]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto-dpif-xlate.c
Add support for connection tracking.
[mirror_ovs.git] / ofproto / ofproto-dpif-xlate.c
1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 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 #include <arpa/inet.h>
21 #include <net/if.h>
22 #include <sys/socket.h>
23 #include <netinet/in.h>
24
25 #include "tnl-arp-cache.h"
26 #include "bfd.h"
27 #include "bitmap.h"
28 #include "bond.h"
29 #include "bundle.h"
30 #include "byte-order.h"
31 #include "cfm.h"
32 #include "connmgr.h"
33 #include "coverage.h"
34 #include "dp-packet.h"
35 #include "dpif.h"
36 #include "dynamic-string.h"
37 #include "in-band.h"
38 #include "lacp.h"
39 #include "learn.h"
40 #include "list.h"
41 #include "ovs-lldp.h"
42 #include "mac-learning.h"
43 #include "mcast-snooping.h"
44 #include "meta-flow.h"
45 #include "multipath.h"
46 #include "netdev-vport.h"
47 #include "netlink.h"
48 #include "nx-match.h"
49 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofproto/ofproto-dpif-ipfix.h"
52 #include "ofproto/ofproto-dpif-mirror.h"
53 #include "ofproto/ofproto-dpif-monitor.h"
54 #include "ofproto/ofproto-dpif-sflow.h"
55 #include "ofproto/ofproto-dpif.h"
56 #include "ofproto/ofproto-provider.h"
57 #include "ovs-router.h"
58 #include "tnl-ports.h"
59 #include "tunnel.h"
60 #include "openvswitch/vlog.h"
61
62 COVERAGE_DEFINE(xlate_actions);
63 COVERAGE_DEFINE(xlate_actions_oversize);
64 COVERAGE_DEFINE(xlate_actions_too_many_output);
65
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
67
68 /* Maximum depth of flow table recursion (due to resubmit actions) in a
69 * flow translation. */
70 #define MAX_RESUBMIT_RECURSION 64
71 #define MAX_INTERNAL_RESUBMITS 1 /* Max resbmits allowed using rules in
72 internal table. */
73
74 /* Maximum number of resubmit actions in a flow translation, whether they are
75 * recursive or not. */
76 #define MAX_RESUBMITS (MAX_RESUBMIT_RECURSION * MAX_RESUBMIT_RECURSION)
77
78 struct xbridge {
79 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
80 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
81
82 struct ovs_list xbundles; /* Owned xbundles. */
83 struct hmap xports; /* Indexed by ofp_port. */
84
85 char *name; /* Name used in log messages. */
86 struct dpif *dpif; /* Datapath interface. */
87 struct mac_learning *ml; /* Mac learning handle. */
88 struct mcast_snooping *ms; /* Multicast Snooping handle. */
89 struct mbridge *mbridge; /* Mirroring. */
90 struct dpif_sflow *sflow; /* SFlow handle, or null. */
91 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
92 struct netflow *netflow; /* Netflow handle, or null. */
93 struct stp *stp; /* STP or null if disabled. */
94 struct rstp *rstp; /* RSTP or null if disabled. */
95
96 bool has_in_band; /* Bridge has in band control? */
97 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
98
99 /* Datapath feature support. */
100 struct dpif_backer_support support;
101 };
102
103 struct xbundle {
104 struct hmap_node hmap_node; /* In global 'xbundles' map. */
105 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
106
107 struct ovs_list list_node; /* In parent 'xbridges' list. */
108 struct xbridge *xbridge; /* Parent xbridge. */
109
110 struct ovs_list xports; /* Contains "struct xport"s. */
111
112 char *name; /* Name used in log messages. */
113 struct bond *bond; /* Nonnull iff more than one port. */
114 struct lacp *lacp; /* LACP handle or null. */
115
116 enum port_vlan_mode vlan_mode; /* VLAN mode. */
117 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
118 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
119 * NULL if all VLANs are trunked. */
120 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
121 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
122 };
123
124 struct xport {
125 struct hmap_node hmap_node; /* Node in global 'xports' map. */
126 struct ofport_dpif *ofport; /* Key in global 'xports map. */
127
128 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
129 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
130
131 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
132
133 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
134 struct xbundle *xbundle; /* Parent xbundle or null. */
135
136 struct netdev *netdev; /* 'ofport''s netdev. */
137
138 struct xbridge *xbridge; /* Parent bridge. */
139 struct xport *peer; /* Patch port peer or null. */
140
141 enum ofputil_port_config config; /* OpenFlow port configuration. */
142 enum ofputil_port_state state; /* OpenFlow port state. */
143 int stp_port_no; /* STP port number or -1 if not in use. */
144 struct rstp_port *rstp_port; /* RSTP port or null. */
145
146 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
147
148 bool may_enable; /* May be enabled in bonds. */
149 bool is_tunnel; /* Is a tunnel port. */
150
151 struct cfm *cfm; /* CFM handle or null. */
152 struct bfd *bfd; /* BFD handle or null. */
153 struct lldp *lldp; /* LLDP handle or null. */
154 };
155
156 struct xlate_ctx {
157 struct xlate_in *xin;
158 struct xlate_out *xout;
159
160 const struct xbridge *xbridge;
161
162 /* Flow tables version at the beginning of the translation. */
163 cls_version_t tables_version;
164
165 /* Flow at the last commit. */
166 struct flow base_flow;
167
168 /* Tunnel IP destination address as received. This is stored separately
169 * as the base_flow.tunnel is cleared on init to reflect the datapath
170 * behavior. Used to make sure not to send tunneled output to ourselves,
171 * which might lead to an infinite loop. This could happen easily
172 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
173 * actually set the tun_dst field. */
174 ovs_be32 orig_tunnel_ip_dst;
175
176 /* Stack for the push and pop actions. Each stack element is of type
177 * "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 /* Flow translation populates this with wildcards relevant in translation.
184 * When 'xin->wc' is nonnull, this is the same pointer. When 'xin->wc' is
185 * null, this is a pointer to uninitialized scratch memory. This allows
186 * code to blindly write to 'ctx->wc' without worrying about whether the
187 * caller really wants wildcards. */
188 struct flow_wildcards *wc;
189
190 /* Output buffer for datapath actions. When 'xin->odp_actions' is nonnull,
191 * this is the same pointer. When 'xin->odp_actions' is null, this points
192 * to a scratch ofpbuf. This allows code to add actions to
193 * 'ctx->odp_actions' without worrying about whether the caller really
194 * wants actions. */
195 struct ofpbuf *odp_actions;
196
197 /* Resubmit statistics, via xlate_table_action(). */
198 int recurse; /* Current resubmit nesting depth. */
199 int resubmits; /* Total number of resubmits. */
200 bool in_group; /* Currently translating ofgroup, if true. */
201 bool in_action_set; /* Currently translating action_set, if true. */
202
203 uint8_t table_id; /* OpenFlow table ID where flow was found. */
204 ovs_be64 rule_cookie; /* Cookie of the rule being translated. */
205 uint32_t orig_skb_priority; /* Priority when packet arrived. */
206 uint32_t sflow_n_outputs; /* Number of output ports. */
207 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
208 ofp_port_t nf_output_iface; /* Output interface index for NetFlow. */
209 bool exit; /* No further actions should be processed. */
210 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
211
212 /* These are used for non-bond recirculation. The recirculation IDs are
213 * stored in xout and must be associated with a datapath flow (ukey),
214 * otherwise they will be freed when the xout is uninitialized.
215 *
216 *
217 * Steps in Recirculation Translation
218 * ==================================
219 *
220 * At some point during translation, the code recognizes the need for
221 * recirculation. For example, recirculation is necessary when, after
222 * popping the last MPLS label, an action or a match tries to examine or
223 * modify a field that has been newly revealed following the MPLS label.
224 *
225 * The simplest part of the work to be done is to commit existing changes to
226 * the packet, which produces datapath actions corresponding to the changes,
227 * and after this, add an OVS_ACTION_ATTR_RECIRC datapath action.
228 *
229 * The main problem here is preserving state. When the datapath executes
230 * OVS_ACTION_ATTR_RECIRC, it will upcall to userspace to get a translation
231 * for the post-recirculation actions. At this point userspace has to
232 * resume the translation where it left off, which means that it has to
233 * execute the following:
234 *
235 * - The action that prompted recirculation, and any actions following
236 * it within the same flow.
237 *
238 * - If the action that prompted recirculation was invoked within a
239 * NXAST_RESUBMIT, then any actions following the resubmit. These
240 * "resubmit"s can be nested, so this has to go all the way up the
241 * control stack.
242 *
243 * - The OpenFlow 1.1+ action set.
244 *
245 * State that actions and flow table lookups can depend on, such as the
246 * following, must also be preserved:
247 *
248 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
249 *
250 * - Action set, stack
251 *
252 * - The table ID and cookie of the flow being translated at each level
253 * of the control stack (since OFPAT_CONTROLLER actions send these to
254 * the controller).
255 *
256 * Translation allows for the control of this state preservation via these
257 * members. When a need for recirculation is identified, the translation
258 * process:
259 *
260 * 1. Sets 'recirc_action_offset' to the current size of 'action_set'. The
261 * action set is part of what needs to be preserved, so this allows the
262 * action set and the additional state to share the 'action_set' buffer.
263 * Later steps can tell that setup for recirculation is in progress from
264 * the nonnegative value of 'recirc_action_offset'.
265 *
266 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
267 * translation process.
268 *
269 * 3. Adds an OFPACT_UNROLL_XLATE action to 'action_set'. This action
270 * holds the current table ID and cookie so that they can be restored
271 * during a post-recirculation upcall translation.
272 *
273 * 4. Adds the action that prompted recirculation and any actions following
274 * it within the same flow to 'action_set', so that they can be executed
275 * during a post-recirculation upcall translation.
276 *
277 * 5. Returns.
278 *
279 * 6. The action that prompted recirculation might be nested in a stack of
280 * nested "resubmit"s that have actions remaining. Each of these notices
281 * that we're exiting (from 'exit') and that recirculation setup is in
282 * progress (from 'recirc_action_offset') and responds by adding more
283 * OFPACT_UNROLL_XLATE actions to 'action_set', as necessary, and any
284 * actions that were yet unprocessed.
285 *
286 * The caller stores all the state produced by this process associated with
287 * the recirculation ID. For post-recirculation upcall translation, the
288 * caller passes it back in for the new translation to execute. The
289 * process yielded a set of ofpacts that can be translated directly, so it
290 * is not much of a special case at that point.
291 */
292 int recirc_action_offset; /* Offset in 'action_set' to actions to be
293 * executed after recirculation, or -1. */
294 int last_unroll_offset; /* Offset in 'action_set' to the latest unroll
295 * action, or -1. */
296
297 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
298 * This is a trigger for recirculation in cases where translating an action
299 * or looking up a flow requires access to the fields of the packet after
300 * the MPLS label stack that was originally present. */
301 bool was_mpls;
302
303 /* True if conntrack has been performed on this packet during processing
304 * on the current bridge. This is used to determine whether conntrack
305 * state from the datapath should be honored after recirculation. */
306 bool conntracked;
307
308 /* OpenFlow 1.1+ action set.
309 *
310 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
311 * When translation is otherwise complete, ofpacts_execute_action_set()
312 * converts it to a set of "struct ofpact"s that can be translated into
313 * datapath actions. */
314 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
315 struct ofpbuf action_set; /* Action set. */
316 };
317
318 static void xlate_action_set(struct xlate_ctx *ctx);
319 static void xlate_commit_actions(struct xlate_ctx *ctx);
320
321 static void
322 ctx_trigger_recirculation(struct xlate_ctx *ctx)
323 {
324 ctx->exit = true;
325 ctx->recirc_action_offset = ctx->action_set.size;
326 }
327
328 static bool
329 ctx_first_recirculation_action(const struct xlate_ctx *ctx)
330 {
331 return ctx->recirc_action_offset == ctx->action_set.size;
332 }
333
334 static inline bool
335 exit_recirculates(const struct xlate_ctx *ctx)
336 {
337 /* When recirculating the 'recirc_action_offset' has a non-negative value.
338 */
339 return ctx->recirc_action_offset >= 0;
340 }
341
342 static void compose_recirculate_action(struct xlate_ctx *ctx);
343
344 /* A controller may use OFPP_NONE as the ingress port to indicate that
345 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
346 * when an input bundle is needed for validation (e.g., mirroring or
347 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
348 * any 'port' structs, so care must be taken when dealing with it. */
349 static struct xbundle ofpp_none_bundle = {
350 .name = "OFPP_NONE",
351 .vlan_mode = PORT_VLAN_TRUNK
352 };
353
354 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
355 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
356 * traffic egressing the 'ofport' with that priority should be marked with. */
357 struct skb_priority_to_dscp {
358 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
359 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
360
361 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
362 };
363
364 enum xc_type {
365 XC_RULE,
366 XC_BOND,
367 XC_NETDEV,
368 XC_NETFLOW,
369 XC_MIRROR,
370 XC_LEARN,
371 XC_NORMAL,
372 XC_FIN_TIMEOUT,
373 XC_GROUP,
374 XC_TNL_ARP,
375 };
376
377 /* xlate_cache entries hold enough information to perform the side effects of
378 * xlate_actions() for a rule, without needing to perform rule translation
379 * from scratch. The primary usage of these is to submit statistics to objects
380 * that a flow relates to, although they may be used for other effects as well
381 * (for instance, refreshing hard timeouts for learned flows). */
382 struct xc_entry {
383 enum xc_type type;
384 union {
385 struct rule_dpif *rule;
386 struct {
387 struct netdev *tx;
388 struct netdev *rx;
389 struct bfd *bfd;
390 } dev;
391 struct {
392 struct netflow *netflow;
393 struct flow *flow;
394 ofp_port_t iface;
395 } nf;
396 struct {
397 struct mbridge *mbridge;
398 mirror_mask_t mirrors;
399 } mirror;
400 struct {
401 struct bond *bond;
402 struct flow *flow;
403 uint16_t vid;
404 } bond;
405 struct {
406 struct ofproto_dpif *ofproto;
407 struct ofputil_flow_mod *fm;
408 struct ofpbuf *ofpacts;
409 } learn;
410 struct {
411 struct ofproto_dpif *ofproto;
412 struct flow *flow;
413 int vlan;
414 } normal;
415 struct {
416 struct rule_dpif *rule;
417 uint16_t idle;
418 uint16_t hard;
419 } fin;
420 struct {
421 struct group_dpif *group;
422 struct ofputil_bucket *bucket;
423 } group;
424 struct {
425 char br_name[IFNAMSIZ];
426 ovs_be32 d_ip;
427 } tnl_arp_cache;
428 } u;
429 };
430
431 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
432 entries = xcache->entries; \
433 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
434 entry; \
435 entry = ofpbuf_try_pull(&entries, sizeof *entry))
436
437 struct xlate_cache {
438 struct ofpbuf entries;
439 };
440
441 /* Xlate config contains hash maps of all bridges, bundles and ports.
442 * Xcfgp contains the pointer to the current xlate configuration.
443 * When the main thread needs to change the configuration, it copies xcfgp to
444 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
445 * does not block handler and revalidator threads. */
446 struct xlate_cfg {
447 struct hmap xbridges;
448 struct hmap xbundles;
449 struct hmap xports;
450 };
451 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
452 static struct xlate_cfg *new_xcfg = NULL;
453
454 static bool may_receive(const struct xport *, struct xlate_ctx *);
455 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
456 struct xlate_ctx *);
457 static void xlate_normal(struct xlate_ctx *);
458 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
459 OVS_PRINTF_FORMAT(2, 3);
460 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
461 uint8_t table_id, bool may_packet_in,
462 bool honor_table_miss);
463 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
464 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
465 static void output_normal(struct xlate_ctx *, const struct xbundle *,
466 uint16_t vlan);
467
468 /* Optional bond recirculation parameter to compose_output_action(). */
469 struct xlate_bond_recirc {
470 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
471 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
472 uint32_t hash_basis; /* Compute hash for recirc before. */
473 };
474
475 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
476 const struct xlate_bond_recirc *xr);
477
478 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
479 const struct ofproto_dpif *);
480 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
481 const struct ofbundle *);
482 static struct xport *xport_lookup(struct xlate_cfg *,
483 const struct ofport_dpif *);
484 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
485 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
486 uint32_t skb_priority);
487 static void clear_skb_priorities(struct xport *);
488 static size_t count_skb_priorities(const struct xport *);
489 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
490 uint8_t *dscp);
491
492 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
493 enum xc_type type);
494 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
495 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
496 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
497 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
498 const struct mac_learning *, struct stp *,
499 struct rstp *, const struct mcast_snooping *,
500 const struct mbridge *,
501 const struct dpif_sflow *,
502 const struct dpif_ipfix *,
503 const struct netflow *,
504 bool forward_bpdu, bool has_in_band,
505 const struct dpif_backer_support *);
506 static void xlate_xbundle_set(struct xbundle *xbundle,
507 enum port_vlan_mode vlan_mode, int vlan,
508 unsigned long *trunks, bool use_priority_tags,
509 const struct bond *bond, const struct lacp *lacp,
510 bool floodable);
511 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
512 const struct netdev *netdev, const struct cfm *cfm,
513 const struct bfd *bfd, const struct lldp *lldp,
514 int stp_port_no, const struct rstp_port *rstp_port,
515 enum ofputil_port_config config,
516 enum ofputil_port_state state, bool is_tunnel,
517 bool may_enable);
518 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
519 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
520 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
521 static void xlate_xbridge_copy(struct xbridge *);
522 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
523 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
524 struct xport *);
525 static void xlate_xcfg_free(struct xlate_cfg *);
526
527 static inline void
528 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
529 {
530 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
531 va_list args;
532
533 va_start(args, format);
534 ctx->xin->report_hook(ctx->xin, ctx->recurse, format, args);
535 va_end(args);
536 }
537 }
538
539 static inline void
540 xlate_report_actions(struct xlate_ctx *ctx, const char *title,
541 const struct ofpact *ofpacts, size_t ofpacts_len)
542 {
543 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
544 struct ds s = DS_EMPTY_INITIALIZER;
545 ofpacts_format(ofpacts, ofpacts_len, &s);
546 xlate_report(ctx, "%s: %s", title, ds_cstr(&s));
547 ds_destroy(&s);
548 }
549 }
550
551 static void
552 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
553 {
554 list_init(&xbridge->xbundles);
555 hmap_init(&xbridge->xports);
556 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
557 hash_pointer(xbridge->ofproto, 0));
558 }
559
560 static void
561 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
562 {
563 list_init(&xbundle->xports);
564 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
565 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
566 hash_pointer(xbundle->ofbundle, 0));
567 }
568
569 static void
570 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
571 {
572 hmap_init(&xport->skb_priorities);
573 hmap_insert(&xcfg->xports, &xport->hmap_node,
574 hash_pointer(xport->ofport, 0));
575 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
576 hash_ofp_port(xport->ofp_port));
577 }
578
579 static void
580 xlate_xbridge_set(struct xbridge *xbridge,
581 struct dpif *dpif,
582 const struct mac_learning *ml, struct stp *stp,
583 struct rstp *rstp, const struct mcast_snooping *ms,
584 const struct mbridge *mbridge,
585 const struct dpif_sflow *sflow,
586 const struct dpif_ipfix *ipfix,
587 const struct netflow *netflow,
588 bool forward_bpdu, bool has_in_band,
589 const struct dpif_backer_support *support)
590 {
591 if (xbridge->ml != ml) {
592 mac_learning_unref(xbridge->ml);
593 xbridge->ml = mac_learning_ref(ml);
594 }
595
596 if (xbridge->ms != ms) {
597 mcast_snooping_unref(xbridge->ms);
598 xbridge->ms = mcast_snooping_ref(ms);
599 }
600
601 if (xbridge->mbridge != mbridge) {
602 mbridge_unref(xbridge->mbridge);
603 xbridge->mbridge = mbridge_ref(mbridge);
604 }
605
606 if (xbridge->sflow != sflow) {
607 dpif_sflow_unref(xbridge->sflow);
608 xbridge->sflow = dpif_sflow_ref(sflow);
609 }
610
611 if (xbridge->ipfix != ipfix) {
612 dpif_ipfix_unref(xbridge->ipfix);
613 xbridge->ipfix = dpif_ipfix_ref(ipfix);
614 }
615
616 if (xbridge->stp != stp) {
617 stp_unref(xbridge->stp);
618 xbridge->stp = stp_ref(stp);
619 }
620
621 if (xbridge->rstp != rstp) {
622 rstp_unref(xbridge->rstp);
623 xbridge->rstp = rstp_ref(rstp);
624 }
625
626 if (xbridge->netflow != netflow) {
627 netflow_unref(xbridge->netflow);
628 xbridge->netflow = netflow_ref(netflow);
629 }
630
631 xbridge->dpif = dpif;
632 xbridge->forward_bpdu = forward_bpdu;
633 xbridge->has_in_band = has_in_band;
634 xbridge->support = *support;
635 }
636
637 static void
638 xlate_xbundle_set(struct xbundle *xbundle,
639 enum port_vlan_mode vlan_mode, int vlan,
640 unsigned long *trunks, bool use_priority_tags,
641 const struct bond *bond, const struct lacp *lacp,
642 bool floodable)
643 {
644 ovs_assert(xbundle->xbridge);
645
646 xbundle->vlan_mode = vlan_mode;
647 xbundle->vlan = vlan;
648 xbundle->trunks = trunks;
649 xbundle->use_priority_tags = use_priority_tags;
650 xbundle->floodable = floodable;
651
652 if (xbundle->bond != bond) {
653 bond_unref(xbundle->bond);
654 xbundle->bond = bond_ref(bond);
655 }
656
657 if (xbundle->lacp != lacp) {
658 lacp_unref(xbundle->lacp);
659 xbundle->lacp = lacp_ref(lacp);
660 }
661 }
662
663 static void
664 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
665 const struct netdev *netdev, const struct cfm *cfm,
666 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
667 const struct rstp_port* rstp_port,
668 enum ofputil_port_config config, enum ofputil_port_state state,
669 bool is_tunnel, bool may_enable)
670 {
671 xport->config = config;
672 xport->state = state;
673 xport->stp_port_no = stp_port_no;
674 xport->is_tunnel = is_tunnel;
675 xport->may_enable = may_enable;
676 xport->odp_port = odp_port;
677
678 if (xport->rstp_port != rstp_port) {
679 rstp_port_unref(xport->rstp_port);
680 xport->rstp_port = rstp_port_ref(rstp_port);
681 }
682
683 if (xport->cfm != cfm) {
684 cfm_unref(xport->cfm);
685 xport->cfm = cfm_ref(cfm);
686 }
687
688 if (xport->bfd != bfd) {
689 bfd_unref(xport->bfd);
690 xport->bfd = bfd_ref(bfd);
691 }
692
693 if (xport->lldp != lldp) {
694 lldp_unref(xport->lldp);
695 xport->lldp = lldp_ref(lldp);
696 }
697
698 if (xport->netdev != netdev) {
699 netdev_close(xport->netdev);
700 xport->netdev = netdev_ref(netdev);
701 }
702 }
703
704 static void
705 xlate_xbridge_copy(struct xbridge *xbridge)
706 {
707 struct xbundle *xbundle;
708 struct xport *xport;
709 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
710 new_xbridge->ofproto = xbridge->ofproto;
711 new_xbridge->name = xstrdup(xbridge->name);
712 xlate_xbridge_init(new_xcfg, new_xbridge);
713
714 xlate_xbridge_set(new_xbridge,
715 xbridge->dpif, xbridge->ml, xbridge->stp,
716 xbridge->rstp, xbridge->ms, xbridge->mbridge,
717 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
718 xbridge->forward_bpdu, xbridge->has_in_band,
719 &xbridge->support);
720 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
721 xlate_xbundle_copy(new_xbridge, xbundle);
722 }
723
724 /* Copy xports which are not part of a xbundle */
725 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
726 if (!xport->xbundle) {
727 xlate_xport_copy(new_xbridge, NULL, xport);
728 }
729 }
730 }
731
732 static void
733 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
734 {
735 struct xport *xport;
736 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
737 new_xbundle->ofbundle = xbundle->ofbundle;
738 new_xbundle->xbridge = xbridge;
739 new_xbundle->name = xstrdup(xbundle->name);
740 xlate_xbundle_init(new_xcfg, new_xbundle);
741
742 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
743 xbundle->vlan, xbundle->trunks,
744 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
745 xbundle->floodable);
746 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
747 xlate_xport_copy(xbridge, new_xbundle, xport);
748 }
749 }
750
751 static void
752 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
753 struct xport *xport)
754 {
755 struct skb_priority_to_dscp *pdscp, *new_pdscp;
756 struct xport *new_xport = xzalloc(sizeof *xport);
757 new_xport->ofport = xport->ofport;
758 new_xport->ofp_port = xport->ofp_port;
759 new_xport->xbridge = xbridge;
760 xlate_xport_init(new_xcfg, new_xport);
761
762 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
763 xport->bfd, xport->lldp, xport->stp_port_no,
764 xport->rstp_port, xport->config, xport->state,
765 xport->is_tunnel, xport->may_enable);
766
767 if (xport->peer) {
768 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
769 if (peer) {
770 new_xport->peer = peer;
771 new_xport->peer->peer = new_xport;
772 }
773 }
774
775 if (xbundle) {
776 new_xport->xbundle = xbundle;
777 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
778 }
779
780 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
781 new_pdscp = xmalloc(sizeof *pdscp);
782 new_pdscp->skb_priority = pdscp->skb_priority;
783 new_pdscp->dscp = pdscp->dscp;
784 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
785 hash_int(new_pdscp->skb_priority, 0));
786 }
787 }
788
789 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
790 * configuration in xcfgp.
791 *
792 * This needs to be called after editing the xlate configuration.
793 *
794 * Functions that edit the new xlate configuration are
795 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
796 *
797 * A sample workflow:
798 *
799 * xlate_txn_start();
800 * ...
801 * edit_xlate_configuration();
802 * ...
803 * xlate_txn_commit(); */
804 void
805 xlate_txn_commit(void)
806 {
807 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
808
809 ovsrcu_set(&xcfgp, new_xcfg);
810 ovsrcu_synchronize();
811 xlate_xcfg_free(xcfg);
812 new_xcfg = NULL;
813 }
814
815 /* Copies the current xlate configuration in xcfgp to new_xcfg.
816 *
817 * This needs to be called prior to editing the xlate configuration. */
818 void
819 xlate_txn_start(void)
820 {
821 struct xbridge *xbridge;
822 struct xlate_cfg *xcfg;
823
824 ovs_assert(!new_xcfg);
825
826 new_xcfg = xmalloc(sizeof *new_xcfg);
827 hmap_init(&new_xcfg->xbridges);
828 hmap_init(&new_xcfg->xbundles);
829 hmap_init(&new_xcfg->xports);
830
831 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
832 if (!xcfg) {
833 return;
834 }
835
836 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
837 xlate_xbridge_copy(xbridge);
838 }
839 }
840
841
842 static void
843 xlate_xcfg_free(struct xlate_cfg *xcfg)
844 {
845 struct xbridge *xbridge, *next_xbridge;
846
847 if (!xcfg) {
848 return;
849 }
850
851 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
852 xlate_xbridge_remove(xcfg, xbridge);
853 }
854
855 hmap_destroy(&xcfg->xbridges);
856 hmap_destroy(&xcfg->xbundles);
857 hmap_destroy(&xcfg->xports);
858 free(xcfg);
859 }
860
861 void
862 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
863 struct dpif *dpif,
864 const struct mac_learning *ml, struct stp *stp,
865 struct rstp *rstp, const struct mcast_snooping *ms,
866 const struct mbridge *mbridge,
867 const struct dpif_sflow *sflow,
868 const struct dpif_ipfix *ipfix,
869 const struct netflow *netflow,
870 bool forward_bpdu, bool has_in_band,
871 const struct dpif_backer_support *support)
872 {
873 struct xbridge *xbridge;
874
875 ovs_assert(new_xcfg);
876
877 xbridge = xbridge_lookup(new_xcfg, ofproto);
878 if (!xbridge) {
879 xbridge = xzalloc(sizeof *xbridge);
880 xbridge->ofproto = ofproto;
881
882 xlate_xbridge_init(new_xcfg, xbridge);
883 }
884
885 free(xbridge->name);
886 xbridge->name = xstrdup(name);
887
888 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
889 netflow, forward_bpdu, has_in_band, support);
890 }
891
892 static void
893 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
894 {
895 struct xbundle *xbundle, *next_xbundle;
896 struct xport *xport, *next_xport;
897
898 if (!xbridge) {
899 return;
900 }
901
902 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
903 xlate_xport_remove(xcfg, xport);
904 }
905
906 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
907 xlate_xbundle_remove(xcfg, xbundle);
908 }
909
910 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
911 mac_learning_unref(xbridge->ml);
912 mcast_snooping_unref(xbridge->ms);
913 mbridge_unref(xbridge->mbridge);
914 dpif_sflow_unref(xbridge->sflow);
915 dpif_ipfix_unref(xbridge->ipfix);
916 stp_unref(xbridge->stp);
917 rstp_unref(xbridge->rstp);
918 hmap_destroy(&xbridge->xports);
919 free(xbridge->name);
920 free(xbridge);
921 }
922
923 void
924 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
925 {
926 struct xbridge *xbridge;
927
928 ovs_assert(new_xcfg);
929
930 xbridge = xbridge_lookup(new_xcfg, ofproto);
931 xlate_xbridge_remove(new_xcfg, xbridge);
932 }
933
934 void
935 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
936 const char *name, enum port_vlan_mode vlan_mode, int vlan,
937 unsigned long *trunks, bool use_priority_tags,
938 const struct bond *bond, const struct lacp *lacp,
939 bool floodable)
940 {
941 struct xbundle *xbundle;
942
943 ovs_assert(new_xcfg);
944
945 xbundle = xbundle_lookup(new_xcfg, ofbundle);
946 if (!xbundle) {
947 xbundle = xzalloc(sizeof *xbundle);
948 xbundle->ofbundle = ofbundle;
949 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
950
951 xlate_xbundle_init(new_xcfg, xbundle);
952 }
953
954 free(xbundle->name);
955 xbundle->name = xstrdup(name);
956
957 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
958 use_priority_tags, bond, lacp, floodable);
959 }
960
961 static void
962 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
963 {
964 struct xport *xport;
965
966 if (!xbundle) {
967 return;
968 }
969
970 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
971 xport->xbundle = NULL;
972 }
973
974 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
975 list_remove(&xbundle->list_node);
976 bond_unref(xbundle->bond);
977 lacp_unref(xbundle->lacp);
978 free(xbundle->name);
979 free(xbundle);
980 }
981
982 void
983 xlate_bundle_remove(struct ofbundle *ofbundle)
984 {
985 struct xbundle *xbundle;
986
987 ovs_assert(new_xcfg);
988
989 xbundle = xbundle_lookup(new_xcfg, ofbundle);
990 xlate_xbundle_remove(new_xcfg, xbundle);
991 }
992
993 void
994 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
995 struct ofport_dpif *ofport, ofp_port_t ofp_port,
996 odp_port_t odp_port, const struct netdev *netdev,
997 const struct cfm *cfm, const struct bfd *bfd,
998 const struct lldp *lldp, struct ofport_dpif *peer,
999 int stp_port_no, const struct rstp_port *rstp_port,
1000 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
1001 enum ofputil_port_config config,
1002 enum ofputil_port_state state, bool is_tunnel,
1003 bool may_enable)
1004 {
1005 size_t i;
1006 struct xport *xport;
1007
1008 ovs_assert(new_xcfg);
1009
1010 xport = xport_lookup(new_xcfg, ofport);
1011 if (!xport) {
1012 xport = xzalloc(sizeof *xport);
1013 xport->ofport = ofport;
1014 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
1015 xport->ofp_port = ofp_port;
1016
1017 xlate_xport_init(new_xcfg, xport);
1018 }
1019
1020 ovs_assert(xport->ofp_port == ofp_port);
1021
1022 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1023 stp_port_no, rstp_port, config, state, is_tunnel,
1024 may_enable);
1025
1026 if (xport->peer) {
1027 xport->peer->peer = NULL;
1028 }
1029 xport->peer = xport_lookup(new_xcfg, peer);
1030 if (xport->peer) {
1031 xport->peer->peer = xport;
1032 }
1033
1034 if (xport->xbundle) {
1035 list_remove(&xport->bundle_node);
1036 }
1037 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1038 if (xport->xbundle) {
1039 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1040 }
1041
1042 clear_skb_priorities(xport);
1043 for (i = 0; i < n_qdscp; i++) {
1044 struct skb_priority_to_dscp *pdscp;
1045 uint32_t skb_priority;
1046
1047 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1048 &skb_priority)) {
1049 continue;
1050 }
1051
1052 pdscp = xmalloc(sizeof *pdscp);
1053 pdscp->skb_priority = skb_priority;
1054 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1055 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1056 hash_int(pdscp->skb_priority, 0));
1057 }
1058 }
1059
1060 static void
1061 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1062 {
1063 if (!xport) {
1064 return;
1065 }
1066
1067 if (xport->peer) {
1068 xport->peer->peer = NULL;
1069 xport->peer = NULL;
1070 }
1071
1072 if (xport->xbundle) {
1073 list_remove(&xport->bundle_node);
1074 }
1075
1076 clear_skb_priorities(xport);
1077 hmap_destroy(&xport->skb_priorities);
1078
1079 hmap_remove(&xcfg->xports, &xport->hmap_node);
1080 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1081
1082 netdev_close(xport->netdev);
1083 rstp_port_unref(xport->rstp_port);
1084 cfm_unref(xport->cfm);
1085 bfd_unref(xport->bfd);
1086 lldp_unref(xport->lldp);
1087 free(xport);
1088 }
1089
1090 void
1091 xlate_ofport_remove(struct ofport_dpif *ofport)
1092 {
1093 struct xport *xport;
1094
1095 ovs_assert(new_xcfg);
1096
1097 xport = xport_lookup(new_xcfg, ofport);
1098 xlate_xport_remove(new_xcfg, xport);
1099 }
1100
1101 static struct ofproto_dpif *
1102 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1103 ofp_port_t *ofp_in_port, const struct xport **xportp)
1104 {
1105 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1106 const struct xport *xport;
1107
1108 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1109 ? tnl_port_receive(flow)
1110 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1111 if (OVS_UNLIKELY(!xport)) {
1112 return NULL;
1113 }
1114 *xportp = xport;
1115 if (ofp_in_port) {
1116 *ofp_in_port = xport->ofp_port;
1117 }
1118 return xport->xbridge->ofproto;
1119 }
1120
1121 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1122 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1123 struct ofproto_dpif *
1124 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1125 ofp_port_t *ofp_in_port)
1126 {
1127 const struct xport *xport;
1128
1129 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1130 }
1131
1132 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1133 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1134 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1135 * handles for those protocols if they're enabled. Caller may use the returned
1136 * pointers until quiescing, for longer term use additional references must
1137 * be taken.
1138 *
1139 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1140 */
1141 int
1142 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1143 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1144 struct dpif_sflow **sflow, struct netflow **netflow,
1145 ofp_port_t *ofp_in_port)
1146 {
1147 struct ofproto_dpif *ofproto;
1148 const struct xport *xport;
1149
1150 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1151
1152 if (!ofproto) {
1153 return ENODEV;
1154 }
1155
1156 if (ofprotop) {
1157 *ofprotop = ofproto;
1158 }
1159
1160 if (ipfix) {
1161 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1162 }
1163
1164 if (sflow) {
1165 *sflow = xport ? xport->xbridge->sflow : NULL;
1166 }
1167
1168 if (netflow) {
1169 *netflow = xport ? xport->xbridge->netflow : NULL;
1170 }
1171
1172 return 0;
1173 }
1174
1175 static struct xbridge *
1176 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1177 {
1178 struct hmap *xbridges;
1179 struct xbridge *xbridge;
1180
1181 if (!ofproto || !xcfg) {
1182 return NULL;
1183 }
1184
1185 xbridges = &xcfg->xbridges;
1186
1187 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1188 xbridges) {
1189 if (xbridge->ofproto == ofproto) {
1190 return xbridge;
1191 }
1192 }
1193 return NULL;
1194 }
1195
1196 static struct xbundle *
1197 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1198 {
1199 struct hmap *xbundles;
1200 struct xbundle *xbundle;
1201
1202 if (!ofbundle || !xcfg) {
1203 return NULL;
1204 }
1205
1206 xbundles = &xcfg->xbundles;
1207
1208 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1209 xbundles) {
1210 if (xbundle->ofbundle == ofbundle) {
1211 return xbundle;
1212 }
1213 }
1214 return NULL;
1215 }
1216
1217 static struct xport *
1218 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1219 {
1220 struct hmap *xports;
1221 struct xport *xport;
1222
1223 if (!ofport || !xcfg) {
1224 return NULL;
1225 }
1226
1227 xports = &xcfg->xports;
1228
1229 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1230 xports) {
1231 if (xport->ofport == ofport) {
1232 return xport;
1233 }
1234 }
1235 return NULL;
1236 }
1237
1238 static struct stp_port *
1239 xport_get_stp_port(const struct xport *xport)
1240 {
1241 return xport->xbridge->stp && xport->stp_port_no != -1
1242 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1243 : NULL;
1244 }
1245
1246 static bool
1247 xport_stp_learn_state(const struct xport *xport)
1248 {
1249 struct stp_port *sp = xport_get_stp_port(xport);
1250 return sp
1251 ? stp_learn_in_state(stp_port_get_state(sp))
1252 : true;
1253 }
1254
1255 static bool
1256 xport_stp_forward_state(const struct xport *xport)
1257 {
1258 struct stp_port *sp = xport_get_stp_port(xport);
1259 return sp
1260 ? stp_forward_in_state(stp_port_get_state(sp))
1261 : true;
1262 }
1263
1264 static bool
1265 xport_stp_should_forward_bpdu(const struct xport *xport)
1266 {
1267 struct stp_port *sp = xport_get_stp_port(xport);
1268 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1269 }
1270
1271 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1272 * were used to make the determination.*/
1273 static bool
1274 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1275 {
1276 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1277 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1278 return is_stp(flow);
1279 }
1280
1281 static void
1282 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1283 {
1284 struct stp_port *sp = xport_get_stp_port(xport);
1285 struct dp_packet payload = *packet;
1286 struct eth_header *eth = dp_packet_data(&payload);
1287
1288 /* Sink packets on ports that have STP disabled when the bridge has
1289 * STP enabled. */
1290 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1291 return;
1292 }
1293
1294 /* Trim off padding on payload. */
1295 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1296 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1297 }
1298
1299 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1300 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1301 }
1302 }
1303
1304 static enum rstp_state
1305 xport_get_rstp_port_state(const struct xport *xport)
1306 {
1307 return xport->rstp_port
1308 ? rstp_port_get_state(xport->rstp_port)
1309 : RSTP_DISABLED;
1310 }
1311
1312 static bool
1313 xport_rstp_learn_state(const struct xport *xport)
1314 {
1315 return xport->xbridge->rstp && xport->rstp_port
1316 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1317 : true;
1318 }
1319
1320 static bool
1321 xport_rstp_forward_state(const struct xport *xport)
1322 {
1323 return xport->xbridge->rstp && xport->rstp_port
1324 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1325 : true;
1326 }
1327
1328 static bool
1329 xport_rstp_should_manage_bpdu(const struct xport *xport)
1330 {
1331 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1332 }
1333
1334 static void
1335 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1336 {
1337 struct dp_packet payload = *packet;
1338 struct eth_header *eth = dp_packet_data(&payload);
1339
1340 /* Sink packets on ports that have no RSTP. */
1341 if (!xport->rstp_port) {
1342 return;
1343 }
1344
1345 /* Trim off padding on payload. */
1346 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1347 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1348 }
1349
1350 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1351 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1352 dp_packet_size(&payload));
1353 }
1354 }
1355
1356 static struct xport *
1357 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1358 {
1359 struct xport *xport;
1360
1361 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1362 &xbridge->xports) {
1363 if (xport->ofp_port == ofp_port) {
1364 return xport;
1365 }
1366 }
1367 return NULL;
1368 }
1369
1370 static odp_port_t
1371 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1372 {
1373 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1374 return xport ? xport->odp_port : ODPP_NONE;
1375 }
1376
1377 static bool
1378 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1379 {
1380 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1381 return xport && xport->may_enable;
1382 }
1383
1384 static struct ofputil_bucket *
1385 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1386 int depth);
1387
1388 static bool
1389 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1390 {
1391 struct group_dpif *group;
1392
1393 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1394 struct ofputil_bucket *bucket;
1395
1396 bucket = group_first_live_bucket(ctx, group, depth);
1397 group_dpif_unref(group);
1398 return bucket == NULL;
1399 }
1400
1401 return false;
1402 }
1403
1404 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1405
1406 static bool
1407 bucket_is_alive(const struct xlate_ctx *ctx,
1408 struct ofputil_bucket *bucket, int depth)
1409 {
1410 if (depth >= MAX_LIVENESS_RECURSION) {
1411 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1412
1413 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1414 MAX_LIVENESS_RECURSION);
1415 return false;
1416 }
1417
1418 return (!ofputil_bucket_has_liveness(bucket)
1419 || (bucket->watch_port != OFPP_ANY
1420 && odp_port_is_alive(ctx, bucket->watch_port))
1421 || (bucket->watch_group != OFPG_ANY
1422 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1423 }
1424
1425 static struct ofputil_bucket *
1426 group_first_live_bucket(const struct xlate_ctx *ctx,
1427 const struct group_dpif *group, int depth)
1428 {
1429 struct ofputil_bucket *bucket;
1430 const struct ovs_list *buckets;
1431
1432 group_dpif_get_buckets(group, &buckets);
1433 LIST_FOR_EACH (bucket, list_node, buckets) {
1434 if (bucket_is_alive(ctx, bucket, depth)) {
1435 return bucket;
1436 }
1437 }
1438
1439 return NULL;
1440 }
1441
1442 static struct ofputil_bucket *
1443 group_best_live_bucket(const struct xlate_ctx *ctx,
1444 const struct group_dpif *group,
1445 uint32_t basis)
1446 {
1447 struct ofputil_bucket *best_bucket = NULL;
1448 uint32_t best_score = 0;
1449 int i = 0;
1450
1451 struct ofputil_bucket *bucket;
1452 const struct ovs_list *buckets;
1453
1454 group_dpif_get_buckets(group, &buckets);
1455 LIST_FOR_EACH (bucket, list_node, buckets) {
1456 if (bucket_is_alive(ctx, bucket, 0)) {
1457 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1458 if (score >= best_score) {
1459 best_bucket = bucket;
1460 best_score = score;
1461 }
1462 }
1463 i++;
1464 }
1465
1466 return best_bucket;
1467 }
1468
1469 static bool
1470 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1471 {
1472 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1473 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1474 }
1475
1476 static bool
1477 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1478 {
1479 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1480 }
1481
1482 static mirror_mask_t
1483 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1484 {
1485 return xbundle != &ofpp_none_bundle
1486 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1487 : 0;
1488 }
1489
1490 static mirror_mask_t
1491 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1492 {
1493 return xbundle != &ofpp_none_bundle
1494 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1495 : 0;
1496 }
1497
1498 static mirror_mask_t
1499 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1500 {
1501 return xbundle != &ofpp_none_bundle
1502 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1503 : 0;
1504 }
1505
1506 static struct xbundle *
1507 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1508 bool warn, struct xport **in_xportp)
1509 {
1510 struct xport *xport;
1511
1512 /* Find the port and bundle for the received packet. */
1513 xport = get_ofp_port(xbridge, in_port);
1514 if (in_xportp) {
1515 *in_xportp = xport;
1516 }
1517 if (xport && xport->xbundle) {
1518 return xport->xbundle;
1519 }
1520
1521 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1522 * which a controller may use as the ingress port for traffic that
1523 * it is sourcing. */
1524 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1525 return &ofpp_none_bundle;
1526 }
1527
1528 /* Odd. A few possible reasons here:
1529 *
1530 * - We deleted a port but there are still a few packets queued up
1531 * from it.
1532 *
1533 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1534 * we don't know about.
1535 *
1536 * - The ofproto client didn't configure the port as part of a bundle.
1537 * This is particularly likely to happen if a packet was received on the
1538 * port after it was created, but before the client had a chance to
1539 * configure its bundle.
1540 */
1541 if (warn) {
1542 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1543
1544 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1545 "port %"PRIu16, xbridge->name, in_port);
1546 }
1547 return NULL;
1548 }
1549
1550 static void
1551 mirror_packet(struct xlate_ctx *ctx, struct xbundle *xbundle,
1552 mirror_mask_t mirrors)
1553 {
1554 bool warn = ctx->xin->packet != NULL;
1555 uint16_t vid = vlan_tci_to_vid(ctx->xin->flow.vlan_tci);
1556 if (!input_vid_is_valid(vid, xbundle, warn)) {
1557 return;
1558 }
1559 uint16_t vlan = input_vid_to_vlan(xbundle, vid);
1560
1561 const struct xbridge *xbridge = ctx->xbridge;
1562
1563 /* Don't mirror to destinations that we've already mirrored to. */
1564 mirrors &= ~ctx->mirrors;
1565 if (!mirrors) {
1566 return;
1567 }
1568
1569 /* Record these mirrors so that we don't mirror to them again. */
1570 ctx->mirrors |= mirrors;
1571
1572 if (ctx->xin->resubmit_stats) {
1573 mirror_update_stats(xbridge->mbridge, mirrors,
1574 ctx->xin->resubmit_stats->n_packets,
1575 ctx->xin->resubmit_stats->n_bytes);
1576 }
1577 if (ctx->xin->xcache) {
1578 struct xc_entry *entry;
1579
1580 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_MIRROR);
1581 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
1582 entry->u.mirror.mirrors = mirrors;
1583 }
1584
1585 while (mirrors) {
1586 const unsigned long *vlans;
1587 mirror_mask_t dup_mirrors;
1588 struct ofbundle *out;
1589 int out_vlan;
1590
1591 bool has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1592 &vlans, &dup_mirrors, &out, &out_vlan);
1593 ovs_assert(has_mirror);
1594
1595 if (vlans) {
1596 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1597 }
1598
1599 if (vlans && !bitmap_is_set(vlans, vlan)) {
1600 mirrors = zero_rightmost_1bit(mirrors);
1601 continue;
1602 }
1603
1604 mirrors &= ~dup_mirrors;
1605 ctx->mirrors |= dup_mirrors;
1606 if (out) {
1607 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1608 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1609 if (out_xbundle) {
1610 output_normal(ctx, out_xbundle, vlan);
1611 }
1612 } else if (vlan != out_vlan
1613 && !eth_addr_is_reserved(ctx->xin->flow.dl_dst)) {
1614 struct xbundle *xbundle;
1615
1616 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1617 if (xbundle_includes_vlan(xbundle, out_vlan)
1618 && !xbundle_mirror_out(xbridge, xbundle)) {
1619 output_normal(ctx, xbundle, out_vlan);
1620 }
1621 }
1622 }
1623 }
1624 }
1625
1626 static void
1627 mirror_ingress_packet(struct xlate_ctx *ctx)
1628 {
1629 if (mbridge_has_mirrors(ctx->xbridge->mbridge)) {
1630 bool warn = ctx->xin->packet != NULL;
1631 struct xbundle *xbundle = lookup_input_bundle(
1632 ctx->xbridge, ctx->xin->flow.in_port.ofp_port, warn, NULL);
1633 if (xbundle) {
1634 mirror_packet(ctx, xbundle,
1635 xbundle_mirror_src(ctx->xbridge, xbundle));
1636 }
1637 }
1638 }
1639
1640 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1641 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1642 * the bundle on which the packet was received, returns the VLAN to which the
1643 * packet belongs.
1644 *
1645 * Both 'vid' and the return value are in the range 0...4095. */
1646 static uint16_t
1647 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1648 {
1649 switch (in_xbundle->vlan_mode) {
1650 case PORT_VLAN_ACCESS:
1651 return in_xbundle->vlan;
1652 break;
1653
1654 case PORT_VLAN_TRUNK:
1655 return vid;
1656
1657 case PORT_VLAN_NATIVE_UNTAGGED:
1658 case PORT_VLAN_NATIVE_TAGGED:
1659 return vid ? vid : in_xbundle->vlan;
1660
1661 default:
1662 OVS_NOT_REACHED();
1663 }
1664 }
1665
1666 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1667 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1668 * a warning.
1669 *
1670 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1671 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1672 * 0...4095. */
1673 static bool
1674 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1675 {
1676 /* Allow any VID on the OFPP_NONE port. */
1677 if (in_xbundle == &ofpp_none_bundle) {
1678 return true;
1679 }
1680
1681 switch (in_xbundle->vlan_mode) {
1682 case PORT_VLAN_ACCESS:
1683 if (vid) {
1684 if (warn) {
1685 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1686 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1687 "packet received on port %s configured as VLAN "
1688 "%"PRIu16" access port", vid, in_xbundle->name,
1689 in_xbundle->vlan);
1690 }
1691 return false;
1692 }
1693 return true;
1694
1695 case PORT_VLAN_NATIVE_UNTAGGED:
1696 case PORT_VLAN_NATIVE_TAGGED:
1697 if (!vid) {
1698 /* Port must always carry its native VLAN. */
1699 return true;
1700 }
1701 /* Fall through. */
1702 case PORT_VLAN_TRUNK:
1703 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1704 if (warn) {
1705 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1706 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1707 "received on port %s not configured for trunking "
1708 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1709 }
1710 return false;
1711 }
1712 return true;
1713
1714 default:
1715 OVS_NOT_REACHED();
1716 }
1717
1718 }
1719
1720 /* Given 'vlan', the VLAN that a packet belongs to, and
1721 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1722 * that should be included in the 802.1Q header. (If the return value is 0,
1723 * then the 802.1Q header should only be included in the packet if there is a
1724 * nonzero PCP.)
1725 *
1726 * Both 'vlan' and the return value are in the range 0...4095. */
1727 static uint16_t
1728 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1729 {
1730 switch (out_xbundle->vlan_mode) {
1731 case PORT_VLAN_ACCESS:
1732 return 0;
1733
1734 case PORT_VLAN_TRUNK:
1735 case PORT_VLAN_NATIVE_TAGGED:
1736 return vlan;
1737
1738 case PORT_VLAN_NATIVE_UNTAGGED:
1739 return vlan == out_xbundle->vlan ? 0 : vlan;
1740
1741 default:
1742 OVS_NOT_REACHED();
1743 }
1744 }
1745
1746 static void
1747 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1748 uint16_t vlan)
1749 {
1750 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1751 uint16_t vid;
1752 ovs_be16 tci, old_tci;
1753 struct xport *xport;
1754 struct xlate_bond_recirc xr;
1755 bool use_recirc = false;
1756
1757 vid = output_vlan_to_vid(out_xbundle, vlan);
1758 if (list_is_empty(&out_xbundle->xports)) {
1759 /* Partially configured bundle with no slaves. Drop the packet. */
1760 return;
1761 } else if (!out_xbundle->bond) {
1762 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1763 bundle_node);
1764 } else {
1765 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1766 struct flow_wildcards *wc = ctx->wc;
1767 struct ofport_dpif *ofport;
1768
1769 if (ctx->xbridge->support.odp.recirc) {
1770 use_recirc = bond_may_recirc(
1771 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1772
1773 if (use_recirc) {
1774 /* Only TCP mode uses recirculation. */
1775 xr.hash_alg = OVS_HASH_ALG_L4;
1776 bond_update_post_recirc_rules(out_xbundle->bond, false);
1777
1778 /* Recirculation does not require unmasking hash fields. */
1779 wc = NULL;
1780 }
1781 }
1782
1783 ofport = bond_choose_output_slave(out_xbundle->bond,
1784 &ctx->xin->flow, wc, vid);
1785 xport = xport_lookup(xcfg, ofport);
1786
1787 if (!xport) {
1788 /* No slaves enabled, so drop packet. */
1789 return;
1790 }
1791
1792 /* If use_recirc is set, the main thread will handle stats
1793 * accounting for this bond. */
1794 if (!use_recirc) {
1795 if (ctx->xin->resubmit_stats) {
1796 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1797 ctx->xin->resubmit_stats->n_bytes);
1798 }
1799 if (ctx->xin->xcache) {
1800 struct xc_entry *entry;
1801 struct flow *flow;
1802
1803 flow = &ctx->xin->flow;
1804 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1805 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1806 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1807 entry->u.bond.vid = vid;
1808 }
1809 }
1810 }
1811
1812 old_tci = *flow_tci;
1813 tci = htons(vid);
1814 if (tci || out_xbundle->use_priority_tags) {
1815 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1816 if (tci) {
1817 tci |= htons(VLAN_CFI);
1818 }
1819 }
1820 *flow_tci = tci;
1821
1822 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1823 *flow_tci = old_tci;
1824 }
1825
1826 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1827 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1828 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1829 static bool
1830 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1831 {
1832 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1833 return false;
1834 }
1835
1836 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1837 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1838 return false;
1839 }
1840
1841 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1842 if (flow->nw_proto == ARP_OP_REPLY) {
1843 return true;
1844 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1845 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1846 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1847
1848 return flow->nw_src == flow->nw_dst;
1849 } else {
1850 return false;
1851 }
1852 }
1853
1854 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1855 * dropped. Returns true if they may be forwarded, false if they should be
1856 * dropped.
1857 *
1858 * 'in_port' must be the xport that corresponds to flow->in_port.
1859 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1860 *
1861 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1862 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1863 * checked by input_vid_is_valid().
1864 *
1865 * May also add tags to '*tags', although the current implementation only does
1866 * so in one special case.
1867 */
1868 static bool
1869 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1870 uint16_t vlan)
1871 {
1872 struct xbundle *in_xbundle = in_port->xbundle;
1873 const struct xbridge *xbridge = ctx->xbridge;
1874 struct flow *flow = &ctx->xin->flow;
1875
1876 /* Drop frames for reserved multicast addresses
1877 * only if forward_bpdu option is absent. */
1878 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1879 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1880 return false;
1881 }
1882
1883 if (in_xbundle->bond) {
1884 struct mac_entry *mac;
1885
1886 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1887 flow->dl_dst)) {
1888 case BV_ACCEPT:
1889 break;
1890
1891 case BV_DROP:
1892 xlate_report(ctx, "bonding refused admissibility, dropping");
1893 return false;
1894
1895 case BV_DROP_IF_MOVED:
1896 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1897 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1898 if (mac
1899 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1900 && (!is_gratuitous_arp(flow, ctx->wc)
1901 || mac_entry_is_grat_arp_locked(mac))) {
1902 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1903 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1904 "dropping");
1905 return false;
1906 }
1907 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1908 break;
1909 }
1910 }
1911
1912 return true;
1913 }
1914
1915 /* Checks whether a MAC learning update is necessary for MAC learning table
1916 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1917 * 'vlan'.
1918 *
1919 * Most packets processed through the MAC learning table do not actually
1920 * change it in any way. This function requires only a read lock on the MAC
1921 * learning table, so it is much cheaper in this common case.
1922 *
1923 * Keep the code here synchronized with that in update_learning_table__()
1924 * below. */
1925 static bool
1926 is_mac_learning_update_needed(const struct mac_learning *ml,
1927 const struct flow *flow,
1928 struct flow_wildcards *wc,
1929 int vlan, struct xbundle *in_xbundle)
1930 OVS_REQ_RDLOCK(ml->rwlock)
1931 {
1932 struct mac_entry *mac;
1933
1934 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1935 return false;
1936 }
1937
1938 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1939 if (!mac || mac_entry_age(ml, mac)) {
1940 return true;
1941 }
1942
1943 if (is_gratuitous_arp(flow, wc)) {
1944 /* We don't want to learn from gratuitous ARP packets that are
1945 * reflected back over bond slaves so we lock the learning table. */
1946 if (!in_xbundle->bond) {
1947 return true;
1948 } else if (mac_entry_is_grat_arp_locked(mac)) {
1949 return false;
1950 }
1951 }
1952
1953 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1954 }
1955
1956
1957 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1958 * received on 'in_xbundle' in 'vlan'.
1959 *
1960 * This code repeats all the checks in is_mac_learning_update_needed() because
1961 * the lock was released between there and here and thus the MAC learning state
1962 * could have changed.
1963 *
1964 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1965 * above. */
1966 static void
1967 update_learning_table__(const struct xbridge *xbridge,
1968 const struct flow *flow, struct flow_wildcards *wc,
1969 int vlan, struct xbundle *in_xbundle)
1970 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1971 {
1972 struct mac_entry *mac;
1973
1974 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1975 return;
1976 }
1977
1978 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1979 if (is_gratuitous_arp(flow, wc)) {
1980 /* We don't want to learn from gratuitous ARP packets that are
1981 * reflected back over bond slaves so we lock the learning table. */
1982 if (!in_xbundle->bond) {
1983 mac_entry_set_grat_arp_lock(mac);
1984 } else if (mac_entry_is_grat_arp_locked(mac)) {
1985 return;
1986 }
1987 }
1988
1989 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1990 /* The log messages here could actually be useful in debugging,
1991 * so keep the rate limit relatively high. */
1992 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1993
1994 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1995 "on port %s in VLAN %d",
1996 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1997 in_xbundle->name, vlan);
1998
1999 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
2000 }
2001 }
2002
2003 static void
2004 update_learning_table(const struct xbridge *xbridge,
2005 const struct flow *flow, struct flow_wildcards *wc,
2006 int vlan, struct xbundle *in_xbundle)
2007 {
2008 bool need_update;
2009
2010 /* Don't learn the OFPP_NONE port. */
2011 if (in_xbundle == &ofpp_none_bundle) {
2012 return;
2013 }
2014
2015 /* First try the common case: no change to MAC learning table. */
2016 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
2017 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
2018 in_xbundle);
2019 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2020
2021 if (need_update) {
2022 /* Slow path: MAC learning table might need an update. */
2023 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2024 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2025 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2026 }
2027 }
2028
2029 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2030 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2031 static void
2032 update_mcast_snooping_table4__(const struct xbridge *xbridge,
2033 const struct flow *flow,
2034 struct mcast_snooping *ms, int vlan,
2035 struct xbundle *in_xbundle,
2036 const struct dp_packet *packet)
2037 OVS_REQ_WRLOCK(ms->rwlock)
2038 {
2039 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2040 int count;
2041 ovs_be32 ip4 = flow->igmp_group_ip4;
2042
2043 switch (ntohs(flow->tp_src)) {
2044 case IGMP_HOST_MEMBERSHIP_REPORT:
2045 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2046 if (mcast_snooping_add_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2047 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2048 IP_FMT" is on port %s in VLAN %d",
2049 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2050 }
2051 break;
2052 case IGMP_HOST_LEAVE_MESSAGE:
2053 if (mcast_snooping_leave_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2054 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2055 IP_FMT" is on port %s in VLAN %d",
2056 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2057 }
2058 break;
2059 case IGMP_HOST_MEMBERSHIP_QUERY:
2060 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2061 in_xbundle->ofbundle)) {
2062 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2063 IP_FMT" is on port %s in VLAN %d",
2064 xbridge->name, IP_ARGS(flow->nw_src),
2065 in_xbundle->name, vlan);
2066 }
2067 break;
2068 case IGMPV3_HOST_MEMBERSHIP_REPORT:
2069 if ((count = mcast_snooping_add_report(ms, packet, vlan,
2070 in_xbundle->ofbundle))) {
2071 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2072 "addresses on port %s in VLAN %d",
2073 xbridge->name, count, in_xbundle->name, vlan);
2074 }
2075 break;
2076 }
2077 }
2078
2079 static void
2080 update_mcast_snooping_table6__(const struct xbridge *xbridge,
2081 const struct flow *flow,
2082 struct mcast_snooping *ms, int vlan,
2083 struct xbundle *in_xbundle,
2084 const struct dp_packet *packet)
2085 OVS_REQ_WRLOCK(ms->rwlock)
2086 {
2087 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2088 int count;
2089
2090 switch (ntohs(flow->tp_src)) {
2091 case MLD_QUERY:
2092 if (!ipv6_addr_equals(&flow->ipv6_src, &in6addr_any)
2093 && mcast_snooping_add_mrouter(ms, vlan, in_xbundle->ofbundle)) {
2094 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query on port %s"
2095 "in VLAN %d",
2096 xbridge->name, in_xbundle->name, vlan);
2097 }
2098 break;
2099 case MLD_REPORT:
2100 case MLD_DONE:
2101 case MLD2_REPORT:
2102 count = mcast_snooping_add_mld(ms, packet, vlan, in_xbundle->ofbundle);
2103 if (count) {
2104 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2105 "addresses on port %s in VLAN %d",
2106 xbridge->name, count, in_xbundle->name, vlan);
2107 }
2108 break;
2109 }
2110 }
2111
2112 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2113 * was received on 'in_xbundle' in 'vlan'. */
2114 static void
2115 update_mcast_snooping_table(const struct xbridge *xbridge,
2116 const struct flow *flow, int vlan,
2117 struct xbundle *in_xbundle,
2118 const struct dp_packet *packet)
2119 {
2120 struct mcast_snooping *ms = xbridge->ms;
2121 struct xlate_cfg *xcfg;
2122 struct xbundle *mcast_xbundle;
2123 struct mcast_port_bundle *fport;
2124
2125 /* Don't learn the OFPP_NONE port. */
2126 if (in_xbundle == &ofpp_none_bundle) {
2127 return;
2128 }
2129
2130 /* Don't learn from flood ports */
2131 mcast_xbundle = NULL;
2132 ovs_rwlock_wrlock(&ms->rwlock);
2133 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2134 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2135 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2136 if (mcast_xbundle == in_xbundle) {
2137 break;
2138 }
2139 }
2140
2141 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2142 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2143 update_mcast_snooping_table4__(xbridge, flow, ms, vlan,
2144 in_xbundle, packet);
2145 } else {
2146 update_mcast_snooping_table6__(xbridge, flow, ms, vlan,
2147 in_xbundle, packet);
2148 }
2149 }
2150 ovs_rwlock_unlock(&ms->rwlock);
2151 }
2152
2153 /* send the packet to ports having the multicast group learned */
2154 static void
2155 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2156 struct mcast_snooping *ms OVS_UNUSED,
2157 struct mcast_group *grp,
2158 struct xbundle *in_xbundle, uint16_t vlan)
2159 OVS_REQ_RDLOCK(ms->rwlock)
2160 {
2161 struct xlate_cfg *xcfg;
2162 struct mcast_group_bundle *b;
2163 struct xbundle *mcast_xbundle;
2164
2165 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2166 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2167 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2168 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2169 xlate_report(ctx, "forwarding to mcast group port");
2170 output_normal(ctx, mcast_xbundle, vlan);
2171 } else if (!mcast_xbundle) {
2172 xlate_report(ctx, "mcast group port is unknown, dropping");
2173 } else {
2174 xlate_report(ctx, "mcast group port is input port, dropping");
2175 }
2176 }
2177 }
2178
2179 /* send the packet to ports connected to multicast routers */
2180 static void
2181 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2182 struct mcast_snooping *ms,
2183 struct xbundle *in_xbundle, uint16_t vlan)
2184 OVS_REQ_RDLOCK(ms->rwlock)
2185 {
2186 struct xlate_cfg *xcfg;
2187 struct mcast_mrouter_bundle *mrouter;
2188 struct xbundle *mcast_xbundle;
2189
2190 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2191 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2192 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2193 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2194 xlate_report(ctx, "forwarding to mcast router port");
2195 output_normal(ctx, mcast_xbundle, vlan);
2196 } else if (!mcast_xbundle) {
2197 xlate_report(ctx, "mcast router port is unknown, dropping");
2198 } else {
2199 xlate_report(ctx, "mcast router port is input port, dropping");
2200 }
2201 }
2202 }
2203
2204 /* send the packet to ports flagged to be flooded */
2205 static void
2206 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2207 struct mcast_snooping *ms,
2208 struct xbundle *in_xbundle, uint16_t vlan)
2209 OVS_REQ_RDLOCK(ms->rwlock)
2210 {
2211 struct xlate_cfg *xcfg;
2212 struct mcast_port_bundle *fport;
2213 struct xbundle *mcast_xbundle;
2214
2215 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2216 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2217 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2218 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2219 xlate_report(ctx, "forwarding to mcast flood port");
2220 output_normal(ctx, mcast_xbundle, vlan);
2221 } else if (!mcast_xbundle) {
2222 xlate_report(ctx, "mcast flood port is unknown, dropping");
2223 } else {
2224 xlate_report(ctx, "mcast flood port is input port, dropping");
2225 }
2226 }
2227 }
2228
2229 /* forward the Reports to configured ports */
2230 static void
2231 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2232 struct mcast_snooping *ms,
2233 struct xbundle *in_xbundle, uint16_t vlan)
2234 OVS_REQ_RDLOCK(ms->rwlock)
2235 {
2236 struct xlate_cfg *xcfg;
2237 struct mcast_port_bundle *rport;
2238 struct xbundle *mcast_xbundle;
2239
2240 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2241 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2242 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2243 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2244 xlate_report(ctx, "forwarding Report to mcast flagged port");
2245 output_normal(ctx, mcast_xbundle, vlan);
2246 } else if (!mcast_xbundle) {
2247 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2248 } else {
2249 xlate_report(ctx, "mcast port is input port, dropping the Report");
2250 }
2251 }
2252 }
2253
2254 static void
2255 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2256 uint16_t vlan)
2257 {
2258 struct xbundle *xbundle;
2259
2260 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2261 if (xbundle != in_xbundle
2262 && xbundle_includes_vlan(xbundle, vlan)
2263 && xbundle->floodable
2264 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2265 output_normal(ctx, xbundle, vlan);
2266 }
2267 }
2268 ctx->nf_output_iface = NF_OUT_FLOOD;
2269 }
2270
2271 static void
2272 xlate_normal(struct xlate_ctx *ctx)
2273 {
2274 struct flow_wildcards *wc = ctx->wc;
2275 struct flow *flow = &ctx->xin->flow;
2276 struct xbundle *in_xbundle;
2277 struct xport *in_port;
2278 struct mac_entry *mac;
2279 void *mac_port;
2280 uint16_t vlan;
2281 uint16_t vid;
2282
2283 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2284 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2285 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2286
2287 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2288 ctx->xin->packet != NULL, &in_port);
2289 if (!in_xbundle) {
2290 xlate_report(ctx, "no input bundle, dropping");
2291 return;
2292 }
2293
2294 /* Drop malformed frames. */
2295 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2296 !(flow->vlan_tci & htons(VLAN_CFI))) {
2297 if (ctx->xin->packet != NULL) {
2298 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2299 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2300 "VLAN tag received on port %s",
2301 ctx->xbridge->name, in_xbundle->name);
2302 }
2303 xlate_report(ctx, "partial VLAN tag, dropping");
2304 return;
2305 }
2306
2307 /* Drop frames on bundles reserved for mirroring. */
2308 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2309 if (ctx->xin->packet != NULL) {
2310 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2311 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2312 "%s, which is reserved exclusively for mirroring",
2313 ctx->xbridge->name, in_xbundle->name);
2314 }
2315 xlate_report(ctx, "input port is mirror output port, dropping");
2316 return;
2317 }
2318
2319 /* Check VLAN. */
2320 vid = vlan_tci_to_vid(flow->vlan_tci);
2321 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2322 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2323 return;
2324 }
2325 vlan = input_vid_to_vlan(in_xbundle, vid);
2326
2327 /* Check other admissibility requirements. */
2328 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2329 return;
2330 }
2331
2332 /* Learn source MAC. */
2333 if (ctx->xin->may_learn) {
2334 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2335 }
2336 if (ctx->xin->xcache) {
2337 struct xc_entry *entry;
2338
2339 /* Save enough info to update mac learning table later. */
2340 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2341 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2342 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2343 entry->u.normal.vlan = vlan;
2344 }
2345
2346 /* Determine output bundle. */
2347 if (mcast_snooping_enabled(ctx->xbridge->ms)
2348 && !eth_addr_is_broadcast(flow->dl_dst)
2349 && eth_addr_is_multicast(flow->dl_dst)
2350 && is_ip_any(flow)) {
2351 struct mcast_snooping *ms = ctx->xbridge->ms;
2352 struct mcast_group *grp = NULL;
2353
2354 if (is_igmp(flow)) {
2355 if (mcast_snooping_is_membership(flow->tp_src) ||
2356 mcast_snooping_is_query(flow->tp_src)) {
2357 if (ctx->xin->may_learn) {
2358 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2359 in_xbundle, ctx->xin->packet);
2360 }
2361 /*
2362 * IGMP packets need to take the slow path, in order to be
2363 * processed for mdb updates. That will prevent expires
2364 * firing off even after hosts have sent reports.
2365 */
2366 ctx->xout->slow |= SLOW_ACTION;
2367 }
2368
2369 if (mcast_snooping_is_membership(flow->tp_src)) {
2370 ovs_rwlock_rdlock(&ms->rwlock);
2371 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2372 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2373 * forward IGMP Membership Reports only to those ports where
2374 * multicast routers are attached. Alternatively stated: a
2375 * snooping switch should not forward IGMP Membership Reports
2376 * to ports on which only hosts are attached.
2377 * An administrative control may be provided to override this
2378 * restriction, allowing the report messages to be flooded to
2379 * other ports. */
2380 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2381 ovs_rwlock_unlock(&ms->rwlock);
2382 } else {
2383 xlate_report(ctx, "multicast traffic, flooding");
2384 xlate_normal_flood(ctx, in_xbundle, vlan);
2385 }
2386 return;
2387 } else if (is_mld(flow)) {
2388 ctx->xout->slow |= SLOW_ACTION;
2389 if (ctx->xin->may_learn) {
2390 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2391 in_xbundle, ctx->xin->packet);
2392 }
2393 if (is_mld_report(flow)) {
2394 ovs_rwlock_rdlock(&ms->rwlock);
2395 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2396 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2397 ovs_rwlock_unlock(&ms->rwlock);
2398 } else {
2399 xlate_report(ctx, "MLD query, flooding");
2400 xlate_normal_flood(ctx, in_xbundle, vlan);
2401 }
2402 } else {
2403 if ((flow->dl_type == htons(ETH_TYPE_IP)
2404 && ip_is_local_multicast(flow->nw_dst))
2405 || (flow->dl_type == htons(ETH_TYPE_IPV6)
2406 && ipv6_is_all_hosts(&flow->ipv6_dst))) {
2407 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2408 * address in the 224.0.0.x range which are not IGMP must
2409 * be forwarded on all ports */
2410 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2411 xlate_normal_flood(ctx, in_xbundle, vlan);
2412 return;
2413 }
2414 }
2415
2416 /* forwarding to group base ports */
2417 ovs_rwlock_rdlock(&ms->rwlock);
2418 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2419 grp = mcast_snooping_lookup4(ms, flow->nw_dst, vlan);
2420 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2421 grp = mcast_snooping_lookup(ms, &flow->ipv6_dst, vlan);
2422 }
2423 if (grp) {
2424 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2425 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2426 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2427 } else {
2428 if (mcast_snooping_flood_unreg(ms)) {
2429 xlate_report(ctx, "unregistered multicast, flooding");
2430 xlate_normal_flood(ctx, in_xbundle, vlan);
2431 } else {
2432 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2433 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2434 }
2435 }
2436 ovs_rwlock_unlock(&ms->rwlock);
2437 } else {
2438 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2439 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2440 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2441 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2442
2443 if (mac_port) {
2444 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2445 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2446 if (mac_xbundle && mac_xbundle != in_xbundle) {
2447 xlate_report(ctx, "forwarding to learned port");
2448 output_normal(ctx, mac_xbundle, vlan);
2449 } else if (!mac_xbundle) {
2450 xlate_report(ctx, "learned port is unknown, dropping");
2451 } else {
2452 xlate_report(ctx, "learned port is input port, dropping");
2453 }
2454 } else {
2455 xlate_report(ctx, "no learned MAC for destination, flooding");
2456 xlate_normal_flood(ctx, in_xbundle, vlan);
2457 }
2458 }
2459 }
2460
2461 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
2462 * 'probability' is the number of packets out of UINT32_MAX to sample. The
2463 * 'cookie' (of length 'cookie_size' bytes) is passed back in the callback for
2464 * each sampled packet. 'tunnel_out_port', if not ODPP_NONE, is added as the
2465 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions', an
2466 * OVS_USERSPACE_ATTR_ACTIONS attribute is added.
2467 */
2468 static size_t
2469 compose_sample_action(struct xlate_ctx *ctx,
2470 const uint32_t probability,
2471 const union user_action_cookie *cookie,
2472 const size_t cookie_size,
2473 const odp_port_t tunnel_out_port,
2474 bool include_actions)
2475 {
2476 size_t sample_offset = nl_msg_start_nested(ctx->odp_actions,
2477 OVS_ACTION_ATTR_SAMPLE);
2478
2479 nl_msg_put_u32(ctx->odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2480
2481 size_t actions_offset = nl_msg_start_nested(ctx->odp_actions,
2482 OVS_SAMPLE_ATTR_ACTIONS);
2483
2484 odp_port_t odp_port = ofp_port_to_odp_port(
2485 ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
2486 uint32_t pid = dpif_port_get_pid(ctx->xbridge->dpif, odp_port,
2487 flow_hash_5tuple(&ctx->xin->flow, 0));
2488 int cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2489 tunnel_out_port,
2490 include_actions,
2491 ctx->odp_actions);
2492
2493 nl_msg_end_nested(ctx->odp_actions, actions_offset);
2494 nl_msg_end_nested(ctx->odp_actions, sample_offset);
2495
2496 return cookie_offset;
2497 }
2498
2499 /* If sFLow is not enabled, returns 0 without doing anything.
2500 *
2501 * If sFlow is enabled, appends a template "sample" action to the ODP actions
2502 * in 'ctx'. This action is a template because some of the information needed
2503 * to fill it out is not available until flow translation is complete. In this
2504 * case, this functions returns an offset, which is always nonzero, to pass
2505 * later to fix_sflow_action() to fill in the rest of the template. */
2506 static size_t
2507 compose_sflow_action(struct xlate_ctx *ctx)
2508 {
2509 struct dpif_sflow *sflow = ctx->xbridge->sflow;
2510 if (!sflow || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2511 return 0;
2512 }
2513
2514 union user_action_cookie cookie = { .type = USER_ACTION_COOKIE_SFLOW };
2515 return compose_sample_action(ctx, dpif_sflow_get_probability(sflow),
2516 &cookie, sizeof cookie.sflow, ODPP_NONE,
2517 true);
2518 }
2519
2520 /* If IPFIX is enabled, this appends a "sample" action to implement IPFIX to
2521 * 'ctx->odp_actions'. */
2522 static void
2523 compose_ipfix_action(struct xlate_ctx *ctx, odp_port_t output_odp_port)
2524 {
2525 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
2526 odp_port_t tunnel_out_port = ODPP_NONE;
2527
2528 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2529 return;
2530 }
2531
2532 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2533 * number. */
2534 if (output_odp_port == ODPP_NONE &&
2535 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix)) {
2536 return;
2537 }
2538
2539 /* For output case, output_odp_port is valid*/
2540 if (output_odp_port != ODPP_NONE) {
2541 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix)) {
2542 return;
2543 }
2544 /* If tunnel sampling is enabled, put an additional option attribute:
2545 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2546 */
2547 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix) &&
2548 dpif_ipfix_get_tunnel_port(ipfix, output_odp_port) ) {
2549 tunnel_out_port = output_odp_port;
2550 }
2551 }
2552
2553 union user_action_cookie cookie = {
2554 .ipfix = {
2555 .type = USER_ACTION_COOKIE_IPFIX,
2556 .output_odp_port = output_odp_port,
2557 }
2558 };
2559 compose_sample_action(ctx,
2560 dpif_ipfix_get_bridge_exporter_probability(ipfix),
2561 &cookie, sizeof cookie.ipfix, tunnel_out_port,
2562 false);
2563 }
2564
2565 /* Fix "sample" action according to data collected while composing ODP actions,
2566 * as described in compose_sflow_action().
2567 *
2568 * 'user_cookie_offset' must be the offset returned by add_sflow_action(). */
2569 static void
2570 fix_sflow_action(struct xlate_ctx *ctx, unsigned int user_cookie_offset)
2571 {
2572 const struct flow *base = &ctx->base_flow;
2573 union user_action_cookie *cookie;
2574
2575 cookie = ofpbuf_at(ctx->odp_actions, user_cookie_offset,
2576 sizeof cookie->sflow);
2577 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2578
2579 cookie->type = USER_ACTION_COOKIE_SFLOW;
2580 cookie->sflow.vlan_tci = base->vlan_tci;
2581
2582 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2583 * port information") for the interpretation of cookie->output. */
2584 switch (ctx->sflow_n_outputs) {
2585 case 0:
2586 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2587 cookie->sflow.output = 0x40000000 | 256;
2588 break;
2589
2590 case 1:
2591 cookie->sflow.output = dpif_sflow_odp_port_to_ifindex(
2592 ctx->xbridge->sflow, ctx->sflow_odp_port);
2593 if (cookie->sflow.output) {
2594 break;
2595 }
2596 /* Fall through. */
2597 default:
2598 /* 0x80000000 means "multiple output ports. */
2599 cookie->sflow.output = 0x80000000 | ctx->sflow_n_outputs;
2600 break;
2601 }
2602 }
2603
2604 static bool
2605 process_special(struct xlate_ctx *ctx, const struct xport *xport)
2606 {
2607 const struct flow *flow = &ctx->xin->flow;
2608 struct flow_wildcards *wc = ctx->wc;
2609 const struct xbridge *xbridge = ctx->xbridge;
2610 const struct dp_packet *packet = ctx->xin->packet;
2611 enum slow_path_reason slow;
2612
2613 if (!xport) {
2614 slow = 0;
2615 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2616 if (packet) {
2617 cfm_process_heartbeat(xport->cfm, packet);
2618 }
2619 slow = SLOW_CFM;
2620 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2621 if (packet) {
2622 bfd_process_packet(xport->bfd, flow, packet);
2623 /* If POLL received, immediately sends FINAL back. */
2624 if (bfd_should_send_packet(xport->bfd)) {
2625 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2626 }
2627 }
2628 slow = SLOW_BFD;
2629 } else if (xport->xbundle && xport->xbundle->lacp
2630 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2631 if (packet) {
2632 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2633 }
2634 slow = SLOW_LACP;
2635 } else if ((xbridge->stp || xbridge->rstp) &&
2636 stp_should_process_flow(flow, wc)) {
2637 if (packet) {
2638 xbridge->stp
2639 ? stp_process_packet(xport, packet)
2640 : rstp_process_packet(xport, packet);
2641 }
2642 slow = SLOW_STP;
2643 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2644 if (packet) {
2645 lldp_process_packet(xport->lldp, packet);
2646 }
2647 slow = SLOW_LLDP;
2648 } else {
2649 slow = 0;
2650 }
2651
2652 if (slow) {
2653 ctx->xout->slow |= slow;
2654 return true;
2655 } else {
2656 return false;
2657 }
2658 }
2659
2660 static int
2661 tnl_route_lookup_flow(const struct flow *oflow,
2662 ovs_be32 *ip, struct xport **out_port)
2663 {
2664 char out_dev[IFNAMSIZ];
2665 struct xbridge *xbridge;
2666 struct xlate_cfg *xcfg;
2667 ovs_be32 gw;
2668
2669 if (!ovs_router_lookup4(oflow->tunnel.ip_dst, out_dev, &gw)) {
2670 return -ENOENT;
2671 }
2672
2673 if (gw) {
2674 *ip = gw;
2675 } else {
2676 *ip = oflow->tunnel.ip_dst;
2677 }
2678
2679 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2680 ovs_assert(xcfg);
2681
2682 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2683 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2684 struct xport *port;
2685
2686 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2687 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2688 *out_port = port;
2689 return 0;
2690 }
2691 }
2692 }
2693 }
2694 return -ENOENT;
2695 }
2696
2697 static int
2698 compose_table_xlate(struct xlate_ctx *ctx, const struct xport *out_dev,
2699 struct dp_packet *packet)
2700 {
2701 struct xbridge *xbridge = out_dev->xbridge;
2702 struct ofpact_output output;
2703 struct flow flow;
2704
2705 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2706 flow_extract(packet, &flow);
2707 flow.in_port.ofp_port = out_dev->ofp_port;
2708 output.port = OFPP_TABLE;
2709 output.max_len = 0;
2710
2711 return ofproto_dpif_execute_actions__(xbridge->ofproto, &flow, NULL,
2712 &output.ofpact, sizeof output,
2713 ctx->recurse, ctx->resubmits, packet);
2714 }
2715
2716 static void
2717 tnl_send_arp_request(struct xlate_ctx *ctx, const struct xport *out_dev,
2718 const struct eth_addr eth_src,
2719 ovs_be32 ip_src, ovs_be32 ip_dst)
2720 {
2721 struct dp_packet packet;
2722
2723 dp_packet_init(&packet, 0);
2724 compose_arp(&packet, ARP_OP_REQUEST,
2725 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2726
2727 compose_table_xlate(ctx, out_dev, &packet);
2728 dp_packet_uninit(&packet);
2729 }
2730
2731 static int
2732 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2733 const struct flow *flow, odp_port_t tunnel_odp_port)
2734 {
2735 struct ovs_action_push_tnl tnl_push_data;
2736 struct xport *out_dev = NULL;
2737 ovs_be32 s_ip, d_ip = 0;
2738 struct eth_addr smac;
2739 struct eth_addr dmac;
2740 int err;
2741
2742 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2743 if (err) {
2744 xlate_report(ctx, "native tunnel routing failed");
2745 return err;
2746 }
2747 xlate_report(ctx, "tunneling to "IP_FMT" via %s",
2748 IP_ARGS(d_ip), netdev_get_name(out_dev->netdev));
2749
2750 /* Use mac addr of bridge port of the peer. */
2751 err = netdev_get_etheraddr(out_dev->netdev, &smac);
2752 if (err) {
2753 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2754 return err;
2755 }
2756
2757 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2758 if (err) {
2759 xlate_report(ctx, "tunnel output device lacks IPv4 address");
2760 return err;
2761 }
2762
2763 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, &dmac);
2764 if (err) {
2765 xlate_report(ctx, "ARP cache miss for "IP_FMT" on bridge %s, "
2766 "sending ARP request",
2767 IP_ARGS(d_ip), out_dev->xbridge->name);
2768 tnl_send_arp_request(ctx, out_dev, smac, s_ip, d_ip);
2769 return err;
2770 }
2771 if (ctx->xin->xcache) {
2772 struct xc_entry *entry;
2773
2774 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2775 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2776 sizeof entry->u.tnl_arp_cache.br_name);
2777 entry->u.tnl_arp_cache.d_ip = d_ip;
2778 }
2779
2780 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" "IP_FMT
2781 " to "ETH_ADDR_FMT" "IP_FMT,
2782 ETH_ADDR_ARGS(smac), IP_ARGS(s_ip),
2783 ETH_ADDR_ARGS(dmac), IP_ARGS(d_ip));
2784 err = tnl_port_build_header(xport->ofport, flow,
2785 dmac, smac, s_ip, &tnl_push_data);
2786 if (err) {
2787 return err;
2788 }
2789 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2790 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2791 odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
2792 return 0;
2793 }
2794
2795 static void
2796 xlate_commit_actions(struct xlate_ctx *ctx)
2797 {
2798 bool use_masked = ctx->xbridge->support.masked_set_action;
2799
2800 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
2801 ctx->odp_actions, ctx->wc,
2802 use_masked);
2803 }
2804
2805 static void
2806 clear_conntrack(struct flow *flow)
2807 {
2808 flow->ct_state = 0;
2809 flow->ct_zone = 0;
2810 }
2811
2812 static void
2813 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2814 const struct xlate_bond_recirc *xr, bool check_stp)
2815 {
2816 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2817 struct flow_wildcards *wc = ctx->wc;
2818 struct flow *flow = &ctx->xin->flow;
2819 struct flow_tnl flow_tnl;
2820 ovs_be16 flow_vlan_tci;
2821 uint32_t flow_pkt_mark;
2822 uint8_t flow_nw_tos;
2823 odp_port_t out_port, odp_port;
2824 bool tnl_push_pop_send = false;
2825 uint8_t dscp;
2826
2827 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2828 * before traversing a patch port. */
2829 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 34);
2830 memset(&flow_tnl, 0, sizeof flow_tnl);
2831
2832 if (!xport) {
2833 xlate_report(ctx, "Nonexistent output port");
2834 return;
2835 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2836 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2837 return;
2838 } else if (check_stp) {
2839 if (is_stp(&ctx->base_flow)) {
2840 if (!xport_stp_should_forward_bpdu(xport) &&
2841 !xport_rstp_should_manage_bpdu(xport)) {
2842 if (ctx->xbridge->stp != NULL) {
2843 xlate_report(ctx, "STP not in listening state, "
2844 "skipping bpdu output");
2845 } else if (ctx->xbridge->rstp != NULL) {
2846 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2847 "skipping bpdu output");
2848 }
2849 return;
2850 }
2851 } else if (!xport_stp_forward_state(xport) ||
2852 !xport_rstp_forward_state(xport)) {
2853 if (ctx->xbridge->stp != NULL) {
2854 xlate_report(ctx, "STP not in forwarding state, "
2855 "skipping output");
2856 } else if (ctx->xbridge->rstp != NULL) {
2857 xlate_report(ctx, "RSTP not in forwarding state, "
2858 "skipping output");
2859 }
2860 return;
2861 }
2862 }
2863
2864 if (xport->peer) {
2865 const struct xport *peer = xport->peer;
2866 struct flow old_flow = ctx->xin->flow;
2867 bool old_conntrack = ctx->conntracked;
2868 bool old_was_mpls = ctx->was_mpls;
2869 cls_version_t old_version = ctx->tables_version;
2870 struct ofpbuf old_stack = ctx->stack;
2871 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2872 struct ofpbuf old_action_set = ctx->action_set;
2873 uint64_t actset_stub[1024 / 8];
2874
2875 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2876 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2877 ctx->xbridge = peer->xbridge;
2878 flow->in_port.ofp_port = peer->ofp_port;
2879 flow->metadata = htonll(0);
2880 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2881 memset(flow->regs, 0, sizeof flow->regs);
2882 flow->actset_output = OFPP_UNSET;
2883 ctx->conntracked = false;
2884 clear_conntrack(flow);
2885
2886 /* The bridge is now known so obtain its table version. */
2887 ctx->tables_version
2888 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
2889
2890 if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
2891 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2892 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2893 if (ctx->action_set.size) {
2894 /* Translate action set only if not dropping the packet and
2895 * not recirculating. */
2896 if (!exit_recirculates(ctx)) {
2897 xlate_action_set(ctx);
2898 }
2899 }
2900 /* Check if need to recirculate. */
2901 if (exit_recirculates(ctx)) {
2902 compose_recirculate_action(ctx);
2903 }
2904 } else {
2905 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2906 * the learning action look at the packet, then drop it. */
2907 struct flow old_base_flow = ctx->base_flow;
2908 size_t old_size = ctx->odp_actions->size;
2909 mirror_mask_t old_mirrors = ctx->mirrors;
2910
2911 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2912 ctx->mirrors = old_mirrors;
2913 ctx->base_flow = old_base_flow;
2914 ctx->odp_actions->size = old_size;
2915
2916 /* Undo changes that may have been done for recirculation. */
2917 if (exit_recirculates(ctx)) {
2918 ctx->action_set.size = ctx->recirc_action_offset;
2919 ctx->recirc_action_offset = -1;
2920 ctx->last_unroll_offset = -1;
2921 }
2922 }
2923 }
2924
2925 ctx->xin->flow = old_flow;
2926 ctx->xbridge = xport->xbridge;
2927 ofpbuf_uninit(&ctx->action_set);
2928 ctx->action_set = old_action_set;
2929 ofpbuf_uninit(&ctx->stack);
2930 ctx->stack = old_stack;
2931
2932 /* Restore calling bridge's lookup version. */
2933 ctx->tables_version = old_version;
2934
2935 /* The peer bridge popping MPLS should have no effect on the original
2936 * bridge. */
2937 ctx->was_mpls = old_was_mpls;
2938
2939 /* The peer bridge's conntrack execution should have no effect on the
2940 * original bridge. */
2941 ctx->conntracked = old_conntrack;
2942
2943 /* The fact that the peer bridge exits (for any reason) does not mean
2944 * that the original bridge should exit. Specifically, if the peer
2945 * bridge recirculates (which typically modifies the packet), the
2946 * original bridge must continue processing with the original, not the
2947 * recirculated packet! */
2948 ctx->exit = false;
2949
2950 if (ctx->xin->resubmit_stats) {
2951 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2952 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2953 if (peer->bfd) {
2954 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2955 }
2956 }
2957 if (ctx->xin->xcache) {
2958 struct xc_entry *entry;
2959
2960 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2961 entry->u.dev.tx = netdev_ref(xport->netdev);
2962 entry->u.dev.rx = netdev_ref(peer->netdev);
2963 entry->u.dev.bfd = bfd_ref(peer->bfd);
2964 }
2965 return;
2966 }
2967
2968 flow_vlan_tci = flow->vlan_tci;
2969 flow_pkt_mark = flow->pkt_mark;
2970 flow_nw_tos = flow->nw_tos;
2971
2972 if (count_skb_priorities(xport)) {
2973 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2974 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2975 wc->masks.nw_tos |= IP_DSCP_MASK;
2976 flow->nw_tos &= ~IP_DSCP_MASK;
2977 flow->nw_tos |= dscp;
2978 }
2979 }
2980
2981 if (xport->is_tunnel) {
2982 /* Save tunnel metadata so that changes made due to
2983 * the Logical (tunnel) Port are not visible for any further
2984 * matches, while explicit set actions on tunnel metadata are.
2985 */
2986 flow_tnl = flow->tunnel;
2987 odp_port = tnl_port_send(xport->ofport, flow, ctx->wc);
2988 if (odp_port == ODPP_NONE) {
2989 xlate_report(ctx, "Tunneling decided against output");
2990 goto out; /* restore flow_nw_tos */
2991 }
2992 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2993 xlate_report(ctx, "Not tunneling to our own address");
2994 goto out; /* restore flow_nw_tos */
2995 }
2996 if (ctx->xin->resubmit_stats) {
2997 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2998 }
2999 if (ctx->xin->xcache) {
3000 struct xc_entry *entry;
3001
3002 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
3003 entry->u.dev.tx = netdev_ref(xport->netdev);
3004 }
3005 out_port = odp_port;
3006 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3007 xlate_report(ctx, "output to native tunnel");
3008 tnl_push_pop_send = true;
3009 } else {
3010 xlate_report(ctx, "output to kernel tunnel");
3011 commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions);
3012 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3013 }
3014 } else {
3015 odp_port = xport->odp_port;
3016 out_port = odp_port;
3017 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
3018 ofp_port_t vlandev_port;
3019
3020 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
3021 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
3022 ofp_port, flow->vlan_tci);
3023 if (vlandev_port != ofp_port) {
3024 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
3025 flow->vlan_tci = htons(0);
3026 }
3027 }
3028 }
3029
3030 if (out_port != ODPP_NONE) {
3031 xlate_commit_actions(ctx);
3032
3033 if (xr) {
3034 struct ovs_action_hash *act_hash;
3035
3036 /* Hash action. */
3037 act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
3038 OVS_ACTION_ATTR_HASH,
3039 sizeof *act_hash);
3040 act_hash->hash_alg = xr->hash_alg;
3041 act_hash->hash_basis = xr->hash_basis;
3042
3043 /* Recirc action. */
3044 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
3045 xr->recirc_id);
3046 } else {
3047
3048 if (tnl_push_pop_send) {
3049 build_tunnel_send(ctx, xport, flow, odp_port);
3050 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3051 } else {
3052 odp_port_t odp_tnl_port = ODPP_NONE;
3053
3054 /* XXX: Write better Filter for tunnel port. We can use inport
3055 * int tunnel-port flow to avoid these checks completely. */
3056 if (ofp_port == OFPP_LOCAL &&
3057 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3058
3059 odp_tnl_port = tnl_port_map_lookup(flow, wc);
3060 }
3061
3062 if (odp_tnl_port != ODPP_NONE) {
3063 nl_msg_put_odp_port(ctx->odp_actions,
3064 OVS_ACTION_ATTR_TUNNEL_POP,
3065 odp_tnl_port);
3066 } else {
3067 /* Tunnel push-pop action is not compatible with
3068 * IPFIX action. */
3069 compose_ipfix_action(ctx, out_port);
3070 nl_msg_put_odp_port(ctx->odp_actions,
3071 OVS_ACTION_ATTR_OUTPUT,
3072 out_port);
3073 }
3074 }
3075 }
3076
3077 ctx->sflow_odp_port = odp_port;
3078 ctx->sflow_n_outputs++;
3079 ctx->nf_output_iface = ofp_port;
3080 }
3081
3082 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
3083 mirror_packet(ctx, xport->xbundle,
3084 xbundle_mirror_dst(xport->xbundle->xbridge,
3085 xport->xbundle));
3086 }
3087
3088 out:
3089 /* Restore flow */
3090 flow->vlan_tci = flow_vlan_tci;
3091 flow->pkt_mark = flow_pkt_mark;
3092 flow->nw_tos = flow_nw_tos;
3093 }
3094
3095 static void
3096 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3097 const struct xlate_bond_recirc *xr)
3098 {
3099 compose_output_action__(ctx, ofp_port, xr, true);
3100 }
3101
3102 static void
3103 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3104 {
3105 struct rule_dpif *old_rule = ctx->rule;
3106 ovs_be64 old_cookie = ctx->rule_cookie;
3107 const struct rule_actions *actions;
3108
3109 if (ctx->xin->resubmit_stats) {
3110 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3111 }
3112
3113 ctx->resubmits++;
3114 ctx->recurse++;
3115 ctx->rule = rule;
3116 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3117 actions = rule_dpif_get_actions(rule);
3118 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3119 ctx->rule_cookie = old_cookie;
3120 ctx->rule = old_rule;
3121 ctx->recurse--;
3122 }
3123
3124 static bool
3125 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3126 {
3127 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3128
3129 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3130 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
3131 MAX_RESUBMIT_RECURSION);
3132 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3133 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
3134 } else if (ctx->odp_actions->size > UINT16_MAX) {
3135 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
3136 } else if (ctx->stack.size >= 65536) {
3137 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
3138 } else {
3139 return true;
3140 }
3141
3142 return false;
3143 }
3144
3145 static void
3146 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3147 bool may_packet_in, bool honor_table_miss)
3148 {
3149 /* Check if we need to recirculate before matching in a table. */
3150 if (ctx->was_mpls) {
3151 ctx_trigger_recirculation(ctx);
3152 return;
3153 }
3154 if (xlate_resubmit_resource_check(ctx)) {
3155 uint8_t old_table_id = ctx->table_id;
3156 struct rule_dpif *rule;
3157
3158 ctx->table_id = table_id;
3159
3160 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3161 ctx->tables_version,
3162 &ctx->xin->flow, ctx->xin->wc,
3163 ctx->xin->resubmit_stats,
3164 &ctx->table_id, in_port,
3165 may_packet_in, honor_table_miss);
3166
3167 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3168 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3169 }
3170
3171 if (rule) {
3172 /* Fill in the cache entry here instead of xlate_recursively
3173 * to make the reference counting more explicit. We take a
3174 * reference in the lookups above if we are going to cache the
3175 * rule. */
3176 if (ctx->xin->xcache) {
3177 struct xc_entry *entry;
3178
3179 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3180 entry->u.rule = rule;
3181 rule_dpif_ref(rule);
3182 }
3183 xlate_recursively(ctx, rule);
3184 }
3185
3186 ctx->table_id = old_table_id;
3187 return;
3188 }
3189
3190 ctx->exit = true;
3191 }
3192
3193 static void
3194 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3195 struct ofputil_bucket *bucket)
3196 {
3197 if (ctx->xin->resubmit_stats) {
3198 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3199 }
3200 if (ctx->xin->xcache) {
3201 struct xc_entry *entry;
3202
3203 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3204 entry->u.group.group = group_dpif_ref(group);
3205 entry->u.group.bucket = bucket;
3206 }
3207 }
3208
3209 static void
3210 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3211 {
3212 uint64_t action_list_stub[1024 / 8];
3213 struct ofpbuf action_list, action_set;
3214 struct flow old_flow = ctx->xin->flow;
3215 bool old_was_mpls = ctx->was_mpls;
3216
3217 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3218 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3219
3220 ofpacts_execute_action_set(&action_list, &action_set);
3221 ctx->recurse++;
3222 do_xlate_actions(action_list.data, action_list.size, ctx);
3223 ctx->recurse--;
3224
3225 ofpbuf_uninit(&action_set);
3226 ofpbuf_uninit(&action_list);
3227
3228 /* Check if need to recirculate. */
3229 if (exit_recirculates(ctx)) {
3230 compose_recirculate_action(ctx);
3231 }
3232
3233 /* Roll back flow to previous state.
3234 * This is equivalent to cloning the packet for each bucket.
3235 *
3236 * As a side effect any subsequently applied actions will
3237 * also effectively be applied to a clone of the packet taken
3238 * just before applying the all or indirect group.
3239 *
3240 * Note that group buckets are action sets, hence they cannot modify the
3241 * main action set. Also any stack actions are ignored when executing an
3242 * action set, so group buckets cannot change the stack either.
3243 * However, we do allow resubmit actions in group buckets, which could
3244 * break the above assumptions. It is up to the controller to not mess up
3245 * with the action_set and stack in the tables resubmitted to from
3246 * group buckets. */
3247 ctx->xin->flow = old_flow;
3248
3249 /* The group bucket popping MPLS should have no effect after bucket
3250 * execution. */
3251 ctx->was_mpls = old_was_mpls;
3252
3253 /* The fact that the group bucket exits (for any reason) does not mean that
3254 * the translation after the group action should exit. Specifically, if
3255 * the group bucket recirculates (which typically modifies the packet), the
3256 * actions after the group action must continue processing with the
3257 * original, not the recirculated packet! */
3258 ctx->exit = false;
3259 }
3260
3261 static void
3262 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3263 {
3264 struct ofputil_bucket *bucket;
3265 const struct ovs_list *buckets;
3266
3267 group_dpif_get_buckets(group, &buckets);
3268
3269 LIST_FOR_EACH (bucket, list_node, buckets) {
3270 xlate_group_bucket(ctx, bucket);
3271 }
3272 xlate_group_stats(ctx, group, NULL);
3273 }
3274
3275 static void
3276 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3277 {
3278 struct ofputil_bucket *bucket;
3279
3280 bucket = group_first_live_bucket(ctx, group, 0);
3281 if (bucket) {
3282 xlate_group_bucket(ctx, bucket);
3283 xlate_group_stats(ctx, group, bucket);
3284 }
3285 }
3286
3287 static void
3288 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3289 {
3290 struct flow_wildcards *wc = ctx->wc;
3291 struct ofputil_bucket *bucket;
3292 uint32_t basis;
3293
3294 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3295 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3296 bucket = group_best_live_bucket(ctx, group, basis);
3297 if (bucket) {
3298 xlate_group_bucket(ctx, bucket);
3299 xlate_group_stats(ctx, group, bucket);
3300 }
3301 }
3302
3303 static void
3304 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3305 {
3306 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3307 const struct field_array *fields;
3308 struct ofputil_bucket *bucket;
3309 uint32_t basis;
3310 int i;
3311
3312 fields = group_dpif_get_fields(group);
3313 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3314
3315 /* Determine which fields to hash */
3316 for (i = 0; i < MFF_N_IDS; i++) {
3317 if (bitmap_is_set(fields->used.bm, i)) {
3318 const struct mf_field *mf;
3319
3320 /* If the field is already present in 'hash_fields' then
3321 * this loop has already checked that it and its pre-requisites
3322 * are present in the flow and its pre-requisites have
3323 * already been added to 'hash_fields'. There is nothing more
3324 * to do here and as an optimisation the loop can continue. */
3325 if (bitmap_is_set(hash_fields.bm, i)) {
3326 continue;
3327 }
3328
3329 mf = mf_from_id(i);
3330
3331 /* Only hash a field if it and its pre-requisites are present
3332 * in the flow. */
3333 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3334 continue;
3335 }
3336
3337 /* Hash both the field and its pre-requisites */
3338 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3339 }
3340 }
3341
3342 /* Hash the fields */
3343 for (i = 0; i < MFF_N_IDS; i++) {
3344 if (bitmap_is_set(hash_fields.bm, i)) {
3345 const struct mf_field *mf = mf_from_id(i);
3346 union mf_value value;
3347 int j;
3348
3349 mf_get_value(mf, &ctx->xin->flow, &value);
3350 /* This seems inefficient but so does apply_mask() */
3351 for (j = 0; j < mf->n_bytes; j++) {
3352 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3353 }
3354 basis = hash_bytes(&value, mf->n_bytes, basis);
3355
3356 /* For tunnels, hash in whether the field is present. */
3357 if (mf_is_tun_metadata(mf)) {
3358 basis = hash_boolean(mf_is_set(mf, &ctx->xin->flow), basis);
3359 }
3360
3361 mf_mask_field(mf, &ctx->wc->masks);
3362 }
3363 }
3364
3365 bucket = group_best_live_bucket(ctx, group, basis);
3366 if (bucket) {
3367 xlate_group_bucket(ctx, bucket);
3368 xlate_group_stats(ctx, group, bucket);
3369 }
3370 }
3371
3372 static void
3373 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3374 {
3375 const char *selection_method = group_dpif_get_selection_method(group);
3376
3377 if (selection_method[0] == '\0') {
3378 xlate_default_select_group(ctx, group);
3379 } else if (!strcasecmp("hash", selection_method)) {
3380 xlate_hash_fields_select_group(ctx, group);
3381 } else {
3382 /* Parsing of groups should ensure this never happens */
3383 OVS_NOT_REACHED();
3384 }
3385 }
3386
3387 static void
3388 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3389 {
3390 bool was_in_group = ctx->in_group;
3391 ctx->in_group = true;
3392
3393 switch (group_dpif_get_type(group)) {
3394 case OFPGT11_ALL:
3395 case OFPGT11_INDIRECT:
3396 xlate_all_group(ctx, group);
3397 break;
3398 case OFPGT11_SELECT:
3399 xlate_select_group(ctx, group);
3400 break;
3401 case OFPGT11_FF:
3402 xlate_ff_group(ctx, group);
3403 break;
3404 default:
3405 OVS_NOT_REACHED();
3406 }
3407 group_dpif_unref(group);
3408
3409 ctx->in_group = was_in_group;
3410 }
3411
3412 static bool
3413 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3414 {
3415 if (xlate_resubmit_resource_check(ctx)) {
3416 struct group_dpif *group;
3417 bool got_group;
3418
3419 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3420 if (got_group) {
3421 xlate_group_action__(ctx, group);
3422 } else {
3423 return true;
3424 }
3425 }
3426
3427 return false;
3428 }
3429
3430 static void
3431 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3432 const struct ofpact_resubmit *resubmit)
3433 {
3434 ofp_port_t in_port;
3435 uint8_t table_id;
3436 bool may_packet_in = false;
3437 bool honor_table_miss = false;
3438
3439 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3440 /* Still allow missed packets to be sent to the controller
3441 * if resubmitting from an internal table. */
3442 may_packet_in = true;
3443 honor_table_miss = true;
3444 }
3445
3446 in_port = resubmit->in_port;
3447 if (in_port == OFPP_IN_PORT) {
3448 in_port = ctx->xin->flow.in_port.ofp_port;
3449 }
3450
3451 table_id = resubmit->table_id;
3452 if (table_id == 255) {
3453 table_id = ctx->table_id;
3454 }
3455
3456 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3457 honor_table_miss);
3458 }
3459
3460 static void
3461 flood_packets(struct xlate_ctx *ctx, bool all)
3462 {
3463 const struct xport *xport;
3464
3465 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3466 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3467 continue;
3468 }
3469
3470 if (all) {
3471 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3472 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3473 compose_output_action(ctx, xport->ofp_port, NULL);
3474 }
3475 }
3476
3477 ctx->nf_output_iface = NF_OUT_FLOOD;
3478 }
3479
3480 static void
3481 execute_controller_action(struct xlate_ctx *ctx, int len,
3482 enum ofp_packet_in_reason reason,
3483 uint16_t controller_id)
3484 {
3485 struct ofproto_packet_in *pin;
3486 struct dp_packet *packet;
3487
3488 ctx->xout->slow |= SLOW_CONTROLLER;
3489 if (!ctx->xin->packet) {
3490 return;
3491 }
3492
3493 packet = dp_packet_clone(ctx->xin->packet);
3494
3495 xlate_commit_actions(ctx);
3496
3497 odp_execute_actions(NULL, &packet, 1, false,
3498 ctx->odp_actions->data, ctx->odp_actions->size, NULL);
3499
3500 pin = xmalloc(sizeof *pin);
3501 pin->up.packet_len = dp_packet_size(packet);
3502 pin->up.packet = dp_packet_steal_data(packet);
3503 pin->up.reason = reason;
3504 pin->up.table_id = ctx->table_id;
3505 pin->up.cookie = ctx->rule_cookie;
3506
3507 flow_get_metadata(&ctx->xin->flow, &pin->up.flow_metadata);
3508
3509 pin->controller_id = controller_id;
3510 pin->send_len = len;
3511 /* If a rule is a table-miss rule then this is
3512 * a table-miss handled by a table-miss rule.
3513 *
3514 * Else, if rule is internal and has a controller action,
3515 * the later being implied by the rule being processed here,
3516 * then this is a table-miss handled without a table-miss rule.
3517 *
3518 * Otherwise this is not a table-miss. */
3519 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3520 if (ctx->rule) {
3521 if (rule_dpif_is_table_miss(ctx->rule)) {
3522 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3523 } else if (rule_dpif_is_internal(ctx->rule)) {
3524 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3525 }
3526 }
3527 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3528 dp_packet_delete(packet);
3529 }
3530
3531 static void
3532 compose_recirculate_action__(struct xlate_ctx *ctx, uint8_t table)
3533 {
3534 struct recirc_metadata md;
3535 uint32_t id;
3536
3537 recirc_metadata_from_flow(&md, &ctx->xin->flow);
3538
3539 ovs_assert(ctx->recirc_action_offset >= 0);
3540
3541 struct recirc_state state = {
3542 .table_id = table,
3543 .ofproto = ctx->xbridge->ofproto,
3544 .metadata = md,
3545 .stack = &ctx->stack,
3546 .mirrors = ctx->mirrors,
3547 .conntracked = ctx->conntracked,
3548 .action_set_len = ctx->recirc_action_offset,
3549 .ofpacts_len = ctx->action_set.size,
3550 .ofpacts = ctx->action_set.data,
3551 };
3552
3553 /* Only allocate recirculation ID if we have a packet. */
3554 if (ctx->xin->packet) {
3555 /* Allocate a unique recirc id for the given metadata state in the
3556 * flow. The life-cycle of this recirc id is managed by associating it
3557 * with the udpif key ('ukey') created for each new datapath flow. */
3558 id = recirc_alloc_id_ctx(&state);
3559 if (!id) {
3560 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3561 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3562 return;
3563 }
3564 xlate_out_add_recirc(ctx->xout, id);
3565 } else {
3566 /* Look up an existing recirc id for the given metadata state in the
3567 * flow. No new reference is taken, as the ID is RCU protected and is
3568 * only required temporarily for verification.
3569 *
3570 * This might fail and return 0. We let zero 'id' to be used in the
3571 * RECIRC action below, which will fail all revalidations as zero is
3572 * not a valid recirculation ID. */
3573 id = recirc_find_id(&state);
3574 }
3575
3576 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3577
3578 /* Undo changes done by recirculation. */
3579 ctx->action_set.size = ctx->recirc_action_offset;
3580 ctx->recirc_action_offset = -1;
3581 ctx->last_unroll_offset = -1;
3582 }
3583
3584 /* Called only when ctx->recirc_action_offset is set. */
3585 static void
3586 compose_recirculate_action(struct xlate_ctx *ctx)
3587 {
3588 xlate_commit_actions(ctx);
3589 compose_recirculate_action__(ctx, 0);
3590 }
3591
3592 static void
3593 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3594 {
3595 struct flow *flow = &ctx->xin->flow;
3596 int n;
3597
3598 ovs_assert(eth_type_mpls(mpls->ethertype));
3599
3600 n = flow_count_mpls_labels(flow, ctx->wc);
3601 if (!n) {
3602 xlate_commit_actions(ctx);
3603 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3604 if (ctx->xin->packet != NULL) {
3605 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3606 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3607 "MPLS push action can't be performed as it would "
3608 "have more MPLS LSEs than the %d supported.",
3609 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3610 }
3611 ctx->exit = true;
3612 return;
3613 }
3614
3615 flow_push_mpls(flow, n, mpls->ethertype, ctx->wc);
3616 }
3617
3618 static void
3619 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3620 {
3621 struct flow *flow = &ctx->xin->flow;
3622 int n = flow_count_mpls_labels(flow, ctx->wc);
3623
3624 if (flow_pop_mpls(flow, n, eth_type, ctx->wc)) {
3625 if (ctx->xbridge->support.odp.recirc) {
3626 ctx->was_mpls = true;
3627 }
3628 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3629 if (ctx->xin->packet != NULL) {
3630 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3631 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3632 "MPLS pop action can't be performed as it has "
3633 "more MPLS LSEs than the %d supported.",
3634 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3635 }
3636 ctx->exit = true;
3637 ofpbuf_clear(ctx->odp_actions);
3638 }
3639 }
3640
3641 static bool
3642 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3643 {
3644 struct flow *flow = &ctx->xin->flow;
3645
3646 if (!is_ip_any(flow)) {
3647 return false;
3648 }
3649
3650 ctx->wc->masks.nw_ttl = 0xff;
3651 if (flow->nw_ttl > 1) {
3652 flow->nw_ttl--;
3653 return false;
3654 } else {
3655 size_t i;
3656
3657 for (i = 0; i < ids->n_controllers; i++) {
3658 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3659 ids->cnt_ids[i]);
3660 }
3661
3662 /* Stop processing for current table. */
3663 return true;
3664 }
3665 }
3666
3667 static void
3668 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3669 {
3670 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3671 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3672 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3673 }
3674 }
3675
3676 static void
3677 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3678 {
3679 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3680 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3681 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3682 }
3683 }
3684
3685 static void
3686 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3687 {
3688 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3689 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3690 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3691 }
3692 }
3693
3694 static bool
3695 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3696 {
3697 struct flow *flow = &ctx->xin->flow;
3698
3699 if (eth_type_mpls(flow->dl_type)) {
3700 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3701
3702 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3703 if (ttl > 1) {
3704 ttl--;
3705 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3706 return false;
3707 } else {
3708 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3709 }
3710 }
3711
3712 /* Stop processing for current table. */
3713 return true;
3714 }
3715
3716 static void
3717 xlate_output_action(struct xlate_ctx *ctx,
3718 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3719 {
3720 ofp_port_t prev_nf_output_iface = ctx->nf_output_iface;
3721
3722 ctx->nf_output_iface = NF_OUT_DROP;
3723
3724 switch (port) {
3725 case OFPP_IN_PORT:
3726 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3727 break;
3728 case OFPP_TABLE:
3729 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3730 0, may_packet_in, true);
3731 break;
3732 case OFPP_NORMAL:
3733 xlate_normal(ctx);
3734 break;
3735 case OFPP_FLOOD:
3736 flood_packets(ctx, false);
3737 break;
3738 case OFPP_ALL:
3739 flood_packets(ctx, true);
3740 break;
3741 case OFPP_CONTROLLER:
3742 execute_controller_action(ctx, max_len,
3743 (ctx->in_group ? OFPR_GROUP
3744 : ctx->in_action_set ? OFPR_ACTION_SET
3745 : OFPR_ACTION),
3746 0);
3747 break;
3748 case OFPP_NONE:
3749 break;
3750 case OFPP_LOCAL:
3751 default:
3752 if (port != ctx->xin->flow.in_port.ofp_port) {
3753 compose_output_action(ctx, port, NULL);
3754 } else {
3755 xlate_report(ctx, "skipping output to input port");
3756 }
3757 break;
3758 }
3759
3760 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3761 ctx->nf_output_iface = NF_OUT_FLOOD;
3762 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
3763 ctx->nf_output_iface = prev_nf_output_iface;
3764 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3765 ctx->nf_output_iface != NF_OUT_FLOOD) {
3766 ctx->nf_output_iface = NF_OUT_MULTI;
3767 }
3768 }
3769
3770 static void
3771 xlate_output_reg_action(struct xlate_ctx *ctx,
3772 const struct ofpact_output_reg *or)
3773 {
3774 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3775 if (port <= UINT16_MAX) {
3776 union mf_subvalue value;
3777
3778 memset(&value, 0xff, sizeof value);
3779 mf_write_subfield_flow(&or->src, &value, &ctx->wc->masks);
3780 xlate_output_action(ctx, u16_to_ofp(port),
3781 or->max_len, false);
3782 }
3783 }
3784
3785 static void
3786 xlate_enqueue_action(struct xlate_ctx *ctx,
3787 const struct ofpact_enqueue *enqueue)
3788 {
3789 ofp_port_t ofp_port = enqueue->port;
3790 uint32_t queue_id = enqueue->queue;
3791 uint32_t flow_priority, priority;
3792 int error;
3793
3794 /* Translate queue to priority. */
3795 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3796 if (error) {
3797 /* Fall back to ordinary output action. */
3798 xlate_output_action(ctx, enqueue->port, 0, false);
3799 return;
3800 }
3801
3802 /* Check output port. */
3803 if (ofp_port == OFPP_IN_PORT) {
3804 ofp_port = ctx->xin->flow.in_port.ofp_port;
3805 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3806 return;
3807 }
3808
3809 /* Add datapath actions. */
3810 flow_priority = ctx->xin->flow.skb_priority;
3811 ctx->xin->flow.skb_priority = priority;
3812 compose_output_action(ctx, ofp_port, NULL);
3813 ctx->xin->flow.skb_priority = flow_priority;
3814
3815 /* Update NetFlow output port. */
3816 if (ctx->nf_output_iface == NF_OUT_DROP) {
3817 ctx->nf_output_iface = ofp_port;
3818 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
3819 ctx->nf_output_iface = NF_OUT_MULTI;
3820 }
3821 }
3822
3823 static void
3824 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3825 {
3826 uint32_t skb_priority;
3827
3828 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3829 ctx->xin->flow.skb_priority = skb_priority;
3830 } else {
3831 /* Couldn't translate queue to a priority. Nothing to do. A warning
3832 * has already been logged. */
3833 }
3834 }
3835
3836 static bool
3837 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3838 {
3839 const struct xbridge *xbridge = xbridge_;
3840 struct xport *port;
3841
3842 switch (ofp_port) {
3843 case OFPP_IN_PORT:
3844 case OFPP_TABLE:
3845 case OFPP_NORMAL:
3846 case OFPP_FLOOD:
3847 case OFPP_ALL:
3848 case OFPP_NONE:
3849 return true;
3850 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3851 return false;
3852 default:
3853 port = get_ofp_port(xbridge, ofp_port);
3854 return port ? port->may_enable : false;
3855 }
3856 }
3857
3858 static void
3859 xlate_bundle_action(struct xlate_ctx *ctx,
3860 const struct ofpact_bundle *bundle)
3861 {
3862 ofp_port_t port;
3863
3864 port = bundle_execute(bundle, &ctx->xin->flow, ctx->wc, slave_enabled_cb,
3865 CONST_CAST(struct xbridge *, ctx->xbridge));
3866 if (bundle->dst.field) {
3867 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow, ctx->wc);
3868 } else {
3869 xlate_output_action(ctx, port, 0, false);
3870 }
3871 }
3872
3873 static void
3874 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3875 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3876 {
3877 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3878 if (ctx->xin->may_learn) {
3879 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3880 }
3881 }
3882
3883 static void
3884 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3885 {
3886 learn_mask(learn, ctx->wc);
3887
3888 if (ctx->xin->xcache) {
3889 struct xc_entry *entry;
3890
3891 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3892 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3893 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3894 entry->u.learn.ofpacts = ofpbuf_new(64);
3895 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3896 entry->u.learn.ofpacts);
3897 } else if (ctx->xin->may_learn) {
3898 uint64_t ofpacts_stub[1024 / 8];
3899 struct ofputil_flow_mod fm;
3900 struct ofpbuf ofpacts;
3901
3902 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3903 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3904 ofpbuf_uninit(&ofpacts);
3905 }
3906 }
3907
3908 static void
3909 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3910 uint16_t idle_timeout, uint16_t hard_timeout)
3911 {
3912 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3913 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3914 }
3915 }
3916
3917 static void
3918 xlate_fin_timeout(struct xlate_ctx *ctx,
3919 const struct ofpact_fin_timeout *oft)
3920 {
3921 if (ctx->rule) {
3922 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3923 oft->fin_idle_timeout, oft->fin_hard_timeout);
3924 if (ctx->xin->xcache) {
3925 struct xc_entry *entry;
3926
3927 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3928 /* XC_RULE already holds a reference on the rule, none is taken
3929 * here. */
3930 entry->u.fin.rule = ctx->rule;
3931 entry->u.fin.idle = oft->fin_idle_timeout;
3932 entry->u.fin.hard = oft->fin_hard_timeout;
3933 }
3934 }
3935 }
3936
3937 static void
3938 xlate_sample_action(struct xlate_ctx *ctx,
3939 const struct ofpact_sample *os)
3940 {
3941 /* Scale the probability from 16-bit to 32-bit while representing
3942 * the same percentage. */
3943 uint32_t probability = (os->probability << 16) | os->probability;
3944
3945 if (!ctx->xbridge->support.variable_length_userdata) {
3946 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3947
3948 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3949 "lacks support (needs Linux 3.10+ or kernel module from "
3950 "OVS 1.11+)");
3951 return;
3952 }
3953
3954 xlate_commit_actions(ctx);
3955
3956 union user_action_cookie cookie = {
3957 .flow_sample = {
3958 .type = USER_ACTION_COOKIE_FLOW_SAMPLE,
3959 .probability = os->probability,
3960 .collector_set_id = os->collector_set_id,
3961 .obs_domain_id = os->obs_domain_id,
3962 .obs_point_id = os->obs_point_id,
3963 }
3964 };
3965 compose_sample_action(ctx, probability, &cookie, sizeof cookie.flow_sample,
3966 ODPP_NONE, false);
3967 }
3968
3969 static bool
3970 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3971 {
3972 if (xport->config & (is_stp(&ctx->xin->flow)
3973 ? OFPUTIL_PC_NO_RECV_STP
3974 : OFPUTIL_PC_NO_RECV)) {
3975 return false;
3976 }
3977
3978 /* Only drop packets here if both forwarding and learning are
3979 * disabled. If just learning is enabled, we need to have
3980 * OFPP_NORMAL and the learning action have a look at the packet
3981 * before we can drop it. */
3982 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3983 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3984 return false;
3985 }
3986
3987 return true;
3988 }
3989
3990 static void
3991 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3992 {
3993 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3994 size_t on_len = ofpact_nest_get_action_len(on);
3995 const struct ofpact *inner;
3996
3997 /* Maintain actset_output depending on the contents of the action set:
3998 *
3999 * - OFPP_UNSET, if there is no "output" action.
4000 *
4001 * - The output port, if there is an "output" action and no "group"
4002 * action.
4003 *
4004 * - OFPP_UNSET, if there is a "group" action.
4005 */
4006 if (!ctx->action_set_has_group) {
4007 OFPACT_FOR_EACH (inner, on->actions, on_len) {
4008 if (inner->type == OFPACT_OUTPUT) {
4009 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
4010 } else if (inner->type == OFPACT_GROUP) {
4011 ctx->xin->flow.actset_output = OFPP_UNSET;
4012 ctx->action_set_has_group = true;
4013 break;
4014 }
4015 }
4016 }
4017
4018 ofpbuf_put(&ctx->action_set, on->actions, on_len);
4019 ofpact_pad(&ctx->action_set);
4020 }
4021
4022 static void
4023 xlate_action_set(struct xlate_ctx *ctx)
4024 {
4025 uint64_t action_list_stub[1024 / 64];
4026 struct ofpbuf action_list;
4027
4028 ctx->in_action_set = true;
4029 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
4030 ofpacts_execute_action_set(&action_list, &ctx->action_set);
4031 /* Clear the action set, as it is not needed any more. */
4032 ofpbuf_clear(&ctx->action_set);
4033 do_xlate_actions(action_list.data, action_list.size, ctx);
4034 ctx->in_action_set = false;
4035 ofpbuf_uninit(&action_list);
4036 }
4037
4038 static void
4039 recirc_put_unroll_xlate(struct xlate_ctx *ctx)
4040 {
4041 struct ofpact_unroll_xlate *unroll;
4042
4043 unroll = ctx->last_unroll_offset < 0
4044 ? NULL
4045 : ALIGNED_CAST(struct ofpact_unroll_xlate *,
4046 (char *)ctx->action_set.data + ctx->last_unroll_offset);
4047
4048 /* Restore the table_id and rule cookie for a potential PACKET
4049 * IN if needed. */
4050 if (!unroll ||
4051 (ctx->table_id != unroll->rule_table_id
4052 || ctx->rule_cookie != unroll->rule_cookie)) {
4053
4054 ctx->last_unroll_offset = ctx->action_set.size;
4055 unroll = ofpact_put_UNROLL_XLATE(&ctx->action_set);
4056 unroll->rule_table_id = ctx->table_id;
4057 unroll->rule_cookie = ctx->rule_cookie;
4058 }
4059 }
4060
4061
4062 /* Copy remaining actions to the action_set to be executed after recirculation.
4063 * UNROLL_XLATE action is inserted, if not already done so, before actions that
4064 * may generate PACKET_INs from the current table and without matching another
4065 * rule. */
4066 static void
4067 recirc_unroll_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4068 struct xlate_ctx *ctx)
4069 {
4070 const struct ofpact *a;
4071
4072 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4073 switch (a->type) {
4074 /* May generate PACKET INs. */
4075 case OFPACT_OUTPUT_REG:
4076 case OFPACT_GROUP:
4077 case OFPACT_OUTPUT:
4078 case OFPACT_CONTROLLER:
4079 case OFPACT_DEC_MPLS_TTL:
4080 case OFPACT_DEC_TTL:
4081 recirc_put_unroll_xlate(ctx);
4082 break;
4083
4084 /* These may not generate PACKET INs. */
4085 case OFPACT_SET_TUNNEL:
4086 case OFPACT_REG_MOVE:
4087 case OFPACT_SET_FIELD:
4088 case OFPACT_STACK_PUSH:
4089 case OFPACT_STACK_POP:
4090 case OFPACT_LEARN:
4091 case OFPACT_WRITE_METADATA:
4092 case OFPACT_RESUBMIT: /* May indirectly generate PACKET INs, */
4093 case OFPACT_GOTO_TABLE: /* but from a different table and rule. */
4094 case OFPACT_ENQUEUE:
4095 case OFPACT_SET_VLAN_VID:
4096 case OFPACT_SET_VLAN_PCP:
4097 case OFPACT_STRIP_VLAN:
4098 case OFPACT_PUSH_VLAN:
4099 case OFPACT_SET_ETH_SRC:
4100 case OFPACT_SET_ETH_DST:
4101 case OFPACT_SET_IPV4_SRC:
4102 case OFPACT_SET_IPV4_DST:
4103 case OFPACT_SET_IP_DSCP:
4104 case OFPACT_SET_IP_ECN:
4105 case OFPACT_SET_IP_TTL:
4106 case OFPACT_SET_L4_SRC_PORT:
4107 case OFPACT_SET_L4_DST_PORT:
4108 case OFPACT_SET_QUEUE:
4109 case OFPACT_POP_QUEUE:
4110 case OFPACT_PUSH_MPLS:
4111 case OFPACT_POP_MPLS:
4112 case OFPACT_SET_MPLS_LABEL:
4113 case OFPACT_SET_MPLS_TC:
4114 case OFPACT_SET_MPLS_TTL:
4115 case OFPACT_MULTIPATH:
4116 case OFPACT_BUNDLE:
4117 case OFPACT_EXIT:
4118 case OFPACT_UNROLL_XLATE:
4119 case OFPACT_FIN_TIMEOUT:
4120 case OFPACT_CLEAR_ACTIONS:
4121 case OFPACT_WRITE_ACTIONS:
4122 case OFPACT_METER:
4123 case OFPACT_SAMPLE:
4124 case OFPACT_DEBUG_RECIRC:
4125 case OFPACT_CT:
4126 break;
4127
4128 /* These need not be copied for restoration. */
4129 case OFPACT_NOTE:
4130 case OFPACT_CONJUNCTION:
4131 continue;
4132 }
4133 /* Copy the action over. */
4134 ofpbuf_put(&ctx->action_set, a, OFPACT_ALIGN(a->len));
4135 }
4136 }
4137
4138 #define CHECK_MPLS_RECIRCULATION() \
4139 if (ctx->was_mpls) { \
4140 ctx_trigger_recirculation(ctx); \
4141 break; \
4142 }
4143 #define CHECK_MPLS_RECIRCULATION_IF(COND) \
4144 if (COND) { \
4145 CHECK_MPLS_RECIRCULATION(); \
4146 }
4147
4148 static void
4149 compose_conntrack_action(struct xlate_ctx *ctx, struct ofpact_conntrack *ofc)
4150 {
4151 size_t ct_offset;
4152 uint16_t zone;
4153
4154 /* Ensure that any prior actions are applied before composing the new
4155 * conntrack action. */
4156 xlate_commit_actions(ctx);
4157
4158 if (ofc->zone_src.field) {
4159 zone = mf_get_subfield(&ofc->zone_src, &ctx->xin->flow);
4160 } else {
4161 zone = ofc->zone_imm;
4162 }
4163
4164 ct_offset = nl_msg_start_nested(ctx->odp_actions, OVS_ACTION_ATTR_CT);
4165 if (ofc->flags & NX_CT_F_COMMIT) {
4166 nl_msg_put_flag(ctx->odp_actions, OVS_CT_ATTR_COMMIT);
4167 }
4168 nl_msg_put_u16(ctx->odp_actions, OVS_CT_ATTR_ZONE, zone);
4169 nl_msg_end_nested(ctx->odp_actions, ct_offset);
4170
4171 if (ofc->recirc_table == NX_CT_RECIRC_NONE) {
4172 /* If we do not recirculate as part of this action, hide the results of
4173 * connection tracking from subsequent recirculations. */
4174 ctx->conntracked = false;
4175 } else {
4176 /* Use ct_* fields from datapath during recirculation upcall. */
4177 ctx->conntracked = true;
4178 ctx_trigger_recirculation(ctx);
4179 compose_recirculate_action__(ctx, ofc->recirc_table);
4180 }
4181 }
4182
4183 static void
4184 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4185 struct xlate_ctx *ctx)
4186 {
4187 struct flow_wildcards *wc = ctx->wc;
4188 struct flow *flow = &ctx->xin->flow;
4189 const struct ofpact *a;
4190
4191 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4192 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
4193 tnl_nd_snoop(flow, wc, ctx->xbridge->name);
4194 }
4195 /* dl_type already in the mask, not set below. */
4196
4197 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4198 struct ofpact_controller *controller;
4199 const struct ofpact_metadata *metadata;
4200 const struct ofpact_set_field *set_field;
4201 const struct mf_field *mf;
4202
4203 if (ctx->exit) {
4204 /* Check if need to store the remaining actions for later
4205 * execution. */
4206 if (exit_recirculates(ctx)) {
4207 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4208 ((uint8_t *)a -
4209 (uint8_t *)ofpacts)),
4210 ctx);
4211 }
4212 break;
4213 }
4214
4215 switch (a->type) {
4216 case OFPACT_OUTPUT:
4217 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4218 ofpact_get_OUTPUT(a)->max_len, true);
4219 break;
4220
4221 case OFPACT_GROUP:
4222 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4223 /* Group could not be found. */
4224 return;
4225 }
4226 break;
4227
4228 case OFPACT_CONTROLLER:
4229 controller = ofpact_get_CONTROLLER(a);
4230 execute_controller_action(ctx, controller->max_len,
4231 controller->reason,
4232 controller->controller_id);
4233 break;
4234
4235 case OFPACT_ENQUEUE:
4236 memset(&wc->masks.skb_priority, 0xff,
4237 sizeof wc->masks.skb_priority);
4238 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4239 break;
4240
4241 case OFPACT_SET_VLAN_VID:
4242 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4243 if (flow->vlan_tci & htons(VLAN_CFI) ||
4244 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4245 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4246 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4247 | htons(VLAN_CFI));
4248 }
4249 break;
4250
4251 case OFPACT_SET_VLAN_PCP:
4252 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4253 if (flow->vlan_tci & htons(VLAN_CFI) ||
4254 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4255 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4256 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4257 << VLAN_PCP_SHIFT) | VLAN_CFI);
4258 }
4259 break;
4260
4261 case OFPACT_STRIP_VLAN:
4262 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4263 flow->vlan_tci = htons(0);
4264 break;
4265
4266 case OFPACT_PUSH_VLAN:
4267 /* XXX 802.1AD(QinQ) */
4268 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4269 flow->vlan_tci = htons(VLAN_CFI);
4270 break;
4271
4272 case OFPACT_SET_ETH_SRC:
4273 WC_MASK_FIELD(wc, dl_src);
4274 flow->dl_src = ofpact_get_SET_ETH_SRC(a)->mac;
4275 break;
4276
4277 case OFPACT_SET_ETH_DST:
4278 WC_MASK_FIELD(wc, dl_dst);
4279 flow->dl_dst = ofpact_get_SET_ETH_DST(a)->mac;
4280 break;
4281
4282 case OFPACT_SET_IPV4_SRC:
4283 CHECK_MPLS_RECIRCULATION();
4284 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4285 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4286 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4287 }
4288 break;
4289
4290 case OFPACT_SET_IPV4_DST:
4291 CHECK_MPLS_RECIRCULATION();
4292 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4293 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4294 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4295 }
4296 break;
4297
4298 case OFPACT_SET_IP_DSCP:
4299 CHECK_MPLS_RECIRCULATION();
4300 if (is_ip_any(flow)) {
4301 wc->masks.nw_tos |= IP_DSCP_MASK;
4302 flow->nw_tos &= ~IP_DSCP_MASK;
4303 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4304 }
4305 break;
4306
4307 case OFPACT_SET_IP_ECN:
4308 CHECK_MPLS_RECIRCULATION();
4309 if (is_ip_any(flow)) {
4310 wc->masks.nw_tos |= IP_ECN_MASK;
4311 flow->nw_tos &= ~IP_ECN_MASK;
4312 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4313 }
4314 break;
4315
4316 case OFPACT_SET_IP_TTL:
4317 CHECK_MPLS_RECIRCULATION();
4318 if (is_ip_any(flow)) {
4319 wc->masks.nw_ttl = 0xff;
4320 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4321 }
4322 break;
4323
4324 case OFPACT_SET_L4_SRC_PORT:
4325 CHECK_MPLS_RECIRCULATION();
4326 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4327 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4328 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4329 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4330 }
4331 break;
4332
4333 case OFPACT_SET_L4_DST_PORT:
4334 CHECK_MPLS_RECIRCULATION();
4335 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4336 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4337 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4338 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4339 }
4340 break;
4341
4342 case OFPACT_RESUBMIT:
4343 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4344 break;
4345
4346 case OFPACT_SET_TUNNEL:
4347 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4348 break;
4349
4350 case OFPACT_SET_QUEUE:
4351 memset(&wc->masks.skb_priority, 0xff,
4352 sizeof wc->masks.skb_priority);
4353 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4354 break;
4355
4356 case OFPACT_POP_QUEUE:
4357 memset(&wc->masks.skb_priority, 0xff,
4358 sizeof wc->masks.skb_priority);
4359 flow->skb_priority = ctx->orig_skb_priority;
4360 break;
4361
4362 case OFPACT_REG_MOVE:
4363 CHECK_MPLS_RECIRCULATION_IF(
4364 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4365 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4366 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4367 break;
4368
4369 case OFPACT_SET_FIELD:
4370 CHECK_MPLS_RECIRCULATION_IF(
4371 mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field));
4372 set_field = ofpact_get_SET_FIELD(a);
4373 mf = set_field->field;
4374
4375 /* Set field action only ever overwrites packet's outermost
4376 * applicable header fields. Do nothing if no header exists. */
4377 if (mf->id == MFF_VLAN_VID) {
4378 wc->masks.vlan_tci |= htons(VLAN_CFI);
4379 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4380 break;
4381 }
4382 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4383 /* 'dl_type' is already unwildcarded. */
4384 && !eth_type_mpls(flow->dl_type)) {
4385 break;
4386 }
4387 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4388 * header field on a packet that does not have them. */
4389 mf_mask_field_and_prereqs(mf, wc);
4390 if (mf_are_prereqs_ok(mf, flow)) {
4391 mf_set_flow_value_masked(mf, &set_field->value,
4392 &set_field->mask, flow);
4393 }
4394 break;
4395
4396 case OFPACT_STACK_PUSH:
4397 CHECK_MPLS_RECIRCULATION_IF(
4398 mf_is_l3_or_higher(ofpact_get_STACK_PUSH(a)->subfield.field));
4399 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4400 &ctx->stack);
4401 break;
4402
4403 case OFPACT_STACK_POP:
4404 CHECK_MPLS_RECIRCULATION_IF(
4405 mf_is_l3_or_higher(ofpact_get_STACK_POP(a)->subfield.field));
4406 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4407 &ctx->stack);
4408 break;
4409
4410 case OFPACT_PUSH_MPLS:
4411 /* Recirculate if it is an IP packet with a zero ttl. This may
4412 * indicate that the packet was previously MPLS and an MPLS pop
4413 * action converted it to IP. In this case recirculating should
4414 * reveal the IP TTL which is used as the basis for a new MPLS
4415 * LSE. */
4416 CHECK_MPLS_RECIRCULATION_IF(
4417 !flow_count_mpls_labels(flow, wc)
4418 && flow->nw_ttl == 0
4419 && is_ip_any(flow));
4420 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4421 break;
4422
4423 case OFPACT_POP_MPLS:
4424 CHECK_MPLS_RECIRCULATION();
4425 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4426 break;
4427
4428 case OFPACT_SET_MPLS_LABEL:
4429 CHECK_MPLS_RECIRCULATION();
4430 compose_set_mpls_label_action(
4431 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4432 break;
4433
4434 case OFPACT_SET_MPLS_TC:
4435 CHECK_MPLS_RECIRCULATION();
4436 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4437 break;
4438
4439 case OFPACT_SET_MPLS_TTL:
4440 CHECK_MPLS_RECIRCULATION();
4441 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4442 break;
4443
4444 case OFPACT_DEC_MPLS_TTL:
4445 CHECK_MPLS_RECIRCULATION();
4446 if (compose_dec_mpls_ttl_action(ctx)) {
4447 return;
4448 }
4449 break;
4450
4451 case OFPACT_DEC_TTL:
4452 CHECK_MPLS_RECIRCULATION();
4453 wc->masks.nw_ttl = 0xff;
4454 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4455 return;
4456 }
4457 break;
4458
4459 case OFPACT_NOTE:
4460 /* Nothing to do. */
4461 break;
4462
4463 case OFPACT_MULTIPATH:
4464 CHECK_MPLS_RECIRCULATION();
4465 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4466 break;
4467
4468 case OFPACT_BUNDLE:
4469 CHECK_MPLS_RECIRCULATION();
4470 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4471 break;
4472
4473 case OFPACT_OUTPUT_REG:
4474 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4475 break;
4476
4477 case OFPACT_LEARN:
4478 CHECK_MPLS_RECIRCULATION();
4479 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4480 break;
4481
4482 case OFPACT_CONJUNCTION: {
4483 /* A flow with a "conjunction" action represents part of a special
4484 * kind of "set membership match". Such a flow should not actually
4485 * get executed, but it could via, say, a "packet-out", even though
4486 * that wouldn't be useful. Log it to help debugging. */
4487 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4488 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4489 break;
4490 }
4491
4492 case OFPACT_EXIT:
4493 ctx->exit = true;
4494 break;
4495
4496 case OFPACT_UNROLL_XLATE: {
4497 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4498
4499 /* Restore translation context data that was stored earlier. */
4500 ctx->table_id = unroll->rule_table_id;
4501 ctx->rule_cookie = unroll->rule_cookie;
4502 break;
4503 }
4504 case OFPACT_FIN_TIMEOUT:
4505 CHECK_MPLS_RECIRCULATION();
4506 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4507 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4508 break;
4509
4510 case OFPACT_CLEAR_ACTIONS:
4511 ofpbuf_clear(&ctx->action_set);
4512 ctx->xin->flow.actset_output = OFPP_UNSET;
4513 ctx->action_set_has_group = false;
4514 break;
4515
4516 case OFPACT_WRITE_ACTIONS:
4517 xlate_write_actions(ctx, a);
4518 break;
4519
4520 case OFPACT_WRITE_METADATA:
4521 metadata = ofpact_get_WRITE_METADATA(a);
4522 flow->metadata &= ~metadata->mask;
4523 flow->metadata |= metadata->metadata & metadata->mask;
4524 break;
4525
4526 case OFPACT_METER:
4527 /* Not implemented yet. */
4528 break;
4529
4530 case OFPACT_GOTO_TABLE: {
4531 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4532
4533 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4534 * than ogt->table_id. This is to allow goto_table actions that
4535 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4536 * after recirculation. */
4537 ovs_assert(ctx->table_id == TBL_INTERNAL
4538 || ctx->table_id < ogt->table_id);
4539 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4540 ogt->table_id, true, true);
4541 break;
4542 }
4543
4544 case OFPACT_SAMPLE:
4545 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4546 break;
4547
4548 case OFPACT_CT:
4549 CHECK_MPLS_RECIRCULATION();
4550 compose_conntrack_action(ctx, ofpact_get_CT(a));
4551 break;
4552
4553 case OFPACT_DEBUG_RECIRC:
4554 ctx_trigger_recirculation(ctx);
4555 a = ofpact_next(a);
4556 break;
4557 }
4558
4559 /* Check if need to store this and the remaining actions for later
4560 * execution. */
4561 if (ctx->exit && ctx_first_recirculation_action(ctx)) {
4562 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4563 ((uint8_t *)a -
4564 (uint8_t *)ofpacts)),
4565 ctx);
4566 break;
4567 }
4568 }
4569 }
4570
4571 void
4572 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4573 const struct flow *flow, ofp_port_t in_port,
4574 struct rule_dpif *rule, uint16_t tcp_flags,
4575 const struct dp_packet *packet, struct flow_wildcards *wc,
4576 struct ofpbuf *odp_actions)
4577 {
4578 xin->ofproto = ofproto;
4579 xin->flow = *flow;
4580 xin->flow.in_port.ofp_port = in_port;
4581 xin->flow.actset_output = OFPP_UNSET;
4582 xin->packet = packet;
4583 xin->may_learn = packet != NULL;
4584 xin->rule = rule;
4585 xin->xcache = NULL;
4586 xin->ofpacts = NULL;
4587 xin->ofpacts_len = 0;
4588 xin->tcp_flags = tcp_flags;
4589 xin->resubmit_hook = NULL;
4590 xin->report_hook = NULL;
4591 xin->resubmit_stats = NULL;
4592 xin->recurse = 0;
4593 xin->resubmits = 0;
4594 xin->wc = wc;
4595 xin->odp_actions = odp_actions;
4596
4597 /* Do recirc lookup. */
4598 xin->recirc = flow->recirc_id
4599 ? recirc_id_node_find(flow->recirc_id)
4600 : NULL;
4601 }
4602
4603 void
4604 xlate_out_uninit(struct xlate_out *xout)
4605 {
4606 if (xout) {
4607 xlate_out_free_recircs(xout);
4608 }
4609 }
4610
4611 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4612 * into datapath actions, using 'ctx', and discards the datapath actions. */
4613 void
4614 xlate_actions_for_side_effects(struct xlate_in *xin)
4615 {
4616 struct xlate_out xout;
4617
4618 xlate_actions(xin, &xout);
4619 xlate_out_uninit(&xout);
4620 }
4621 \f
4622 static struct skb_priority_to_dscp *
4623 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4624 {
4625 struct skb_priority_to_dscp *pdscp;
4626 uint32_t hash;
4627
4628 hash = hash_int(skb_priority, 0);
4629 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4630 if (pdscp->skb_priority == skb_priority) {
4631 return pdscp;
4632 }
4633 }
4634 return NULL;
4635 }
4636
4637 static bool
4638 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4639 uint8_t *dscp)
4640 {
4641 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4642 *dscp = pdscp ? pdscp->dscp : 0;
4643 return pdscp != NULL;
4644 }
4645
4646 static size_t
4647 count_skb_priorities(const struct xport *xport)
4648 {
4649 return hmap_count(&xport->skb_priorities);
4650 }
4651
4652 static void
4653 clear_skb_priorities(struct xport *xport)
4654 {
4655 struct skb_priority_to_dscp *pdscp, *next;
4656
4657 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4658 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4659 free(pdscp);
4660 }
4661 }
4662
4663 static bool
4664 actions_output_to_local_port(const struct xlate_ctx *ctx)
4665 {
4666 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4667 const struct nlattr *a;
4668 unsigned int left;
4669
4670 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->odp_actions->data,
4671 ctx->odp_actions->size) {
4672 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4673 && nl_attr_get_odp_port(a) == local_odp_port) {
4674 return true;
4675 }
4676 }
4677 return false;
4678 }
4679
4680 #if defined(__linux__)
4681 /* Returns the maximum number of packets that the Linux kernel is willing to
4682 * queue up internally to certain kinds of software-implemented ports, or the
4683 * default (and rarely modified) value if it cannot be determined. */
4684 static int
4685 netdev_max_backlog(void)
4686 {
4687 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4688 static int max_backlog = 1000; /* The normal default value. */
4689
4690 if (ovsthread_once_start(&once)) {
4691 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4692 FILE *stream;
4693 int n;
4694
4695 stream = fopen(filename, "r");
4696 if (!stream) {
4697 VLOG_INFO("%s: open failed (%s)", filename, ovs_strerror(errno));
4698 } else {
4699 if (fscanf(stream, "%d", &n) != 1) {
4700 VLOG_WARN("%s: read error", filename);
4701 } else if (n <= 100) {
4702 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4703 } else {
4704 max_backlog = n;
4705 }
4706 fclose(stream);
4707 }
4708 ovsthread_once_done(&once);
4709
4710 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4711 }
4712
4713 return max_backlog;
4714 }
4715
4716 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4717 * 'odp_actions'. */
4718 static int
4719 count_output_actions(const struct ofpbuf *odp_actions)
4720 {
4721 const struct nlattr *a;
4722 size_t left;
4723 int n = 0;
4724
4725 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4726 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4727 n++;
4728 }
4729 }
4730 return n;
4731 }
4732 #endif /* defined(__linux__) */
4733
4734 /* Returns true if 'odp_actions' contains more output actions than the datapath
4735 * can reliably handle in one go. On Linux, this is the value of the
4736 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4737 * packets that the kernel is willing to queue up for processing while the
4738 * datapath is processing a set of actions. */
4739 static bool
4740 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4741 {
4742 #ifdef __linux__
4743 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4744 && count_output_actions(odp_actions) > netdev_max_backlog());
4745 #else
4746 /* OSes other than Linux might have similar limits, but we don't know how
4747 * to determine them.*/
4748 return false;
4749 #endif
4750 }
4751
4752 static void
4753 xlate_wc_init(struct xlate_ctx *ctx)
4754 {
4755 flow_wildcards_init_catchall(ctx->wc);
4756
4757 /* Some fields we consider to always be examined. */
4758 WC_MASK_FIELD(ctx->wc, in_port);
4759 WC_MASK_FIELD(ctx->wc, dl_type);
4760 if (is_ip_any(&ctx->xin->flow)) {
4761 WC_MASK_FIELD_MASK(ctx->wc, nw_frag, FLOW_NW_FRAG_MASK);
4762 }
4763
4764 if (ctx->xbridge->support.odp.recirc) {
4765 /* Always exactly match recirc_id when datapath supports
4766 * recirculation. */
4767 WC_MASK_FIELD(ctx->wc, recirc_id);
4768 }
4769
4770 if (ctx->xbridge->netflow) {
4771 netflow_mask_wc(&ctx->xin->flow, ctx->wc);
4772 }
4773
4774 tnl_wc_init(&ctx->xin->flow, ctx->wc);
4775 }
4776
4777 static void
4778 xlate_wc_finish(struct xlate_ctx *ctx)
4779 {
4780 /* Clear the metadata and register wildcard masks, because we won't
4781 * use non-header fields as part of the cache. */
4782 flow_wildcards_clear_non_packet_fields(ctx->wc);
4783
4784 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
4785 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
4786 * represent these fields. The datapath interface, on the other hand,
4787 * represents them with just 8 bits each. This means that if the high
4788 * 8 bits of the masks for these fields somehow become set, then they
4789 * will get chopped off by a round trip through the datapath, and
4790 * revalidation will spot that as an inconsistency and delete the flow.
4791 * Avoid the problem here by making sure that only the low 8 bits of
4792 * either field can be unwildcarded for ICMP.
4793 */
4794 if (is_icmpv4(&ctx->xin->flow) || is_icmpv6(&ctx->xin->flow)) {
4795 ctx->wc->masks.tp_src &= htons(UINT8_MAX);
4796 ctx->wc->masks.tp_dst &= htons(UINT8_MAX);
4797 }
4798 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
4799 if (ctx->wc->masks.vlan_tci) {
4800 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI);
4801 }
4802 }
4803
4804 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
4805 * 'xout'.
4806 * The caller must take responsibility for eventually freeing 'xout', with
4807 * xlate_out_uninit(). */
4808 void
4809 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4810 {
4811 *xout = (struct xlate_out) {
4812 .slow = 0,
4813 .fail_open = false,
4814 .n_recircs = 0,
4815 };
4816
4817 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4818 struct xbridge *xbridge = xbridge_lookup(xcfg, xin->ofproto);
4819 if (!xbridge) {
4820 return;
4821 }
4822
4823 struct flow *flow = &xin->flow;
4824
4825 union mf_subvalue stack_stub[1024 / sizeof(union mf_subvalue)];
4826 uint64_t action_set_stub[1024 / 8];
4827 struct flow_wildcards scratch_wc;
4828 uint64_t actions_stub[256 / 8];
4829 struct ofpbuf scratch_actions = OFPBUF_STUB_INITIALIZER(actions_stub);
4830 struct xlate_ctx ctx = {
4831 .xin = xin,
4832 .xout = xout,
4833 .base_flow = *flow,
4834 .orig_tunnel_ip_dst = flow->tunnel.ip_dst,
4835 .xbridge = xbridge,
4836 .stack = OFPBUF_STUB_INITIALIZER(stack_stub),
4837 .rule = xin->rule,
4838 .wc = xin->wc ? xin->wc : &scratch_wc,
4839 .odp_actions = xin->odp_actions ? xin->odp_actions : &scratch_actions,
4840
4841 .recurse = xin->recurse,
4842 .resubmits = xin->resubmits,
4843 .in_group = false,
4844 .in_action_set = false,
4845
4846 .table_id = 0,
4847 .rule_cookie = OVS_BE64_MAX,
4848 .orig_skb_priority = flow->skb_priority,
4849 .sflow_n_outputs = 0,
4850 .sflow_odp_port = 0,
4851 .nf_output_iface = NF_OUT_DROP,
4852 .exit = false,
4853 .mirrors = 0,
4854
4855 .recirc_action_offset = -1,
4856 .last_unroll_offset = -1,
4857
4858 .was_mpls = false,
4859 .conntracked = false,
4860
4861 .action_set_has_group = false,
4862 .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
4863 };
4864
4865 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
4866 * the packet as the datapath will treat it for output actions:
4867 *
4868 * - Our datapath doesn't retain tunneling information without us
4869 * re-setting it, so clear the tunnel data.
4870 *
4871 * - For VLAN splinters, a higher layer may pretend that the packet
4872 * came in on 'flow->in_port.ofp_port' with 'flow->vlan_tci'
4873 * attached, because that's how we want to treat it from an OpenFlow
4874 * perspective. But from the datapath's perspective it actually came
4875 * in on a VLAN device without any VLAN attached. So here we put the
4876 * datapath's view of the VLAN information in 'base_flow' to ensure
4877 * correct treatment.
4878 */
4879 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4880 if (flow->in_port.ofp_port
4881 != vsp_realdev_to_vlandev(xbridge->ofproto,
4882 flow->in_port.ofp_port,
4883 flow->vlan_tci)) {
4884 ctx.base_flow.vlan_tci = 0;
4885 }
4886
4887 ofpbuf_reserve(ctx.odp_actions, NL_A_U32_SIZE);
4888 if (xin->wc) {
4889 xlate_wc_init(&ctx);
4890 }
4891
4892 COVERAGE_INC(xlate_actions);
4893
4894 if (xin->recirc) {
4895 const struct recirc_state *state = &xin->recirc->state;
4896
4897 xlate_report(&ctx, "Restoring state post-recirculation:");
4898
4899 if (xin->ofpacts_len > 0 || ctx.rule) {
4900 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4901 const char *conflict = xin->ofpacts_len ? "actions" : "rule";
4902
4903 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!", conflict);
4904 xlate_report(&ctx, "- Recirculation conflict (%s)!", conflict);
4905 goto exit;
4906 }
4907
4908 /* Set the bridge for post-recirculation processing if needed. */
4909 if (ctx.xbridge->ofproto != state->ofproto) {
4910 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4911 const struct xbridge *new_bridge
4912 = xbridge_lookup(xcfg, state->ofproto);
4913
4914 if (OVS_UNLIKELY(!new_bridge)) {
4915 /* Drop the packet if the bridge cannot be found. */
4916 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4917 VLOG_WARN_RL(&rl, "Recirculation bridge no longer exists.");
4918 xlate_report(&ctx, "- Recirculation bridge no longer exists.");
4919 goto exit;
4920 }
4921 ctx.xbridge = new_bridge;
4922 }
4923
4924 /* Set the post-recirculation table id. Note: A table lookup is done
4925 * only if there are no post-recirculation actions. */
4926 ctx.table_id = state->table_id;
4927 xlate_report(&ctx, "- Resuming from table %"PRIu8, ctx.table_id);
4928
4929 if (!state->conntracked) {
4930 clear_conntrack(flow);
4931 }
4932
4933 /* Restore pipeline metadata. May change flow's in_port and other
4934 * metadata to the values that existed when recirculation was
4935 * triggered. */
4936 recirc_metadata_to_flow(&state->metadata, flow);
4937
4938 /* Restore stack, if any. */
4939 if (state->stack) {
4940 ofpbuf_put(&ctx.stack, state->stack->data, state->stack->size);
4941 }
4942
4943 /* Restore mirror state. */
4944 ctx.mirrors = state->mirrors;
4945
4946 /* Restore action set, if any. */
4947 if (state->action_set_len) {
4948 const struct ofpact *a;
4949
4950 xlate_report_actions(&ctx, "- Restoring action set",
4951 state->ofpacts, state->action_set_len);
4952
4953 ofpbuf_put(&ctx.action_set, state->ofpacts, state->action_set_len);
4954
4955 OFPACT_FOR_EACH(a, state->ofpacts, state->action_set_len) {
4956 if (a->type == OFPACT_GROUP) {
4957 ctx.action_set_has_group = true;
4958 break;
4959 }
4960 }
4961 }
4962
4963 /* Restore recirculation actions. If there are no actions, processing
4964 * will start with a lookup in the table set above. */
4965 if (state->ofpacts_len > state->action_set_len) {
4966 xin->ofpacts_len = state->ofpacts_len - state->action_set_len;
4967 xin->ofpacts = state->ofpacts +
4968 state->action_set_len / sizeof *state->ofpacts;
4969
4970 xlate_report_actions(&ctx, "- Restoring actions",
4971 xin->ofpacts, xin->ofpacts_len);
4972 }
4973 } else if (OVS_UNLIKELY(flow->recirc_id)) {
4974 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4975
4976 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
4977 flow->recirc_id);
4978 goto exit;
4979 }
4980 /* The bridge is now known so obtain its table version. */
4981 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
4982
4983 if (!xin->ofpacts && !ctx.rule) {
4984 ctx.rule = rule_dpif_lookup_from_table(
4985 ctx.xbridge->ofproto, ctx.tables_version, flow, xin->wc,
4986 ctx.xin->resubmit_stats, &ctx.table_id,
4987 flow->in_port.ofp_port, true, true);
4988 if (ctx.xin->resubmit_stats) {
4989 rule_dpif_credit_stats(ctx.rule, ctx.xin->resubmit_stats);
4990 }
4991 if (ctx.xin->xcache) {
4992 struct xc_entry *entry;
4993
4994 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4995 entry->u.rule = ctx.rule;
4996 rule_dpif_ref(ctx.rule);
4997 }
4998
4999 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
5000 ctx.xin->resubmit_hook(ctx.xin, ctx.rule, 0);
5001 }
5002 }
5003 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
5004
5005 /* Get the proximate input port of the packet. (If xin->recirc,
5006 * flow->in_port is the ultimate input port of the packet.) */
5007 struct xport *in_port = get_ofp_port(xbridge,
5008 ctx.base_flow.in_port.ofp_port);
5009
5010 /* Tunnel stats only for non-recirculated packets. */
5011 if (!xin->recirc && in_port && in_port->is_tunnel) {
5012 if (ctx.xin->resubmit_stats) {
5013 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
5014 if (in_port->bfd) {
5015 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
5016 }
5017 }
5018 if (ctx.xin->xcache) {
5019 struct xc_entry *entry;
5020
5021 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
5022 entry->u.dev.rx = netdev_ref(in_port->netdev);
5023 entry->u.dev.bfd = bfd_ref(in_port->bfd);
5024 }
5025 }
5026
5027 if (!xin->recirc && process_special(&ctx, in_port)) {
5028 /* process_special() did all the processing for this packet.
5029 *
5030 * We do not perform special processing on recirculated packets, as
5031 * recirculated packets are not really received by the bridge.*/
5032 } else if (in_port && in_port->xbundle
5033 && xbundle_mirror_out(xbridge, in_port->xbundle)) {
5034 if (ctx.xin->packet != NULL) {
5035 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
5036 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
5037 "%s, which is reserved exclusively for mirroring",
5038 ctx.xbridge->name, in_port->xbundle->name);
5039 }
5040 } else {
5041 /* Sampling is done only for packets really received by the bridge. */
5042 unsigned int user_cookie_offset = 0;
5043 if (!xin->recirc) {
5044 user_cookie_offset = compose_sflow_action(&ctx);
5045 compose_ipfix_action(&ctx, ODPP_NONE);
5046 }
5047 size_t sample_actions_len = ctx.odp_actions->size;
5048
5049 if (tnl_process_ecn(flow)
5050 && (!in_port || may_receive(in_port, &ctx))) {
5051 const struct ofpact *ofpacts;
5052 size_t ofpacts_len;
5053
5054 if (xin->ofpacts) {
5055 ofpacts = xin->ofpacts;
5056 ofpacts_len = xin->ofpacts_len;
5057 } else if (ctx.rule) {
5058 const struct rule_actions *actions
5059 = rule_dpif_get_actions(ctx.rule);
5060 ofpacts = actions->ofpacts;
5061 ofpacts_len = actions->ofpacts_len;
5062 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
5063 } else {
5064 OVS_NOT_REACHED();
5065 }
5066
5067 mirror_ingress_packet(&ctx);
5068 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
5069
5070 /* We've let OFPP_NORMAL and the learning action look at the
5071 * packet, so drop it now if forwarding is disabled. */
5072 if (in_port && (!xport_stp_forward_state(in_port) ||
5073 !xport_rstp_forward_state(in_port))) {
5074 /* Drop all actions added by do_xlate_actions() above. */
5075 ctx.odp_actions->size = sample_actions_len;
5076
5077 /* Undo changes that may have been done for recirculation. */
5078 if (exit_recirculates(&ctx)) {
5079 ctx.action_set.size = ctx.recirc_action_offset;
5080 ctx.recirc_action_offset = -1;
5081 ctx.last_unroll_offset = -1;
5082 }
5083 } else if (ctx.action_set.size) {
5084 /* Translate action set only if not dropping the packet and
5085 * not recirculating. */
5086 if (!exit_recirculates(&ctx)) {
5087 xlate_action_set(&ctx);
5088 }
5089 }
5090 /* Check if need to recirculate. */
5091 if (exit_recirculates(&ctx)) {
5092 compose_recirculate_action(&ctx);
5093 }
5094 }
5095
5096 /* Output only fully processed packets. */
5097 if (!exit_recirculates(&ctx)
5098 && xbridge->has_in_band
5099 && in_band_must_output_to_local_port(flow)
5100 && !actions_output_to_local_port(&ctx)) {
5101 compose_output_action(&ctx, OFPP_LOCAL, NULL);
5102 }
5103
5104 if (user_cookie_offset) {
5105 fix_sflow_action(&ctx, user_cookie_offset);
5106 }
5107 }
5108
5109 if (nl_attr_oversized(ctx.odp_actions->size)) {
5110 /* These datapath actions are too big for a Netlink attribute, so we
5111 * can't hand them to the kernel directly. dpif_execute() can execute
5112 * them one by one with help, so just mark the result as SLOW_ACTION to
5113 * prevent the flow from being installed. */
5114 COVERAGE_INC(xlate_actions_oversize);
5115 ctx.xout->slow |= SLOW_ACTION;
5116 } else if (too_many_output_actions(ctx.odp_actions)) {
5117 COVERAGE_INC(xlate_actions_too_many_output);
5118 ctx.xout->slow |= SLOW_ACTION;
5119 }
5120
5121 /* Do netflow only for packets really received by the bridge and not sent
5122 * to the controller. We consider packets sent to the controller to be
5123 * part of the control plane rather than the data plane. */
5124 if (!xin->recirc && xbridge->netflow && !(xout->slow & SLOW_CONTROLLER)) {
5125 if (ctx.xin->resubmit_stats) {
5126 netflow_flow_update(xbridge->netflow, flow,
5127 ctx.nf_output_iface,
5128 ctx.xin->resubmit_stats);
5129 }
5130 if (ctx.xin->xcache) {
5131 struct xc_entry *entry;
5132
5133 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5134 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5135 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5136 entry->u.nf.iface = ctx.nf_output_iface;
5137 }
5138 }
5139
5140 if (xin->wc) {
5141 xlate_wc_finish(&ctx);
5142 }
5143
5144 exit:
5145 ofpbuf_uninit(&ctx.stack);
5146 ofpbuf_uninit(&ctx.action_set);
5147 ofpbuf_uninit(&scratch_actions);
5148 }
5149
5150 /* Sends 'packet' out 'ofport'.
5151 * May modify 'packet'.
5152 * Returns 0 if successful, otherwise a positive errno value. */
5153 int
5154 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5155 {
5156 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5157 struct xport *xport;
5158 struct ofpact_output output;
5159 struct flow flow;
5160
5161 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5162 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5163 flow_extract(packet, &flow);
5164 flow.in_port.ofp_port = OFPP_NONE;
5165
5166 xport = xport_lookup(xcfg, ofport);
5167 if (!xport) {
5168 return EINVAL;
5169 }
5170 output.port = xport->ofp_port;
5171 output.max_len = 0;
5172
5173 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5174 &output.ofpact, sizeof output,
5175 packet);
5176 }
5177
5178 struct xlate_cache *
5179 xlate_cache_new(void)
5180 {
5181 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5182
5183 ofpbuf_init(&xcache->entries, 512);
5184 return xcache;
5185 }
5186
5187 static struct xc_entry *
5188 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5189 {
5190 struct xc_entry *entry;
5191
5192 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5193 entry->type = type;
5194
5195 return entry;
5196 }
5197
5198 static void
5199 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5200 {
5201 if (entry->u.dev.tx) {
5202 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5203 }
5204 if (entry->u.dev.rx) {
5205 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5206 }
5207 if (entry->u.dev.bfd) {
5208 bfd_account_rx(entry->u.dev.bfd, stats);
5209 }
5210 }
5211
5212 static void
5213 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5214 {
5215 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5216 struct xbridge *xbridge;
5217 struct xbundle *xbundle;
5218 struct flow_wildcards wc;
5219
5220 xbridge = xbridge_lookup(xcfg, ofproto);
5221 if (!xbridge) {
5222 return;
5223 }
5224
5225 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5226 NULL);
5227 if (!xbundle) {
5228 return;
5229 }
5230
5231 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5232 }
5233
5234 /* Push stats and perform side effects of flow translation. */
5235 void
5236 xlate_push_stats(struct xlate_cache *xcache,
5237 const struct dpif_flow_stats *stats)
5238 {
5239 struct xc_entry *entry;
5240 struct ofpbuf entries = xcache->entries;
5241 struct eth_addr dmac;
5242
5243 if (!stats->n_packets) {
5244 return;
5245 }
5246
5247 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5248 switch (entry->type) {
5249 case XC_RULE:
5250 rule_dpif_credit_stats(entry->u.rule, stats);
5251 break;
5252 case XC_BOND:
5253 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5254 entry->u.bond.vid, stats->n_bytes);
5255 break;
5256 case XC_NETDEV:
5257 xlate_cache_netdev(entry, stats);
5258 break;
5259 case XC_NETFLOW:
5260 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5261 entry->u.nf.iface, stats);
5262 break;
5263 case XC_MIRROR:
5264 mirror_update_stats(entry->u.mirror.mbridge,
5265 entry->u.mirror.mirrors,
5266 stats->n_packets, stats->n_bytes);
5267 break;
5268 case XC_LEARN:
5269 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5270 break;
5271 case XC_NORMAL:
5272 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5273 entry->u.normal.vlan);
5274 break;
5275 case XC_FIN_TIMEOUT:
5276 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5277 entry->u.fin.idle, entry->u.fin.hard);
5278 break;
5279 case XC_GROUP:
5280 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5281 stats);
5282 break;
5283 case XC_TNL_ARP:
5284 /* Lookup arp to avoid arp timeout. */
5285 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name,
5286 entry->u.tnl_arp_cache.d_ip, &dmac);
5287 break;
5288 default:
5289 OVS_NOT_REACHED();
5290 }
5291 }
5292 }
5293
5294 static void
5295 xlate_dev_unref(struct xc_entry *entry)
5296 {
5297 if (entry->u.dev.tx) {
5298 netdev_close(entry->u.dev.tx);
5299 }
5300 if (entry->u.dev.rx) {
5301 netdev_close(entry->u.dev.rx);
5302 }
5303 if (entry->u.dev.bfd) {
5304 bfd_unref(entry->u.dev.bfd);
5305 }
5306 }
5307
5308 static void
5309 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5310 {
5311 netflow_flow_clear(netflow, flow);
5312 netflow_unref(netflow);
5313 free(flow);
5314 }
5315
5316 void
5317 xlate_cache_clear(struct xlate_cache *xcache)
5318 {
5319 struct xc_entry *entry;
5320 struct ofpbuf entries;
5321
5322 if (!xcache) {
5323 return;
5324 }
5325
5326 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5327 switch (entry->type) {
5328 case XC_RULE:
5329 rule_dpif_unref(entry->u.rule);
5330 break;
5331 case XC_BOND:
5332 free(entry->u.bond.flow);
5333 bond_unref(entry->u.bond.bond);
5334 break;
5335 case XC_NETDEV:
5336 xlate_dev_unref(entry);
5337 break;
5338 case XC_NETFLOW:
5339 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5340 break;
5341 case XC_MIRROR:
5342 mbridge_unref(entry->u.mirror.mbridge);
5343 break;
5344 case XC_LEARN:
5345 free(entry->u.learn.fm);
5346 ofpbuf_delete(entry->u.learn.ofpacts);
5347 break;
5348 case XC_NORMAL:
5349 free(entry->u.normal.flow);
5350 break;
5351 case XC_FIN_TIMEOUT:
5352 /* 'u.fin.rule' is always already held as a XC_RULE, which
5353 * has already released it's reference above. */
5354 break;
5355 case XC_GROUP:
5356 group_dpif_unref(entry->u.group.group);
5357 break;
5358 case XC_TNL_ARP:
5359 break;
5360 default:
5361 OVS_NOT_REACHED();
5362 }
5363 }
5364
5365 ofpbuf_clear(&xcache->entries);
5366 }
5367
5368 void
5369 xlate_cache_delete(struct xlate_cache *xcache)
5370 {
5371 xlate_cache_clear(xcache);
5372 ofpbuf_uninit(&xcache->entries);
5373 free(xcache);
5374 }