1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License. */
17 #include "ofproto/ofproto-dpif-xlate.h"
20 #include <arpa/inet.h>
22 #include <sys/socket.h>
23 #include <netinet/in.h>
25 #include "tnl-arp-cache.h"
30 #include "byte-order.h"
34 #include "dp-packet.h"
36 #include "dynamic-string.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"
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"
60 #include "openvswitch/vlog.h"
62 COVERAGE_DEFINE(xlate_actions);
63 COVERAGE_DEFINE(xlate_actions_oversize);
64 COVERAGE_DEFINE(xlate_actions_too_many_output);
66 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
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
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)
79 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
80 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
82 struct ovs_list xbundles; /* Owned xbundles. */
83 struct hmap xports; /* Indexed by ofp_port. */
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. */
96 bool has_in_band; /* Bridge has in band control? */
97 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
99 /* Datapath feature support. */
100 struct dpif_backer_support support;
104 struct hmap_node hmap_node; /* In global 'xbundles' map. */
105 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
107 struct ovs_list list_node; /* In parent 'xbridges' list. */
108 struct xbridge *xbridge; /* Parent xbridge. */
110 struct ovs_list xports; /* Contains "struct xport"s. */
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. */
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? */
125 struct hmap_node hmap_node; /* Node in global 'xports' map. */
126 struct ofport_dpif *ofport; /* Key in global 'xports map. */
128 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
129 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
131 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
133 struct ovs_list bundle_node; /* In parent xbundle (if it exists). */
134 struct xbundle *xbundle; /* Parent xbundle or null. */
136 struct netdev *netdev; /* 'ofport''s netdev. */
138 struct xbridge *xbridge; /* Parent bridge. */
139 struct xport *peer; /* Patch port peer or null. */
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. */
146 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
148 bool may_enable; /* May be enabled in bonds. */
149 bool is_tunnel; /* Is a tunnel port. */
151 struct cfm *cfm; /* CFM handle or null. */
152 struct bfd *bfd; /* BFD handle or null. */
153 struct lldp *lldp; /* LLDP handle or null. */
157 struct xlate_in *xin;
158 struct xlate_out *xout;
160 const struct xbridge *xbridge;
162 /* Flow tables version at the beginning of the translation. */
163 cls_version_t tables_version;
165 /* Flow at the last commit. */
166 struct flow base_flow;
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;
176 /* Stack for the push and pop actions. Each stack element is of type
177 * "union mf_subvalue". */
180 /* The rule that we are currently translating, or NULL. */
181 struct rule_dpif *rule;
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;
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
195 struct ofpbuf *odp_actions;
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. */
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 bool exit; /* No further actions should be processed. */
210 /* These are used for non-bond recirculation. The recirculation IDs are
211 * stored in xout and must be associated with a datapath flow (ukey),
212 * otherwise they will be freed when the xout is uninitialized.
215 * Steps in Recirculation Translation
216 * ==================================
218 * At some point during translation, the code recognizes the need for
219 * recirculation. For example, recirculation is necessary when, after
220 * popping the last MPLS label, an action or a match tries to examine or
221 * modify a field that has been newly revealed following the MPLS label.
223 * The simplest part of the work to be done is to commit existing changes to
224 * the packet, which produces datapath actions corresponding to the changes,
225 * and after this, add an OVS_ACTION_ATTR_RECIRC datapath action.
227 * The main problem here is preserving state. When the datapath executes
228 * OVS_ACTION_ATTR_RECIRC, it will upcall to userspace to get a translation
229 * for the post-recirculation actions. At this point userspace has to
230 * resume the translation where it left off, which means that it has to
231 * execute the following:
233 * - The action that prompted recirculation, and any actions following
234 * it within the same flow.
236 * - If the action that prompted recirculation was invoked within a
237 * NXAST_RESUBMIT, then any actions following the resubmit. These
238 * "resubmit"s can be nested, so this has to go all the way up the
241 * - The OpenFlow 1.1+ action set.
243 * State that actions and flow table lookups can depend on, such as the
244 * following, must also be preserved:
246 * - Metadata fields (input port, registers, OF1.1+ metadata, ...).
248 * - Action set, stack
250 * - The table ID and cookie of the flow being translated at each level
251 * of the control stack (since OFPAT_CONTROLLER actions send these to
254 * Translation allows for the control of this state preservation via these
255 * members. When a need for recirculation is identified, the translation
258 * 1. Sets 'recirc_action_offset' to the current size of 'action_set'. The
259 * action set is part of what needs to be preserved, so this allows the
260 * action set and the additional state to share the 'action_set' buffer.
261 * Later steps can tell that setup for recirculation is in progress from
262 * the nonnegative value of 'recirc_action_offset'.
264 * 2. Sets 'exit' to true to tell later steps that we're exiting from the
265 * translation process.
267 * 3. Adds an OFPACT_UNROLL_XLATE action to 'action_set'. This action
268 * holds the current table ID and cookie so that they can be restored
269 * during a post-recirculation upcall translation.
271 * 4. Adds the action that prompted recirculation and any actions following
272 * it within the same flow to 'action_set', so that they can be executed
273 * during a post-recirculation upcall translation.
277 * 6. The action that prompted recirculation might be nested in a stack of
278 * nested "resubmit"s that have actions remaining. Each of these notices
279 * that we're exiting (from 'exit') and that recirculation setup is in
280 * progress (from 'recirc_action_offset') and responds by adding more
281 * OFPACT_UNROLL_XLATE actions to 'action_set', as necessary, and any
282 * actions that were yet unprocessed.
284 * The caller stores all the state produced by this process associated with
285 * the recirculation ID. For post-recirculation upcall translation, the
286 * caller passes it back in for the new translation to execute. The
287 * process yielded a set of ofpacts that can be translated directly, so it
288 * is not much of a special case at that point.
290 int recirc_action_offset; /* Offset in 'action_set' to actions to be
291 * executed after recirculation, or -1. */
292 int last_unroll_offset; /* Offset in 'action_set' to the latest unroll
295 /* True if a packet was but is no longer MPLS (due to an MPLS pop action).
296 * This is a trigger for recirculation in cases where translating an action
297 * or looking up a flow requires access to the fields of the packet after
298 * the MPLS label stack that was originally present. */
301 /* OpenFlow 1.1+ action set.
303 * 'action_set' accumulates "struct ofpact"s added by OFPACT_WRITE_ACTIONS.
304 * When translation is otherwise complete, ofpacts_execute_action_set()
305 * converts it to a set of "struct ofpact"s that can be translated into
306 * datapath actions. */
307 bool action_set_has_group; /* Action set contains OFPACT_GROUP? */
308 struct ofpbuf action_set; /* Action set. */
311 static void xlate_action_set(struct xlate_ctx *ctx);
314 ctx_trigger_recirculation(struct xlate_ctx *ctx)
317 ctx->recirc_action_offset = ctx->action_set.size;
321 ctx_first_recirculation_action(const struct xlate_ctx *ctx)
323 return ctx->recirc_action_offset == ctx->action_set.size;
327 exit_recirculates(const struct xlate_ctx *ctx)
329 /* When recirculating the 'recirc_action_offset' has a non-negative value.
331 return ctx->recirc_action_offset >= 0;
334 static void compose_recirculate_action(struct xlate_ctx *ctx);
336 /* A controller may use OFPP_NONE as the ingress port to indicate that
337 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
338 * when an input bundle is needed for validation (e.g., mirroring or
339 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
340 * any 'port' structs, so care must be taken when dealing with it. */
341 static struct xbundle ofpp_none_bundle = {
343 .vlan_mode = PORT_VLAN_TRUNK
346 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
347 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
348 * traffic egressing the 'ofport' with that priority should be marked with. */
349 struct skb_priority_to_dscp {
350 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
351 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
353 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
369 /* xlate_cache entries hold enough information to perform the side effects of
370 * xlate_actions() for a rule, without needing to perform rule translation
371 * from scratch. The primary usage of these is to submit statistics to objects
372 * that a flow relates to, although they may be used for other effects as well
373 * (for instance, refreshing hard timeouts for learned flows). */
377 struct rule_dpif *rule;
384 struct netflow *netflow;
389 struct mbridge *mbridge;
390 mirror_mask_t mirrors;
398 struct ofproto_dpif *ofproto;
399 struct ofputil_flow_mod *fm;
400 struct ofpbuf *ofpacts;
403 struct ofproto_dpif *ofproto;
408 struct rule_dpif *rule;
413 struct group_dpif *group;
414 struct ofputil_bucket *bucket;
417 char br_name[IFNAMSIZ];
423 #define XC_ENTRY_FOR_EACH(entry, entries, xcache) \
424 entries = xcache->entries; \
425 for (entry = ofpbuf_try_pull(&entries, sizeof *entry); \
427 entry = ofpbuf_try_pull(&entries, sizeof *entry))
430 struct ofpbuf entries;
433 /* Xlate config contains hash maps of all bridges, bundles and ports.
434 * Xcfgp contains the pointer to the current xlate configuration.
435 * When the main thread needs to change the configuration, it copies xcfgp to
436 * new_xcfg and edits new_xcfg. This enables the use of RCU locking which
437 * does not block handler and revalidator threads. */
439 struct hmap xbridges;
440 struct hmap xbundles;
443 static OVSRCU_TYPE(struct xlate_cfg *) xcfgp = OVSRCU_INITIALIZER(NULL);
444 static struct xlate_cfg *new_xcfg = NULL;
446 static bool may_receive(const struct xport *, struct xlate_ctx *);
447 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
449 static void xlate_normal(struct xlate_ctx *);
450 static inline void xlate_report(struct xlate_ctx *, const char *, ...)
451 OVS_PRINTF_FORMAT(2, 3);
452 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
453 uint8_t table_id, bool may_packet_in,
454 bool honor_table_miss);
455 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
456 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
457 static void output_normal(struct xlate_ctx *, const struct xbundle *,
460 /* Optional bond recirculation parameter to compose_output_action(). */
461 struct xlate_bond_recirc {
462 uint32_t recirc_id; /* !0 Use recirculation instead of output. */
463 uint8_t hash_alg; /* !0 Compute hash for recirc before. */
464 uint32_t hash_basis; /* Compute hash for recirc before. */
467 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port,
468 const struct xlate_bond_recirc *xr);
470 static struct xbridge *xbridge_lookup(struct xlate_cfg *,
471 const struct ofproto_dpif *);
472 static struct xbundle *xbundle_lookup(struct xlate_cfg *,
473 const struct ofbundle *);
474 static struct xport *xport_lookup(struct xlate_cfg *,
475 const struct ofport_dpif *);
476 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
477 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
478 uint32_t skb_priority);
479 static void clear_skb_priorities(struct xport *);
480 static size_t count_skb_priorities(const struct xport *);
481 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
484 static struct xc_entry *xlate_cache_add_entry(struct xlate_cache *xc,
486 static void xlate_xbridge_init(struct xlate_cfg *, struct xbridge *);
487 static void xlate_xbundle_init(struct xlate_cfg *, struct xbundle *);
488 static void xlate_xport_init(struct xlate_cfg *, struct xport *);
489 static void xlate_xbridge_set(struct xbridge *, struct dpif *,
490 const struct mac_learning *, struct stp *,
491 struct rstp *, const struct mcast_snooping *,
492 const struct mbridge *,
493 const struct dpif_sflow *,
494 const struct dpif_ipfix *,
495 const struct netflow *,
496 bool forward_bpdu, bool has_in_band,
497 const struct dpif_backer_support *);
498 static void xlate_xbundle_set(struct xbundle *xbundle,
499 enum port_vlan_mode vlan_mode, int vlan,
500 unsigned long *trunks, bool use_priority_tags,
501 const struct bond *bond, const struct lacp *lacp,
503 static void xlate_xport_set(struct xport *xport, odp_port_t odp_port,
504 const struct netdev *netdev, const struct cfm *cfm,
505 const struct bfd *bfd, const struct lldp *lldp,
506 int stp_port_no, const struct rstp_port *rstp_port,
507 enum ofputil_port_config config,
508 enum ofputil_port_state state, bool is_tunnel,
510 static void xlate_xbridge_remove(struct xlate_cfg *, struct xbridge *);
511 static void xlate_xbundle_remove(struct xlate_cfg *, struct xbundle *);
512 static void xlate_xport_remove(struct xlate_cfg *, struct xport *);
513 static void xlate_xbridge_copy(struct xbridge *);
514 static void xlate_xbundle_copy(struct xbridge *, struct xbundle *);
515 static void xlate_xport_copy(struct xbridge *, struct xbundle *,
517 static void xlate_xcfg_free(struct xlate_cfg *);
520 xlate_report(struct xlate_ctx *ctx, const char *format, ...)
522 if (OVS_UNLIKELY(ctx->xin->report_hook)) {
525 va_start(args, format);
526 ctx->xin->report_hook(ctx->xin, ctx->recurse, format, args);
532 xlate_xbridge_init(struct xlate_cfg *xcfg, struct xbridge *xbridge)
534 list_init(&xbridge->xbundles);
535 hmap_init(&xbridge->xports);
536 hmap_insert(&xcfg->xbridges, &xbridge->hmap_node,
537 hash_pointer(xbridge->ofproto, 0));
541 xlate_xbundle_init(struct xlate_cfg *xcfg, struct xbundle *xbundle)
543 list_init(&xbundle->xports);
544 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
545 hmap_insert(&xcfg->xbundles, &xbundle->hmap_node,
546 hash_pointer(xbundle->ofbundle, 0));
550 xlate_xport_init(struct xlate_cfg *xcfg, struct xport *xport)
552 hmap_init(&xport->skb_priorities);
553 hmap_insert(&xcfg->xports, &xport->hmap_node,
554 hash_pointer(xport->ofport, 0));
555 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
556 hash_ofp_port(xport->ofp_port));
560 xlate_xbridge_set(struct xbridge *xbridge,
562 const struct mac_learning *ml, struct stp *stp,
563 struct rstp *rstp, const struct mcast_snooping *ms,
564 const struct mbridge *mbridge,
565 const struct dpif_sflow *sflow,
566 const struct dpif_ipfix *ipfix,
567 const struct netflow *netflow,
568 bool forward_bpdu, bool has_in_band,
569 const struct dpif_backer_support *support)
571 if (xbridge->ml != ml) {
572 mac_learning_unref(xbridge->ml);
573 xbridge->ml = mac_learning_ref(ml);
576 if (xbridge->ms != ms) {
577 mcast_snooping_unref(xbridge->ms);
578 xbridge->ms = mcast_snooping_ref(ms);
581 if (xbridge->mbridge != mbridge) {
582 mbridge_unref(xbridge->mbridge);
583 xbridge->mbridge = mbridge_ref(mbridge);
586 if (xbridge->sflow != sflow) {
587 dpif_sflow_unref(xbridge->sflow);
588 xbridge->sflow = dpif_sflow_ref(sflow);
591 if (xbridge->ipfix != ipfix) {
592 dpif_ipfix_unref(xbridge->ipfix);
593 xbridge->ipfix = dpif_ipfix_ref(ipfix);
596 if (xbridge->stp != stp) {
597 stp_unref(xbridge->stp);
598 xbridge->stp = stp_ref(stp);
601 if (xbridge->rstp != rstp) {
602 rstp_unref(xbridge->rstp);
603 xbridge->rstp = rstp_ref(rstp);
606 if (xbridge->netflow != netflow) {
607 netflow_unref(xbridge->netflow);
608 xbridge->netflow = netflow_ref(netflow);
611 xbridge->dpif = dpif;
612 xbridge->forward_bpdu = forward_bpdu;
613 xbridge->has_in_band = has_in_band;
614 xbridge->support = *support;
618 xlate_xbundle_set(struct xbundle *xbundle,
619 enum port_vlan_mode vlan_mode, int vlan,
620 unsigned long *trunks, bool use_priority_tags,
621 const struct bond *bond, const struct lacp *lacp,
624 ovs_assert(xbundle->xbridge);
626 xbundle->vlan_mode = vlan_mode;
627 xbundle->vlan = vlan;
628 xbundle->trunks = trunks;
629 xbundle->use_priority_tags = use_priority_tags;
630 xbundle->floodable = floodable;
632 if (xbundle->bond != bond) {
633 bond_unref(xbundle->bond);
634 xbundle->bond = bond_ref(bond);
637 if (xbundle->lacp != lacp) {
638 lacp_unref(xbundle->lacp);
639 xbundle->lacp = lacp_ref(lacp);
644 xlate_xport_set(struct xport *xport, odp_port_t odp_port,
645 const struct netdev *netdev, const struct cfm *cfm,
646 const struct bfd *bfd, const struct lldp *lldp, int stp_port_no,
647 const struct rstp_port* rstp_port,
648 enum ofputil_port_config config, enum ofputil_port_state state,
649 bool is_tunnel, bool may_enable)
651 xport->config = config;
652 xport->state = state;
653 xport->stp_port_no = stp_port_no;
654 xport->is_tunnel = is_tunnel;
655 xport->may_enable = may_enable;
656 xport->odp_port = odp_port;
658 if (xport->rstp_port != rstp_port) {
659 rstp_port_unref(xport->rstp_port);
660 xport->rstp_port = rstp_port_ref(rstp_port);
663 if (xport->cfm != cfm) {
664 cfm_unref(xport->cfm);
665 xport->cfm = cfm_ref(cfm);
668 if (xport->bfd != bfd) {
669 bfd_unref(xport->bfd);
670 xport->bfd = bfd_ref(bfd);
673 if (xport->lldp != lldp) {
674 lldp_unref(xport->lldp);
675 xport->lldp = lldp_ref(lldp);
678 if (xport->netdev != netdev) {
679 netdev_close(xport->netdev);
680 xport->netdev = netdev_ref(netdev);
685 xlate_xbridge_copy(struct xbridge *xbridge)
687 struct xbundle *xbundle;
689 struct xbridge *new_xbridge = xzalloc(sizeof *xbridge);
690 new_xbridge->ofproto = xbridge->ofproto;
691 new_xbridge->name = xstrdup(xbridge->name);
692 xlate_xbridge_init(new_xcfg, new_xbridge);
694 xlate_xbridge_set(new_xbridge,
695 xbridge->dpif, xbridge->ml, xbridge->stp,
696 xbridge->rstp, xbridge->ms, xbridge->mbridge,
697 xbridge->sflow, xbridge->ipfix, xbridge->netflow,
698 xbridge->forward_bpdu, xbridge->has_in_band,
700 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
701 xlate_xbundle_copy(new_xbridge, xbundle);
704 /* Copy xports which are not part of a xbundle */
705 HMAP_FOR_EACH (xport, ofp_node, &xbridge->xports) {
706 if (!xport->xbundle) {
707 xlate_xport_copy(new_xbridge, NULL, xport);
713 xlate_xbundle_copy(struct xbridge *xbridge, struct xbundle *xbundle)
716 struct xbundle *new_xbundle = xzalloc(sizeof *xbundle);
717 new_xbundle->ofbundle = xbundle->ofbundle;
718 new_xbundle->xbridge = xbridge;
719 new_xbundle->name = xstrdup(xbundle->name);
720 xlate_xbundle_init(new_xcfg, new_xbundle);
722 xlate_xbundle_set(new_xbundle, xbundle->vlan_mode,
723 xbundle->vlan, xbundle->trunks,
724 xbundle->use_priority_tags, xbundle->bond, xbundle->lacp,
726 LIST_FOR_EACH (xport, bundle_node, &xbundle->xports) {
727 xlate_xport_copy(xbridge, new_xbundle, xport);
732 xlate_xport_copy(struct xbridge *xbridge, struct xbundle *xbundle,
735 struct skb_priority_to_dscp *pdscp, *new_pdscp;
736 struct xport *new_xport = xzalloc(sizeof *xport);
737 new_xport->ofport = xport->ofport;
738 new_xport->ofp_port = xport->ofp_port;
739 new_xport->xbridge = xbridge;
740 xlate_xport_init(new_xcfg, new_xport);
742 xlate_xport_set(new_xport, xport->odp_port, xport->netdev, xport->cfm,
743 xport->bfd, xport->lldp, xport->stp_port_no,
744 xport->rstp_port, xport->config, xport->state,
745 xport->is_tunnel, xport->may_enable);
748 struct xport *peer = xport_lookup(new_xcfg, xport->peer->ofport);
750 new_xport->peer = peer;
751 new_xport->peer->peer = new_xport;
756 new_xport->xbundle = xbundle;
757 list_insert(&new_xport->xbundle->xports, &new_xport->bundle_node);
760 HMAP_FOR_EACH (pdscp, hmap_node, &xport->skb_priorities) {
761 new_pdscp = xmalloc(sizeof *pdscp);
762 new_pdscp->skb_priority = pdscp->skb_priority;
763 new_pdscp->dscp = pdscp->dscp;
764 hmap_insert(&new_xport->skb_priorities, &new_pdscp->hmap_node,
765 hash_int(new_pdscp->skb_priority, 0));
769 /* Sets the current xlate configuration to new_xcfg and frees the old xlate
770 * configuration in xcfgp.
772 * This needs to be called after editing the xlate configuration.
774 * Functions that edit the new xlate configuration are
775 * xlate_<ofport/bundle/ofport>_set and xlate_<ofport/bundle/ofport>_remove.
781 * edit_xlate_configuration();
783 * xlate_txn_commit(); */
785 xlate_txn_commit(void)
787 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
789 ovsrcu_set(&xcfgp, new_xcfg);
790 ovsrcu_synchronize();
791 xlate_xcfg_free(xcfg);
795 /* Copies the current xlate configuration in xcfgp to new_xcfg.
797 * This needs to be called prior to editing the xlate configuration. */
799 xlate_txn_start(void)
801 struct xbridge *xbridge;
802 struct xlate_cfg *xcfg;
804 ovs_assert(!new_xcfg);
806 new_xcfg = xmalloc(sizeof *new_xcfg);
807 hmap_init(&new_xcfg->xbridges);
808 hmap_init(&new_xcfg->xbundles);
809 hmap_init(&new_xcfg->xports);
811 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
816 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
817 xlate_xbridge_copy(xbridge);
823 xlate_xcfg_free(struct xlate_cfg *xcfg)
825 struct xbridge *xbridge, *next_xbridge;
831 HMAP_FOR_EACH_SAFE (xbridge, next_xbridge, hmap_node, &xcfg->xbridges) {
832 xlate_xbridge_remove(xcfg, xbridge);
835 hmap_destroy(&xcfg->xbridges);
836 hmap_destroy(&xcfg->xbundles);
837 hmap_destroy(&xcfg->xports);
842 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
844 const struct mac_learning *ml, struct stp *stp,
845 struct rstp *rstp, const struct mcast_snooping *ms,
846 const struct mbridge *mbridge,
847 const struct dpif_sflow *sflow,
848 const struct dpif_ipfix *ipfix,
849 const struct netflow *netflow,
850 bool forward_bpdu, bool has_in_band,
851 const struct dpif_backer_support *support)
853 struct xbridge *xbridge;
855 ovs_assert(new_xcfg);
857 xbridge = xbridge_lookup(new_xcfg, ofproto);
859 xbridge = xzalloc(sizeof *xbridge);
860 xbridge->ofproto = ofproto;
862 xlate_xbridge_init(new_xcfg, xbridge);
866 xbridge->name = xstrdup(name);
868 xlate_xbridge_set(xbridge, dpif, ml, stp, rstp, ms, mbridge, sflow, ipfix,
869 netflow, forward_bpdu, has_in_band, support);
873 xlate_xbridge_remove(struct xlate_cfg *xcfg, struct xbridge *xbridge)
875 struct xbundle *xbundle, *next_xbundle;
876 struct xport *xport, *next_xport;
882 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
883 xlate_xport_remove(xcfg, xport);
886 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
887 xlate_xbundle_remove(xcfg, xbundle);
890 hmap_remove(&xcfg->xbridges, &xbridge->hmap_node);
891 mac_learning_unref(xbridge->ml);
892 mcast_snooping_unref(xbridge->ms);
893 mbridge_unref(xbridge->mbridge);
894 dpif_sflow_unref(xbridge->sflow);
895 dpif_ipfix_unref(xbridge->ipfix);
896 stp_unref(xbridge->stp);
897 rstp_unref(xbridge->rstp);
898 hmap_destroy(&xbridge->xports);
904 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
906 struct xbridge *xbridge;
908 ovs_assert(new_xcfg);
910 xbridge = xbridge_lookup(new_xcfg, ofproto);
911 xlate_xbridge_remove(new_xcfg, xbridge);
915 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
916 const char *name, enum port_vlan_mode vlan_mode, int vlan,
917 unsigned long *trunks, bool use_priority_tags,
918 const struct bond *bond, const struct lacp *lacp,
921 struct xbundle *xbundle;
923 ovs_assert(new_xcfg);
925 xbundle = xbundle_lookup(new_xcfg, ofbundle);
927 xbundle = xzalloc(sizeof *xbundle);
928 xbundle->ofbundle = ofbundle;
929 xbundle->xbridge = xbridge_lookup(new_xcfg, ofproto);
931 xlate_xbundle_init(new_xcfg, xbundle);
935 xbundle->name = xstrdup(name);
937 xlate_xbundle_set(xbundle, vlan_mode, vlan, trunks,
938 use_priority_tags, bond, lacp, floodable);
942 xlate_xbundle_remove(struct xlate_cfg *xcfg, struct xbundle *xbundle)
950 LIST_FOR_EACH_POP (xport, bundle_node, &xbundle->xports) {
951 xport->xbundle = NULL;
954 hmap_remove(&xcfg->xbundles, &xbundle->hmap_node);
955 list_remove(&xbundle->list_node);
956 bond_unref(xbundle->bond);
957 lacp_unref(xbundle->lacp);
963 xlate_bundle_remove(struct ofbundle *ofbundle)
965 struct xbundle *xbundle;
967 ovs_assert(new_xcfg);
969 xbundle = xbundle_lookup(new_xcfg, ofbundle);
970 xlate_xbundle_remove(new_xcfg, xbundle);
974 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
975 struct ofport_dpif *ofport, ofp_port_t ofp_port,
976 odp_port_t odp_port, const struct netdev *netdev,
977 const struct cfm *cfm, const struct bfd *bfd,
978 const struct lldp *lldp, struct ofport_dpif *peer,
979 int stp_port_no, const struct rstp_port *rstp_port,
980 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
981 enum ofputil_port_config config,
982 enum ofputil_port_state state, bool is_tunnel,
988 ovs_assert(new_xcfg);
990 xport = xport_lookup(new_xcfg, ofport);
992 xport = xzalloc(sizeof *xport);
993 xport->ofport = ofport;
994 xport->xbridge = xbridge_lookup(new_xcfg, ofproto);
995 xport->ofp_port = ofp_port;
997 xlate_xport_init(new_xcfg, xport);
1000 ovs_assert(xport->ofp_port == ofp_port);
1002 xlate_xport_set(xport, odp_port, netdev, cfm, bfd, lldp,
1003 stp_port_no, rstp_port, config, state, is_tunnel,
1007 xport->peer->peer = NULL;
1009 xport->peer = xport_lookup(new_xcfg, peer);
1011 xport->peer->peer = xport;
1014 if (xport->xbundle) {
1015 list_remove(&xport->bundle_node);
1017 xport->xbundle = xbundle_lookup(new_xcfg, ofbundle);
1018 if (xport->xbundle) {
1019 list_insert(&xport->xbundle->xports, &xport->bundle_node);
1022 clear_skb_priorities(xport);
1023 for (i = 0; i < n_qdscp; i++) {
1024 struct skb_priority_to_dscp *pdscp;
1025 uint32_t skb_priority;
1027 if (dpif_queue_to_priority(xport->xbridge->dpif, qdscp_list[i].queue,
1032 pdscp = xmalloc(sizeof *pdscp);
1033 pdscp->skb_priority = skb_priority;
1034 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
1035 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
1036 hash_int(pdscp->skb_priority, 0));
1041 xlate_xport_remove(struct xlate_cfg *xcfg, struct xport *xport)
1048 xport->peer->peer = NULL;
1052 if (xport->xbundle) {
1053 list_remove(&xport->bundle_node);
1056 clear_skb_priorities(xport);
1057 hmap_destroy(&xport->skb_priorities);
1059 hmap_remove(&xcfg->xports, &xport->hmap_node);
1060 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
1062 netdev_close(xport->netdev);
1063 rstp_port_unref(xport->rstp_port);
1064 cfm_unref(xport->cfm);
1065 bfd_unref(xport->bfd);
1066 lldp_unref(xport->lldp);
1071 xlate_ofport_remove(struct ofport_dpif *ofport)
1073 struct xport *xport;
1075 ovs_assert(new_xcfg);
1077 xport = xport_lookup(new_xcfg, ofport);
1078 xlate_xport_remove(new_xcfg, xport);
1081 static struct ofproto_dpif *
1082 xlate_lookup_ofproto_(const struct dpif_backer *backer, const struct flow *flow,
1083 ofp_port_t *ofp_in_port, const struct xport **xportp)
1085 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1086 const struct xport *xport;
1088 xport = xport_lookup(xcfg, tnl_port_should_receive(flow)
1089 ? tnl_port_receive(flow)
1090 : odp_port_to_ofport(backer, flow->in_port.odp_port));
1091 if (OVS_UNLIKELY(!xport)) {
1096 *ofp_in_port = xport->ofp_port;
1098 return xport->xbridge->ofproto;
1101 /* Given a datapath and flow metadata ('backer', and 'flow' respectively)
1102 * returns the corresponding struct ofproto_dpif and OpenFlow port number. */
1103 struct ofproto_dpif *
1104 xlate_lookup_ofproto(const struct dpif_backer *backer, const struct flow *flow,
1105 ofp_port_t *ofp_in_port)
1107 const struct xport *xport;
1109 return xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1112 /* Given a datapath and flow metadata ('backer', and 'flow' respectively),
1113 * optionally populates 'ofproto' with the ofproto_dpif, 'ofp_in_port' with the
1114 * openflow in_port, and 'ipfix', 'sflow', and 'netflow' with the appropriate
1115 * handles for those protocols if they're enabled. Caller may use the returned
1116 * pointers until quiescing, for longer term use additional references must
1119 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofproto.
1122 xlate_lookup(const struct dpif_backer *backer, const struct flow *flow,
1123 struct ofproto_dpif **ofprotop, struct dpif_ipfix **ipfix,
1124 struct dpif_sflow **sflow, struct netflow **netflow,
1125 ofp_port_t *ofp_in_port)
1127 struct ofproto_dpif *ofproto;
1128 const struct xport *xport;
1130 ofproto = xlate_lookup_ofproto_(backer, flow, ofp_in_port, &xport);
1137 *ofprotop = ofproto;
1141 *ipfix = xport ? xport->xbridge->ipfix : NULL;
1145 *sflow = xport ? xport->xbridge->sflow : NULL;
1149 *netflow = xport ? xport->xbridge->netflow : NULL;
1155 static struct xbridge *
1156 xbridge_lookup(struct xlate_cfg *xcfg, const struct ofproto_dpif *ofproto)
1158 struct hmap *xbridges;
1159 struct xbridge *xbridge;
1161 if (!ofproto || !xcfg) {
1165 xbridges = &xcfg->xbridges;
1167 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
1169 if (xbridge->ofproto == ofproto) {
1176 static struct xbundle *
1177 xbundle_lookup(struct xlate_cfg *xcfg, const struct ofbundle *ofbundle)
1179 struct hmap *xbundles;
1180 struct xbundle *xbundle;
1182 if (!ofbundle || !xcfg) {
1186 xbundles = &xcfg->xbundles;
1188 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
1190 if (xbundle->ofbundle == ofbundle) {
1197 static struct xport *
1198 xport_lookup(struct xlate_cfg *xcfg, const struct ofport_dpif *ofport)
1200 struct hmap *xports;
1201 struct xport *xport;
1203 if (!ofport || !xcfg) {
1207 xports = &xcfg->xports;
1209 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
1211 if (xport->ofport == ofport) {
1218 static struct stp_port *
1219 xport_get_stp_port(const struct xport *xport)
1221 return xport->xbridge->stp && xport->stp_port_no != -1
1222 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
1227 xport_stp_learn_state(const struct xport *xport)
1229 struct stp_port *sp = xport_get_stp_port(xport);
1231 ? stp_learn_in_state(stp_port_get_state(sp))
1236 xport_stp_forward_state(const struct xport *xport)
1238 struct stp_port *sp = xport_get_stp_port(xport);
1240 ? stp_forward_in_state(stp_port_get_state(sp))
1245 xport_stp_should_forward_bpdu(const struct xport *xport)
1247 struct stp_port *sp = xport_get_stp_port(xport);
1248 return stp_should_forward_bpdu(sp ? stp_port_get_state(sp) : STP_DISABLED);
1251 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
1252 * were used to make the determination.*/
1254 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
1256 /* is_stp() also checks dl_type, but dl_type is always set in 'wc'. */
1257 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1258 return is_stp(flow);
1262 stp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1264 struct stp_port *sp = xport_get_stp_port(xport);
1265 struct dp_packet payload = *packet;
1266 struct eth_header *eth = dp_packet_data(&payload);
1268 /* Sink packets on ports that have STP disabled when the bridge has
1270 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
1274 /* Trim off padding on payload. */
1275 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1276 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1279 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1280 stp_received_bpdu(sp, dp_packet_data(&payload), dp_packet_size(&payload));
1284 static enum rstp_state
1285 xport_get_rstp_port_state(const struct xport *xport)
1287 return xport->rstp_port
1288 ? rstp_port_get_state(xport->rstp_port)
1293 xport_rstp_learn_state(const struct xport *xport)
1295 return xport->xbridge->rstp && xport->rstp_port
1296 ? rstp_learn_in_state(xport_get_rstp_port_state(xport))
1301 xport_rstp_forward_state(const struct xport *xport)
1303 return xport->xbridge->rstp && xport->rstp_port
1304 ? rstp_forward_in_state(xport_get_rstp_port_state(xport))
1309 xport_rstp_should_manage_bpdu(const struct xport *xport)
1311 return rstp_should_manage_bpdu(xport_get_rstp_port_state(xport));
1315 rstp_process_packet(const struct xport *xport, const struct dp_packet *packet)
1317 struct dp_packet payload = *packet;
1318 struct eth_header *eth = dp_packet_data(&payload);
1320 /* Sink packets on ports that have no RSTP. */
1321 if (!xport->rstp_port) {
1325 /* Trim off padding on payload. */
1326 if (dp_packet_size(&payload) > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
1327 dp_packet_set_size(&payload, ntohs(eth->eth_type) + ETH_HEADER_LEN);
1330 if (dp_packet_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
1331 rstp_port_received_bpdu(xport->rstp_port, dp_packet_data(&payload),
1332 dp_packet_size(&payload));
1336 static struct xport *
1337 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1339 struct xport *xport;
1341 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
1343 if (xport->ofp_port == ofp_port) {
1351 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
1353 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
1354 return xport ? xport->odp_port : ODPP_NONE;
1358 odp_port_is_alive(const struct xlate_ctx *ctx, ofp_port_t ofp_port)
1360 struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1361 return xport && xport->may_enable;
1364 static struct ofputil_bucket *
1365 group_first_live_bucket(const struct xlate_ctx *, const struct group_dpif *,
1369 group_is_alive(const struct xlate_ctx *ctx, uint32_t group_id, int depth)
1371 struct group_dpif *group;
1373 if (group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group)) {
1374 struct ofputil_bucket *bucket;
1376 bucket = group_first_live_bucket(ctx, group, depth);
1377 group_dpif_unref(group);
1378 return bucket == NULL;
1384 #define MAX_LIVENESS_RECURSION 128 /* Arbitrary limit */
1387 bucket_is_alive(const struct xlate_ctx *ctx,
1388 struct ofputil_bucket *bucket, int depth)
1390 if (depth >= MAX_LIVENESS_RECURSION) {
1391 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
1393 VLOG_WARN_RL(&rl, "bucket chaining exceeded %d links",
1394 MAX_LIVENESS_RECURSION);
1398 return (!ofputil_bucket_has_liveness(bucket)
1399 || (bucket->watch_port != OFPP_ANY
1400 && odp_port_is_alive(ctx, bucket->watch_port))
1401 || (bucket->watch_group != OFPG_ANY
1402 && group_is_alive(ctx, bucket->watch_group, depth + 1)));
1405 static struct ofputil_bucket *
1406 group_first_live_bucket(const struct xlate_ctx *ctx,
1407 const struct group_dpif *group, int depth)
1409 struct ofputil_bucket *bucket;
1410 const struct ovs_list *buckets;
1412 group_dpif_get_buckets(group, &buckets);
1413 LIST_FOR_EACH (bucket, list_node, buckets) {
1414 if (bucket_is_alive(ctx, bucket, depth)) {
1422 static struct ofputil_bucket *
1423 group_best_live_bucket(const struct xlate_ctx *ctx,
1424 const struct group_dpif *group,
1427 struct ofputil_bucket *best_bucket = NULL;
1428 uint32_t best_score = 0;
1431 struct ofputil_bucket *bucket;
1432 const struct ovs_list *buckets;
1434 group_dpif_get_buckets(group, &buckets);
1435 LIST_FOR_EACH (bucket, list_node, buckets) {
1436 if (bucket_is_alive(ctx, bucket, 0)) {
1437 uint32_t score = (hash_int(i, basis) & 0xffff) * bucket->weight;
1438 if (score >= best_score) {
1439 best_bucket = bucket;
1450 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
1452 return (bundle->vlan_mode != PORT_VLAN_ACCESS
1453 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
1457 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
1459 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
1462 static mirror_mask_t
1463 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
1465 return xbundle != &ofpp_none_bundle
1466 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
1470 static mirror_mask_t
1471 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
1473 return xbundle != &ofpp_none_bundle
1474 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
1478 static mirror_mask_t
1479 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
1481 return xbundle != &ofpp_none_bundle
1482 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
1486 static struct xbundle *
1487 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
1488 bool warn, struct xport **in_xportp)
1490 struct xport *xport;
1492 /* Find the port and bundle for the received packet. */
1493 xport = get_ofp_port(xbridge, in_port);
1497 if (xport && xport->xbundle) {
1498 return xport->xbundle;
1501 /* Special-case OFPP_NONE (OF1.0) and OFPP_CONTROLLER (OF1.1+),
1502 * which a controller may use as the ingress port for traffic that
1503 * it is sourcing. */
1504 if (in_port == OFPP_CONTROLLER || in_port == OFPP_NONE) {
1505 return &ofpp_none_bundle;
1508 /* Odd. A few possible reasons here:
1510 * - We deleted a port but there are still a few packets queued up
1513 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
1514 * we don't know about.
1516 * - The ofproto client didn't configure the port as part of a bundle.
1517 * This is particularly likely to happen if a packet was received on the
1518 * port after it was created, but before the client had a chance to
1519 * configure its bundle.
1522 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1524 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
1525 "port %"PRIu16, xbridge->name, in_port);
1531 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
1533 const struct xbridge *xbridge = ctx->xbridge;
1534 mirror_mask_t mirrors;
1535 struct xbundle *in_xbundle;
1539 mirrors = ctx->xout->mirrors;
1540 ctx->xout->mirrors = 0;
1542 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
1543 ctx->xin->packet != NULL, NULL);
1547 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
1549 /* Drop frames on bundles reserved for mirroring. */
1550 if (xbundle_mirror_out(xbridge, in_xbundle)) {
1551 if (ctx->xin->packet != NULL) {
1552 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1553 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1554 "%s, which is reserved exclusively for mirroring",
1555 ctx->xbridge->name, in_xbundle->name);
1557 ofpbuf_clear(ctx->odp_actions);
1562 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
1563 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1566 vlan = input_vid_to_vlan(in_xbundle, vid);
1572 /* Restore the original packet before adding the mirror actions. */
1573 ctx->xin->flow = *orig_flow;
1576 mirror_mask_t dup_mirrors;
1577 struct ofbundle *out;
1578 const unsigned long *vlans;
1583 has_mirror = mirror_get(xbridge->mbridge, raw_ctz(mirrors),
1584 &vlans, &dup_mirrors, &out, &out_vlan);
1585 ovs_assert(has_mirror);
1588 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
1590 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
1592 if (!vlan_mirrored) {
1593 mirrors = zero_rightmost_1bit(mirrors);
1597 mirrors &= ~dup_mirrors;
1598 ctx->xout->mirrors |= dup_mirrors;
1600 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1601 struct xbundle *out_xbundle = xbundle_lookup(xcfg, out);
1603 output_normal(ctx, out_xbundle, vlan);
1605 } else if (vlan != out_vlan
1606 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
1607 struct xbundle *xbundle;
1609 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
1610 if (xbundle_includes_vlan(xbundle, out_vlan)
1611 && !xbundle_mirror_out(xbridge, xbundle)) {
1612 output_normal(ctx, xbundle, out_vlan);
1619 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
1620 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
1621 * the bundle on which the packet was received, returns the VLAN to which the
1624 * Both 'vid' and the return value are in the range 0...4095. */
1626 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
1628 switch (in_xbundle->vlan_mode) {
1629 case PORT_VLAN_ACCESS:
1630 return in_xbundle->vlan;
1633 case PORT_VLAN_TRUNK:
1636 case PORT_VLAN_NATIVE_UNTAGGED:
1637 case PORT_VLAN_NATIVE_TAGGED:
1638 return vid ? vid : in_xbundle->vlan;
1645 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
1646 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
1649 * 'vid' should be the VID obtained from the 802.1Q header that was received as
1650 * part of a packet (specify 0 if there was no 802.1Q header), in the range
1653 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
1655 /* Allow any VID on the OFPP_NONE port. */
1656 if (in_xbundle == &ofpp_none_bundle) {
1660 switch (in_xbundle->vlan_mode) {
1661 case PORT_VLAN_ACCESS:
1664 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1665 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
1666 "packet received on port %s configured as VLAN "
1667 "%"PRIu16" access port", vid, in_xbundle->name,
1674 case PORT_VLAN_NATIVE_UNTAGGED:
1675 case PORT_VLAN_NATIVE_TAGGED:
1677 /* Port must always carry its native VLAN. */
1681 case PORT_VLAN_TRUNK:
1682 if (!xbundle_includes_vlan(in_xbundle, vid)) {
1684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1685 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
1686 "received on port %s not configured for trunking "
1687 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
1699 /* Given 'vlan', the VLAN that a packet belongs to, and
1700 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
1701 * that should be included in the 802.1Q header. (If the return value is 0,
1702 * then the 802.1Q header should only be included in the packet if there is a
1705 * Both 'vlan' and the return value are in the range 0...4095. */
1707 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
1709 switch (out_xbundle->vlan_mode) {
1710 case PORT_VLAN_ACCESS:
1713 case PORT_VLAN_TRUNK:
1714 case PORT_VLAN_NATIVE_TAGGED:
1717 case PORT_VLAN_NATIVE_UNTAGGED:
1718 return vlan == out_xbundle->vlan ? 0 : vlan;
1726 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
1729 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
1731 ovs_be16 tci, old_tci;
1732 struct xport *xport;
1733 struct xlate_bond_recirc xr;
1734 bool use_recirc = false;
1736 vid = output_vlan_to_vid(out_xbundle, vlan);
1737 if (list_is_empty(&out_xbundle->xports)) {
1738 /* Partially configured bundle with no slaves. Drop the packet. */
1740 } else if (!out_xbundle->bond) {
1741 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
1744 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
1745 struct flow_wildcards *wc = ctx->wc;
1746 struct ofport_dpif *ofport;
1748 if (ctx->xbridge->support.odp.recirc) {
1749 use_recirc = bond_may_recirc(
1750 out_xbundle->bond, &xr.recirc_id, &xr.hash_basis);
1753 /* Only TCP mode uses recirculation. */
1754 xr.hash_alg = OVS_HASH_ALG_L4;
1755 bond_update_post_recirc_rules(out_xbundle->bond, false);
1757 /* Recirculation does not require unmasking hash fields. */
1762 ofport = bond_choose_output_slave(out_xbundle->bond,
1763 &ctx->xin->flow, wc, vid);
1764 xport = xport_lookup(xcfg, ofport);
1767 /* No slaves enabled, so drop packet. */
1771 /* If use_recirc is set, the main thread will handle stats
1772 * accounting for this bond. */
1774 if (ctx->xin->resubmit_stats) {
1775 bond_account(out_xbundle->bond, &ctx->xin->flow, vid,
1776 ctx->xin->resubmit_stats->n_bytes);
1778 if (ctx->xin->xcache) {
1779 struct xc_entry *entry;
1782 flow = &ctx->xin->flow;
1783 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_BOND);
1784 entry->u.bond.bond = bond_ref(out_xbundle->bond);
1785 entry->u.bond.flow = xmemdup(flow, sizeof *flow);
1786 entry->u.bond.vid = vid;
1791 old_tci = *flow_tci;
1793 if (tci || out_xbundle->use_priority_tags) {
1794 tci |= *flow_tci & htons(VLAN_PCP_MASK);
1796 tci |= htons(VLAN_CFI);
1801 compose_output_action(ctx, xport->ofp_port, use_recirc ? &xr : NULL);
1802 *flow_tci = old_tci;
1805 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
1806 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
1807 * indicate this; newer upstream kernels use gratuitous ARP requests. */
1809 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
1811 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
1815 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1816 if (!eth_addr_is_broadcast(flow->dl_dst)) {
1820 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
1821 if (flow->nw_proto == ARP_OP_REPLY) {
1823 } else if (flow->nw_proto == ARP_OP_REQUEST) {
1824 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
1825 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
1827 return flow->nw_src == flow->nw_dst;
1833 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
1834 * dropped. Returns true if they may be forwarded, false if they should be
1837 * 'in_port' must be the xport that corresponds to flow->in_port.
1838 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
1840 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
1841 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
1842 * checked by input_vid_is_valid().
1844 * May also add tags to '*tags', although the current implementation only does
1845 * so in one special case.
1848 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
1851 struct xbundle *in_xbundle = in_port->xbundle;
1852 const struct xbridge *xbridge = ctx->xbridge;
1853 struct flow *flow = &ctx->xin->flow;
1855 /* Drop frames for reserved multicast addresses
1856 * only if forward_bpdu option is absent. */
1857 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
1858 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1862 if (in_xbundle->bond) {
1863 struct mac_entry *mac;
1865 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1871 xlate_report(ctx, "bonding refused admissibility, dropping");
1874 case BV_DROP_IF_MOVED:
1875 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1876 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1878 && mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle
1879 && (!is_gratuitous_arp(flow, ctx->wc)
1880 || mac_entry_is_grat_arp_locked(mac))) {
1881 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1882 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1886 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1894 /* Checks whether a MAC learning update is necessary for MAC learning table
1895 * 'ml' given that a packet matching 'flow' was received on 'in_xbundle' in
1898 * Most packets processed through the MAC learning table do not actually
1899 * change it in any way. This function requires only a read lock on the MAC
1900 * learning table, so it is much cheaper in this common case.
1902 * Keep the code here synchronized with that in update_learning_table__()
1905 is_mac_learning_update_needed(const struct mac_learning *ml,
1906 const struct flow *flow,
1907 struct flow_wildcards *wc,
1908 int vlan, struct xbundle *in_xbundle)
1909 OVS_REQ_RDLOCK(ml->rwlock)
1911 struct mac_entry *mac;
1913 if (!mac_learning_may_learn(ml, flow->dl_src, vlan)) {
1917 mac = mac_learning_lookup(ml, flow->dl_src, vlan);
1918 if (!mac || mac_entry_age(ml, mac)) {
1922 if (is_gratuitous_arp(flow, wc)) {
1923 /* We don't want to learn from gratuitous ARP packets that are
1924 * reflected back over bond slaves so we lock the learning table. */
1925 if (!in_xbundle->bond) {
1927 } else if (mac_entry_is_grat_arp_locked(mac)) {
1932 return mac_entry_get_port(ml, mac) != in_xbundle->ofbundle;
1936 /* Updates MAC learning table 'ml' given that a packet matching 'flow' was
1937 * received on 'in_xbundle' in 'vlan'.
1939 * This code repeats all the checks in is_mac_learning_update_needed() because
1940 * the lock was released between there and here and thus the MAC learning state
1941 * could have changed.
1943 * Keep the code here synchronized with that in is_mac_learning_update_needed()
1946 update_learning_table__(const struct xbridge *xbridge,
1947 const struct flow *flow, struct flow_wildcards *wc,
1948 int vlan, struct xbundle *in_xbundle)
1949 OVS_REQ_WRLOCK(xbridge->ml->rwlock)
1951 struct mac_entry *mac;
1953 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
1957 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
1958 if (is_gratuitous_arp(flow, wc)) {
1959 /* We don't want to learn from gratuitous ARP packets that are
1960 * reflected back over bond slaves so we lock the learning table. */
1961 if (!in_xbundle->bond) {
1962 mac_entry_set_grat_arp_lock(mac);
1963 } else if (mac_entry_is_grat_arp_locked(mac)) {
1968 if (mac_entry_get_port(xbridge->ml, mac) != in_xbundle->ofbundle) {
1969 /* The log messages here could actually be useful in debugging,
1970 * so keep the rate limit relatively high. */
1971 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
1973 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
1974 "on port %s in VLAN %d",
1975 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
1976 in_xbundle->name, vlan);
1978 mac_entry_set_port(xbridge->ml, mac, in_xbundle->ofbundle);
1983 update_learning_table(const struct xbridge *xbridge,
1984 const struct flow *flow, struct flow_wildcards *wc,
1985 int vlan, struct xbundle *in_xbundle)
1989 /* Don't learn the OFPP_NONE port. */
1990 if (in_xbundle == &ofpp_none_bundle) {
1994 /* First try the common case: no change to MAC learning table. */
1995 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1996 need_update = is_mac_learning_update_needed(xbridge->ml, flow, wc, vlan,
1998 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2001 /* Slow path: MAC learning table might need an update. */
2002 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
2003 update_learning_table__(xbridge, flow, wc, vlan, in_xbundle);
2004 ovs_rwlock_unlock(&xbridge->ml->rwlock);
2008 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2009 * was received on 'in_xbundle' in 'vlan' and is either Report or Query. */
2011 update_mcast_snooping_table4__(const struct xbridge *xbridge,
2012 const struct flow *flow,
2013 struct mcast_snooping *ms, int vlan,
2014 struct xbundle *in_xbundle,
2015 const struct dp_packet *packet)
2016 OVS_REQ_WRLOCK(ms->rwlock)
2018 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2020 ovs_be32 ip4 = flow->igmp_group_ip4;
2022 switch (ntohs(flow->tp_src)) {
2023 case IGMP_HOST_MEMBERSHIP_REPORT:
2024 case IGMPV2_HOST_MEMBERSHIP_REPORT:
2025 if (mcast_snooping_add_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2026 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping learned that "
2027 IP_FMT" is on port %s in VLAN %d",
2028 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2031 case IGMP_HOST_LEAVE_MESSAGE:
2032 if (mcast_snooping_leave_group4(ms, ip4, vlan, in_xbundle->ofbundle)) {
2033 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping leaving "
2034 IP_FMT" is on port %s in VLAN %d",
2035 xbridge->name, IP_ARGS(ip4), in_xbundle->name, vlan);
2038 case IGMP_HOST_MEMBERSHIP_QUERY:
2039 if (flow->nw_src && mcast_snooping_add_mrouter(ms, vlan,
2040 in_xbundle->ofbundle)) {
2041 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query from "
2042 IP_FMT" is on port %s in VLAN %d",
2043 xbridge->name, IP_ARGS(flow->nw_src),
2044 in_xbundle->name, vlan);
2047 case IGMPV3_HOST_MEMBERSHIP_REPORT:
2048 if ((count = mcast_snooping_add_report(ms, packet, vlan,
2049 in_xbundle->ofbundle))) {
2050 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2051 "addresses on port %s in VLAN %d",
2052 xbridge->name, count, in_xbundle->name, vlan);
2059 update_mcast_snooping_table6__(const struct xbridge *xbridge,
2060 const struct flow *flow,
2061 struct mcast_snooping *ms, int vlan,
2062 struct xbundle *in_xbundle,
2063 const struct dp_packet *packet)
2064 OVS_REQ_WRLOCK(ms->rwlock)
2066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 30);
2069 switch (ntohs(flow->tp_src)) {
2071 if (!ipv6_addr_equals(&flow->ipv6_src, &in6addr_any)
2072 && mcast_snooping_add_mrouter(ms, vlan, in_xbundle->ofbundle)) {
2073 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping query on port %s"
2075 xbridge->name, in_xbundle->name, vlan);
2081 count = mcast_snooping_add_mld(ms, packet, vlan, in_xbundle->ofbundle);
2083 VLOG_DBG_RL(&rl, "bridge %s: multicast snooping processed %d "
2084 "addresses on port %s in VLAN %d",
2085 xbridge->name, count, in_xbundle->name, vlan);
2091 /* Updates multicast snooping table 'ms' given that a packet matching 'flow'
2092 * was received on 'in_xbundle' in 'vlan'. */
2094 update_mcast_snooping_table(const struct xbridge *xbridge,
2095 const struct flow *flow, int vlan,
2096 struct xbundle *in_xbundle,
2097 const struct dp_packet *packet)
2099 struct mcast_snooping *ms = xbridge->ms;
2100 struct xlate_cfg *xcfg;
2101 struct xbundle *mcast_xbundle;
2102 struct mcast_port_bundle *fport;
2104 /* Don't learn the OFPP_NONE port. */
2105 if (in_xbundle == &ofpp_none_bundle) {
2109 /* Don't learn from flood ports */
2110 mcast_xbundle = NULL;
2111 ovs_rwlock_wrlock(&ms->rwlock);
2112 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2113 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2114 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2115 if (mcast_xbundle == in_xbundle) {
2120 if (!mcast_xbundle || mcast_xbundle != in_xbundle) {
2121 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2122 update_mcast_snooping_table4__(xbridge, flow, ms, vlan,
2123 in_xbundle, packet);
2125 update_mcast_snooping_table6__(xbridge, flow, ms, vlan,
2126 in_xbundle, packet);
2129 ovs_rwlock_unlock(&ms->rwlock);
2132 /* send the packet to ports having the multicast group learned */
2134 xlate_normal_mcast_send_group(struct xlate_ctx *ctx,
2135 struct mcast_snooping *ms OVS_UNUSED,
2136 struct mcast_group *grp,
2137 struct xbundle *in_xbundle, uint16_t vlan)
2138 OVS_REQ_RDLOCK(ms->rwlock)
2140 struct xlate_cfg *xcfg;
2141 struct mcast_group_bundle *b;
2142 struct xbundle *mcast_xbundle;
2144 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2145 LIST_FOR_EACH(b, bundle_node, &grp->bundle_lru) {
2146 mcast_xbundle = xbundle_lookup(xcfg, b->port);
2147 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2148 xlate_report(ctx, "forwarding to mcast group port");
2149 output_normal(ctx, mcast_xbundle, vlan);
2150 } else if (!mcast_xbundle) {
2151 xlate_report(ctx, "mcast group port is unknown, dropping");
2153 xlate_report(ctx, "mcast group port is input port, dropping");
2158 /* send the packet to ports connected to multicast routers */
2160 xlate_normal_mcast_send_mrouters(struct xlate_ctx *ctx,
2161 struct mcast_snooping *ms,
2162 struct xbundle *in_xbundle, uint16_t vlan)
2163 OVS_REQ_RDLOCK(ms->rwlock)
2165 struct xlate_cfg *xcfg;
2166 struct mcast_mrouter_bundle *mrouter;
2167 struct xbundle *mcast_xbundle;
2169 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2170 LIST_FOR_EACH(mrouter, mrouter_node, &ms->mrouter_lru) {
2171 mcast_xbundle = xbundle_lookup(xcfg, mrouter->port);
2172 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2173 xlate_report(ctx, "forwarding to mcast router port");
2174 output_normal(ctx, mcast_xbundle, vlan);
2175 } else if (!mcast_xbundle) {
2176 xlate_report(ctx, "mcast router port is unknown, dropping");
2178 xlate_report(ctx, "mcast router port is input port, dropping");
2183 /* send the packet to ports flagged to be flooded */
2185 xlate_normal_mcast_send_fports(struct xlate_ctx *ctx,
2186 struct mcast_snooping *ms,
2187 struct xbundle *in_xbundle, uint16_t vlan)
2188 OVS_REQ_RDLOCK(ms->rwlock)
2190 struct xlate_cfg *xcfg;
2191 struct mcast_port_bundle *fport;
2192 struct xbundle *mcast_xbundle;
2194 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2195 LIST_FOR_EACH(fport, node, &ms->fport_list) {
2196 mcast_xbundle = xbundle_lookup(xcfg, fport->port);
2197 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2198 xlate_report(ctx, "forwarding to mcast flood port");
2199 output_normal(ctx, mcast_xbundle, vlan);
2200 } else if (!mcast_xbundle) {
2201 xlate_report(ctx, "mcast flood port is unknown, dropping");
2203 xlate_report(ctx, "mcast flood port is input port, dropping");
2208 /* forward the Reports to configured ports */
2210 xlate_normal_mcast_send_rports(struct xlate_ctx *ctx,
2211 struct mcast_snooping *ms,
2212 struct xbundle *in_xbundle, uint16_t vlan)
2213 OVS_REQ_RDLOCK(ms->rwlock)
2215 struct xlate_cfg *xcfg;
2216 struct mcast_port_bundle *rport;
2217 struct xbundle *mcast_xbundle;
2219 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2220 LIST_FOR_EACH(rport, node, &ms->rport_list) {
2221 mcast_xbundle = xbundle_lookup(xcfg, rport->port);
2222 if (mcast_xbundle && mcast_xbundle != in_xbundle) {
2223 xlate_report(ctx, "forwarding Report to mcast flagged port");
2224 output_normal(ctx, mcast_xbundle, vlan);
2225 } else if (!mcast_xbundle) {
2226 xlate_report(ctx, "mcast port is unknown, dropping the Report");
2228 xlate_report(ctx, "mcast port is input port, dropping the Report");
2234 xlate_normal_flood(struct xlate_ctx *ctx, struct xbundle *in_xbundle,
2237 struct xbundle *xbundle;
2239 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
2240 if (xbundle != in_xbundle
2241 && xbundle_includes_vlan(xbundle, vlan)
2242 && xbundle->floodable
2243 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
2244 output_normal(ctx, xbundle, vlan);
2247 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
2251 xlate_normal(struct xlate_ctx *ctx)
2253 struct flow_wildcards *wc = ctx->wc;
2254 struct flow *flow = &ctx->xin->flow;
2255 struct xbundle *in_xbundle;
2256 struct xport *in_port;
2257 struct mac_entry *mac;
2262 ctx->xout->has_normal = true;
2264 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
2265 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
2266 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2268 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
2269 ctx->xin->packet != NULL, &in_port);
2271 xlate_report(ctx, "no input bundle, dropping");
2275 /* Drop malformed frames. */
2276 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
2277 !(flow->vlan_tci & htons(VLAN_CFI))) {
2278 if (ctx->xin->packet != NULL) {
2279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2280 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
2281 "VLAN tag received on port %s",
2282 ctx->xbridge->name, in_xbundle->name);
2284 xlate_report(ctx, "partial VLAN tag, dropping");
2288 /* Drop frames on bundles reserved for mirroring. */
2289 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
2290 if (ctx->xin->packet != NULL) {
2291 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2292 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
2293 "%s, which is reserved exclusively for mirroring",
2294 ctx->xbridge->name, in_xbundle->name);
2296 xlate_report(ctx, "input port is mirror output port, dropping");
2301 vid = vlan_tci_to_vid(flow->vlan_tci);
2302 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
2303 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
2306 vlan = input_vid_to_vlan(in_xbundle, vid);
2308 /* Check other admissibility requirements. */
2309 if (in_port && !is_admissible(ctx, in_port, vlan)) {
2313 /* Learn source MAC. */
2314 if (ctx->xin->may_learn) {
2315 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
2317 if (ctx->xin->xcache) {
2318 struct xc_entry *entry;
2320 /* Save enough info to update mac learning table later. */
2321 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NORMAL);
2322 entry->u.normal.ofproto = ctx->xbridge->ofproto;
2323 entry->u.normal.flow = xmemdup(flow, sizeof *flow);
2324 entry->u.normal.vlan = vlan;
2327 /* Determine output bundle. */
2328 if (mcast_snooping_enabled(ctx->xbridge->ms)
2329 && !eth_addr_is_broadcast(flow->dl_dst)
2330 && eth_addr_is_multicast(flow->dl_dst)
2331 && is_ip_any(flow)) {
2332 struct mcast_snooping *ms = ctx->xbridge->ms;
2333 struct mcast_group *grp = NULL;
2335 if (is_igmp(flow)) {
2336 if (mcast_snooping_is_membership(flow->tp_src) ||
2337 mcast_snooping_is_query(flow->tp_src)) {
2338 if (ctx->xin->may_learn) {
2339 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2340 in_xbundle, ctx->xin->packet);
2343 * IGMP packets need to take the slow path, in order to be
2344 * processed for mdb updates. That will prevent expires
2345 * firing off even after hosts have sent reports.
2347 ctx->xout->slow |= SLOW_ACTION;
2350 if (mcast_snooping_is_membership(flow->tp_src)) {
2351 ovs_rwlock_rdlock(&ms->rwlock);
2352 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2353 /* RFC4541: section 2.1.1, item 1: A snooping switch should
2354 * forward IGMP Membership Reports only to those ports where
2355 * multicast routers are attached. Alternatively stated: a
2356 * snooping switch should not forward IGMP Membership Reports
2357 * to ports on which only hosts are attached.
2358 * An administrative control may be provided to override this
2359 * restriction, allowing the report messages to be flooded to
2361 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2362 ovs_rwlock_unlock(&ms->rwlock);
2364 xlate_report(ctx, "multicast traffic, flooding");
2365 xlate_normal_flood(ctx, in_xbundle, vlan);
2368 } else if (is_mld(flow)) {
2369 ctx->xout->slow |= SLOW_ACTION;
2370 if (ctx->xin->may_learn) {
2371 update_mcast_snooping_table(ctx->xbridge, flow, vlan,
2372 in_xbundle, ctx->xin->packet);
2374 if (is_mld_report(flow)) {
2375 ovs_rwlock_rdlock(&ms->rwlock);
2376 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2377 xlate_normal_mcast_send_rports(ctx, ms, in_xbundle, vlan);
2378 ovs_rwlock_unlock(&ms->rwlock);
2380 xlate_report(ctx, "MLD query, flooding");
2381 xlate_normal_flood(ctx, in_xbundle, vlan);
2384 if ((flow->dl_type == htons(ETH_TYPE_IP)
2385 && ip_is_local_multicast(flow->nw_dst))
2386 || (flow->dl_type == htons(ETH_TYPE_IPV6)
2387 && ipv6_is_all_hosts(&flow->ipv6_dst))) {
2388 /* RFC4541: section 2.1.2, item 2: Packets with a dst IP
2389 * address in the 224.0.0.x range which are not IGMP must
2390 * be forwarded on all ports */
2391 xlate_report(ctx, "RFC4541: section 2.1.2, item 2, flooding");
2392 xlate_normal_flood(ctx, in_xbundle, vlan);
2397 /* forwarding to group base ports */
2398 ovs_rwlock_rdlock(&ms->rwlock);
2399 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2400 grp = mcast_snooping_lookup4(ms, flow->nw_dst, vlan);
2401 } else if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
2402 grp = mcast_snooping_lookup(ms, &flow->ipv6_dst, vlan);
2405 xlate_normal_mcast_send_group(ctx, ms, grp, in_xbundle, vlan);
2406 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2407 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2409 if (mcast_snooping_flood_unreg(ms)) {
2410 xlate_report(ctx, "unregistered multicast, flooding");
2411 xlate_normal_flood(ctx, in_xbundle, vlan);
2413 xlate_normal_mcast_send_mrouters(ctx, ms, in_xbundle, vlan);
2414 xlate_normal_mcast_send_fports(ctx, ms, in_xbundle, vlan);
2417 ovs_rwlock_unlock(&ms->rwlock);
2419 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
2420 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
2421 mac_port = mac ? mac_entry_get_port(ctx->xbridge->ml, mac) : NULL;
2422 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
2425 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2426 struct xbundle *mac_xbundle = xbundle_lookup(xcfg, mac_port);
2427 if (mac_xbundle && mac_xbundle != in_xbundle) {
2428 xlate_report(ctx, "forwarding to learned port");
2429 output_normal(ctx, mac_xbundle, vlan);
2430 } else if (!mac_xbundle) {
2431 xlate_report(ctx, "learned port is unknown, dropping");
2433 xlate_report(ctx, "learned port is input port, dropping");
2436 xlate_report(ctx, "no learned MAC for destination, flooding");
2437 xlate_normal_flood(ctx, in_xbundle, vlan);
2442 /* Appends a "sample" action for sFlow or IPFIX to 'ctx->odp_actions'. The
2443 * 'probability' is the number of packets out of UINT32_MAX to sample. The
2444 * 'cookie' (of length 'cookie_size' bytes) is passed back in the callback for
2445 * each sampled packet. 'tunnel_out_port', if not ODPP_NONE, is added as the
2446 * OVS_USERSPACE_ATTR_EGRESS_TUN_PORT attribute. If 'include_actions', an
2447 * OVS_USERSPACE_ATTR_ACTIONS attribute is added.
2450 compose_sample_action(struct xlate_ctx *ctx,
2451 const uint32_t probability,
2452 const union user_action_cookie *cookie,
2453 const size_t cookie_size,
2454 const odp_port_t tunnel_out_port,
2455 bool include_actions)
2457 size_t sample_offset = nl_msg_start_nested(ctx->odp_actions,
2458 OVS_ACTION_ATTR_SAMPLE);
2460 nl_msg_put_u32(ctx->odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
2462 size_t actions_offset = nl_msg_start_nested(ctx->odp_actions,
2463 OVS_SAMPLE_ATTR_ACTIONS);
2465 odp_port_t odp_port = ofp_port_to_odp_port(
2466 ctx->xbridge, ctx->xin->flow.in_port.ofp_port);
2467 uint32_t pid = dpif_port_get_pid(ctx->xbridge->dpif, odp_port,
2468 flow_hash_5tuple(&ctx->xin->flow, 0));
2469 int cookie_offset = odp_put_userspace_action(pid, cookie, cookie_size,
2474 nl_msg_end_nested(ctx->odp_actions, actions_offset);
2475 nl_msg_end_nested(ctx->odp_actions, sample_offset);
2477 return cookie_offset;
2480 /* If sFLow is not enabled, returns 0 without doing anything.
2482 * If sFlow is enabled, appends a template "sample" action to the ODP actions
2483 * in 'ctx'. This action is a template because some of the information needed
2484 * to fill it out is not available until flow translation is complete. In this
2485 * case, this functions returns an offset, which is always nonzero, to pass
2486 * later to fix_sflow_action() to fill in the rest of the template. */
2488 compose_sflow_action(struct xlate_ctx *ctx)
2490 struct dpif_sflow *sflow = ctx->xbridge->sflow;
2491 if (!sflow || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2495 union user_action_cookie cookie = { .type = USER_ACTION_COOKIE_SFLOW };
2496 return compose_sample_action(ctx, dpif_sflow_get_probability(sflow),
2497 &cookie, sizeof cookie.sflow, ODPP_NONE,
2501 /* If IPFIX is enabled, this appends a "sample" action to implement IPFIX to
2502 * 'ctx->odp_actions'. */
2504 compose_ipfix_action(struct xlate_ctx *ctx, odp_port_t output_odp_port)
2506 struct dpif_ipfix *ipfix = ctx->xbridge->ipfix;
2507 odp_port_t tunnel_out_port = ODPP_NONE;
2509 if (!ipfix || ctx->xin->flow.in_port.ofp_port == OFPP_NONE) {
2513 /* For input case, output_odp_port is ODPP_NONE, which is an invalid port
2515 if (output_odp_port == ODPP_NONE &&
2516 !dpif_ipfix_get_bridge_exporter_input_sampling(ipfix)) {
2520 /* For output case, output_odp_port is valid*/
2521 if (output_odp_port != ODPP_NONE) {
2522 if (!dpif_ipfix_get_bridge_exporter_output_sampling(ipfix)) {
2525 /* If tunnel sampling is enabled, put an additional option attribute:
2526 * OVS_USERSPACE_ATTR_TUNNEL_OUT_PORT
2528 if (dpif_ipfix_get_bridge_exporter_tunnel_sampling(ipfix) &&
2529 dpif_ipfix_get_tunnel_port(ipfix, output_odp_port) ) {
2530 tunnel_out_port = output_odp_port;
2534 union user_action_cookie cookie = {
2536 .type = USER_ACTION_COOKIE_IPFIX,
2537 .output_odp_port = output_odp_port,
2540 compose_sample_action(ctx,
2541 dpif_ipfix_get_bridge_exporter_probability(ipfix),
2542 &cookie, sizeof cookie.ipfix, tunnel_out_port,
2546 /* Fix "sample" action according to data collected while composing ODP actions,
2547 * as described in compose_sflow_action().
2549 * 'user_cookie_offset' must be the offset returned by add_sflow_action(). */
2551 fix_sflow_action(struct xlate_ctx *ctx, unsigned int user_cookie_offset)
2553 const struct flow *base = &ctx->base_flow;
2554 union user_action_cookie *cookie;
2556 cookie = ofpbuf_at(ctx->odp_actions, user_cookie_offset,
2557 sizeof cookie->sflow);
2558 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
2560 cookie->type = USER_ACTION_COOKIE_SFLOW;
2561 cookie->sflow.vlan_tci = base->vlan_tci;
2563 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
2564 * port information") for the interpretation of cookie->output. */
2565 switch (ctx->sflow_n_outputs) {
2567 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
2568 cookie->sflow.output = 0x40000000 | 256;
2572 cookie->sflow.output = dpif_sflow_odp_port_to_ifindex(
2573 ctx->xbridge->sflow, ctx->sflow_odp_port);
2574 if (cookie->sflow.output) {
2579 /* 0x80000000 means "multiple output ports. */
2580 cookie->sflow.output = 0x80000000 | ctx->sflow_n_outputs;
2586 process_special(struct xlate_ctx *ctx, const struct xport *xport)
2588 const struct flow *flow = &ctx->xin->flow;
2589 struct flow_wildcards *wc = ctx->wc;
2590 const struct xbridge *xbridge = ctx->xbridge;
2591 const struct dp_packet *packet = ctx->xin->packet;
2592 enum slow_path_reason slow;
2596 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
2598 cfm_process_heartbeat(xport->cfm, packet);
2601 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
2603 bfd_process_packet(xport->bfd, flow, packet);
2604 /* If POLL received, immediately sends FINAL back. */
2605 if (bfd_should_send_packet(xport->bfd)) {
2606 ofproto_dpif_monitor_port_send_soon(xport->ofport);
2610 } else if (xport->xbundle && xport->xbundle->lacp
2611 && flow->dl_type == htons(ETH_TYPE_LACP)) {
2613 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
2616 } else if ((xbridge->stp || xbridge->rstp) &&
2617 stp_should_process_flow(flow, wc)) {
2620 ? stp_process_packet(xport, packet)
2621 : rstp_process_packet(xport, packet);
2624 } else if (xport->lldp && lldp_should_process_flow(xport->lldp, flow)) {
2626 lldp_process_packet(xport->lldp, packet);
2634 ctx->xout->slow |= slow;
2642 tnl_route_lookup_flow(const struct flow *oflow,
2643 ovs_be32 *ip, struct xport **out_port)
2645 char out_dev[IFNAMSIZ];
2646 struct xbridge *xbridge;
2647 struct xlate_cfg *xcfg;
2650 if (!ovs_router_lookup(oflow->tunnel.ip_dst, out_dev, &gw)) {
2657 *ip = oflow->tunnel.ip_dst;
2660 xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
2663 HMAP_FOR_EACH (xbridge, hmap_node, &xcfg->xbridges) {
2664 if (!strncmp(xbridge->name, out_dev, IFNAMSIZ)) {
2667 HMAP_FOR_EACH (port, ofp_node, &xbridge->xports) {
2668 if (!strncmp(netdev_get_name(port->netdev), out_dev, IFNAMSIZ)) {
2679 xlate_flood_packet(struct xbridge *xbridge, struct dp_packet *packet)
2681 struct ofpact_output output;
2684 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
2685 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
2686 flow_extract(packet, &flow);
2687 flow.in_port.ofp_port = OFPP_NONE;
2688 output.port = OFPP_FLOOD;
2691 return ofproto_dpif_execute_actions(xbridge->ofproto, &flow, NULL,
2692 &output.ofpact, sizeof output,
2697 tnl_send_arp_request(const struct xport *out_dev, const uint8_t eth_src[ETH_ADDR_LEN],
2698 ovs_be32 ip_src, ovs_be32 ip_dst)
2700 struct xbridge *xbridge = out_dev->xbridge;
2701 struct dp_packet packet;
2703 dp_packet_init(&packet, 0);
2704 compose_arp(&packet, ARP_OP_REQUEST,
2705 eth_src, eth_addr_zero, true, ip_src, ip_dst);
2707 xlate_flood_packet(xbridge, &packet);
2708 dp_packet_uninit(&packet);
2712 build_tunnel_send(struct xlate_ctx *ctx, const struct xport *xport,
2713 const struct flow *flow, odp_port_t tunnel_odp_port)
2715 struct ovs_action_push_tnl tnl_push_data;
2716 struct xport *out_dev = NULL;
2717 ovs_be32 s_ip, d_ip = 0;
2718 uint8_t smac[ETH_ADDR_LEN];
2719 uint8_t dmac[ETH_ADDR_LEN];
2722 err = tnl_route_lookup_flow(flow, &d_ip, &out_dev);
2724 xlate_report(ctx, "native tunnel routing failed");
2727 xlate_report(ctx, "tunneling to "IP_FMT" via %s",
2728 IP_ARGS(d_ip), netdev_get_name(out_dev->netdev));
2730 /* Use mac addr of bridge port of the peer. */
2731 err = netdev_get_etheraddr(out_dev->netdev, smac);
2733 xlate_report(ctx, "tunnel output device lacks Ethernet address");
2737 err = netdev_get_in4(out_dev->netdev, (struct in_addr *) &s_ip, NULL);
2739 xlate_report(ctx, "tunnel output device lacks IPv4 address");
2743 err = tnl_arp_lookup(out_dev->xbridge->name, d_ip, dmac);
2745 xlate_report(ctx, "ARP cache miss for "IP_FMT" on bridge %s, "
2746 "sending ARP request",
2747 IP_ARGS(d_ip), out_dev->xbridge->name);
2748 tnl_send_arp_request(out_dev, smac, s_ip, d_ip);
2751 if (ctx->xin->xcache) {
2752 struct xc_entry *entry;
2754 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_TNL_ARP);
2755 ovs_strlcpy(entry->u.tnl_arp_cache.br_name, out_dev->xbridge->name,
2756 sizeof entry->u.tnl_arp_cache.br_name);
2757 entry->u.tnl_arp_cache.d_ip = d_ip;
2760 xlate_report(ctx, "tunneling from "ETH_ADDR_FMT" "IP_FMT
2761 " to "ETH_ADDR_FMT" "IP_FMT,
2762 ETH_ADDR_ARGS(smac), IP_ARGS(s_ip),
2763 ETH_ADDR_ARGS(dmac), IP_ARGS(d_ip));
2764 err = tnl_port_build_header(xport->ofport, flow,
2765 dmac, smac, s_ip, &tnl_push_data);
2769 tnl_push_data.tnl_port = odp_to_u32(tunnel_odp_port);
2770 tnl_push_data.out_port = odp_to_u32(out_dev->odp_port);
2771 odp_put_tnl_push_action(ctx->odp_actions, &tnl_push_data);
2776 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
2777 const struct xlate_bond_recirc *xr, bool check_stp)
2779 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
2780 struct flow_wildcards *wc = ctx->wc;
2781 struct flow *flow = &ctx->xin->flow;
2782 struct flow_tnl flow_tnl;
2783 ovs_be16 flow_vlan_tci;
2784 uint32_t flow_pkt_mark;
2785 uint8_t flow_nw_tos;
2786 odp_port_t out_port, odp_port;
2787 bool tnl_push_pop_send = false;
2790 /* If 'struct flow' gets additional metadata, we'll need to zero it out
2791 * before traversing a patch port. */
2792 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 33);
2793 memset(&flow_tnl, 0, sizeof flow_tnl);
2796 xlate_report(ctx, "Nonexistent output port");
2798 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
2799 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
2801 } else if (check_stp) {
2802 if (is_stp(&ctx->base_flow)) {
2803 if (!xport_stp_should_forward_bpdu(xport) &&
2804 !xport_rstp_should_manage_bpdu(xport)) {
2805 if (ctx->xbridge->stp != NULL) {
2806 xlate_report(ctx, "STP not in listening state, "
2807 "skipping bpdu output");
2808 } else if (ctx->xbridge->rstp != NULL) {
2809 xlate_report(ctx, "RSTP not managing BPDU in this state, "
2810 "skipping bpdu output");
2814 } else if (!xport_stp_forward_state(xport) ||
2815 !xport_rstp_forward_state(xport)) {
2816 if (ctx->xbridge->stp != NULL) {
2817 xlate_report(ctx, "STP not in forwarding state, "
2819 } else if (ctx->xbridge->rstp != NULL) {
2820 xlate_report(ctx, "RSTP not in forwarding state, "
2827 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
2828 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
2833 const struct xport *peer = xport->peer;
2834 struct flow old_flow = ctx->xin->flow;
2835 bool old_was_mpls = ctx->was_mpls;
2836 cls_version_t old_version = ctx->tables_version;
2837 struct ofpbuf old_stack = ctx->stack;
2838 union mf_subvalue new_stack[1024 / sizeof(union mf_subvalue)];
2839 struct ofpbuf old_action_set = ctx->action_set;
2840 uint64_t actset_stub[1024 / 8];
2842 ofpbuf_use_stub(&ctx->stack, new_stack, sizeof new_stack);
2843 ofpbuf_use_stub(&ctx->action_set, actset_stub, sizeof actset_stub);
2844 ctx->xbridge = peer->xbridge;
2845 flow->in_port.ofp_port = peer->ofp_port;
2846 flow->metadata = htonll(0);
2847 memset(&flow->tunnel, 0, sizeof flow->tunnel);
2848 memset(flow->regs, 0, sizeof flow->regs);
2849 flow->actset_output = OFPP_UNSET;
2851 /* The bridge is now known so obtain its table version. */
2853 = ofproto_dpif_get_tables_version(ctx->xbridge->ofproto);
2855 if (!process_special(ctx, peer) && may_receive(peer, ctx)) {
2856 if (xport_stp_forward_state(peer) && xport_rstp_forward_state(peer)) {
2857 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2858 if (ctx->action_set.size) {
2859 /* Translate action set only if not dropping the packet and
2860 * not recirculating. */
2861 if (!exit_recirculates(ctx)) {
2862 xlate_action_set(ctx);
2865 /* Check if need to recirculate. */
2866 if (exit_recirculates(ctx)) {
2867 compose_recirculate_action(ctx);
2870 /* Forwarding is disabled by STP and RSTP. Let OFPP_NORMAL and
2871 * the learning action look at the packet, then drop it. */
2872 struct flow old_base_flow = ctx->base_flow;
2873 size_t old_size = ctx->odp_actions->size;
2874 mirror_mask_t old_mirrors = ctx->xout->mirrors;
2876 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true, true);
2877 ctx->xout->mirrors = old_mirrors;
2878 ctx->base_flow = old_base_flow;
2879 ctx->odp_actions->size = old_size;
2881 /* Undo changes that may have been done for recirculation. */
2882 if (exit_recirculates(ctx)) {
2883 ctx->action_set.size = ctx->recirc_action_offset;
2884 ctx->recirc_action_offset = -1;
2885 ctx->last_unroll_offset = -1;
2890 ctx->xin->flow = old_flow;
2891 ctx->xbridge = xport->xbridge;
2892 ofpbuf_uninit(&ctx->action_set);
2893 ctx->action_set = old_action_set;
2894 ofpbuf_uninit(&ctx->stack);
2895 ctx->stack = old_stack;
2897 /* Restore calling bridge's lookup version. */
2898 ctx->tables_version = old_version;
2900 /* The peer bridge popping MPLS should have no effect on the original
2902 ctx->was_mpls = old_was_mpls;
2904 /* The fact that the peer bridge exits (for any reason) does not mean
2905 * that the original bridge should exit. Specifically, if the peer
2906 * bridge recirculates (which typically modifies the packet), the
2907 * original bridge must continue processing with the original, not the
2908 * recirculated packet! */
2911 if (ctx->xin->resubmit_stats) {
2912 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2913 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
2915 bfd_account_rx(peer->bfd, ctx->xin->resubmit_stats);
2918 if (ctx->xin->xcache) {
2919 struct xc_entry *entry;
2921 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2922 entry->u.dev.tx = netdev_ref(xport->netdev);
2923 entry->u.dev.rx = netdev_ref(peer->netdev);
2924 entry->u.dev.bfd = bfd_ref(peer->bfd);
2929 flow_vlan_tci = flow->vlan_tci;
2930 flow_pkt_mark = flow->pkt_mark;
2931 flow_nw_tos = flow->nw_tos;
2933 if (count_skb_priorities(xport)) {
2934 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2935 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
2936 wc->masks.nw_tos |= IP_DSCP_MASK;
2937 flow->nw_tos &= ~IP_DSCP_MASK;
2938 flow->nw_tos |= dscp;
2942 if (xport->is_tunnel) {
2943 /* Save tunnel metadata so that changes made due to
2944 * the Logical (tunnel) Port are not visible for any further
2945 * matches, while explicit set actions on tunnel metadata are.
2947 flow_tnl = flow->tunnel;
2948 odp_port = tnl_port_send(xport->ofport, flow, ctx->wc);
2949 if (odp_port == ODPP_NONE) {
2950 xlate_report(ctx, "Tunneling decided against output");
2951 goto out; /* restore flow_nw_tos */
2953 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
2954 xlate_report(ctx, "Not tunneling to our own address");
2955 goto out; /* restore flow_nw_tos */
2957 if (ctx->xin->resubmit_stats) {
2958 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
2960 if (ctx->xin->xcache) {
2961 struct xc_entry *entry;
2963 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_NETDEV);
2964 entry->u.dev.tx = netdev_ref(xport->netdev);
2966 out_port = odp_port;
2967 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
2968 xlate_report(ctx, "output to native tunnel");
2969 tnl_push_pop_send = true;
2971 xlate_report(ctx, "output to kernel tunnel");
2972 commit_odp_tunnel_action(flow, &ctx->base_flow, ctx->odp_actions);
2973 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
2976 odp_port = xport->odp_port;
2977 out_port = odp_port;
2978 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
2979 ofp_port_t vlandev_port;
2981 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
2982 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto,
2983 ofp_port, flow->vlan_tci);
2984 if (vlandev_port != ofp_port) {
2985 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
2986 flow->vlan_tci = htons(0);
2991 if (out_port != ODPP_NONE) {
2992 bool use_masked = ctx->xbridge->support.masked_set_action;
2994 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
2999 struct ovs_action_hash *act_hash;
3002 act_hash = nl_msg_put_unspec_uninit(ctx->odp_actions,
3003 OVS_ACTION_ATTR_HASH,
3005 act_hash->hash_alg = xr->hash_alg;
3006 act_hash->hash_basis = xr->hash_basis;
3008 /* Recirc action. */
3009 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC,
3013 if (tnl_push_pop_send) {
3014 build_tunnel_send(ctx, xport, flow, odp_port);
3015 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
3017 odp_port_t odp_tnl_port = ODPP_NONE;
3019 /* XXX: Write better Filter for tunnel port. We can use inport
3020 * int tunnel-port flow to avoid these checks completely. */
3021 if (ofp_port == OFPP_LOCAL &&
3022 ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
3024 odp_tnl_port = tnl_port_map_lookup(flow, wc);
3027 if (odp_tnl_port != ODPP_NONE) {
3028 nl_msg_put_odp_port(ctx->odp_actions,
3029 OVS_ACTION_ATTR_TUNNEL_POP,
3032 /* Tunnel push-pop action is not compatible with
3034 compose_ipfix_action(ctx, out_port);
3035 nl_msg_put_odp_port(ctx->odp_actions,
3036 OVS_ACTION_ATTR_OUTPUT,
3042 ctx->sflow_odp_port = odp_port;
3043 ctx->sflow_n_outputs++;
3044 ctx->xout->nf_output_iface = ofp_port;
3049 flow->vlan_tci = flow_vlan_tci;
3050 flow->pkt_mark = flow_pkt_mark;
3051 flow->nw_tos = flow_nw_tos;
3055 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port,
3056 const struct xlate_bond_recirc *xr)
3058 compose_output_action__(ctx, ofp_port, xr, true);
3062 xlate_recursively(struct xlate_ctx *ctx, struct rule_dpif *rule)
3064 struct rule_dpif *old_rule = ctx->rule;
3065 ovs_be64 old_cookie = ctx->rule_cookie;
3066 const struct rule_actions *actions;
3068 if (ctx->xin->resubmit_stats) {
3069 rule_dpif_credit_stats(rule, ctx->xin->resubmit_stats);
3075 ctx->rule_cookie = rule_dpif_get_flow_cookie(rule);
3076 actions = rule_dpif_get_actions(rule);
3077 do_xlate_actions(actions->ofpacts, actions->ofpacts_len, ctx);
3078 ctx->rule_cookie = old_cookie;
3079 ctx->rule = old_rule;
3084 xlate_resubmit_resource_check(struct xlate_ctx *ctx)
3086 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3088 if (ctx->recurse >= MAX_RESUBMIT_RECURSION + MAX_INTERNAL_RESUBMITS) {
3089 VLOG_ERR_RL(&rl, "resubmit actions recursed over %d times",
3090 MAX_RESUBMIT_RECURSION);
3091 } else if (ctx->resubmits >= MAX_RESUBMITS + MAX_INTERNAL_RESUBMITS) {
3092 VLOG_ERR_RL(&rl, "over %d resubmit actions", MAX_RESUBMITS);
3093 } else if (ctx->odp_actions->size > UINT16_MAX) {
3094 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of actions");
3095 } else if (ctx->stack.size >= 65536) {
3096 VLOG_ERR_RL(&rl, "resubmits yielded over 64 kB of stack");
3105 xlate_table_action(struct xlate_ctx *ctx, ofp_port_t in_port, uint8_t table_id,
3106 bool may_packet_in, bool honor_table_miss)
3108 /* Check if we need to recirculate before matching in a table. */
3109 if (ctx->was_mpls) {
3110 ctx_trigger_recirculation(ctx);
3113 if (xlate_resubmit_resource_check(ctx)) {
3114 uint8_t old_table_id = ctx->table_id;
3115 struct rule_dpif *rule;
3117 ctx->table_id = table_id;
3119 rule = rule_dpif_lookup_from_table(ctx->xbridge->ofproto,
3120 ctx->tables_version,
3121 &ctx->xin->flow, ctx->xin->wc,
3122 ctx->xin->xcache != NULL,
3123 ctx->xin->resubmit_stats,
3124 &ctx->table_id, in_port,
3125 may_packet_in, honor_table_miss);
3127 if (OVS_UNLIKELY(ctx->xin->resubmit_hook)) {
3128 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse + 1);
3132 /* Fill in the cache entry here instead of xlate_recursively
3133 * to make the reference counting more explicit. We take a
3134 * reference in the lookups above if we are going to cache the
3136 if (ctx->xin->xcache) {
3137 struct xc_entry *entry;
3139 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_RULE);
3140 entry->u.rule = rule;
3142 xlate_recursively(ctx, rule);
3145 ctx->table_id = old_table_id;
3153 xlate_group_stats(struct xlate_ctx *ctx, struct group_dpif *group,
3154 struct ofputil_bucket *bucket)
3156 if (ctx->xin->resubmit_stats) {
3157 group_dpif_credit_stats(group, bucket, ctx->xin->resubmit_stats);
3159 if (ctx->xin->xcache) {
3160 struct xc_entry *entry;
3162 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_GROUP);
3163 entry->u.group.group = group_dpif_ref(group);
3164 entry->u.group.bucket = bucket;
3169 xlate_group_bucket(struct xlate_ctx *ctx, struct ofputil_bucket *bucket)
3171 uint64_t action_list_stub[1024 / 8];
3172 struct ofpbuf action_list, action_set;
3173 struct flow old_flow = ctx->xin->flow;
3174 bool old_was_mpls = ctx->was_mpls;
3176 ofpbuf_use_const(&action_set, bucket->ofpacts, bucket->ofpacts_len);
3177 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3179 ofpacts_execute_action_set(&action_list, &action_set);
3181 do_xlate_actions(action_list.data, action_list.size, ctx);
3184 ofpbuf_uninit(&action_set);
3185 ofpbuf_uninit(&action_list);
3187 /* Check if need to recirculate. */
3188 if (exit_recirculates(ctx)) {
3189 compose_recirculate_action(ctx);
3192 /* Roll back flow to previous state.
3193 * This is equivalent to cloning the packet for each bucket.
3195 * As a side effect any subsequently applied actions will
3196 * also effectively be applied to a clone of the packet taken
3197 * just before applying the all or indirect group.
3199 * Note that group buckets are action sets, hence they cannot modify the
3200 * main action set. Also any stack actions are ignored when executing an
3201 * action set, so group buckets cannot change the stack either.
3202 * However, we do allow resubmit actions in group buckets, which could
3203 * break the above assumptions. It is up to the controller to not mess up
3204 * with the action_set and stack in the tables resubmitted to from
3206 ctx->xin->flow = old_flow;
3208 /* The group bucket popping MPLS should have no effect after bucket
3210 ctx->was_mpls = old_was_mpls;
3212 /* The fact that the group bucket exits (for any reason) does not mean that
3213 * the translation after the group action should exit. Specifically, if
3214 * the group bucket recirculates (which typically modifies the packet), the
3215 * actions after the group action must continue processing with the
3216 * original, not the recirculated packet! */
3221 xlate_all_group(struct xlate_ctx *ctx, struct group_dpif *group)
3223 struct ofputil_bucket *bucket;
3224 const struct ovs_list *buckets;
3226 group_dpif_get_buckets(group, &buckets);
3228 LIST_FOR_EACH (bucket, list_node, buckets) {
3229 xlate_group_bucket(ctx, bucket);
3231 xlate_group_stats(ctx, group, NULL);
3235 xlate_ff_group(struct xlate_ctx *ctx, struct group_dpif *group)
3237 struct ofputil_bucket *bucket;
3239 bucket = group_first_live_bucket(ctx, group, 0);
3241 xlate_group_bucket(ctx, bucket);
3242 xlate_group_stats(ctx, group, bucket);
3247 xlate_default_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3249 struct flow_wildcards *wc = ctx->wc;
3250 struct ofputil_bucket *bucket;
3253 basis = flow_hash_symmetric_l4(&ctx->xin->flow, 0);
3254 flow_mask_hash_fields(&ctx->xin->flow, wc, NX_HASH_FIELDS_SYMMETRIC_L4);
3255 bucket = group_best_live_bucket(ctx, group, basis);
3257 xlate_group_bucket(ctx, bucket);
3258 xlate_group_stats(ctx, group, bucket);
3263 xlate_hash_fields_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3265 struct mf_bitmap hash_fields = MF_BITMAP_INITIALIZER;
3266 const struct field_array *fields;
3267 struct ofputil_bucket *bucket;
3271 fields = group_dpif_get_fields(group);
3272 basis = hash_uint64(group_dpif_get_selection_method_param(group));
3274 /* Determine which fields to hash */
3275 for (i = 0; i < MFF_N_IDS; i++) {
3276 if (bitmap_is_set(fields->used.bm, i)) {
3277 const struct mf_field *mf;
3279 /* If the field is already present in 'hash_fields' then
3280 * this loop has already checked that it and its pre-requisites
3281 * are present in the flow and its pre-requisites have
3282 * already been added to 'hash_fields'. There is nothing more
3283 * to do here and as an optimisation the loop can continue. */
3284 if (bitmap_is_set(hash_fields.bm, i)) {
3290 /* Only hash a field if it and its pre-requisites are present
3292 if (!mf_are_prereqs_ok(mf, &ctx->xin->flow)) {
3296 /* Hash both the field and its pre-requisites */
3297 mf_bitmap_set_field_and_prereqs(mf, &hash_fields);
3301 /* Hash the fields */
3302 for (i = 0; i < MFF_N_IDS; i++) {
3303 if (bitmap_is_set(hash_fields.bm, i)) {
3304 const struct mf_field *mf = mf_from_id(i);
3305 union mf_value value;
3308 mf_get_value(mf, &ctx->xin->flow, &value);
3309 /* This seems inefficient but so does apply_mask() */
3310 for (j = 0; j < mf->n_bytes; j++) {
3311 ((uint8_t *) &value)[j] &= ((uint8_t *) &fields->value[i])[j];
3313 basis = hash_bytes(&value, mf->n_bytes, basis);
3315 mf_mask_field(mf, &ctx->wc->masks);
3319 bucket = group_best_live_bucket(ctx, group, basis);
3321 xlate_group_bucket(ctx, bucket);
3322 xlate_group_stats(ctx, group, bucket);
3327 xlate_select_group(struct xlate_ctx *ctx, struct group_dpif *group)
3329 const char *selection_method = group_dpif_get_selection_method(group);
3331 if (selection_method[0] == '\0') {
3332 xlate_default_select_group(ctx, group);
3333 } else if (!strcasecmp("hash", selection_method)) {
3334 xlate_hash_fields_select_group(ctx, group);
3336 /* Parsing of groups should ensure this never happens */
3342 xlate_group_action__(struct xlate_ctx *ctx, struct group_dpif *group)
3344 bool was_in_group = ctx->in_group;
3345 ctx->in_group = true;
3347 switch (group_dpif_get_type(group)) {
3349 case OFPGT11_INDIRECT:
3350 xlate_all_group(ctx, group);
3352 case OFPGT11_SELECT:
3353 xlate_select_group(ctx, group);
3356 xlate_ff_group(ctx, group);
3361 group_dpif_unref(group);
3363 ctx->in_group = was_in_group;
3367 xlate_group_action(struct xlate_ctx *ctx, uint32_t group_id)
3369 if (xlate_resubmit_resource_check(ctx)) {
3370 struct group_dpif *group;
3373 got_group = group_dpif_lookup(ctx->xbridge->ofproto, group_id, &group);
3375 xlate_group_action__(ctx, group);
3385 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
3386 const struct ofpact_resubmit *resubmit)
3390 bool may_packet_in = false;
3391 bool honor_table_miss = false;
3393 if (ctx->rule && rule_dpif_is_internal(ctx->rule)) {
3394 /* Still allow missed packets to be sent to the controller
3395 * if resubmitting from an internal table. */
3396 may_packet_in = true;
3397 honor_table_miss = true;
3400 in_port = resubmit->in_port;
3401 if (in_port == OFPP_IN_PORT) {
3402 in_port = ctx->xin->flow.in_port.ofp_port;
3405 table_id = resubmit->table_id;
3406 if (table_id == 255) {
3407 table_id = ctx->table_id;
3410 xlate_table_action(ctx, in_port, table_id, may_packet_in,
3415 flood_packets(struct xlate_ctx *ctx, bool all)
3417 const struct xport *xport;
3419 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
3420 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
3425 compose_output_action__(ctx, xport->ofp_port, NULL, false);
3426 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
3427 compose_output_action(ctx, xport->ofp_port, NULL);
3431 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3435 execute_controller_action(struct xlate_ctx *ctx, int len,
3436 enum ofp_packet_in_reason reason,
3437 uint16_t controller_id)
3439 struct ofproto_packet_in *pin;
3440 struct dp_packet *packet;
3443 ctx->xout->slow |= SLOW_CONTROLLER;
3444 if (!ctx->xin->packet) {
3448 packet = dp_packet_clone(ctx->xin->packet);
3450 use_masked = ctx->xbridge->support.masked_set_action;
3451 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3453 ctx->wc, use_masked);
3455 odp_execute_actions(NULL, &packet, 1, false,
3456 ctx->odp_actions->data, ctx->odp_actions->size, NULL);
3458 pin = xmalloc(sizeof *pin);
3459 pin->up.packet_len = dp_packet_size(packet);
3460 pin->up.packet = dp_packet_steal_data(packet);
3461 pin->up.reason = reason;
3462 pin->up.table_id = ctx->table_id;
3463 pin->up.cookie = ctx->rule_cookie;
3465 flow_get_metadata(&ctx->xin->flow, &pin->up.flow_metadata);
3467 pin->controller_id = controller_id;
3468 pin->send_len = len;
3469 /* If a rule is a table-miss rule then this is
3470 * a table-miss handled by a table-miss rule.
3472 * Else, if rule is internal and has a controller action,
3473 * the later being implied by the rule being processed here,
3474 * then this is a table-miss handled without a table-miss rule.
3476 * Otherwise this is not a table-miss. */
3477 pin->miss_type = OFPROTO_PACKET_IN_NO_MISS;
3479 if (rule_dpif_is_table_miss(ctx->rule)) {
3480 pin->miss_type = OFPROTO_PACKET_IN_MISS_FLOW;
3481 } else if (rule_dpif_is_internal(ctx->rule)) {
3482 pin->miss_type = OFPROTO_PACKET_IN_MISS_WITHOUT_FLOW;
3485 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, pin);
3486 dp_packet_delete(packet);
3489 /* Called only when ctx->recirc_action_offset is set. */
3491 compose_recirculate_action(struct xlate_ctx *ctx)
3493 struct recirc_metadata md;
3497 use_masked = ctx->xbridge->support.masked_set_action;
3498 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3500 ctx->wc, use_masked);
3502 recirc_metadata_from_flow(&md, &ctx->xin->flow);
3504 ovs_assert(ctx->recirc_action_offset >= 0);
3506 /* Only allocate recirculation ID if we have a packet. */
3507 if (ctx->xin->packet) {
3508 /* Allocate a unique recirc id for the given metadata state in the
3509 * flow. The life-cycle of this recirc id is managed by associating it
3510 * with the udpif key ('ukey') created for each new datapath flow. */
3511 id = recirc_alloc_id_ctx(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3512 ctx->recirc_action_offset,
3513 ctx->action_set.size, ctx->action_set.data);
3515 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3516 VLOG_ERR_RL(&rl, "Failed to allocate recirculation id");
3519 xlate_out_add_recirc(ctx->xout, id);
3521 /* Look up an existing recirc id for the given metadata state in the
3522 * flow. No new reference is taken, as the ID is RCU protected and is
3523 * only required temporarily for verification. */
3524 id = recirc_find_id(ctx->xbridge->ofproto, 0, &md, &ctx->stack,
3525 ctx->recirc_action_offset,
3526 ctx->action_set.size, ctx->action_set.data);
3527 /* We let zero 'id' to be used in the RECIRC action below, which will
3528 * fail all revalidations as zero is not a valid recirculation ID. */
3531 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_RECIRC, id);
3533 /* Undo changes done by recirculation. */
3534 ctx->action_set.size = ctx->recirc_action_offset;
3535 ctx->recirc_action_offset = -1;
3536 ctx->last_unroll_offset = -1;
3540 compose_mpls_push_action(struct xlate_ctx *ctx, struct ofpact_push_mpls *mpls)
3542 struct flow *flow = &ctx->xin->flow;
3545 ovs_assert(eth_type_mpls(mpls->ethertype));
3547 n = flow_count_mpls_labels(flow, ctx->wc);
3549 bool use_masked = ctx->xbridge->support.masked_set_action;
3551 ctx->xout->slow |= commit_odp_actions(flow, &ctx->base_flow,
3553 ctx->wc, use_masked);
3554 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3555 if (ctx->xin->packet != NULL) {
3556 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3557 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3558 "MPLS push action can't be performed as it would "
3559 "have more MPLS LSEs than the %d supported.",
3560 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3566 flow_push_mpls(flow, n, mpls->ethertype, ctx->wc);
3570 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
3572 struct flow *flow = &ctx->xin->flow;
3573 int n = flow_count_mpls_labels(flow, ctx->wc);
3575 if (flow_pop_mpls(flow, n, eth_type, ctx->wc)) {
3576 if (ctx->xbridge->support.odp.recirc) {
3577 ctx->was_mpls = true;
3579 } else if (n >= FLOW_MAX_MPLS_LABELS) {
3580 if (ctx->xin->packet != NULL) {
3581 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
3582 VLOG_WARN_RL(&rl, "bridge %s: dropping packet on which an "
3583 "MPLS pop action can't be performed as it has "
3584 "more MPLS LSEs than the %d supported.",
3585 ctx->xbridge->name, FLOW_MAX_MPLS_LABELS);
3588 ofpbuf_clear(ctx->odp_actions);
3593 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
3595 struct flow *flow = &ctx->xin->flow;
3597 if (!is_ip_any(flow)) {
3601 ctx->wc->masks.nw_ttl = 0xff;
3602 if (flow->nw_ttl > 1) {
3608 for (i = 0; i < ids->n_controllers; i++) {
3609 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
3613 /* Stop processing for current table. */
3619 compose_set_mpls_label_action(struct xlate_ctx *ctx, ovs_be32 label)
3621 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3622 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_LABEL_MASK);
3623 set_mpls_lse_label(&ctx->xin->flow.mpls_lse[0], label);
3628 compose_set_mpls_tc_action(struct xlate_ctx *ctx, uint8_t tc)
3630 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3631 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TC_MASK);
3632 set_mpls_lse_tc(&ctx->xin->flow.mpls_lse[0], tc);
3637 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
3639 if (eth_type_mpls(ctx->xin->flow.dl_type)) {
3640 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3641 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse[0], ttl);
3646 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
3648 struct flow *flow = &ctx->xin->flow;
3650 if (eth_type_mpls(flow->dl_type)) {
3651 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse[0]);
3653 ctx->wc->masks.mpls_lse[0] |= htonl(MPLS_TTL_MASK);
3656 set_mpls_lse_ttl(&flow->mpls_lse[0], ttl);
3659 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
3663 /* Stop processing for current table. */
3668 xlate_output_action(struct xlate_ctx *ctx,
3669 ofp_port_t port, uint16_t max_len, bool may_packet_in)
3671 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
3673 ctx->xout->nf_output_iface = NF_OUT_DROP;
3677 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port, NULL);
3680 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
3681 0, may_packet_in, true);
3687 flood_packets(ctx, false);
3690 flood_packets(ctx, true);
3692 case OFPP_CONTROLLER:
3693 execute_controller_action(ctx, max_len,
3694 (ctx->in_group ? OFPR_GROUP
3695 : ctx->in_action_set ? OFPR_ACTION_SET
3703 if (port != ctx->xin->flow.in_port.ofp_port) {
3704 compose_output_action(ctx, port, NULL);
3706 xlate_report(ctx, "skipping output to input port");
3711 if (prev_nf_output_iface == NF_OUT_FLOOD) {
3712 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
3713 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3714 ctx->xout->nf_output_iface = prev_nf_output_iface;
3715 } else if (prev_nf_output_iface != NF_OUT_DROP &&
3716 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3717 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3722 xlate_output_reg_action(struct xlate_ctx *ctx,
3723 const struct ofpact_output_reg *or)
3725 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
3726 if (port <= UINT16_MAX) {
3727 union mf_subvalue value;
3729 memset(&value, 0xff, sizeof value);
3730 mf_write_subfield_flow(&or->src, &value, &ctx->wc->masks);
3731 xlate_output_action(ctx, u16_to_ofp(port),
3732 or->max_len, false);
3737 xlate_enqueue_action(struct xlate_ctx *ctx,
3738 const struct ofpact_enqueue *enqueue)
3740 ofp_port_t ofp_port = enqueue->port;
3741 uint32_t queue_id = enqueue->queue;
3742 uint32_t flow_priority, priority;
3745 /* Translate queue to priority. */
3746 error = dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &priority);
3748 /* Fall back to ordinary output action. */
3749 xlate_output_action(ctx, enqueue->port, 0, false);
3753 /* Check output port. */
3754 if (ofp_port == OFPP_IN_PORT) {
3755 ofp_port = ctx->xin->flow.in_port.ofp_port;
3756 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
3760 /* Add datapath actions. */
3761 flow_priority = ctx->xin->flow.skb_priority;
3762 ctx->xin->flow.skb_priority = priority;
3763 compose_output_action(ctx, ofp_port, NULL);
3764 ctx->xin->flow.skb_priority = flow_priority;
3766 /* Update NetFlow output port. */
3767 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
3768 ctx->xout->nf_output_iface = ofp_port;
3769 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
3770 ctx->xout->nf_output_iface = NF_OUT_MULTI;
3775 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
3777 uint32_t skb_priority;
3779 if (!dpif_queue_to_priority(ctx->xbridge->dpif, queue_id, &skb_priority)) {
3780 ctx->xin->flow.skb_priority = skb_priority;
3782 /* Couldn't translate queue to a priority. Nothing to do. A warning
3783 * has already been logged. */
3788 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
3790 const struct xbridge *xbridge = xbridge_;
3801 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
3804 port = get_ofp_port(xbridge, ofp_port);
3805 return port ? port->may_enable : false;
3810 xlate_bundle_action(struct xlate_ctx *ctx,
3811 const struct ofpact_bundle *bundle)
3815 port = bundle_execute(bundle, &ctx->xin->flow, ctx->wc, slave_enabled_cb,
3816 CONST_CAST(struct xbridge *, ctx->xbridge));
3817 if (bundle->dst.field) {
3818 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow, ctx->wc);
3820 xlate_output_action(ctx, port, 0, false);
3825 xlate_learn_action__(struct xlate_ctx *ctx, const struct ofpact_learn *learn,
3826 struct ofputil_flow_mod *fm, struct ofpbuf *ofpacts)
3828 learn_execute(learn, &ctx->xin->flow, fm, ofpacts);
3829 if (ctx->xin->may_learn) {
3830 ofproto_dpif_flow_mod(ctx->xbridge->ofproto, fm);
3835 xlate_learn_action(struct xlate_ctx *ctx, const struct ofpact_learn *learn)
3837 ctx->xout->has_learn = true;
3838 learn_mask(learn, ctx->wc);
3840 if (ctx->xin->xcache) {
3841 struct xc_entry *entry;
3843 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_LEARN);
3844 entry->u.learn.ofproto = ctx->xbridge->ofproto;
3845 entry->u.learn.fm = xmalloc(sizeof *entry->u.learn.fm);
3846 entry->u.learn.ofpacts = ofpbuf_new(64);
3847 xlate_learn_action__(ctx, learn, entry->u.learn.fm,
3848 entry->u.learn.ofpacts);
3849 } else if (ctx->xin->may_learn) {
3850 uint64_t ofpacts_stub[1024 / 8];
3851 struct ofputil_flow_mod fm;
3852 struct ofpbuf ofpacts;
3854 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3855 xlate_learn_action__(ctx, learn, &fm, &ofpacts);
3856 ofpbuf_uninit(&ofpacts);
3861 xlate_fin_timeout__(struct rule_dpif *rule, uint16_t tcp_flags,
3862 uint16_t idle_timeout, uint16_t hard_timeout)
3864 if (tcp_flags & (TCP_FIN | TCP_RST)) {
3865 rule_dpif_reduce_timeouts(rule, idle_timeout, hard_timeout);
3870 xlate_fin_timeout(struct xlate_ctx *ctx,
3871 const struct ofpact_fin_timeout *oft)
3874 xlate_fin_timeout__(ctx->rule, ctx->xin->tcp_flags,
3875 oft->fin_idle_timeout, oft->fin_hard_timeout);
3876 if (ctx->xin->xcache) {
3877 struct xc_entry *entry;
3879 entry = xlate_cache_add_entry(ctx->xin->xcache, XC_FIN_TIMEOUT);
3880 /* XC_RULE already holds a reference on the rule, none is taken
3882 entry->u.fin.rule = ctx->rule;
3883 entry->u.fin.idle = oft->fin_idle_timeout;
3884 entry->u.fin.hard = oft->fin_hard_timeout;
3890 xlate_sample_action(struct xlate_ctx *ctx,
3891 const struct ofpact_sample *os)
3893 /* Scale the probability from 16-bit to 32-bit while representing
3894 * the same percentage. */
3895 uint32_t probability = (os->probability << 16) | os->probability;
3898 if (!ctx->xbridge->support.variable_length_userdata) {
3899 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
3901 VLOG_ERR_RL(&rl, "ignoring NXAST_SAMPLE action because datapath "
3902 "lacks support (needs Linux 3.10+ or kernel module from "
3907 use_masked = ctx->xbridge->support.masked_set_action;
3908 ctx->xout->slow |= commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
3910 ctx->wc, use_masked);
3912 union user_action_cookie cookie = {
3914 .type = USER_ACTION_COOKIE_FLOW_SAMPLE,
3915 .probability = os->probability,
3916 .collector_set_id = os->collector_set_id,
3917 .obs_domain_id = os->obs_domain_id,
3918 .obs_point_id = os->obs_point_id,
3921 compose_sample_action(ctx, probability, &cookie, sizeof cookie.flow_sample,
3926 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
3928 if (xport->config & (is_stp(&ctx->xin->flow)
3929 ? OFPUTIL_PC_NO_RECV_STP
3930 : OFPUTIL_PC_NO_RECV)) {
3934 /* Only drop packets here if both forwarding and learning are
3935 * disabled. If just learning is enabled, we need to have
3936 * OFPP_NORMAL and the learning action have a look at the packet
3937 * before we can drop it. */
3938 if ((!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) ||
3939 (!xport_rstp_forward_state(xport) && !xport_rstp_learn_state(xport))) {
3947 xlate_write_actions(struct xlate_ctx *ctx, const struct ofpact *a)
3949 const struct ofpact_nest *on = ofpact_get_WRITE_ACTIONS(a);
3950 size_t on_len = ofpact_nest_get_action_len(on);
3951 const struct ofpact *inner;
3953 /* Maintain actset_output depending on the contents of the action set:
3955 * - OFPP_UNSET, if there is no "output" action.
3957 * - The output port, if there is an "output" action and no "group"
3960 * - OFPP_UNSET, if there is a "group" action.
3962 if (!ctx->action_set_has_group) {
3963 OFPACT_FOR_EACH (inner, on->actions, on_len) {
3964 if (inner->type == OFPACT_OUTPUT) {
3965 ctx->xin->flow.actset_output = ofpact_get_OUTPUT(inner)->port;
3966 } else if (inner->type == OFPACT_GROUP) {
3967 ctx->xin->flow.actset_output = OFPP_UNSET;
3968 ctx->action_set_has_group = true;
3974 ofpbuf_put(&ctx->action_set, on->actions, on_len);
3975 ofpact_pad(&ctx->action_set);
3979 xlate_action_set(struct xlate_ctx *ctx)
3981 uint64_t action_list_stub[1024 / 64];
3982 struct ofpbuf action_list;
3984 ctx->in_action_set = true;
3985 ofpbuf_use_stub(&action_list, action_list_stub, sizeof action_list_stub);
3986 ofpacts_execute_action_set(&action_list, &ctx->action_set);
3987 /* Clear the action set, as it is not needed any more. */
3988 ofpbuf_clear(&ctx->action_set);
3989 do_xlate_actions(action_list.data, action_list.size, ctx);
3990 ctx->in_action_set = false;
3991 ofpbuf_uninit(&action_list);
3995 recirc_put_unroll_xlate(struct xlate_ctx *ctx)
3997 struct ofpact_unroll_xlate *unroll;
3999 unroll = ctx->last_unroll_offset < 0
4001 : ALIGNED_CAST(struct ofpact_unroll_xlate *,
4002 (char *)ctx->action_set.data + ctx->last_unroll_offset);
4004 /* Restore the table_id and rule cookie for a potential PACKET
4007 (ctx->table_id != unroll->rule_table_id
4008 || ctx->rule_cookie != unroll->rule_cookie)) {
4010 ctx->last_unroll_offset = ctx->action_set.size;
4011 unroll = ofpact_put_UNROLL_XLATE(&ctx->action_set);
4012 unroll->rule_table_id = ctx->table_id;
4013 unroll->rule_cookie = ctx->rule_cookie;
4018 /* Copy remaining actions to the action_set to be executed after recirculation.
4019 * UNROLL_XLATE action is inserted, if not already done so, before actions that
4020 * may generate PACKET_INs from the current table and without matching another
4023 recirc_unroll_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4024 struct xlate_ctx *ctx)
4026 const struct ofpact *a;
4028 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4030 /* May generate PACKET INs. */
4031 case OFPACT_OUTPUT_REG:
4034 case OFPACT_CONTROLLER:
4035 case OFPACT_DEC_MPLS_TTL:
4036 case OFPACT_DEC_TTL:
4037 recirc_put_unroll_xlate(ctx);
4040 /* These may not generate PACKET INs. */
4041 case OFPACT_SET_TUNNEL:
4042 case OFPACT_REG_MOVE:
4043 case OFPACT_SET_FIELD:
4044 case OFPACT_STACK_PUSH:
4045 case OFPACT_STACK_POP:
4047 case OFPACT_WRITE_METADATA:
4048 case OFPACT_RESUBMIT: /* May indirectly generate PACKET INs, */
4049 case OFPACT_GOTO_TABLE: /* but from a different table and rule. */
4050 case OFPACT_ENQUEUE:
4051 case OFPACT_SET_VLAN_VID:
4052 case OFPACT_SET_VLAN_PCP:
4053 case OFPACT_STRIP_VLAN:
4054 case OFPACT_PUSH_VLAN:
4055 case OFPACT_SET_ETH_SRC:
4056 case OFPACT_SET_ETH_DST:
4057 case OFPACT_SET_IPV4_SRC:
4058 case OFPACT_SET_IPV4_DST:
4059 case OFPACT_SET_IP_DSCP:
4060 case OFPACT_SET_IP_ECN:
4061 case OFPACT_SET_IP_TTL:
4062 case OFPACT_SET_L4_SRC_PORT:
4063 case OFPACT_SET_L4_DST_PORT:
4064 case OFPACT_SET_QUEUE:
4065 case OFPACT_POP_QUEUE:
4066 case OFPACT_PUSH_MPLS:
4067 case OFPACT_POP_MPLS:
4068 case OFPACT_SET_MPLS_LABEL:
4069 case OFPACT_SET_MPLS_TC:
4070 case OFPACT_SET_MPLS_TTL:
4071 case OFPACT_MULTIPATH:
4074 case OFPACT_UNROLL_XLATE:
4075 case OFPACT_FIN_TIMEOUT:
4076 case OFPACT_CLEAR_ACTIONS:
4077 case OFPACT_WRITE_ACTIONS:
4082 /* These need not be copied for restoration. */
4084 case OFPACT_CONJUNCTION:
4087 /* Copy the action over. */
4088 ofpbuf_put(&ctx->action_set, a, OFPACT_ALIGN(a->len));
4092 #define CHECK_MPLS_RECIRCULATION() \
4093 if (ctx->was_mpls) { \
4094 ctx_trigger_recirculation(ctx); \
4097 #define CHECK_MPLS_RECIRCULATION_IF(COND) \
4099 CHECK_MPLS_RECIRCULATION(); \
4103 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
4104 struct xlate_ctx *ctx)
4106 struct flow_wildcards *wc = ctx->wc;
4107 struct flow *flow = &ctx->xin->flow;
4108 const struct ofpact *a;
4110 if (ovs_native_tunneling_is_on(ctx->xbridge->ofproto)) {
4111 tnl_arp_snoop(flow, wc, ctx->xbridge->name);
4113 /* dl_type already in the mask, not set below. */
4115 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
4116 struct ofpact_controller *controller;
4117 const struct ofpact_metadata *metadata;
4118 const struct ofpact_set_field *set_field;
4119 const struct mf_field *mf;
4122 /* Check if need to store the remaining actions for later
4124 if (exit_recirculates(ctx)) {
4125 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4127 (uint8_t *)ofpacts)),
4135 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
4136 ofpact_get_OUTPUT(a)->max_len, true);
4140 if (xlate_group_action(ctx, ofpact_get_GROUP(a)->group_id)) {
4141 /* Group could not be found. */
4146 case OFPACT_CONTROLLER:
4147 controller = ofpact_get_CONTROLLER(a);
4148 execute_controller_action(ctx, controller->max_len,
4150 controller->controller_id);
4153 case OFPACT_ENQUEUE:
4154 memset(&wc->masks.skb_priority, 0xff,
4155 sizeof wc->masks.skb_priority);
4156 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
4159 case OFPACT_SET_VLAN_VID:
4160 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
4161 if (flow->vlan_tci & htons(VLAN_CFI) ||
4162 ofpact_get_SET_VLAN_VID(a)->push_vlan_if_needed) {
4163 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
4164 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
4169 case OFPACT_SET_VLAN_PCP:
4170 wc->masks.vlan_tci |= htons(VLAN_PCP_MASK | VLAN_CFI);
4171 if (flow->vlan_tci & htons(VLAN_CFI) ||
4172 ofpact_get_SET_VLAN_PCP(a)->push_vlan_if_needed) {
4173 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
4174 flow->vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
4175 << VLAN_PCP_SHIFT) | VLAN_CFI);
4179 case OFPACT_STRIP_VLAN:
4180 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4181 flow->vlan_tci = htons(0);
4184 case OFPACT_PUSH_VLAN:
4185 /* XXX 802.1AD(QinQ) */
4186 memset(&wc->masks.vlan_tci, 0xff, sizeof wc->masks.vlan_tci);
4187 flow->vlan_tci = htons(VLAN_CFI);
4190 case OFPACT_SET_ETH_SRC:
4191 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
4192 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
4195 case OFPACT_SET_ETH_DST:
4196 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
4197 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
4200 case OFPACT_SET_IPV4_SRC:
4201 CHECK_MPLS_RECIRCULATION();
4202 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4203 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
4204 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
4208 case OFPACT_SET_IPV4_DST:
4209 CHECK_MPLS_RECIRCULATION();
4210 if (flow->dl_type == htons(ETH_TYPE_IP)) {
4211 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
4212 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
4216 case OFPACT_SET_IP_DSCP:
4217 CHECK_MPLS_RECIRCULATION();
4218 if (is_ip_any(flow)) {
4219 wc->masks.nw_tos |= IP_DSCP_MASK;
4220 flow->nw_tos &= ~IP_DSCP_MASK;
4221 flow->nw_tos |= ofpact_get_SET_IP_DSCP(a)->dscp;
4225 case OFPACT_SET_IP_ECN:
4226 CHECK_MPLS_RECIRCULATION();
4227 if (is_ip_any(flow)) {
4228 wc->masks.nw_tos |= IP_ECN_MASK;
4229 flow->nw_tos &= ~IP_ECN_MASK;
4230 flow->nw_tos |= ofpact_get_SET_IP_ECN(a)->ecn;
4234 case OFPACT_SET_IP_TTL:
4235 CHECK_MPLS_RECIRCULATION();
4236 if (is_ip_any(flow)) {
4237 wc->masks.nw_ttl = 0xff;
4238 flow->nw_ttl = ofpact_get_SET_IP_TTL(a)->ttl;
4242 case OFPACT_SET_L4_SRC_PORT:
4243 CHECK_MPLS_RECIRCULATION();
4244 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4245 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4246 memset(&wc->masks.tp_src, 0xff, sizeof wc->masks.tp_src);
4247 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
4251 case OFPACT_SET_L4_DST_PORT:
4252 CHECK_MPLS_RECIRCULATION();
4253 if (is_ip_any(flow) && !(flow->nw_frag & FLOW_NW_FRAG_LATER)) {
4254 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4255 memset(&wc->masks.tp_dst, 0xff, sizeof wc->masks.tp_dst);
4256 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
4260 case OFPACT_RESUBMIT:
4261 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
4264 case OFPACT_SET_TUNNEL:
4265 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
4268 case OFPACT_SET_QUEUE:
4269 memset(&wc->masks.skb_priority, 0xff,
4270 sizeof wc->masks.skb_priority);
4271 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
4274 case OFPACT_POP_QUEUE:
4275 memset(&wc->masks.skb_priority, 0xff,
4276 sizeof wc->masks.skb_priority);
4277 flow->skb_priority = ctx->orig_skb_priority;
4280 case OFPACT_REG_MOVE:
4281 CHECK_MPLS_RECIRCULATION_IF(
4282 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->dst.field) ||
4283 mf_is_l3_or_higher(ofpact_get_REG_MOVE(a)->src.field));
4284 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
4287 case OFPACT_SET_FIELD:
4288 CHECK_MPLS_RECIRCULATION_IF(
4289 mf_is_l3_or_higher(ofpact_get_SET_FIELD(a)->field));
4290 set_field = ofpact_get_SET_FIELD(a);
4291 mf = set_field->field;
4293 /* Set field action only ever overwrites packet's outermost
4294 * applicable header fields. Do nothing if no header exists. */
4295 if (mf->id == MFF_VLAN_VID) {
4296 wc->masks.vlan_tci |= htons(VLAN_CFI);
4297 if (!(flow->vlan_tci & htons(VLAN_CFI))) {
4300 } else if ((mf->id == MFF_MPLS_LABEL || mf->id == MFF_MPLS_TC)
4301 /* 'dl_type' is already unwildcarded. */
4302 && !eth_type_mpls(flow->dl_type)) {
4305 /* A flow may wildcard nw_frag. Do nothing if setting a trasport
4306 * header field on a packet that does not have them. */
4307 mf_mask_field_and_prereqs(mf, &wc->masks);
4308 if (mf_are_prereqs_ok(mf, flow)) {
4309 mf_set_flow_value_masked(mf, &set_field->value,
4310 &set_field->mask, flow);
4314 case OFPACT_STACK_PUSH:
4315 CHECK_MPLS_RECIRCULATION_IF(
4316 mf_is_l3_or_higher(ofpact_get_STACK_PUSH(a)->subfield.field));
4317 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
4321 case OFPACT_STACK_POP:
4322 CHECK_MPLS_RECIRCULATION_IF(
4323 mf_is_l3_or_higher(ofpact_get_STACK_POP(a)->subfield.field));
4324 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, wc,
4328 case OFPACT_PUSH_MPLS:
4329 /* Recirculate if it is an IP packet with a zero ttl. This may
4330 * indicate that the packet was previously MPLS and an MPLS pop
4331 * action converted it to IP. In this case recirculating should
4332 * reveal the IP TTL which is used as the basis for a new MPLS
4334 CHECK_MPLS_RECIRCULATION_IF(
4335 !flow_count_mpls_labels(flow, wc)
4336 && flow->nw_ttl == 0
4337 && is_ip_any(flow));
4338 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a));
4341 case OFPACT_POP_MPLS:
4342 CHECK_MPLS_RECIRCULATION();
4343 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
4346 case OFPACT_SET_MPLS_LABEL:
4347 CHECK_MPLS_RECIRCULATION();
4348 compose_set_mpls_label_action(
4349 ctx, ofpact_get_SET_MPLS_LABEL(a)->label);
4352 case OFPACT_SET_MPLS_TC:
4353 CHECK_MPLS_RECIRCULATION();
4354 compose_set_mpls_tc_action(ctx, ofpact_get_SET_MPLS_TC(a)->tc);
4357 case OFPACT_SET_MPLS_TTL:
4358 CHECK_MPLS_RECIRCULATION();
4359 compose_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl);
4362 case OFPACT_DEC_MPLS_TTL:
4363 CHECK_MPLS_RECIRCULATION();
4364 if (compose_dec_mpls_ttl_action(ctx)) {
4369 case OFPACT_DEC_TTL:
4370 CHECK_MPLS_RECIRCULATION();
4371 wc->masks.nw_ttl = 0xff;
4372 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
4378 /* Nothing to do. */
4381 case OFPACT_MULTIPATH:
4382 CHECK_MPLS_RECIRCULATION();
4383 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
4387 CHECK_MPLS_RECIRCULATION();
4388 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
4391 case OFPACT_OUTPUT_REG:
4392 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
4396 CHECK_MPLS_RECIRCULATION();
4397 xlate_learn_action(ctx, ofpact_get_LEARN(a));
4400 case OFPACT_CONJUNCTION: {
4401 /* A flow with a "conjunction" action represents part of a special
4402 * kind of "set membership match". Such a flow should not actually
4403 * get executed, but it could via, say, a "packet-out", even though
4404 * that wouldn't be useful. Log it to help debugging. */
4405 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4406 VLOG_INFO_RL(&rl, "executing no-op conjunction action");
4414 case OFPACT_UNROLL_XLATE: {
4415 struct ofpact_unroll_xlate *unroll = ofpact_get_UNROLL_XLATE(a);
4417 /* Restore translation context data that was stored earlier. */
4418 ctx->table_id = unroll->rule_table_id;
4419 ctx->rule_cookie = unroll->rule_cookie;
4422 case OFPACT_FIN_TIMEOUT:
4423 CHECK_MPLS_RECIRCULATION();
4424 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
4425 ctx->xout->has_fin_timeout = true;
4426 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
4429 case OFPACT_CLEAR_ACTIONS:
4430 ofpbuf_clear(&ctx->action_set);
4431 ctx->xin->flow.actset_output = OFPP_UNSET;
4432 ctx->action_set_has_group = false;
4435 case OFPACT_WRITE_ACTIONS:
4436 xlate_write_actions(ctx, a);
4439 case OFPACT_WRITE_METADATA:
4440 metadata = ofpact_get_WRITE_METADATA(a);
4441 flow->metadata &= ~metadata->mask;
4442 flow->metadata |= metadata->metadata & metadata->mask;
4446 /* Not implemented yet. */
4449 case OFPACT_GOTO_TABLE: {
4450 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
4452 /* Allow ctx->table_id == TBL_INTERNAL, which will be greater
4453 * than ogt->table_id. This is to allow goto_table actions that
4454 * triggered recirculation: ctx->table_id will be TBL_INTERNAL
4455 * after recirculation. */
4456 ovs_assert(ctx->table_id == TBL_INTERNAL
4457 || ctx->table_id < ogt->table_id);
4458 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
4459 ogt->table_id, true, true);
4464 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
4468 /* Check if need to store this and the remaining actions for later
4470 if (ctx->exit && ctx_first_recirculation_action(ctx)) {
4471 recirc_unroll_actions(a, OFPACT_ALIGN(ofpacts_len -
4473 (uint8_t *)ofpacts)),
4481 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
4482 const struct flow *flow, ofp_port_t in_port,
4483 struct rule_dpif *rule, uint16_t tcp_flags,
4484 const struct dp_packet *packet, struct flow_wildcards *wc,
4485 struct ofpbuf *odp_actions)
4487 xin->ofproto = ofproto;
4489 xin->flow.in_port.ofp_port = in_port;
4490 xin->flow.actset_output = OFPP_UNSET;
4491 xin->packet = packet;
4492 xin->may_learn = packet != NULL;
4495 xin->ofpacts = NULL;
4496 xin->ofpacts_len = 0;
4497 xin->tcp_flags = tcp_flags;
4498 xin->resubmit_hook = NULL;
4499 xin->report_hook = NULL;
4500 xin->resubmit_stats = NULL;
4502 xin->odp_actions = odp_actions;
4504 /* Do recirc lookup. */
4505 xin->recirc = flow->recirc_id
4506 ? recirc_id_node_find(flow->recirc_id)
4511 xlate_out_uninit(struct xlate_out *xout)
4514 xlate_out_free_recircs(xout);
4518 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
4519 * into datapath actions, using 'ctx', and discards the datapath actions. */
4521 xlate_actions_for_side_effects(struct xlate_in *xin)
4523 struct xlate_out xout;
4525 xlate_actions(xin, &xout);
4526 xlate_out_uninit(&xout);
4529 static struct skb_priority_to_dscp *
4530 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
4532 struct skb_priority_to_dscp *pdscp;
4535 hash = hash_int(skb_priority, 0);
4536 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
4537 if (pdscp->skb_priority == skb_priority) {
4545 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
4548 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
4549 *dscp = pdscp ? pdscp->dscp : 0;
4550 return pdscp != NULL;
4554 count_skb_priorities(const struct xport *xport)
4556 return hmap_count(&xport->skb_priorities);
4560 clear_skb_priorities(struct xport *xport)
4562 struct skb_priority_to_dscp *pdscp, *next;
4564 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
4565 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
4571 actions_output_to_local_port(const struct xlate_ctx *ctx)
4573 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
4574 const struct nlattr *a;
4577 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->odp_actions->data,
4578 ctx->odp_actions->size) {
4579 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
4580 && nl_attr_get_odp_port(a) == local_odp_port) {
4587 #if defined(__linux__)
4588 /* Returns the maximum number of packets that the Linux kernel is willing to
4589 * queue up internally to certain kinds of software-implemented ports, or the
4590 * default (and rarely modified) value if it cannot be determined. */
4592 netdev_max_backlog(void)
4594 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
4595 static int max_backlog = 1000; /* The normal default value. */
4597 if (ovsthread_once_start(&once)) {
4598 static const char filename[] = "/proc/sys/net/core/netdev_max_backlog";
4602 stream = fopen(filename, "r");
4604 VLOG_WARN("%s: open failed (%s)", filename, ovs_strerror(errno));
4606 if (fscanf(stream, "%d", &n) != 1) {
4607 VLOG_WARN("%s: read error", filename);
4608 } else if (n <= 100) {
4609 VLOG_WARN("%s: unexpectedly small value %d", filename, n);
4615 ovsthread_once_done(&once);
4617 VLOG_DBG("%s: using %d max_backlog", filename, max_backlog);
4623 /* Counts and returns the number of OVS_ACTION_ATTR_OUTPUT actions in
4626 count_output_actions(const struct ofpbuf *odp_actions)
4628 const struct nlattr *a;
4632 NL_ATTR_FOR_EACH_UNSAFE (a, left, odp_actions->data, odp_actions->size) {
4633 if (a->nla_type == OVS_ACTION_ATTR_OUTPUT) {
4639 #endif /* defined(__linux__) */
4641 /* Returns true if 'odp_actions' contains more output actions than the datapath
4642 * can reliably handle in one go. On Linux, this is the value of the
4643 * net.core.netdev_max_backlog sysctl, which limits the maximum number of
4644 * packets that the kernel is willing to queue up for processing while the
4645 * datapath is processing a set of actions. */
4647 too_many_output_actions(const struct ofpbuf *odp_actions OVS_UNUSED)
4650 return (odp_actions->size / NL_A_U32_SIZE > netdev_max_backlog()
4651 && count_output_actions(odp_actions) > netdev_max_backlog());
4653 /* OSes other than Linux might have similar limits, but we don't know how
4654 * to determine them.*/
4660 xlate_wc_init(struct xlate_ctx *ctx)
4662 flow_wildcards_init_catchall(ctx->wc);
4664 /* Some fields we consider to always be examined. */
4665 memset(&ctx->wc->masks.in_port, 0xff, sizeof ctx->wc->masks.in_port);
4666 memset(&ctx->wc->masks.dl_type, 0xff, sizeof ctx->wc->masks.dl_type);
4667 if (is_ip_any(&ctx->xin->flow)) {
4668 ctx->wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
4671 if (ctx->xbridge->support.odp.recirc) {
4672 /* Always exactly match recirc_id when datapath supports
4674 ctx->wc->masks.recirc_id = UINT32_MAX;
4677 if (ctx->xbridge->netflow) {
4678 netflow_mask_wc(&ctx->xin->flow, ctx->wc);
4681 tnl_wc_init(&ctx->xin->flow, ctx->wc);
4685 xlate_wc_finish(struct xlate_ctx *ctx)
4687 /* Clear the metadata and register wildcard masks, because we won't
4688 * use non-header fields as part of the cache. */
4689 flow_wildcards_clear_non_packet_fields(ctx->wc);
4691 /* ICMPv4 and ICMPv6 have 8-bit "type" and "code" fields. struct flow
4692 * uses the low 8 bits of the 16-bit tp_src and tp_dst members to
4693 * represent these fields. The datapath interface, on the other hand,
4694 * represents them with just 8 bits each. This means that if the high
4695 * 8 bits of the masks for these fields somehow become set, then they
4696 * will get chopped off by a round trip through the datapath, and
4697 * revalidation will spot that as an inconsistency and delete the flow.
4698 * Avoid the problem here by making sure that only the low 8 bits of
4699 * either field can be unwildcarded for ICMP.
4701 if (is_icmpv4(&ctx->xin->flow) || is_icmpv6(&ctx->xin->flow)) {
4702 ctx->wc->masks.tp_src &= htons(UINT8_MAX);
4703 ctx->wc->masks.tp_dst &= htons(UINT8_MAX);
4705 /* VLAN_TCI CFI bit must be matched if any of the TCI is matched. */
4706 if (ctx->wc->masks.vlan_tci) {
4707 ctx->wc->masks.vlan_tci |= htons(VLAN_CFI);
4711 /* Translates the flow, actions, or rule in 'xin' into datapath actions in
4713 * The caller must take responsibility for eventually freeing 'xout', with
4714 * xlate_out_uninit(). */
4716 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
4718 *xout = (struct xlate_out) {
4722 .has_normal = false,
4723 .has_fin_timeout = false,
4724 .nf_output_iface = NF_OUT_DROP,
4729 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4730 struct xbridge *xbridge = xbridge_lookup(xcfg, xin->ofproto);
4735 struct flow *flow = &xin->flow;
4737 union mf_subvalue stack_stub[1024 / sizeof(union mf_subvalue)];
4738 uint64_t action_set_stub[1024 / 8];
4739 struct flow_wildcards scratch_wc;
4740 uint64_t actions_stub[256 / 8];
4741 struct ofpbuf scratch_actions = OFPBUF_STUB_INITIALIZER(actions_stub);
4742 struct xlate_ctx ctx = {
4746 .orig_tunnel_ip_dst = flow->tunnel.ip_dst,
4748 .stack = OFPBUF_STUB_INITIALIZER(stack_stub),
4750 .wc = xin->wc ? xin->wc : &scratch_wc,
4751 .odp_actions = xin->odp_actions ? xin->odp_actions : &scratch_actions,
4756 .in_action_set = false,
4759 .rule_cookie = OVS_BE64_MAX,
4760 .orig_skb_priority = flow->skb_priority,
4761 .sflow_n_outputs = 0,
4762 .sflow_odp_port = 0,
4765 .recirc_action_offset = -1,
4766 .last_unroll_offset = -1,
4770 .action_set_has_group = false,
4771 .action_set = OFPBUF_STUB_INITIALIZER(action_set_stub),
4774 /* 'base_flow' reflects the packet as it came in, but we need it to reflect
4775 * the packet as the datapath will treat it for output actions:
4777 * - Our datapath doesn't retain tunneling information without us
4778 * re-setting it, so clear the tunnel data.
4780 * - For VLAN splinters, a higher layer may pretend that the packet
4781 * came in on 'flow->in_port.ofp_port' with 'flow->vlan_tci'
4782 * attached, because that's how we want to treat it from an OpenFlow
4783 * perspective. But from the datapath's perspective it actually came
4784 * in on a VLAN device without any VLAN attached. So here we put the
4785 * datapath's view of the VLAN information in 'base_flow' to ensure
4786 * correct treatment.
4788 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
4789 if (flow->in_port.ofp_port
4790 != vsp_realdev_to_vlandev(xbridge->ofproto,
4791 flow->in_port.ofp_port,
4793 ctx.base_flow.vlan_tci = 0;
4796 ofpbuf_reserve(ctx.odp_actions, NL_A_U32_SIZE);
4798 xlate_wc_init(&ctx);
4801 struct xport *in_port;
4803 COVERAGE_INC(xlate_actions);
4805 /* The in_port of the original packet before recirculation. */
4806 in_port = get_ofp_port(xbridge, flow->in_port.ofp_port);
4809 const struct recirc_id_node *recirc = xin->recirc;
4811 if (xin->ofpacts_len > 0 || ctx.rule) {
4812 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4814 VLOG_WARN_RL(&rl, "Recirculation conflict (%s)!",
4815 xin->ofpacts_len > 0
4821 /* Set the bridge for post-recirculation processing if needed. */
4822 if (ctx.xbridge->ofproto != recirc->ofproto) {
4823 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
4824 const struct xbridge *new_bridge = xbridge_lookup(xcfg,
4827 if (OVS_UNLIKELY(!new_bridge)) {
4828 /* Drop the packet if the bridge cannot be found. */
4829 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4830 VLOG_WARN_RL(&rl, "Recirculation bridge no longer exists.");
4833 ctx.xbridge = new_bridge;
4836 /* Set the post-recirculation table id. Note: A table lookup is done
4837 * only if there are no post-recirculation actions. */
4838 ctx.table_id = recirc->table_id;
4840 /* Restore pipeline metadata. May change flow's in_port and other
4841 * metadata to the values that existed when recirculation was
4843 recirc_metadata_to_flow(&recirc->metadata, flow);
4845 /* Restore stack, if any. */
4846 if (recirc->stack) {
4847 ofpbuf_put(&ctx.stack, recirc->stack->data, recirc->stack->size);
4850 /* Restore action set, if any. */
4851 if (recirc->action_set_len) {
4852 const struct ofpact *a;
4854 ofpbuf_put(&ctx.action_set, recirc->ofpacts,
4855 recirc->action_set_len);
4857 OFPACT_FOR_EACH(a, recirc->ofpacts, recirc->action_set_len) {
4858 if (a->type == OFPACT_GROUP) {
4859 ctx.action_set_has_group = true;
4865 /* Restore recirculation actions. If there are no actions, processing
4866 * will start with a lookup in the table set above. */
4867 if (recirc->ofpacts_len > recirc->action_set_len) {
4868 xin->ofpacts_len = recirc->ofpacts_len - recirc->action_set_len;
4869 xin->ofpacts = recirc->ofpacts +
4870 recirc->action_set_len / sizeof *recirc->ofpacts;
4872 } else if (OVS_UNLIKELY(flow->recirc_id)) {
4873 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
4875 VLOG_WARN_RL(&rl, "Recirculation context not found for ID %"PRIx32,
4879 /* The bridge is now known so obtain its table version. */
4880 ctx.tables_version = ofproto_dpif_get_tables_version(ctx.xbridge->ofproto);
4882 if (!xin->ofpacts && !ctx.rule) {
4883 ctx.rule = rule_dpif_lookup_from_table(
4884 ctx.xbridge->ofproto, ctx.tables_version, flow, xin->wc,
4885 ctx.xin->xcache != NULL, ctx.xin->resubmit_stats, &ctx.table_id,
4886 flow->in_port.ofp_port, true, true);
4887 if (ctx.xin->resubmit_stats) {
4888 rule_dpif_credit_stats(ctx.rule, ctx.xin->resubmit_stats);
4890 if (ctx.xin->xcache) {
4891 struct xc_entry *entry;
4893 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_RULE);
4894 entry->u.rule = ctx.rule;
4897 if (OVS_UNLIKELY(ctx.xin->resubmit_hook)) {
4898 ctx.xin->resubmit_hook(ctx.xin, ctx.rule, 0);
4901 xout->fail_open = ctx.rule && rule_dpif_is_fail_open(ctx.rule);
4903 struct flow orig_flow;
4904 if (mbridge_has_mirrors(xbridge->mbridge)) {
4905 /* Do this conditionally because the copy is expensive enough that it
4906 * shows up in profiles. */
4910 /* Tunnel stats only for non-recirculated packets. */
4911 if (!xin->recirc && in_port && in_port->is_tunnel) {
4912 if (ctx.xin->resubmit_stats) {
4913 netdev_vport_inc_rx(in_port->netdev, ctx.xin->resubmit_stats);
4915 bfd_account_rx(in_port->bfd, ctx.xin->resubmit_stats);
4918 if (ctx.xin->xcache) {
4919 struct xc_entry *entry;
4921 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETDEV);
4922 entry->u.dev.rx = netdev_ref(in_port->netdev);
4923 entry->u.dev.bfd = bfd_ref(in_port->bfd);
4927 /* Do not perform special processing on recirculated packets,
4928 * as recirculated packets are not really received by the bridge. */
4929 if (xin->recirc || !process_special(&ctx, in_port)) {
4930 /* Sampling is done only for packets really received by the bridge. */
4931 unsigned int user_cookie_offset = 0;
4933 user_cookie_offset = compose_sflow_action(&ctx);
4934 compose_ipfix_action(&ctx, ODPP_NONE);
4936 size_t sample_actions_len = ctx.odp_actions->size;
4938 if (tnl_process_ecn(flow)
4939 && (!in_port || may_receive(in_port, &ctx))) {
4940 const struct ofpact *ofpacts;
4944 ofpacts = xin->ofpacts;
4945 ofpacts_len = xin->ofpacts_len;
4946 } else if (ctx.rule) {
4947 const struct rule_actions *actions
4948 = rule_dpif_get_actions(ctx.rule);
4949 ofpacts = actions->ofpacts;
4950 ofpacts_len = actions->ofpacts_len;
4951 ctx.rule_cookie = rule_dpif_get_flow_cookie(ctx.rule);
4956 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
4958 /* We've let OFPP_NORMAL and the learning action look at the
4959 * packet, so drop it now if forwarding is disabled. */
4960 if (in_port && (!xport_stp_forward_state(in_port) ||
4961 !xport_rstp_forward_state(in_port))) {
4962 /* Drop all actions added by do_xlate_actions() above. */
4963 ctx.odp_actions->size = sample_actions_len;
4965 /* Undo changes that may have been done for recirculation. */
4966 if (exit_recirculates(&ctx)) {
4967 ctx.action_set.size = ctx.recirc_action_offset;
4968 ctx.recirc_action_offset = -1;
4969 ctx.last_unroll_offset = -1;
4971 } else if (ctx.action_set.size) {
4972 /* Translate action set only if not dropping the packet and
4973 * not recirculating. */
4974 if (!exit_recirculates(&ctx)) {
4975 xlate_action_set(&ctx);
4978 /* Check if need to recirculate. */
4979 if (exit_recirculates(&ctx)) {
4980 compose_recirculate_action(&ctx);
4984 /* Output only fully processed packets. */
4985 if (!exit_recirculates(&ctx)
4986 && xbridge->has_in_band
4987 && in_band_must_output_to_local_port(flow)
4988 && !actions_output_to_local_port(&ctx)) {
4989 compose_output_action(&ctx, OFPP_LOCAL, NULL);
4992 if (user_cookie_offset) {
4993 fix_sflow_action(&ctx, user_cookie_offset);
4995 /* Only mirror fully processed packets. */
4996 if (!exit_recirculates(&ctx)
4997 && mbridge_has_mirrors(xbridge->mbridge)) {
4998 add_mirror_actions(&ctx, &orig_flow);
5002 if (nl_attr_oversized(ctx.odp_actions->size)) {
5003 /* These datapath actions are too big for a Netlink attribute, so we
5004 * can't hand them to the kernel directly. dpif_execute() can execute
5005 * them one by one with help, so just mark the result as SLOW_ACTION to
5006 * prevent the flow from being installed. */
5007 COVERAGE_INC(xlate_actions_oversize);
5008 ctx.xout->slow |= SLOW_ACTION;
5009 } else if (too_many_output_actions(ctx.odp_actions)) {
5010 COVERAGE_INC(xlate_actions_too_many_output);
5011 ctx.xout->slow |= SLOW_ACTION;
5014 /* Update mirror stats only for packets really received by the bridge. */
5015 if (!xin->recirc && mbridge_has_mirrors(xbridge->mbridge)) {
5016 if (ctx.xin->resubmit_stats) {
5017 mirror_update_stats(xbridge->mbridge, xout->mirrors,
5018 ctx.xin->resubmit_stats->n_packets,
5019 ctx.xin->resubmit_stats->n_bytes);
5021 if (ctx.xin->xcache) {
5022 struct xc_entry *entry;
5024 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_MIRROR);
5025 entry->u.mirror.mbridge = mbridge_ref(xbridge->mbridge);
5026 entry->u.mirror.mirrors = xout->mirrors;
5030 /* Do netflow only for packets really received by the bridge and not sent
5031 * to the controller. We consider packets sent to the controller to be
5032 * part of the control plane rather than the data plane. */
5033 if (!xin->recirc && xbridge->netflow && !(xout->slow & SLOW_CONTROLLER)) {
5034 if (ctx.xin->resubmit_stats) {
5035 netflow_flow_update(xbridge->netflow, flow,
5036 xout->nf_output_iface,
5037 ctx.xin->resubmit_stats);
5039 if (ctx.xin->xcache) {
5040 struct xc_entry *entry;
5042 entry = xlate_cache_add_entry(ctx.xin->xcache, XC_NETFLOW);
5043 entry->u.nf.netflow = netflow_ref(xbridge->netflow);
5044 entry->u.nf.flow = xmemdup(flow, sizeof *flow);
5045 entry->u.nf.iface = xout->nf_output_iface;
5050 xlate_wc_finish(&ctx);
5054 ofpbuf_uninit(&ctx.stack);
5055 ofpbuf_uninit(&ctx.action_set);
5056 ofpbuf_uninit(&scratch_actions);
5059 /* Sends 'packet' out 'ofport'.
5060 * May modify 'packet'.
5061 * Returns 0 if successful, otherwise a positive errno value. */
5063 xlate_send_packet(const struct ofport_dpif *ofport, struct dp_packet *packet)
5065 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5066 struct xport *xport;
5067 struct ofpact_output output;
5070 ofpact_init(&output.ofpact, OFPACT_OUTPUT, sizeof output);
5071 /* Use OFPP_NONE as the in_port to avoid special packet processing. */
5072 flow_extract(packet, &flow);
5073 flow.in_port.ofp_port = OFPP_NONE;
5075 xport = xport_lookup(xcfg, ofport);
5079 output.port = xport->ofp_port;
5082 return ofproto_dpif_execute_actions(xport->xbridge->ofproto, &flow, NULL,
5083 &output.ofpact, sizeof output,
5087 struct xlate_cache *
5088 xlate_cache_new(void)
5090 struct xlate_cache *xcache = xmalloc(sizeof *xcache);
5092 ofpbuf_init(&xcache->entries, 512);
5096 static struct xc_entry *
5097 xlate_cache_add_entry(struct xlate_cache *xcache, enum xc_type type)
5099 struct xc_entry *entry;
5101 entry = ofpbuf_put_zeros(&xcache->entries, sizeof *entry);
5108 xlate_cache_netdev(struct xc_entry *entry, const struct dpif_flow_stats *stats)
5110 if (entry->u.dev.tx) {
5111 netdev_vport_inc_tx(entry->u.dev.tx, stats);
5113 if (entry->u.dev.rx) {
5114 netdev_vport_inc_rx(entry->u.dev.rx, stats);
5116 if (entry->u.dev.bfd) {
5117 bfd_account_rx(entry->u.dev.bfd, stats);
5122 xlate_cache_normal(struct ofproto_dpif *ofproto, struct flow *flow, int vlan)
5124 struct xlate_cfg *xcfg = ovsrcu_get(struct xlate_cfg *, &xcfgp);
5125 struct xbridge *xbridge;
5126 struct xbundle *xbundle;
5127 struct flow_wildcards wc;
5129 xbridge = xbridge_lookup(xcfg, ofproto);
5134 xbundle = lookup_input_bundle(xbridge, flow->in_port.ofp_port, false,
5140 update_learning_table(xbridge, flow, &wc, vlan, xbundle);
5143 /* Push stats and perform side effects of flow translation. */
5145 xlate_push_stats(struct xlate_cache *xcache,
5146 const struct dpif_flow_stats *stats)
5148 struct xc_entry *entry;
5149 struct ofpbuf entries = xcache->entries;
5150 uint8_t dmac[ETH_ADDR_LEN];
5152 if (!stats->n_packets) {
5156 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5157 switch (entry->type) {
5159 rule_dpif_credit_stats(entry->u.rule, stats);
5162 bond_account(entry->u.bond.bond, entry->u.bond.flow,
5163 entry->u.bond.vid, stats->n_bytes);
5166 xlate_cache_netdev(entry, stats);
5169 netflow_flow_update(entry->u.nf.netflow, entry->u.nf.flow,
5170 entry->u.nf.iface, stats);
5173 mirror_update_stats(entry->u.mirror.mbridge,
5174 entry->u.mirror.mirrors,
5175 stats->n_packets, stats->n_bytes);
5178 ofproto_dpif_flow_mod(entry->u.learn.ofproto, entry->u.learn.fm);
5181 xlate_cache_normal(entry->u.normal.ofproto, entry->u.normal.flow,
5182 entry->u.normal.vlan);
5184 case XC_FIN_TIMEOUT:
5185 xlate_fin_timeout__(entry->u.fin.rule, stats->tcp_flags,
5186 entry->u.fin.idle, entry->u.fin.hard);
5189 group_dpif_credit_stats(entry->u.group.group, entry->u.group.bucket,
5193 /* Lookup arp to avoid arp timeout. */
5194 tnl_arp_lookup(entry->u.tnl_arp_cache.br_name, entry->u.tnl_arp_cache.d_ip, dmac);
5203 xlate_dev_unref(struct xc_entry *entry)
5205 if (entry->u.dev.tx) {
5206 netdev_close(entry->u.dev.tx);
5208 if (entry->u.dev.rx) {
5209 netdev_close(entry->u.dev.rx);
5211 if (entry->u.dev.bfd) {
5212 bfd_unref(entry->u.dev.bfd);
5217 xlate_cache_clear_netflow(struct netflow *netflow, struct flow *flow)
5219 netflow_flow_clear(netflow, flow);
5220 netflow_unref(netflow);
5225 xlate_cache_clear(struct xlate_cache *xcache)
5227 struct xc_entry *entry;
5228 struct ofpbuf entries;
5234 XC_ENTRY_FOR_EACH (entry, entries, xcache) {
5235 switch (entry->type) {
5237 rule_dpif_unref(entry->u.rule);
5240 free(entry->u.bond.flow);
5241 bond_unref(entry->u.bond.bond);
5244 xlate_dev_unref(entry);
5247 xlate_cache_clear_netflow(entry->u.nf.netflow, entry->u.nf.flow);
5250 mbridge_unref(entry->u.mirror.mbridge);
5253 free(entry->u.learn.fm);
5254 ofpbuf_delete(entry->u.learn.ofpacts);
5257 free(entry->u.normal.flow);
5259 case XC_FIN_TIMEOUT:
5260 /* 'u.fin.rule' is always already held as a XC_RULE, which
5261 * has already released it's reference above. */
5264 group_dpif_unref(entry->u.group.group);
5273 ofpbuf_clear(&xcache->entries);
5277 xlate_cache_delete(struct xlate_cache *xcache)
5279 xlate_cache_clear(xcache);
5280 ofpbuf_uninit(&xcache->entries);