]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto-dpif-xlate.c
ofproto: Fix wildcard masking with nw_tos.
[mirror_ovs.git] / ofproto / ofproto-dpif-xlate.c
1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
2 *
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:
6 *
7 * http://www.apache.org/licenses/LICENSE-2.0
8 *
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. */
14
15 #include <config.h>
16
17 #include "ofproto/ofproto-dpif-xlate.h"
18
19 #include <errno.h>
20
21 #include "bfd.h"
22 #include "bitmap.h"
23 #include "bond.h"
24 #include "bundle.h"
25 #include "byte-order.h"
26 #include "cfm.h"
27 #include "connmgr.h"
28 #include "coverage.h"
29 #include "dpif.h"
30 #include "dynamic-string.h"
31 #include "in-band.h"
32 #include "lacp.h"
33 #include "learn.h"
34 #include "list.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
39 #include "netlink.h"
40 #include "nx-match.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"
49 #include "tunnel.h"
50 #include "vlog.h"
51
52 COVERAGE_DEFINE(xlate_actions);
53 COVERAGE_DEFINE(xlate_actions_oversize);
54 COVERAGE_DEFINE(xlate_actions_mpls_overflow);
55
56 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
57
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
62 internal table. */
63
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)
67
68 struct ovs_rwlock xlate_rwlock = OVS_RWLOCK_INITIALIZER;
69
70 struct xbridge {
71 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
72 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
73
74 struct list xbundles; /* Owned xbundles. */
75 struct hmap xports; /* Indexed by ofp_port. */
76
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. */
85
86 /* Special rules installed by ofproto-dpif. */
87 struct rule_dpif *miss_rule;
88 struct rule_dpif *no_packet_in_rule;
89
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? */
93
94 /* True if the datapath supports recirculation. */
95 bool enable_recirc;
96
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;
101
102 /* Number of MPLS label stack entries that the datapath supports
103 * in matches. */
104 size_t max_mpls_depth;
105 };
106
107 struct xbundle {
108 struct hmap_node hmap_node; /* In global 'xbundles' map. */
109 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
110
111 struct list list_node; /* In parent 'xbridges' list. */
112 struct xbridge *xbridge; /* Parent xbridge. */
113
114 struct list xports; /* Contains "struct xport"s. */
115
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. */
119
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? */
126 };
127
128 struct xport {
129 struct hmap_node hmap_node; /* Node in global 'xports' map. */
130 struct ofport_dpif *ofport; /* Key in global 'xports map. */
131
132 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
133 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
134
135 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
136
137 struct list bundle_node; /* In parent xbundle (if it exists). */
138 struct xbundle *xbundle; /* Parent xbundle or null. */
139
140 struct netdev *netdev; /* 'ofport''s netdev. */
141
142 struct xbridge *xbridge; /* Parent bridge. */
143 struct xport *peer; /* Patch port peer or null. */
144
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. */
148
149 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
150
151 bool may_enable; /* May be enabled in bonds. */
152 bool is_tunnel; /* Is a tunnel port. */
153
154 struct cfm *cfm; /* CFM handle or null. */
155 struct bfd *bfd; /* BFD handle or null. */
156 };
157
158 struct xlate_ctx {
159 struct xlate_in *xin;
160 struct xlate_out *xout;
161
162 const struct xbridge *xbridge;
163
164 /* Flow at the last commit. */
165 struct flow base_flow;
166
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;
174
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)];
178 struct ofpbuf stack;
179
180 /* The rule that we are currently translating, or NULL. */
181 struct rule_dpif *rule;
182
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. */
187
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. */
194
195 /* OpenFlow 1.1+ action set.
196 *
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];
203 };
204
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 = {
211 .name = "OFPP_NONE",
212 .vlan_mode = PORT_VLAN_TRUNK
213 };
214
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). */
221
222 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
223 };
224
225 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
226 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
227 static struct hmap xports = HMAP_INITIALIZER(&xports);
228
229 static bool may_receive(const struct xport *, struct xlate_ctx *);
230 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
231 struct xlate_ctx *);
232 static void xlate_actions__(struct xlate_in *, struct xlate_out *)
233 OVS_REQ_RDLOCK(xlate_rwlock);
234 static void xlate_normal(struct xlate_ctx *);
235 static void xlate_report(struct xlate_ctx *, const char *);
236 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
237 uint8_t table_id, bool may_packet_in,
238 bool honor_table_miss);
239 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
240 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
241 static void output_normal(struct xlate_ctx *, const struct xbundle *,
242 uint16_t vlan);
243 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
244
245 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
246 static struct xbundle *xbundle_lookup(const struct ofbundle *);
247 static struct xport *xport_lookup(const struct ofport_dpif *);
248 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
249 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
250 uint32_t skb_priority);
251 static void clear_skb_priorities(struct xport *);
252 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
253 uint8_t *dscp);
254
255 void
256 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
257 struct dpif *dpif, struct rule_dpif *miss_rule,
258 struct rule_dpif *no_packet_in_rule,
259 const struct mac_learning *ml, struct stp *stp,
260 const struct mbridge *mbridge,
261 const struct dpif_sflow *sflow,
262 const struct dpif_ipfix *ipfix,
263 const struct netflow *netflow, enum ofp_config_flags frag,
264 bool forward_bpdu, bool has_in_band,
265 bool enable_recirc,
266 bool variable_length_userdata,
267 size_t max_mpls_depth)
268 {
269 struct xbridge *xbridge = xbridge_lookup(ofproto);
270
271 if (!xbridge) {
272 xbridge = xzalloc(sizeof *xbridge);
273 xbridge->ofproto = ofproto;
274
275 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
276 hmap_init(&xbridge->xports);
277 list_init(&xbridge->xbundles);
278 }
279
280 if (xbridge->ml != ml) {
281 mac_learning_unref(xbridge->ml);
282 xbridge->ml = mac_learning_ref(ml);
283 }
284
285 if (xbridge->mbridge != mbridge) {
286 mbridge_unref(xbridge->mbridge);
287 xbridge->mbridge = mbridge_ref(mbridge);
288 }
289
290 if (xbridge->sflow != sflow) {
291 dpif_sflow_unref(xbridge->sflow);
292 xbridge->sflow = dpif_sflow_ref(sflow);
293 }
294
295 if (xbridge->ipfix != ipfix) {
296 dpif_ipfix_unref(xbridge->ipfix);
297 xbridge->ipfix = dpif_ipfix_ref(ipfix);
298 }
299
300 if (xbridge->stp != stp) {
301 stp_unref(xbridge->stp);
302 xbridge->stp = stp_ref(stp);
303 }
304
305 if (xbridge->netflow != netflow) {
306 netflow_unref(xbridge->netflow);
307 xbridge->netflow = netflow_ref(netflow);
308 }
309
310 free(xbridge->name);
311 xbridge->name = xstrdup(name);
312
313 xbridge->dpif = dpif;
314 xbridge->forward_bpdu = forward_bpdu;
315 xbridge->has_in_band = has_in_band;
316 xbridge->frag = frag;
317 xbridge->miss_rule = miss_rule;
318 xbridge->no_packet_in_rule = no_packet_in_rule;
319 xbridge->enable_recirc = enable_recirc;
320 xbridge->variable_length_userdata = variable_length_userdata;
321 xbridge->max_mpls_depth = max_mpls_depth;
322 }
323
324 void
325 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
326 {
327 struct xbridge *xbridge = xbridge_lookup(ofproto);
328 struct xbundle *xbundle, *next_xbundle;
329 struct xport *xport, *next_xport;
330
331 if (!xbridge) {
332 return;
333 }
334
335 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
336 xlate_ofport_remove(xport->ofport);
337 }
338
339 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
340 xlate_bundle_remove(xbundle->ofbundle);
341 }
342
343 hmap_remove(&xbridges, &xbridge->hmap_node);
344 mac_learning_unref(xbridge->ml);
345 mbridge_unref(xbridge->mbridge);
346 dpif_sflow_unref(xbridge->sflow);
347 dpif_ipfix_unref(xbridge->ipfix);
348 stp_unref(xbridge->stp);
349 hmap_destroy(&xbridge->xports);
350 free(xbridge->name);
351 free(xbridge);
352 }
353
354 void
355 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
356 const char *name, enum port_vlan_mode vlan_mode, int vlan,
357 unsigned long *trunks, bool use_priority_tags,
358 const struct bond *bond, const struct lacp *lacp,
359 bool floodable)
360 {
361 struct xbundle *xbundle = xbundle_lookup(ofbundle);
362
363 if (!xbundle) {
364 xbundle = xzalloc(sizeof *xbundle);
365 xbundle->ofbundle = ofbundle;
366 xbundle->xbridge = xbridge_lookup(ofproto);
367
368 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
369 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
370 list_init(&xbundle->xports);
371 }
372
373 ovs_assert(xbundle->xbridge);
374
375 free(xbundle->name);
376 xbundle->name = xstrdup(name);
377
378 xbundle->vlan_mode = vlan_mode;
379 xbundle->vlan = vlan;
380 xbundle->trunks = trunks;
381 xbundle->use_priority_tags = use_priority_tags;
382 xbundle->floodable = floodable;
383
384 if (xbundle->bond != bond) {
385 bond_unref(xbundle->bond);
386 xbundle->bond = bond_ref(bond);
387 }
388
389 if (xbundle->lacp != lacp) {
390 lacp_unref(xbundle->lacp);
391 xbundle->lacp = lacp_ref(lacp);
392 }
393 }
394
395 void
396 xlate_bundle_remove(struct ofbundle *ofbundle)
397 {
398 struct xbundle *xbundle = xbundle_lookup(ofbundle);
399 struct xport *xport, *next;
400
401 if (!xbundle) {
402 return;
403 }
404
405 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
406 list_remove(&xport->bundle_node);
407 xport->xbundle = NULL;
408 }
409
410 hmap_remove(&xbundles, &xbundle->hmap_node);
411 list_remove(&xbundle->list_node);
412 bond_unref(xbundle->bond);
413 lacp_unref(xbundle->lacp);
414 free(xbundle->name);
415 free(xbundle);
416 }
417
418 void
419 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
420 struct ofport_dpif *ofport, ofp_port_t ofp_port,
421 odp_port_t odp_port, const struct netdev *netdev,
422 const struct cfm *cfm, const struct bfd *bfd,
423 struct ofport_dpif *peer, int stp_port_no,
424 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
425 enum ofputil_port_config config,
426 enum ofputil_port_state state, bool is_tunnel,
427 bool may_enable)
428 {
429 struct xport *xport = xport_lookup(ofport);
430 size_t i;
431
432 if (!xport) {
433 xport = xzalloc(sizeof *xport);
434 xport->ofport = ofport;
435 xport->xbridge = xbridge_lookup(ofproto);
436 xport->ofp_port = ofp_port;
437
438 hmap_init(&xport->skb_priorities);
439 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
440 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
441 hash_ofp_port(xport->ofp_port));
442 }
443
444 ovs_assert(xport->ofp_port == ofp_port);
445
446 xport->config = config;
447 xport->state = state;
448 xport->stp_port_no = stp_port_no;
449 xport->is_tunnel = is_tunnel;
450 xport->may_enable = may_enable;
451 xport->odp_port = odp_port;
452
453 if (xport->netdev != netdev) {
454 netdev_close(xport->netdev);
455 xport->netdev = netdev_ref(netdev);
456 }
457
458 if (xport->cfm != cfm) {
459 cfm_unref(xport->cfm);
460 xport->cfm = cfm_ref(cfm);
461 }
462
463 if (xport->bfd != bfd) {
464 bfd_unref(xport->bfd);
465 xport->bfd = bfd_ref(bfd);
466 }
467
468 if (xport->peer) {
469 xport->peer->peer = NULL;
470 }
471 xport->peer = xport_lookup(peer);
472 if (xport->peer) {
473 xport->peer->peer = xport;
474 }
475
476 if (xport->xbundle) {
477 list_remove(&xport->bundle_node);
478 }
479 xport->xbundle = xbundle_lookup(ofbundle);
480 if (xport->xbundle) {
481 list_insert(&xport->xbundle->xports, &xport->bundle_node);
482 }
483
484 clear_skb_priorities(xport);
485 for (i = 0; i < n_qdscp; i++) {
486 struct skb_priority_to_dscp *pdscp;
487 uint32_t skb_priority;
488
489 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
490 &skb_priority)) {
491 continue;
492 }
493
494 pdscp = xmalloc(sizeof *pdscp);
495 pdscp->skb_priority = skb_priority;
496 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
497 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
498 hash_int(pdscp->skb_priority, 0));
499 }
500 }
501
502 void
503 xlate_ofport_remove(struct ofport_dpif *ofport)
504 {
505 struct xport *xport = xport_lookup(ofport);
506
507 if (!xport) {
508 return;
509 }
510
511 if (xport->peer) {
512 xport->peer->peer = NULL;
513 xport->peer = NULL;
514 }
515
516 if (xport->xbundle) {
517 list_remove(&xport->bundle_node);
518 }
519
520 clear_skb_priorities(xport);
521 hmap_destroy(&xport->skb_priorities);
522
523 hmap_remove(&xports, &xport->hmap_node);
524 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
525
526 netdev_close(xport->netdev);
527 cfm_unref(xport->cfm);
528 bfd_unref(xport->bfd);
529 free(xport);
530 }
531
532 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
533 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
534 * Optionally populates 'ofproto' with the ofproto_dpif, 'odp_in_port' with
535 * the datapath in_port, that 'packet' ingressed, and 'ipfix', 'sflow', and
536 * 'netflow' with the appropriate handles for those protocols if they're
537 * enabled. Caller is responsible for unrefing them.
538 *
539 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
540 * 'flow''s in_port to OFPP_NONE.
541 *
542 * This function does post-processing on data returned from
543 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
544 * of the upcall processing logic. In particular, if the extracted in_port is
545 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
546 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
547 * a VLAN header onto 'packet' (if it is nonnull).
548 *
549 * Similarly, this function also includes some logic to help with tunnels. It
550 * may modify 'flow' as necessary to make the tunneling implementation
551 * transparent to the upcall processing logic.
552 *
553 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
554 * or some other positive errno if there are other problems. */
555 int
556 xlate_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
557 const struct nlattr *key, size_t key_len, struct flow *flow,
558 struct ofproto_dpif **ofproto, struct dpif_ipfix **ipfix,
559 struct dpif_sflow **sflow, struct netflow **netflow,
560 odp_port_t *odp_in_port)
561 {
562 const struct xport *xport;
563 int error = ENODEV;
564
565 ovs_rwlock_rdlock(&xlate_rwlock);
566 if (odp_flow_key_to_flow(key, key_len, flow) == ODP_FIT_ERROR) {
567 error = EINVAL;
568 goto exit;
569 }
570
571 if (odp_in_port) {
572 *odp_in_port = flow->in_port.odp_port;
573 }
574
575 xport = xport_lookup(tnl_port_should_receive(flow)
576 ? tnl_port_receive(flow)
577 : odp_port_to_ofport(backer, flow->in_port.odp_port));
578
579 flow->in_port.ofp_port = xport ? xport->ofp_port : OFPP_NONE;
580 if (!xport) {
581 goto exit;
582 }
583
584 if (vsp_adjust_flow(xport->xbridge->ofproto, flow)) {
585 if (packet) {
586 /* Make the packet resemble the flow, so that it gets sent to
587 * an OpenFlow controller properly, so that it looks correct
588 * for sFlow, and so that flow_extract() will get the correct
589 * vlan_tci if it is called on 'packet'. */
590 eth_push_vlan(packet, htons(ETH_TYPE_VLAN), flow->vlan_tci);
591 }
592 }
593 error = 0;
594
595 if (ofproto) {
596 *ofproto = xport->xbridge->ofproto;
597 }
598
599 if (ipfix) {
600 *ipfix = dpif_ipfix_ref(xport->xbridge->ipfix);
601 }
602
603 if (sflow) {
604 *sflow = dpif_sflow_ref(xport->xbridge->sflow);
605 }
606
607 if (netflow) {
608 *netflow = netflow_ref(xport->xbridge->netflow);
609 }
610
611 exit:
612 ovs_rwlock_unlock(&xlate_rwlock);
613 return error;
614 }
615
616 static struct xbridge *
617 xbridge_lookup(const struct ofproto_dpif *ofproto)
618 {
619 struct xbridge *xbridge;
620
621 if (!ofproto) {
622 return NULL;
623 }
624
625 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
626 &xbridges) {
627 if (xbridge->ofproto == ofproto) {
628 return xbridge;
629 }
630 }
631 return NULL;
632 }
633
634 static struct xbundle *
635 xbundle_lookup(const struct ofbundle *ofbundle)
636 {
637 struct xbundle *xbundle;
638
639 if (!ofbundle) {
640 return NULL;
641 }
642
643 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
644 &xbundles) {
645 if (xbundle->ofbundle == ofbundle) {
646 return xbundle;
647 }
648 }
649 return NULL;
650 }
651
652 static struct xport *
653 xport_lookup(const struct ofport_dpif *ofport)
654 {
655 struct xport *xport;
656
657 if (!ofport) {
658 return NULL;
659 }
660
661 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
662 &xports) {
663 if (xport->ofport == ofport) {
664 return xport;
665 }
666 }
667 return NULL;
668 }
669
670 static struct stp_port *
671 xport_get_stp_port(const struct xport *xport)
672 {
673 return xport->xbridge->stp && xport->stp_port_no != -1
674 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
675 : NULL;
676 }
677
678 static bool
679 xport_stp_learn_state(const struct xport *xport)
680 {
681 struct stp_port *sp = xport_get_stp_port(xport);
682 return stp_learn_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
683 }
684
685 static bool
686 xport_stp_forward_state(const struct xport *xport)
687 {
688 struct stp_port *sp = xport_get_stp_port(xport);
689 return stp_forward_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
690 }
691
692 static bool
693 xport_stp_listen_state(const struct xport *xport)
694 {
695 struct stp_port *sp = xport_get_stp_port(xport);
696 return stp_listen_in_state(sp ? stp_port_get_state(sp) : STP_DISABLED);
697 }
698
699 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
700 * were used to make the determination.*/
701 static bool
702 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
703 {
704 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
705 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
706 }
707
708 static void
709 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
710 {
711 struct stp_port *sp = xport_get_stp_port(xport);
712 struct ofpbuf payload = *packet;
713 struct eth_header *eth = ofpbuf_data(&payload);
714
715 /* Sink packets on ports that have STP disabled when the bridge has
716 * STP enabled. */
717 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
718 return;
719 }
720
721 /* Trim off padding on payload. */
722 if (ofpbuf_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
723 ofpbuf_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
724 }
725
726 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
727 stp_received_bpdu(sp, ofpbuf_data(&payload), ofpbuf_size(&payload));
728 }
729 }
730
731 static struct xport *
732 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
733 {
734 struct xport *xport;
735
736 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
737 &xbridge->xports) {
738 if (xport->ofp_port == ofp_port) {
739 return xport;
740 }
741 }
742 return NULL;
743 }
744
745 static odp_port_t
746 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
747 {
748 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
749 return xport ? xport->odp_port : ODPP_NONE;
750 }
751
752 static bool
753 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
754 {
755 struct xport *xport;
756
757 xport = get_ofp_port(ctx->xbridge, ofp_port);
758 if (!xport || xport->config & OFPUTIL_PC_PORT_DOWN ||
759 xport->state & OFPUTIL_PS_LINK_DOWN) {
760 return false;
761 }
762
763 return true;
764 }
765
766 static const struct ofputil_bucket *
767 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
768 int depth);
769
770 static bool
771 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
772 {
773 struct group_dpif *group;
774 bool hit;
775
776 hit = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
777 if (!hit) {
778 return false;
779 }
780
781 hit = group_first_live_bucket(ctx, group, depth) != NULL;
782
783 group_dpif_release(group);
784 return hit;
785 }
786
787 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
788
789 static bool
790 bucket_is_alive(const struct xlate_ctx *ctx,
791 const struct ofputil_bucket *bucket, int depth)
792 {
793 if (depth >= MAX_LIVENESS_RECURSION) {
794 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
795
796 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
797 MAX_LIVENESS_RECURSION);
798 return false;
799 }
800
801 return !ofputil_bucket_has_liveness(bucket) ||
802 (bucket->watch_port != OFPP_ANY &&
803 odp_port_is_alive(ctx, bucket->watch_port)) ||
804 (bucket->watch_group != OFPG_ANY &&
805 group_is_alive(ctx, bucket->watch_group, depth + 1));
806 }
807
808 static const struct ofputil_bucket *
809 group_first_live_bucket(const struct xlate_ctx *ctx,
810 const struct group_dpif *group, int depth)
811 {
812 struct ofputil_bucket *bucket;
813 const struct list *buckets;
814
815 group_dpif_get_buckets(group, &buckets);
816 LIST_FOR_EACH (bucket, list_node, buckets) {
817 if (bucket_is_alive(ctx, bucket, depth)) {
818 return bucket;
819 }
820 }
821
822 return NULL;
823 }
824
825 static const struct ofputil_bucket *
826 group_best_live_bucket(const struct xlate_ctx *ctx,
827 const struct group_dpif *group,
828 uint32_t basis)
829 {
830 const struct ofputil_bucket *best_bucket = NULL;
831 uint32_t best_score = 0;
832 int i = 0;
833
834 const struct ofputil_bucket *bucket;
835 const struct list *buckets;
836
837 group_dpif_get_buckets(group, &buckets);
838 LIST_FOR_EACH (bucket, list_node, buckets) {
839 if (bucket_is_alive(ctx, bucket, 0)) {
840 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
841 if (score >= best_score) {
842 best_bucket = bucket;
843 best_score = score;
844 }
845 }
846 i++;
847 }
848
849 return best_bucket;
850 }
851
852 static bool
853 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
854 {
855 return (bundle->vlan_mode != PORT_VLAN_ACCESS
856 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
857 }
858
859 static bool
860 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
861 {
862 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
863 }
864
865 static mirror_mask_t
866 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
867 {
868 return xbundle != &ofpp_none_bundle
869 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
870 : 0;
871 }
872
873 static mirror_mask_t
874 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
875 {
876 return xbundle != &ofpp_none_bundle
877 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
878 : 0;
879 }
880
881 static mirror_mask_t
882 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
883 {
884 return xbundle != &ofpp_none_bundle
885 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
886 : 0;
887 }
888
889 static struct xbundle *
890 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
891 bool warn, struct xport **in_xportp)
892 {
893 struct xport *xport;
894
895 /* Find the port and bundle for the received packet. */
896 xport = get_ofp_port(xbridge, in_port);
897 if (in_xportp) {
898 *in_xportp = xport;
899 }
900 if (xport && xport->xbundle) {
901 return xport->xbundle;
902 }
903
904 /* Special-case OFPP_NONE, which a controller may use as the ingress
905 * port for traffic that it is sourcing. */
906 if (in_port == OFPP_NONE) {
907 return &ofpp_none_bundle;
908 }
909
910 /* Odd. A few possible reasons here:
911 *
912 * - We deleted a port but there are still a few packets queued up
913 * from it.
914 *
915 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
916 * we don't know about.
917 *
918 * - The ofproto client didn't configure the port as part of a bundle.
919 * This is particularly likely to happen if a packet was received on the
920 * port after it was created, but before the client had a chance to
921 * configure its bundle.
922 */
923 if (warn) {
924 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
925
926 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
927 "port %"PRIu16, xbridge->name, in_port);
928 }
929 return NULL;
930 }
931
932 static void
933 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
934 {
935 const struct xbridge *xbridge = ctx->xbridge;
936 mirror_mask_t mirrors;
937 struct xbundle *in_xbundle;
938 uint16_t vlan;
939 uint16_t vid;
940
941 mirrors = ctx->xout->mirrors;
942 ctx->xout->mirrors = 0;
943
944 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
945 ctx->xin->packet != NULL, NULL);
946 if (!in_xbundle) {
947 return;
948 }
949 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
950
951 /* Drop frames on bundles reserved for mirroring. */
952 if (xbundle_mirror_out(xbridge, in_xbundle)) {
953 if (ctx->xin->packet != NULL) {
954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
955 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
956 "%s, which is reserved exclusively for mirroring",
957 ctx->xbridge->name, in_xbundle->name);
958 }
959 ofpbuf_clear(&ctx->xout->odp_actions);
960 return;
961 }
962
963 /* Check VLAN. */
964 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
965 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
966 return;
967 }
968 vlan = input_vid_to_vlan(in_xbundle, vid);
969
970 if (!mirrors) {
971 return;
972 }
973
974 /* Restore the original packet before adding the mirror actions. */
975 ctx->xin->flow = *orig_flow;
976
977 while (mirrors) {
978 mirror_mask_t dup_mirrors;
979 struct ofbundle *out;
980 unsigned long *vlans;
981 bool vlan_mirrored;
982 bool has_mirror;
983 int out_vlan;
984
985 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
986 &vlans, &dup_mirrors, &out, &out_vlan);
987 ovs_assert(has_mirror);
988
989 if (vlans) {
990 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
991 }
992 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
993 free(vlans);
994
995 if (!vlan_mirrored) {
996 mirrors = zero_rightmost_1bit(mirrors);
997 continue;
998 }
999
1000 mirrors &= ~dup_mirrors;
1001 ctx->xout->mirrors |= dup_mirrors;
1002 if (out) {
1003 struct xbundle *out_xbundle = xbundle_lookup(out);
1004 if (out_xbundle) {
1005 output_normal(ctx, out_xbundle, vlan);
1006 }
1007 } else if (vlan != out_vlan
1008 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1009 struct xbundle *xbundle;
1010
1011 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1012 if (xbundle_includes_vlan(xbundle, out_vlan)
1013 && !xbundle_mirror_out(xbridge, xbundle)) {
1014 output_normal(ctx, xbundle, out_vlan);
1015 }
1016 }
1017 }
1018 }
1019 }
1020
1021 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1022 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1023 * the bundle on which the packet was received, returns the VLAN to which the
1024 * packet belongs.
1025 *
1026 * Both 'vid' and the return value are in the range 0...4095. */
1027 static uint16_t
1028 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1029 {
1030 switch (in_xbundle->vlan_mode) {
1031 case PORT_VLAN_ACCESS:
1032 return in_xbundle->vlan;
1033 break;
1034
1035 case PORT_VLAN_TRUNK:
1036 return vid;
1037
1038 case PORT_VLAN_NATIVE_UNTAGGED:
1039 case PORT_VLAN_NATIVE_TAGGED:
1040 return vid ? vid : in_xbundle->vlan;
1041
1042 default:
1043 OVS_NOT_REACHED();
1044 }
1045 }
1046
1047 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1048 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1049 * a warning.
1050 *
1051 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1052 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1053 * 0...4095. */
1054 static bool
1055 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1056 {
1057 /* Allow any VID on the OFPP_NONE port. */
1058 if (in_xbundle == &ofpp_none_bundle) {
1059 return true;
1060 }
1061
1062 switch (in_xbundle->vlan_mode) {
1063 case PORT_VLAN_ACCESS:
1064 if (vid) {
1065 if (warn) {
1066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1067 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1068 "packet received on port %s configured as VLAN "
1069 "%"PRIu16" access port", vid, in_xbundle->name,
1070 in_xbundle->vlan);
1071 }
1072 return false;
1073 }
1074 return true;
1075
1076 case PORT_VLAN_NATIVE_UNTAGGED:
1077 case PORT_VLAN_NATIVE_TAGGED:
1078 if (!vid) {
1079 /* Port must always carry its native VLAN. */
1080 return true;
1081 }
1082 /* Fall through. */
1083 case PORT_VLAN_TRUNK:
1084 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1085 if (warn) {
1086 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1087 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1088 "received on port %s not configured for trunking "
1089 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1090 }
1091 return false;
1092 }
1093 return true;
1094
1095 default:
1096 OVS_NOT_REACHED();
1097 }
1098
1099 }
1100
1101 /* Given 'vlan', the VLAN that a packet belongs to, and
1102 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1103 * that should be included in the 802.1Q header. (If the return value is 0,
1104 * then the 802.1Q header should only be included in the packet if there is a
1105 * nonzero PCP.)
1106 *
1107 * Both 'vlan' and the return value are in the range 0...4095. */
1108 static uint16_t
1109 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1110 {
1111 switch (out_xbundle->vlan_mode) {
1112 case PORT_VLAN_ACCESS:
1113 return 0;
1114
1115 case PORT_VLAN_TRUNK:
1116 case PORT_VLAN_NATIVE_TAGGED:
1117 return vlan;
1118
1119 case PORT_VLAN_NATIVE_UNTAGGED:
1120 return vlan == out_xbundle->vlan ? 0 : vlan;
1121
1122 default:
1123 OVS_NOT_REACHED();
1124 }
1125 }
1126
1127 static void
1128 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1129 uint16_t vlan)
1130 {
1131 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1132 uint16_t vid;
1133 ovs_be16 tci, old_tci;
1134 struct xport *xport;
1135
1136 vid = output_vlan_to_vid(out_xbundle, vlan);
1137 if (list_is_empty(&out_xbundle->xports)) {
1138 /* Partially configured bundle with no slaves. Drop the packet. */
1139 return;
1140 } else if (!out_xbundle->bond) {
1141 ctx->xout->use_recirc = false;
1142 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1143 bundle_node);
1144 } else {
1145 struct ofport_dpif *ofport;
1146 struct xlate_recirc *xr = &ctx->xout->recirc;
1147 struct flow_wildcards *wc = &ctx->xout->wc;
1148
1149 if (ctx->xbridge->enable_recirc) {
1150 ctx->xout->use_recirc = bond_may_recirc(
1151 out_xbundle->bond, &xr->recirc_id, &xr->hash_bias);
1152
1153 if (ctx->xout->use_recirc) {
1154 /* Only TCP mode uses recirculation. */
1155 xr->hash_alg = OVS_RECIRC_HASH_ALG_L4;
1156 bond_update_post_recirc_rules(out_xbundle->bond, false);
1157
1158 /* Recirculation does not require unmasking hash fields. */
1159 wc = NULL;
1160 }
1161 }
1162
1163 ofport = bond_choose_output_slave(out_xbundle->bond,
1164 &ctx->xin->flow, wc, vid);
1165 xport = xport_lookup(ofport);
1166
1167 if (!xport) {
1168 /* No slaves enabled, so drop packet. */
1169 return;
1170 }
1171
1172 if (ctx->xin->resubmit_stats) {
1173 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1174 ctx->xin->resubmit_stats->n_bytes);
1175 }
1176 }
1177
1178 old_tci = *flow_tci;
1179 tci = htons(vid);
1180 if (tci || out_xbundle->use_priority_tags) {
1181 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1182 if (tci) {
1183 tci |= htons(VLAN_CFI);
1184 }
1185 }
1186 *flow_tci = tci;
1187
1188 compose_output_action(ctx, xport->ofp_port);
1189 *flow_tci = old_tci;
1190 }
1191
1192 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1193 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1194 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1195 static bool
1196 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1197 {
1198 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1199 return false;
1200 }
1201
1202 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1203 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1204 return false;
1205 }
1206
1207 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1208 if (flow->nw_proto == ARP_OP_REPLY) {
1209 return true;
1210 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1211 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1212 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1213
1214 return flow->nw_src == flow->nw_dst;
1215 } else {
1216 return false;
1217 }
1218 }
1219
1220 /* Checks whether a MAC learning update is necessary for MAC learning table
1221 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1222 * 'vlan'.
1223 *
1224 * Most packets processed through the MAC learning table do not actually
1225 * change it in any way. This function requires only a read lock on the MAC
1226 * learning table, so it is much cheaper in this common case.
1227 *
1228 * Keep the code here synchronized with that in update_learning_table__()
1229 * below. */
1230 static bool
1231 is_mac_learning_update_needed(const struct mac_learning *ml,
1232 const struct flow *flow,
1233 struct flow_wildcards *wc,
1234 int vlan, struct xbundle *in_xbundle)
1235 OVS_REQ_RDLOCK(ml->rwlock)
1236 {
1237 struct mac_entry *mac;
1238
1239 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1240 return false;
1241 }
1242
1243 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1244 if (!mac || mac_entry_age(ml, mac)) {
1245 return true;
1246 }
1247
1248 if (is_gratuitous_arp(flow, wc)) {
1249 /* We don't want to learn from gratuitous ARP packets that are
1250 * reflected back over bond slaves so we lock the learning table. */
1251 if (!in_xbundle->bond) {
1252 return true;
1253 } else if (mac_entry_is_grat_arp_locked(mac)) {
1254 return false;
1255 }
1256 }
1257
1258 return mac->port.p != in_xbundle->ofbundle;
1259 }
1260
1261
1262 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1263 * received on 'in_xbundle' in 'vlan'.
1264 *
1265 * This code repeats all the checks in is_mac_learning_update_needed() because
1266 * the lock was released between there and here and thus the MAC learning state
1267 * could have changed.
1268 *
1269 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1270 * above. */
1271 static void
1272 update_learning_table__(const struct xbridge *xbridge,
1273 const struct flow *flow, struct flow_wildcards *wc,
1274 int vlan, struct xbundle *in_xbundle)
1275 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1276 {
1277 struct mac_entry *mac;
1278
1279 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1280 return;
1281 }
1282
1283 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1284 if (is_gratuitous_arp(flow, wc)) {
1285 /* We don't want to learn from gratuitous ARP packets that are
1286 * reflected back over bond slaves so we lock the learning table. */
1287 if (!in_xbundle->bond) {
1288 mac_entry_set_grat_arp_lock(mac);
1289 } else if (mac_entry_is_grat_arp_locked(mac)) {
1290 return;
1291 }
1292 }
1293
1294 if (mac->port.p != in_xbundle->ofbundle) {
1295 /* The log messages here could actually be useful in debugging,
1296 * so keep the rate limit relatively high. */
1297 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1298
1299 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1300 "on port %s in VLAN %d",
1301 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1302 in_xbundle->name, vlan);
1303
1304 mac->port.p = in_xbundle->ofbundle;
1305 mac_learning_changed(xbridge->ml);
1306 }
1307 }
1308
1309 static void
1310 update_learning_table(const struct xbridge *xbridge,
1311 const struct flow *flow, struct flow_wildcards *wc,
1312 int vlan, struct xbundle *in_xbundle)
1313 {
1314 bool need_update;
1315
1316 /* Don't learn the OFPP_NONE port. */
1317 if (in_xbundle == &ofpp_none_bundle) {
1318 return;
1319 }
1320
1321 /* First try the common case: no change to MAC learning table. */
1322 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1323 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1324 in_xbundle);
1325 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1326
1327 if (need_update) {
1328 /* Slow path: MAC learning table might need an update. */
1329 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
1330 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
1331 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1332 }
1333 }
1334
1335 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1336 * dropped. Returns true if they may be forwarded, false if they should be
1337 * dropped.
1338 *
1339 * 'in_port' must be the xport that corresponds to flow->in_port.
1340 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1341 *
1342 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1343 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1344 * checked by input_vid_is_valid().
1345 *
1346 * May also add tags to '*tags', although the current implementation only does
1347 * so in one special case.
1348 */
1349 static bool
1350 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1351 uint16_t vlan)
1352 {
1353 struct xbundle *in_xbundle = in_port->xbundle;
1354 const struct xbridge *xbridge = ctx->xbridge;
1355 struct flow *flow = &ctx->xin->flow;
1356
1357 /* Drop frames for reserved multicast addresses
1358 * only if forward_bpdu option is absent. */
1359 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1360 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1361 return false;
1362 }
1363
1364 if (in_xbundle->bond) {
1365 struct mac_entry *mac;
1366
1367 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1368 flow->dl_dst)) {
1369 case BV_ACCEPT:
1370 break;
1371
1372 case BV_DROP:
1373 xlate_report(ctx, "bonding refused admissibility, dropping");
1374 return false;
1375
1376 case BV_DROP_IF_MOVED:
1377 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1378 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1379 if (mac && mac->port.p != in_xbundle->ofbundle &&
1380 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1381 || mac_entry_is_grat_arp_locked(mac))) {
1382 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1383 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1384 "dropping");
1385 return false;
1386 }
1387 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1388 break;
1389 }
1390 }
1391
1392 return true;
1393 }
1394
1395 static void
1396 xlate_normal(struct xlate_ctx *ctx)
1397 {
1398 struct flow_wildcards *wc = &ctx->xout->wc;
1399 struct flow *flow = &ctx->xin->flow;
1400 struct xbundle *in_xbundle;
1401 struct xport *in_port;
1402 struct mac_entry *mac;
1403 void *mac_port;
1404 uint16_t vlan;
1405 uint16_t vid;
1406
1407 ctx->xout->has_normal = true;
1408
1409 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1410 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1411 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1412
1413 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1414 ctx->xin->packet != NULL, &in_port);
1415 if (!in_xbundle) {
1416 xlate_report(ctx, "no input bundle, dropping");
1417 return;
1418 }
1419
1420 /* Drop malformed frames. */
1421 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1422 !(flow->vlan_tci & htons(VLAN_CFI))) {
1423 if (ctx->xin->packet != NULL) {
1424 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1425 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1426 "VLAN tag received on port %s",
1427 ctx->xbridge->name, in_xbundle->name);
1428 }
1429 xlate_report(ctx, "partial VLAN tag, dropping");
1430 return;
1431 }
1432
1433 /* Drop frames on bundles reserved for mirroring. */
1434 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1435 if (ctx->xin->packet != NULL) {
1436 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1437 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1438 "%s, which is reserved exclusively for mirroring",
1439 ctx->xbridge->name, in_xbundle->name);
1440 }
1441 xlate_report(ctx, "input port is mirror output port, dropping");
1442 return;
1443 }
1444
1445 /* Check VLAN. */
1446 vid = vlan_tci_to_vid(flow->vlan_tci);
1447 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1448 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1449 return;
1450 }
1451 vlan = input_vid_to_vlan(in_xbundle, vid);
1452
1453 /* Check other admissibility requirements. */
1454 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1455 return;
1456 }
1457
1458 /* Learn source MAC. */
1459 if (ctx->xin->may_learn) {
1460 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1461 }
1462
1463 /* Determine output bundle. */
1464 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1465 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1466 mac_port = mac ? mac->port.p : NULL;
1467 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1468
1469 if (mac_port) {
1470 struct xbundle *mac_xbundle = xbundle_lookup(mac_port);
1471 if (mac_xbundle && mac_xbundle != in_xbundle) {
1472 xlate_report(ctx, "forwarding to learned port");
1473 output_normal(ctx, mac_xbundle, vlan);
1474 } else if (!mac_xbundle) {
1475 xlate_report(ctx, "learned port is unknown, dropping");
1476 } else {
1477 xlate_report(ctx, "learned port is input port, dropping");
1478 }
1479 } else {
1480 struct xbundle *xbundle;
1481
1482 xlate_report(ctx, "no learned MAC for destination, flooding");
1483 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1484 if (xbundle != in_xbundle
1485 && xbundle_includes_vlan(xbundle, vlan)
1486 && xbundle->floodable
1487 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1488 output_normal(ctx, xbundle, vlan);
1489 }
1490 }
1491 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1492 }
1493 }
1494
1495 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1496 * the number of packets out of UINT32_MAX to sample. The given
1497 * cookie is passed back in the callback for each sampled packet.
1498 */
1499 static size_t
1500 compose_sample_action(const struct xbridge *xbridge,
1501 struct ofpbuf *odp_actions,
1502 const struct flow *flow,
1503 const uint32_t probability,
1504 const union user_action_cookie *cookie,
1505 const size_t cookie_size)
1506 {
1507 size_t sample_offset, actions_offset;
1508 odp_port_t odp_port;
1509 int cookie_offset;
1510 uint32_t pid;
1511
1512 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1513
1514 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1515
1516 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1517
1518 odp_port = ofp_port_to_odp_port(xbridge, flow->in_port.ofp_port);
1519 pid = dpif_port_get_pid(xbridge->dpif, odp_port, 0);
1520 cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
1521
1522 nl_msg_end_nested(odp_actions, actions_offset);
1523 nl_msg_end_nested(odp_actions, sample_offset);
1524 return cookie_offset;
1525 }
1526
1527 static void
1528 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1529 odp_port_t odp_port, unsigned int n_outputs,
1530 union user_action_cookie *cookie)
1531 {
1532 int ifindex;
1533
1534 cookie->type = USER_ACTION_COOKIE_SFLOW;
1535 cookie->sflow.vlan_tci = vlan_tci;
1536
1537 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1538 * port information") for the interpretation of cookie->output. */
1539 switch (n_outputs) {
1540 case 0:
1541 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1542 cookie->sflow.output = 0x40000000 | 256;
1543 break;
1544
1545 case 1:
1546 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1547 if (ifindex) {
1548 cookie->sflow.output = ifindex;
1549 break;
1550 }
1551 /* Fall through. */
1552 default:
1553 /* 0x80000000 means "multiple output ports. */
1554 cookie->sflow.output = 0x80000000 | n_outputs;
1555 break;
1556 }
1557 }
1558
1559 /* Compose SAMPLE action for sFlow bridge sampling. */
1560 static size_t
1561 compose_sflow_action(const struct xbridge *xbridge,
1562 struct ofpbuf *odp_actions,
1563 const struct flow *flow,
1564 odp_port_t odp_port)
1565 {
1566 uint32_t probability;
1567 union user_action_cookie cookie;
1568
1569 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1570 return 0;
1571 }
1572
1573 probability = dpif_sflow_get_probability(xbridge->sflow);
1574 compose_sflow_cookie(xbridge, htons(0), odp_port,
1575 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1576
1577 return compose_sample_action(xbridge, odp_actions, flow, probability,
1578 &cookie, sizeof cookie.sflow);
1579 }
1580
1581 static void
1582 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1583 uint32_t obs_domain_id, uint32_t obs_point_id,
1584 union user_action_cookie *cookie)
1585 {
1586 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1587 cookie->flow_sample.probability = probability;
1588 cookie->flow_sample.collector_set_id = collector_set_id;
1589 cookie->flow_sample.obs_domain_id = obs_domain_id;
1590 cookie->flow_sample.obs_point_id = obs_point_id;
1591 }
1592
1593 static void
1594 compose_ipfix_cookie(union user_action_cookie *cookie)
1595 {
1596 cookie->type = USER_ACTION_COOKIE_IPFIX;
1597 }
1598
1599 /* Compose SAMPLE action for IPFIX bridge sampling. */
1600 static void
1601 compose_ipfix_action(const struct xbridge *xbridge,
1602 struct ofpbuf *odp_actions,
1603 const struct flow *flow)
1604 {
1605 uint32_t probability;
1606 union user_action_cookie cookie;
1607
1608 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1609 return;
1610 }
1611
1612 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1613 compose_ipfix_cookie(&cookie);
1614
1615 compose_sample_action(xbridge, odp_actions, flow, probability,
1616 &cookie, sizeof cookie.ipfix);
1617 }
1618
1619 /* SAMPLE action for sFlow must be first action in any given list of
1620 * actions. At this point we do not have all information required to
1621 * build it. So try to build sample action as complete as possible. */
1622 static void
1623 add_sflow_action(struct xlate_ctx *ctx)
1624 {
1625 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1626 &ctx->xout->odp_actions,
1627 &ctx->xin->flow, ODPP_NONE);
1628 ctx->sflow_odp_port = 0;
1629 ctx->sflow_n_outputs = 0;
1630 }
1631
1632 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1633 * of actions, eventually after the SAMPLE action for sFlow. */
1634 static void
1635 add_ipfix_action(struct xlate_ctx *ctx)
1636 {
1637 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1638 &ctx->xin->flow);
1639 }
1640
1641 /* Fix SAMPLE action according to data collected while composing ODP actions.
1642 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1643 * USERSPACE action's user-cookie which is required for sflow. */
1644 static void
1645 fix_sflow_action(struct xlate_ctx *ctx)
1646 {
1647 const struct flow *base = &ctx->base_flow;
1648 union user_action_cookie *cookie;
1649
1650 if (!ctx->user_cookie_offset) {
1651 return;
1652 }
1653
1654 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1655 sizeof cookie->sflow);
1656 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1657
1658 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1659 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1660 }
1661
1662 static enum slow_path_reason
1663 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1664 const struct xport *xport, const struct ofpbuf *packet)
1665 {
1666 struct flow_wildcards *wc = &ctx->xout->wc;
1667 const struct xbridge *xbridge = ctx->xbridge;
1668
1669 if (!xport) {
1670 return 0;
1671 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1672 if (packet) {
1673 cfm_process_heartbeat(xport->cfm, packet);
1674 }
1675 return SLOW_CFM;
1676 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1677 if (packet) {
1678 bfd_process_packet(xport->bfd, flow, packet);
1679 /* If POLL received, immediately sends FINAL back. */
1680 if (bfd_should_send_packet(xport->bfd)) {
1681 if (xport->peer) {
1682 ofproto_dpif_monitor_port_send_soon(xport->ofport);
1683 } else {
1684 ofproto_dpif_monitor_port_send_soon_safe(xport->ofport);
1685 }
1686 }
1687 }
1688 return SLOW_BFD;
1689 } else if (xport->xbundle && xport->xbundle->lacp
1690 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1691 if (packet) {
1692 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1693 }
1694 return SLOW_LACP;
1695 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1696 if (packet) {
1697 stp_process_packet(xport, packet);
1698 }
1699 return SLOW_STP;
1700 } else {
1701 return 0;
1702 }
1703 }
1704
1705 static void
1706 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1707 bool check_stp)
1708 {
1709 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1710 struct flow_wildcards *wc = &ctx->xout->wc;
1711 struct flow *flow = &ctx->xin->flow;
1712 ovs_be16 flow_vlan_tci;
1713 uint32_t flow_pkt_mark;
1714 uint8_t flow_nw_tos;
1715 odp_port_t out_port, odp_port;
1716 uint8_t dscp;
1717
1718 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1719 * before traversing a patch port. */
1720 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 25);
1721
1722 if (!xport) {
1723 xlate_report(ctx, "Nonexistent output port");
1724 return;
1725 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1726 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1727 return;
1728 } else if (check_stp) {
1729 if (eth_addr_equals(ctx->base_flow.dl_dst, eth_addr_stp)) {
1730 if (!xport_stp_listen_state(xport)) {
1731 xlate_report(ctx, "STP not in listening state, "
1732 "skipping bpdu output");
1733 return;
1734 }
1735 } else if (!xport_stp_forward_state(xport)) {
1736 xlate_report(ctx, "STP not in forwarding state, "
1737 "skipping output");
1738 return;
1739 }
1740 }
1741
1742 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1743 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1744 xport->xbundle);
1745 }
1746
1747 if (xport->peer) {
1748 const struct xport *peer = xport->peer;
1749 struct flow old_flow = ctx->xin->flow;
1750 enum slow_path_reason special;
1751
1752 ctx->xbridge = peer->xbridge;
1753 flow->in_port.ofp_port = peer->ofp_port;
1754 flow->metadata = htonll(0);
1755 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1756 memset(flow->regs, 0, sizeof flow->regs);
1757
1758 special = process_special(ctx, &ctx->xin->flow, peer,
1759 ctx->xin->packet);
1760 if (special) {
1761 ctx->xout->slow |= special;
1762 } else if (may_receive(peer, ctx)) {
1763 if (xport_stp_forward_state(peer)) {
1764 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1765 } else {
1766 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1767 * learning action look at the packet, then drop it. */
1768 struct flow old_base_flow = ctx->base_flow;
1769 size_t old_size = ofpbuf_size(&ctx->xout->odp_actions);
1770 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1771 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
1772 ctx->xout->mirrors = old_mirrors;
1773 ctx->base_flow = old_base_flow;
1774 ofpbuf_set_size(&ctx->xout->odp_actions, old_size);
1775 }
1776 }
1777
1778 ctx->xin->flow = old_flow;
1779 ctx->xbridge = xport->xbridge;
1780
1781 if (ctx->xin->resubmit_stats) {
1782 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1783 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1784 if (peer->bfd) {
1785 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
1786 }
1787 }
1788
1789 return;
1790 }
1791
1792 flow_vlan_tci = flow->vlan_tci;
1793 flow_pkt_mark = flow->pkt_mark;
1794 flow_nw_tos = flow->nw_tos;
1795
1796 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1797 wc->masks.nw_tos |= IP_DSCP_MASK;
1798 flow->nw_tos &= ~IP_DSCP_MASK;
1799 flow->nw_tos |= dscp;
1800 }
1801
1802 if (xport->is_tunnel) {
1803 /* Save tunnel metadata so that changes made due to
1804 * the Logical (tunnel) Port are not visible for any further
1805 * matches, while explicit set actions on tunnel metadata are.
1806 */
1807 struct flow_tnl flow_tnl = flow->tunnel;
1808 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1809 if (odp_port == ODPP_NONE) {
1810 xlate_report(ctx, "Tunneling decided against output");
1811 goto out; /* restore flow_nw_tos */
1812 }
1813 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1814 xlate_report(ctx, "Not tunneling to our own address");
1815 goto out; /* restore flow_nw_tos */
1816 }
1817 if (ctx->xin->resubmit_stats) {
1818 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1819 }
1820 out_port = odp_port;
1821 commit_odp_tunnel_action(flow, &ctx->base_flow,
1822 &ctx->xout->odp_actions);
1823 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1824 } else {
1825 odp_port = xport->odp_port;
1826 out_port = odp_port;
1827 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1828 ofp_port_t vlandev_port;
1829
1830 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1831 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
1832 ofp_port, flow->vlan_tci);
1833 if (vlandev_port != ofp_port) {
1834 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1835 flow->vlan_tci = htons(0);
1836 }
1837 }
1838 }
1839
1840 if (out_port != ODPP_NONE) {
1841 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
1842 &ctx->xout->odp_actions,
1843 &ctx->xout->wc);
1844
1845 if (ctx->xout->use_recirc) {
1846 struct ovs_action_recirc *act_recirc;
1847 struct xlate_recirc *xr = &ctx->xout->recirc;
1848
1849 act_recirc = nl_msg_put_unspec_uninit(&ctx->xout->odp_actions,
1850 OVS_ACTION_ATTR_RECIRC, sizeof *act_recirc);
1851 act_recirc->recirc_id = xr->recirc_id;
1852 act_recirc->hash_alg = xr->hash_alg;
1853 act_recirc->hash_bias = xr->hash_bias;
1854 } else {
1855 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1856 out_port);
1857 }
1858
1859 ctx->sflow_odp_port = odp_port;
1860 ctx->sflow_n_outputs++;
1861 ctx->xout->nf_output_iface = ofp_port;
1862 }
1863
1864 out:
1865 /* Restore flow */
1866 flow->vlan_tci = flow_vlan_tci;
1867 flow->pkt_mark = flow_pkt_mark;
1868 flow->nw_tos = flow_nw_tos;
1869 }
1870
1871 static void
1872 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1873 {
1874 compose_output_action__(ctx, ofp_port, true);
1875 }
1876
1877 static void
1878 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
1879 {
1880 struct rule_dpif *old_rule = ctx->rule;
1881 struct rule_actions *actions;
1882
1883 if (ctx->xin->resubmit_stats) {
1884 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
1885 }
1886
1887 ctx->resubmits++;
1888 ctx->recurse++;
1889 ctx->rule = rule;
1890 actions = rule_dpif_get_actions(rule);
1891 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
1892 ctx->rule = old_rule;
1893 ctx->recurse--;
1894 }
1895
1896 static bool
1897 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
1898 {
1899 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1900
1901 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
1902 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
1903 MAX_RESUBMIT_RECURSION);
1904 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
1905 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
1906 } else if (ofpbuf_size(&ctx->xout->odp_actions) > UINT16_MAX) {
1907 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
1908 } else if (ofpbuf_size(&ctx->stack) >= 65536) {
1909 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
1910 } else {
1911 return true;
1912 }
1913
1914 return false;
1915 }
1916
1917 static void
1918 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
1919 bool may_packet_in, bool honor_table_miss)
1920 {
1921 if (xlate_resubmit_resource_check(ctx)) {
1922 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1923 bool skip_wildcards = ctx->xin->skip_wildcards;
1924 uint8_t old_table_id = ctx->table_id;
1925 struct rule_dpif *rule;
1926 enum rule_dpif_lookup_verdict verdict;
1927 enum ofputil_port_config config = 0;
1928
1929 ctx->table_id = table_id;
1930
1931 /* Look up a flow with 'in_port' as the input port. Then restore the
1932 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
1933 * have surprising behavior). */
1934 ctx->xin->flow.in_port.ofp_port = in_port;
1935 verdict = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
1936 &ctx->xin->flow,
1937 !skip_wildcards
1938 ? &ctx->xout->wc : NULL,
1939 honor_table_miss,
1940 &ctx->table_id, &rule);
1941 ctx->xin->flow.in_port.ofp_port = old_in_port;
1942
1943 if (ctx->xin->resubmit_hook) {
1944 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1945 }
1946
1947 switch (verdict) {
1948 case RULE_DPIF_LOOKUP_VERDICT_MATCH:
1949 goto match;
1950 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER:
1951 if (may_packet_in) {
1952 struct xport *xport;
1953
1954 xport = get_ofp_port(ctx->xbridge,
1955 ctx->xin->flow.in_port.ofp_port);
1956 config = xport ? xport->config : 0;
1957 break;
1958 }
1959 /* Fall through to drop */
1960 case RULE_DPIF_LOOKUP_VERDICT_DROP:
1961 config = OFPUTIL_PC_NO_PACKET_IN;
1962 break;
1963 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT:
1964 if (!ofproto_dpif_wants_packet_in_on_miss(ctx->xbridge->ofproto)) {
1965 config = OFPUTIL_PC_NO_PACKET_IN;
1966 }
1967 break;
1968 default:
1969 OVS_NOT_REACHED();
1970 }
1971
1972 choose_miss_rule(config, ctx->xbridge->miss_rule,
1973 ctx->xbridge->no_packet_in_rule, &rule);
1974
1975 match:
1976 if (rule) {
1977 xlate_recursively(ctx, rule);
1978 rule_dpif_unref(rule);
1979 }
1980
1981 ctx->table_id = old_table_id;
1982 return;
1983 }
1984
1985 ctx->exit = true;
1986 }
1987
1988 static void
1989 xlate_group_bucket(struct xlate_ctx *ctx, const struct ofputil_bucket *bucket)
1990 {
1991 uint64_t action_list_stub[1024 / 8];
1992 struct ofpbuf action_list, action_set;
1993
1994 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
1995 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
1996
1997 ofpacts_execute_action_set(&action_list, &action_set);
1998 ctx->recurse++;
1999 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2000 ctx->recurse--;
2001
2002 ofpbuf_uninit(&action_set);
2003 ofpbuf_uninit(&action_list);
2004 }
2005
2006 static void
2007 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
2008 {
2009 const struct ofputil_bucket *bucket;
2010 const struct list *buckets;
2011 struct flow old_flow = ctx->xin->flow;
2012
2013 group_dpif_get_buckets(group, &buckets);
2014
2015 LIST_FOR_EACH (bucket, list_node, buckets) {
2016 xlate_group_bucket(ctx, bucket);
2017 /* Roll back flow to previous state.
2018 * This is equivalent to cloning the packet for each bucket.
2019 *
2020 * As a side effect any subsequently applied actions will
2021 * also effectively be applied to a clone of the packet taken
2022 * just before applying the all or indirect group. */
2023 ctx->xin->flow = old_flow;
2024 }
2025 }
2026
2027 static void
2028 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
2029 {
2030 const struct ofputil_bucket *bucket;
2031
2032 bucket = group_first_live_bucket(ctx, group, 0);
2033 if (bucket) {
2034 xlate_group_bucket(ctx, bucket);
2035 }
2036 }
2037
2038 static void
2039 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
2040 {
2041 struct flow_wildcards *wc = &ctx->xout->wc;
2042 const struct ofputil_bucket *bucket;
2043 uint32_t basis;
2044
2045 basis = hash_mac(ctx->xin->flow.dl_dst, 0, 0);
2046 bucket = group_best_live_bucket(ctx, group, basis);
2047 if (bucket) {
2048 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2049 xlate_group_bucket(ctx, bucket);
2050 }
2051 }
2052
2053 static void
2054 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
2055 {
2056 ctx->in_group = true;
2057
2058 switch (group_dpif_get_type(group)) {
2059 case OFPGT11_ALL:
2060 case OFPGT11_INDIRECT:
2061 xlate_all_group(ctx, group);
2062 break;
2063 case OFPGT11_SELECT:
2064 xlate_select_group(ctx, group);
2065 break;
2066 case OFPGT11_FF:
2067 xlate_ff_group(ctx, group);
2068 break;
2069 default:
2070 OVS_NOT_REACHED();
2071 }
2072 group_dpif_release(group);
2073
2074 ctx->in_group = false;
2075 }
2076
2077 static bool
2078 xlate_group_resource_check(struct xlate_ctx *ctx)
2079 {
2080 if (!xlate_resubmit_resource_check(ctx)) {
2081 return false;
2082 } else if (ctx->in_group) {
2083 /* Prevent nested translation of OpenFlow groups.
2084 *
2085 * OpenFlow allows this restriction. We enforce this restriction only
2086 * because, with the current architecture, we would otherwise have to
2087 * take a possibly recursive read lock on the ofgroup rwlock, which is
2088 * unsafe given that POSIX allows taking a read lock to block if there
2089 * is a thread blocked on taking the write lock. Other solutions
2090 * without this restriction are also possible, but seem unwarranted
2091 * given the current limited use of groups. */
2092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2093
2094 VLOG_ERR_RL(&rl, "cannot recursively translate OpenFlow group");
2095 return false;
2096 } else {
2097 return true;
2098 }
2099 }
2100
2101 static bool
2102 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
2103 {
2104 if (xlate_group_resource_check(ctx)) {
2105 struct group_dpif *group;
2106 bool got_group;
2107
2108 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
2109 if (got_group) {
2110 xlate_group_action__(ctx, group);
2111 } else {
2112 return true;
2113 }
2114 }
2115
2116 return false;
2117 }
2118
2119 static void
2120 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
2121 const struct ofpact_resubmit *resubmit)
2122 {
2123 ofp_port_t in_port;
2124 uint8_t table_id;
2125 bool may_packet_in = false;
2126 bool honor_table_miss = false;
2127
2128 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
2129 /* Still allow missed packets to be sent to the controller
2130 * if resubmitting from an internal table. */
2131 may_packet_in = true;
2132 honor_table_miss = true;
2133 }
2134
2135 in_port = resubmit->in_port;
2136 if (in_port == OFPP_IN_PORT) {
2137 in_port = ctx->xin->flow.in_port.ofp_port;
2138 }
2139
2140 table_id = resubmit->table_id;
2141 if (table_id == 255) {
2142 table_id = ctx->table_id;
2143 }
2144
2145 xlate_table_action(ctx, in_port, table_id, may_packet_in,
2146 honor_table_miss);
2147 }
2148
2149 static void
2150 flood_packets(struct xlate_ctx *ctx, bool all)
2151 {
2152 const struct xport *xport;
2153
2154 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
2155 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
2156 continue;
2157 }
2158
2159 if (all) {
2160 compose_output_action__(ctx, xport->ofp_port, false);
2161 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
2162 compose_output_action(ctx, xport->ofp_port);
2163 }
2164 }
2165
2166 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2167 }
2168
2169 static void
2170 execute_controller_action(struct xlate_ctx *ctx, int len,
2171 enum ofp_packet_in_reason reason,
2172 uint16_t controller_id)
2173 {
2174 struct ofproto_packet_in *pin;
2175 struct ofpbuf *packet;
2176 struct pkt_metadata md = PKT_METADATA_INITIALIZER(0);
2177
2178 ctx->xout->slow |= SLOW_CONTROLLER;
2179 if (!ctx->xin->packet) {
2180 return;
2181 }
2182
2183 packet = ofpbuf_clone(ctx->xin->packet);
2184
2185 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2186 &ctx->xout->odp_actions,
2187 &ctx->xout->wc);
2188
2189 odp_execute_actions(NULL, packet, false, &md,
2190 ofpbuf_data(&ctx->xout->odp_actions),
2191 ofpbuf_size(&ctx->xout->odp_actions), NULL);
2192
2193 pin = xmalloc(sizeof *pin);
2194 pin->up.packet_len = ofpbuf_size(packet);
2195 pin->up.packet = ofpbuf_steal_data(packet);
2196 pin->up.reason = reason;
2197 pin->up.table_id = ctx->table_id;
2198 pin->up.cookie = (ctx->rule
2199 ? rule_dpif_get_flow_cookie(ctx->rule)
2200 : OVS_BE64_MAX);
2201
2202 flow_get_metadata(&ctx->xin->flow, &pin->up.fmd);
2203
2204 pin->controller_id = controller_id;
2205 pin->send_len = len;
2206 /* If a rule is a table-miss rule then this is
2207 * a table-miss handled by a table-miss rule.
2208 *
2209 * Else, if rule is internal and has a controller action,
2210 * the later being implied by the rule being processed here,
2211 * then this is a table-miss handled without a table-miss rule.
2212 *
2213 * Otherwise this is not a table-miss. */
2214 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
2215 if (ctx->rule) {
2216 if (rule_dpif_is_table_miss(ctx->rule)) {
2217 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
2218 } else if (rule_dpif_is_internal(ctx->rule)) {
2219 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
2220 }
2221 }
2222 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
2223 ofpbuf_delete(packet);
2224 }
2225
2226 static void
2227 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
2228 {
2229 struct flow_wildcards *wc = &ctx->xout->wc;
2230 struct flow *flow = &ctx->xin->flow;
2231 int n;
2232
2233 ovs_assert(eth_type_mpls(mpls->ethertype));
2234
2235 n = flow_count_mpls_labels(flow, wc);
2236 if (!n) {
2237 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2238 &ctx->xout->odp_actions,
2239 &ctx->xout->wc);
2240 } else if (n >= FLOW_MAX_MPLS_LABELS) {
2241 if (ctx->xin->packet != NULL) {
2242 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2243 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2244 "MPLS push action can't be performed as it would "
2245 "have more MPLS LSEs than the %d supported.",
2246 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2247 }
2248 ctx->exit = true;
2249 return;
2250 } else if (n >= ctx->xbridge->max_mpls_depth) {
2251 COVERAGE_INC(xlate_actions_mpls_overflow);
2252 ctx->xout->slow |= SLOW_ACTION;
2253 }
2254
2255 flow_push_mpls(flow, n, mpls->ethertype, wc);
2256 }
2257
2258 static void
2259 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
2260 {
2261 struct flow_wildcards *wc = &ctx->xout->wc;
2262 struct flow *flow = &ctx->xin->flow;
2263 int n = flow_count_mpls_labels(flow, wc);
2264
2265 if (!flow_pop_mpls(flow, n, eth_type, wc) && n >= FLOW_MAX_MPLS_LABELS) {
2266 if (ctx->xin->packet != NULL) {
2267 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2268 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
2269 "MPLS pop action can't be performed as it has "
2270 "more MPLS LSEs than the %d supported.",
2271 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
2272 }
2273 ctx->exit = true;
2274 ofpbuf_clear(&ctx->xout->odp_actions);
2275 }
2276 }
2277
2278 static bool
2279 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
2280 {
2281 struct flow *flow = &ctx->xin->flow;
2282
2283 if (!is_ip_any(flow)) {
2284 return false;
2285 }
2286
2287 ctx->xout->wc.masks.nw_ttl = 0xff;
2288 if (flow->nw_ttl > 1) {
2289 flow->nw_ttl--;
2290 return false;
2291 } else {
2292 size_t i;
2293
2294 for (i = 0; i < ids->n_controllers; i++) {
2295 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
2296 ids->cnt_ids[i]);
2297 }
2298
2299 /* Stop processing for current table. */
2300 return true;
2301 }
2302 }
2303
2304 static void
2305 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
2306 {
2307 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2308 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
2309 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
2310 }
2311 }
2312
2313 static void
2314 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
2315 {
2316 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2317 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
2318 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
2319 }
2320 }
2321
2322 static void
2323 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
2324 {
2325 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
2326 ctx->xout->wc.masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
2327 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
2328 }
2329 }
2330
2331 static bool
2332 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
2333 {
2334 struct flow *flow = &ctx->xin->flow;
2335 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
2336 struct flow_wildcards *wc = &ctx->xout->wc;
2337
2338 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
2339 if (eth_type_mpls(flow->dl_type)) {
2340 if (ttl > 1) {
2341 ttl--;
2342 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
2343 return false;
2344 } else {
2345 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
2346
2347 /* Stop processing for current table. */
2348 return true;
2349 }
2350 } else {
2351 return true;
2352 }
2353 }
2354
2355 static void
2356 xlate_output_action(struct xlate_ctx *ctx,
2357 ofp_port_t port, uint16_t max_len, bool may_packet_in)
2358 {
2359 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
2360
2361 ctx->xout->nf_output_iface = NF_OUT_DROP;
2362
2363 switch (port) {
2364 case OFPP_IN_PORT:
2365 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
2366 break;
2367 case OFPP_TABLE:
2368 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2369 0, may_packet_in, true);
2370 break;
2371 case OFPP_NORMAL:
2372 xlate_normal(ctx);
2373 break;
2374 case OFPP_FLOOD:
2375 flood_packets(ctx, false);
2376 break;
2377 case OFPP_ALL:
2378 flood_packets(ctx, true);
2379 break;
2380 case OFPP_CONTROLLER:
2381 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
2382 break;
2383 case OFPP_NONE:
2384 break;
2385 case OFPP_LOCAL:
2386 default:
2387 if (port != ctx->xin->flow.in_port.ofp_port) {
2388 compose_output_action(ctx, port);
2389 } else {
2390 xlate_report(ctx, "skipping output to input port");
2391 }
2392 break;
2393 }
2394
2395 if (prev_nf_output_iface == NF_OUT_FLOOD) {
2396 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2397 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2398 ctx->xout->nf_output_iface = prev_nf_output_iface;
2399 } else if (prev_nf_output_iface != NF_OUT_DROP &&
2400 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2401 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2402 }
2403 }
2404
2405 static void
2406 xlate_output_reg_action(struct xlate_ctx *ctx,
2407 const struct ofpact_output_reg *or)
2408 {
2409 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
2410 if (port <= UINT16_MAX) {
2411 union mf_subvalue value;
2412
2413 memset(&value, 0xff, sizeof value);
2414 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
2415 xlate_output_action(ctx, u16_to_ofp(port),
2416 or->max_len, false);
2417 }
2418 }
2419
2420 static void
2421 xlate_enqueue_action(struct xlate_ctx *ctx,
2422 const struct ofpact_enqueue *enqueue)
2423 {
2424 ofp_port_t ofp_port = enqueue->port;
2425 uint32_t queue_id = enqueue->queue;
2426 uint32_t flow_priority, priority;
2427 int error;
2428
2429 /* Translate queue to priority. */
2430 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
2431 if (error) {
2432 /* Fall back to ordinary output action. */
2433 xlate_output_action(ctx, enqueue->port, 0, false);
2434 return;
2435 }
2436
2437 /* Check output port. */
2438 if (ofp_port == OFPP_IN_PORT) {
2439 ofp_port = ctx->xin->flow.in_port.ofp_port;
2440 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
2441 return;
2442 }
2443
2444 /* Add datapath actions. */
2445 flow_priority = ctx->xin->flow.skb_priority;
2446 ctx->xin->flow.skb_priority = priority;
2447 compose_output_action(ctx, ofp_port);
2448 ctx->xin->flow.skb_priority = flow_priority;
2449
2450 /* Update NetFlow output port. */
2451 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
2452 ctx->xout->nf_output_iface = ofp_port;
2453 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
2454 ctx->xout->nf_output_iface = NF_OUT_MULTI;
2455 }
2456 }
2457
2458 static void
2459 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
2460 {
2461 uint32_t skb_priority;
2462
2463 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
2464 ctx->xin->flow.skb_priority = skb_priority;
2465 } else {
2466 /* Couldn't translate queue to a priority. Nothing to do. A warning
2467 * has already been logged. */
2468 }
2469 }
2470
2471 static bool
2472 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
2473 {
2474 const struct xbridge *xbridge = xbridge_;
2475 struct xport *port;
2476
2477 switch (ofp_port) {
2478 case OFPP_IN_PORT:
2479 case OFPP_TABLE:
2480 case OFPP_NORMAL:
2481 case OFPP_FLOOD:
2482 case OFPP_ALL:
2483 case OFPP_NONE:
2484 return true;
2485 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
2486 return false;
2487 default:
2488 port = get_ofp_port(xbridge, ofp_port);
2489 return port ? port->may_enable : false;
2490 }
2491 }
2492
2493 static void
2494 xlate_bundle_action(struct xlate_ctx *ctx,
2495 const struct ofpact_bundle *bundle)
2496 {
2497 ofp_port_t port;
2498
2499 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
2500 slave_enabled_cb,
2501 CONST_CAST(struct xbridge *, ctx->xbridge));
2502 if (bundle->dst.field) {
2503 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow,
2504 &ctx->xout->wc);
2505 } else {
2506 xlate_output_action(ctx, port, 0, false);
2507 }
2508 }
2509
2510 static void
2511 xlate_learn_action(struct xlate_ctx *ctx,
2512 const struct ofpact_learn *learn)
2513 {
2514 uint64_t ofpacts_stub[1024 / 8];
2515 struct ofputil_flow_mod fm;
2516 struct ofpbuf ofpacts;
2517
2518 ctx->xout->has_learn = true;
2519
2520 learn_mask(learn, &ctx->xout->wc);
2521
2522 if (!ctx->xin->may_learn) {
2523 return;
2524 }
2525
2526 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2527 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
2528 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
2529 ofpbuf_uninit(&ofpacts);
2530 }
2531
2532 static void
2533 xlate_fin_timeout(struct xlate_ctx *ctx,
2534 const struct ofpact_fin_timeout *oft)
2535 {
2536 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
2537 rule_dpif_reduce_timeouts(ctx->rule, oft->fin_idle_timeout,
2538 oft->fin_hard_timeout);
2539 }
2540 }
2541
2542 static void
2543 xlate_sample_action(struct xlate_ctx *ctx,
2544 const struct ofpact_sample *os)
2545 {
2546 union user_action_cookie cookie;
2547 /* Scale the probability from 16-bit to 32-bit while representing
2548 * the same percentage. */
2549 uint32_t probability = (os->probability << 16) | os->probability;
2550
2551 if (!ctx->xbridge->variable_length_userdata) {
2552 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
2553
2554 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
2555 "lacks support (needs Linux 3.10+ or kernel module from "
2556 "OVS 1.11+)");
2557 return;
2558 }
2559
2560 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2561 &ctx->xout->odp_actions,
2562 &ctx->xout->wc);
2563
2564 compose_flow_sample_cookie(os->probability, os->collector_set_id,
2565 os->obs_domain_id, os->obs_point_id, &cookie);
2566 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
2567 probability, &cookie, sizeof cookie.flow_sample);
2568 }
2569
2570 static bool
2571 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
2572 {
2573 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
2574 ? OFPUTIL_PC_NO_RECV_STP
2575 : OFPUTIL_PC_NO_RECV)) {
2576 return false;
2577 }
2578
2579 /* Only drop packets here if both forwarding and learning are
2580 * disabled. If just learning is enabled, we need to have
2581 * OFPP_NORMAL and the learning action have a look at the packet
2582 * before we can drop it. */
2583 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
2584 return false;
2585 }
2586
2587 return true;
2588 }
2589
2590 static void
2591 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
2592 {
2593 struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
2594 ofpbuf_put(&ctx->action_set, on->actions, ofpact_nest_get_action_len(on));
2595 ofpact_pad(&ctx->action_set);
2596 }
2597
2598 static void
2599 xlate_action_set(struct xlate_ctx *ctx)
2600 {
2601 uint64_t action_list_stub[1024 / 64];
2602 struct ofpbuf action_list;
2603
2604 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
2605 ofpacts_execute_action_set(&action_list, &ctx->action_set);
2606 do_xlate_actions(ofpbuf_data(&action_list), ofpbuf_size(&action_list), ctx);
2607 ofpbuf_uninit(&action_list);
2608 }
2609
2610 static void
2611 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2612 struct xlate_ctx *ctx)
2613 {
2614 struct flow_wildcards *wc = &ctx->xout->wc;
2615 struct flow *flow = &ctx->xin->flow;
2616 const struct ofpact *a;
2617
2618 /* dl_type already in the mask, not set below. */
2619
2620 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2621 struct ofpact_controller *controller;
2622 const struct ofpact_metadata *metadata;
2623 const struct ofpact_set_field *set_field;
2624 const struct mf_field *mf;
2625
2626 if (ctx->exit) {
2627 break;
2628 }
2629
2630 switch (a->type) {
2631 case OFPACT_OUTPUT:
2632 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2633 ofpact_get_OUTPUT(a)->max_len, true);
2634 break;
2635
2636 case OFPACT_GROUP:
2637 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
2638 return;
2639 }
2640 break;
2641
2642 case OFPACT_CONTROLLER:
2643 controller = ofpact_get_CONTROLLER(a);
2644 execute_controller_action(ctx, controller->max_len,
2645 controller->reason,
2646 controller->controller_id);
2647 break;
2648
2649 case OFPACT_ENQUEUE:
2650 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2651 break;
2652
2653 case OFPACT_SET_VLAN_VID:
2654 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2655 if (flow->vlan_tci & htons(VLAN_CFI) ||
2656 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
2657 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2658 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2659 | htons(VLAN_CFI));
2660 }
2661 break;
2662
2663 case OFPACT_SET_VLAN_PCP:
2664 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
2665 if (flow->vlan_tci & htons(VLAN_CFI) ||
2666 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
2667 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2668 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
2669 << VLAN_PCP_SHIFT) | VLAN_CFI);
2670 }
2671 break;
2672
2673 case OFPACT_STRIP_VLAN:
2674 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2675 flow->vlan_tci = htons(0);
2676 break;
2677
2678 case OFPACT_PUSH_VLAN:
2679 /* XXX 802.1AD(QinQ) */
2680 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
2681 flow->vlan_tci = htons(VLAN_CFI);
2682 break;
2683
2684 case OFPACT_SET_ETH_SRC:
2685 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2686 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2687 break;
2688
2689 case OFPACT_SET_ETH_DST:
2690 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2691 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2692 break;
2693
2694 case OFPACT_SET_IPV4_SRC:
2695 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2696 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
2697 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2698 }
2699 break;
2700
2701 case OFPACT_SET_IPV4_DST:
2702 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2703 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
2704 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2705 }
2706 break;
2707
2708 case OFPACT_SET_IP_DSCP:
2709 if (is_ip_any(flow)) {
2710 wc->masks.nw_tos |= IP_DSCP_MASK;
2711 flow->nw_tos &= ~IP_DSCP_MASK;
2712 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
2713 }
2714 break;
2715
2716 case OFPACT_SET_IP_ECN:
2717 if (is_ip_any(flow)) {
2718 wc->masks.nw_tos |= IP_ECN_MASK;
2719 flow->nw_tos &= ~IP_ECN_MASK;
2720 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
2721 }
2722 break;
2723
2724 case OFPACT_SET_IP_TTL:
2725 if (is_ip_any(flow)) {
2726 wc->masks.nw_ttl = 0xff;
2727 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
2728 }
2729 break;
2730
2731 case OFPACT_SET_L4_SRC_PORT:
2732 if (is_ip_any(flow)) {
2733 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2734 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
2735 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2736 }
2737 break;
2738
2739 case OFPACT_SET_L4_DST_PORT:
2740 if (is_ip_any(flow)) {
2741 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2742 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
2743 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2744 }
2745 break;
2746
2747 case OFPACT_RESUBMIT:
2748 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2749 break;
2750
2751 case OFPACT_SET_TUNNEL:
2752 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2753 break;
2754
2755 case OFPACT_SET_QUEUE:
2756 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2757 break;
2758
2759 case OFPACT_POP_QUEUE:
2760 flow->skb_priority = ctx->orig_skb_priority;
2761 break;
2762
2763 case OFPACT_REG_MOVE:
2764 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2765 break;
2766
2767 case OFPACT_REG_LOAD:
2768 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow, wc);
2769 break;
2770
2771 case OFPACT_SET_FIELD:
2772 set_field = ofpact_get_SET_FIELD(a);
2773 mf = set_field->field;
2774 mf_mask_field_and_prereqs(mf, &wc->masks);
2775
2776 /* Set field action only ever overwrites packet's outermost
2777 * applicable header fields. Do nothing if no header exists. */
2778 if ((mf->id != MFF_VLAN_VID || flow->vlan_tci & htons(VLAN_CFI))
2779 && ((mf->id != MFF_MPLS_LABEL && mf->id != MFF_MPLS_TC)
2780 || eth_type_mpls(flow->dl_type))) {
2781 mf_set_flow_value(mf, &set_field->value, flow);
2782 }
2783 break;
2784
2785 case OFPACT_STACK_PUSH:
2786 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2787 &ctx->stack);
2788 break;
2789
2790 case OFPACT_STACK_POP:
2791 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
2792 &ctx->stack);
2793 break;
2794
2795 case OFPACT_PUSH_MPLS:
2796 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
2797 break;
2798
2799 case OFPACT_POP_MPLS:
2800 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2801 break;
2802
2803 case OFPACT_SET_MPLS_LABEL:
2804 compose_set_mpls_label_action(
2805 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
2806 break;
2807
2808 case OFPACT_SET_MPLS_TC:
2809 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
2810 break;
2811
2812 case OFPACT_SET_MPLS_TTL:
2813 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
2814 break;
2815
2816 case OFPACT_DEC_MPLS_TTL:
2817 if (compose_dec_mpls_ttl_action(ctx)) {
2818 return;
2819 }
2820 break;
2821
2822 case OFPACT_DEC_TTL:
2823 wc->masks.nw_ttl = 0xff;
2824 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2825 return;
2826 }
2827 break;
2828
2829 case OFPACT_NOTE:
2830 /* Nothing to do. */
2831 break;
2832
2833 case OFPACT_MULTIPATH:
2834 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2835 break;
2836
2837 case OFPACT_BUNDLE:
2838 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2839 break;
2840
2841 case OFPACT_OUTPUT_REG:
2842 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2843 break;
2844
2845 case OFPACT_LEARN:
2846 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2847 break;
2848
2849 case OFPACT_EXIT:
2850 ctx->exit = true;
2851 break;
2852
2853 case OFPACT_FIN_TIMEOUT:
2854 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2855 ctx->xout->has_fin_timeout = true;
2856 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2857 break;
2858
2859 case OFPACT_CLEAR_ACTIONS:
2860 ofpbuf_clear(&ctx->action_set);
2861 break;
2862
2863 case OFPACT_WRITE_ACTIONS:
2864 xlate_write_actions(ctx, a);
2865 break;
2866
2867 case OFPACT_WRITE_METADATA:
2868 metadata = ofpact_get_WRITE_METADATA(a);
2869 flow->metadata &= ~metadata->mask;
2870 flow->metadata |= metadata->metadata & metadata->mask;
2871 break;
2872
2873 case OFPACT_METER:
2874 /* Not implemented yet. */
2875 break;
2876
2877 case OFPACT_GOTO_TABLE: {
2878 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2879
2880 ovs_assert(ctx->table_id < ogt->table_id);
2881 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2882 ogt->table_id, true, true);
2883 break;
2884 }
2885
2886 case OFPACT_SAMPLE:
2887 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2888 break;
2889 }
2890 }
2891 }
2892
2893 void
2894 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2895 const struct flow *flow, struct rule_dpif *rule,
2896 uint16_t tcp_flags, const struct ofpbuf *packet)
2897 {
2898 xin->ofproto = ofproto;
2899 xin->flow = *flow;
2900 xin->packet = packet;
2901 xin->may_learn = packet != NULL;
2902 xin->rule = rule;
2903 xin->ofpacts = NULL;
2904 xin->ofpacts_len = 0;
2905 xin->tcp_flags = tcp_flags;
2906 xin->resubmit_hook = NULL;
2907 xin->report_hook = NULL;
2908 xin->resubmit_stats = NULL;
2909 xin->skip_wildcards = false;
2910 }
2911
2912 void
2913 xlate_out_uninit(struct xlate_out *xout)
2914 {
2915 if (xout) {
2916 ofpbuf_uninit(&xout->odp_actions);
2917 }
2918 }
2919
2920 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2921 * into datapath actions, using 'ctx', and discards the datapath actions. */
2922 void
2923 xlate_actions_for_side_effects(struct xlate_in *xin)
2924 {
2925 struct xlate_out xout;
2926
2927 xlate_actions(xin, &xout);
2928 xlate_out_uninit(&xout);
2929 }
2930
2931 static void
2932 xlate_report(struct xlate_ctx *ctx, const char *s)
2933 {
2934 if (ctx->xin->report_hook) {
2935 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2936 }
2937 }
2938
2939 void
2940 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2941 {
2942 dst->wc = src->wc;
2943 dst->slow = src->slow;
2944 dst->has_learn = src->has_learn;
2945 dst->has_normal = src->has_normal;
2946 dst->has_fin_timeout = src->has_fin_timeout;
2947 dst->nf_output_iface = src->nf_output_iface;
2948 dst->mirrors = src->mirrors;
2949
2950 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2951 sizeof dst->odp_actions_stub);
2952 ofpbuf_put(&dst->odp_actions, ofpbuf_data(&src->odp_actions),
2953 ofpbuf_size(&src->odp_actions));
2954 }
2955 \f
2956 static struct skb_priority_to_dscp *
2957 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2958 {
2959 struct skb_priority_to_dscp *pdscp;
2960 uint32_t hash;
2961
2962 hash = hash_int(skb_priority, 0);
2963 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2964 if (pdscp->skb_priority == skb_priority) {
2965 return pdscp;
2966 }
2967 }
2968 return NULL;
2969 }
2970
2971 static bool
2972 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2973 uint8_t *dscp)
2974 {
2975 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2976 *dscp = pdscp ? pdscp->dscp : 0;
2977 return pdscp != NULL;
2978 }
2979
2980 static void
2981 clear_skb_priorities(struct xport *xport)
2982 {
2983 struct skb_priority_to_dscp *pdscp, *next;
2984
2985 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2986 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2987 free(pdscp);
2988 }
2989 }
2990
2991 static bool
2992 actions_output_to_local_port(const struct xlate_ctx *ctx)
2993 {
2994 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2995 const struct nlattr *a;
2996 unsigned int left;
2997
2998 NL_ATTR_FOR_EACH_UNSAFE (a, left, ofpbuf_data(&ctx->xout->odp_actions),
2999 ofpbuf_size(&ctx->xout->odp_actions)) {
3000 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
3001 && nl_attr_get_odp_port(a) == local_odp_port) {
3002 return true;
3003 }
3004 }
3005 return false;
3006 }
3007
3008 /* Thread safe call to xlate_actions__(). */
3009 void
3010 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
3011 OVS_EXCLUDED(xlate_rwlock)
3012 {
3013 ovs_rwlock_rdlock(&xlate_rwlock);
3014 xlate_actions__(xin, xout);
3015 ovs_rwlock_unlock(&xlate_rwlock);
3016 }
3017
3018 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
3019 * into datapath actions in 'odp_actions', using 'ctx'.
3020 *
3021 * The caller must take responsibility for eventually freeing 'xout', with
3022 * xlate_out_uninit(). */
3023 static void
3024 xlate_actions__(struct xlate_in *xin, struct xlate_out *xout)
3025 OVS_REQ_RDLOCK(xlate_rwlock)
3026 {
3027 struct flow_wildcards *wc = &xout->wc;
3028 struct flow *flow = &xin->flow;
3029 struct rule_dpif *rule = NULL;
3030
3031 struct rule_actions *actions = NULL;
3032 enum slow_path_reason special;
3033 const struct ofpact *ofpacts;
3034 struct xport *in_port;
3035 struct flow orig_flow;
3036 struct xlate_ctx ctx;
3037 size_t ofpacts_len;
3038 bool tnl_may_send;
3039 bool is_icmp;
3040
3041 COVERAGE_INC(xlate_actions);
3042
3043 /* Flow initialization rules:
3044 * - 'base_flow' must match the kernel's view of the packet at the
3045 * time that action processing starts. 'flow' represents any
3046 * transformations we wish to make through actions.
3047 * - By default 'base_flow' and 'flow' are the same since the input
3048 * packet matches the output before any actions are applied.
3049 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
3050 * of the received packet as seen by the kernel. If we later output
3051 * to another device without any modifications this will cause us to
3052 * insert a new tag since the original one was stripped off by the
3053 * VLAN device.
3054 * - Tunnel metadata as received is retained in 'flow'. This allows
3055 * tunnel metadata matching also in later tables.
3056 * Since a kernel action for setting the tunnel metadata will only be
3057 * generated with actual tunnel output, changing the tunnel metadata
3058 * values in 'flow' (such as tun_id) will only have effect with a later
3059 * tunnel output action.
3060 * - Tunnel 'base_flow' is completely cleared since that is what the
3061 * kernel does. If we wish to maintain the original values an action
3062 * needs to be generated. */
3063
3064 ctx.xin = xin;
3065 ctx.xout = xout;
3066 ctx.xout->slow = 0;
3067 ctx.xout->has_learn = false;
3068 ctx.xout->has_normal = false;
3069 ctx.xout->has_fin_timeout = false;
3070 ctx.xout->nf_output_iface = NF_OUT_DROP;
3071 ctx.xout->mirrors = 0;
3072 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
3073 sizeof ctx.xout->odp_actions_stub);
3074 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
3075
3076 ctx.xbridge = xbridge_lookup(xin->ofproto);
3077 if (!ctx.xbridge) {
3078 goto out;
3079 }
3080
3081 ctx.rule = xin->rule;
3082
3083 ctx.base_flow = *flow;
3084 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
3085 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
3086
3087 flow_wildcards_init_catchall(wc);
3088 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
3089 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
3090 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
3091 if (is_ip_any(flow)) {
3092 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
3093 }
3094 is_icmp = is_icmpv4(flow) || is_icmpv6(flow);
3095
3096 tnl_may_send = tnl_xlate_init(&ctx.base_flow, flow, wc);
3097 if (ctx.xbridge->netflow) {
3098 netflow_mask_wc(flow, wc);
3099 }
3100
3101 ctx.recurse = 0;
3102 ctx.resubmits = 0;
3103 ctx.in_group = false;
3104 ctx.orig_skb_priority = flow->skb_priority;
3105 ctx.table_id = 0;
3106 ctx.exit = false;
3107
3108 if (!xin->ofpacts && !ctx.rule) {
3109 ctx.table_id = rule_dpif_lookup(ctx.xbridge->ofproto, flow,
3110 !xin->skip_wildcards ? wc : NULL,
3111 &rule);
3112 if (ctx.xin->resubmit_stats) {
3113 rule_dpif_credit_stats(rule, ctx.xin->resubmit_stats);
3114 }
3115 ctx.rule = rule;
3116 }
3117 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
3118 xout->use_recirc = false;
3119
3120 if (xin->ofpacts) {
3121 ofpacts = xin->ofpacts;
3122 ofpacts_len = xin->ofpacts_len;
3123 } else if (ctx.rule) {
3124 actions = rule_dpif_get_actions(ctx.rule);
3125 ofpacts = actions->ofpacts;
3126 ofpacts_len = actions->ofpacts_len;
3127 } else {
3128 OVS_NOT_REACHED();
3129 }
3130
3131 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
3132 ofpbuf_use_stub(&ctx.action_set,
3133 ctx.action_set_stub, sizeof ctx.action_set_stub);
3134
3135 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3136 /* Do this conditionally because the copy is expensive enough that it
3137 * shows up in profiles. */
3138 orig_flow = *flow;
3139 }
3140
3141 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
3142 switch (ctx.xbridge->frag) {
3143 case OFPC_FRAG_NORMAL:
3144 /* We must pretend that transport ports are unavailable. */
3145 flow->tp_src = ctx.base_flow.tp_src = htons(0);
3146 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
3147 break;
3148
3149 case OFPC_FRAG_DROP:
3150 goto out;
3151
3152 case OFPC_FRAG_REASM:
3153 OVS_NOT_REACHED();
3154
3155 case OFPC_FRAG_NX_MATCH:
3156 /* Nothing to do. */
3157 break;
3158
3159 case OFPC_INVALID_TTL_TO_CONTROLLER:
3160 OVS_NOT_REACHED();
3161 }
3162 }
3163
3164 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
3165 if (in_port && in_port->is_tunnel && ctx.xin->resubmit_stats) {
3166 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
3167 if (in_port->bfd) {
3168 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
3169 }
3170 }
3171
3172 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
3173 if (special) {
3174 ctx.xout->slow |= special;
3175 } else {
3176 size_t sample_actions_len;
3177
3178 if (flow->in_port.ofp_port
3179 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
3180 flow->in_port.ofp_port,
3181 flow->vlan_tci)) {
3182 ctx.base_flow.vlan_tci = 0;
3183 }
3184
3185 add_sflow_action(&ctx);
3186 add_ipfix_action(&ctx);
3187 sample_actions_len = ofpbuf_size(&ctx.xout->odp_actions);
3188
3189 if (tnl_may_send && (!in_port || may_receive(in_port, &ctx))) {
3190 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
3191
3192 /* We've let OFPP_NORMAL and the learning action look at the
3193 * packet, so drop it now if forwarding is disabled. */
3194 if (in_port && !xport_stp_forward_state(in_port)) {
3195 ofpbuf_set_size(&ctx.xout->odp_actions, sample_actions_len);
3196 }
3197 }
3198
3199 if (ofpbuf_size(&ctx.action_set)) {
3200 xlate_action_set(&ctx);
3201 }
3202
3203 if (ctx.xbridge->has_in_band
3204 && in_band_must_output_to_local_port(flow)
3205 && !actions_output_to_local_port(&ctx)) {
3206 compose_output_action(&ctx, OFPP_LOCAL);
3207 }
3208
3209 fix_sflow_action(&ctx);
3210
3211 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
3212 add_mirror_actions(&ctx, &orig_flow);
3213 }
3214 }
3215
3216 if (nl_attr_oversized(ofpbuf_size(&ctx.xout->odp_actions))) {
3217 /* These datapath actions are too big for a Netlink attribute, so we
3218 * can't hand them to the kernel directly. dpif_execute() can execute
3219 * them one by one with help, so just mark the result as SLOW_ACTION to
3220 * prevent the flow from being installed. */
3221 COVERAGE_INC(xlate_actions_oversize);
3222 ctx.xout->slow |= SLOW_ACTION;
3223 }
3224
3225 if (ctx.xin->resubmit_stats) {
3226 mirror_update_stats(ctx.xbridge->mbridge, xout->mirrors,
3227 ctx.xin->resubmit_stats->n_packets,
3228 ctx.xin->resubmit_stats->n_bytes);
3229
3230 if (ctx.xbridge->netflow) {
3231 const struct ofpact *ofpacts;
3232 size_t ofpacts_len;
3233
3234 ofpacts_len = actions->ofpacts_len;
3235 ofpacts = actions->ofpacts;
3236 if (ofpacts_len == 0
3237 || ofpacts->type != OFPACT_CONTROLLER
3238 || ofpact_next(ofpacts) < ofpact_end(ofpacts, ofpacts_len)) {
3239 /* Only update netflow if we don't have controller flow. We don't
3240 * report NetFlow expiration messages for such facets because they
3241 * are just part of the control logic for the network, not real
3242 * traffic. */
3243 netflow_flow_update(ctx.xbridge->netflow, flow,
3244 xout->nf_output_iface,
3245 ctx.xin->resubmit_stats);
3246 }
3247 }
3248 }
3249
3250 ofpbuf_uninit(&ctx.stack);
3251 ofpbuf_uninit(&ctx.action_set);
3252
3253 /* Clear the metadata and register wildcard masks, because we won't
3254 * use non-header fields as part of the cache. */
3255 flow_wildcards_clear_non_packet_fields(wc);
3256
3257 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow uses
3258 * the low 8 bits of the 16-bit tp_src and tp_dst members to represent
3259 * these fields. The datapath interface, on the other hand, represents
3260 * them with just 8 bits each. This means that if the high 8 bits of the
3261 * masks for these fields somehow become set, then they will get chopped
3262 * off by a round trip through the datapath, and revalidation will spot
3263 * that as an inconsistency and delete the flow. Avoid the problem here by
3264 * making sure that only the low 8 bits of either field can be unwildcarded
3265 * for ICMP.
3266 */
3267 if (is_icmp) {
3268 wc->masks.tp_src &= htons(UINT8_MAX);
3269 wc->masks.tp_dst &= htons(UINT8_MAX);
3270 }
3271
3272 out:
3273 rule_dpif_unref(rule);
3274 }
3275
3276 /* Sends 'packet' out 'ofport'.
3277 * May modify 'packet'.
3278 * Returns 0 if successful, otherwise a positive errno value. */
3279 int
3280 xlate_send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
3281 {
3282 struct xport *xport;
3283 struct ofpact_output output;
3284 struct flow flow;
3285
3286 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
3287 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
3288 flow_extract(packet, NULL, &flow);
3289 flow.in_port.ofp_port = OFPP_NONE;
3290
3291 ovs_rwlock_rdlock(&xlate_rwlock);
3292 xport = xport_lookup(ofport);
3293 if (!xport) {
3294 ovs_rwlock_unlock(&xlate_rwlock);
3295 return EINVAL;
3296 }
3297 output.port = xport->ofp_port;
3298 output.max_len = 0;
3299 ovs_rwlock_unlock(&xlate_rwlock);
3300
3301 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
3302 &output.ofpact, sizeof output,
3303 packet);
3304 }