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 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
774 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
778 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
780 struct stp_port *sp = xport_get_stp_port(xport);
781 struct ofpbuf payload = *packet;
782 struct eth_header *eth = ofpbuf_data(&payload);
784 /* Sink packets on ports that have STP disabled when the bridge has
786 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
790 /* Trim off padding on payload. */
791 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
792 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
795 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
796 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
800 static struct xport *
801 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
805 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
807 if (xport->ofp_port == ofp_port) {
815 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
817 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
818 return xport ? xport->odp_port : ODPP_NONE;
822 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
826 xport = get_ofp_port(ctx->xbridge, ofp_port);
827 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
828 xport->state & OFPUTIL_PS_LINK_DOWN) {
835 static const struct ofputil_bucket *
836 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
840 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
842 struct group_dpif *group;
845 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
850 hit = group_first_live_bucket(ctx, group, depth) != NULL;
852 group_dpif_release(group);
856 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
859 bucket_is_alive(const struct xlate_ctx *ctx,
860 const struct ofputil_bucket *bucket, int depth)
862 if (depth >= MAX_LIVENESS_RECURSION) {
863 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
865 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
866 MAX_LIVENESS_RECURSION);
870 return !ofputil_bucket_has_liveness(bucket) ||
871 (bucket->watch_port != OFPP_ANY &&
872 odp_port_is_alive(ctx, bucket->watch_port)) ||
873 (bucket->watch_group != OFPG_ANY &&
874 group_is_alive(ctx, bucket->watch_group, depth + 1));
877 static const struct ofputil_bucket *
878 group_first_live_bucket(const struct xlate_ctx *ctx,
879 const struct group_dpif *group, int depth)
881 struct ofputil_bucket *bucket;
882 const struct list *buckets;
884 group_dpif_get_buckets(group, &buckets);
885 LIST_FOR_EACH (bucket, list_node, buckets) {
886 if (bucket_is_alive(ctx, bucket, depth)) {
894 static const struct ofputil_bucket *
895 group_best_live_bucket(const struct xlate_ctx *ctx,
896 const struct group_dpif *group,
899 const struct ofputil_bucket *best_bucket = NULL;
900 uint32_t best_score = 0;
903 const struct ofputil_bucket *bucket;
904 const struct list *buckets;
906 group_dpif_get_buckets(group, &buckets);
907 LIST_FOR_EACH (bucket, list_node, buckets) {
908 if (bucket_is_alive(ctx, bucket, 0)) {
909 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
910 if (score >= best_score) {
911 best_bucket = bucket;
922 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
924 return (bundle->vlan_mode != PORT_VLAN_ACCESS
925 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
929 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
931 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
935 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
937 return xbundle != &ofpp_none_bundle
938 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
943 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
945 return xbundle != &ofpp_none_bundle
946 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
951 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
953 return xbundle != &ofpp_none_bundle
954 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
958 static struct xbundle *
959 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
960 bool warn, struct xport **in_xportp)
964 /* Find the port and bundle for the received packet. */
965 xport = get_ofp_port(xbridge, in_port);
969 if (xport && xport->xbundle) {
970 return xport->xbundle;
973 /* Special-case OFPP_NONE, which a controller may use as the ingress
974 * port for traffic that it is sourcing. */
975 if (in_port == OFPP_NONE) {
976 return &ofpp_none_bundle;
979 /* Odd. A few possible reasons here:
981 * - We deleted a port but there are still a few packets queued up
984 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
985 * we don't know about.
987 * - The ofproto client didn't configure the port as part of a bundle.
988 * This is particularly likely to happen if a packet was received on the
989 * port after it was created, but before the client had a chance to
990 * configure its bundle.
993 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
995 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
996 "port %"PRIu16, xbridge->name, in_port);
1002 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1004 const struct xbridge *xbridge = ctx->xbridge;
1005 mirror_mask_t mirrors;
1006 struct xbundle *in_xbundle;
1010 mirrors = ctx->xout->mirrors;
1011 ctx->xout->mirrors = 0;
1013 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1014 ctx->xin->packet != NULL, NULL);
1018 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1020 /* Drop frames on bundles reserved for mirroring. */
1021 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1022 if (ctx->xin->packet != NULL) {
1023 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1024 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1025 "%s, which is reserved exclusively for mirroring",
1026 ctx->xbridge->name, in_xbundle->name);
1028 ofpbuf_clear(&ctx->xout->odp_actions);
1033 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1034 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1037 vlan = input_vid_to_vlan(in_xbundle, vid);
1043 /* Restore the original packet before adding the mirror actions. */
1044 ctx->xin->flow = *orig_flow;
1047 mirror_mask_t dup_mirrors;
1048 struct ofbundle *out;
1049 unsigned long *vlans;
1054 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1055 &vlans, &dup_mirrors, &out, &out_vlan);
1056 ovs_assert(has_mirror);
1059 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1061 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1064 if (!vlan_mirrored) {
1065 mirrors = zero_rightmost_1bit(mirrors);
1069 mirrors &= ~dup_mirrors;
1070 ctx->xout->mirrors |= dup_mirrors;
1072 struct xbundle *out_xbundle = xbundle_lookup(out);
1074 output_normal(ctx, out_xbundle, vlan);
1076 } else if (vlan != out_vlan
1077 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1078 struct xbundle *xbundle;
1080 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1081 if (xbundle_includes_vlan(xbundle, out_vlan)
1082 && !xbundle_mirror_out(xbridge, xbundle)) {
1083 output_normal(ctx, xbundle, out_vlan);
1090 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1091 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1092 * the bundle on which the packet was received, returns the VLAN to which the
1095 * Both 'vid' and the return value are in the range 0...4095. */
1097 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1099 switch (in_xbundle->vlan_mode) {
1100 case PORT_VLAN_ACCESS:
1101 return in_xbundle->vlan;
1104 case PORT_VLAN_TRUNK:
1107 case PORT_VLAN_NATIVE_UNTAGGED:
1108 case PORT_VLAN_NATIVE_TAGGED:
1109 return vid ? vid : in_xbundle->vlan;
1116 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1117 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1120 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1121 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1124 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1126 /* Allow any VID on the OFPP_NONE port. */
1127 if (in_xbundle == &ofpp_none_bundle) {
1131 switch (in_xbundle->vlan_mode) {
1132 case PORT_VLAN_ACCESS:
1135 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1136 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1137 "packet received on port %s configured as VLAN "
1138 "%"PRIu16" access port", vid, in_xbundle->name,
1145 case PORT_VLAN_NATIVE_UNTAGGED:
1146 case PORT_VLAN_NATIVE_TAGGED:
1148 /* Port must always carry its native VLAN. */
1152 case PORT_VLAN_TRUNK:
1153 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1155 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1156 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1157 "received on port %s not configured for trunking "
1158 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1170 /* Given 'vlan', the VLAN that a packet belongs to, and
1171 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1172 * that should be included in the 802.1Q header. (If the return value is 0,
1173 * then the 802.1Q header should only be included in the packet if there is a
1176 * Both 'vlan' and the return value are in the range 0...4095. */
1178 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1180 switch (out_xbundle->vlan_mode) {
1181 case PORT_VLAN_ACCESS:
1184 case PORT_VLAN_TRUNK:
1185 case PORT_VLAN_NATIVE_TAGGED:
1188 case PORT_VLAN_NATIVE_UNTAGGED:
1189 return vlan == out_xbundle->vlan ? 0 : vlan;
1197 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1200 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1202 ovs_be16 tci, old_tci;
1203 struct xport *xport;
1205 vid = output_vlan_to_vid(out_xbundle, vlan);
1206 if (list_is_empty(&out_xbundle->xports)) {
1207 /* Partially configured bundle with no slaves. Drop the packet. */
1209 } else if (!out_xbundle->bond) {
1210 ctx->xout->use_recirc = false;
1211 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1214 struct ofport_dpif *ofport;
1215 struct xlate_recirc *xr = &ctx->xout->recirc;
1216 struct flow_wildcards *wc = &ctx->xout->wc;
1218 if (ctx->xbridge->enable_recirc) {
1219 ctx->xout->use_recirc = bond_may_recirc(
1220 out_xbundle->bond, &xr->recirc_id, &xr->hash_basis);
1222 if (ctx->xout->use_recirc) {
1223 /* Only TCP mode uses recirculation. */
1224 xr->hash_alg = OVS_HASH_ALG_L4;
1225 bond_update_post_recirc_rules(out_xbundle->bond, false);
1227 /* Recirculation does not require unmasking hash fields. */
1232 ofport = bond_choose_output_slave(out_xbundle->bond,
1233 &ctx->xin->flow, wc, vid);
1234 xport = xport_lookup(ofport);
1237 /* No slaves enabled, so drop packet. */
1241 /* If ctx->xout->use_recirc is set, the main thread will handle stats
1242 * accounting for this bond. */
1243 if (!ctx->xout->use_recirc) {
1244 if (ctx->xin->resubmit_stats) {
1245 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1246 ctx->xin->resubmit_stats->n_bytes);
1248 if (ctx->xin->xcache) {
1249 struct xc_entry *entry;
1252 flow = &ctx->xin->flow;
1253 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1254 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1255 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1256 entry->u.bond.vid = vid;
1261 old_tci = *flow_tci;
1263 if (tci || out_xbundle->use_priority_tags) {
1264 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1266 tci |= htons(VLAN_CFI);
1271 compose_output_action(ctx, xport->ofp_port);
1272 *flow_tci = old_tci;
1275 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1276 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1277 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1279 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1281 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1285 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1286 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1290 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1291 if (flow->nw_proto == ARP_OP_REPLY) {
1293 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1294 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1295 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1297 return flow->nw_src == flow->nw_dst;
1303 /* Checks whether a MAC learning update is necessary for MAC learning table
1304 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1307 * Most packets processed through the MAC learning table do not actually
1308 * change it in any way. This function requires only a read lock on the MAC
1309 * learning table, so it is much cheaper in this common case.
1311 * Keep the code here synchronized with that in update_learning_table__()
1314 is_mac_learning_update_needed(const struct mac_learning *ml,
1315 const struct flow *flow,
1316 struct flow_wildcards *wc,
1317 int vlan, struct xbundle *in_xbundle)
1318 OVS_REQ_RDLOCK(ml->rwlock)
1320 struct mac_entry *mac;
1322 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1326 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1327 if (!mac || mac_entry_age(ml, mac)) {
1331 if (is_gratuitous_arp(flow, wc)) {
1332 /* We don't want to learn from gratuitous ARP packets that are
1333 * reflected back over bond slaves so we lock the learning table. */
1334 if (!in_xbundle->bond) {
1336 } else if (mac_entry_is_grat_arp_locked(mac)) {
1341 return mac->port.p != in_xbundle->ofbundle;
1345 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1346 * received on 'in_xbundle' in 'vlan'.
1348 * This code repeats all the checks in is_mac_learning_update_needed() because
1349 * the lock was released between there and here and thus the MAC learning state
1350 * could have changed.
1352 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1355 update_learning_table__(const struct xbridge *xbridge,
1356 const struct flow *flow, struct flow_wildcards *wc,
1357 int vlan, struct xbundle *in_xbundle)
1358 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1360 struct mac_entry *mac;
1362 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1366 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1367 if (is_gratuitous_arp(flow, wc)) {
1368 /* We don't want to learn from gratuitous ARP packets that are
1369 * reflected back over bond slaves so we lock the learning table. */
1370 if (!in_xbundle->bond) {
1371 mac_entry_set_grat_arp_lock(mac);
1372 } else if (mac_entry_is_grat_arp_locked(mac)) {
1377 if (mac->port.p != in_xbundle->ofbundle) {
1378 /* The log messages here could actually be useful in debugging,
1379 * so keep the rate limit relatively high. */
1380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1382 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1383 "on port %s in VLAN %d",
1384 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1385 in_xbundle->name, vlan);
1387 mac->port.p = in_xbundle->ofbundle;
1388 mac_learning_changed(xbridge->ml);
1393 update_learning_table(const struct xbridge *xbridge,
1394 const struct flow *flow, struct flow_wildcards *wc,
1395 int vlan, struct xbundle *in_xbundle)
1399 /* Don't learn the OFPP_NONE port. */
1400 if (in_xbundle == &ofpp_none_bundle) {
1404 /* First try the common case: no change to MAC learning table. */
1405 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1406 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1408 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1411 /* Slow path: MAC learning table might need an update. */
1412 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1413 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1414 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1418 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1419 * dropped. Returns true if they may be forwarded, false if they should be
1422 * 'in_port' must be the xport that corresponds to flow->in_port.
1423 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1425 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1426 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1427 * checked by input_vid_is_valid().
1429 * May also add tags to '*tags', although the current implementation only does
1430 * so in one special case.
1433 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1436 struct xbundle *in_xbundle = in_port->xbundle;
1437 const struct xbridge *xbridge = ctx->xbridge;
1438 struct flow *flow = &ctx->xin->flow;
1440 /* Drop frames for reserved multicast addresses
1441 * only if forward_bpdu option is absent. */
1442 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1443 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1447 if (in_xbundle->bond) {
1448 struct mac_entry *mac;
1450 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1456 xlate_report(ctx, "bonding refused admissibility, dropping");
1459 case BV_DROP_IF_MOVED:
1460 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1461 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1462 if (mac && mac->port.p != in_xbundle->ofbundle &&
1463 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1464 || mac_entry_is_grat_arp_locked(mac))) {
1465 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1466 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1470 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1479 xlate_normal(struct xlate_ctx *ctx)
1481 struct flow_wildcards *wc = &ctx->xout->wc;
1482 struct flow *flow = &ctx->xin->flow;
1483 struct xbundle *in_xbundle;
1484 struct xport *in_port;
1485 struct mac_entry *mac;
1490 ctx->xout->has_normal = true;
1492 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1493 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1494 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1496 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1497 ctx->xin->packet != NULL, &in_port);
1499 xlate_report(ctx, "no input bundle, dropping");
1503 /* Drop malformed frames. */
1504 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1505 !(flow->vlan_tci & htons(VLAN_CFI))) {
1506 if (ctx->xin->packet != NULL) {
1507 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1508 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1509 "VLAN tag received on port %s",
1510 ctx->xbridge->name, in_xbundle->name);
1512 xlate_report(ctx, "partial VLAN tag, dropping");
1516 /* Drop frames on bundles reserved for mirroring. */
1517 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1518 if (ctx->xin->packet != NULL) {
1519 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1520 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1521 "%s, which is reserved exclusively for mirroring",
1522 ctx->xbridge->name, in_xbundle->name);
1524 xlate_report(ctx, "input port is mirror output port, dropping");
1529 vid = vlan_tci_to_vid(flow->vlan_tci);
1530 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1531 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1534 vlan = input_vid_to_vlan(in_xbundle, vid);
1536 /* Check other admissibility requirements. */
1537 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1541 /* Learn source MAC. */
1542 if (ctx->xin->may_learn) {
1543 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1545 if (ctx->xin->xcache) {
1546 struct xc_entry *entry;
1548 /* Save enough info to update mac learning table later. */
1549 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
1550 entry->u.normal.ofproto = ctx->xin->ofproto;
1551 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
1552 entry->u.normal.vlan = vlan;
1555 /* Determine output bundle. */
1556 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1557 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1558 mac_port = mac ? mac->port.p : NULL;
1559 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1562 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1563 if (mac_xbundle && mac_xbundle != in_xbundle) {
1564 xlate_report(ctx, "forwarding to learned port");
1565 output_normal(ctx, mac_xbundle, vlan);
1566 } else if (!mac_xbundle) {
1567 xlate_report(ctx, "learned port is unknown, dropping");
1569 xlate_report(ctx, "learned port is input port, dropping");
1572 struct xbundle *xbundle;
1574 xlate_report(ctx, "no learned MAC for destination, flooding");
1575 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1576 if (xbundle != in_xbundle
1577 && xbundle_includes_vlan(xbundle, vlan)
1578 && xbundle->floodable
1579 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1580 output_normal(ctx, xbundle, vlan);
1583 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1587 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1588 * the number of packets out of UINT32_MAX to sample. The given
1589 * cookie is passed back in the callback for each sampled packet.
1592 compose_sample_action(const struct xbridge *xbridge,
1593 struct ofpbuf *odp_actions,
1594 const struct flow *flow,
1595 const uint32_t probability,
1596 const union user_action_cookie *cookie,
1597 const size_t cookie_size)
1599 size_t sample_offset, actions_offset;
1600 odp_port_t odp_port;
1604 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1606 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1608 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1610 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1611 pid = dpif_port_get_pid(xbridge->dpif, odp_port,
1612 flow_hash_5tuple(flow, 0));
1613 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
1616 nl_msg_end_nested(odp_actions, actions_offset);
1617 nl_msg_end_nested(odp_actions, sample_offset);
1618 return cookie_offset;
1622 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1623 odp_port_t odp_port, unsigned int n_outputs,
1624 union user_action_cookie *cookie)
1628 cookie->type = USER_ACTION_COOKIE_SFLOW;
1629 cookie->sflow.vlan_tci = vlan_tci;
1631 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1632 * port information") for the interpretation of cookie->output. */
1633 switch (n_outputs) {
1635 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1636 cookie->sflow.output = 0x40000000 | 256;
1640 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1642 cookie->sflow.output = ifindex;
1647 /* 0x80000000 means "multiple output ports. */
1648 cookie->sflow.output = 0x80000000 | n_outputs;
1653 /* Compose SAMPLE action for sFlow bridge sampling. */
1655 compose_sflow_action(const struct xbridge *xbridge,
1656 struct ofpbuf *odp_actions,
1657 const struct flow *flow,
1658 odp_port_t odp_port)
1660 uint32_t probability;
1661 union user_action_cookie cookie;
1663 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1667 probability = dpif_sflow_get_probability(xbridge->sflow);
1668 compose_sflow_cookie(xbridge, htons(0), odp_port,
1669 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1671 return compose_sample_action(xbridge, odp_actions, flow, probability,
1672 &cookie, sizeof cookie.sflow);
1676 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1677 uint32_t obs_domain_id, uint32_t obs_point_id,
1678 union user_action_cookie *cookie)
1680 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1681 cookie->flow_sample.probability = probability;
1682 cookie->flow_sample.collector_set_id = collector_set_id;
1683 cookie->flow_sample.obs_domain_id = obs_domain_id;
1684 cookie->flow_sample.obs_point_id = obs_point_id;
1688 compose_ipfix_cookie(union user_action_cookie *cookie)
1690 cookie->type = USER_ACTION_COOKIE_IPFIX;
1693 /* Compose SAMPLE action for IPFIX bridge sampling. */
1695 compose_ipfix_action(const struct xbridge *xbridge,
1696 struct ofpbuf *odp_actions,
1697 const struct flow *flow)
1699 uint32_t probability;
1700 union user_action_cookie cookie;
1702 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1706 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1707 compose_ipfix_cookie(&cookie);
1709 compose_sample_action(xbridge, odp_actions, flow, probability,
1710 &cookie, sizeof cookie.ipfix);
1713 /* SAMPLE action for sFlow must be first action in any given list of
1714 * actions. At this point we do not have all information required to
1715 * build it. So try to build sample action as complete as possible. */
1717 add_sflow_action(struct xlate_ctx *ctx)
1719 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1720 &ctx->xout->odp_actions,
1721 &ctx->xin->flow, ODPP_NONE);
1722 ctx->sflow_odp_port = 0;
1723 ctx->sflow_n_outputs = 0;
1726 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1727 * of actions, eventually after the SAMPLE action for sFlow. */
1729 add_ipfix_action(struct xlate_ctx *ctx)
1731 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1735 /* Fix SAMPLE action according to data collected while composing ODP actions.
1736 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1737 * USERSPACE action's user-cookie which is required for sflow. */
1739 fix_sflow_action(struct xlate_ctx *ctx)
1741 const struct flow *base = &ctx->base_flow;
1742 union user_action_cookie *cookie;
1744 if (!ctx->user_cookie_offset) {
1748 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1749 sizeof cookie->sflow);
1750 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1752 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1753 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1756 static enum slow_path_reason
1757 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1758 const struct xport *xport, const struct ofpbuf *packet)
1760 struct flow_wildcards *wc = &ctx->xout->wc;
1761 const struct xbridge *xbridge = ctx->xbridge;
1765 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1767 cfm_process_heartbeat(xport->cfm, packet);
1770 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1772 bfd_process_packet(xport->bfd, flow, packet);
1773 /* If POLL received, immediately sends FINAL back. */
1774 if (bfd_should_send_packet(xport->bfd)) {
1776 ofproto_dpif_monitor_port_send_soon(xport->ofport);
1778 ofproto_dpif_monitor_port_send_soon_safe(xport->ofport);
1783 } else if (xport->xbundle && xport->xbundle->lacp
1784 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1786 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1789 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1791 stp_process_packet(xport, packet);
1800 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1803 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1804 struct flow_wildcards *wc = &ctx->xout->wc;
1805 struct flow *flow = &ctx->xin->flow;
1806 ovs_be16 flow_vlan_tci;
1807 uint32_t flow_pkt_mark;
1808 uint8_t flow_nw_tos;
1809 odp_port_t out_port, odp_port;
1812 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1813 * before traversing a patch port. */
1814 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 26);
1817 xlate_report(ctx, "Nonexistent output port");
1819 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1820 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1822 } else if (check_stp) {
1823 if (eth_addr_equals(ctx->base_flow.dl_dst, eth_addr_stp)) {
1824 if (!xport_stp_listen_state(xport)) {
1825 xlate_report(ctx, "STP not in listening state, "
1826 "skipping bpdu output");
1829 } else if (!xport_stp_forward_state(xport)) {
1830 xlate_report(ctx, "STP not in forwarding state, "
1836 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1837 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1842 const struct xport *peer = xport->peer;
1843 struct flow old_flow = ctx->xin->flow;
1844 enum slow_path_reason special;
1846 ctx->xbridge = peer->xbridge;
1847 flow->in_port.ofp_port = peer->ofp_port;
1848 flow->metadata = htonll(0);
1849 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1850 memset(flow->regs, 0, sizeof flow->regs);
1852 special = process_special(ctx, &ctx->xin->flow, peer,
1855 ctx->xout->slow |= special;
1856 } else if (may_receive(peer, ctx)) {
1857 if (xport_stp_forward_state(peer)) {
1858 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1860 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1861 * learning action look at the packet, then drop it. */
1862 struct flow old_base_flow = ctx->base_flow;
1863 size_t old_size = ofpbuf_size(&ctx->xout->odp_actions);
1864 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1865 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1866 ctx->xout->mirrors = old_mirrors;
1867 ctx->base_flow = old_base_flow;
1868 ofpbuf_set_size(&ctx->xout->odp_actions, old_size);
1872 ctx->xin->flow = old_flow;
1873 ctx->xbridge = xport->xbridge;
1875 if (ctx->xin->resubmit_stats) {
1876 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1877 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1879 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1882 if (ctx->xin->xcache) {
1883 struct xc_entry *entry;
1885 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1886 entry->u.dev.tx = netdev_ref(xport->netdev);
1887 entry->u.dev.rx = netdev_ref(peer->netdev);
1888 entry->u.dev.bfd = bfd_ref(peer->bfd);
1894 flow_vlan_tci = flow->vlan_tci;
1895 flow_pkt_mark = flow->pkt_mark;
1896 flow_nw_tos = flow->nw_tos;
1898 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1899 wc->masks.nw_tos |= IP_DSCP_MASK;
1900 flow->nw_tos &= ~IP_DSCP_MASK;
1901 flow->nw_tos |= dscp;
1904 if (xport->is_tunnel) {
1905 /* Save tunnel metadata so that changes made due to
1906 * the Logical (tunnel) Port are not visible for any further
1907 * matches, while explicit set actions on tunnel metadata are.
1909 struct flow_tnl flow_tnl = flow->tunnel;
1910 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1911 if (odp_port == ODPP_NONE) {
1912 xlate_report(ctx, "Tunneling decided against output");
1913 goto out; /* restore flow_nw_tos */
1915 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1916 xlate_report(ctx, "Not tunneling to our own address");
1917 goto out; /* restore flow_nw_tos */
1919 if (ctx->xin->resubmit_stats) {
1920 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1922 if (ctx->xin->xcache) {
1923 struct xc_entry *entry;
1925 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
1926 entry->u.dev.tx = netdev_ref(xport->netdev);
1928 out_port = odp_port;
1929 commit_odp_tunnel_action(flow, &ctx->base_flow,
1930 &ctx->xout->odp_actions);
1931 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1933 odp_port = xport->odp_port;
1934 out_port = odp_port;
1935 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1936 ofp_port_t vlandev_port;
1938 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1939 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1940 ofp_port, flow->vlan_tci);
1941 if (vlandev_port != ofp_port) {
1942 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1943 flow->vlan_tci = htons(0);
1948 if (out_port != ODPP_NONE) {
1949 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1950 &ctx->xout->odp_actions,
1953 if (ctx->xout->use_recirc) {
1954 struct ovs_action_hash *act_hash;
1955 struct xlate_recirc *xr = &ctx->xout->recirc;
1958 act_hash = nl_msg_put_unspec_uninit(&ctx->xout->odp_actions,
1959 OVS_ACTION_ATTR_HASH,
1961 act_hash->hash_alg = xr->hash_alg;
1962 act_hash->hash_basis = xr->hash_basis;
1964 /* Recirc action. */
1965 nl_msg_put_u32(&ctx->xout->odp_actions, OVS_ACTION_ATTR_RECIRC,
1968 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1972 ctx->sflow_odp_port = odp_port;
1973 ctx->sflow_n_outputs++;
1974 ctx->xout->nf_output_iface = ofp_port;
1979 flow->vlan_tci = flow_vlan_tci;
1980 flow->pkt_mark = flow_pkt_mark;
1981 flow->nw_tos = flow_nw_tos;
1985 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1987 compose_output_action__(ctx, ofp_port, true);
1991 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1993 struct rule_dpif *old_rule = ctx->rule;
1994 struct rule_actions *actions;
1996 if (ctx->xin->resubmit_stats) {
1997 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1999 if (ctx->xin->xcache) {
2000 struct xc_entry *entry;
2002 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
2003 entry->u.rule = rule;
2004 rule_dpif_ref(rule);
2010 actions = rule_dpif_get_actions(rule);
2011 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
2012 ctx->rule = old_rule;
2017 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
2019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2021 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
2022 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
2023 MAX_RESUBMIT_RECURSION);
2024 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
2025 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
2026 } else if (ofpbuf_size(&ctx->xout->odp_actions) > UINT16_MAX) {
2027 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
2028 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
2029 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
2038 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
2039 bool may_packet_in, bool honor_table_miss)
2041 if (xlate_resubmit_resource_check(ctx)) {
2042 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
2043 bool skip_wildcards = ctx->xin->skip_wildcards;
2044 uint8_t old_table_id = ctx->table_id;
2045 struct rule_dpif *rule;
2046 enum rule_dpif_lookup_verdict verdict;
2047 enum ofputil_port_config config = 0;
2049 ctx->table_id = table_id;
2051 /* Look up a flow with 'in_port' as the input port. Then restore the
2052 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2053 * have surprising behavior). */
2054 ctx->xin->flow.in_port.ofp_port = in_port;
2055 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
2058 ? &ctx->xout->wc : NULL,
2060 &ctx->table_id, &rule);
2061 ctx->xin->flow.in_port.ofp_port = old_in_port;
2063 if (ctx->xin->resubmit_hook) {
2064 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
2068 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
2070 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
2071 if (may_packet_in) {
2072 struct xport *xport;
2074 xport = get_ofp_port(ctx->xbridge,
2075 ctx->xin->flow.in_port.ofp_port);
2076 config = xport ? xport->config : 0;
2079 /* Fall through to drop */
2080 case RULE_DPIF_LOOKUP_VERDICT_DROP:
2081 config = OFPUTIL_PC_NO_PACKET_IN;
2083 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
2084 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
2085 config = OFPUTIL_PC_NO_PACKET_IN;
2092 choose_miss_rule(config, ctx->xbridge->miss_rule,
2093 ctx->xbridge->no_packet_in_rule, &rule);
2097 xlate_recursively(ctx, rule);
2098 rule_dpif_unref(rule);
2101 ctx->table_id = old_table_id;
2109 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
2111 uint64_t action_list_stub[1024 / 8];
2112 struct ofpbuf action_list, action_set;
2114 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
2115 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2117 ofpacts_execute_action_set(&action_list, &action_set);
2119 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2122 ofpbuf_uninit(&action_set);
2123 ofpbuf_uninit(&action_list);
2127 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2129 const struct ofputil_bucket *bucket;
2130 const struct list *buckets;
2131 struct flow old_flow = ctx->xin->flow;
2133 group_dpif_get_buckets(group, &buckets);
2135 LIST_FOR_EACH (bucket, list_node, buckets) {
2136 xlate_group_bucket(ctx, bucket);
2137 /* Roll back flow to previous state.
2138 * This is equivalent to cloning the packet for each bucket.
2140 * As a side effect any subsequently applied actions will
2141 * also effectively be applied to a clone of the packet taken
2142 * just before applying the all or indirect group. */
2143 ctx->xin->flow = old_flow;
2148 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2150 const struct ofputil_bucket *bucket;
2152 bucket = group_first_live_bucket(ctx, group, 0);
2154 xlate_group_bucket(ctx, bucket);
2159 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2161 struct flow_wildcards *wc = &ctx->xout->wc;
2162 const struct ofputil_bucket *bucket;
2165 basis = hash_mac(ctx->xin->flow.dl_dst, 0, 0);
2166 bucket = group_best_live_bucket(ctx, group, basis);
2168 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2169 xlate_group_bucket(ctx, bucket);
2174 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2176 ctx->in_group = true;
2178 switch (group_dpif_get_type(group)) {
2180 case OFPGT11_INDIRECT:
2181 xlate_all_group(ctx, group);
2183 case OFPGT11_SELECT:
2184 xlate_select_group(ctx, group);
2187 xlate_ff_group(ctx, group);
2192 group_dpif_release(group);
2194 ctx->in_group = false;
2198 xlate_group_resource_check(struct xlate_ctx *ctx)
2200 if (!xlate_resubmit_resource_check(ctx)) {
2202 } else if (ctx->in_group) {
2203 /* Prevent nested translation of OpenFlow groups.
2205 * OpenFlow allows this restriction. We enforce this restriction only
2206 * because, with the current architecture, we would otherwise have to
2207 * take a possibly recursive read lock on the ofgroup rwlock, which is
2208 * unsafe given that POSIX allows taking a read lock to block if there
2209 * is a thread blocked on taking the write lock. Other solutions
2210 * without this restriction are also possible, but seem unwarranted
2211 * given the current limited use of groups. */
2212 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2214 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2222 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2224 if (xlate_group_resource_check(ctx)) {
2225 struct group_dpif *group;
2228 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2230 xlate_group_action__(ctx, group);
2240 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2241 const struct ofpact_resubmit *resubmit)
2245 bool may_packet_in = false;
2246 bool honor_table_miss = false;
2248 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2249 /* Still allow missed packets to be sent to the controller
2250 * if resubmitting from an internal table. */
2251 may_packet_in = true;
2252 honor_table_miss = true;
2255 in_port = resubmit->in_port;
2256 if (in_port == OFPP_IN_PORT) {
2257 in_port = ctx->xin->flow.in_port.ofp_port;
2260 table_id = resubmit->table_id;
2261 if (table_id == 255) {
2262 table_id = ctx->table_id;
2265 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2270 flood_packets(struct xlate_ctx *ctx, bool all)
2272 const struct xport *xport;
2274 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2275 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2280 compose_output_action__(ctx, xport->ofp_port, false);
2281 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2282 compose_output_action(ctx, xport->ofp_port);
2286 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2290 execute_controller_action(struct xlate_ctx *ctx, int len,
2291 enum ofp_packet_in_reason reason,
2292 uint16_t controller_id)
2294 struct ofproto_packet_in *pin;
2295 struct ofpbuf *packet;
2296 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2298 ctx->xout->slow |= SLOW_CONTROLLER;
2299 if (!ctx->xin->packet) {
2303 packet = ofpbuf_clone(ctx->xin->packet);
2305 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2306 &ctx->xout->odp_actions,
2309 odp_execute_actions(NULL, packet, false, &md,
2310 ofpbuf_data(&ctx->xout->odp_actions),
2311 ofpbuf_size(&ctx->xout->odp_actions), NULL);
2313 pin = xmalloc(sizeof *pin);
2314 pin->up.packet_len = ofpbuf_size(packet);
2315 pin->up.packet = ofpbuf_steal_data(packet);
2316 pin->up.reason = reason;
2317 pin->up.table_id = ctx->table_id;
2318 pin->up.cookie = (ctx->rule
2319 ? rule_dpif_get_flow_cookie(ctx->rule)
2322 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2324 pin->controller_id = controller_id;
2325 pin->send_len = len;
2326 /* If a rule is a table-miss rule then this is
2327 * a table-miss handled by a table-miss rule.
2329 * Else, if rule is internal and has a controller action,
2330 * the later being implied by the rule being processed here,
2331 * then this is a table-miss handled without a table-miss rule.
2333 * Otherwise this is not a table-miss. */
2334 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
2336 if (rule_dpif_is_table_miss(ctx->rule)) {
2337 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
2338 } else if (rule_dpif_is_internal(ctx->rule)) {
2339 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
2342 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2343 ofpbuf_delete(packet);
2347 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2349 struct flow_wildcards *wc = &ctx->xout->wc;
2350 struct flow *flow = &ctx->xin->flow;
2353 ovs_assert(eth_type_mpls(mpls->ethertype));
2355 n = flow_count_mpls_labels(flow, wc);
2357 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2358 &ctx->xout->odp_actions,
2360 } else if (n >= FLOW_MAX_MPLS_LABELS) {
2361 if (ctx->xin->packet != NULL) {
2362 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2363 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2364 "MPLS push action can't be performed as it would "
2365 "have more MPLS LSEs than the %d supported.",
2366 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2370 } else if (n >= ctx->xbridge->max_mpls_depth) {
2371 COVERAGE_INC(xlate_actions_mpls_overflow);
2372 ctx->xout->slow |= SLOW_ACTION;
2375 flow_push_mpls(flow, n, mpls->ethertype, wc);
2379 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2381 struct flow_wildcards *wc = &ctx->xout->wc;
2382 struct flow *flow = &ctx->xin->flow;
2383 int n = flow_count_mpls_labels(flow, wc);
2385 if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2386 if (ctx->xin->packet != NULL) {
2387 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2388 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2389 "MPLS pop action can't be performed as it has "
2390 "more MPLS LSEs than the %d supported.",
2391 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2394 ofpbuf_clear(&ctx->xout->odp_actions);
2399 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2401 struct flow *flow = &ctx->xin->flow;
2403 if (!is_ip_any(flow)) {
2407 ctx->xout->wc.masks.nw_ttl = 0xff;
2408 if (flow->nw_ttl > 1) {
2414 for (i = 0; i < ids->n_controllers; i++) {
2415 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2419 /* Stop processing for current table. */
2425 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2427 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2428 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2429 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2434 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2436 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2437 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2438 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2443 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2445 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2446 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2447 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2452 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2454 struct flow *flow = &ctx->xin->flow;
2455 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2456 struct flow_wildcards *wc = &ctx->xout->wc;
2458 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2459 if (eth_type_mpls(flow->dl_type)) {
2462 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2465 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2467 /* Stop processing for current table. */
2476 xlate_output_action(struct xlate_ctx *ctx,
2477 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2479 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2481 ctx->xout->nf_output_iface = NF_OUT_DROP;
2485 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2488 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2489 0, may_packet_in, true);
2495 flood_packets(ctx, false);
2498 flood_packets(ctx, true);
2500 case OFPP_CONTROLLER:
2501 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2507 if (port != ctx->xin->flow.in_port.ofp_port) {
2508 compose_output_action(ctx, port);
2510 xlate_report(ctx, "skipping output to input port");
2515 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2516 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2517 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2518 ctx->xout->nf_output_iface = prev_nf_output_iface;
2519 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2520 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2521 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2526 xlate_output_reg_action(struct xlate_ctx *ctx,
2527 const struct ofpact_output_reg *or)
2529 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2530 if (port <= UINT16_MAX) {
2531 union mf_subvalue value;
2533 memset(&value, 0xff, sizeof value);
2534 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2535 xlate_output_action(ctx, u16_to_ofp(port),
2536 or->max_len, false);
2541 xlate_enqueue_action(struct xlate_ctx *ctx,
2542 const struct ofpact_enqueue *enqueue)
2544 ofp_port_t ofp_port = enqueue->port;
2545 uint32_t queue_id = enqueue->queue;
2546 uint32_t flow_priority, priority;
2549 /* Translate queue to priority. */
2550 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2552 /* Fall back to ordinary output action. */
2553 xlate_output_action(ctx, enqueue->port, 0, false);
2557 /* Check output port. */
2558 if (ofp_port == OFPP_IN_PORT) {
2559 ofp_port = ctx->xin->flow.in_port.ofp_port;
2560 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2564 /* Add datapath actions. */
2565 flow_priority = ctx->xin->flow.skb_priority;
2566 ctx->xin->flow.skb_priority = priority;
2567 compose_output_action(ctx, ofp_port);
2568 ctx->xin->flow.skb_priority = flow_priority;
2570 /* Update NetFlow output port. */
2571 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2572 ctx->xout->nf_output_iface = ofp_port;
2573 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2574 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2579 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2581 uint32_t skb_priority;
2583 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2584 ctx->xin->flow.skb_priority = skb_priority;
2586 /* Couldn't translate queue to a priority. Nothing to do. A warning
2587 * has already been logged. */
2592 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2594 const struct xbridge *xbridge = xbridge_;
2605 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2608 port = get_ofp_port(xbridge, ofp_port);
2609 return port ? port->may_enable : false;
2614 xlate_bundle_action(struct xlate_ctx *ctx,
2615 const struct ofpact_bundle *bundle)
2619 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2621 CONST_CAST(struct xbridge *, ctx->xbridge));
2622 if (bundle->dst.field) {
2623 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2626 xlate_output_action(ctx, port, 0, false);
2631 xlate_learn_action(struct xlate_ctx *ctx,
2632 const struct ofpact_learn *learn)
2634 uint64_t ofpacts_stub[1024 / 8];
2635 struct ofputil_flow_mod fm;
2636 struct ofpbuf ofpacts;
2638 ctx->xout->has_learn = true;
2640 learn_mask(learn, &ctx->xout->wc);
2642 if (!ctx->xin->may_learn) {
2646 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2647 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2648 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2649 ofpbuf_uninit(&ofpacts);
2651 if (ctx->xin->xcache) {
2652 struct xc_entry *entry;
2654 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
2655 entry->u.learn.ofproto = ctx->xin->ofproto;
2656 rule_dpif_lookup(ctx->xbridge->ofproto, &ctx->xin->flow, NULL,
2657 &entry->u.learn.rule);
2662 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
2663 uint16_t idle_timeout, uint16_t hard_timeout)
2665 if (tcp_flags & (TCP_FIN | TCP_RST)) {
2666 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
2671 xlate_fin_timeout(struct xlate_ctx *ctx,
2672 const struct ofpact_fin_timeout *oft)
2675 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
2676 oft->fin_idle_timeout, oft->fin_hard_timeout);
2677 if (ctx->xin->xcache) {
2678 struct xc_entry *entry;
2680 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
2681 entry->u.fin.rule = ctx->rule;
2682 entry->u.fin.idle = oft->fin_idle_timeout;
2683 entry->u.fin.hard = oft->fin_hard_timeout;
2684 rule_dpif_ref(ctx->rule);
2690 xlate_sample_action(struct xlate_ctx *ctx,
2691 const struct ofpact_sample *os)
2693 union user_action_cookie cookie;
2694 /* Scale the probability from 16-bit to 32-bit while representing
2695 * the same percentage. */
2696 uint32_t probability = (os->probability << 16) | os->probability;
2698 if (!ctx->xbridge->variable_length_userdata) {
2699 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2701 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2702 "lacks support (needs Linux 3.10+ or kernel module from "
2707 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2708 &ctx->xout->odp_actions,
2711 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2712 os->obs_domain_id, os->obs_point_id, &cookie);
2713 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2714 probability, &cookie, sizeof cookie.flow_sample);
2718 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2720 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2721 ? OFPUTIL_PC_NO_RECV_STP
2722 : OFPUTIL_PC_NO_RECV)) {
2726 /* Only drop packets here if both forwarding and learning are
2727 * disabled. If just learning is enabled, we need to have
2728 * OFPP_NORMAL and the learning action have a look at the packet
2729 * before we can drop it. */
2730 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2738 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2740 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2741 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2742 ofpact_pad(&ctx->action_set);
2746 xlate_action_set(struct xlate_ctx *ctx)
2748 uint64_t action_list_stub[1024 / 64];
2749 struct ofpbuf action_list;
2751 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2752 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2753 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2754 ofpbuf_uninit(&action_list);
2758 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2759 struct xlate_ctx *ctx)
2761 struct flow_wildcards *wc = &ctx->xout->wc;
2762 struct flow *flow = &ctx->xin->flow;
2763 const struct ofpact *a;
2765 /* dl_type already in the mask, not set below. */
2767 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2768 struct ofpact_controller *controller;
2769 const struct ofpact_metadata *metadata;
2770 const struct ofpact_set_field *set_field;
2771 const struct mf_field *mf;
2779 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2780 ofpact_get_OUTPUT(a)->max_len, true);
2784 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2789 case OFPACT_CONTROLLER:
2790 controller = ofpact_get_CONTROLLER(a);
2791 execute_controller_action(ctx, controller->max_len,
2793 controller->controller_id);
2796 case OFPACT_ENQUEUE:
2797 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2800 case OFPACT_SET_VLAN_VID:
2801 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2802 if (flow->vlan_tci & htons(VLAN_CFI) ||
2803 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2804 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2805 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2810 case OFPACT_SET_VLAN_PCP:
2811 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2812 if (flow->vlan_tci & htons(VLAN_CFI) ||
2813 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2814 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2815 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2816 << VLAN_PCP_SHIFT) | VLAN_CFI);
2820 case OFPACT_STRIP_VLAN:
2821 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2822 flow->vlan_tci = htons(0);
2825 case OFPACT_PUSH_VLAN:
2826 /* XXX 802.1AD(QinQ) */
2827 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2828 flow->vlan_tci = htons(VLAN_CFI);
2831 case OFPACT_SET_ETH_SRC:
2832 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2833 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2836 case OFPACT_SET_ETH_DST:
2837 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2838 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2841 case OFPACT_SET_IPV4_SRC:
2842 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2843 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2844 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2848 case OFPACT_SET_IPV4_DST:
2849 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2850 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2851 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2855 case OFPACT_SET_IP_DSCP:
2856 if (is_ip_any(flow)) {
2857 wc->masks.nw_tos |= IP_DSCP_MASK;
2858 flow->nw_tos &= ~IP_DSCP_MASK;
2859 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2863 case OFPACT_SET_IP_ECN:
2864 if (is_ip_any(flow)) {
2865 wc->masks.nw_tos |= IP_ECN_MASK;
2866 flow->nw_tos &= ~IP_ECN_MASK;
2867 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2871 case OFPACT_SET_IP_TTL:
2872 if (is_ip_any(flow)) {
2873 wc->masks.nw_ttl = 0xff;
2874 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2878 case OFPACT_SET_L4_SRC_PORT:
2879 if (is_ip_any(flow)) {
2880 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2881 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2882 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2886 case OFPACT_SET_L4_DST_PORT:
2887 if (is_ip_any(flow)) {
2888 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2889 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2890 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2894 case OFPACT_RESUBMIT:
2895 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2898 case OFPACT_SET_TUNNEL:
2899 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2902 case OFPACT_SET_QUEUE:
2903 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2906 case OFPACT_POP_QUEUE:
2907 flow->skb_priority = ctx->orig_skb_priority;
2910 case OFPACT_REG_MOVE:
2911 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2914 case OFPACT_REG_LOAD:
2915 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2918 case OFPACT_SET_FIELD:
2919 set_field = ofpact_get_SET_FIELD(a);
2920 mf = set_field->field;
2922 /* Set field action only ever overwrites packet's outermost
2923 * applicable header fields. Do nothing if no header exists. */
2924 if (mf->id == MFF_VLAN_VID) {
2925 wc->masks.vlan_tci |= htons(VLAN_CFI);
2926 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
2929 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
2930 /* 'dl_type' is already unwildcarded. */
2931 && !eth_type_mpls(flow->dl_type)) {
2935 mf_mask_field_and_prereqs(mf, &wc->masks);
2936 mf_set_flow_value(mf, &set_field->value, flow);
2939 case OFPACT_STACK_PUSH:
2940 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2944 case OFPACT_STACK_POP:
2945 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2949 case OFPACT_PUSH_MPLS:
2950 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2953 case OFPACT_POP_MPLS:
2954 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2957 case OFPACT_SET_MPLS_LABEL:
2958 compose_set_mpls_label_action(
2959 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2962 case OFPACT_SET_MPLS_TC:
2963 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2966 case OFPACT_SET_MPLS_TTL:
2967 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2970 case OFPACT_DEC_MPLS_TTL:
2971 if (compose_dec_mpls_ttl_action(ctx)) {
2976 case OFPACT_DEC_TTL:
2977 wc->masks.nw_ttl = 0xff;
2978 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2984 /* Nothing to do. */
2987 case OFPACT_MULTIPATH:
2988 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2992 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2995 case OFPACT_OUTPUT_REG:
2996 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
3000 xlate_learn_action(ctx, ofpact_get_LEARN(a));
3007 case OFPACT_FIN_TIMEOUT:
3008 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
3009 ctx->xout->has_fin_timeout = true;
3010 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
3013 case OFPACT_CLEAR_ACTIONS:
3014 ofpbuf_clear(&ctx->action_set);
3017 case OFPACT_WRITE_ACTIONS:
3018 xlate_write_actions(ctx, a);
3021 case OFPACT_WRITE_METADATA:
3022 metadata = ofpact_get_WRITE_METADATA(a);
3023 flow->metadata &= ~metadata->mask;
3024 flow->metadata |= metadata->metadata & metadata->mask;
3028 /* Not implemented yet. */
3031 case OFPACT_GOTO_TABLE: {
3032 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
3034 ovs_assert(ctx->table_id < ogt->table_id);
3035 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3036 ogt->table_id, true, true);
3041 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
3048 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
3049 const struct flow *flow, struct rule_dpif *rule,
3050 uint16_t tcp_flags, const struct ofpbuf *packet)
3052 xin->ofproto = ofproto;
3054 xin->packet = packet;
3055 xin->may_learn = packet != NULL;
3058 xin->ofpacts = NULL;
3059 xin->ofpacts_len = 0;
3060 xin->tcp_flags = tcp_flags;
3061 xin->resubmit_hook = NULL;
3062 xin->report_hook = NULL;
3063 xin->resubmit_stats = NULL;
3064 xin->skip_wildcards = false;
3068 xlate_out_uninit(struct xlate_out *xout)
3071 ofpbuf_uninit(&xout->odp_actions);
3075 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
3076 * into datapath actions, using 'ctx', and discards the datapath actions. */
3078 xlate_actions_for_side_effects(struct xlate_in *xin)
3080 struct xlate_out xout;
3082 xlate_actions(xin, &xout);
3083 xlate_out_uninit(&xout);
3087 xlate_report(struct xlate_ctx *ctx, const char *s)
3089 if (ctx->xin->report_hook) {
3090 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
3095 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
3098 dst->slow = src->slow;
3099 dst->has_learn = src->has_learn;
3100 dst->has_normal = src->has_normal;
3101 dst->has_fin_timeout = src->has_fin_timeout;
3102 dst->nf_output_iface = src->nf_output_iface;
3103 dst->mirrors = src->mirrors;
3105 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
3106 sizeof dst->odp_actions_stub);
3107 ofpbuf_put(&dst->odp_actions, ofpbuf_data(&src->odp_actions),
3108 ofpbuf_size(&src->odp_actions));
3111 static struct skb_priority_to_dscp *
3112 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
3114 struct skb_priority_to_dscp *pdscp;
3117 hash = hash_int(skb_priority, 0);
3118 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
3119 if (pdscp->skb_priority == skb_priority) {
3127 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
3130 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
3131 *dscp = pdscp ? pdscp->dscp : 0;
3132 return pdscp != NULL;
3136 clear_skb_priorities(struct xport *xport)
3138 struct skb_priority_to_dscp *pdscp, *next;
3140 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
3141 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
3147 actions_output_to_local_port(const struct xlate_ctx *ctx)
3149 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
3150 const struct nlattr *a;
3153 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(&ctx->xout->odp_actions),
3154 ofpbuf_size(&ctx->xout->odp_actions)) {
3155 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
3156 && nl_attr_get_odp_port(a) == local_odp_port) {
3163 /* Thread safe call to xlate_actions__(). */
3165 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
3166 OVS_EXCLUDED(xlate_rwlock)
3168 ovs_rwlock_rdlock(&xlate_rwlock);
3169 xlate_actions__(xin, xout);
3170 ovs_rwlock_unlock(&xlate_rwlock);
3173 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
3174 * into datapath actions in 'odp_actions', using 'ctx'.
3176 * The caller must take responsibility for eventually freeing 'xout', with
3177 * xlate_out_uninit(). */
3179 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
3180 OVS_REQ_RDLOCK(xlate_rwlock)
3182 struct flow_wildcards *wc = &xout->wc;
3183 struct flow *flow = &xin->flow;
3184 struct rule_dpif *rule = NULL;
3186 struct rule_actions *actions = NULL;
3187 enum slow_path_reason special;
3188 const struct ofpact *ofpacts;
3189 struct xport *in_port;
3190 struct flow orig_flow;
3191 struct xlate_ctx ctx;
3196 COVERAGE_INC(xlate_actions);
3198 /* Flow initialization rules:
3199 * - 'base_flow' must match the kernel's view of the packet at the
3200 * time that action processing starts. 'flow' represents any
3201 * transformations we wish to make through actions.
3202 * - By default 'base_flow' and 'flow' are the same since the input
3203 * packet matches the output before any actions are applied.
3204 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3205 * of the received packet as seen by the kernel. If we later output
3206 * to another device without any modifications this will cause us to
3207 * insert a new tag since the original one was stripped off by the
3209 * - Tunnel metadata as received is retained in 'flow'. This allows
3210 * tunnel metadata matching also in later tables.
3211 * Since a kernel action for setting the tunnel metadata will only be
3212 * generated with actual tunnel output, changing the tunnel metadata
3213 * values in 'flow' (such as tun_id) will only have effect with a later
3214 * tunnel output action.
3215 * - Tunnel 'base_flow' is completely cleared since that is what the
3216 * kernel does. If we wish to maintain the original values an action
3217 * needs to be generated. */
3222 ctx.xout->has_learn = false;
3223 ctx.xout->has_normal = false;
3224 ctx.xout->has_fin_timeout = false;
3225 ctx.xout->nf_output_iface = NF_OUT_DROP;
3226 ctx.xout->mirrors = 0;
3227 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3228 sizeof ctx.xout->odp_actions_stub);
3229 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3231 ctx.xbridge = xbridge_lookup(xin->ofproto);
3236 ctx.rule = xin->rule;
3238 ctx.base_flow = *flow;
3239 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3240 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3242 flow_wildcards_init_catchall(wc);
3243 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3244 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3245 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3246 if (is_ip_any(flow)) {
3247 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3249 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
3251 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3252 if (ctx.xbridge->netflow) {
3253 netflow_mask_wc(flow, wc);
3258 ctx.in_group = false;
3259 ctx.orig_skb_priority = flow->skb_priority;
3263 if (!xin->ofpacts && !ctx.rule) {
3264 ctx.table_id = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3265 !xin->skip_wildcards ? wc : NULL,
3267 if (ctx.xin->resubmit_stats) {
3268 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3270 if (ctx.xin->xcache) {
3271 struct xc_entry *entry;
3273 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
3274 rule_dpif_ref(rule);
3275 entry->u.rule = rule;
3279 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3280 xout->use_recirc = false;
3283 ofpacts = xin->ofpacts;
3284 ofpacts_len = xin->ofpacts_len;
3285 } else if (ctx.rule) {
3286 actions = rule_dpif_get_actions(ctx.rule);
3287 ofpacts = actions->ofpacts;
3288 ofpacts_len = actions->ofpacts_len;
3293 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3294 ofpbuf_use_stub(&ctx.action_set,
3295 ctx.action_set_stub, sizeof ctx.action_set_stub);
3297 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3298 /* Do this conditionally because the copy is expensive enough that it
3299 * shows up in profiles. */
3303 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3304 switch (ctx.xbridge->frag) {
3305 case OFPC_FRAG_NORMAL:
3306 /* We must pretend that transport ports are unavailable. */
3307 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3308 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3311 case OFPC_FRAG_DROP:
3314 case OFPC_FRAG_REASM:
3317 case OFPC_FRAG_NX_MATCH:
3318 /* Nothing to do. */
3321 case OFPC_INVALID_TTL_TO_CONTROLLER:
3326 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3327 if (in_port && in_port->is_tunnel) {
3328 if (ctx.xin->resubmit_stats) {
3329 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3331 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3334 if (ctx.xin->xcache) {
3335 struct xc_entry *entry;
3337 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
3338 entry->u.dev.rx = netdev_ref(in_port->netdev);
3339 entry->u.dev.bfd = bfd_ref(in_port->bfd);
3343 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3345 ctx.xout->slow |= special;
3347 size_t sample_actions_len;
3349 if (flow->in_port.ofp_port
3350 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3351 flow->in_port.ofp_port,
3353 ctx.base_flow.vlan_tci = 0;
3356 add_sflow_action(&ctx);
3357 add_ipfix_action(&ctx);
3358 sample_actions_len = ofpbuf_size(&ctx.xout->odp_actions);
3360 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3361 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3363 /* We've let OFPP_NORMAL and the learning action look at the
3364 * packet, so drop it now if forwarding is disabled. */
3365 if (in_port && !xport_stp_forward_state(in_port)) {
3366 ofpbuf_set_size(&ctx.xout->odp_actions, sample_actions_len);
3370 if (ofpbuf_size(&ctx.action_set)) {
3371 xlate_action_set(&ctx);
3374 if (ctx.xbridge->has_in_band
3375 && in_band_must_output_to_local_port(flow)
3376 && !actions_output_to_local_port(&ctx)) {
3377 compose_output_action(&ctx, OFPP_LOCAL);
3380 fix_sflow_action(&ctx);
3382 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3383 add_mirror_actions(&ctx, &orig_flow);
3387 if (nl_attr_oversized(ofpbuf_size(&ctx.xout->odp_actions))) {
3388 /* These datapath actions are too big for a Netlink attribute, so we
3389 * can't hand them to the kernel directly. dpif_execute() can execute
3390 * them one by one with help, so just mark the result as SLOW_ACTION to
3391 * prevent the flow from being installed. */
3392 COVERAGE_INC(xlate_actions_oversize);
3393 ctx.xout->slow |= SLOW_ACTION;
3396 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3397 if (ctx.xin->resubmit_stats) {
3398 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3399 ctx.xin->resubmit_stats->n_packets,
3400 ctx.xin->resubmit_stats->n_bytes);
3402 if (ctx.xin->xcache) {
3403 struct xc_entry *entry;
3405 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
3406 entry->u.mirror.mbridge = mbridge_ref(ctx.xbridge->mbridge);
3407 entry->u.mirror.mirrors = xout->mirrors;
3411 if (ctx.xbridge->netflow) {
3412 const struct ofpact *ofpacts = actions->ofpacts;
3413 size_t ofpacts_len = actions->ofpacts_len;
3415 /* Only update netflow if we don't have controller flow. We don't
3416 * report NetFlow expiration messages for such facets because they
3417 * are just part of the control logic for the network, not real
3419 if (ofpacts_len == 0
3420 || ofpacts->type != OFPACT_CONTROLLER
3421 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3422 if (ctx.xin->resubmit_stats) {
3423 netflow_flow_update(ctx.xbridge->netflow, flow,
3424 xout->nf_output_iface,
3425 ctx.xin->resubmit_stats);
3427 if (ctx.xin->xcache) {
3428 struct xc_entry *entry;
3430 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
3431 entry->u.nf.netflow = netflow_ref(ctx.xbridge->netflow);
3432 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
3433 entry->u.nf.iface = xout->nf_output_iface;
3438 ofpbuf_uninit(&ctx.stack);
3439 ofpbuf_uninit(&ctx.action_set);
3441 /* Clear the metadata and register wildcard masks, because we won't
3442 * use non-header fields as part of the cache. */
3443 flow_wildcards_clear_non_packet_fields(wc);
3445 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3446 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3447 * these fields. The datapath interface, on the other hand, represents
3448 * them with just 8 bits each. This means that if the high 8 bits of the
3449 * masks for these fields somehow become set, then they will get chopped
3450 * off by a round trip through the datapath, and revalidation will spot
3451 * that as an inconsistency and delete the flow. Avoid the problem here by
3452 * making sure that only the low 8 bits of either field can be unwildcarded
3456 wc->masks.tp_src &= htons(UINT8_MAX);
3457 wc->masks.tp_dst &= htons(UINT8_MAX);
3461 rule_dpif_unref(rule);
3464 /* Sends 'packet' out 'ofport'.
3465 * May modify 'packet'.
3466 * Returns 0 if successful, otherwise a positive errno value. */
3468 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3470 struct xport *xport;
3471 struct ofpact_output output;
3474 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3475 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3476 flow_extract(packet, NULL, &flow);
3477 flow.in_port.ofp_port = OFPP_NONE;
3479 ovs_rwlock_rdlock(&xlate_rwlock);
3480 xport = xport_lookup(ofport);
3482 ovs_rwlock_unlock(&xlate_rwlock);
3485 output.port = xport->ofp_port;
3487 ovs_rwlock_unlock(&xlate_rwlock);
3489 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3490 &output.ofpact, sizeof output,
3494 struct xlate_cache *
3495 xlate_cache_new(void)
3497 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
3499 ofpbuf_init(&xcache->entries, 512);
3503 static struct xc_entry *
3504 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
3506 struct xc_entry *entry;
3508 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
3515 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
3517 if (entry->u.dev.tx) {
3518 netdev_vport_inc_tx(entry->u.dev.tx, stats);
3520 if (entry->u.dev.rx) {
3521 netdev_vport_inc_rx(entry->u.dev.rx, stats);
3523 if (entry->u.dev.bfd) {
3524 bfd_account_rx(entry->u.dev.bfd, stats);
3529 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
3531 struct xbridge *xbridge;
3532 struct xbundle *xbundle;
3533 struct flow_wildcards wc;
3535 xbridge = xbridge_lookup(ofproto);
3540 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
3546 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
3549 /* Push stats and perform side effects of flow translation. */
3551 xlate_push_stats(struct xlate_cache *xcache, bool may_learn,
3552 const struct dpif_flow_stats *stats)
3554 struct xc_entry *entry;
3555 struct ofpbuf entries = xcache->entries;
3557 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3558 switch (entry->type) {
3560 rule_dpif_credit_stats(entry->u.rule, stats);
3563 bond_account(entry->u.bond.bond, entry->u.bond.flow,
3564 entry->u.bond.vid, stats->n_bytes);
3567 xlate_cache_netdev(entry, stats);
3570 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
3571 entry->u.nf.iface, stats);
3574 mirror_update_stats(entry->u.mirror.mbridge,
3575 entry->u.mirror.mirrors,
3576 stats->n_packets, stats->n_bytes);
3580 struct rule_dpif *rule = entry->u.learn.rule;
3582 /* Reset the modified time for a rule that is equivalent to
3583 * the currently cached rule. If the rule is not the exact
3584 * rule wehave cached, update the reference that we have. */
3585 entry->u.learn.rule = ofproto_dpif_refresh_rule(rule);
3589 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
3590 entry->u.normal.vlan);
3592 case XC_FIN_TIMEOUT:
3593 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
3594 entry->u.fin.idle, entry->u.fin.hard);
3603 xlate_dev_unref(struct xc_entry *entry)
3605 if (entry->u.dev.tx) {
3606 netdev_close(entry->u.dev.tx);
3608 if (entry->u.dev.rx) {
3609 netdev_close(entry->u.dev.rx);
3611 if (entry->u.dev.bfd) {
3612 bfd_unref(entry->u.dev.bfd);
3617 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
3619 netflow_expire(netflow, flow);
3620 netflow_flow_clear(netflow, flow);
3621 netflow_unref(netflow);
3626 xlate_cache_clear(struct xlate_cache *xcache)
3628 struct xc_entry *entry;
3629 struct ofpbuf entries;
3635 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
3636 switch (entry->type) {
3638 rule_dpif_unref(entry->u.rule);
3641 free(entry->u.bond.flow);
3642 bond_unref(entry->u.bond.bond);
3645 xlate_dev_unref(entry);
3648 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
3651 mbridge_unref(entry->u.mirror.mbridge);
3654 rule_dpif_unref(entry->u.learn.rule);
3657 free(entry->u.normal.flow);
3659 case XC_FIN_TIMEOUT:
3660 rule_dpif_unref(entry->u.fin.rule);
3667 ofpbuf_clear(&xcache->entries);
3671 xlate_cache_delete(struct xlate_cache *xcache)
3673 xlate_cache_clear(xcache);
3674 ofpbuf_uninit(&xcache->entries);