc;
1219 if (ctx->xbridge->enable_recirc) {
1220 ctx->xout->use_recirc = bond_may_recirc(
1221 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1223 if (ctx->xout->use_recirc) {
1224 /* Only TCP mode uses recirculation. */
1225 xr->hash_alg = OVS_HASH_ALG_L4;
1226 bond_update_post_recirc_rules(out_xbundle->bond, false);
1228 /* Recirculation does not require unmasking hash fields. */
1233 ofport = bond_choose_output_slave(out_xbundle->bond,
1234 &ctx->xin->flow, wc, vid);
1235 xport = xport_lookup(ofport);
1238 /* No slaves enabled, so drop packet. */
1242 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1243 * accounting for this bond. */
1244 if (!ctx->xout->use_recirc) {
1245 if (ctx->xin->resubmit_stats) {
1246 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1247 ctx->xin->resubmit_stats->n_bytes);
1249 if (ctx->xin->xcache) {
1250 struct xc_entry *entry;
1253 flow = &ctx->xin->flow;
1254 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1255 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1256 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1257 entry->u.bond.vid = vid;
1262 old_tci = *flow_tci;
1264 if (tci || out_xbundle->use_priority_tags) {
1265 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1267 tci |= htons(VLAN_CFI);
1272 compose_output_action(ctx, xport->ofp_port);
1273 *flow_tci = old_tci;
1276 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1277 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1278 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1280 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1282 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1286 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1287 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1291 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1292 if (flow->nw_proto == ARP_OP_REPLY) {
1294 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1295 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1296 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1298 return flow->nw_src == flow->nw_dst;
1304 /* Checks whether a MAC learning update is necessary for MAC learning table
1305 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1308 * Most packets processed through the MAC learning table do not actually
1309 * change it in any way. This function requires only a read lock on the MAC
1310 * learning table, so it is much cheaper in this common case.
1312 * Keep the code here synchronized with that in update_learning_table__()
1315 is_mac_learning_update_needed(const struct mac_learning *ml,
1316 const struct flow *flow,
1317 struct flow_wildcards *wc,
1318 int vlan, struct xbundle *in_xbundle)
1319 OVS_REQ_RDLOCK(ml->rwlock)
1321 struct mac_entry *mac;
1323 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1327 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1328 if (!mac || mac_entry_age(ml, mac)) {
1332 if (is_gratuitous_arp(flow, wc)) {
1333 /* We don't want to learn from gratuitous ARP packets that are
1334 * reflected back over bond slaves so we lock the learning table. */
1335 if (!in_xbundle->bond) {
1337 } else if (mac_entry_is_grat_arp_locked(mac)) {
1342 return mac->port.p != in_xbundle->ofbundle;
1346 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1347 * received on 'in_xbundle' in 'vlan'.
1349 * This code repeats all the checks in is_mac_learning_update_needed() because
1350 * the lock was released between there and here and thus the MAC learning state
1351 * could have changed.
1353 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1356 update_learning_table__(const struct xbridge *xbridge,
1357 const struct flow *flow, struct flow_wildcards *wc,
1358 int vlan, struct xbundle *in_xbundle)
1359 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1361 struct mac_entry *mac;
1363 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1367 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1368 if (is_gratuitous_arp(flow, wc)) {
1369 /* We don't want to learn from gratuitous ARP packets that are
1370 * reflected back over bond slaves so we lock the learning table. */
1371 if (!in_xbundle->bond) {
1372 mac_entry_set_grat_arp_lock(mac);
1373 } else if (mac_entry_is_grat_arp_locked(mac)) {
1378 if (mac->port.p != in_xbundle->ofbundle) {
1379 /* The log messages here could actually be useful in debugging,
1380 * so keep the rate limit relatively high. */
1381 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1383 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1384 "on port %s in VLAN %d",
1385 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1386 in_xbundle->name, vlan);
1388 mac->port.p = in_xbundle->ofbundle;
1389 mac_learning_changed(xbridge->ml);
1394 update_learning_table(const struct xbridge *xbridge,
1395 const struct flow *flow, struct flow_wildcards *wc,
1396 int vlan, struct xbundle *in_xbundle)
1400 /* Don't learn the OFPP_NONE port. */
1401 if (in_xbundle == &ofpp_none_bundle) {
1405 /* First try the common case: no change to MAC learning table. */
1406 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1407 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1409 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1412 /* Slow path: MAC learning table might need an update. */
1413 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1414 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1415 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1419 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1420 * dropped. Returns true if they may be forwarded, false if they should be
1423 * 'in_port' must be the xport that corresponds to flow->in_port.
1424 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1426 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1427 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1428 * checked by input_vid_is_valid().
1430 * May also add tags to '*tags', although the current implementation only does
1431 * so in one special case.
1434 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1437 struct xbundle *in_xbundle = in_port->xbundle;
1438 const struct xbridge *xbridge = ctx->xbridge;
1439 struct flow *flow = &ctx->xin->flow;
1441 /* Drop frames for reserved multicast addresses
1442 * only if forward_bpdu option is absent. */
1443 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1444 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1448 if (in_xbundle->bond) {
1449 struct mac_entry *mac;
1451 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1457 xlate_report(ctx, "bonding refused admissibility, dropping");
1460 case BV_DROP_IF_MOVED:
1461 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1462 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1463 if (mac && mac->port.p != in_xbundle->ofbundle &&
1464 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1465 || mac_entry_is_grat_arp_locked(mac))) {
1466 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1467 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1471 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1480 xlate_normal(struct xlate_ctx *ctx)
1482 struct flow_wildcards *wc = &ctx->xout->wc;
1483 struct flow *flow = &ctx->xin->flow;
1484 struct xbundle *in_xbundle;
1485 struct xport *in_port;
1486 struct mac_entry *mac;
1491 ctx->xout->has_normal = true;
1493 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1494 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1495 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1497 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1498 ctx->xin->packet != NULL, &in_port);
1500 xlate_report(ctx, "no input bundle, dropping");
1504 /* Drop malformed frames. */
1505 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1506 !(flow->vlan_tci & htons(VLAN_CFI))) {
1507 if (ctx->xin->packet != NULL) {
1508 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1509 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1510 "VLAN tag received on port %s",
1511 ctx->xbridge->name, in_xbundle->name);
1513 xlate_report(ctx, "partial VLAN tag, dropping");
1517 /* Drop frames on bundles reserved for mirroring. */
1518 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1519 if (ctx->xin->packet != NULL) {
1520 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1521 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1522 "%s, which is reserved exclusively for mirroring",
1523 ctx->xbridge->name, in_xbundle->name);
1525 xlate_report(ctx, "input port is mirror output port, dropping");
1530 vid = vlan_tci_to_vid(flow->vlan_tci);
1531 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1532 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1535 vlan = input_vid_to_vlan(in_xbundle, vid);
1537 /* Check other admissibility requirements. */
1538 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1542 /* Learn source MAC. */
1543 if (ctx->xin->may_learn) {
1544 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1546 if (ctx->xin->xcache) {
1547 struct xc_entry *entry;
1549 /* Save enough info to update mac learning table later. */
1550 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
1551 entry->u.normal.ofproto = ctx->xin->ofproto;
1552 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
1553 entry->u.normal.vlan = vlan;
1556 /* Determine output bundle. */
1557 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1558 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1559 mac_port = mac ? mac->port.p : NULL;
1560 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1563 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1564 if (mac_xbundle && mac_xbundle != in_xbundle) {
1565 xlate_report(ctx, "forwarding to learned port");
1566 output_normal(ctx, mac_xbundle, vlan);
1567 } else if (!mac_xbundle) {
1568 xlate_report(ctx, "learned port is unknown, dropping");
1570 xlate_report(ctx, "learned port is input port, dropping");
1573 struct xbundle *xbundle;
1575 xlate_report(ctx, "no learned MAC for destination, flooding");
1576 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1577 if (xbundle != in_xbundle
1578 && xbundle_includes_vlan(xbundle, vlan)
1579 && xbundle->floodable
1580 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1581 output_normal(ctx, xbundle, vlan);
1584 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1588 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1589 * the number of packets out of UINT32_MAX to sample. The given
1590 * cookie is passed back in the callback for each sampled packet.
1593 compose_sample_action(const struct xbridge *xbridge,
1594 struct ofpbuf *odp_actions,
1595 const struct flow *flow,
1596 const uint32_t probability,
1597 const union user_action_cookie *cookie,
1598 const size_t cookie_size)
1600 size_t sample_offset, actions_offset;
1601 odp_port_t odp_port;
1605 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1607 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1609 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1611 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1612 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
1613 flow_hash_5tuple(flow, 0));
1614 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
1617 nl_msg_end_nested(odp_actions, actions_offset);
1618 nl_msg_end_nested(odp_actions, sample_offset);
1619 return cookie_offset;
1623 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1624 odp_port_t odp_port, unsigned int n_outputs,
1625 union user_action_cookie *cookie)
1629 cookie->type = USER_ACTION_COOKIE_SFLOW;
1630 cookie->sflow.vlan_tci = vlan_tci;
1632 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1633 * port information") for the interpretation of cookie->output. */
1634 switch (n_outputs) {
1636 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1637 cookie->sflow.output = 0x40000000 | 256;
1641 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1643 cookie->sflow.output = ifindex;
1648 /* 0x80000000 means "multiple output ports. */
1649 cookie->sflow.output = 0x80000000 | n_outputs;
1654 /* Compose SAMPLE action for sFlow bridge sampling. */
1656 compose_sflow_action(const struct xbridge *xbridge,
1657 struct ofpbuf *odp_actions,
1658 const struct flow *flow,
1659 odp_port_t odp_port)
1661 uint32_t probability;
1662 union user_action_cookie cookie;
1664 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1668 probability = dpif_sflow_get_probability(xbridge->sflow);
1669 compose_sflow_cookie(xbridge, htons(0), odp_port,
1670 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1672 return compose_sample_action(xbridge, odp_actions, flow, probability,
1673 &cookie, sizeof cookie.sflow);
1677 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1678 uint32_t obs_domain_id, uint32_t obs_point_id,
1679 union user_action_cookie *cookie)
1681 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1682 cookie->flow_sample.probability = probability;
1683 cookie->flow_sample.collector_set_id = collector_set_id;
1684 cookie->flow_sample.obs_domain_id = obs_domain_id;
1685 cookie->flow_sample.obs_point_id = obs_point_id;
1689 compose_ipfix_cookie(union user_action_cookie *cookie)
1691 cookie->type = USER_ACTION_COOKIE_IPFIX;
1694 /* Compose SAMPLE action for IPFIX bridge sampling. */
1696 compose_ipfix_action(const struct xbridge *xbridge,
1697 struct ofpbuf *odp_actions,
1698 const struct flow *flow)
1700 uint32_t probability;
1701 union user_action_cookie cookie;
1703 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1707 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1708 compose_ipfix_cookie(&cookie);
1710 compose_sample_action(xbridge, odp_actions, flow, probability,
1711 &cookie, sizeof cookie.ipfix);
1714 /* SAMPLE action for sFlow must be first action in any given list of
1715 * actions. At this point we do not have all information required to
1716 * build it. So try to build sample action as complete as possible. */
1718 add_sflow_action(struct xlate_ctx *ctx)
1720 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1721 &ctx->xout->odp_actions,
1722 &ctx->xin->flow, ODPP_NONE);
1723 ctx->sflow_odp_port = 0;
1724 ctx->sflow_n_outputs = 0;
1727 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1728 * of actions, eventually after the SAMPLE action for sFlow. */
1730 add_ipfix_action(struct xlate_ctx *ctx)
1732 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1736 /* Fix SAMPLE action according to data collected while composing ODP actions.
1737 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1738 * USERSPACE action's user-cookie which is required for sflow. */
1740 fix_sflow_action(struct xlate_ctx *ctx)
1742 const struct flow *base = &ctx->base_flow;
1743 union user_action_cookie *cookie;
1745 if (!ctx->user_cookie_offset) {
1749 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1750 sizeof cookie->sflow);
1751 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1753 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1754 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1757 static enum slow_path_reason
1758 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1759 const struct xport *xport, const struct ofpbuf *packet)
1761 struct flow_wildcards *wc = &ctx->xout->wc;
1762 const struct xbridge *xbridge = ctx->xbridge;
1766 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1768 cfm_process_heartbeat(xport->cfm, packet);
1771 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1773 bfd_process_packet(xport->bfd, flow, packet);
1774 /* If POLL received, immediately sends FINAL back. */
1775 if (bfd_should_send_packet(xport->bfd)) {
1777 ofproto_dpif_monitor_port_send_soon(xport->ofport);
1779 ofproto_dpif_monitor_port_send_soon_safe(xport->ofport);
1784 } else if (xport->xbundle && xport->xbundle->lacp
1785 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1787 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1790 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1792 stp_process_packet(xport, packet);
1801 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1804 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1805 struct flow_wildcards *wc = &ctx->xout->wc;
1806 struct flow *flow = &ctx->xin->flow;
1807 ovs_be16 flow_vlan_tci;
1808 uint32_t flow_pkt_mark;
1809 uint8_t flow_nw_tos;
1810 odp_port_t out_port, odp_port;
1813 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1814 * before traversing a patch port. */
1815 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 26);
1818 xlate_report(ctx, "Nonexistent output port");
1820 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1821 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1823 } else if (check_stp) {
1824 if (is_stp(&ctx->base_flow)) {
1825 if (!xport_stp_listen_state(xport)) {
1826 xlate_report(ctx, "STP not in listening state, "
1827 "skipping bpdu output");
1830 } else if (!xport_stp_forward_state(xport)) {
1831 xlate_report(ctx, "STP not in forwarding state, "
1837 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1838 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1843 const struct xport *peer = xport->peer;
1844 struct flow old_flow = ctx->xin->flow;
1845 enum slow_path_reason special;
1847 ctx->xbridge = peer->xbridge;
1848 flow->in_port.ofp_port = peer->ofp_port;
1849 flow->metadata = htonll(0);
1850 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1851 memset(flow->regs, 0, sizeof flow->regs);
1853 special = process_special(ctx, &ctx->xin->flow, peer,
1856 ctx->xout->slow |= special;
1857 } else if (may_receive(peer, ctx)) {
1858 if (xport_stp_forward_state(peer)) {
1859 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1861 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1862 * learning action look at the packet, then drop it. */
1863 struct flow old_base_flow = ctx->base_flow;
1864 size_t old_size = ofpbuf_size(&ctx->xout->odp_actions);
1865 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1866 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1867 ctx->xout->mirrors = old_mirrors;
1868 ctx->base_flow = old_base_flow;
1869 ofpbuf_set_size(&ctx->xout->odp_actions, old_size);
1873 ctx->xin->flow = old_flow;
1874 ctx->xbridge = xport->xbridge;
1876 if (ctx->xin->resubmit_stats) {
1877 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1878 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1880 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1883 if (ctx->xin->xcache) {
1884 struct xc_entry *entry;
1886 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1887 entry->u.dev.tx = netdev_ref(xport->netdev);
1888 entry->u.dev.rx = netdev_ref(peer->netdev);
1889 entry->u.dev.bfd = bfd_ref(peer->bfd);
1895 flow_vlan_tci = flow->vlan_tci;
1896 flow_pkt_mark = flow->pkt_mark;
1897 flow_nw_tos = flow->nw_tos;
1899 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1900 wc->masks.nw_tos |= IP_DSCP_MASK;
1901 flow->nw_tos &= ~IP_DSCP_MASK;
1902 flow->nw_tos |= dscp;
1905 if (xport->is_tunnel) {
1906 /* Save tunnel metadata so that changes made due to
1907 * the Logical (tunnel) Port are not visible for any further
1908 * matches, while explicit set actions on tunnel metadata are.
1910 struct flow_tnl flow_tnl = flow->tunnel;
1911 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1912 if (odp_port == ODPP_NONE) {
1913 xlate_report(ctx, "Tunneling decided against output");
1914 goto out; /* restore flow_nw_tos */
1916 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1917 xlate_report(ctx, "Not tunneling to our own address");
1918 goto out; /* restore flow_nw_tos */
1920 if (ctx->xin->resubmit_stats) {
1921 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1923 if (ctx->xin->xcache) {
1924 struct xc_entry *entry;
1926 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1927 entry->u.dev.tx = netdev_ref(xport->netdev);
1929 out_port = odp_port;
1930 commit_odp_tunnel_action(flow, &ctx->base_flow,
1931 &ctx->xout->odp_actions);
1932 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1934 odp_port = xport->odp_port;
1935 out_port = odp_port;
1936 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1937 ofp_port_t vlandev_port;
1939 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1940 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1941 ofp_port, flow->vlan_tci);
1942 if (vlandev_port != ofp_port) {
1943 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1944 flow->vlan_tci = htons(0);
1949 if (out_port != ODPP_NONE) {
1950 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1951 &ctx->xout->odp_actions,
1954 if (ctx->xout->use_recirc) {
1955 struct ovs_action_hash *act_hash;
1956 struct xlate_recirc *xr = &ctx->xout->recirc;
1959 act_hash = nl_msg_put_unspec_uninit(&ctx->xout->odp_actions,
1960 OVS_ACTION_ATTR_HASH,
1962 act_hash->hash_alg = xr->hash_alg;
1963 act_hash->hash_basis = xr->hash_basis;
1965 /* Recirc action. */
1966 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
1969 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1973 ctx->sflow_odp_port = odp_port;
1974 ctx->sflow_n_outputs++;
1975 ctx->xout->nf_output_iface = ofp_port;
1980 flow->vlan_tci = flow_vlan_tci;
1981 flow->pkt_mark = flow_pkt_mark;
1982 flow->nw_tos = flow_nw_tos;
1986 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1988 compose_output_action__(ctx, ofp_port, true);
1992 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1994 struct rule_dpif *old_rule = ctx->rule;
1995 struct rule_actions *actions;
1997 if (ctx->xin->resubmit_stats) {
1998 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
2004 actions = rule_dpif_get_actions(rule);
2005 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2006 ctx->rule = old_rule;
2011 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2013 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2015 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2016 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2017 MAX_RESUBMIT_RECURSION);
2018 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2019 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2020 } else if (ofpbuf_size(&ctx->xout->odp_actions) > UINT16_MAX) {
2021 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2022 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2023 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2032 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2033 bool may_packet_in, bool honor_table_miss)
2035 if (xlate_resubmit_resource_check(ctx)) {
2036 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
2037 bool skip_wildcards = ctx->xin->skip_wildcards;
2038 uint8_t old_table_id = ctx->table_id;
2039 struct rule_dpif *rule;
2040 enum rule_dpif_lookup_verdict verdict;
2041 enum ofputil_port_config config = 0;
2043 ctx->table_id = table_id;
2045 /* Look up a flow with 'in_port' as the input port. Then restore the
2046 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2047 * have surprising behavior). */
2048 ctx->xin->flow.in_port.ofp_port = in_port;
2049 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2052 ? &ctx->xout->wc : NULL,
2054 &ctx->table_id, &rule,
2055 ctx->xin->xcache != NULL);
2056 ctx->xin->flow.in_port.ofp_port = old_in_port;
2058 if (ctx->xin->resubmit_hook) {
2059 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
2063 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
2065 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
2066 if (may_packet_in) {
2067 struct xport *xport;
2069 xport = get_ofp_port(ctx->xbridge,
2070 ctx->xin->flow.in_port.ofp_port);
2071 config = xport ? xport->config : 0;
2074 /* Fall through to drop */
2075 case RULE_DPIF_LOOKUP_VERDICT_DROP:
2076 config = OFPUTIL_PC_NO_PACKET_IN;
2078 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
2079 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
2080 config = OFPUTIL_PC_NO_PACKET_IN;
2087 choose_miss_rule(config, ctx->xbridge->miss_rule,
2088 ctx->xbridge->no_packet_in_rule, &rule,
2089 ctx->xin->xcache != NULL);
2093 /* Fill in the cache entry here instead of xlate_recursively
2094 * to make the reference counting more explicit. We take a
2095 * reference in the lookups above if we are going to cache the
2097 if (ctx->xin->xcache) {
2098 struct xc_entry *entry;
2100 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2101 entry->u.rule = rule;
2103 xlate_recursively(ctx, rule);
2106 ctx->table_id = old_table_id;
2114 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
2116 uint64_t action_list_stub[1024 / 8];
2117 struct ofpbuf action_list, action_set;
2119 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2120 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2122 ofpacts_execute_action_set(&action_list, &action_set);
2124 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2127 ofpbuf_uninit(&action_set);
2128 ofpbuf_uninit(&action_list);
2132 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2134 const struct ofputil_bucket *bucket;
2135 const struct list *buckets;
2136 struct flow old_flow = ctx->xin->flow;
2138 group_dpif_get_buckets(group, &buckets);
2140 LIST_FOR_EACH (bucket, list_node, buckets) {
2141 xlate_group_bucket(ctx, bucket);
2142 /* Roll back flow to previous state.
2143 * This is equivalent to cloning the packet for each bucket.
2145 * As a side effect any subsequently applied actions will
2146 * also effectively be applied to a clone of the packet taken
2147 * just before applying the all or indirect group. */
2148 ctx->xin->flow = old_flow;
2153 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2155 const struct ofputil_bucket *bucket;
2157 bucket = group_first_live_bucket(ctx, group, 0);
2159 xlate_group_bucket(ctx, bucket);
2164 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2166 struct flow_wildcards *wc = &ctx->xout->wc;
2167 const struct ofputil_bucket *bucket;
2170 basis = hash_mac(ctx->xin->flow.dl_dst, 0, 0);
2171 bucket = group_best_live_bucket(ctx, group, basis);
2173 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2174 xlate_group_bucket(ctx, bucket);
2179 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2181 ctx->in_group = true;
2183 switch (group_dpif_get_type(group)) {
2185 case OFPGT11_INDIRECT:
2186 xlate_all_group(ctx, group);
2188 case OFPGT11_SELECT:
2189 xlate_select_group(ctx, group);
2192 xlate_ff_group(ctx, group);
2197 group_dpif_release(group);
2199 ctx->in_group = false;
2203 xlate_group_resource_check(struct xlate_ctx *ctx)
2205 if (!xlate_resubmit_resource_check(ctx)) {
2207 } else if (ctx->in_group) {
2208 /* Prevent nested translation of OpenFlow groups.
2210 * OpenFlow allows this restriction. We enforce this restriction only
2211 * because, with the current architecture, we would otherwise have to
2212 * take a possibly recursive read lock on the ofgroup rwlock, which is
2213 * unsafe given that POSIX allows taking a read lock to block if there
2214 * is a thread blocked on taking the write lock. Other solutions
2215 * without this restriction are also possible, but seem unwarranted
2216 * given the current limited use of groups. */
2217 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2219 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2227 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2229 if (xlate_group_resource_check(ctx)) {
2230 struct group_dpif *group;
2233 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2235 xlate_group_action__(ctx, group);
2245 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2246 const struct ofpact_resubmit *resubmit)
2250 bool may_packet_in = false;
2251 bool honor_table_miss = false;
2253 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2254 /* Still allow missed packets to be sent to the controller
2255 * if resubmitting from an internal table. */
2256 may_packet_in = true;
2257 honor_table_miss = true;
2260 in_port = resubmit->in_port;
2261 if (in_port == OFPP_IN_PORT) {
2262 in_port = ctx->xin->flow.in_port.ofp_port;
2265 table_id = resubmit->table_id;
2266 if (table_id == 255) {
2267 table_id = ctx->table_id;
2270 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2275 flood_packets(struct xlate_ctx *ctx, bool all)
2277 const struct xport *xport;
2279 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2280 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2285 compose_output_action__(ctx, xport->ofp_port, false);
2286 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2287 compose_output_action(ctx, xport->ofp_port);
2291 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2295 execute_controller_action(struct xlate_ctx *ctx, int len,
2296 enum ofp_packet_in_reason reason,
2297 uint16_t controller_id)
2299 struct ofproto_packet_in *pin;
2300 struct ofpbuf *packet;
2301 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2303 ctx->xout->slow |= SLOW_CONTROLLER;
2304 if (!ctx->xin->packet) {
2308 packet = ofpbuf_clone(ctx->xin->packet);
2310 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2311 &ctx->xout->odp_actions,
2314 odp_execute_actions(NULL, packet, false, &md,
2315 ofpbuf_data(&ctx->xout->odp_actions),
2316 ofpbuf_size(&ctx->xout->odp_actions), NULL);
2318 pin = xmalloc(sizeof *pin);
2319 pin->up.packet_len = ofpbuf_size(packet);
2320 pin->up.packet = ofpbuf_steal_data(packet);
2321 pin->up.reason = reason;
2322 pin->up.table_id = ctx->table_id;
2323 pin->up.cookie = (ctx->rule
2324 ? rule_dpif_get_flow_cookie(ctx->rule)
2327 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2329 pin->controller_id = controller_id;
2330 pin->send_len = len;
2331 /* If a rule is a table-miss rule then this is
2332 * a table-miss handled by a table-miss rule.
2334 * Else, if rule is internal and has a controller action,
2335 * the later being implied by the rule being processed here,
2336 * then this is a table-miss handled without a table-miss rule.
2338 * Otherwise this is not a table-miss. */
2339 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
2341 if (rule_dpif_is_table_miss(ctx->rule)) {
2342 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
2343 } else if (rule_dpif_is_internal(ctx->rule)) {
2344 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
2347 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2348 ofpbuf_delete(packet);
2352 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2354 struct flow_wildcards *wc = &ctx->xout->wc;
2355 struct flow *flow = &ctx->xin->flow;
2358 ovs_assert(eth_type_mpls(mpls->ethertype));
2360 n = flow_count_mpls_labels(flow, wc);
2362 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2363 &ctx->xout->odp_actions,
2365 } else if (n >= FLOW_MAX_MPLS_LABELS) {
2366 if (ctx->xin->packet != NULL) {
2367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2368 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2369 "MPLS push action can't be performed as it would "
2370 "have more MPLS LSEs than the %d supported.",
2371 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2375 } else if (n >= ctx->xbridge->max_mpls_depth) {
2376 COVERAGE_INC(xlate_actions_mpls_overflow);
2377 ctx->xout->slow |= SLOW_ACTION;
2380 flow_push_mpls(flow, n, mpls->ethertype, wc);
2384 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2386 struct flow_wildcards *wc = &ctx->xout->wc;
2387 struct flow *flow = &ctx->xin->flow;
2388 int n = flow_count_mpls_labels(flow, wc);
2390 if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2391 if (ctx->xin->packet != NULL) {
2392 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2393 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2394 "MPLS pop action can't be performed as it has "
2395 "more MPLS LSEs than the %d supported.",
2396 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2399 ofpbuf_clear(&ctx->xout->odp_actions);
2404 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2406 struct flow *flow = &ctx->xin->flow;
2408 if (!is_ip_any(flow)) {
2412 ctx->xout->wc.masks.nw_ttl = 0xff;
2413 if (flow->nw_ttl > 1) {
2419 for (i = 0; i < ids->n_controllers; i++) {
2420 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2424 /* Stop processing for current table. */
2430 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2432 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2433 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2434 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2439 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2441 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2442 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2443 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2448 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2450 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2451 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2452 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2457 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2459 struct flow *flow = &ctx->xin->flow;
2460 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2461 struct flow_wildcards *wc = &ctx->xout->wc;
2463 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2464 if (eth_type_mpls(flow->dl_type)) {
2467 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2470 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2472 /* Stop processing for current table. */
2481 xlate_output_action(struct xlate_ctx *ctx,
2482 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2484 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2486 ctx->xout->nf_output_iface = NF_OUT_DROP;
2490 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2493 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2494 0, may_packet_in, true);
2500 flood_packets(ctx, false);
2503 flood_packets(ctx, true);
2505 case OFPP_CONTROLLER:
2506 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2512 if (port != ctx->xin->flow.in_port.ofp_port) {
2513 compose_output_action(ctx, port);
2515 xlate_report(ctx, "skipping output to input port");
2520 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2521 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2522 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2523 ctx->xout->nf_output_iface = prev_nf_output_iface;
2524 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2525 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2526 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2531 xlate_output_reg_action(struct xlate_ctx *ctx,
2532 const struct ofpact_output_reg *or)
2534 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2535 if (port <= UINT16_MAX) {
2536 union mf_subvalue value;
2538 memset(&value, 0xff, sizeof value);
2539 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2540 xlate_output_action(ctx, u16_to_ofp(port),
2541 or->max_len, false);
2546 xlate_enqueue_action(struct xlate_ctx *ctx,
2547 const struct ofpact_enqueue *enqueue)
2549 ofp_port_t ofp_port = enqueue->port;
2550 uint32_t queue_id = enqueue->queue;
2551 uint32_t flow_priority, priority;
2554 /* Translate queue to priority. */
2555 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2557 /* Fall back to ordinary output action. */
2558 xlate_output_action(ctx, enqueue->port, 0, false);
2562 /* Check output port. */
2563 if (ofp_port == OFPP_IN_PORT) {
2564 ofp_port = ctx->xin->flow.in_port.ofp_port;
2565 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2569 /* Add datapath actions. */
2570 flow_priority = ctx->xin->flow.skb_priority;
2571 ctx->xin->flow.skb_priority = priority;
2572 compose_output_action(ctx, ofp_port);
2573 ctx->xin->flow.skb_priority = flow_priority;
2575 /* Update NetFlow output port. */
2576 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2577 ctx->xout->nf_output_iface = ofp_port;
2578 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2579 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2584 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2586 uint32_t skb_priority;
2588 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2589 ctx->xin->flow.skb_priority = skb_priority;
2591 /* Couldn't translate queue to a priority. Nothing to do. A warning
2592 * has already been logged. */
2597 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2599 const struct xbridge *xbridge = xbridge_;
2610 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2613 port = get_ofp_port(xbridge, ofp_port);
2614 return port ? port->may_enable : false;
2619 xlate_bundle_action(struct xlate_ctx *ctx,
2620 const struct ofpact_bundle *bundle)
2624 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2626 CONST_CAST(struct xbridge *, ctx->xbridge));
2627 if (bundle->dst.field) {
2628 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2631 xlate_output_action(ctx, port, 0, false);
2636 xlate_learn_action(struct xlate_ctx *ctx,
2637 const struct ofpact_learn *learn)
2639 uint64_t ofpacts_stub[1024 / 8];
2640 struct ofputil_flow_mod fm;
2641 struct ofpbuf ofpacts;
2643 ctx->xout->has_learn = true;
2645 learn_mask(learn, &ctx->xout->wc);
2647 if (!ctx->xin->may_learn) {
2651 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2652 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2653 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2654 ofpbuf_uninit(&ofpacts);
2656 if (ctx->xin->xcache) {
2657 struct xc_entry *entry;
2659 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
2660 entry->u.learn.ofproto = ctx->xin->ofproto;
2661 /* Lookup the learned rule, taking a reference on it. The reference
2662 * is released when this cache entry is deleted. */
2663 rule_dpif_lookup(ctx->xbridge->ofproto, &ctx->xin->flow, NULL,
2664 &entry->u.learn.rule, true);
2669 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
2670 uint16_t idle_timeout, uint16_t hard_timeout)
2672 if (tcp_flags & (TCP_FIN | TCP_RST)) {
2673 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
2678 xlate_fin_timeout(struct xlate_ctx *ctx,
2679 const struct ofpact_fin_timeout *oft)
2682 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
2683 oft->fin_idle_timeout, oft->fin_hard_timeout);
2684 if (ctx->xin->xcache) {
2685 struct xc_entry *entry;
2687 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
2688 /* XC_RULE already holds a reference on the rule, none is taken
2690 entry->u.fin.rule = ctx->rule;
2691 entry->u.fin.idle = oft->fin_idle_timeout;
2692 entry->u.fin.hard = oft->fin_hard_timeout;
2698 xlate_sample_action(struct xlate_ctx *ctx,
2699 const struct ofpact_sample *os)
2701 union user_action_cookie cookie;
2702 /* Scale the probability from 16-bit to 32-bit while representing
2703 * the same percentage. */
2704 uint32_t probability = (os->probability << 16) | os->probability;
2706 if (!ctx->xbridge->variable_length_userdata) {
2707 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2709 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2710 "lacks support (needs Linux 3.10+ or kernel module from "
2715 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2716 &ctx->xout->odp_actions,
2719 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2720 os->obs_domain_id, os->obs_point_id, &cookie);
2721 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2722 probability, &cookie, sizeof cookie.flow_sample);
2726 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2728 if (xport->config & (is_stp(&ctx->xin->flow)
2729 ? OFPUTIL_PC_NO_RECV_STP
2730 : OFPUTIL_PC_NO_RECV)) {
2734 /* Only drop packets here if both forwarding and learning are
2735 * disabled. If just learning is enabled, we need to have
2736 * OFPP_NORMAL and the learning action have a look at the packet
2737 * before we can drop it. */
2738 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2746 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2748 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2749 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2750 ofpact_pad(&ctx->action_set);
2754 xlate_action_set(struct xlate_ctx *ctx)
2756 uint64_t action_list_stub[1024 / 64];
2757 struct ofpbuf action_list;
2759 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2760 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2761 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2762 ofpbuf_uninit(&action_list);
2766 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2767 struct xlate_ctx *ctx)
2769 struct flow_wildcards *wc = &ctx->xout->wc;
2770 struct flow *flow = &ctx->xin->flow;
2771 const struct ofpact *a;
2773 /* dl_type already in the mask, not set below. */
2775 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2776 struct ofpact_controller *controller;
2777 const struct ofpact_metadata *metadata;
2778 const struct ofpact_set_field *set_field;
2779 const struct mf_field *mf;
2787 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2788 ofpact_get_OUTPUT(a)->max_len, true);
2792 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2797 case OFPACT_CONTROLLER:
2798 controller = ofpact_get_CONTROLLER(a);
2799 execute_controller_action(ctx, controller->max_len,
2801 controller->controller_id);
2804 case OFPACT_ENQUEUE:
2805 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2808 case OFPACT_SET_VLAN_VID:
2809 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2810 if (flow->vlan_tci & htons(VLAN_CFI) ||
2811 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2812 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2813 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2818 case OFPACT_SET_VLAN_PCP:
2819 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2820 if (flow->vlan_tci & htons(VLAN_CFI) ||
2821 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2822 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2823 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2824 << VLAN_PCP_SHIFT) | VLAN_CFI);
2828 case OFPACT_STRIP_VLAN:
2829 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2830 flow->vlan_tci = htons(0);
2833 case OFPACT_PUSH_VLAN:
2834 /* XXX 802.1AD(QinQ) */
2835 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2836 flow->vlan_tci = htons(VLAN_CFI);
2839 case OFPACT_SET_ETH_SRC:
2840 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2841 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2844 case OFPACT_SET_ETH_DST:
2845 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2846 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2849 case OFPACT_SET_IPV4_SRC:
2850 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2851 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2852 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2856 case OFPACT_SET_IPV4_DST:
2857 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2858 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2859 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2863 case OFPACT_SET_IP_DSCP:
2864 if (is_ip_any(flow)) {
2865 wc->masks.nw_tos |= IP_DSCP_MASK;
2866 flow->nw_tos &= ~IP_DSCP_MASK;
2867 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2871 case OFPACT_SET_IP_ECN:
2872 if (is_ip_any(flow)) {
2873 wc->masks.nw_tos |= IP_ECN_MASK;
2874 flow->nw_tos &= ~IP_ECN_MASK;
2875 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2879 case OFPACT_SET_IP_TTL:
2880 if (is_ip_any(flow)) {
2881 wc->masks.nw_ttl = 0xff;
2882 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2886 case OFPACT_SET_L4_SRC_PORT:
2887 if (is_ip_any(flow)) {
2888 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2889 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2890 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2894 case OFPACT_SET_L4_DST_PORT:
2895 if (is_ip_any(flow)) {
2896 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2897 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2898 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2902 case OFPACT_RESUBMIT:
2903 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2906 case OFPACT_SET_TUNNEL:
2907 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2910 case OFPACT_SET_QUEUE:
2911 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2914 case OFPACT_POP_QUEUE:
2915 flow->skb_priority = ctx->orig_skb_priority;
2918 case OFPACT_REG_MOVE:
2919 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2922 case OFPACT_REG_LOAD:
2923 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2926 case OFPACT_SET_FIELD:
2927 set_field = ofpact_get_SET_FIELD(a);
2928 mf = set_field->field;
2930 /* Set field action only ever overwrites packet's outermost
2931 * applicable header fields. Do nothing if no header exists. */
2932 if (mf->id == MFF_VLAN_VID) {
2933 wc->masks.vlan_tci |= htons(VLAN_CFI);
2934 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2937 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
2938 /* 'dl_type' is already unwildcarded. */
2939 && !eth_type_mpls(flow->dl_type)) {
2943 mf_mask_field_and_prereqs(mf, &wc->masks);
2944 mf_set_flow_value(mf, &set_field->value, flow);
2947 case OFPACT_STACK_PUSH:
2948 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2952 case OFPACT_STACK_POP:
2953 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2957 case OFPACT_PUSH_MPLS:
2958 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2961 case OFPACT_POP_MPLS:
2962 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2965 case OFPACT_SET_MPLS_LABEL:
2966 compose_set_mpls_label_action(
2967 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2970 case OFPACT_SET_MPLS_TC:
2971 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2974 case OFPACT_SET_MPLS_TTL:
2975 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2978 case OFPACT_DEC_MPLS_TTL:
2979 if (compose_dec_mpls_ttl_action(ctx)) {
2984 case OFPACT_DEC_TTL:
2985 wc->masks.nw_ttl = 0xff;
2986 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2992 /* Nothing to do. */
2995 case OFPACT_MULTIPATH:
2996 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
3000 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
3003 case OFPACT_OUTPUT_REG:
3004 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
3008 xlate_learn_action(ctx, ofpact_get_LEARN(a));
3015 case OFPACT_FIN_TIMEOUT:
3016 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3017 ctx->xout->has_fin_timeout = true;
3018 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
3021 case OFPACT_CLEAR_ACTIONS:
3022 ofpbuf_clear(&ctx->action_set);
3025 case OFPACT_WRITE_ACTIONS:
3026 xlate_write_actions(ctx, a);
3029 case OFPACT_WRITE_METADATA:
3030 metadata = ofpact_get_WRITE_METADATA(a);
3031 flow->metadata &= ~metadata->mask;
3032 flow->metadata |= metadata->metadata & metadata->mask;
3036 /* Not implemented yet. */
3039 case OFPACT_GOTO_TABLE: {
3040 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
3042 ovs_assert(ctx->table_id < ogt->table_id);
3043 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3044 ogt->table_id, true, true);
3049 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
3056 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
3057 const struct flow *flow, struct rule_dpif *rule,
3058 uint16_t tcp_flags, const struct ofpbuf *packet)
3060 xin->ofproto = ofproto;
3062 xin->packet = packet;
3063 xin->may_learn = packet != NULL;
3066 xin->ofpacts = NULL;
3067 xin->ofpacts_len = 0;
3068 xin->tcp_flags = tcp_flags;
3069 xin->resubmit_hook = NULL;
3070 xin->report_hook = NULL;
3071 xin->resubmit_stats = NULL;
3072 xin->skip_wildcards = false;
3076 xlate_out_uninit(struct xlate_out *xout)
3079 ofpbuf_uninit(&xout->odp_actions);
3083 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
3084 * into datapath actions, using 'ctx', and discards the datapath actions. */
3086 xlate_actions_for_side_effects(struct xlate_in *xin)
3088 struct xlate_out xout;
3090 xlate_actions(xin, &xout);
3091 xlate_out_uninit(&xout);
3095 xlate_report(struct xlate_ctx *ctx, const char *s)
3097 if (ctx->xin->report_hook) {
3098 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
3103 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
3106 dst->slow = src->slow;
3107 dst->has_learn = src->has_learn;
3108 dst->has_normal = src->has_normal;
3109 dst->has_fin_timeout = src->has_fin_timeout;
3110 dst->nf_output_iface = src->nf_output_iface;
3111 dst->mirrors = src->mirrors;
3113 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
3114 sizeof dst->odp_actions_stub);
3115 ofpbuf_put(&dst->odp_actions, ofpbuf_data(&src->odp_actions),
3116 ofpbuf_size(&src->odp_actions));
3119 static struct skb_priority_to_dscp *
3120 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
3122 struct skb_priority_to_dscp *pdscp;
3125 hash = hash_int(skb_priority, 0);
3126 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
3127 if (pdscp->skb_priority == skb_priority) {
3135 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
3138 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
3139 *dscp = pdscp ? pdscp->dscp : 0;
3140 return pdscp != NULL;
3144 clear_skb_priorities(struct xport *xport)
3146 struct skb_priority_to_dscp *pdscp, *next;
3148 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
3149 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
3155 actions_output_to_local_port(const struct xlate_ctx *ctx)
3157 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
3158 const struct nlattr *a;
3161 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(&ctx->xout->odp_actions),
3162 ofpbuf_size(&ctx->xout->odp_actions)) {
3163 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
3164 && nl_attr_get_odp_port(a) == local_odp_port) {
3171 /* Thread safe call to xlate_actions__(). */
3173 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
3174 OVS_EXCLUDED(xlate_rwlock)
3176 ovs_rwlock_rdlock(&xlate_rwlock);
3177 xlate_actions__(xin, xout);
3178 ovs_rwlock_unlock(&xlate_rwlock);
3181 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
3182 * into datapath actions in 'odp_actions', using 'ctx'.
3184 * The caller must take responsibility for eventually freeing 'xout', with
3185 * xlate_out_uninit(). */
3187 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
3188 OVS_REQ_RDLOCK(xlate_rwlock)
3190 struct flow_wildcards *wc = &xout->wc;
3191 struct flow *flow = &xin->flow;
3192 struct rule_dpif *rule = NULL;
3194 struct rule_actions *actions = NULL;
3195 enum slow_path_reason special;
3196 const struct ofpact *ofpacts;
3197 struct xport *in_port;
3198 struct flow orig_flow;
3199 struct xlate_ctx ctx;
3204 COVERAGE_INC(xlate_actions);
3206 /* Flow initialization rules:
3207 * - 'base_flow' must match the kernel's view of the packet at the
3208 * time that action processing starts. 'flow' represents any
3209 * transformations we wish to make through actions.
3210 * - By default 'base_flow' and 'flow' are the same since the input
3211 * packet matches the output before any actions are applied.
3212 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3213 * of the received packet as seen by the kernel. If we later output
3214 * to another device without any modifications this will cause us to
3215 * insert a new tag since the original one was stripped off by the
3217 * - Tunnel metadata as received is retained in 'flow'. This allows
3218 * tunnel metadata matching also in later tables.
3219 * Since a kernel action for setting the tunnel metadata will only be
3220 * generated with actual tunnel output, changing the tunnel metadata
3221 * values in 'flow' (such as tun_id) will only have effect with a later
3222 * tunnel output action.
3223 * - Tunnel 'base_flow' is completely cleared since that is what the
3224 * kernel does. If we wish to maintain the original values an action
3225 * needs to be generated. */
3230 ctx.xout->has_learn = false;
3231 ctx.xout->has_normal = false;
3232 ctx.xout->has_fin_timeout = false;
3233 ctx.xout->nf_output_iface = NF_OUT_DROP;
3234 ctx.xout->mirrors = 0;
3235 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3236 sizeof ctx.xout->odp_actions_stub);
3237 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3239 ctx.xbridge = xbridge_lookup(xin->ofproto);
3244 ctx.rule = xin->rule;
3246 ctx.base_flow = *flow;
3247 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3248 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3250 flow_wildcards_init_catchall(wc);
3251 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3252 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3253 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3254 if (is_ip_any(flow)) {
3255 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3257 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
3259 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3260 if (ctx.xbridge->netflow) {
3261 netflow_mask_wc(flow, wc);
3266 ctx.in_group = false;
3267 ctx.orig_skb_priority = flow->skb_priority;
3271 if (!xin->ofpacts && !ctx.rule) {
3272 ctx.table_id = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3273 !xin->skip_wildcards ? wc : NULL,
3274 &rule, ctx.xin->xcache != NULL);
3275 if (ctx.xin->resubmit_stats) {
3276 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3278 if (ctx.xin->xcache) {
3279 struct xc_entry *entry;
3281 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
3282 entry->u.rule = rule;
3286 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3287 xout->use_recirc = false;
3290 ofpacts = xin->ofpacts;
3291 ofpacts_len = xin->ofpacts_len;
3292 } else if (ctx.rule) {
3293 actions = rule_dpif_get_actions(ctx.rule);
3294 ofpacts = actions->ofpacts;
3295 ofpacts_len = actions->ofpacts_len;
3300 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3301 ofpbuf_use_stub(&ctx.action_set,
3302 ctx.action_set_stub, sizeof ctx.action_set_stub);
3304 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3305 /* Do this conditionally because the copy is expensive enough that it
3306 * shows up in profiles. */
3310 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3311 switch (ctx.xbridge->frag) {
3312 case OFPC_FRAG_NORMAL:
3313 /* We must pretend that transport ports are unavailable. */
3314 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3315 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3318 case OFPC_FRAG_DROP:
3321 case OFPC_FRAG_REASM:
3324 case OFPC_FRAG_NX_MATCH:
3325 /* Nothing to do. */
3328 case OFPC_INVALID_TTL_TO_CONTROLLER:
3333 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3334 if (in_port && in_port->is_tunnel) {
3335 if (ctx.xin->resubmit_stats) {
3336 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3338 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3341 if (ctx.xin->xcache) {
3342 struct xc_entry *entry;
3344 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
3345 entry->u.dev.rx = netdev_ref(in_port->netdev);
3346 entry->u.dev.bfd = bfd_ref(in_port->bfd);
3350 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3352 ctx.xout->slow |= special;
3354 size_t sample_actions_len;
3356 if (flow->in_port.ofp_port
3357 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3358 flow->in_port.ofp_port,
3360 ctx.base_flow.vlan_tci = 0;
3363 add_sflow_action(&ctx);
3364 add_ipfix_action(&ctx);
3365 sample_actions_len = ofpbuf_size(&ctx.xout->odp_actions);
3367 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3368 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3370 /* We've let OFPP_NORMAL and the learning action look at the
3371 * packet, so drop it now if forwarding is disabled. */
3372 if (in_port && !xport_stp_forward_state(in_port)) {
3373 ofpbuf_set_size(&ctx.xout->odp_actions, sample_actions_len);
3377 if (ofpbuf_size(&ctx.action_set)) {
3378 xlate_action_set(&ctx);
3381 if (ctx.xbridge->has_in_band
3382 && in_band_must_output_to_local_port(flow)
3383 && !actions_output_to_local_port(&ctx)) {
3384 compose_output_action(&ctx, OFPP_LOCAL);
3387 fix_sflow_action(&ctx);
3389 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3390 add_mirror_actions(&ctx, &orig_flow);
3394 if (nl_attr_oversized(ofpbuf_size(&ctx.xout->odp_actions))) {
3395 /* These datapath actions are too big for a Netlink attribute, so we
3396 * can't hand them to the kernel directly. dpif_execute() can execute
3397 * them one by one with help, so just mark the result as SLOW_ACTION to
3398 * prevent the flow from being installed. */
3399 COVERAGE_INC(xlate_actions_oversize);
3400 ctx.xout->slow |= SLOW_ACTION;
3403 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3404 if (ctx.xin->resubmit_stats) {
3405 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3406 ctx.xin->resubmit_stats->n_packets,
3407 ctx.xin->resubmit_stats->n_bytes);
3409 if (ctx.xin->xcache) {
3410 struct xc_entry *entry;
3412 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
3413 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
3414 entry->u.mirror.mirrors = xout->mirrors;
3418 if (ctx.xbridge->netflow) {
3419 const struct ofpact *ofpacts = actions->ofpacts;
3420 size_t ofpacts_len = actions->ofpacts_len;
3422 /* Only update netflow if we don't have controller flow. We don't
3423 * report NetFlow expiration messages for such facets because they
3424 * are just part of the control logic for the network, not real
3426 if (ofpacts_len == 0
3427 || ofpacts->type != OFPACT_CONTROLLER
3428 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3429 if (ctx.xin->resubmit_stats) {
3430 netflow_flow_update(ctx.xbridge->netflow, flow,
3431 xout->nf_output_iface,
3432 ctx.xin->resubmit_stats);
3434 if (ctx.xin->xcache) {
3435 struct xc_entry *entry;
3437 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
3438 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
3439 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
3440 entry->u.nf.iface = xout->nf_output_iface;
3445 ofpbuf_uninit(&ctx.stack);
3446 ofpbuf_uninit(&ctx.action_set);
3448 /* Clear the metadata and register wildcard masks, because we won't
3449 * use non-header fields as part of the cache. */
3450 flow_wildcards_clear_non_packet_fields(wc);
3452 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3453 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3454 * these fields. The datapath interface, on the other hand, represents
3455 * them with just 8 bits each. This means that if the high 8 bits of the
3456 * masks for these fields somehow become set, then they will get chopped
3457 * off by a round trip through the datapath, and revalidation will spot
3458 * that as an inconsistency and delete the flow. Avoid the problem here by
3459 * making sure that only the low 8 bits of either field can be unwildcarded
3463 wc->masks.tp_src &= htons(UINT8_MAX);
3464 wc->masks.tp_dst &= htons(UINT8_MAX);
3468 /* Sends 'packet' out 'ofport'.
3469 * May modify 'packet'.
3470 * Returns 0 if successful, otherwise a positive errno value. */
3472 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3474 struct xport *xport;
3475 struct ofpact_output output;
3478 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3479 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3480 flow_extract(packet, NULL, &flow);
3481 flow.in_port.ofp_port = OFPP_NONE;
3483 ovs_rwlock_rdlock(&xlate_rwlock);
3484 xport = xport_lookup(ofport);
3486 ovs_rwlock_unlock(&xlate_rwlock);
3489 output.port = xport->ofp_port;
3491 ovs_rwlock_unlock(&xlate_rwlock);
3493 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3494 &output.ofpact, sizeof output,
3498 struct xlate_cache *
3499 xlate_cache_new(void)
3501 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
3503 ofpbuf_init(&xcache->entries, 512);
3507 static struct xc_entry *
3508 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
3510 struct xc_entry *entry;
3512 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
3519 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
3521 if (entry->u.dev.tx) {
3522 netdev_vport_inc_tx(entry->u.dev.tx, stats);
3524 if (entry->u.dev.rx) {
3525 netdev_vport_inc_rx(entry->u.dev.rx, stats);
3527 if (entry->u.dev.bfd) {
3528 bfd_account_rx(entry->u.dev.bfd, stats);
3533 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
3535 struct xbridge *xbridge;
3536 struct xbundle *xbundle;
3537 struct flow_wildcards wc;
3539 xbridge = xbridge_lookup(ofproto);
3544 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
3550 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
3553 /* Push stats and perform side effects of flow translation. */
3555 xlate_push_stats(struct xlate_cache *xcache, bool may_learn,
3556 const struct dpif_flow_stats *stats)
3558 struct xc_entry *entry;
3559 struct ofpbuf entries = xcache->entries;
3561 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3562 switch (entry->type) {
3564 rule_dpif_credit_stats(entry->u.rule, stats);
3567 bond_account(entry->u.bond.bond, entry->u.bond.flow,
3568 entry->u.bond.vid, stats->n_bytes);
3571 xlate_cache_netdev(entry, stats);
3574 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
3575 entry->u.nf.iface, stats);
3578 mirror_update_stats(entry->u.mirror.mbridge,
3579 entry->u.mirror.mirrors,
3580 stats->n_packets, stats->n_bytes);
3584 struct rule_dpif *rule = entry->u.learn.rule;
3586 /* Reset the modified time for a rule that is equivalent to
3587 * the currently cached rule. If the rule is not the exact
3588 * rule we have cached, update the reference that we have. */
3589 entry->u.learn.rule = ofproto_dpif_refresh_rule(rule);
3593 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
3594 entry->u.normal.vlan);
3596 case XC_FIN_TIMEOUT:
3597 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
3598 entry->u.fin.idle, entry->u.fin.hard);
3607 xlate_dev_unref(struct xc_entry *entry)
3609 if (entry->u.dev.tx) {
3610 netdev_close(entry->u.dev.tx);
3612 if (entry->u.dev.rx) {
3613 netdev_close(entry->u.dev.rx);
3615 if (entry->u.dev.bfd) {
3616 bfd_unref(entry->u.dev.bfd);
3621 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
3623 netflow_expire(netflow, flow);
3624 netflow_flow_clear(netflow, flow);
3625 netflow_unref(netflow);
3630 xlate_cache_clear(struct xlate_cache *xcache)
3632 struct xc_entry *entry;
3633 struct ofpbuf entries;
3639 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3640 switch (entry->type) {
3642 rule_dpif_unref(entry->u.rule);
3645 free(entry->u.bond.flow);
3646 bond_unref(entry->u.bond.bond);
3649 xlate_dev_unref(entry);
3652 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
3655 mbridge_unref(entry->u.mirror.mbridge);
3658 /* 'u.learn.rule' is the learned rule. */
3659 rule_dpif_unref(entry->u.learn.rule);
3662 free(entry->u.normal.flow);
3664 case XC_FIN_TIMEOUT:
3665 /* 'u.fin.rule' is always already held as a XC_RULE, which
3666 * has already released it's reference above. */
3673 ofpbuf_clear(&xcache->entries);
3677 xlate_cache_delete(struct xlate_cache *xcache)
3679 xlate_cache_clear(xcache);
3680 ofpbuf_uninit(&xcache->entries);