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
62 /* Maximum number of resubmit actions in a flow translation, whether they are
63 * recursive or not. */
64 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
66 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
69 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
70 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
72 struct list xbundles; /* Owned xbundles. */
73 struct hmap xports; /* Indexed by ofp_port. */
75 char *name; /* Name used in log messages. */
76 struct dpif *dpif; /* Datapath interface. */
77 struct mac_learning *ml; /* Mac learning handle. */
78 struct mbridge *mbridge; /* Mirroring. */
79 struct dpif_sflow *sflow; /* SFlow handle, or null. */
80 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
81 struct netflow *netflow; /* Netflow handle, or null. */
82 struct stp *stp; /* STP or null if disabled. */
84 /* Special rules installed by ofproto-dpif. */
85 struct rule_dpif *miss_rule;
86 struct rule_dpif *no_packet_in_rule;
88 enum ofp_config_flags frag; /* Fragmentation handling. */
89 bool has_in_band; /* Bridge has in band control? */
90 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
92 /* True if the datapath supports variable-length
93 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
94 * False if the datapath supports only 8-byte (or shorter) userdata. */
95 bool variable_length_userdata;
97 /* Number of MPLS label stack entries that the datapath supports
99 size_t max_mpls_depth;
103 struct hmap_node hmap_node; /* In global 'xbundles' map. */
104 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
106 struct list list_node; /* In parent 'xbridges' list. */
107 struct xbridge *xbridge; /* Parent xbridge. */
109 struct list xports; /* Contains "struct xport"s. */
111 char *name; /* Name used in log messages. */
112 struct bond *bond; /* Nonnull iff more than one port. */
113 struct lacp *lacp; /* LACP handle or null. */
115 enum port_vlan_mode vlan_mode; /* VLAN mode. */
116 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
117 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
118 * NULL if all VLANs are trunked. */
119 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
120 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
124 struct hmap_node hmap_node; /* Node in global 'xports' map. */
125 struct ofport_dpif *ofport; /* Key in global 'xports map. */
127 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
128 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
130 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
132 struct list bundle_node; /* In parent xbundle (if it exists). */
133 struct xbundle *xbundle; /* Parent xbundle or null. */
135 struct netdev *netdev; /* 'ofport''s netdev. */
137 struct xbridge *xbridge; /* Parent bridge. */
138 struct xport *peer; /* Patch port peer or null. */
140 enum ofputil_port_config config; /* OpenFlow port configuration. */
141 enum ofputil_port_state state; /* OpenFlow port state. */
142 int stp_port_no; /* STP port number or -1 if not in use. */
144 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
146 bool may_enable; /* May be enabled in bonds. */
147 bool is_tunnel; /* Is a tunnel port. */
149 struct cfm *cfm; /* CFM handle or null. */
150 struct bfd *bfd; /* BFD handle or null. */
154 struct xlate_in *xin;
155 struct xlate_out *xout;
157 const struct xbridge *xbridge;
159 /* Flow at the last commit. */
160 struct flow base_flow;
162 /* Tunnel IP destination address as received. This is stored separately
163 * as the base_flow.tunnel is cleared on init to reflect the datapath
164 * behavior. Used to make sure not to send tunneled output to ourselves,
165 * which might lead to an infinite loop. This could happen easily
166 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
167 * actually set the tun_dst field. */
168 ovs_be32 orig_tunnel_ip_dst;
170 /* Stack for the push and pop actions. Each stack element is of type
171 * "union mf_subvalue". */
172 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
175 /* The rule that we are currently translating, or NULL. */
176 struct rule_dpif *rule;
178 /* Resubmit statistics, via xlate_table_action(). */
179 int recurse; /* Current resubmit nesting depth. */
180 int resubmits; /* Total number of resubmits. */
182 uint32_t orig_skb_priority; /* Priority when packet arrived. */
183 uint8_t table_id; /* OpenFlow table ID where flow was found. */
184 uint32_t sflow_n_outputs; /* Number of output ports. */
185 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
186 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
187 bool exit; /* No further actions should be processed. */
189 /* OpenFlow 1.1+ action set.
191 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
192 * When translation is otherwise complete, ofpacts_execute_action_set()
193 * converts it to a set of "struct ofpact"s that can be translated into
194 * datapath actions. */
195 struct ofpbuf action_set; /* Action set. */
196 uint64_t action_set_stub[1024 / 8];
199 /* A controller may use OFPP_NONE as the ingress port to indicate that
200 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
201 * when an input bundle is needed for validation (e.g., mirroring or
202 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
203 * any 'port' structs, so care must be taken when dealing with it.
204 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
205 static struct xbundle ofpp_none_bundle;
207 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
208 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
209 * traffic egressing the 'ofport' with that priority should be marked with. */
210 struct skb_priority_to_dscp {
211 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
212 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
214 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
217 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
218 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
219 static struct hmap xports = HMAP_INITIALIZER(&xports);
221 static bool may_receive(const struct xport *, struct xlate_ctx *);
222 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
224 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
225 OVS_REQ_RDLOCK(xlate_rwlock);
226 static void xlate_normal(struct xlate_ctx *);
227 static void xlate_report(struct xlate_ctx *, const char *);
228 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
229 uint8_t table_id, bool may_packet_in);
230 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
231 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
232 static void output_normal(struct xlate_ctx *, const struct xbundle *,
234 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
236 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
237 static struct xbundle *xbundle_lookup(const struct ofbundle *);
238 static struct xport *xport_lookup(const struct ofport_dpif *);
239 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
240 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
241 uint32_t skb_priority);
242 static void clear_skb_priorities(struct xport *);
243 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
247 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
248 struct dpif *dpif, struct rule_dpif *miss_rule,
249 struct rule_dpif *no_packet_in_rule,
250 const struct mac_learning *ml, struct stp *stp,
251 const struct mbridge *mbridge,
252 const struct dpif_sflow *sflow,
253 const struct dpif_ipfix *ipfix,
254 const struct netflow *netflow, enum ofp_config_flags frag,
255 bool forward_bpdu, bool has_in_band,
256 bool variable_length_userdata,
257 size_t max_mpls_depth)
259 struct xbridge *xbridge = xbridge_lookup(ofproto);
262 xbridge = xzalloc(sizeof *xbridge);
263 xbridge->ofproto = ofproto;
265 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
266 hmap_init(&xbridge->xports);
267 list_init(&xbridge->xbundles);
270 if (xbridge->ml != ml) {
271 mac_learning_unref(xbridge->ml);
272 xbridge->ml = mac_learning_ref(ml);
275 if (xbridge->mbridge != mbridge) {
276 mbridge_unref(xbridge->mbridge);
277 xbridge->mbridge = mbridge_ref(mbridge);
280 if (xbridge->sflow != sflow) {
281 dpif_sflow_unref(xbridge->sflow);
282 xbridge->sflow = dpif_sflow_ref(sflow);
285 if (xbridge->ipfix != ipfix) {
286 dpif_ipfix_unref(xbridge->ipfix);
287 xbridge->ipfix = dpif_ipfix_ref(ipfix);
290 if (xbridge->stp != stp) {
291 stp_unref(xbridge->stp);
292 xbridge->stp = stp_ref(stp);
295 if (xbridge->netflow != netflow) {
296 netflow_unref(xbridge->netflow);
297 xbridge->netflow = netflow_ref(netflow);
301 xbridge->name = xstrdup(name);
303 xbridge->dpif = dpif;
304 xbridge->forward_bpdu = forward_bpdu;
305 xbridge->has_in_band = has_in_band;
306 xbridge->frag = frag;
307 xbridge->miss_rule = miss_rule;
308 xbridge->no_packet_in_rule = no_packet_in_rule;
309 xbridge->variable_length_userdata = variable_length_userdata;
310 xbridge->max_mpls_depth = max_mpls_depth;
314 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
316 struct xbridge *xbridge = xbridge_lookup(ofproto);
317 struct xbundle *xbundle, *next_xbundle;
318 struct xport *xport, *next_xport;
324 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
325 xlate_ofport_remove(xport->ofport);
328 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
329 xlate_bundle_remove(xbundle->ofbundle);
332 hmap_remove(&xbridges, &xbridge->hmap_node);
333 mac_learning_unref(xbridge->ml);
334 mbridge_unref(xbridge->mbridge);
335 dpif_sflow_unref(xbridge->sflow);
336 dpif_ipfix_unref(xbridge->ipfix);
337 stp_unref(xbridge->stp);
338 hmap_destroy(&xbridge->xports);
344 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
345 const char *name, enum port_vlan_mode vlan_mode, int vlan,
346 unsigned long *trunks, bool use_priority_tags,
347 const struct bond *bond, const struct lacp *lacp,
350 struct xbundle *xbundle = xbundle_lookup(ofbundle);
353 xbundle = xzalloc(sizeof *xbundle);
354 xbundle->ofbundle = ofbundle;
355 xbundle->xbridge = xbridge_lookup(ofproto);
357 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
358 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
359 list_init(&xbundle->xports);
362 ovs_assert(xbundle->xbridge);
365 xbundle->name = xstrdup(name);
367 xbundle->vlan_mode = vlan_mode;
368 xbundle->vlan = vlan;
369 xbundle->trunks = trunks;
370 xbundle->use_priority_tags = use_priority_tags;
371 xbundle->floodable = floodable;
373 if (xbundle->bond != bond) {
374 bond_unref(xbundle->bond);
375 xbundle->bond = bond_ref(bond);
378 if (xbundle->lacp != lacp) {
379 lacp_unref(xbundle->lacp);
380 xbundle->lacp = lacp_ref(lacp);
385 xlate_bundle_remove(struct ofbundle *ofbundle)
387 struct xbundle *xbundle = xbundle_lookup(ofbundle);
388 struct xport *xport, *next;
394 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
395 list_remove(&xport->bundle_node);
396 xport->xbundle = NULL;
399 hmap_remove(&xbundles, &xbundle->hmap_node);
400 list_remove(&xbundle->list_node);
401 bond_unref(xbundle->bond);
402 lacp_unref(xbundle->lacp);
408 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
409 struct ofport_dpif *ofport, ofp_port_t ofp_port,
410 odp_port_t odp_port, const struct netdev *netdev,
411 const struct cfm *cfm, const struct bfd *bfd,
412 struct ofport_dpif *peer, int stp_port_no,
413 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
414 enum ofputil_port_config config,
415 enum ofputil_port_state state, bool is_tunnel,
418 struct xport *xport = xport_lookup(ofport);
422 xport = xzalloc(sizeof *xport);
423 xport->ofport = ofport;
424 xport->xbridge = xbridge_lookup(ofproto);
425 xport->ofp_port = ofp_port;
427 hmap_init(&xport->skb_priorities);
428 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
429 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
430 hash_ofp_port(xport->ofp_port));
433 ovs_assert(xport->ofp_port == ofp_port);
435 xport->config = config;
436 xport->state = state;
437 xport->stp_port_no = stp_port_no;
438 xport->is_tunnel = is_tunnel;
439 xport->may_enable = may_enable;
440 xport->odp_port = odp_port;
442 if (xport->netdev != netdev) {
443 netdev_close(xport->netdev);
444 xport->netdev = netdev_ref(netdev);
447 if (xport->cfm != cfm) {
448 cfm_unref(xport->cfm);
449 xport->cfm = cfm_ref(cfm);
452 if (xport->bfd != bfd) {
453 bfd_unref(xport->bfd);
454 xport->bfd = bfd_ref(bfd);
458 xport->peer->peer = NULL;
460 xport->peer = xport_lookup(peer);
462 xport->peer->peer = xport;
465 if (xport->xbundle) {
466 list_remove(&xport->bundle_node);
468 xport->xbundle = xbundle_lookup(ofbundle);
469 if (xport->xbundle) {
470 list_insert(&xport->xbundle->xports, &xport->bundle_node);
473 clear_skb_priorities(xport);
474 for (i = 0; i < n_qdscp; i++) {
475 struct skb_priority_to_dscp *pdscp;
476 uint32_t skb_priority;
478 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
483 pdscp = xmalloc(sizeof *pdscp);
484 pdscp->skb_priority = skb_priority;
485 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
486 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
487 hash_int(pdscp->skb_priority, 0));
492 xlate_ofport_remove(struct ofport_dpif *ofport)
494 struct xport *xport = xport_lookup(ofport);
501 xport->peer->peer = NULL;
505 if (xport->xbundle) {
506 list_remove(&xport->bundle_node);
509 clear_skb_priorities(xport);
510 hmap_destroy(&xport->skb_priorities);
512 hmap_remove(&xports, &xport->hmap_node);
513 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
515 netdev_close(xport->netdev);
516 cfm_unref(xport->cfm);
517 bfd_unref(xport->bfd);
521 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
522 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
523 * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
524 * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
525 * 'netflow' with the appropriate handles for those protocols if they're
526 * enabled. Caller is responsible for unrefing them.
528 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
529 * 'flow''s in_port to OFPP_NONE.
531 * This function does post-processing on data returned from
532 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
533 * of the upcall processing logic. In particular, if the extracted in_port is
534 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
535 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
536 * a VLAN header onto 'packet' (if it is nonnull).
538 * Similarly, this function also includes some logic to help with tunnels. It
539 * may modify 'flow' as necessary to make the tunneling implementation
540 * transparent to the upcall processing logic.
542 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
543 * or some other positive errno if there are other problems. */
545 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
546 const struct nlattr *key, size_t key_len, struct flow *flow,
547 struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
548 struct dpif_sflow **sflow, struct netflow **netflow,
549 odp_port_t *odp_in_port)
551 const struct xport *xport;
554 ovs_rwlock_rdlock(&xlate_rwlock);
555 if (odp_flow_key_to_flow(key, key_len, flow) == ODP_FIT_ERROR) {
561 *odp_in_port = flow->in_port.odp_port;
564 xport = xport_lookup(tnl_port_should_receive(flow)
565 ? tnl_port_receive(flow)
566 : odp_port_to_ofport(backer, flow->in_port.odp_port));
568 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
573 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
575 /* Make the packet resemble the flow, so that it gets sent to
576 * an OpenFlow controller properly, so that it looks correct
577 * for sFlow, and so that flow_extract() will get the correct
578 * vlan_tci if it is called on 'packet'. */
579 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
585 *ofproto = xport->xbridge->ofproto;
589 *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
593 *sflow = dpif_sflow_ref(xport->xbridge->sflow);
597 *netflow = netflow_ref(xport->xbridge->netflow);
601 ovs_rwlock_unlock(&xlate_rwlock);
605 static struct xbridge *
606 xbridge_lookup(const struct ofproto_dpif *ofproto)
608 struct xbridge *xbridge;
614 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
616 if (xbridge->ofproto == ofproto) {
623 static struct xbundle *
624 xbundle_lookup(const struct ofbundle *ofbundle)
626 struct xbundle *xbundle;
632 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
634 if (xbundle->ofbundle == ofbundle) {
641 static struct xport *
642 xport_lookup(const struct ofport_dpif *ofport)
650 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
652 if (xport->ofport == ofport) {
659 static struct stp_port *
660 xport_get_stp_port(const struct xport *xport)
662 return xport->xbridge->stp && xport->stp_port_no != -1
663 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
667 static enum stp_state
668 xport_stp_learn_state(const struct xport *xport)
670 struct stp_port *sp = xport_get_stp_port(xport);
671 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
675 xport_stp_forward_state(const struct xport *xport)
677 struct stp_port *sp = xport_get_stp_port(xport);
678 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
681 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
682 * were used to make the determination.*/
684 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
686 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
687 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
691 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
693 struct stp_port *sp = xport_get_stp_port(xport);
694 struct ofpbuf payload = *packet;
695 struct eth_header *eth = payload.data;
697 /* Sink packets on ports that have STP disabled when the bridge has
699 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
703 /* Trim off padding on payload. */
704 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
705 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
708 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
709 stp_received_bpdu(sp, payload.data, payload.size);
713 static struct xport *
714 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
718 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
720 if (xport->ofp_port == ofp_port) {
728 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
730 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
731 return xport ? xport->odp_port : ODPP_NONE;
735 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
739 xport = get_ofp_port(ctx->xbridge, ofp_port);
740 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
741 xport->state & OFPUTIL_PS_LINK_DOWN) {
748 static const struct ofputil_bucket *
749 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
753 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
755 struct group_dpif *group;
758 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
763 hit = group_first_live_bucket(ctx, group, depth) != NULL;
765 group_dpif_release(group);
769 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
772 bucket_is_alive(const struct xlate_ctx *ctx,
773 const struct ofputil_bucket *bucket, int depth)
775 if (depth >= MAX_LIVENESS_RECURSION) {
776 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
778 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
779 MAX_LIVENESS_RECURSION);
783 return !ofputil_bucket_has_liveness(bucket) ||
784 (bucket->watch_port != OFPP_ANY &&
785 odp_port_is_alive(ctx, bucket->watch_port)) ||
786 (bucket->watch_group != OFPG_ANY &&
787 group_is_alive(ctx, bucket->watch_group, depth + 1));
790 static const struct ofputil_bucket *
791 group_first_live_bucket(const struct xlate_ctx *ctx,
792 const struct group_dpif *group, int depth)
794 struct ofputil_bucket *bucket;
795 const struct list *buckets;
797 group_dpif_get_buckets(group, &buckets);
798 LIST_FOR_EACH (bucket, list_node, buckets) {
799 if (bucket_is_alive(ctx, bucket, depth)) {
807 static const struct ofputil_bucket *
808 group_best_live_bucket(const struct xlate_ctx *ctx,
809 const struct group_dpif *group,
812 const struct ofputil_bucket *best_bucket = NULL;
813 uint32_t best_score = 0;
816 const struct ofputil_bucket *bucket;
817 const struct list *buckets;
819 group_dpif_get_buckets(group, &buckets);
820 LIST_FOR_EACH (bucket, list_node, buckets) {
821 if (bucket_is_alive(ctx, bucket, 0)) {
822 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
823 if (score >= best_score) {
824 best_bucket = bucket;
835 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
837 return (bundle->vlan_mode != PORT_VLAN_ACCESS
838 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
842 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
844 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
848 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
850 return xbundle != &ofpp_none_bundle
851 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
856 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
858 return xbundle != &ofpp_none_bundle
859 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
864 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
866 return xbundle != &ofpp_none_bundle
867 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
871 static struct xbundle *
872 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
873 bool warn, struct xport **in_xportp)
877 /* Find the port and bundle for the received packet. */
878 xport = get_ofp_port(xbridge, in_port);
882 if (xport && xport->xbundle) {
883 return xport->xbundle;
886 /* Special-case OFPP_NONE, which a controller may use as the ingress
887 * port for traffic that it is sourcing. */
888 if (in_port == OFPP_NONE) {
889 ofpp_none_bundle.name = "OFPP_NONE";
890 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
891 return &ofpp_none_bundle;
894 /* Odd. A few possible reasons here:
896 * - We deleted a port but there are still a few packets queued up
899 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
900 * we don't know about.
902 * - The ofproto client didn't configure the port as part of a bundle.
903 * This is particularly likely to happen if a packet was received on the
904 * port after it was created, but before the client had a chance to
905 * configure its bundle.
908 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
910 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
911 "port %"PRIu16, xbridge->name, in_port);
917 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
919 const struct xbridge *xbridge = ctx->xbridge;
920 mirror_mask_t mirrors;
921 struct xbundle *in_xbundle;
925 mirrors = ctx->xout->mirrors;
926 ctx->xout->mirrors = 0;
928 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
929 ctx->xin->packet != NULL, NULL);
933 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
935 /* Drop frames on bundles reserved for mirroring. */
936 if (xbundle_mirror_out(xbridge, in_xbundle)) {
937 if (ctx->xin->packet != NULL) {
938 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
939 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
940 "%s, which is reserved exclusively for mirroring",
941 ctx->xbridge->name, in_xbundle->name);
943 ofpbuf_clear(&ctx->xout->odp_actions);
948 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
949 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
952 vlan = input_vid_to_vlan(in_xbundle, vid);
958 /* Restore the original packet before adding the mirror actions. */
959 ctx->xin->flow = *orig_flow;
962 mirror_mask_t dup_mirrors;
963 struct ofbundle *out;
964 unsigned long *vlans;
969 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
970 &vlans, &dup_mirrors, &out, &out_vlan);
971 ovs_assert(has_mirror);
974 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
976 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
979 if (!vlan_mirrored) {
980 mirrors = zero_rightmost_1bit(mirrors);
984 mirrors &= ~dup_mirrors;
985 ctx->xout->mirrors |= dup_mirrors;
987 struct xbundle *out_xbundle = xbundle_lookup(out);
989 output_normal(ctx, out_xbundle, vlan);
991 } else if (vlan != out_vlan
992 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
993 struct xbundle *xbundle;
995 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
996 if (xbundle_includes_vlan(xbundle, out_vlan)
997 && !xbundle_mirror_out(xbridge, xbundle)) {
998 output_normal(ctx, xbundle, out_vlan);
1005 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1006 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1007 * the bundle on which the packet was received, returns the VLAN to which the
1010 * Both 'vid' and the return value are in the range 0...4095. */
1012 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1014 switch (in_xbundle->vlan_mode) {
1015 case PORT_VLAN_ACCESS:
1016 return in_xbundle->vlan;
1019 case PORT_VLAN_TRUNK:
1022 case PORT_VLAN_NATIVE_UNTAGGED:
1023 case PORT_VLAN_NATIVE_TAGGED:
1024 return vid ? vid : in_xbundle->vlan;
1031 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1032 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1035 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1036 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1039 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1041 /* Allow any VID on the OFPP_NONE port. */
1042 if (in_xbundle == &ofpp_none_bundle) {
1046 switch (in_xbundle->vlan_mode) {
1047 case PORT_VLAN_ACCESS:
1050 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1051 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1052 "packet received on port %s configured as VLAN "
1053 "%"PRIu16" access port", vid, in_xbundle->name,
1060 case PORT_VLAN_NATIVE_UNTAGGED:
1061 case PORT_VLAN_NATIVE_TAGGED:
1063 /* Port must always carry its native VLAN. */
1067 case PORT_VLAN_TRUNK:
1068 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1070 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1071 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1072 "received on port %s not configured for trunking "
1073 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1085 /* Given 'vlan', the VLAN that a packet belongs to, and
1086 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1087 * that should be included in the 802.1Q header. (If the return value is 0,
1088 * then the 802.1Q header should only be included in the packet if there is a
1091 * Both 'vlan' and the return value are in the range 0...4095. */
1093 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1095 switch (out_xbundle->vlan_mode) {
1096 case PORT_VLAN_ACCESS:
1099 case PORT_VLAN_TRUNK:
1100 case PORT_VLAN_NATIVE_TAGGED:
1103 case PORT_VLAN_NATIVE_UNTAGGED:
1104 return vlan == out_xbundle->vlan ? 0 : vlan;
1112 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1115 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1117 ovs_be16 tci, old_tci;
1118 struct xport *xport;
1120 vid = output_vlan_to_vid(out_xbundle, vlan);
1121 if (list_is_empty(&out_xbundle->xports)) {
1122 /* Partially configured bundle with no slaves. Drop the packet. */
1124 } else if (!out_xbundle->bond) {
1125 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1128 struct ofport_dpif *ofport;
1130 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
1131 &ctx->xout->wc, vid);
1132 xport = xport_lookup(ofport);
1135 /* No slaves enabled, so drop packet. */
1139 if (ctx->xin->resubmit_stats) {
1140 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1141 ctx->xin->resubmit_stats->n_bytes);
1145 old_tci = *flow_tci;
1147 if (tci || out_xbundle->use_priority_tags) {
1148 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1150 tci |= htons(VLAN_CFI);
1155 compose_output_action(ctx, xport->ofp_port);
1156 *flow_tci = old_tci;
1159 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1160 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1161 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1163 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1165 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1169 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1170 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1174 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1175 if (flow->nw_proto == ARP_OP_REPLY) {
1177 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1178 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1179 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1181 return flow->nw_src == flow->nw_dst;
1187 /* Checks whether a MAC learning update is necessary for MAC learning table
1188 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1191 * Most packets processed through the MAC learning table do not actually
1192 * change it in any way. This function requires only a read lock on the MAC
1193 * learning table, so it is much cheaper in this common case.
1195 * Keep the code here synchronized with that in update_learning_table__()
1198 is_mac_learning_update_needed(const struct mac_learning *ml,
1199 const struct flow *flow,
1200 struct flow_wildcards *wc,
1201 int vlan, struct xbundle *in_xbundle)
1202 OVS_REQ_RDLOCK(ml->rwlock)
1204 struct mac_entry *mac;
1206 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1210 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1211 if (!mac || mac_entry_age(ml, mac)) {
1215 if (is_gratuitous_arp(flow, wc)) {
1216 /* We don't want to learn from gratuitous ARP packets that are
1217 * reflected back over bond slaves so we lock the learning table. */
1218 if (!in_xbundle->bond) {
1220 } else if (mac_entry_is_grat_arp_locked(mac)) {
1225 return mac->port.p != in_xbundle->ofbundle;
1229 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1230 * received on 'in_xbundle' in 'vlan'.
1232 * This code repeats all the checks in is_mac_learning_update_needed() because
1233 * the lock was released between there and here and thus the MAC learning state
1234 * could have changed.
1236 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1239 update_learning_table__(const struct xbridge *xbridge,
1240 const struct flow *flow, struct flow_wildcards *wc,
1241 int vlan, struct xbundle *in_xbundle)
1242 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1244 struct mac_entry *mac;
1246 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1250 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1251 if (is_gratuitous_arp(flow, wc)) {
1252 /* We don't want to learn from gratuitous ARP packets that are
1253 * reflected back over bond slaves so we lock the learning table. */
1254 if (!in_xbundle->bond) {
1255 mac_entry_set_grat_arp_lock(mac);
1256 } else if (mac_entry_is_grat_arp_locked(mac)) {
1261 if (mac->port.p != in_xbundle->ofbundle) {
1262 /* The log messages here could actually be useful in debugging,
1263 * so keep the rate limit relatively high. */
1264 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1266 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1267 "on port %s in VLAN %d",
1268 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1269 in_xbundle->name, vlan);
1271 mac->port.p = in_xbundle->ofbundle;
1272 mac_learning_changed(xbridge->ml);
1277 update_learning_table(const struct xbridge *xbridge,
1278 const struct flow *flow, struct flow_wildcards *wc,
1279 int vlan, struct xbundle *in_xbundle)
1283 /* Don't learn the OFPP_NONE port. */
1284 if (in_xbundle == &ofpp_none_bundle) {
1288 /* First try the common case: no change to MAC learning table. */
1289 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1290 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1292 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1295 /* Slow path: MAC learning table might need an update. */
1296 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1297 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1298 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1302 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1303 * dropped. Returns true if they may be forwarded, false if they should be
1306 * 'in_port' must be the xport that corresponds to flow->in_port.
1307 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1309 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1310 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1311 * checked by input_vid_is_valid().
1313 * May also add tags to '*tags', although the current implementation only does
1314 * so in one special case.
1317 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1320 struct xbundle *in_xbundle = in_port->xbundle;
1321 const struct xbridge *xbridge = ctx->xbridge;
1322 struct flow *flow = &ctx->xin->flow;
1324 /* Drop frames for reserved multicast addresses
1325 * only if forward_bpdu option is absent. */
1326 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1327 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1331 if (in_xbundle->bond) {
1332 struct mac_entry *mac;
1334 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1340 xlate_report(ctx, "bonding refused admissibility, dropping");
1343 case BV_DROP_IF_MOVED:
1344 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1345 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1346 if (mac && mac->port.p != in_xbundle->ofbundle &&
1347 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1348 || mac_entry_is_grat_arp_locked(mac))) {
1349 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1350 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1354 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1363 xlate_normal(struct xlate_ctx *ctx)
1365 struct flow_wildcards *wc = &ctx->xout->wc;
1366 struct flow *flow = &ctx->xin->flow;
1367 struct xbundle *in_xbundle;
1368 struct xport *in_port;
1369 struct mac_entry *mac;
1374 ctx->xout->has_normal = true;
1376 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1377 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1378 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1380 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1381 ctx->xin->packet != NULL, &in_port);
1383 xlate_report(ctx, "no input bundle, dropping");
1387 /* Drop malformed frames. */
1388 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1389 !(flow->vlan_tci & htons(VLAN_CFI))) {
1390 if (ctx->xin->packet != NULL) {
1391 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1392 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1393 "VLAN tag received on port %s",
1394 ctx->xbridge->name, in_xbundle->name);
1396 xlate_report(ctx, "partial VLAN tag, dropping");
1400 /* Drop frames on bundles reserved for mirroring. */
1401 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1402 if (ctx->xin->packet != NULL) {
1403 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1404 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1405 "%s, which is reserved exclusively for mirroring",
1406 ctx->xbridge->name, in_xbundle->name);
1408 xlate_report(ctx, "input port is mirror output port, dropping");
1413 vid = vlan_tci_to_vid(flow->vlan_tci);
1414 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1415 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1418 vlan = input_vid_to_vlan(in_xbundle, vid);
1420 /* Check other admissibility requirements. */
1421 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1425 /* Learn source MAC. */
1426 if (ctx->xin->may_learn) {
1427 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1430 /* Determine output bundle. */
1431 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1432 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1433 mac_port = mac ? mac->port.p : NULL;
1434 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1437 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1438 if (mac_xbundle && mac_xbundle != in_xbundle) {
1439 xlate_report(ctx, "forwarding to learned port");
1440 output_normal(ctx, mac_xbundle, vlan);
1441 } else if (!mac_xbundle) {
1442 xlate_report(ctx, "learned port is unknown, dropping");
1444 xlate_report(ctx, "learned port is input port, dropping");
1447 struct xbundle *xbundle;
1449 xlate_report(ctx, "no learned MAC for destination, flooding");
1450 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1451 if (xbundle != in_xbundle
1452 && xbundle_includes_vlan(xbundle, vlan)
1453 && xbundle->floodable
1454 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1455 output_normal(ctx, xbundle, vlan);
1458 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1462 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1463 * the number of packets out of UINT32_MAX to sample. The given
1464 * cookie is passed back in the callback for each sampled packet.
1467 compose_sample_action(const struct xbridge *xbridge,
1468 struct ofpbuf *odp_actions,
1469 const struct flow *flow,
1470 const uint32_t probability,
1471 const union user_action_cookie *cookie,
1472 const size_t cookie_size)
1474 size_t sample_offset, actions_offset;
1475 odp_port_t odp_port;
1479 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1481 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1483 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1485 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1486 pid = dpif_port_get_pid(xbridge->dpif, odp_port);
1487 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1489 nl_msg_end_nested(odp_actions, actions_offset);
1490 nl_msg_end_nested(odp_actions, sample_offset);
1491 return cookie_offset;
1495 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1496 odp_port_t odp_port, unsigned int n_outputs,
1497 union user_action_cookie *cookie)
1501 cookie->type = USER_ACTION_COOKIE_SFLOW;
1502 cookie->sflow.vlan_tci = vlan_tci;
1504 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1505 * port information") for the interpretation of cookie->output. */
1506 switch (n_outputs) {
1508 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1509 cookie->sflow.output = 0x40000000 | 256;
1513 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1515 cookie->sflow.output = ifindex;
1520 /* 0x80000000 means "multiple output ports. */
1521 cookie->sflow.output = 0x80000000 | n_outputs;
1526 /* Compose SAMPLE action for sFlow bridge sampling. */
1528 compose_sflow_action(const struct xbridge *xbridge,
1529 struct ofpbuf *odp_actions,
1530 const struct flow *flow,
1531 odp_port_t odp_port)
1533 uint32_t probability;
1534 union user_action_cookie cookie;
1536 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1540 probability = dpif_sflow_get_probability(xbridge->sflow);
1541 compose_sflow_cookie(xbridge, htons(0), odp_port,
1542 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1544 return compose_sample_action(xbridge, odp_actions, flow, probability,
1545 &cookie, sizeof cookie.sflow);
1549 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1550 uint32_t obs_domain_id, uint32_t obs_point_id,
1551 union user_action_cookie *cookie)
1553 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1554 cookie->flow_sample.probability = probability;
1555 cookie->flow_sample.collector_set_id = collector_set_id;
1556 cookie->flow_sample.obs_domain_id = obs_domain_id;
1557 cookie->flow_sample.obs_point_id = obs_point_id;
1561 compose_ipfix_cookie(union user_action_cookie *cookie)
1563 cookie->type = USER_ACTION_COOKIE_IPFIX;
1566 /* Compose SAMPLE action for IPFIX bridge sampling. */
1568 compose_ipfix_action(const struct xbridge *xbridge,
1569 struct ofpbuf *odp_actions,
1570 const struct flow *flow)
1572 uint32_t probability;
1573 union user_action_cookie cookie;
1575 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1579 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1580 compose_ipfix_cookie(&cookie);
1582 compose_sample_action(xbridge, odp_actions, flow, probability,
1583 &cookie, sizeof cookie.ipfix);
1586 /* SAMPLE action for sFlow must be first action in any given list of
1587 * actions. At this point we do not have all information required to
1588 * build it. So try to build sample action as complete as possible. */
1590 add_sflow_action(struct xlate_ctx *ctx)
1592 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1593 &ctx->xout->odp_actions,
1594 &ctx->xin->flow, ODPP_NONE);
1595 ctx->sflow_odp_port = 0;
1596 ctx->sflow_n_outputs = 0;
1599 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1600 * of actions, eventually after the SAMPLE action for sFlow. */
1602 add_ipfix_action(struct xlate_ctx *ctx)
1604 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1608 /* Fix SAMPLE action according to data collected while composing ODP actions.
1609 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1610 * USERSPACE action's user-cookie which is required for sflow. */
1612 fix_sflow_action(struct xlate_ctx *ctx)
1614 const struct flow *base = &ctx->base_flow;
1615 union user_action_cookie *cookie;
1617 if (!ctx->user_cookie_offset) {
1621 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1622 sizeof cookie->sflow);
1623 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1625 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1626 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1629 static enum slow_path_reason
1630 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1631 const struct xport *xport, const struct ofpbuf *packet)
1633 struct flow_wildcards *wc = &ctx->xout->wc;
1634 const struct xbridge *xbridge = ctx->xbridge;
1638 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1640 cfm_process_heartbeat(xport->cfm, packet);
1643 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1645 bfd_process_packet(xport->bfd, flow, packet);
1646 /* If POLL received, immediately sends FINAL back. */
1647 if (bfd_should_send_packet(xport->bfd)) {
1649 ofproto_dpif_monitor_port_send_soon(xport->ofport);
1651 ofproto_dpif_monitor_port_send_soon_safe(xport->ofport);
1656 } else if (xport->xbundle && xport->xbundle->lacp
1657 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1659 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1662 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1664 stp_process_packet(xport, packet);
1673 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1676 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1677 struct flow_wildcards *wc = &ctx->xout->wc;
1678 struct flow *flow = &ctx->xin->flow;
1679 ovs_be16 flow_vlan_tci;
1680 uint32_t flow_pkt_mark;
1681 uint8_t flow_nw_tos;
1682 odp_port_t out_port, odp_port;
1685 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1686 * before traversing a patch port. */
1687 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 24);
1690 xlate_report(ctx, "Nonexistent output port");
1692 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1693 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1695 } else if (check_stp && !xport_stp_forward_state(xport)) {
1696 xlate_report(ctx, "STP not in forwarding state, skipping output");
1700 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1701 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1706 const struct xport *peer = xport->peer;
1707 struct flow old_flow = ctx->xin->flow;
1708 enum slow_path_reason special;
1710 ctx->xbridge = peer->xbridge;
1711 flow->in_port.ofp_port = peer->ofp_port;
1712 flow->metadata = htonll(0);
1713 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1714 memset(flow->regs, 0, sizeof flow->regs);
1716 special = process_special(ctx, &ctx->xin->flow, peer,
1719 ctx->xout->slow |= special;
1720 } else if (may_receive(peer, ctx)) {
1721 if (xport_stp_forward_state(peer)) {
1722 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1724 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1725 * learning action look at the packet, then drop it. */
1726 struct flow old_base_flow = ctx->base_flow;
1727 size_t old_size = ctx->xout->odp_actions.size;
1728 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1729 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1730 ctx->xout->mirrors = old_mirrors;
1731 ctx->base_flow = old_base_flow;
1732 ctx->xout->odp_actions.size = old_size;
1736 ctx->xin->flow = old_flow;
1737 ctx->xbridge = xport->xbridge;
1739 if (ctx->xin->resubmit_stats) {
1740 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1741 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1743 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1750 flow_vlan_tci = flow->vlan_tci;
1751 flow_pkt_mark = flow->pkt_mark;
1752 flow_nw_tos = flow->nw_tos;
1754 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1755 wc->masks.nw_tos |= IP_ECN_MASK;
1756 flow->nw_tos &= ~IP_DSCP_MASK;
1757 flow->nw_tos |= dscp;
1760 if (xport->is_tunnel) {
1761 /* Save tunnel metadata so that changes made due to
1762 * the Logical (tunnel) Port are not visible for any further
1763 * matches, while explicit set actions on tunnel metadata are.
1765 struct flow_tnl flow_tnl = flow->tunnel;
1766 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1767 if (odp_port == ODPP_NONE) {
1768 xlate_report(ctx, "Tunneling decided against output");
1769 goto out; /* restore flow_nw_tos */
1771 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1772 xlate_report(ctx, "Not tunneling to our own address");
1773 goto out; /* restore flow_nw_tos */
1775 if (ctx->xin->resubmit_stats) {
1776 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1778 out_port = odp_port;
1779 commit_odp_tunnel_action(flow, &ctx->base_flow,
1780 &ctx->xout->odp_actions);
1781 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1783 odp_port = xport->odp_port;
1784 out_port = odp_port;
1785 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1786 ofp_port_t vlandev_port;
1788 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1789 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1790 ofp_port, flow->vlan_tci);
1791 if (vlandev_port != ofp_port) {
1792 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1793 flow->vlan_tci = htons(0);
1798 if (out_port != ODPP_NONE) {
1799 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1800 &ctx->xout->odp_actions,
1802 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1805 ctx->sflow_odp_port = odp_port;
1806 ctx->sflow_n_outputs++;
1807 ctx->xout->nf_output_iface = ofp_port;
1812 flow->vlan_tci = flow_vlan_tci;
1813 flow->pkt_mark = flow_pkt_mark;
1814 flow->nw_tos = flow_nw_tos;
1818 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1820 compose_output_action__(ctx, ofp_port, true);
1824 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1826 struct rule_dpif *old_rule = ctx->rule;
1827 struct rule_actions *actions;
1829 if (ctx->xin->resubmit_stats) {
1830 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1836 actions = rule_dpif_get_actions(rule);
1837 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1838 rule_actions_unref(actions);
1839 ctx->rule = old_rule;
1844 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
1846 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1848 if (ctx->recurse >= MAX_RESUBMIT_RECURSION) {
1849 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
1850 MAX_RESUBMIT_RECURSION);
1851 } else if (ctx->resubmits >= MAX_RESUBMITS) {
1852 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
1853 } else if (ctx->xout->odp_actions.size > UINT16_MAX) {
1854 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
1855 } else if (ctx->stack.size >= 65536) {
1856 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
1865 xlate_table_action(struct xlate_ctx *ctx,
1866 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1868 if (xlate_resubmit_resource_check(ctx)) {
1869 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1870 bool skip_wildcards = ctx->xin->skip_wildcards;
1871 uint8_t old_table_id = ctx->table_id;
1872 struct rule_dpif *rule;
1874 ctx->table_id = table_id;
1876 /* Look up a flow with 'in_port' as the input port. Then restore the
1877 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1878 * have surprising behavior). */
1879 ctx->xin->flow.in_port.ofp_port = in_port;
1880 rule_dpif_lookup_in_table(ctx->xbridge->ofproto, &ctx->xin->flow,
1881 !skip_wildcards ? &ctx->xout->wc : NULL,
1883 ctx->xin->flow.in_port.ofp_port = old_in_port;
1885 if (ctx->xin->resubmit_hook) {
1886 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1889 if (!rule && may_packet_in) {
1890 struct xport *xport;
1893 * check if table configuration flags
1894 * OFPTC11_TABLE_MISS_CONTROLLER, default.
1895 * OFPTC11_TABLE_MISS_CONTINUE,
1896 * OFPTC11_TABLE_MISS_DROP
1897 * When OF1.0, OFPTC11_TABLE_MISS_CONTINUE is used. What to do? */
1898 xport = get_ofp_port(ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
1899 choose_miss_rule(xport ? xport->config : 0,
1900 ctx->xbridge->miss_rule,
1901 ctx->xbridge->no_packet_in_rule, &rule);
1904 xlate_recursively(ctx, rule);
1905 rule_dpif_unref(rule);
1908 ctx->table_id = old_table_id;
1916 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
1918 uint64_t action_list_stub[1024 / 8];
1919 struct ofpbuf action_list, action_set;
1921 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
1922 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
1924 ofpacts_execute_action_set(&action_list, &action_set);
1926 do_xlate_actions(action_list.data, action_list.size, ctx);
1929 ofpbuf_uninit(&action_set);
1930 ofpbuf_uninit(&action_list);
1934 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
1936 const struct ofputil_bucket *bucket;
1937 const struct list *buckets;
1938 struct flow old_flow = ctx->xin->flow;
1940 group_dpif_get_buckets(group, &buckets);
1942 LIST_FOR_EACH (bucket, list_node, buckets) {
1943 xlate_group_bucket(ctx, bucket);
1944 /* Roll back flow to previous state.
1945 * This is equivalent to cloning the packet for each bucket.
1947 * As a side effect any subsequently applied actions will
1948 * also effectively be applied to a clone of the packet taken
1949 * just before applying the all or indirect group. */
1950 ctx->xin->flow = old_flow;
1955 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
1957 const struct ofputil_bucket *bucket;
1959 bucket = group_first_live_bucket(ctx, group, 0);
1961 xlate_group_bucket(ctx, bucket);
1966 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
1968 struct flow_wildcards *wc = &ctx->xout->wc;
1969 const struct ofputil_bucket *bucket;
1972 basis = hash_bytes(ctx->xin->flow.dl_dst, sizeof ctx->xin->flow.dl_dst, 0);
1973 bucket = group_best_live_bucket(ctx, group, basis);
1975 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1976 xlate_group_bucket(ctx, bucket);
1981 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
1983 switch (group_dpif_get_type(group)) {
1985 case OFPGT11_INDIRECT:
1986 xlate_all_group(ctx, group);
1988 case OFPGT11_SELECT:
1989 xlate_select_group(ctx, group);
1992 xlate_ff_group(ctx, group);
1997 group_dpif_release(group);
2001 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2003 if (xlate_resubmit_resource_check(ctx)) {
2004 struct group_dpif *group;
2007 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2009 xlate_group_action__(ctx, group);
2019 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2020 const struct ofpact_resubmit *resubmit)
2025 in_port = resubmit->in_port;
2026 if (in_port == OFPP_IN_PORT) {
2027 in_port = ctx->xin->flow.in_port.ofp_port;
2030 table_id = resubmit->table_id;
2031 if (table_id == 255) {
2032 table_id = ctx->table_id;
2035 xlate_table_action(ctx, in_port, table_id, false);
2039 flood_packets(struct xlate_ctx *ctx, bool all)
2041 const struct xport *xport;
2043 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2044 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2049 compose_output_action__(ctx, xport->ofp_port, false);
2050 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2051 compose_output_action(ctx, xport->ofp_port);
2055 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2059 execute_controller_action(struct xlate_ctx *ctx, int len,
2060 enum ofp_packet_in_reason reason,
2061 uint16_t controller_id)
2063 struct ofproto_packet_in *pin;
2064 struct ofpbuf *packet;
2065 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2067 ctx->xout->slow |= SLOW_CONTROLLER;
2068 if (!ctx->xin->packet) {
2072 packet = ofpbuf_clone(ctx->xin->packet);
2074 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2075 &ctx->xout->odp_actions,
2078 odp_execute_actions(NULL, packet, &md, ctx->xout->odp_actions.data,
2079 ctx->xout->odp_actions.size, NULL);
2081 pin = xmalloc(sizeof *pin);
2082 pin->up.packet_len = packet->size;
2083 pin->up.packet = ofpbuf_steal_data(packet);
2084 pin->up.reason = reason;
2085 pin->up.table_id = ctx->table_id;
2086 pin->up.cookie = (ctx->rule
2087 ? rule_dpif_get_flow_cookie(ctx->rule)
2090 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2092 pin->controller_id = controller_id;
2093 pin->send_len = len;
2094 pin->generated_by_table_miss = (ctx->rule
2095 && rule_dpif_is_table_miss(ctx->rule));
2096 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2097 ofpbuf_delete(packet);
2101 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2103 struct flow_wildcards *wc = &ctx->xout->wc;
2104 struct flow *flow = &ctx->xin->flow;
2107 ovs_assert(eth_type_mpls(mpls->ethertype));
2109 n = flow_count_mpls_labels(flow, wc);
2111 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2112 &ctx->xout->odp_actions,
2114 } else if (n >= FLOW_MAX_MPLS_LABELS) {
2115 if (ctx->xin->packet != NULL) {
2116 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2117 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2118 "MPLS push action can't be performed as it would "
2119 "have more MPLS LSEs than the %d supported.",
2120 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2124 } else if (n >= ctx->xbridge->max_mpls_depth) {
2125 COVERAGE_INC(xlate_actions_mpls_overflow);
2126 ctx->xout->slow |= SLOW_ACTION;
2129 flow_push_mpls(flow, n, mpls->ethertype, wc);
2133 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2135 struct flow_wildcards *wc = &ctx->xout->wc;
2136 struct flow *flow = &ctx->xin->flow;
2137 int n = flow_count_mpls_labels(flow, wc);
2139 if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2140 if (ctx->xin->packet != NULL) {
2141 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2142 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2143 "MPLS pop action can't be performed as it has "
2144 "more MPLS LSEs than the %d supported.",
2145 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2148 ofpbuf_clear(&ctx->xout->odp_actions);
2153 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2155 struct flow *flow = &ctx->xin->flow;
2157 if (!is_ip_any(flow)) {
2161 ctx->xout->wc.masks.nw_ttl = 0xff;
2162 if (flow->nw_ttl > 1) {
2168 for (i = 0; i < ids->n_controllers; i++) {
2169 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2173 /* Stop processing for current table. */
2179 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2181 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2182 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2183 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2188 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2190 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2191 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2192 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2197 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2199 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2200 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2201 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2206 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2208 struct flow *flow = &ctx->xin->flow;
2209 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2210 struct flow_wildcards *wc = &ctx->xout->wc;
2212 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2213 if (eth_type_mpls(flow->dl_type)) {
2216 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2219 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2221 /* Stop processing for current table. */
2230 xlate_output_action(struct xlate_ctx *ctx,
2231 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2233 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2235 ctx->xout->nf_output_iface = NF_OUT_DROP;
2239 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2242 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2249 flood_packets(ctx, false);
2252 flood_packets(ctx, true);
2254 case OFPP_CONTROLLER:
2255 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2261 if (port != ctx->xin->flow.in_port.ofp_port) {
2262 compose_output_action(ctx, port);
2264 xlate_report(ctx, "skipping output to input port");
2269 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2270 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2271 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2272 ctx->xout->nf_output_iface = prev_nf_output_iface;
2273 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2274 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2275 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2280 xlate_output_reg_action(struct xlate_ctx *ctx,
2281 const struct ofpact_output_reg *or)
2283 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2284 if (port <= UINT16_MAX) {
2285 union mf_subvalue value;
2287 memset(&value, 0xff, sizeof value);
2288 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2289 xlate_output_action(ctx, u16_to_ofp(port),
2290 or->max_len, false);
2295 xlate_enqueue_action(struct xlate_ctx *ctx,
2296 const struct ofpact_enqueue *enqueue)
2298 ofp_port_t ofp_port = enqueue->port;
2299 uint32_t queue_id = enqueue->queue;
2300 uint32_t flow_priority, priority;
2303 /* Translate queue to priority. */
2304 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2306 /* Fall back to ordinary output action. */
2307 xlate_output_action(ctx, enqueue->port, 0, false);
2311 /* Check output port. */
2312 if (ofp_port == OFPP_IN_PORT) {
2313 ofp_port = ctx->xin->flow.in_port.ofp_port;
2314 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2318 /* Add datapath actions. */
2319 flow_priority = ctx->xin->flow.skb_priority;
2320 ctx->xin->flow.skb_priority = priority;
2321 compose_output_action(ctx, ofp_port);
2322 ctx->xin->flow.skb_priority = flow_priority;
2324 /* Update NetFlow output port. */
2325 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2326 ctx->xout->nf_output_iface = ofp_port;
2327 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2328 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2333 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2335 uint32_t skb_priority;
2337 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2338 ctx->xin->flow.skb_priority = skb_priority;
2340 /* Couldn't translate queue to a priority. Nothing to do. A warning
2341 * has already been logged. */
2346 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2348 const struct xbridge *xbridge = xbridge_;
2359 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2362 port = get_ofp_port(xbridge, ofp_port);
2363 return port ? port->may_enable : false;
2368 xlate_bundle_action(struct xlate_ctx *ctx,
2369 const struct ofpact_bundle *bundle)
2373 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2375 CONST_CAST(struct xbridge *, ctx->xbridge));
2376 if (bundle->dst.field) {
2377 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2380 xlate_output_action(ctx, port, 0, false);
2385 xlate_learn_action(struct xlate_ctx *ctx,
2386 const struct ofpact_learn *learn)
2388 uint64_t ofpacts_stub[1024 / 8];
2389 struct ofputil_flow_mod fm;
2390 struct ofpbuf ofpacts;
2392 ctx->xout->has_learn = true;
2394 learn_mask(learn, &ctx->xout->wc);
2396 if (!ctx->xin->may_learn) {
2400 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2401 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2402 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2403 ofpbuf_uninit(&ofpacts);
2407 xlate_fin_timeout(struct xlate_ctx *ctx,
2408 const struct ofpact_fin_timeout *oft)
2410 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2411 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2412 oft->fin_hard_timeout);
2417 xlate_sample_action(struct xlate_ctx *ctx,
2418 const struct ofpact_sample *os)
2420 union user_action_cookie cookie;
2421 /* Scale the probability from 16-bit to 32-bit while representing
2422 * the same percentage. */
2423 uint32_t probability = (os->probability << 16) | os->probability;
2425 if (!ctx->xbridge->variable_length_userdata) {
2426 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2428 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2429 "lacks support (needs Linux 3.10+ or kernel module from "
2434 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2435 &ctx->xout->odp_actions,
2438 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2439 os->obs_domain_id, os->obs_point_id, &cookie);
2440 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2441 probability, &cookie, sizeof cookie.flow_sample);
2445 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2447 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2448 ? OFPUTIL_PC_NO_RECV_STP
2449 : OFPUTIL_PC_NO_RECV)) {
2453 /* Only drop packets here if both forwarding and learning are
2454 * disabled. If just learning is enabled, we need to have
2455 * OFPP_NORMAL and the learning action have a look at the packet
2456 * before we can drop it. */
2457 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2465 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2467 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2468 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2469 ofpact_pad(&ctx->action_set);
2473 xlate_action_set(struct xlate_ctx *ctx)
2475 uint64_t action_list_stub[1024 / 64];
2476 struct ofpbuf action_list;
2478 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2479 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2480 do_xlate_actions(action_list.data, action_list.size, ctx);
2481 ofpbuf_uninit(&action_list);
2485 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2486 struct xlate_ctx *ctx)
2488 struct flow_wildcards *wc = &ctx->xout->wc;
2489 struct flow *flow = &ctx->xin->flow;
2490 const struct ofpact *a;
2492 /* dl_type already in the mask, not set below. */
2494 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2495 struct ofpact_controller *controller;
2496 const struct ofpact_metadata *metadata;
2497 const struct ofpact_set_field *set_field;
2498 const struct mf_field *mf;
2506 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2507 ofpact_get_OUTPUT(a)->max_len, true);
2511 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2516 case OFPACT_CONTROLLER:
2517 controller = ofpact_get_CONTROLLER(a);
2518 execute_controller_action(ctx, controller->max_len,
2520 controller->controller_id);
2523 case OFPACT_ENQUEUE:
2524 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2527 case OFPACT_SET_VLAN_VID:
2528 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2529 if (flow->vlan_tci & htons(VLAN_CFI) ||
2530 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2531 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2532 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2537 case OFPACT_SET_VLAN_PCP:
2538 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2539 if (flow->vlan_tci & htons(VLAN_CFI) ||
2540 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2541 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2542 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2543 << VLAN_PCP_SHIFT) | VLAN_CFI);
2547 case OFPACT_STRIP_VLAN:
2548 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2549 flow->vlan_tci = htons(0);
2552 case OFPACT_PUSH_VLAN:
2553 /* XXX 802.1AD(QinQ) */
2554 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2555 flow->vlan_tci = htons(VLAN_CFI);
2558 case OFPACT_SET_ETH_SRC:
2559 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2560 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2563 case OFPACT_SET_ETH_DST:
2564 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2565 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2568 case OFPACT_SET_IPV4_SRC:
2569 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2570 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2571 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2575 case OFPACT_SET_IPV4_DST:
2576 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2577 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2578 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2582 case OFPACT_SET_IP_DSCP:
2583 if (is_ip_any(flow)) {
2584 wc->masks.nw_tos |= IP_DSCP_MASK;
2585 flow->nw_tos &= ~IP_DSCP_MASK;
2586 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2590 case OFPACT_SET_IP_ECN:
2591 if (is_ip_any(flow)) {
2592 wc->masks.nw_tos |= IP_ECN_MASK;
2593 flow->nw_tos &= ~IP_ECN_MASK;
2594 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2598 case OFPACT_SET_IP_TTL:
2599 if (is_ip_any(flow)) {
2600 wc->masks.nw_ttl = 0xff;
2601 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2605 case OFPACT_SET_L4_SRC_PORT:
2606 if (is_ip_any(flow)) {
2607 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2608 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2609 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2613 case OFPACT_SET_L4_DST_PORT:
2614 if (is_ip_any(flow)) {
2615 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2616 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2617 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2621 case OFPACT_RESUBMIT:
2622 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2625 case OFPACT_SET_TUNNEL:
2626 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2629 case OFPACT_SET_QUEUE:
2630 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2633 case OFPACT_POP_QUEUE:
2634 flow->skb_priority = ctx->orig_skb_priority;
2637 case OFPACT_REG_MOVE:
2638 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2641 case OFPACT_REG_LOAD:
2642 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2645 case OFPACT_SET_FIELD:
2646 set_field = ofpact_get_SET_FIELD(a);
2647 mf = set_field->field;
2648 mf_mask_field_and_prereqs(mf, &wc->masks);
2650 /* Set field action only ever overwrites packet's outermost
2651 * applicable header fields. Do nothing if no header exists. */
2652 if ((mf->id != MFF_VLAN_VID || flow->vlan_tci & htons(VLAN_CFI))
2653 && ((mf->id != MFF_MPLS_LABEL && mf->id != MFF_MPLS_TC)
2654 || eth_type_mpls(flow->dl_type))) {
2655 mf_set_flow_value(mf, &set_field->value, flow);
2659 case OFPACT_STACK_PUSH:
2660 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2664 case OFPACT_STACK_POP:
2665 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2669 case OFPACT_PUSH_MPLS:
2670 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2673 case OFPACT_POP_MPLS:
2674 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2677 case OFPACT_SET_MPLS_LABEL:
2678 compose_set_mpls_label_action(
2679 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2682 case OFPACT_SET_MPLS_TC:
2683 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2686 case OFPACT_SET_MPLS_TTL:
2687 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2690 case OFPACT_DEC_MPLS_TTL:
2691 if (compose_dec_mpls_ttl_action(ctx)) {
2696 case OFPACT_DEC_TTL:
2697 wc->masks.nw_ttl = 0xff;
2698 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2704 /* Nothing to do. */
2707 case OFPACT_MULTIPATH:
2708 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2712 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2715 case OFPACT_OUTPUT_REG:
2716 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2720 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2727 case OFPACT_FIN_TIMEOUT:
2728 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2729 ctx->xout->has_fin_timeout = true;
2730 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2733 case OFPACT_CLEAR_ACTIONS:
2734 ofpbuf_clear(&ctx->action_set);
2737 case OFPACT_WRITE_ACTIONS:
2738 xlate_write_actions(ctx, a);
2741 case OFPACT_WRITE_METADATA:
2742 metadata = ofpact_get_WRITE_METADATA(a);
2743 flow->metadata &= ~metadata->mask;
2744 flow->metadata |= metadata->metadata & metadata->mask;
2748 /* Not implemented yet. */
2751 case OFPACT_GOTO_TABLE: {
2752 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2754 ovs_assert(ctx->table_id < ogt->table_id);
2755 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2756 ogt->table_id, true);
2761 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2768 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2769 const struct flow *flow, struct rule_dpif *rule,
2770 uint16_t tcp_flags, const struct ofpbuf *packet)
2772 xin->ofproto = ofproto;
2774 xin->packet = packet;
2775 xin->may_learn = packet != NULL;
2777 xin->ofpacts = NULL;
2778 xin->ofpacts_len = 0;
2779 xin->tcp_flags = tcp_flags;
2780 xin->resubmit_hook = NULL;
2781 xin->report_hook = NULL;
2782 xin->resubmit_stats = NULL;
2783 xin->skip_wildcards = false;
2787 xlate_out_uninit(struct xlate_out *xout)
2790 ofpbuf_uninit(&xout->odp_actions);
2794 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2795 * into datapath actions, using 'ctx', and discards the datapath actions. */
2797 xlate_actions_for_side_effects(struct xlate_in *xin)
2799 struct xlate_out xout;
2801 xlate_actions(xin, &xout);
2802 xlate_out_uninit(&xout);
2806 xlate_report(struct xlate_ctx *ctx, const char *s)
2808 if (ctx->xin->report_hook) {
2809 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2814 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2817 dst->slow = src->slow;
2818 dst->has_learn = src->has_learn;
2819 dst->has_normal = src->has_normal;
2820 dst->has_fin_timeout = src->has_fin_timeout;
2821 dst->nf_output_iface = src->nf_output_iface;
2822 dst->mirrors = src->mirrors;
2824 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2825 sizeof dst->odp_actions_stub);
2826 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2827 src->odp_actions.size);
2830 static struct skb_priority_to_dscp *
2831 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2833 struct skb_priority_to_dscp *pdscp;
2836 hash = hash_int(skb_priority, 0);
2837 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2838 if (pdscp->skb_priority == skb_priority) {
2846 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2849 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2850 *dscp = pdscp ? pdscp->dscp : 0;
2851 return pdscp != NULL;
2855 clear_skb_priorities(struct xport *xport)
2857 struct skb_priority_to_dscp *pdscp, *next;
2859 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2860 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2866 actions_output_to_local_port(const struct xlate_ctx *ctx)
2868 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2869 const struct nlattr *a;
2872 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2873 ctx->xout->odp_actions.size) {
2874 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2875 && nl_attr_get_odp_port(a) == local_odp_port) {
2882 /* Thread safe call to xlate_actions__(). */
2884 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2885 OVS_EXCLUDED(xlate_rwlock)
2887 ovs_rwlock_rdlock(&xlate_rwlock);
2888 xlate_actions__(xin, xout);
2889 ovs_rwlock_unlock(&xlate_rwlock);
2892 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2893 * into datapath actions in 'odp_actions', using 'ctx'.
2895 * The caller must take responsibility for eventually freeing 'xout', with
2896 * xlate_out_uninit(). */
2898 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
2899 OVS_REQ_RDLOCK(xlate_rwlock)
2901 struct flow_wildcards *wc = &xout->wc;
2902 struct flow *flow = &xin->flow;
2903 struct rule_dpif *rule = NULL;
2905 struct rule_actions *actions = NULL;
2906 enum slow_path_reason special;
2907 const struct ofpact *ofpacts;
2908 struct xport *in_port;
2909 struct flow orig_flow;
2910 struct xlate_ctx ctx;
2915 COVERAGE_INC(xlate_actions);
2917 /* Flow initialization rules:
2918 * - 'base_flow' must match the kernel's view of the packet at the
2919 * time that action processing starts. 'flow' represents any
2920 * transformations we wish to make through actions.
2921 * - By default 'base_flow' and 'flow' are the same since the input
2922 * packet matches the output before any actions are applied.
2923 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2924 * of the received packet as seen by the kernel. If we later output
2925 * to another device without any modifications this will cause us to
2926 * insert a new tag since the original one was stripped off by the
2928 * - Tunnel metadata as received is retained in 'flow'. This allows
2929 * tunnel metadata matching also in later tables.
2930 * Since a kernel action for setting the tunnel metadata will only be
2931 * generated with actual tunnel output, changing the tunnel metadata
2932 * values in 'flow' (such as tun_id) will only have effect with a later
2933 * tunnel output action.
2934 * - Tunnel 'base_flow' is completely cleared since that is what the
2935 * kernel does. If we wish to maintain the original values an action
2936 * needs to be generated. */
2941 ctx.xout->has_learn = false;
2942 ctx.xout->has_normal = false;
2943 ctx.xout->has_fin_timeout = false;
2944 ctx.xout->nf_output_iface = NF_OUT_DROP;
2945 ctx.xout->mirrors = 0;
2946 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2947 sizeof ctx.xout->odp_actions_stub);
2948 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2950 ctx.xbridge = xbridge_lookup(xin->ofproto);
2955 ctx.rule = xin->rule;
2957 ctx.base_flow = *flow;
2958 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2959 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2961 flow_wildcards_init_catchall(wc);
2962 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2963 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2964 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2965 if (is_ip_any(flow)) {
2966 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2968 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
2970 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
2971 if (ctx.xbridge->netflow) {
2972 netflow_mask_wc(flow, wc);
2977 ctx.orig_skb_priority = flow->skb_priority;
2981 if (!xin->ofpacts && !ctx.rule) {
2982 rule_dpif_lookup(ctx.xbridge->ofproto, flow,
2983 !xin->skip_wildcards ? wc : NULL, &rule);
2984 if (ctx.xin->resubmit_stats) {
2985 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
2989 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
2992 ofpacts = xin->ofpacts;
2993 ofpacts_len = xin->ofpacts_len;
2994 } else if (ctx.rule) {
2995 actions = rule_dpif_get_actions(ctx.rule);
2996 ofpacts = actions->ofpacts;
2997 ofpacts_len = actions->ofpacts_len;
3002 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3003 ofpbuf_use_stub(&ctx.action_set,
3004 ctx.action_set_stub, sizeof ctx.action_set_stub);
3006 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3007 /* Do this conditionally because the copy is expensive enough that it
3008 * shows up in profiles. */
3012 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3013 switch (ctx.xbridge->frag) {
3014 case OFPC_FRAG_NORMAL:
3015 /* We must pretend that transport ports are unavailable. */
3016 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3017 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3020 case OFPC_FRAG_DROP:
3023 case OFPC_FRAG_REASM:
3026 case OFPC_FRAG_NX_MATCH:
3027 /* Nothing to do. */
3030 case OFPC_INVALID_TTL_TO_CONTROLLER:
3035 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3036 if (in_port && in_port->is_tunnel && ctx.xin->resubmit_stats) {
3037 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3039 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3043 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3045 ctx.xout->slow |= special;
3047 size_t sample_actions_len;
3049 if (flow->in_port.ofp_port
3050 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3051 flow->in_port.ofp_port,
3053 ctx.base_flow.vlan_tci = 0;
3056 add_sflow_action(&ctx);
3057 add_ipfix_action(&ctx);
3058 sample_actions_len = ctx.xout->odp_actions.size;
3060 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3061 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3063 /* We've let OFPP_NORMAL and the learning action look at the
3064 * packet, so drop it now if forwarding is disabled. */
3065 if (in_port && !xport_stp_forward_state(in_port)) {
3066 ctx.xout->odp_actions.size = sample_actions_len;
3070 if (ctx.action_set.size) {
3071 xlate_action_set(&ctx);
3074 if (ctx.xbridge->has_in_band
3075 && in_band_must_output_to_local_port(flow)
3076 && !actions_output_to_local_port(&ctx)) {
3077 compose_output_action(&ctx, OFPP_LOCAL);
3080 fix_sflow_action(&ctx);
3082 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3083 add_mirror_actions(&ctx, &orig_flow);
3087 if (nl_attr_oversized(ctx.xout->odp_actions.size)) {
3088 /* These datapath actions are too big for a Netlink attribute, so we
3089 * can't hand them to the kernel directly. dpif_execute() can execute
3090 * them one by one with help, so just mark the result as SLOW_ACTION to
3091 * prevent the flow from being installed. */
3092 COVERAGE_INC(xlate_actions_oversize);
3093 ctx.xout->slow |= SLOW_ACTION;
3096 if (ctx.xin->resubmit_stats) {
3097 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3098 ctx.xin->resubmit_stats->n_packets,
3099 ctx.xin->resubmit_stats->n_bytes);
3101 if (ctx.xbridge->netflow) {
3102 const struct ofpact *ofpacts;
3105 ofpacts_len = actions->ofpacts_len;
3106 ofpacts = actions->ofpacts;
3107 if (ofpacts_len == 0
3108 || ofpacts->type != OFPACT_CONTROLLER
3109 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3110 /* Only update netflow if we don't have controller flow. We don't
3111 * report NetFlow expiration messages for such facets because they
3112 * are just part of the control logic for the network, not real
3114 netflow_flow_update(ctx.xbridge->netflow, flow,
3115 xout->nf_output_iface,
3116 ctx.xin->resubmit_stats);
3121 ofpbuf_uninit(&ctx.stack);
3122 ofpbuf_uninit(&ctx.action_set);
3124 /* Clear the metadata and register wildcard masks, because we won't
3125 * use non-header fields as part of the cache. */
3126 flow_wildcards_clear_non_packet_fields(wc);
3128 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3129 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3130 * these fields. The datapath interface, on the other hand, represents
3131 * them with just 8 bits each. This means that if the high 8 bits of the
3132 * masks for these fields somehow become set, then they will get chopped
3133 * off by a round trip through the datapath, and revalidation will spot
3134 * that as an inconsistency and delete the flow. Avoid the problem here by
3135 * making sure that only the low 8 bits of either field can be unwildcarded
3139 wc->masks.tp_src &= htons(UINT8_MAX);
3140 wc->masks.tp_dst &= htons(UINT8_MAX);
3144 rule_actions_unref(actions);
3145 rule_dpif_unref(rule);
3148 /* Sends 'packet' out 'ofport'.
3149 * May modify 'packet'.
3150 * Returns 0 if successful, otherwise a positive errno value. */
3152 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3154 struct xport *xport;
3155 struct ofpact_output output;
3157 union flow_in_port in_port_;
3159 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3160 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3161 in_port_.ofp_port = OFPP_NONE;
3162 flow_extract(packet, 0, 0, NULL, &in_port_, &flow);
3164 ovs_rwlock_rdlock(&xlate_rwlock);
3165 xport = xport_lookup(ofport);
3167 ovs_rwlock_unlock(&xlate_rwlock);
3170 output.port = xport->ofp_port;
3172 ovs_rwlock_unlock(&xlate_rwlock);
3174 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3175 &output.ofpact, sizeof output,