1 /* Copyright (c) 2009, 2010, 2011, 2012, 2013 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"
23 #include "byte-order.h"
28 #include "dynamic-string.h"
33 #include "mac-learning.h"
34 #include "meta-flow.h"
35 #include "multipath.h"
36 #include "netdev-vport.h"
39 #include "odp-execute.h"
40 #include "ofp-actions.h"
41 #include "ofproto/ofproto-dpif-ipfix.h"
42 #include "ofproto/ofproto-dpif-mirror.h"
43 #include "ofproto/ofproto-dpif-sflow.h"
44 #include "ofproto/ofproto-dpif.h"
48 COVERAGE_DEFINE(xlate_actions);
50 VLOG_DEFINE_THIS_MODULE(ofproto_dpif_xlate);
52 /* Maximum depth of flow table recursion (due to resubmit actions) in a
53 * flow translation. */
54 #define MAX_RESUBMIT_RECURSION 64
57 struct hmap_node hmap_node; /* Node in global 'xbridges' map. */
58 struct ofproto_dpif *ofproto; /* Key in global 'xbridges' map. */
60 struct list xbundles; /* Owned xbundles. */
61 struct hmap xports; /* Indexed by ofp_port. */
63 char *name; /* Name used in log messages. */
64 struct mac_learning *ml; /* Mac learning handle. */
65 struct mbridge *mbridge; /* Mirroring. */
66 struct dpif_sflow *sflow; /* SFlow handle, or null. */
67 struct dpif_ipfix *ipfix; /* Ipfix handle, or null. */
68 struct stp *stp; /* STP or null if disabled. */
70 enum ofp_config_flags frag; /* Fragmentation handling. */
71 bool has_netflow; /* Bridge runs netflow? */
72 bool has_in_band; /* Bridge has in band control? */
73 bool forward_bpdu; /* Bridge forwards STP BPDUs? */
77 struct hmap_node hmap_node; /* In global 'xbundles' map. */
78 struct ofbundle *ofbundle; /* Key in global 'xbundles' map. */
80 struct list list_node; /* In parent 'xbridges' list. */
81 struct xbridge *xbridge; /* Parent xbridge. */
83 struct list xports; /* Contains "struct xport"s. */
85 char *name; /* Name used in log messages. */
86 struct bond *bond; /* Nonnull iff more than one port. */
87 struct lacp *lacp; /* LACP handle or null. */
89 enum port_vlan_mode vlan_mode; /* VLAN mode. */
90 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
91 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
92 * NULL if all VLANs are trunked. */
93 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
94 bool floodable; /* No port has OFPUTIL_PC_NO_FLOOD set? */
98 struct hmap_node hmap_node; /* Node in global 'xports' map. */
99 struct ofport_dpif *ofport; /* Key in global 'xports map. */
101 struct hmap_node ofp_node; /* Node in parent xbridge 'xports' map. */
102 ofp_port_t ofp_port; /* Key in parent xbridge 'xports' map. */
104 odp_port_t odp_port; /* Datapath port number or ODPP_NONE. */
106 struct list bundle_node; /* In parent xbundle (if it exists). */
107 struct xbundle *xbundle; /* Parent xbundle or null. */
109 struct netdev *netdev; /* 'ofport''s netdev. */
111 struct xbridge *xbridge; /* Parent bridge. */
112 struct xport *peer; /* Patch port peer or null. */
114 enum ofputil_port_config config; /* OpenFlow port configuration. */
115 int stp_port_no; /* STP port number or 0 if not in use. */
117 struct hmap skb_priorities; /* Map of 'skb_priority_to_dscp's. */
119 bool may_enable; /* May be enabled in bonds. */
120 bool is_tunnel; /* Is a tunnel port. */
122 struct cfm *cfm; /* CFM handle or null. */
123 struct bfd *bfd; /* BFD handle or null. */
127 struct xlate_in *xin;
128 struct xlate_out *xout;
130 const struct xbridge *xbridge;
132 /* Flow at the last commit. */
133 struct flow base_flow;
135 /* Tunnel IP destination address as received. This is stored separately
136 * as the base_flow.tunnel is cleared on init to reflect the datapath
137 * behavior. Used to make sure not to send tunneled output to ourselves,
138 * which might lead to an infinite loop. This could happen easily
139 * if a tunnel is marked as 'ip_remote=flow', and the flow does not
140 * actually set the tun_dst field. */
141 ovs_be32 orig_tunnel_ip_dst;
143 /* Stack for the push and pop actions. Each stack element is of type
144 * "union mf_subvalue". */
145 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
148 /* The rule that we are currently translating, or NULL. */
149 struct rule_dpif *rule;
151 int recurse; /* Recursion level, via xlate_table_action. */
152 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
153 uint32_t orig_skb_priority; /* Priority when packet arrived. */
154 uint8_t table_id; /* OpenFlow table ID where flow was found. */
155 uint32_t sflow_n_outputs; /* Number of output ports. */
156 odp_port_t sflow_odp_port; /* Output port for composing sFlow action. */
157 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
158 bool exit; /* No further actions should be processed. */
161 /* A controller may use OFPP_NONE as the ingress port to indicate that
162 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
163 * when an input bundle is needed for validation (e.g., mirroring or
164 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
165 * any 'port' structs, so care must be taken when dealing with it.
166 * The bundle's name and vlan mode are initialized in lookup_input_bundle() */
167 static struct xbundle ofpp_none_bundle;
169 /* Node in 'xport''s 'skb_priorities' map. Used to maintain a map from
170 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
171 * traffic egressing the 'ofport' with that priority should be marked with. */
172 struct skb_priority_to_dscp {
173 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'skb_priorities'. */
174 uint32_t skb_priority; /* Priority of this queue (see struct flow). */
176 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
179 static struct hmap xbridges = HMAP_INITIALIZER(&xbridges);
180 static struct hmap xbundles = HMAP_INITIALIZER(&xbundles);
181 static struct hmap xports = HMAP_INITIALIZER(&xports);
183 static bool may_receive(const struct xport *, struct xlate_ctx *);
184 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
186 static void xlate_normal(struct xlate_ctx *);
187 static void xlate_report(struct xlate_ctx *, const char *);
188 static void xlate_table_action(struct xlate_ctx *, ofp_port_t in_port,
189 uint8_t table_id, bool may_packet_in);
190 static bool input_vid_is_valid(uint16_t vid, struct xbundle *, bool warn);
191 static uint16_t input_vid_to_vlan(const struct xbundle *, uint16_t vid);
192 static void output_normal(struct xlate_ctx *, const struct xbundle *,
194 static void compose_output_action(struct xlate_ctx *, ofp_port_t ofp_port);
196 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
198 static struct xbridge *xbridge_lookup(const struct ofproto_dpif *);
199 static struct xbundle *xbundle_lookup(const struct ofbundle *);
200 static struct xport *xport_lookup(struct ofport_dpif *);
201 static struct xport *get_ofp_port(const struct xbridge *, ofp_port_t ofp_port);
202 static struct skb_priority_to_dscp *get_skb_priority(const struct xport *,
203 uint32_t skb_priority);
204 static void clear_skb_priorities(struct xport *);
205 static bool dscp_from_skb_priority(const struct xport *, uint32_t skb_priority,
209 xlate_ofproto_set(struct ofproto_dpif *ofproto, const char *name,
210 const struct mac_learning *ml, struct stp *stp,
211 const struct mbridge *mbridge,
212 const struct dpif_sflow *sflow,
213 const struct dpif_ipfix *ipfix, enum ofp_config_flags frag,
214 bool forward_bpdu, bool has_in_band, bool has_netflow)
216 struct xbridge *xbridge = xbridge_lookup(ofproto);
219 xbridge = xzalloc(sizeof *xbridge);
220 xbridge->ofproto = ofproto;
222 hmap_insert(&xbridges, &xbridge->hmap_node, hash_pointer(ofproto, 0));
223 hmap_init(&xbridge->xports);
224 list_init(&xbridge->xbundles);
227 if (xbridge->ml != ml) {
228 mac_learning_unref(xbridge->ml);
229 xbridge->ml = mac_learning_ref(ml);
232 if (xbridge->mbridge != mbridge) {
233 mbridge_unref(xbridge->mbridge);
234 xbridge->mbridge = mbridge_ref(mbridge);
237 if (xbridge->sflow != sflow) {
238 dpif_sflow_unref(xbridge->sflow);
239 xbridge->sflow = dpif_sflow_ref(sflow);
242 if (xbridge->ipfix != ipfix) {
243 dpif_ipfix_unref(xbridge->ipfix);
244 xbridge->ipfix = dpif_ipfix_ref(ipfix);
247 if (xbridge->stp != stp) {
248 stp_unref(xbridge->stp);
249 xbridge->stp = stp_ref(stp);
253 xbridge->name = xstrdup(name);
255 xbridge->forward_bpdu = forward_bpdu;
256 xbridge->has_in_band = has_in_band;
257 xbridge->has_netflow = has_netflow;
258 xbridge->frag = frag;
262 xlate_remove_ofproto(struct ofproto_dpif *ofproto)
264 struct xbridge *xbridge = xbridge_lookup(ofproto);
265 struct xbundle *xbundle, *next_xbundle;
266 struct xport *xport, *next_xport;
272 HMAP_FOR_EACH_SAFE (xport, next_xport, ofp_node, &xbridge->xports) {
273 xlate_ofport_remove(xport->ofport);
276 LIST_FOR_EACH_SAFE (xbundle, next_xbundle, list_node, &xbridge->xbundles) {
277 xlate_bundle_remove(xbundle->ofbundle);
280 hmap_remove(&xbridges, &xbridge->hmap_node);
286 xlate_bundle_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
287 const char *name, enum port_vlan_mode vlan_mode, int vlan,
288 unsigned long *trunks, bool use_priority_tags,
289 const struct bond *bond, const struct lacp *lacp,
292 struct xbundle *xbundle = xbundle_lookup(ofbundle);
295 xbundle = xzalloc(sizeof *xbundle);
296 xbundle->ofbundle = ofbundle;
297 xbundle->xbridge = xbridge_lookup(ofproto);
299 hmap_insert(&xbundles, &xbundle->hmap_node, hash_pointer(ofbundle, 0));
300 list_insert(&xbundle->xbridge->xbundles, &xbundle->list_node);
301 list_init(&xbundle->xports);
304 ovs_assert(xbundle->xbridge);
307 xbundle->name = xstrdup(name);
309 xbundle->vlan_mode = vlan_mode;
310 xbundle->vlan = vlan;
311 xbundle->trunks = trunks;
312 xbundle->use_priority_tags = use_priority_tags;
313 xbundle->floodable = floodable;
315 if (xbundle->bond != bond) {
316 bond_unref(xbundle->bond);
317 xbundle->bond = bond_ref(bond);
320 if (xbundle->lacp != lacp) {
321 lacp_unref(xbundle->lacp);
322 xbundle->lacp = lacp_ref(lacp);
327 xlate_bundle_remove(struct ofbundle *ofbundle)
329 struct xbundle *xbundle = xbundle_lookup(ofbundle);
330 struct xport *xport, *next;
336 LIST_FOR_EACH_SAFE (xport, next, bundle_node, &xbundle->xports) {
337 list_remove(&xport->bundle_node);
338 xport->xbundle = NULL;
341 hmap_remove(&xbundles, &xbundle->hmap_node);
342 list_remove(&xbundle->list_node);
343 bond_unref(xbundle->bond);
344 lacp_unref(xbundle->lacp);
350 xlate_ofport_set(struct ofproto_dpif *ofproto, struct ofbundle *ofbundle,
351 struct ofport_dpif *ofport, ofp_port_t ofp_port,
352 odp_port_t odp_port, const struct netdev *netdev,
353 const struct cfm *cfm, const struct bfd *bfd,
354 struct ofport_dpif *peer, int stp_port_no,
355 const struct ofproto_port_queue *qdscp_list, size_t n_qdscp,
356 enum ofputil_port_config config, bool is_tunnel,
359 struct xport *xport = xport_lookup(ofport);
363 xport = xzalloc(sizeof *xport);
364 xport->ofport = ofport;
365 xport->xbridge = xbridge_lookup(ofproto);
366 xport->ofp_port = ofp_port;
368 hmap_init(&xport->skb_priorities);
369 hmap_insert(&xports, &xport->hmap_node, hash_pointer(ofport, 0));
370 hmap_insert(&xport->xbridge->xports, &xport->ofp_node,
371 hash_ofp_port(xport->ofp_port));
374 ovs_assert(xport->ofp_port == ofp_port);
376 xport->config = config;
377 xport->stp_port_no = stp_port_no;
378 xport->is_tunnel = is_tunnel;
379 xport->may_enable = may_enable;
380 xport->odp_port = odp_port;
382 if (xport->netdev != netdev) {
383 netdev_close(xport->netdev);
384 xport->netdev = netdev_ref(netdev);
387 if (xport->cfm != cfm) {
388 cfm_unref(xport->cfm);
389 xport->cfm = cfm_ref(cfm);
392 if (xport->bfd != bfd) {
393 bfd_unref(xport->bfd);
394 xport->bfd = bfd_ref(bfd);
398 xport->peer->peer = NULL;
400 xport->peer = peer ? xport_lookup(peer) : NULL;
402 xport->peer->peer = xport;
405 if (xport->xbundle) {
406 list_remove(&xport->bundle_node);
408 xport->xbundle = ofbundle ? xbundle_lookup(ofbundle) : NULL;
409 if (xport->xbundle) {
410 list_insert(&xport->xbundle->xports, &xport->bundle_node);
413 clear_skb_priorities(xport);
414 for (i = 0; i < n_qdscp; i++) {
415 struct skb_priority_to_dscp *pdscp;
416 uint32_t skb_priority;
418 if (ofproto_dpif_queue_to_priority(xport->xbridge->ofproto,
424 pdscp = xmalloc(sizeof *pdscp);
425 pdscp->skb_priority = skb_priority;
426 pdscp->dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
427 hmap_insert(&xport->skb_priorities, &pdscp->hmap_node,
428 hash_int(pdscp->skb_priority, 0));
433 xlate_ofport_remove(struct ofport_dpif *ofport)
435 struct xport *xport = xport_lookup(ofport);
442 xport->peer->peer = NULL;
446 if (xport->xbundle) {
447 list_remove(&xport->bundle_node);
450 clear_skb_priorities(xport);
451 hmap_destroy(&xport->skb_priorities);
453 hmap_remove(&xports, &xport->hmap_node);
454 hmap_remove(&xport->xbridge->xports, &xport->ofp_node);
456 netdev_close(xport->netdev);
457 cfm_unref(xport->cfm);
458 bfd_unref(xport->bfd);
462 static struct xbridge *
463 xbridge_lookup(const struct ofproto_dpif *ofproto)
465 struct xbridge *xbridge;
467 HMAP_FOR_EACH_IN_BUCKET (xbridge, hmap_node, hash_pointer(ofproto, 0),
469 if (xbridge->ofproto == ofproto) {
476 static struct xbundle *
477 xbundle_lookup(const struct ofbundle *ofbundle)
479 struct xbundle *xbundle;
481 HMAP_FOR_EACH_IN_BUCKET (xbundle, hmap_node, hash_pointer(ofbundle, 0),
483 if (xbundle->ofbundle == ofbundle) {
490 static struct xport *
491 xport_lookup(struct ofport_dpif *ofport)
495 HMAP_FOR_EACH_IN_BUCKET (xport, hmap_node, hash_pointer(ofport, 0),
497 if (xport->ofport == ofport) {
505 static enum stp_state
506 xport_stp_learn_state(const struct xport *xport)
508 enum stp_state stp_state = xport->xbridge->stp && xport->stp_port_no
509 ? stp_port_get_state(stp_get_port(xport->xbridge->stp,
512 return stp_learn_in_state(stp_state);
516 xport_stp_forward_state(const struct xport *xport)
518 enum stp_state stp_state = xport->xbridge->stp && xport->stp_port_no
519 ? stp_port_get_state(stp_get_port(xport->xbridge->stp,
522 return stp_forward_in_state(stp_state);
525 /* Returns true if STP should process 'flow'. Sets fields in 'wc' that
526 * were used to make the determination.*/
528 stp_should_process_flow(const struct flow *flow, struct flow_wildcards *wc)
530 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
531 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
535 stp_process_packet(const struct xport *xport, const struct ofpbuf *packet)
537 struct ofpbuf payload = *packet;
538 struct eth_header *eth = payload.data;
539 struct stp_port *sp = xport->xbridge->stp && xport->stp_port_no
540 ? stp_get_port(xport->xbridge->stp, xport->stp_port_no)
543 /* Sink packets on ports that have STP disabled when the bridge has
545 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
549 /* Trim off padding on payload. */
550 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
551 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
554 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
555 stp_received_bpdu(sp, payload.data, payload.size);
559 static struct xport *
560 get_ofp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
564 HMAP_FOR_EACH_IN_BUCKET (xport, ofp_node, hash_ofp_port(ofp_port),
566 if (xport->ofp_port == ofp_port) {
574 ofp_port_to_odp_port(const struct xbridge *xbridge, ofp_port_t ofp_port)
576 const struct xport *xport = get_ofp_port(xbridge, ofp_port);
577 return xport ? xport->odp_port : ODPP_NONE;
581 xbundle_trunks_vlan(const struct xbundle *bundle, uint16_t vlan)
583 return (bundle->vlan_mode != PORT_VLAN_ACCESS
584 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
588 xbundle_includes_vlan(const struct xbundle *xbundle, uint16_t vlan)
590 return vlan == xbundle->vlan || xbundle_trunks_vlan(xbundle, vlan);
594 xbundle_mirror_out(const struct xbridge *xbridge, struct xbundle *xbundle)
596 return xbundle != &ofpp_none_bundle
597 ? mirror_bundle_out(xbridge->mbridge, xbundle->ofbundle)
602 xbundle_mirror_src(const struct xbridge *xbridge, struct xbundle *xbundle)
604 return xbundle != &ofpp_none_bundle
605 ? mirror_bundle_src(xbridge->mbridge, xbundle->ofbundle)
610 xbundle_mirror_dst(const struct xbridge *xbridge, struct xbundle *xbundle)
612 return xbundle != &ofpp_none_bundle
613 ? mirror_bundle_dst(xbridge->mbridge, xbundle->ofbundle)
617 static struct xbundle *
618 lookup_input_bundle(const struct xbridge *xbridge, ofp_port_t in_port,
619 bool warn, struct xport **in_xportp)
623 /* Find the port and bundle for the received packet. */
624 xport = get_ofp_port(xbridge, in_port);
628 if (xport && xport->xbundle) {
629 return xport->xbundle;
632 /* Special-case OFPP_NONE, which a controller may use as the ingress
633 * port for traffic that it is sourcing. */
634 if (in_port == OFPP_NONE) {
635 ofpp_none_bundle.name = "OFPP_NONE";
636 ofpp_none_bundle.vlan_mode = PORT_VLAN_TRUNK;
637 return &ofpp_none_bundle;
640 /* Odd. A few possible reasons here:
642 * - We deleted a port but there are still a few packets queued up
645 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
646 * we don't know about.
648 * - The ofproto client didn't configure the port as part of a bundle.
649 * This is particularly likely to happen if a packet was received on the
650 * port after it was created, but before the client had a chance to
651 * configure its bundle.
654 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
656 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
657 "port %"PRIu16, xbridge->name, in_port);
663 add_mirror_actions(struct xlate_ctx *ctx, const struct flow *orig_flow)
665 const struct xbridge *xbridge = ctx->xbridge;
666 mirror_mask_t mirrors;
667 struct xbundle *in_xbundle;
671 mirrors = ctx->xout->mirrors;
672 ctx->xout->mirrors = 0;
674 in_xbundle = lookup_input_bundle(xbridge, orig_flow->in_port.ofp_port,
675 ctx->xin->packet != NULL, NULL);
679 mirrors |= xbundle_mirror_src(xbridge, in_xbundle);
681 /* Drop frames on bundles reserved for mirroring. */
682 if (xbundle_mirror_out(xbridge, in_xbundle)) {
683 if (ctx->xin->packet != NULL) {
684 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
685 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
686 "%s, which is reserved exclusively for mirroring",
687 ctx->xbridge->name, in_xbundle->name);
689 ofpbuf_clear(&ctx->xout->odp_actions);
694 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
695 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
698 vlan = input_vid_to_vlan(in_xbundle, vid);
704 /* Restore the original packet before adding the mirror actions. */
705 ctx->xin->flow = *orig_flow;
708 mirror_mask_t dup_mirrors;
709 struct ofbundle *out;
710 unsigned long *vlans;
715 has_mirror = mirror_get(xbridge->mbridge, mirror_mask_ffs(mirrors) - 1,
716 &vlans, &dup_mirrors, &out, &out_vlan);
717 ovs_assert(has_mirror);
720 ctx->xout->wc.masks.vlan_tci |= htons(VLAN_CFI | VLAN_VID_MASK);
722 vlan_mirrored = !vlans || bitmap_is_set(vlans, vlan);
725 if (!vlan_mirrored) {
726 mirrors = zero_rightmost_1bit(mirrors);
730 mirrors &= ~dup_mirrors;
731 ctx->xout->mirrors |= dup_mirrors;
733 struct xbundle *out_xbundle = xbundle_lookup(out);
735 output_normal(ctx, out_xbundle, vlan);
737 } else if (vlan != out_vlan
738 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
739 struct xbundle *xbundle;
741 LIST_FOR_EACH (xbundle, list_node, &xbridge->xbundles) {
742 if (xbundle_includes_vlan(xbundle, out_vlan)
743 && !xbundle_mirror_out(xbridge, xbundle)) {
744 output_normal(ctx, xbundle, out_vlan);
751 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
752 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_xbundle',
753 * the bundle on which the packet was received, returns the VLAN to which the
756 * Both 'vid' and the return value are in the range 0...4095. */
758 input_vid_to_vlan(const struct xbundle *in_xbundle, uint16_t vid)
760 switch (in_xbundle->vlan_mode) {
761 case PORT_VLAN_ACCESS:
762 return in_xbundle->vlan;
765 case PORT_VLAN_TRUNK:
768 case PORT_VLAN_NATIVE_UNTAGGED:
769 case PORT_VLAN_NATIVE_TAGGED:
770 return vid ? vid : in_xbundle->vlan;
777 /* Checks whether a packet with the given 'vid' may ingress on 'in_xbundle'.
778 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
781 * 'vid' should be the VID obtained from the 802.1Q header that was received as
782 * part of a packet (specify 0 if there was no 802.1Q header), in the range
785 input_vid_is_valid(uint16_t vid, struct xbundle *in_xbundle, bool warn)
787 /* Allow any VID on the OFPP_NONE port. */
788 if (in_xbundle == &ofpp_none_bundle) {
792 switch (in_xbundle->vlan_mode) {
793 case PORT_VLAN_ACCESS:
796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
797 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" tagged "
798 "packet received on port %s configured as VLAN "
799 "%"PRIu16" access port", vid, in_xbundle->name,
806 case PORT_VLAN_NATIVE_UNTAGGED:
807 case PORT_VLAN_NATIVE_TAGGED:
809 /* Port must always carry its native VLAN. */
813 case PORT_VLAN_TRUNK:
814 if (!xbundle_includes_vlan(in_xbundle, vid)) {
816 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
817 VLOG_WARN_RL(&rl, "dropping VLAN %"PRIu16" packet "
818 "received on port %s not configured for trunking "
819 "VLAN %"PRIu16, vid, in_xbundle->name, vid);
831 /* Given 'vlan', the VLAN that a packet belongs to, and
832 * 'out_xbundle', a bundle on which the packet is to be output, returns the VID
833 * that should be included in the 802.1Q header. (If the return value is 0,
834 * then the 802.1Q header should only be included in the packet if there is a
837 * Both 'vlan' and the return value are in the range 0...4095. */
839 output_vlan_to_vid(const struct xbundle *out_xbundle, uint16_t vlan)
841 switch (out_xbundle->vlan_mode) {
842 case PORT_VLAN_ACCESS:
845 case PORT_VLAN_TRUNK:
846 case PORT_VLAN_NATIVE_TAGGED:
849 case PORT_VLAN_NATIVE_UNTAGGED:
850 return vlan == out_xbundle->vlan ? 0 : vlan;
858 output_normal(struct xlate_ctx *ctx, const struct xbundle *out_xbundle,
861 ovs_be16 *flow_tci = &ctx->xin->flow.vlan_tci;
863 ovs_be16 tci, old_tci;
866 vid = output_vlan_to_vid(out_xbundle, vlan);
867 if (list_is_empty(&out_xbundle->xports)) {
868 /* Partially configured bundle with no slaves. Drop the packet. */
870 } else if (!out_xbundle->bond) {
871 xport = CONTAINER_OF(list_front(&out_xbundle->xports), struct xport,
874 struct ofport_dpif *ofport;
876 ofport = bond_choose_output_slave(out_xbundle->bond, &ctx->xin->flow,
877 &ctx->xout->wc, vid);
878 xport = ofport ? xport_lookup(ofport) : NULL;
881 /* No slaves enabled, so drop packet. */
888 if (tci || out_xbundle->use_priority_tags) {
889 tci |= *flow_tci & htons(VLAN_PCP_MASK);
891 tci |= htons(VLAN_CFI);
896 compose_output_action(ctx, xport->ofp_port);
900 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
901 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
902 * indicate this; newer upstream kernels use gratuitous ARP requests. */
904 is_gratuitous_arp(const struct flow *flow, struct flow_wildcards *wc)
906 if (flow->dl_type != htons(ETH_TYPE_ARP)) {
910 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
911 if (!eth_addr_is_broadcast(flow->dl_dst)) {
915 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
916 if (flow->nw_proto == ARP_OP_REPLY) {
918 } else if (flow->nw_proto == ARP_OP_REQUEST) {
919 memset(&wc->masks.nw_src, 0xff, sizeof wc->masks.nw_src);
920 memset(&wc->masks.nw_dst, 0xff, sizeof wc->masks.nw_dst);
922 return flow->nw_src == flow->nw_dst;
929 update_learning_table(const struct xbridge *xbridge,
930 const struct flow *flow, struct flow_wildcards *wc,
931 int vlan, struct xbundle *in_xbundle)
933 struct mac_entry *mac;
935 /* Don't learn the OFPP_NONE port. */
936 if (in_xbundle == &ofpp_none_bundle) {
940 ovs_rwlock_wrlock(&xbridge->ml->rwlock);
941 if (!mac_learning_may_learn(xbridge->ml, flow->dl_src, vlan)) {
945 mac = mac_learning_insert(xbridge->ml, flow->dl_src, vlan);
946 if (is_gratuitous_arp(flow, wc)) {
947 /* We don't want to learn from gratuitous ARP packets that are
948 * reflected back over bond slaves so we lock the learning table. */
949 if (!in_xbundle->bond) {
950 mac_entry_set_grat_arp_lock(mac);
951 } else if (mac_entry_is_grat_arp_locked(mac)) {
956 if (mac->port.p != in_xbundle->ofbundle) {
957 /* The log messages here could actually be useful in debugging,
958 * so keep the rate limit relatively high. */
959 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
960 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
961 "on port %s in VLAN %d",
962 xbridge->name, ETH_ADDR_ARGS(flow->dl_src),
963 in_xbundle->name, vlan);
965 mac->port.p = in_xbundle->ofbundle;
966 mac_learning_changed(xbridge->ml);
969 ovs_rwlock_unlock(&xbridge->ml->rwlock);
972 /* Determines whether packets in 'flow' within 'xbridge' should be forwarded or
973 * dropped. Returns true if they may be forwarded, false if they should be
976 * 'in_port' must be the xport that corresponds to flow->in_port.
977 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
979 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
980 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
981 * checked by input_vid_is_valid().
983 * May also add tags to '*tags', although the current implementation only does
984 * so in one special case.
987 is_admissible(struct xlate_ctx *ctx, struct xport *in_port,
990 struct xbundle *in_xbundle = in_port->xbundle;
991 const struct xbridge *xbridge = ctx->xbridge;
992 struct flow *flow = &ctx->xin->flow;
994 /* Drop frames for reserved multicast addresses
995 * only if forward_bpdu option is absent. */
996 if (!xbridge->forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
997 xlate_report(ctx, "packet has reserved destination MAC, dropping");
1001 if (in_xbundle->bond) {
1002 struct mac_entry *mac;
1004 switch (bond_check_admissibility(in_xbundle->bond, in_port->ofport,
1010 xlate_report(ctx, "bonding refused admissibility, dropping");
1013 case BV_DROP_IF_MOVED:
1014 ovs_rwlock_rdlock(&xbridge->ml->rwlock);
1015 mac = mac_learning_lookup(xbridge->ml, flow->dl_src, vlan);
1016 if (mac && mac->port.p != in_xbundle->ofbundle &&
1017 (!is_gratuitous_arp(flow, &ctx->xout->wc)
1018 || mac_entry_is_grat_arp_locked(mac))) {
1019 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1020 xlate_report(ctx, "SLB bond thinks this packet looped back, "
1024 ovs_rwlock_unlock(&xbridge->ml->rwlock);
1033 xlate_normal(struct xlate_ctx *ctx)
1035 struct flow_wildcards *wc = &ctx->xout->wc;
1036 struct flow *flow = &ctx->xin->flow;
1037 struct xbundle *in_xbundle;
1038 struct xport *in_port;
1039 struct mac_entry *mac;
1043 ctx->xout->has_normal = true;
1045 memset(&wc->masks.dl_src, 0xff, sizeof wc->masks.dl_src);
1046 memset(&wc->masks.dl_dst, 0xff, sizeof wc->masks.dl_dst);
1047 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1049 in_xbundle = lookup_input_bundle(ctx->xbridge, flow->in_port.ofp_port,
1050 ctx->xin->packet != NULL, &in_port);
1052 xlate_report(ctx, "no input bundle, dropping");
1056 /* Drop malformed frames. */
1057 if (flow->dl_type == htons(ETH_TYPE_VLAN) &&
1058 !(flow->vlan_tci & htons(VLAN_CFI))) {
1059 if (ctx->xin->packet != NULL) {
1060 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1061 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
1062 "VLAN tag received on port %s",
1063 ctx->xbridge->name, in_xbundle->name);
1065 xlate_report(ctx, "partial VLAN tag, dropping");
1069 /* Drop frames on bundles reserved for mirroring. */
1070 if (xbundle_mirror_out(ctx->xbridge, in_xbundle)) {
1071 if (ctx->xin->packet != NULL) {
1072 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1073 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
1074 "%s, which is reserved exclusively for mirroring",
1075 ctx->xbridge->name, in_xbundle->name);
1077 xlate_report(ctx, "input port is mirror output port, dropping");
1082 vid = vlan_tci_to_vid(flow->vlan_tci);
1083 if (!input_vid_is_valid(vid, in_xbundle, ctx->xin->packet != NULL)) {
1084 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
1087 vlan = input_vid_to_vlan(in_xbundle, vid);
1089 /* Check other admissibility requirements. */
1090 if (in_port && !is_admissible(ctx, in_port, vlan)) {
1094 /* Learn source MAC. */
1095 if (ctx->xin->may_learn) {
1096 update_learning_table(ctx->xbridge, flow, wc, vlan, in_xbundle);
1099 /* Determine output bundle. */
1100 ovs_rwlock_rdlock(&ctx->xbridge->ml->rwlock);
1101 mac = mac_learning_lookup(ctx->xbridge->ml, flow->dl_dst, vlan);
1103 struct xbundle *mac_xbundle = xbundle_lookup(mac->port.p);
1104 if (mac_xbundle && mac_xbundle != in_xbundle) {
1105 xlate_report(ctx, "forwarding to learned port");
1106 output_normal(ctx, mac_xbundle, vlan);
1107 } else if (!mac_xbundle) {
1108 xlate_report(ctx, "learned port is unknown, dropping");
1110 xlate_report(ctx, "learned port is input port, dropping");
1113 struct xbundle *xbundle;
1115 xlate_report(ctx, "no learned MAC for destination, flooding");
1116 LIST_FOR_EACH (xbundle, list_node, &ctx->xbridge->xbundles) {
1117 if (xbundle != in_xbundle
1118 && xbundle_includes_vlan(xbundle, vlan)
1119 && xbundle->floodable
1120 && !xbundle_mirror_out(ctx->xbridge, xbundle)) {
1121 output_normal(ctx, xbundle, vlan);
1124 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1126 ovs_rwlock_unlock(&ctx->xbridge->ml->rwlock);
1129 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
1130 * the number of packets out of UINT32_MAX to sample. The given
1131 * cookie is passed back in the callback for each sampled packet.
1134 compose_sample_action(const struct xbridge *xbridge,
1135 struct ofpbuf *odp_actions,
1136 const struct flow *flow,
1137 const uint32_t probability,
1138 const union user_action_cookie *cookie,
1139 const size_t cookie_size)
1141 size_t sample_offset, actions_offset;
1144 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
1146 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
1148 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
1149 cookie_offset = put_userspace_action(xbridge->ofproto, odp_actions, flow,
1150 cookie, cookie_size);
1152 nl_msg_end_nested(odp_actions, actions_offset);
1153 nl_msg_end_nested(odp_actions, sample_offset);
1154 return cookie_offset;
1158 compose_sflow_cookie(const struct xbridge *xbridge, ovs_be16 vlan_tci,
1159 odp_port_t odp_port, unsigned int n_outputs,
1160 union user_action_cookie *cookie)
1164 cookie->type = USER_ACTION_COOKIE_SFLOW;
1165 cookie->sflow.vlan_tci = vlan_tci;
1167 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
1168 * port information") for the interpretation of cookie->output. */
1169 switch (n_outputs) {
1171 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
1172 cookie->sflow.output = 0x40000000 | 256;
1176 ifindex = dpif_sflow_odp_port_to_ifindex(xbridge->sflow, odp_port);
1178 cookie->sflow.output = ifindex;
1183 /* 0x80000000 means "multiple output ports. */
1184 cookie->sflow.output = 0x80000000 | n_outputs;
1189 /* Compose SAMPLE action for sFlow bridge sampling. */
1191 compose_sflow_action(const struct xbridge *xbridge,
1192 struct ofpbuf *odp_actions,
1193 const struct flow *flow,
1194 odp_port_t odp_port)
1196 uint32_t probability;
1197 union user_action_cookie cookie;
1199 if (!xbridge->sflow || flow->in_port.ofp_port == OFPP_NONE) {
1203 probability = dpif_sflow_get_probability(xbridge->sflow);
1204 compose_sflow_cookie(xbridge, htons(0), odp_port,
1205 odp_port == ODPP_NONE ? 0 : 1, &cookie);
1207 return compose_sample_action(xbridge, odp_actions, flow, probability,
1208 &cookie, sizeof cookie.sflow);
1212 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
1213 uint32_t obs_domain_id, uint32_t obs_point_id,
1214 union user_action_cookie *cookie)
1216 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
1217 cookie->flow_sample.probability = probability;
1218 cookie->flow_sample.collector_set_id = collector_set_id;
1219 cookie->flow_sample.obs_domain_id = obs_domain_id;
1220 cookie->flow_sample.obs_point_id = obs_point_id;
1224 compose_ipfix_cookie(union user_action_cookie *cookie)
1226 cookie->type = USER_ACTION_COOKIE_IPFIX;
1229 /* Compose SAMPLE action for IPFIX bridge sampling. */
1231 compose_ipfix_action(const struct xbridge *xbridge,
1232 struct ofpbuf *odp_actions,
1233 const struct flow *flow)
1235 uint32_t probability;
1236 union user_action_cookie cookie;
1238 if (!xbridge->ipfix || flow->in_port.ofp_port == OFPP_NONE) {
1242 probability = dpif_ipfix_get_bridge_exporter_probability(xbridge->ipfix);
1243 compose_ipfix_cookie(&cookie);
1245 compose_sample_action(xbridge, odp_actions, flow, probability,
1246 &cookie, sizeof cookie.ipfix);
1249 /* SAMPLE action for sFlow must be first action in any given list of
1250 * actions. At this point we do not have all information required to
1251 * build it. So try to build sample action as complete as possible. */
1253 add_sflow_action(struct xlate_ctx *ctx)
1255 ctx->user_cookie_offset = compose_sflow_action(ctx->xbridge,
1256 &ctx->xout->odp_actions,
1257 &ctx->xin->flow, ODPP_NONE);
1258 ctx->sflow_odp_port = 0;
1259 ctx->sflow_n_outputs = 0;
1262 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
1263 * of actions, eventually after the SAMPLE action for sFlow. */
1265 add_ipfix_action(struct xlate_ctx *ctx)
1267 compose_ipfix_action(ctx->xbridge, &ctx->xout->odp_actions,
1271 /* Fix SAMPLE action according to data collected while composing ODP actions.
1272 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
1273 * USERSPACE action's user-cookie which is required for sflow. */
1275 fix_sflow_action(struct xlate_ctx *ctx)
1277 const struct flow *base = &ctx->base_flow;
1278 union user_action_cookie *cookie;
1280 if (!ctx->user_cookie_offset) {
1284 cookie = ofpbuf_at(&ctx->xout->odp_actions, ctx->user_cookie_offset,
1285 sizeof cookie->sflow);
1286 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
1288 compose_sflow_cookie(ctx->xbridge, base->vlan_tci,
1289 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
1292 static enum slow_path_reason
1293 process_special(struct xlate_ctx *ctx, const struct flow *flow,
1294 const struct xport *xport, const struct ofpbuf *packet)
1296 struct flow_wildcards *wc = &ctx->xout->wc;
1297 const struct xbridge *xbridge = ctx->xbridge;
1301 } else if (xport->cfm && cfm_should_process_flow(xport->cfm, flow, wc)) {
1303 cfm_process_heartbeat(xport->cfm, packet);
1306 } else if (xport->bfd && bfd_should_process_flow(xport->bfd, flow, wc)) {
1308 bfd_process_packet(xport->bfd, flow, packet);
1311 } else if (xport->xbundle && xport->xbundle->lacp
1312 && flow->dl_type == htons(ETH_TYPE_LACP)) {
1314 lacp_process_packet(xport->xbundle->lacp, xport->ofport, packet);
1317 } else if (xbridge->stp && stp_should_process_flow(flow, wc)) {
1319 stp_process_packet(xport, packet);
1328 compose_output_action__(struct xlate_ctx *ctx, ofp_port_t ofp_port,
1331 const struct xport *xport = get_ofp_port(ctx->xbridge, ofp_port);
1332 struct flow_wildcards *wc = &ctx->xout->wc;
1333 struct flow *flow = &ctx->xin->flow;
1334 ovs_be16 flow_vlan_tci;
1335 uint32_t flow_skb_mark;
1336 uint8_t flow_nw_tos;
1337 odp_port_t out_port, odp_port;
1340 /* If 'struct flow' gets additional metadata, we'll need to zero it out
1341 * before traversing a patch port. */
1342 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
1345 xlate_report(ctx, "Nonexistent output port");
1347 } else if (xport->config & OFPUTIL_PC_NO_FWD) {
1348 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
1350 } else if (check_stp && !xport_stp_forward_state(xport)) {
1351 xlate_report(ctx, "STP not in forwarding state, skipping output");
1355 if (mbridge_has_mirrors(ctx->xbridge->mbridge) && xport->xbundle) {
1356 ctx->xout->mirrors |= xbundle_mirror_dst(xport->xbundle->xbridge,
1361 const struct xport *peer = xport->peer;
1362 struct flow old_flow = ctx->xin->flow;
1363 enum slow_path_reason special;
1365 ctx->xbridge = peer->xbridge;
1366 flow->in_port.ofp_port = peer->ofp_port;
1367 flow->metadata = htonll(0);
1368 memset(&flow->tunnel, 0, sizeof flow->tunnel);
1369 memset(flow->regs, 0, sizeof flow->regs);
1371 special = process_special(ctx, &ctx->xin->flow, peer,
1374 ctx->xout->slow = special;
1375 } else if (may_receive(peer, ctx)) {
1376 if (xport_stp_forward_state(peer)) {
1377 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1379 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
1380 * learning action look at the packet, then drop it. */
1381 struct flow old_base_flow = ctx->base_flow;
1382 size_t old_size = ctx->xout->odp_actions.size;
1383 mirror_mask_t old_mirrors = ctx->xout->mirrors;
1384 xlate_table_action(ctx, flow->in_port.ofp_port, 0, true);
1385 ctx->xout->mirrors = old_mirrors;
1386 ctx->base_flow = old_base_flow;
1387 ctx->xout->odp_actions.size = old_size;
1391 ctx->xin->flow = old_flow;
1392 ctx->xbridge = xport->xbundle->xbridge;
1394 if (ctx->xin->resubmit_stats) {
1395 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1396 netdev_vport_inc_rx(peer->netdev, ctx->xin->resubmit_stats);
1402 flow_vlan_tci = flow->vlan_tci;
1403 flow_skb_mark = flow->skb_mark;
1404 flow_nw_tos = flow->nw_tos;
1406 if (dscp_from_skb_priority(xport, flow->skb_priority, &dscp)) {
1407 wc->masks.nw_tos |= IP_ECN_MASK;
1408 flow->nw_tos &= ~IP_DSCP_MASK;
1409 flow->nw_tos |= dscp;
1412 if (xport->is_tunnel) {
1413 /* Save tunnel metadata so that changes made due to
1414 * the Logical (tunnel) Port are not visible for any further
1415 * matches, while explicit set actions on tunnel metadata are.
1417 struct flow_tnl flow_tnl = flow->tunnel;
1418 odp_port = tnl_port_send(xport->ofport, flow, &ctx->xout->wc);
1419 if (odp_port == ODPP_NONE) {
1420 xlate_report(ctx, "Tunneling decided against output");
1421 goto out; /* restore flow_nw_tos */
1423 if (flow->tunnel.ip_dst == ctx->orig_tunnel_ip_dst) {
1424 xlate_report(ctx, "Not tunneling to our own address");
1425 goto out; /* restore flow_nw_tos */
1427 if (ctx->xin->resubmit_stats) {
1428 netdev_vport_inc_tx(xport->netdev, ctx->xin->resubmit_stats);
1430 out_port = odp_port;
1431 commit_odp_tunnel_action(flow, &ctx->base_flow,
1432 &ctx->xout->odp_actions);
1433 flow->tunnel = flow_tnl; /* Restore tunnel metadata */
1435 ofp_port_t vlandev_port;
1437 odp_port = xport->odp_port;
1438 if (ofproto_has_vlan_splinters(ctx->xbridge->ofproto)) {
1439 wc->masks.vlan_tci |= htons(VLAN_VID_MASK | VLAN_CFI);
1441 vlandev_port = vsp_realdev_to_vlandev(ctx->xbridge->ofproto, ofp_port,
1443 if (vlandev_port == ofp_port) {
1444 out_port = odp_port;
1446 out_port = ofp_port_to_odp_port(ctx->xbridge, vlandev_port);
1447 flow->vlan_tci = htons(0);
1449 flow->skb_mark &= ~IPSEC_MARK;
1452 if (out_port != ODPP_NONE) {
1453 commit_odp_actions(flow, &ctx->base_flow,
1454 &ctx->xout->odp_actions, &ctx->xout->wc);
1455 nl_msg_put_odp_port(&ctx->xout->odp_actions, OVS_ACTION_ATTR_OUTPUT,
1458 ctx->sflow_odp_port = odp_port;
1459 ctx->sflow_n_outputs++;
1460 ctx->xout->nf_output_iface = ofp_port;
1465 flow->vlan_tci = flow_vlan_tci;
1466 flow->skb_mark = flow_skb_mark;
1467 flow->nw_tos = flow_nw_tos;
1471 compose_output_action(struct xlate_ctx *ctx, ofp_port_t ofp_port)
1473 compose_output_action__(ctx, ofp_port, true);
1476 /* Common rule processing in one place to avoid duplicating code. */
1477 static struct rule_dpif *
1478 ctx_rule_hooks(struct xlate_ctx *ctx, struct rule_dpif *rule,
1481 if (ctx->xin->resubmit_hook) {
1482 ctx->xin->resubmit_hook(ctx->xin, rule, ctx->recurse);
1484 if (rule == NULL && may_packet_in) {
1486 * check if table configuration flags
1487 * OFPTC_TABLE_MISS_CONTROLLER, default.
1488 * OFPTC_TABLE_MISS_CONTINUE,
1489 * OFPTC_TABLE_MISS_DROP
1490 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
1492 rule = rule_dpif_miss_rule(ctx->xbridge->ofproto, &ctx->xin->flow);
1494 if (rule && ctx->xin->resubmit_stats) {
1495 rule_credit_stats(rule, ctx->xin->resubmit_stats);
1501 xlate_table_action(struct xlate_ctx *ctx,
1502 ofp_port_t in_port, uint8_t table_id, bool may_packet_in)
1504 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
1505 struct rule_dpif *rule;
1506 ofp_port_t old_in_port = ctx->xin->flow.in_port.ofp_port;
1507 uint8_t old_table_id = ctx->table_id;
1509 ctx->table_id = table_id;
1511 /* Look up a flow with 'in_port' as the input port. */
1512 ctx->xin->flow.in_port.ofp_port = in_port;
1513 rule = rule_dpif_lookup_in_table(ctx->xbridge->ofproto,
1514 &ctx->xin->flow, &ctx->xout->wc,
1517 /* Restore the original input port. Otherwise OFPP_NORMAL and
1518 * OFPP_IN_PORT will have surprising behavior. */
1519 ctx->xin->flow.in_port.ofp_port = old_in_port;
1521 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
1524 struct rule_dpif *old_rule = ctx->rule;
1528 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
1529 ctx->rule = old_rule;
1533 ctx->table_id = old_table_id;
1535 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
1537 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
1538 MAX_RESUBMIT_RECURSION);
1539 ctx->max_resubmit_trigger = true;
1544 xlate_ofpact_resubmit(struct xlate_ctx *ctx,
1545 const struct ofpact_resubmit *resubmit)
1550 in_port = resubmit->in_port;
1551 if (in_port == OFPP_IN_PORT) {
1552 in_port = ctx->xin->flow.in_port.ofp_port;
1555 table_id = resubmit->table_id;
1556 if (table_id == 255) {
1557 table_id = ctx->table_id;
1560 xlate_table_action(ctx, in_port, table_id, false);
1564 flood_packets(struct xlate_ctx *ctx, bool all)
1566 const struct xport *xport;
1568 HMAP_FOR_EACH (xport, ofp_node, &ctx->xbridge->xports) {
1569 if (xport->ofp_port == ctx->xin->flow.in_port.ofp_port) {
1574 compose_output_action__(ctx, xport->ofp_port, false);
1575 } else if (!(xport->config & OFPUTIL_PC_NO_FLOOD)) {
1576 compose_output_action(ctx, xport->ofp_port);
1580 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1584 execute_controller_action(struct xlate_ctx *ctx, int len,
1585 enum ofp_packet_in_reason reason,
1586 uint16_t controller_id)
1588 struct ofputil_packet_in pin;
1589 struct ofpbuf *packet;
1592 ovs_assert(!ctx->xout->slow || ctx->xout->slow == SLOW_CONTROLLER);
1593 ctx->xout->slow = SLOW_CONTROLLER;
1594 if (!ctx->xin->packet) {
1598 packet = ofpbuf_clone(ctx->xin->packet);
1600 key.skb_priority = 0;
1602 memset(&key.tunnel, 0, sizeof key.tunnel);
1604 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1605 &ctx->xout->odp_actions, &ctx->xout->wc);
1607 odp_execute_actions(NULL, packet, &key, ctx->xout->odp_actions.data,
1608 ctx->xout->odp_actions.size, NULL, NULL);
1610 pin.packet = packet->data;
1611 pin.packet_len = packet->size;
1612 pin.reason = reason;
1613 pin.controller_id = controller_id;
1614 pin.table_id = ctx->table_id;
1615 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
1618 flow_get_metadata(&ctx->xin->flow, &pin.fmd);
1620 ofproto_dpif_send_packet_in(ctx->xbridge->ofproto, &pin);
1621 ofpbuf_delete(packet);
1625 compose_mpls_push_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1627 struct flow_wildcards *wc = &ctx->xout->wc;
1628 struct flow *flow = &ctx->xin->flow;
1630 ovs_assert(eth_type_mpls(eth_type));
1632 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1633 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1635 if (flow->mpls_depth) {
1636 flow->mpls_lse &= ~htonl(MPLS_BOS_MASK);
1642 if (flow->dl_type == htons(ETH_TYPE_IPV6)) {
1643 label = htonl(0x2); /* IPV6 Explicit Null. */
1645 label = htonl(0x0); /* IPV4 Explicit Null. */
1647 wc->masks.nw_tos |= IP_DSCP_MASK;
1648 wc->masks.nw_ttl = 0xff;
1649 tc = (flow->nw_tos & IP_DSCP_MASK) >> 2;
1650 ttl = flow->nw_ttl ? flow->nw_ttl : 0x40;
1651 flow->mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
1652 flow->mpls_depth = 1;
1654 flow->dl_type = eth_type;
1658 compose_mpls_pop_action(struct xlate_ctx *ctx, ovs_be16 eth_type)
1660 struct flow_wildcards *wc = &ctx->xout->wc;
1661 struct flow *flow = &ctx->xin->flow;
1663 ovs_assert(eth_type_mpls(ctx->xin->flow.dl_type));
1664 ovs_assert(!eth_type_mpls(eth_type));
1666 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1667 memset(&wc->masks.mpls_depth, 0xff, sizeof wc->masks.mpls_depth);
1669 if (flow->mpls_depth) {
1671 flow->mpls_lse = htonl(0);
1672 if (!flow->mpls_depth) {
1673 flow->dl_type = eth_type;
1679 compose_dec_ttl(struct xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
1681 struct flow *flow = &ctx->xin->flow;
1683 if (!is_ip_any(flow)) {
1687 ctx->xout->wc.masks.nw_ttl = 0xff;
1688 if (flow->nw_ttl > 1) {
1694 for (i = 0; i < ids->n_controllers; i++) {
1695 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
1699 /* Stop processing for current table. */
1705 compose_set_mpls_ttl_action(struct xlate_ctx *ctx, uint8_t ttl)
1707 if (!eth_type_mpls(ctx->xin->flow.dl_type)) {
1711 set_mpls_lse_ttl(&ctx->xin->flow.mpls_lse, ttl);
1716 compose_dec_mpls_ttl_action(struct xlate_ctx *ctx)
1718 struct flow *flow = &ctx->xin->flow;
1719 uint8_t ttl = mpls_lse_to_ttl(flow->mpls_lse);
1720 struct flow_wildcards *wc = &ctx->xout->wc;
1722 memset(&wc->masks.mpls_lse, 0xff, sizeof wc->masks.mpls_lse);
1724 if (!eth_type_mpls(flow->dl_type)) {
1730 set_mpls_lse_ttl(&flow->mpls_lse, ttl);
1733 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
1735 /* Stop processing for current table. */
1741 xlate_output_action(struct xlate_ctx *ctx,
1742 ofp_port_t port, uint16_t max_len, bool may_packet_in)
1744 ofp_port_t prev_nf_output_iface = ctx->xout->nf_output_iface;
1746 ctx->xout->nf_output_iface = NF_OUT_DROP;
1750 compose_output_action(ctx, ctx->xin->flow.in_port.ofp_port);
1753 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
1760 flood_packets(ctx, false);
1763 flood_packets(ctx, true);
1765 case OFPP_CONTROLLER:
1766 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
1772 if (port != ctx->xin->flow.in_port.ofp_port) {
1773 compose_output_action(ctx, port);
1775 xlate_report(ctx, "skipping output to input port");
1780 if (prev_nf_output_iface == NF_OUT_FLOOD) {
1781 ctx->xout->nf_output_iface = NF_OUT_FLOOD;
1782 } else if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1783 ctx->xout->nf_output_iface = prev_nf_output_iface;
1784 } else if (prev_nf_output_iface != NF_OUT_DROP &&
1785 ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1786 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1791 xlate_output_reg_action(struct xlate_ctx *ctx,
1792 const struct ofpact_output_reg *or)
1794 uint64_t port = mf_get_subfield(&or->src, &ctx->xin->flow);
1795 if (port <= UINT16_MAX) {
1796 union mf_subvalue value;
1798 memset(&value, 0xff, sizeof value);
1799 mf_write_subfield_flow(&or->src, &value, &ctx->xout->wc.masks);
1800 xlate_output_action(ctx, u16_to_ofp(port),
1801 or->max_len, false);
1806 xlate_enqueue_action(struct xlate_ctx *ctx,
1807 const struct ofpact_enqueue *enqueue)
1809 ofp_port_t ofp_port = enqueue->port;
1810 uint32_t queue_id = enqueue->queue;
1811 uint32_t flow_priority, priority;
1814 /* Translate queue to priority. */
1815 error = ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1818 /* Fall back to ordinary output action. */
1819 xlate_output_action(ctx, enqueue->port, 0, false);
1823 /* Check output port. */
1824 if (ofp_port == OFPP_IN_PORT) {
1825 ofp_port = ctx->xin->flow.in_port.ofp_port;
1826 } else if (ofp_port == ctx->xin->flow.in_port.ofp_port) {
1830 /* Add datapath actions. */
1831 flow_priority = ctx->xin->flow.skb_priority;
1832 ctx->xin->flow.skb_priority = priority;
1833 compose_output_action(ctx, ofp_port);
1834 ctx->xin->flow.skb_priority = flow_priority;
1836 /* Update NetFlow output port. */
1837 if (ctx->xout->nf_output_iface == NF_OUT_DROP) {
1838 ctx->xout->nf_output_iface = ofp_port;
1839 } else if (ctx->xout->nf_output_iface != NF_OUT_FLOOD) {
1840 ctx->xout->nf_output_iface = NF_OUT_MULTI;
1845 xlate_set_queue_action(struct xlate_ctx *ctx, uint32_t queue_id)
1847 uint32_t skb_priority;
1849 if (!ofproto_dpif_queue_to_priority(ctx->xbridge->ofproto, queue_id,
1851 ctx->xin->flow.skb_priority = skb_priority;
1853 /* Couldn't translate queue to a priority. Nothing to do. A warning
1854 * has already been logged. */
1859 slave_enabled_cb(ofp_port_t ofp_port, void *xbridge_)
1861 const struct xbridge *xbridge = xbridge_;
1872 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
1875 port = get_ofp_port(xbridge, ofp_port);
1876 return port ? port->may_enable : false;
1881 xlate_bundle_action(struct xlate_ctx *ctx,
1882 const struct ofpact_bundle *bundle)
1886 port = bundle_execute(bundle, &ctx->xin->flow, &ctx->xout->wc,
1888 CONST_CAST(struct xbridge *, ctx->xbridge));
1889 if (bundle->dst.field) {
1890 nxm_reg_load(&bundle->dst, ofp_to_u16(port), &ctx->xin->flow);
1892 xlate_output_action(ctx, port, 0, false);
1897 xlate_learn_action(struct xlate_ctx *ctx,
1898 const struct ofpact_learn *learn)
1900 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
1901 struct ofputil_flow_mod fm;
1902 uint64_t ofpacts_stub[1024 / 8];
1903 struct ofpbuf ofpacts;
1906 ctx->xout->has_learn = true;
1908 learn_mask(learn, &ctx->xout->wc);
1910 if (!ctx->xin->may_learn) {
1914 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1915 learn_execute(learn, &ctx->xin->flow, &fm, &ofpacts);
1917 error = ofproto_dpif_flow_mod(ctx->xbridge->ofproto, &fm);
1918 if (error && !VLOG_DROP_WARN(&rl)) {
1919 VLOG_WARN("learning action failed to modify flow table (%s)",
1920 ofperr_get_name(error));
1923 ofpbuf_uninit(&ofpacts);
1926 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
1927 * means "infinite". */
1929 reduce_timeout(uint16_t max, uint16_t *timeout)
1931 if (max && (!*timeout || *timeout > max)) {
1937 xlate_fin_timeout(struct xlate_ctx *ctx,
1938 const struct ofpact_fin_timeout *oft)
1940 if (ctx->xin->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
1941 struct rule_dpif *rule = ctx->rule;
1943 if (list_is_empty(&rule->up.expirable)) {
1944 list_insert(&rule->up.ofproto->expirable, &rule->up.expirable);
1947 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
1948 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
1953 xlate_sample_action(struct xlate_ctx *ctx,
1954 const struct ofpact_sample *os)
1956 union user_action_cookie cookie;
1957 /* Scale the probability from 16-bit to 32-bit while representing
1958 * the same percentage. */
1959 uint32_t probability = (os->probability << 16) | os->probability;
1961 commit_odp_actions(&ctx->xin->flow, &ctx->base_flow,
1962 &ctx->xout->odp_actions, &ctx->xout->wc);
1964 compose_flow_sample_cookie(os->probability, os->collector_set_id,
1965 os->obs_domain_id, os->obs_point_id, &cookie);
1966 compose_sample_action(ctx->xbridge, &ctx->xout->odp_actions, &ctx->xin->flow,
1967 probability, &cookie, sizeof cookie.flow_sample);
1971 may_receive(const struct xport *xport, struct xlate_ctx *ctx)
1973 if (xport->config & (eth_addr_equals(ctx->xin->flow.dl_dst, eth_addr_stp)
1974 ? OFPUTIL_PC_NO_RECV_STP
1975 : OFPUTIL_PC_NO_RECV)) {
1979 /* Only drop packets here if both forwarding and learning are
1980 * disabled. If just learning is enabled, we need to have
1981 * OFPP_NORMAL and the learning action have a look at the packet
1982 * before we can drop it. */
1983 if (!xport_stp_forward_state(xport) && !xport_stp_learn_state(xport)) {
1991 tunnel_ecn_ok(struct xlate_ctx *ctx)
1993 if (is_ip_any(&ctx->base_flow)
1994 && (ctx->xin->flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
1995 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
1996 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
1997 " but is not ECN capable");
2000 /* Set the ECN CE value in the tunneled packet. */
2001 ctx->xin->flow.nw_tos |= IP_ECN_CE;
2009 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
2010 struct xlate_ctx *ctx)
2012 struct flow_wildcards *wc = &ctx->xout->wc;
2013 struct flow *flow = &ctx->xin->flow;
2014 bool was_evictable = true;
2015 const struct ofpact *a;
2018 /* Don't let the rule we're working on get evicted underneath us. */
2019 was_evictable = ctx->rule->up.evictable;
2020 ctx->rule->up.evictable = false;
2023 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2024 struct ofpact_controller *controller;
2025 const struct ofpact_metadata *metadata;
2033 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
2034 ofpact_get_OUTPUT(a)->max_len, true);
2037 case OFPACT_CONTROLLER:
2038 controller = ofpact_get_CONTROLLER(a);
2039 execute_controller_action(ctx, controller->max_len,
2041 controller->controller_id);
2044 case OFPACT_ENQUEUE:
2045 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
2048 case OFPACT_SET_VLAN_VID:
2049 flow->vlan_tci &= ~htons(VLAN_VID_MASK);
2050 flow->vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
2054 case OFPACT_SET_VLAN_PCP:
2055 flow->vlan_tci &= ~htons(VLAN_PCP_MASK);
2057 htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp << VLAN_PCP_SHIFT)
2061 case OFPACT_STRIP_VLAN:
2062 flow->vlan_tci = htons(0);
2065 case OFPACT_PUSH_VLAN:
2066 /* XXX 802.1AD(QinQ) */
2067 flow->vlan_tci = htons(VLAN_CFI);
2070 case OFPACT_SET_ETH_SRC:
2071 memcpy(flow->dl_src, ofpact_get_SET_ETH_SRC(a)->mac, ETH_ADDR_LEN);
2074 case OFPACT_SET_ETH_DST:
2075 memcpy(flow->dl_dst, ofpact_get_SET_ETH_DST(a)->mac, ETH_ADDR_LEN);
2078 case OFPACT_SET_IPV4_SRC:
2079 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2080 flow->nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
2084 case OFPACT_SET_IPV4_DST:
2085 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2086 flow->nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
2090 case OFPACT_SET_IPV4_DSCP:
2091 /* OpenFlow 1.0 only supports IPv4. */
2092 if (flow->dl_type == htons(ETH_TYPE_IP)) {
2093 flow->nw_tos &= ~IP_DSCP_MASK;
2094 flow->nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
2098 case OFPACT_SET_L4_SRC_PORT:
2099 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2100 if (is_ip_any(flow)) {
2101 flow->tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
2105 case OFPACT_SET_L4_DST_PORT:
2106 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2107 if (is_ip_any(flow)) {
2108 flow->tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
2112 case OFPACT_RESUBMIT:
2113 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
2116 case OFPACT_SET_TUNNEL:
2117 flow->tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
2120 case OFPACT_SET_QUEUE:
2121 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
2124 case OFPACT_POP_QUEUE:
2125 flow->skb_priority = ctx->orig_skb_priority;
2128 case OFPACT_REG_MOVE:
2129 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), flow, wc);
2132 case OFPACT_REG_LOAD:
2133 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), flow);
2136 case OFPACT_STACK_PUSH:
2137 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), flow, wc,
2141 case OFPACT_STACK_POP:
2142 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), flow, &ctx->stack);
2145 case OFPACT_PUSH_MPLS:
2146 compose_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
2149 case OFPACT_POP_MPLS:
2150 compose_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
2153 case OFPACT_SET_MPLS_TTL:
2154 if (compose_set_mpls_ttl_action(ctx,
2155 ofpact_get_SET_MPLS_TTL(a)->ttl)) {
2160 case OFPACT_DEC_MPLS_TTL:
2161 if (compose_dec_mpls_ttl_action(ctx)) {
2166 case OFPACT_DEC_TTL:
2167 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
2173 /* Nothing to do. */
2176 case OFPACT_MULTIPATH:
2177 multipath_execute(ofpact_get_MULTIPATH(a), flow, wc);
2181 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
2184 case OFPACT_OUTPUT_REG:
2185 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
2189 xlate_learn_action(ctx, ofpact_get_LEARN(a));
2196 case OFPACT_FIN_TIMEOUT:
2197 memset(&wc->masks.nw_proto, 0xff, sizeof wc->masks.nw_proto);
2198 ctx->xout->has_fin_timeout = true;
2199 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
2202 case OFPACT_CLEAR_ACTIONS:
2204 * Nothing to do because writa-actions is not supported for now.
2205 * When writa-actions is supported, clear-actions also must
2206 * be supported at the same time.
2210 case OFPACT_WRITE_METADATA:
2211 metadata = ofpact_get_WRITE_METADATA(a);
2212 flow->metadata &= ~metadata->mask;
2213 flow->metadata |= metadata->metadata & metadata->mask;
2217 /* Not implemented yet. */
2220 case OFPACT_GOTO_TABLE: {
2221 /* It is assumed that goto-table is the last action. */
2222 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
2224 ovs_assert(ctx->table_id < ogt->table_id);
2225 xlate_table_action(ctx, ctx->xin->flow.in_port.ofp_port,
2226 ogt->table_id, true);
2231 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
2238 ctx->rule->up.evictable = was_evictable;
2243 xlate_in_init(struct xlate_in *xin, struct ofproto_dpif *ofproto,
2244 const struct flow *flow, struct rule_dpif *rule,
2245 uint8_t tcp_flags, const struct ofpbuf *packet)
2247 xin->ofproto = ofproto;
2249 xin->packet = packet;
2250 xin->may_learn = packet != NULL;
2252 xin->ofpacts = NULL;
2253 xin->ofpacts_len = 0;
2254 xin->tcp_flags = tcp_flags;
2255 xin->resubmit_hook = NULL;
2256 xin->report_hook = NULL;
2257 xin->resubmit_stats = NULL;
2261 xlate_out_uninit(struct xlate_out *xout)
2264 ofpbuf_uninit(&xout->odp_actions);
2268 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
2269 * into datapath actions, using 'ctx', and discards the datapath actions. */
2271 xlate_actions_for_side_effects(struct xlate_in *xin)
2273 struct xlate_out xout;
2275 xlate_actions(xin, &xout);
2276 xlate_out_uninit(&xout);
2280 xlate_report(struct xlate_ctx *ctx, const char *s)
2282 if (ctx->xin->report_hook) {
2283 ctx->xin->report_hook(ctx->xin, s, ctx->recurse);
2288 xlate_out_copy(struct xlate_out *dst, const struct xlate_out *src)
2291 dst->slow = src->slow;
2292 dst->has_learn = src->has_learn;
2293 dst->has_normal = src->has_normal;
2294 dst->has_fin_timeout = src->has_fin_timeout;
2295 dst->nf_output_iface = src->nf_output_iface;
2296 dst->mirrors = src->mirrors;
2298 ofpbuf_use_stub(&dst->odp_actions, dst->odp_actions_stub,
2299 sizeof dst->odp_actions_stub);
2300 ofpbuf_put(&dst->odp_actions, src->odp_actions.data,
2301 src->odp_actions.size);
2304 static struct skb_priority_to_dscp *
2305 get_skb_priority(const struct xport *xport, uint32_t skb_priority)
2307 struct skb_priority_to_dscp *pdscp;
2310 hash = hash_int(skb_priority, 0);
2311 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &xport->skb_priorities) {
2312 if (pdscp->skb_priority == skb_priority) {
2320 dscp_from_skb_priority(const struct xport *xport, uint32_t skb_priority,
2323 struct skb_priority_to_dscp *pdscp = get_skb_priority(xport, skb_priority);
2324 *dscp = pdscp ? pdscp->dscp : 0;
2325 return pdscp != NULL;
2329 clear_skb_priorities(struct xport *xport)
2331 struct skb_priority_to_dscp *pdscp, *next;
2333 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &xport->skb_priorities) {
2334 hmap_remove(&xport->skb_priorities, &pdscp->hmap_node);
2340 actions_output_to_local_port(const struct xlate_ctx *ctx)
2342 odp_port_t local_odp_port = ofp_port_to_odp_port(ctx->xbridge, OFPP_LOCAL);
2343 const struct nlattr *a;
2346 NL_ATTR_FOR_EACH_UNSAFE (a, left, ctx->xout->odp_actions.data,
2347 ctx->xout->odp_actions.size) {
2348 if (nl_attr_type(a) == OVS_ACTION_ATTR_OUTPUT
2349 && nl_attr_get_odp_port(a) == local_odp_port) {
2356 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
2357 * into datapath actions in 'odp_actions', using 'ctx'. */
2359 xlate_actions(struct xlate_in *xin, struct xlate_out *xout)
2361 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
2362 * that in the future we always keep a copy of the original flow for
2363 * tracing purposes. */
2364 static bool hit_resubmit_limit;
2366 struct flow_wildcards *wc = &xout->wc;
2367 struct flow *flow = &xin->flow;
2369 enum slow_path_reason special;
2370 const struct ofpact *ofpacts;
2371 struct xport *in_port;
2372 struct flow orig_flow;
2373 struct xlate_ctx ctx;
2376 COVERAGE_INC(xlate_actions);
2378 /* Flow initialization rules:
2379 * - 'base_flow' must match the kernel's view of the packet at the
2380 * time that action processing starts. 'flow' represents any
2381 * transformations we wish to make through actions.
2382 * - By default 'base_flow' and 'flow' are the same since the input
2383 * packet matches the output before any actions are applied.
2384 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
2385 * of the received packet as seen by the kernel. If we later output
2386 * to another device without any modifications this will cause us to
2387 * insert a new tag since the original one was stripped off by the
2389 * - Tunnel metadata as received is retained in 'flow'. This allows
2390 * tunnel metadata matching also in later tables.
2391 * Since a kernel action for setting the tunnel metadata will only be
2392 * generated with actual tunnel output, changing the tunnel metadata
2393 * values in 'flow' (such as tun_id) will only have effect with a later
2394 * tunnel output action.
2395 * - Tunnel 'base_flow' is completely cleared since that is what the
2396 * kernel does. If we wish to maintain the original values an action
2397 * needs to be generated. */
2402 ctx.xout->has_learn = false;
2403 ctx.xout->has_normal = false;
2404 ctx.xout->has_fin_timeout = false;
2405 ctx.xout->nf_output_iface = NF_OUT_DROP;
2406 ctx.xout->mirrors = 0;
2407 ofpbuf_use_stub(&ctx.xout->odp_actions, ctx.xout->odp_actions_stub,
2408 sizeof ctx.xout->odp_actions_stub);
2409 ofpbuf_reserve(&ctx.xout->odp_actions, NL_A_U32_SIZE);
2411 ctx.xbridge = xbridge_lookup(xin->ofproto);
2416 ctx.rule = xin->rule;
2418 ctx.base_flow = *flow;
2419 memset(&ctx.base_flow.tunnel, 0, sizeof ctx.base_flow.tunnel);
2420 ctx.orig_tunnel_ip_dst = flow->tunnel.ip_dst;
2422 flow_wildcards_init_catchall(wc);
2423 memset(&wc->masks.in_port, 0xff, sizeof wc->masks.in_port);
2424 memset(&wc->masks.skb_priority, 0xff, sizeof wc->masks.skb_priority);
2425 memset(&wc->masks.dl_type, 0xff, sizeof wc->masks.dl_type);
2426 wc->masks.nw_frag |= FLOW_NW_FRAG_MASK;
2428 if (tnl_port_should_receive(&ctx.xin->flow)) {
2429 memset(&wc->masks.tunnel, 0xff, sizeof wc->masks.tunnel);
2430 /* skb_mark is currently used only by tunnels but that will likely
2431 * change in the future. */
2432 memset(&wc->masks.skb_mark, 0xff, sizeof wc->masks.skb_mark);
2434 if (ctx.xbridge->has_netflow) {
2435 netflow_mask_wc(flow, wc);
2439 ctx.max_resubmit_trigger = false;
2440 ctx.orig_skb_priority = flow->skb_priority;
2445 ofpacts = xin->ofpacts;
2446 ofpacts_len = xin->ofpacts_len;
2447 } else if (xin->rule) {
2448 ofpacts = xin->rule->up.ofpacts;
2449 ofpacts_len = xin->rule->up.ofpacts_len;
2454 ofpbuf_use_stub(&ctx.stack, ctx.init_stack, sizeof ctx.init_stack);
2456 if (mbridge_has_mirrors(ctx.xbridge->mbridge) || hit_resubmit_limit) {
2457 /* Do this conditionally because the copy is expensive enough that it
2458 * shows up in profiles. */
2462 if (flow->nw_frag & FLOW_NW_FRAG_ANY) {
2463 switch (ctx.xbridge->frag) {
2464 case OFPC_FRAG_NORMAL:
2465 /* We must pretend that transport ports are unavailable. */
2466 flow->tp_src = ctx.base_flow.tp_src = htons(0);
2467 flow->tp_dst = ctx.base_flow.tp_dst = htons(0);
2470 case OFPC_FRAG_DROP:
2473 case OFPC_FRAG_REASM:
2476 case OFPC_FRAG_NX_MATCH:
2477 /* Nothing to do. */
2480 case OFPC_INVALID_TTL_TO_CONTROLLER:
2485 in_port = get_ofp_port(ctx.xbridge, flow->in_port.ofp_port);
2486 special = process_special(&ctx, flow, in_port, ctx.xin->packet);
2488 ctx.xout->slow = special;
2490 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
2491 size_t sample_actions_len;
2493 if (flow->in_port.ofp_port
2494 != vsp_realdev_to_vlandev(ctx.xbridge->ofproto,
2495 flow->in_port.ofp_port,
2497 ctx.base_flow.vlan_tci = 0;
2500 add_sflow_action(&ctx);
2501 add_ipfix_action(&ctx);
2502 sample_actions_len = ctx.xout->odp_actions.size;
2504 if (tunnel_ecn_ok(&ctx) && (!in_port || may_receive(in_port, &ctx))) {
2505 do_xlate_actions(ofpacts, ofpacts_len, &ctx);
2507 /* We've let OFPP_NORMAL and the learning action look at the
2508 * packet, so drop it now if forwarding is disabled. */
2509 if (in_port && !xport_stp_forward_state(in_port)) {
2510 ctx.xout->odp_actions.size = sample_actions_len;
2514 if (ctx.max_resubmit_trigger && !ctx.xin->resubmit_hook) {
2515 if (!hit_resubmit_limit) {
2516 /* We didn't record the original flow. Make sure we do from
2518 hit_resubmit_limit = true;
2519 } else if (!VLOG_DROP_ERR(&trace_rl)) {
2520 struct ds ds = DS_EMPTY_INITIALIZER;
2522 ofproto_trace(ctx.xbridge->ofproto, &orig_flow,
2523 ctx.xin->packet, &ds);
2524 VLOG_ERR("Trace triggered by excessive resubmit "
2525 "recursion:\n%s", ds_cstr(&ds));
2530 if (ctx.xbridge->has_in_band
2531 && in_band_must_output_to_local_port(flow)
2532 && !actions_output_to_local_port(&ctx)) {
2533 compose_output_action(&ctx, OFPP_LOCAL);
2536 fix_sflow_action(&ctx);
2538 if (mbridge_has_mirrors(ctx.xbridge->mbridge)) {
2539 add_mirror_actions(&ctx, &orig_flow);
2543 ofpbuf_uninit(&ctx.stack);
2545 /* Clear the metadata and register wildcard masks, because we won't
2546 * use non-header fields as part of the cache. */
2547 memset(&wc->masks.metadata, 0, sizeof wc->masks.metadata);
2548 memset(&wc->masks.regs, 0, sizeof wc->masks.regs);