2 * Copyright (c) 2009, 2010 Nicira Networks.
3 * Copyright (c) 2010 Jean Tourrilhes - HP-Labs.
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
22 #include <sys/socket.h>
24 #include <netinet/in.h>
27 #include "byte-order.h"
28 #include "classifier.h"
30 #include "discovery.h"
32 #include "dynamic-string.h"
33 #include "fail-open.h"
37 #include "mac-learning.h"
42 #include "ofp-print.h"
44 #include "ofproto-sflow.h"
46 #include "openflow/nicira-ext.h"
47 #include "openflow/openflow.h"
48 #include "openvswitch/datapath-protocol.h"
52 #include "poll-loop.h"
56 #include "stream-ssl.h"
64 VLOG_DEFINE_THIS_MODULE(ofproto
);
66 COVERAGE_DEFINE(facet_changed_rule
);
67 COVERAGE_DEFINE(facet_revalidate
);
68 COVERAGE_DEFINE(odp_overflow
);
69 COVERAGE_DEFINE(ofproto_agg_request
);
70 COVERAGE_DEFINE(ofproto_costly_flags
);
71 COVERAGE_DEFINE(ofproto_ctlr_action
);
72 COVERAGE_DEFINE(ofproto_del_rule
);
73 COVERAGE_DEFINE(ofproto_error
);
74 COVERAGE_DEFINE(ofproto_expiration
);
75 COVERAGE_DEFINE(ofproto_expired
);
76 COVERAGE_DEFINE(ofproto_flows_req
);
77 COVERAGE_DEFINE(ofproto_flush
);
78 COVERAGE_DEFINE(ofproto_invalidated
);
79 COVERAGE_DEFINE(ofproto_no_packet_in
);
80 COVERAGE_DEFINE(ofproto_ofconn_stuck
);
81 COVERAGE_DEFINE(ofproto_ofp2odp
);
82 COVERAGE_DEFINE(ofproto_packet_in
);
83 COVERAGE_DEFINE(ofproto_packet_out
);
84 COVERAGE_DEFINE(ofproto_queue_req
);
85 COVERAGE_DEFINE(ofproto_recv_openflow
);
86 COVERAGE_DEFINE(ofproto_reinit_ports
);
87 COVERAGE_DEFINE(ofproto_unexpected_rule
);
88 COVERAGE_DEFINE(ofproto_uninstallable
);
89 COVERAGE_DEFINE(ofproto_update_port
);
91 #include "sflow_api.h"
96 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
97 struct netdev
*netdev
;
98 struct ofp_phy_port opp
; /* In host byte order. */
102 static void ofport_free(struct ofport
*);
103 static void hton_ofp_phy_port(struct ofp_phy_port
*);
105 struct action_xlate_ctx
{
106 /* action_xlate_ctx_init() initializes these members. */
109 struct ofproto
*ofproto
;
111 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
112 * this flow when actions change header fields. */
115 /* The packet corresponding to 'flow', or a null pointer if we are
116 * revalidating without a packet to refer to. */
117 const struct ofpbuf
*packet
;
119 /* If nonnull, called just before executing a resubmit action.
121 * This is normally null so the client has to set it manually after
122 * calling action_xlate_ctx_init(). */
123 void (*resubmit_hook
)(struct action_xlate_ctx
*, const struct rule
*);
125 /* xlate_actions() initializes and uses these members. The client might want
126 * to look at them after it returns. */
128 /* Datapath action set. This is xlate_actions()'s primary output. */
129 struct odp_actions out
;
131 tag_type tags
; /* Tags associated with OFPP_NORMAL actions. */
132 bool may_set_up_flow
; /* True ordinarily; false if the actions must
133 * be reassessed for every packet. */
134 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
136 /* xlate_actions() initializes and uses these members, but the client has no
137 * reason to look at them. */
139 int recurse
; /* Recursion level, via xlate_table_action. */
142 static void action_xlate_ctx_init(struct action_xlate_ctx
*,
143 struct ofproto
*, const struct flow
*,
144 const struct ofpbuf
*);
145 static int xlate_actions(struct action_xlate_ctx
*ctx
,
146 const union ofp_action
*in
, size_t n_in
);
148 /* An OpenFlow flow. */
150 long long int used
; /* Time last used; time created if not used. */
151 long long int created
; /* Creation time. */
155 * - Do include packets and bytes from facets that have been deleted or
156 * whose own statistics have been folded into the rule.
158 * - Do include packets and bytes sent "by hand" that were accounted to
159 * the rule without any facet being involved (this is a rare corner
160 * case in rule_execute()).
162 * - Do not include packet or bytes that can be obtained from any facet's
163 * packet_count or byte_count member or that can be obtained from the
164 * datapath by, e.g., dpif_flow_get() for any facet.
166 uint64_t packet_count
; /* Number of packets received. */
167 uint64_t byte_count
; /* Number of bytes received. */
169 ovs_be64 flow_cookie
; /* Controller-issued identifier. */
171 struct cls_rule cr
; /* In owning ofproto's classifier. */
172 uint16_t idle_timeout
; /* In seconds from time of last use. */
173 uint16_t hard_timeout
; /* In seconds from time of creation. */
174 bool send_flow_removed
; /* Send a flow removed message? */
175 int n_actions
; /* Number of elements in actions[]. */
176 union ofp_action
*actions
; /* OpenFlow actions. */
177 struct list facets
; /* List of "struct facet"s. */
180 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
181 static bool rule_is_hidden(const struct rule
*);
183 static struct rule
*rule_create(const struct cls_rule
*,
184 const union ofp_action
*, size_t n_actions
,
185 uint16_t idle_timeout
, uint16_t hard_timeout
,
186 ovs_be64 flow_cookie
, bool send_flow_removed
);
187 static void rule_destroy(struct ofproto
*, struct rule
*);
188 static void rule_free(struct rule
*);
190 static struct rule
*rule_lookup(struct ofproto
*, const struct flow
*);
191 static void rule_insert(struct ofproto
*, struct rule
*);
192 static void rule_remove(struct ofproto
*, struct rule
*);
194 static void rule_send_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
196 /* An exact-match instantiation of an OpenFlow flow. */
198 long long int used
; /* Time last used; time created if not used. */
202 * - Do include packets and bytes sent "by hand", e.g. with
205 * - Do include packets and bytes that were obtained from the datapath
206 * when a flow was deleted (e.g. dpif_flow_del()) or when its
207 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
209 * - Do not include any packets or bytes that can currently be obtained
210 * from the datapath by, e.g., dpif_flow_get().
212 uint64_t packet_count
; /* Number of packets received. */
213 uint64_t byte_count
; /* Number of bytes received. */
215 /* Number of bytes passed to account_cb. This may include bytes that can
216 * currently obtained from the datapath (thus, it can be greater than
218 uint64_t accounted_bytes
;
220 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
221 struct list list_node
; /* In owning rule's 'facets' list. */
222 struct rule
*rule
; /* Owning rule. */
223 struct flow flow
; /* Exact-match flow. */
224 bool installed
; /* Installed in datapath? */
225 bool may_install
; /* True ordinarily; false if actions must
226 * be reassessed for every packet. */
227 int n_actions
; /* Number of elements in actions[]. */
228 union odp_action
*actions
; /* Datapath actions. */
229 tag_type tags
; /* Tags (set only by hooks). */
230 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
233 static struct facet
*facet_create(struct ofproto
*, struct rule
*,
235 const struct ofpbuf
*packet
);
236 static void facet_remove(struct ofproto
*, struct facet
*);
237 static void facet_free(struct facet
*);
239 static struct facet
*facet_lookup_valid(struct ofproto
*, const struct flow
*);
240 static bool facet_revalidate(struct ofproto
*, struct facet
*);
242 static void facet_install(struct ofproto
*, struct facet
*, bool zero_stats
);
243 static void facet_uninstall(struct ofproto
*, struct facet
*);
244 static void facet_flush_stats(struct ofproto
*, struct facet
*);
246 static void facet_make_actions(struct ofproto
*, struct facet
*,
247 const struct ofpbuf
*packet
);
248 static void facet_update_stats(struct ofproto
*, struct facet
*,
249 const struct odp_flow_stats
*);
251 /* ofproto supports two kinds of OpenFlow connections:
253 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
254 * maintains persistent connections to these controllers and by default
255 * sends them asynchronous messages such as packet-ins.
257 * - "Service" connections, e.g. from ovs-ofctl. When these connections
258 * drop, it is the other side's responsibility to reconnect them if
259 * necessary. ofproto does not send them asynchronous messages by default.
261 * Currently, active (tcp, ssl, unix) connections are always "primary"
262 * connections and passive (ptcp, pssl, punix) connections are always "service"
263 * connections. There is no inherent reason for this, but it reflects the
267 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
268 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
271 /* A listener for incoming OpenFlow "service" connections. */
273 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
274 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
276 /* These are not used by ofservice directly. They are settings for
277 * accepted "struct ofconn"s from the pvconn. */
278 int probe_interval
; /* Max idle time before probing, in seconds. */
279 int rate_limit
; /* Max packet-in rate in packets per second. */
280 int burst_limit
; /* Limit on accumulating packet credits. */
283 static struct ofservice
*ofservice_lookup(struct ofproto
*,
285 static int ofservice_create(struct ofproto
*,
286 const struct ofproto_controller
*);
287 static void ofservice_reconfigure(struct ofservice
*,
288 const struct ofproto_controller
*);
289 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
291 /* An OpenFlow connection. */
293 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
294 struct list node
; /* In struct ofproto's "all_conns" list. */
295 struct rconn
*rconn
; /* OpenFlow connection. */
296 enum ofconn_type type
; /* Type. */
297 enum nx_flow_format flow_format
; /* Currently selected flow format. */
299 /* OFPT_PACKET_IN related data. */
300 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
301 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
302 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
303 int miss_send_len
; /* Bytes to send of buffered packets. */
305 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
306 * requests, and the maximum number before we stop reading OpenFlow
308 #define OFCONN_REPLY_MAX 100
309 struct rconn_packet_counter
*reply_counter
;
311 /* type == OFCONN_PRIMARY only. */
312 enum nx_role role
; /* Role. */
313 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
314 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
315 struct status_category
*ss
; /* Switch status category. */
316 enum ofproto_band band
; /* In-band or out-of-band? */
319 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
320 * "schedulers" array. Their values are 0 and 1, and their meanings and values
321 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
322 * case anything ever changes, check their values here. */
323 #define N_SCHEDULERS 2
324 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
325 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
326 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
327 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
329 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
331 static void ofconn_destroy(struct ofconn
*);
332 static void ofconn_run(struct ofconn
*);
333 static void ofconn_wait(struct ofconn
*);
334 static bool ofconn_receives_async_msgs(const struct ofconn
*);
335 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
336 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
338 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
339 struct rconn_packet_counter
*counter
);
341 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
342 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
346 uint64_t datapath_id
; /* Datapath ID. */
347 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
348 char *mfr_desc
; /* Manufacturer. */
349 char *hw_desc
; /* Hardware. */
350 char *sw_desc
; /* Software version. */
351 char *serial_desc
; /* Serial number. */
352 char *dp_desc
; /* Datapath description. */
356 struct netdev_monitor
*netdev_monitor
;
357 struct hmap ports
; /* Contains "struct ofport"s. */
358 struct shash port_by_name
;
362 struct switch_status
*switch_status
;
363 struct fail_open
*fail_open
;
364 struct netflow
*netflow
;
365 struct ofproto_sflow
*sflow
;
367 /* In-band control. */
368 struct in_band
*in_band
;
369 long long int next_in_band_update
;
370 struct sockaddr_in
*extra_in_band_remotes
;
371 size_t n_extra_remotes
;
375 struct classifier cls
;
376 long long int next_expiration
;
380 bool need_revalidate
;
381 struct tag_set revalidate_set
;
383 /* OpenFlow connections. */
384 struct hmap controllers
; /* Controller "struct ofconn"s. */
385 struct list all_conns
; /* Contains "struct ofconn"s. */
386 enum ofproto_fail_mode fail_mode
;
388 /* OpenFlow listeners. */
389 struct hmap services
; /* Contains "struct ofservice"s. */
390 struct pvconn
**snoops
;
393 /* Hooks for ovs-vswitchd. */
394 const struct ofhooks
*ofhooks
;
397 /* Used by default ofhooks. */
398 struct mac_learning
*ml
;
401 /* Map from dpif name to struct ofproto, for use by unixctl commands. */
402 static struct shash all_ofprotos
= SHASH_INITIALIZER(&all_ofprotos
);
404 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
406 static const struct ofhooks default_ofhooks
;
408 static uint64_t pick_datapath_id(const struct ofproto
*);
409 static uint64_t pick_fallback_dpid(void);
411 static int ofproto_expire(struct ofproto
*);
413 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
415 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
417 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
418 static void update_port(struct ofproto
*, const char *devname
);
419 static int init_ports(struct ofproto
*);
420 static void reinit_ports(struct ofproto
*);
422 static void ofproto_unixctl_init(void);
425 ofproto_create(const char *datapath
, const char *datapath_type
,
426 const struct ofhooks
*ofhooks
, void *aux
,
427 struct ofproto
**ofprotop
)
429 struct odp_stats stats
;
436 ofproto_unixctl_init();
438 /* Connect to datapath and start listening for messages. */
439 error
= dpif_open(datapath
, datapath_type
, &dpif
);
441 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
444 error
= dpif_get_dp_stats(dpif
, &stats
);
446 VLOG_ERR("failed to obtain stats for datapath %s: %s",
447 datapath
, strerror(error
));
451 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
453 VLOG_ERR("failed to listen on datapath %s: %s",
454 datapath
, strerror(error
));
458 dpif_flow_flush(dpif
);
459 dpif_recv_purge(dpif
);
461 /* Initialize settings. */
462 p
= xzalloc(sizeof *p
);
463 p
->fallback_dpid
= pick_fallback_dpid();
464 p
->datapath_id
= p
->fallback_dpid
;
465 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
466 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
467 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
468 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
469 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
471 /* Initialize datapath. */
473 p
->netdev_monitor
= netdev_monitor_create();
474 hmap_init(&p
->ports
);
475 shash_init(&p
->port_by_name
);
476 p
->max_ports
= stats
.max_ports
;
478 /* Initialize submodules. */
479 p
->switch_status
= switch_status_create(p
);
484 /* Initialize in-band control. */
486 p
->in_band_queue
= -1;
488 /* Initialize flow table. */
489 classifier_init(&p
->cls
);
490 p
->next_expiration
= time_msec() + 1000;
492 /* Initialize facet table. */
493 hmap_init(&p
->facets
);
494 p
->need_revalidate
= false;
495 tag_set_init(&p
->revalidate_set
);
497 /* Initialize OpenFlow connections. */
498 list_init(&p
->all_conns
);
499 hmap_init(&p
->controllers
);
500 hmap_init(&p
->services
);
504 /* Initialize hooks. */
506 p
->ofhooks
= ofhooks
;
510 p
->ofhooks
= &default_ofhooks
;
512 p
->ml
= mac_learning_create();
515 /* Pick final datapath ID. */
516 p
->datapath_id
= pick_datapath_id(p
);
517 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
519 shash_add_once(&all_ofprotos
, dpif_name(p
->dpif
), p
);
526 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
528 uint64_t old_dpid
= p
->datapath_id
;
529 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
530 if (p
->datapath_id
!= old_dpid
) {
531 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
533 /* Force all active connections to reconnect, since there is no way to
534 * notify a controller that the datapath ID has changed. */
535 ofproto_reconnect_controllers(p
);
540 is_discovery_controller(const struct ofproto_controller
*c
)
542 return !strcmp(c
->target
, "discover");
546 is_in_band_controller(const struct ofproto_controller
*c
)
548 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
551 /* Creates a new controller in 'ofproto'. Some of the settings are initially
552 * drawn from 'c', but update_controller() needs to be called later to finish
553 * the new ofconn's configuration. */
555 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
557 struct discovery
*discovery
;
558 struct ofconn
*ofconn
;
560 if (is_discovery_controller(c
)) {
561 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
562 ofproto
->dpif
, ofproto
->switch_status
,
571 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
572 ofconn
->pktbuf
= pktbuf_create();
573 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
575 ofconn
->discovery
= discovery
;
577 char *name
= ofconn_make_name(ofproto
, c
->target
);
578 rconn_connect(ofconn
->rconn
, c
->target
, name
);
581 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
582 hash_string(c
->target
, 0));
585 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
586 * target or turn discovery on or off (these are done by creating new ofconns
587 * and deleting old ones), but it can update the rest of an ofconn's
590 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
594 ofconn
->band
= (is_in_band_controller(c
)
595 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
597 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
599 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
600 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
602 if (ofconn
->discovery
) {
603 discovery_set_update_resolv_conf(ofconn
->discovery
,
604 c
->update_resolv_conf
);
605 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
608 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
612 ofconn_get_target(const struct ofconn
*ofconn
)
614 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
617 static struct ofconn
*
618 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
620 struct ofconn
*ofconn
;
622 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
623 hash_string(target
, 0), &ofproto
->controllers
) {
624 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
632 update_in_band_remotes(struct ofproto
*ofproto
)
634 const struct ofconn
*ofconn
;
635 struct sockaddr_in
*addrs
;
636 size_t max_addrs
, n_addrs
;
640 /* Allocate enough memory for as many remotes as we could possibly have. */
641 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
642 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
645 /* Add all the remotes. */
647 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
648 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
650 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
654 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
655 if (sin
->sin_addr
.s_addr
) {
656 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
659 if (ofconn
->discovery
) {
663 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
664 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
667 /* Create or update or destroy in-band.
669 * Ordinarily we only enable in-band if there's at least one remote
670 * address, but discovery needs the in-band rules for DHCP to be installed
671 * even before we know any remote addresses. */
672 if (n_addrs
|| discovery
) {
673 if (!ofproto
->in_band
) {
674 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
677 if (ofproto
->in_band
) {
678 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
680 in_band_set_queue(ofproto
->in_band
, ofproto
->in_band_queue
);
681 ofproto
->next_in_band_update
= time_msec() + 1000;
683 in_band_destroy(ofproto
->in_band
);
684 ofproto
->in_band
= NULL
;
692 update_fail_open(struct ofproto
*p
)
694 struct ofconn
*ofconn
;
696 if (!hmap_is_empty(&p
->controllers
)
697 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
698 struct rconn
**rconns
;
702 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
706 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
707 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
708 rconns
[n
++] = ofconn
->rconn
;
711 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
712 /* p->fail_open takes ownership of 'rconns'. */
714 fail_open_destroy(p
->fail_open
);
720 ofproto_set_controllers(struct ofproto
*p
,
721 const struct ofproto_controller
*controllers
,
722 size_t n_controllers
)
724 struct shash new_controllers
;
725 struct ofconn
*ofconn
, *next_ofconn
;
726 struct ofservice
*ofservice
, *next_ofservice
;
730 /* Create newly configured controllers and services.
731 * Create a name to ofproto_controller mapping in 'new_controllers'. */
732 shash_init(&new_controllers
);
733 for (i
= 0; i
< n_controllers
; i
++) {
734 const struct ofproto_controller
*c
= &controllers
[i
];
736 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
737 if (!find_controller_by_target(p
, c
->target
)) {
738 add_controller(p
, c
);
740 } else if (!pvconn_verify_name(c
->target
)) {
741 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
745 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
746 dpif_name(p
->dpif
), c
->target
);
750 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
753 /* Delete controllers that are no longer configured.
754 * Update configuration of all now-existing controllers. */
756 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
757 struct ofproto_controller
*c
;
759 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
761 ofconn_destroy(ofconn
);
763 update_controller(ofconn
, c
);
770 /* Delete services that are no longer configured.
771 * Update configuration of all now-existing services. */
772 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
773 struct ofproto_controller
*c
;
775 c
= shash_find_data(&new_controllers
,
776 pvconn_get_name(ofservice
->pvconn
));
778 ofservice_destroy(p
, ofservice
);
780 ofservice_reconfigure(ofservice
, c
);
784 shash_destroy(&new_controllers
);
786 update_in_band_remotes(p
);
789 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
790 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
791 struct ofconn
, hmap_node
);
792 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
793 rconn_status_cb
, ofconn
->rconn
);
798 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
800 p
->fail_mode
= fail_mode
;
804 /* Drops the connections between 'ofproto' and all of its controllers, forcing
805 * them to reconnect. */
807 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
809 struct ofconn
*ofconn
;
811 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
812 rconn_reconnect(ofconn
->rconn
);
817 any_extras_changed(const struct ofproto
*ofproto
,
818 const struct sockaddr_in
*extras
, size_t n
)
822 if (n
!= ofproto
->n_extra_remotes
) {
826 for (i
= 0; i
< n
; i
++) {
827 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
828 const struct sockaddr_in
*new = &extras
[i
];
830 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
831 old
->sin_port
!= new->sin_port
) {
839 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
840 * in-band control should guarantee access, in the same way that in-band
841 * control guarantees access to OpenFlow controllers. */
843 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
844 const struct sockaddr_in
*extras
, size_t n
)
846 if (!any_extras_changed(ofproto
, extras
, n
)) {
850 free(ofproto
->extra_in_band_remotes
);
851 ofproto
->n_extra_remotes
= n
;
852 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
854 update_in_band_remotes(ofproto
);
857 /* Sets the OpenFlow queue used by flows set up by in-band control on
858 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
859 * flows will use the default queue. */
861 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
863 if (queue_id
!= ofproto
->in_band_queue
) {
864 ofproto
->in_band_queue
= queue_id
;
865 update_in_band_remotes(ofproto
);
870 ofproto_set_desc(struct ofproto
*p
,
871 const char *mfr_desc
, const char *hw_desc
,
872 const char *sw_desc
, const char *serial_desc
,
875 struct ofp_desc_stats
*ods
;
878 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
879 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
880 sizeof ods
->mfr_desc
);
883 p
->mfr_desc
= xstrdup(mfr_desc
);
886 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
887 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
888 sizeof ods
->hw_desc
);
891 p
->hw_desc
= xstrdup(hw_desc
);
894 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
895 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
896 sizeof ods
->sw_desc
);
899 p
->sw_desc
= xstrdup(sw_desc
);
902 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
903 VLOG_WARN("truncating serial_desc, must be less than %zu "
905 sizeof ods
->serial_num
);
907 free(p
->serial_desc
);
908 p
->serial_desc
= xstrdup(serial_desc
);
911 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
912 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
913 sizeof ods
->dp_desc
);
916 p
->dp_desc
= xstrdup(dp_desc
);
921 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
922 const struct svec
*svec
)
924 struct pvconn
**pvconns
= *pvconnsp
;
925 size_t n_pvconns
= *n_pvconnsp
;
929 for (i
= 0; i
< n_pvconns
; i
++) {
930 pvconn_close(pvconns
[i
]);
934 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
936 for (i
= 0; i
< svec
->n
; i
++) {
937 const char *name
= svec
->names
[i
];
938 struct pvconn
*pvconn
;
941 error
= pvconn_open(name
, &pvconn
);
943 pvconns
[n_pvconns
++] = pvconn
;
945 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
953 *n_pvconnsp
= n_pvconns
;
959 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
961 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
965 ofproto_set_netflow(struct ofproto
*ofproto
,
966 const struct netflow_options
*nf_options
)
968 if (nf_options
&& nf_options
->collectors
.n
) {
969 if (!ofproto
->netflow
) {
970 ofproto
->netflow
= netflow_create();
972 return netflow_set_options(ofproto
->netflow
, nf_options
);
974 netflow_destroy(ofproto
->netflow
);
975 ofproto
->netflow
= NULL
;
981 ofproto_set_sflow(struct ofproto
*ofproto
,
982 const struct ofproto_sflow_options
*oso
)
984 struct ofproto_sflow
*os
= ofproto
->sflow
;
987 struct ofport
*ofport
;
989 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
990 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
991 ofproto_sflow_add_port(os
, ofport
->odp_port
,
992 netdev_get_name(ofport
->netdev
));
995 ofproto_sflow_set_options(os
, oso
);
997 ofproto_sflow_destroy(os
);
998 ofproto
->sflow
= NULL
;
1003 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
1005 return ofproto
->datapath_id
;
1009 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
1011 return !hmap_is_empty(&ofproto
->controllers
);
1014 enum ofproto_fail_mode
1015 ofproto_get_fail_mode(const struct ofproto
*p
)
1017 return p
->fail_mode
;
1021 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
1025 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
1026 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
1031 ofproto_destroy(struct ofproto
*p
)
1033 struct ofservice
*ofservice
, *next_ofservice
;
1034 struct ofconn
*ofconn
, *next_ofconn
;
1035 struct ofport
*ofport
, *next_ofport
;
1042 shash_find_and_delete(&all_ofprotos
, dpif_name(p
->dpif
));
1044 /* Destroy fail-open and in-band early, since they touch the classifier. */
1045 fail_open_destroy(p
->fail_open
);
1046 p
->fail_open
= NULL
;
1048 in_band_destroy(p
->in_band
);
1050 free(p
->extra_in_band_remotes
);
1052 ofproto_flush_flows(p
);
1053 classifier_destroy(&p
->cls
);
1054 hmap_destroy(&p
->facets
);
1056 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1057 ofconn_destroy(ofconn
);
1059 hmap_destroy(&p
->controllers
);
1061 dpif_close(p
->dpif
);
1062 netdev_monitor_destroy(p
->netdev_monitor
);
1063 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1064 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1065 ofport_free(ofport
);
1067 shash_destroy(&p
->port_by_name
);
1069 switch_status_destroy(p
->switch_status
);
1070 netflow_destroy(p
->netflow
);
1071 ofproto_sflow_destroy(p
->sflow
);
1073 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
1074 ofservice_destroy(p
, ofservice
);
1076 hmap_destroy(&p
->services
);
1078 for (i
= 0; i
< p
->n_snoops
; i
++) {
1079 pvconn_close(p
->snoops
[i
]);
1083 mac_learning_destroy(p
->ml
);
1088 free(p
->serial_desc
);
1091 hmap_destroy(&p
->ports
);
1097 ofproto_run(struct ofproto
*p
)
1099 int error
= ofproto_run1(p
);
1101 error
= ofproto_run2(p
, false);
1107 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1109 if (error
== ENOBUFS
) {
1110 reinit_ports(ofproto
);
1111 } else if (!error
) {
1112 update_port(ofproto
, devname
);
1117 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1118 * means that 'ofconn' is more interesting for monitoring than a lower return
1121 snoop_preference(const struct ofconn
*ofconn
)
1123 switch (ofconn
->role
) {
1124 case NX_ROLE_MASTER
:
1131 /* Shouldn't happen. */
1136 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1137 * Connects this vconn to a controller. */
1139 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1141 struct ofconn
*ofconn
, *best
;
1143 /* Pick a controller for monitoring. */
1145 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1146 if (ofconn
->type
== OFCONN_PRIMARY
1147 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1153 rconn_add_monitor(best
->rconn
, vconn
);
1155 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1161 ofproto_run1(struct ofproto
*p
)
1163 struct ofconn
*ofconn
, *next_ofconn
;
1164 struct ofservice
*ofservice
;
1169 if (shash_is_empty(&p
->port_by_name
)) {
1173 for (i
= 0; i
< 50; i
++) {
1176 error
= dpif_recv(p
->dpif
, &buf
);
1178 if (error
== ENODEV
) {
1179 /* Someone destroyed the datapath behind our back. The caller
1180 * better destroy us and give up, because we're just going to
1181 * spin from here on out. */
1182 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1183 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1184 dpif_name(p
->dpif
));
1190 handle_odp_msg(p
, buf
);
1193 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1194 process_port_change(p
, error
, devname
);
1196 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1197 &devname
)) != EAGAIN
) {
1198 process_port_change(p
, error
, devname
);
1202 if (time_msec() >= p
->next_in_band_update
) {
1203 update_in_band_remotes(p
);
1205 in_band_run(p
->in_band
);
1208 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1212 /* Fail-open maintenance. Do this after processing the ofconns since
1213 * fail-open checks the status of the controller rconn. */
1215 fail_open_run(p
->fail_open
);
1218 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1219 struct vconn
*vconn
;
1222 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1224 struct rconn
*rconn
;
1227 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1228 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1229 rconn_connect_unreliably(rconn
, vconn
, name
);
1232 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1233 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1234 ofservice
->burst_limit
);
1235 } else if (retval
!= EAGAIN
) {
1236 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1240 for (i
= 0; i
< p
->n_snoops
; i
++) {
1241 struct vconn
*vconn
;
1244 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1246 add_snooper(p
, vconn
);
1247 } else if (retval
!= EAGAIN
) {
1248 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1252 if (time_msec() >= p
->next_expiration
) {
1253 int delay
= ofproto_expire(p
);
1254 p
->next_expiration
= time_msec() + delay
;
1255 COVERAGE_INC(ofproto_expiration
);
1259 netflow_run(p
->netflow
);
1262 ofproto_sflow_run(p
->sflow
);
1269 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1271 /* Figure out what we need to revalidate now, if anything. */
1272 struct tag_set revalidate_set
= p
->revalidate_set
;
1273 if (p
->need_revalidate
) {
1274 revalidate_all
= true;
1277 /* Clear the revalidation flags. */
1278 tag_set_init(&p
->revalidate_set
);
1279 p
->need_revalidate
= false;
1281 /* Now revalidate if there's anything to do. */
1282 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1283 struct facet
*facet
, *next
;
1285 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1287 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1288 facet_revalidate(p
, facet
);
1297 ofproto_wait(struct ofproto
*p
)
1299 struct ofservice
*ofservice
;
1300 struct ofconn
*ofconn
;
1303 dpif_recv_wait(p
->dpif
);
1304 dpif_port_poll_wait(p
->dpif
);
1305 netdev_monitor_poll_wait(p
->netdev_monitor
);
1306 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1307 ofconn_wait(ofconn
);
1310 poll_timer_wait_until(p
->next_in_band_update
);
1311 in_band_wait(p
->in_band
);
1314 fail_open_wait(p
->fail_open
);
1317 ofproto_sflow_wait(p
->sflow
);
1319 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1320 poll_immediate_wake();
1322 if (p
->need_revalidate
) {
1323 /* Shouldn't happen, but if it does just go around again. */
1324 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1325 poll_immediate_wake();
1326 } else if (p
->next_expiration
!= LLONG_MAX
) {
1327 poll_timer_wait_until(p
->next_expiration
);
1329 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1330 pvconn_wait(ofservice
->pvconn
);
1332 for (i
= 0; i
< p
->n_snoops
; i
++) {
1333 pvconn_wait(p
->snoops
[i
]);
1338 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1340 tag_set_add(&ofproto
->revalidate_set
, tag
);
1344 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1346 return &ofproto
->revalidate_set
;
1350 ofproto_is_alive(const struct ofproto
*p
)
1352 return !hmap_is_empty(&p
->controllers
);
1355 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1357 * This is almost the same as calling dpif_port_del() directly on the
1358 * datapath, but it also makes 'ofproto' close its open netdev for the port
1359 * (if any). This makes it possible to create a new netdev of a different
1360 * type under the same name, which otherwise the netdev library would refuse
1361 * to do because of the conflict. (The netdev would eventually get closed on
1362 * the next trip through ofproto_run(), but this interface is more direct.)
1364 * Returns 0 if successful, otherwise a positive errno. */
1366 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1368 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1369 const char *name
= ofport
? ofport
->opp
.name
: "<unknown>";
1372 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1374 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1375 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1376 } else if (ofport
) {
1377 /* 'name' is ofport->opp.name and update_port() is going to destroy
1378 * 'ofport'. Just in case update_port() refers to 'name' after it
1379 * destroys 'ofport', make a copy of it around the update_port()
1381 char *devname
= xstrdup(name
);
1382 update_port(ofproto
, devname
);
1388 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1389 * true if 'odp_port' exists and should be included, false otherwise. */
1391 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1393 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1394 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1398 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1399 const union ofp_action
*actions
, size_t n_actions
,
1400 const struct ofpbuf
*packet
)
1402 struct action_xlate_ctx ctx
;
1405 action_xlate_ctx_init(&ctx
, p
, flow
, packet
);
1406 error
= xlate_actions(&ctx
, actions
, n_actions
);
1411 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1413 dpif_execute(p
->dpif
, ctx
.out
.actions
, ctx
.out
.n_actions
, packet
);
1417 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1418 * performs the 'n_actions' actions in 'actions'. The new flow will not
1421 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1422 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1423 * controllers; otherwise, it will be hidden.
1425 * The caller retains ownership of 'cls_rule' and 'actions'. */
1427 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1428 const union ofp_action
*actions
, size_t n_actions
)
1431 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1432 rule_insert(p
, rule
);
1436 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1440 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1443 rule_remove(ofproto
, rule
);
1448 ofproto_flush_flows(struct ofproto
*ofproto
)
1450 struct facet
*facet
, *next_facet
;
1451 struct rule
*rule
, *next_rule
;
1452 struct cls_cursor cursor
;
1454 COVERAGE_INC(ofproto_flush
);
1456 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1457 /* Mark the facet as not installed so that facet_remove() doesn't
1458 * bother trying to uninstall it. There is no point in uninstalling it
1459 * individually since we are about to blow away all the facets with
1460 * dpif_flow_flush(). */
1461 facet
->installed
= false;
1462 facet_remove(ofproto
, facet
);
1465 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1466 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1467 rule_remove(ofproto
, rule
);
1470 dpif_flow_flush(ofproto
->dpif
);
1471 if (ofproto
->in_band
) {
1472 in_band_flushed(ofproto
->in_band
);
1474 if (ofproto
->fail_open
) {
1475 fail_open_flushed(ofproto
->fail_open
);
1480 reinit_ports(struct ofproto
*p
)
1482 struct svec devnames
;
1483 struct ofport
*ofport
;
1484 struct odp_port
*odp_ports
;
1488 COVERAGE_INC(ofproto_reinit_ports
);
1490 svec_init(&devnames
);
1491 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1492 svec_add (&devnames
, ofport
->opp
.name
);
1494 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1495 for (i
= 0; i
< n_odp_ports
; i
++) {
1496 svec_add (&devnames
, odp_ports
[i
].devname
);
1500 svec_sort_unique(&devnames
);
1501 for (i
= 0; i
< devnames
.n
; i
++) {
1502 update_port(p
, devnames
.names
[i
]);
1504 svec_destroy(&devnames
);
1507 static struct ofport
*
1508 make_ofport(const struct odp_port
*odp_port
)
1510 struct netdev_options netdev_options
;
1511 enum netdev_flags flags
;
1512 struct ofport
*ofport
;
1513 struct netdev
*netdev
;
1516 memset(&netdev_options
, 0, sizeof netdev_options
);
1517 netdev_options
.name
= odp_port
->devname
;
1518 netdev_options
.type
= odp_port
->type
;
1519 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1521 error
= netdev_open(&netdev_options
, &netdev
);
1523 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1524 "cannot be opened (%s)",
1525 odp_port
->devname
, odp_port
->port
,
1526 odp_port
->devname
, strerror(error
));
1530 ofport
= xmalloc(sizeof *ofport
);
1531 ofport
->netdev
= netdev
;
1532 ofport
->odp_port
= odp_port
->port
;
1533 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1534 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1535 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1536 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1537 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1539 netdev_get_flags(netdev
, &flags
);
1540 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1542 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1544 netdev_get_features(netdev
,
1545 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1546 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1551 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1553 if (get_port(p
, odp_port
->port
)) {
1554 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1557 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1558 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1567 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1569 const struct ofp_phy_port
*a
= &a_
->opp
;
1570 const struct ofp_phy_port
*b
= &b_
->opp
;
1572 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1573 return (a
->port_no
== b
->port_no
1574 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1575 && !strcmp(a
->name
, b
->name
)
1576 && a
->state
== b
->state
1577 && a
->config
== b
->config
1578 && a
->curr
== b
->curr
1579 && a
->advertised
== b
->advertised
1580 && a
->supported
== b
->supported
1581 && a
->peer
== b
->peer
);
1585 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1588 /* XXX Should limit the number of queued port status change messages. */
1589 struct ofconn
*ofconn
;
1590 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1591 struct ofp_port_status
*ops
;
1594 /* Primary controllers, even slaves, should always get port status
1595 updates. Otherwise obey ofconn_receives_async_msgs(). */
1596 if (ofconn
->type
!= OFCONN_PRIMARY
1597 && !ofconn_receives_async_msgs(ofconn
)) {
1601 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1602 ops
->reason
= reason
;
1603 ops
->desc
= ofport
->opp
;
1604 hton_ofp_phy_port(&ops
->desc
);
1605 queue_tx(b
, ofconn
, NULL
);
1610 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1612 const char *netdev_name
= ofport
->opp
.name
;
1614 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1615 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1616 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1618 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1623 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1625 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1626 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1627 shash_delete(&p
->port_by_name
,
1628 shash_find(&p
->port_by_name
, ofport
->opp
.name
));
1630 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1635 ofport_free(struct ofport
*ofport
)
1638 netdev_close(ofport
->netdev
);
1643 static struct ofport
*
1644 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1646 struct ofport
*port
;
1648 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1649 hash_int(odp_port
, 0), &ofproto
->ports
) {
1650 if (port
->odp_port
== odp_port
) {
1658 update_port(struct ofproto
*p
, const char *devname
)
1660 struct odp_port odp_port
;
1661 struct ofport
*old_ofport
;
1662 struct ofport
*new_ofport
;
1665 COVERAGE_INC(ofproto_update_port
);
1667 /* Query the datapath for port information. */
1668 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1670 /* Find the old ofport. */
1671 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1674 /* There's no port named 'devname' but there might be a port with
1675 * the same port number. This could happen if a port is deleted
1676 * and then a new one added in its place very quickly, or if a port
1677 * is renamed. In the former case we want to send an OFPPR_DELETE
1678 * and an OFPPR_ADD, and in the latter case we want to send a
1679 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1680 * the old port's ifindex against the new port, or perhaps less
1681 * reliably but more portably by comparing the old port's MAC
1682 * against the new port's MAC. However, this code isn't that smart
1683 * and always sends an OFPPR_MODIFY (XXX). */
1684 old_ofport
= get_port(p
, odp_port
.port
);
1686 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1687 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1688 "%s", strerror(error
));
1692 /* Create a new ofport. */
1693 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1695 /* Eliminate a few pathological cases. */
1696 if (!old_ofport
&& !new_ofport
) {
1698 } else if (old_ofport
&& new_ofport
) {
1699 /* Most of the 'config' bits are OpenFlow soft state, but
1700 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1701 * OpenFlow bits from old_ofport. (make_ofport() only sets
1702 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1703 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1705 if (ofport_equal(old_ofport
, new_ofport
)) {
1706 /* False alarm--no change. */
1707 ofport_free(new_ofport
);
1712 /* Now deal with the normal cases. */
1714 ofport_remove(p
, old_ofport
);
1717 ofport_install(p
, new_ofport
);
1719 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1720 (!old_ofport
? OFPPR_ADD
1721 : !new_ofport
? OFPPR_DELETE
1723 ofport_free(old_ofport
);
1727 init_ports(struct ofproto
*p
)
1729 struct odp_port
*ports
;
1734 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1739 for (i
= 0; i
< n_ports
; i
++) {
1740 const struct odp_port
*odp_port
= &ports
[i
];
1741 if (!ofport_conflicts(p
, odp_port
)) {
1742 struct ofport
*ofport
= make_ofport(odp_port
);
1744 ofport_install(p
, ofport
);
1752 static struct ofconn
*
1753 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1755 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1756 ofconn
->ofproto
= p
;
1757 list_push_back(&p
->all_conns
, &ofconn
->node
);
1758 ofconn
->rconn
= rconn
;
1759 ofconn
->type
= type
;
1760 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1761 ofconn
->role
= NX_ROLE_OTHER
;
1762 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1763 ofconn
->pktbuf
= NULL
;
1764 ofconn
->miss_send_len
= 0;
1765 ofconn
->reply_counter
= rconn_packet_counter_create ();
1770 ofconn_destroy(struct ofconn
*ofconn
)
1772 if (ofconn
->type
== OFCONN_PRIMARY
) {
1773 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1775 discovery_destroy(ofconn
->discovery
);
1777 list_remove(&ofconn
->node
);
1778 switch_status_unregister(ofconn
->ss
);
1779 rconn_destroy(ofconn
->rconn
);
1780 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1781 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1782 pktbuf_destroy(ofconn
->pktbuf
);
1787 ofconn_run(struct ofconn
*ofconn
)
1789 struct ofproto
*p
= ofconn
->ofproto
;
1793 if (ofconn
->discovery
) {
1794 char *controller_name
;
1795 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1796 discovery_question_connectivity(ofconn
->discovery
);
1798 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1799 if (controller_name
) {
1800 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1801 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1804 rconn_disconnect(ofconn
->rconn
);
1809 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1810 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1813 rconn_run(ofconn
->rconn
);
1815 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1816 /* Limit the number of iterations to prevent other tasks from
1818 for (iteration
= 0; iteration
< 50; iteration
++) {
1819 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1824 fail_open_maybe_recover(p
->fail_open
);
1826 handle_openflow(ofconn
, of_msg
);
1827 ofpbuf_delete(of_msg
);
1831 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1832 ofconn_destroy(ofconn
);
1837 ofconn_wait(struct ofconn
*ofconn
)
1841 if (ofconn
->discovery
) {
1842 discovery_wait(ofconn
->discovery
);
1844 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1845 pinsched_wait(ofconn
->schedulers
[i
]);
1847 rconn_run_wait(ofconn
->rconn
);
1848 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1849 rconn_recv_wait(ofconn
->rconn
);
1851 COVERAGE_INC(ofproto_ofconn_stuck
);
1855 /* Returns true if 'ofconn' should receive asynchronous messages. */
1857 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1859 if (ofconn
->type
== OFCONN_PRIMARY
) {
1860 /* Primary controllers always get asynchronous messages unless they
1861 * have configured themselves as "slaves". */
1862 return ofconn
->role
!= NX_ROLE_SLAVE
;
1864 /* Service connections don't get asynchronous messages unless they have
1865 * explicitly asked for them by setting a nonzero miss send length. */
1866 return ofconn
->miss_send_len
> 0;
1870 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1871 * and 'target', suitable for use in log messages for identifying the
1874 * The name is dynamically allocated. The caller should free it (with free())
1875 * when it is no longer needed. */
1877 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1879 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1883 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1887 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1888 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1892 *s
= pinsched_create(rate
, burst
,
1893 ofconn
->ofproto
->switch_status
);
1895 pinsched_set_limits(*s
, rate
, burst
);
1898 pinsched_destroy(*s
);
1905 ofservice_reconfigure(struct ofservice
*ofservice
,
1906 const struct ofproto_controller
*c
)
1908 ofservice
->probe_interval
= c
->probe_interval
;
1909 ofservice
->rate_limit
= c
->rate_limit
;
1910 ofservice
->burst_limit
= c
->burst_limit
;
1913 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1914 * positive errno value. */
1916 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1918 struct ofservice
*ofservice
;
1919 struct pvconn
*pvconn
;
1922 error
= pvconn_open(c
->target
, &pvconn
);
1927 ofservice
= xzalloc(sizeof *ofservice
);
1928 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1929 hash_string(c
->target
, 0));
1930 ofservice
->pvconn
= pvconn
;
1932 ofservice_reconfigure(ofservice
, c
);
1938 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1940 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1941 pvconn_close(ofservice
->pvconn
);
1945 /* Finds and returns the ofservice within 'ofproto' that has the given
1946 * 'target', or a null pointer if none exists. */
1947 static struct ofservice
*
1948 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1950 struct ofservice
*ofservice
;
1952 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1953 &ofproto
->services
) {
1954 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1961 /* Returns true if 'rule' should be hidden from the controller.
1963 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1964 * (e.g. by in-band control) and are intentionally hidden from the
1967 rule_is_hidden(const struct rule
*rule
)
1969 return rule
->cr
.priority
> UINT16_MAX
;
1972 /* Creates and returns a new rule initialized as specified.
1974 * The caller is responsible for inserting the rule into the classifier (with
1975 * rule_insert()). */
1976 static struct rule
*
1977 rule_create(const struct cls_rule
*cls_rule
,
1978 const union ofp_action
*actions
, size_t n_actions
,
1979 uint16_t idle_timeout
, uint16_t hard_timeout
,
1980 ovs_be64 flow_cookie
, bool send_flow_removed
)
1982 struct rule
*rule
= xzalloc(sizeof *rule
);
1983 rule
->cr
= *cls_rule
;
1984 rule
->idle_timeout
= idle_timeout
;
1985 rule
->hard_timeout
= hard_timeout
;
1986 rule
->flow_cookie
= flow_cookie
;
1987 rule
->used
= rule
->created
= time_msec();
1988 rule
->send_flow_removed
= send_flow_removed
;
1989 list_init(&rule
->facets
);
1990 if (n_actions
> 0) {
1991 rule
->n_actions
= n_actions
;
1992 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1998 static struct rule
*
1999 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
2001 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
2005 rule_free(struct rule
*rule
)
2007 free(rule
->actions
);
2011 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
2012 * destroying any that no longer has a rule (which is probably all of them).
2014 * The caller must have already removed 'rule' from the classifier. */
2016 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
2018 struct facet
*facet
, *next_facet
;
2019 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
2020 facet_revalidate(ofproto
, facet
);
2025 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2026 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
2029 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
2031 const union ofp_action
*oa
;
2032 struct actions_iterator i
;
2034 if (out_port
== htons(OFPP_NONE
)) {
2037 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
2038 oa
= actions_next(&i
)) {
2039 if (action_outputs_to_port(oa
, out_port
)) {
2046 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
2047 * 'packet', which arrived on 'in_port'.
2049 * Takes ownership of 'packet'. */
2051 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
2052 const union odp_action
*actions
, size_t n_actions
,
2053 struct ofpbuf
*packet
)
2055 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
2056 /* As an optimization, avoid a round-trip from userspace to kernel to
2057 * userspace. This also avoids possibly filling up kernel packet
2058 * buffers along the way. */
2059 struct odp_msg
*msg
;
2061 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
2062 msg
->type
= _ODPL_ACTION_NR
;
2063 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
2064 msg
->port
= in_port
;
2066 msg
->arg
= actions
[0].controller
.arg
;
2068 send_packet_in(ofproto
, packet
);
2074 error
= dpif_execute(ofproto
->dpif
, actions
, n_actions
, packet
);
2075 ofpbuf_delete(packet
);
2080 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2081 * statistics appropriately. 'packet' must have at least sizeof(struct
2082 * ofp_packet_in) bytes of headroom.
2084 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2085 * applying flow_extract() to 'packet' would yield the same flow as
2088 * 'facet' must have accurately composed ODP actions; that is, it must not be
2089 * in need of revalidation.
2091 * Takes ownership of 'packet'. */
2093 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2094 struct ofpbuf
*packet
)
2096 struct odp_flow_stats stats
;
2098 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2100 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2101 if (execute_odp_actions(ofproto
, facet
->flow
.in_port
,
2102 facet
->actions
, facet
->n_actions
, packet
)) {
2103 facet_update_stats(ofproto
, facet
, &stats
);
2104 facet
->used
= time_msec();
2105 netflow_flow_update_time(ofproto
->netflow
,
2106 &facet
->nf_flow
, facet
->used
);
2110 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2111 * statistics (or the statistics for one of its facets) appropriately.
2112 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2114 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2115 * with statistics for 'packet' either way.
2117 * Takes ownership of 'packet'. */
2119 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2120 struct ofpbuf
*packet
)
2122 struct action_xlate_ctx ctx
;
2123 struct facet
*facet
;
2127 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2129 flow_extract(packet
, 0, in_port
, &flow
);
2131 /* First look for a related facet. If we find one, account it to that. */
2132 facet
= facet_lookup_valid(ofproto
, &flow
);
2133 if (facet
&& facet
->rule
== rule
) {
2134 facet_execute(ofproto
, facet
, packet
);
2138 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2139 * create a new facet for it and use that. */
2140 if (rule_lookup(ofproto
, &flow
) == rule
) {
2141 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2142 facet_execute(ofproto
, facet
, packet
);
2143 facet_install(ofproto
, facet
, true);
2147 /* We can't account anything to a facet. If we were to try, then that
2148 * facet would have a non-matching rule, busting our invariants. */
2149 action_xlate_ctx_init(&ctx
, ofproto
, &flow
, packet
);
2150 if (xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
)) {
2151 ofpbuf_delete(packet
);
2155 size
= packet
->size
;
2156 if (execute_odp_actions(ofproto
, in_port
,
2157 ctx
.out
.actions
, ctx
.out
.n_actions
, packet
)) {
2158 rule
->used
= time_msec();
2159 rule
->packet_count
++;
2160 rule
->byte_count
+= size
;
2164 /* Inserts 'rule' into 'p''s flow table. */
2166 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2168 struct rule
*displaced_rule
;
2170 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2171 if (displaced_rule
) {
2172 rule_destroy(p
, displaced_rule
);
2174 p
->need_revalidate
= true;
2177 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2178 * 'flow' and an example 'packet' within that flow.
2180 * The caller must already have determined that no facet with an identical
2181 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2182 * 'ofproto''s classifier table. */
2183 static struct facet
*
2184 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2185 const struct flow
*flow
, const struct ofpbuf
*packet
)
2187 struct facet
*facet
;
2189 facet
= xzalloc(sizeof *facet
);
2190 facet
->used
= time_msec();
2191 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2192 list_push_back(&rule
->facets
, &facet
->list_node
);
2194 facet
->flow
= *flow
;
2195 netflow_flow_init(&facet
->nf_flow
);
2196 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2198 facet_make_actions(ofproto
, facet
, packet
);
2204 facet_free(struct facet
*facet
)
2206 free(facet
->actions
);
2210 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2212 * - Removes 'rule' from the classifier.
2214 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2215 * destroys them), via rule_destroy().
2218 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2220 COVERAGE_INC(ofproto_del_rule
);
2221 ofproto
->need_revalidate
= true;
2222 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2223 rule_destroy(ofproto
, rule
);
2226 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2228 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2229 * rule's statistics, via facet_uninstall().
2231 * - Removes 'facet' from its rule and from ofproto->facets.
2234 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2236 facet_uninstall(ofproto
, facet
);
2237 facet_flush_stats(ofproto
, facet
);
2238 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2239 list_remove(&facet
->list_node
);
2243 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2245 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2246 const struct ofpbuf
*packet
)
2248 const struct rule
*rule
= facet
->rule
;
2249 struct action_xlate_ctx ctx
;
2252 action_xlate_ctx_init(&ctx
, p
, &facet
->flow
, packet
);
2253 xlate_actions(&ctx
, rule
->actions
, rule
->n_actions
);
2254 facet
->tags
= ctx
.tags
;
2255 facet
->may_install
= ctx
.may_set_up_flow
;
2256 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2258 actions_len
= ctx
.out
.n_actions
* sizeof *ctx
.out
.actions
;
2259 if (facet
->n_actions
!= ctx
.out
.n_actions
2260 || memcmp(facet
->actions
, ctx
.out
.actions
, actions_len
)) {
2261 free(facet
->actions
);
2262 facet
->n_actions
= ctx
.out
.n_actions
;
2263 facet
->actions
= xmemdup(ctx
.out
.actions
, actions_len
);
2268 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
, int flags
,
2269 struct odp_flow_put
*put
)
2271 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2272 odp_flow_key_from_flow(&put
->flow
.key
, &facet
->flow
);
2273 put
->flow
.actions
= facet
->actions
;
2274 put
->flow
.n_actions
= facet
->n_actions
;
2275 put
->flow
.flags
= 0;
2277 return dpif_flow_put(ofproto
->dpif
, put
);
2280 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2281 * 'zero_stats' is true, clears any existing statistics from the datapath for
2284 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2286 if (facet
->may_install
) {
2287 struct odp_flow_put put
;
2290 flags
= ODPPF_CREATE
| ODPPF_MODIFY
;
2292 flags
|= ODPPF_ZERO_STATS
;
2294 if (!facet_put__(p
, facet
, flags
, &put
)) {
2295 facet
->installed
= true;
2300 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2301 * to the accounting hook function in the ofhooks structure. */
2303 facet_account(struct ofproto
*ofproto
,
2304 struct facet
*facet
, uint64_t extra_bytes
)
2306 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2308 if (ofproto
->ofhooks
->account_flow_cb
2309 && total_bytes
> facet
->accounted_bytes
)
2311 ofproto
->ofhooks
->account_flow_cb(
2312 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->n_actions
,
2313 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2314 facet
->accounted_bytes
= total_bytes
;
2318 /* If 'rule' is installed in the datapath, uninstalls it. */
2320 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2322 if (facet
->installed
) {
2323 struct odp_flow odp_flow
;
2325 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
2326 odp_flow
.actions
= NULL
;
2327 odp_flow
.n_actions
= 0;
2329 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2330 facet_update_stats(p
, facet
, &odp_flow
.stats
);
2332 facet
->installed
= false;
2336 /* Returns true if the only action for 'facet' is to send to the controller.
2337 * (We don't report NetFlow expiration messages for such facets because they
2338 * are just part of the control logic for the network, not real traffic). */
2340 facet_is_controller_flow(struct facet
*facet
)
2343 && facet
->rule
->n_actions
== 1
2344 && action_outputs_to_port(&facet
->rule
->actions
[0],
2345 htons(OFPP_CONTROLLER
)));
2348 /* Folds all of 'facet''s statistics into its rule. Also updates the
2349 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2351 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2353 facet_account(ofproto
, facet
, 0);
2355 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2356 struct ofexpired expired
;
2357 expired
.flow
= facet
->flow
;
2358 expired
.packet_count
= facet
->packet_count
;
2359 expired
.byte_count
= facet
->byte_count
;
2360 expired
.used
= facet
->used
;
2361 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2364 facet
->rule
->packet_count
+= facet
->packet_count
;
2365 facet
->rule
->byte_count
+= facet
->byte_count
;
2367 /* Reset counters to prevent double counting if 'facet' ever gets
2369 facet
->packet_count
= 0;
2370 facet
->byte_count
= 0;
2371 facet
->accounted_bytes
= 0;
2373 netflow_flow_clear(&facet
->nf_flow
);
2376 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2377 * Returns it if found, otherwise a null pointer.
2379 * The returned facet might need revalidation; use facet_lookup_valid()
2380 * instead if that is important. */
2381 static struct facet
*
2382 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2384 struct facet
*facet
;
2386 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2388 if (flow_equal(flow
, &facet
->flow
)) {
2396 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2397 * Returns it if found, otherwise a null pointer.
2399 * The returned facet is guaranteed to be valid. */
2400 static struct facet
*
2401 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2403 struct facet
*facet
= facet_find(ofproto
, flow
);
2405 /* The facet we found might not be valid, since we could be in need of
2406 * revalidation. If it is not valid, don't return it. */
2408 && ofproto
->need_revalidate
2409 && !facet_revalidate(ofproto
, facet
)) {
2410 COVERAGE_INC(ofproto_invalidated
);
2417 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2419 * - If the rule found is different from 'facet''s current rule, moves
2420 * 'facet' to the new rule and recompiles its actions.
2422 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2423 * where it is and recompiles its actions anyway.
2425 * - If there is none, destroys 'facet'.
2427 * Returns true if 'facet' still exists, false if it has been destroyed. */
2429 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2431 struct action_xlate_ctx ctx
;
2432 struct rule
*new_rule
;
2434 bool actions_changed
;
2436 COVERAGE_INC(facet_revalidate
);
2438 /* Determine the new rule. */
2439 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2441 /* No new rule, so delete the facet. */
2442 facet_remove(ofproto
, facet
);
2446 /* Calculate new ODP actions.
2448 * We are very cautious about actually modifying 'facet' state at this
2449 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2450 * so, we need to have the old state around to properly compose it. */
2451 action_xlate_ctx_init(&ctx
, ofproto
, &facet
->flow
, NULL
);
2452 xlate_actions(&ctx
, new_rule
->actions
, new_rule
->n_actions
);
2453 actions_len
= ctx
.out
.n_actions
* sizeof *ctx
.out
.actions
;
2454 actions_changed
= (facet
->n_actions
!= ctx
.out
.n_actions
2455 || memcmp(facet
->actions
, ctx
.out
.actions
,
2458 /* If the ODP actions changed or the installability changed, then we need
2459 * to talk to the datapath. */
2460 if (actions_changed
|| facet
->may_install
!= facet
->installed
) {
2461 if (facet
->may_install
) {
2462 struct odp_flow_put put
;
2464 memset(&put
.flow
.stats
, 0, sizeof put
.flow
.stats
);
2465 odp_flow_key_from_flow(&put
.flow
.key
, &facet
->flow
);
2466 put
.flow
.actions
= ctx
.out
.actions
;
2467 put
.flow
.n_actions
= ctx
.out
.n_actions
;
2469 put
.flags
= ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
;
2470 dpif_flow_put(ofproto
->dpif
, &put
);
2472 facet_update_stats(ofproto
, facet
, &put
.flow
.stats
);
2474 facet_uninstall(ofproto
, facet
);
2477 /* The datapath flow is gone or has zeroed stats, so push stats out of
2478 * 'facet' into 'rule'. */
2479 facet_flush_stats(ofproto
, facet
);
2482 /* Update 'facet' now that we've taken care of all the old state. */
2483 facet
->tags
= ctx
.tags
;
2484 facet
->nf_flow
.output_iface
= ctx
.nf_output_iface
;
2485 facet
->may_install
= ctx
.may_set_up_flow
;
2486 if (actions_changed
) {
2487 free(facet
->actions
);
2488 facet
->n_actions
= ctx
.out
.n_actions
;
2489 facet
->actions
= xmemdup(ctx
.out
.actions
, actions_len
);
2491 if (facet
->rule
!= new_rule
) {
2492 COVERAGE_INC(facet_changed_rule
);
2493 list_remove(&facet
->list_node
);
2494 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2495 facet
->rule
= new_rule
;
2496 facet
->used
= new_rule
->created
;
2503 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2504 struct rconn_packet_counter
*counter
)
2506 update_openflow_length(msg
);
2507 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2513 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2516 struct ofpbuf
*buf
= make_ofp_error_msg(error
, oh
);
2518 COVERAGE_INC(ofproto_error
);
2519 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2524 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2526 opp
->port_no
= htons(opp
->port_no
);
2527 opp
->config
= htonl(opp
->config
);
2528 opp
->state
= htonl(opp
->state
);
2529 opp
->curr
= htonl(opp
->curr
);
2530 opp
->advertised
= htonl(opp
->advertised
);
2531 opp
->supported
= htonl(opp
->supported
);
2532 opp
->peer
= htonl(opp
->peer
);
2536 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2538 queue_tx(make_echo_reply(oh
), ofconn
, ofconn
->reply_counter
);
2543 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2545 struct ofp_switch_features
*osf
;
2547 struct ofport
*port
;
2549 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2550 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2551 osf
->n_buffers
= htonl(pktbuf_capacity());
2553 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2554 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2555 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2556 (1u << OFPAT_SET_VLAN_VID
) |
2557 (1u << OFPAT_SET_VLAN_PCP
) |
2558 (1u << OFPAT_STRIP_VLAN
) |
2559 (1u << OFPAT_SET_DL_SRC
) |
2560 (1u << OFPAT_SET_DL_DST
) |
2561 (1u << OFPAT_SET_NW_SRC
) |
2562 (1u << OFPAT_SET_NW_DST
) |
2563 (1u << OFPAT_SET_NW_TOS
) |
2564 (1u << OFPAT_SET_TP_SRC
) |
2565 (1u << OFPAT_SET_TP_DST
) |
2566 (1u << OFPAT_ENQUEUE
));
2568 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2569 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2572 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2577 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2580 struct ofp_switch_config
*osc
;
2584 /* Figure out flags. */
2585 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2586 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2589 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2590 osc
->flags
= htons(flags
);
2591 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2592 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2598 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
2600 uint16_t flags
= ntohs(osc
->flags
);
2602 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2603 switch (flags
& OFPC_FRAG_MASK
) {
2604 case OFPC_FRAG_NORMAL
:
2605 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2607 case OFPC_FRAG_DROP
:
2608 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2611 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2617 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2623 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2625 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2626 a
->controller
.arg
= max_len
;
2629 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2630 * flow translation. */
2631 #define MAX_RESUBMIT_RECURSION 8
2633 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2634 struct action_xlate_ctx
*ctx
);
2637 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2639 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2642 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2643 /* Forwarding disabled on port. */
2648 * We don't have an ofport record for this port, but it doesn't hurt to
2649 * allow forwarding to it anyhow. Maybe such a port will appear later
2650 * and we're pre-populating the flow table.
2654 odp_actions_add(&ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2655 ctx
->nf_output_iface
= port
;
2658 static struct rule
*
2659 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2661 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2665 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2667 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2668 uint16_t old_in_port
;
2671 /* Look up a flow with 'in_port' as the input port. Then restore the
2672 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2673 * have surprising behavior). */
2674 old_in_port
= ctx
->flow
.in_port
;
2675 ctx
->flow
.in_port
= in_port
;
2676 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2677 ctx
->flow
.in_port
= old_in_port
;
2679 if (ctx
->resubmit_hook
) {
2680 ctx
->resubmit_hook(ctx
, rule
);
2685 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2689 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2691 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2692 MAX_RESUBMIT_RECURSION
);
2697 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2698 uint16_t *nf_output_iface
, struct odp_actions
*actions
)
2700 struct ofport
*ofport
;
2702 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2703 uint16_t odp_port
= ofport
->odp_port
;
2704 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2705 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= odp_port
;
2708 *nf_output_iface
= NF_OUT_FLOOD
;
2712 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2713 uint16_t port
, uint16_t max_len
)
2716 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2718 ctx
->nf_output_iface
= NF_OUT_DROP
;
2722 add_output_action(ctx
, ctx
->flow
.in_port
);
2725 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2728 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2729 &ctx
->out
, &ctx
->tags
,
2730 &ctx
->nf_output_iface
,
2731 ctx
->ofproto
->aux
)) {
2732 COVERAGE_INC(ofproto_uninstallable
);
2733 ctx
->may_set_up_flow
= false;
2737 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2738 &ctx
->nf_output_iface
, &ctx
->out
);
2741 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2742 &ctx
->nf_output_iface
, &ctx
->out
);
2744 case OFPP_CONTROLLER
:
2745 add_controller_action(&ctx
->out
, max_len
);
2748 add_output_action(ctx
, ODPP_LOCAL
);
2751 odp_port
= ofp_port_to_odp_port(port
);
2752 if (odp_port
!= ctx
->flow
.in_port
) {
2753 add_output_action(ctx
, odp_port
);
2758 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2759 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2760 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2761 ctx
->nf_output_iface
= prev_nf_output_iface
;
2762 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2763 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2764 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2769 xlate_output_action(struct action_xlate_ctx
*ctx
,
2770 const struct ofp_action_output
*oao
)
2772 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2775 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2776 * optimization, because we're going to add another action that sets the
2777 * priority immediately after, or because there are no actions following the
2780 remove_pop_action(struct action_xlate_ctx
*ctx
)
2782 size_t n
= ctx
->out
.n_actions
;
2783 if (n
> 0 && ctx
->out
.actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2784 ctx
->out
.n_actions
--;
2789 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2790 const struct ofp_action_enqueue
*oae
)
2792 uint16_t ofp_port
, odp_port
;
2796 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2799 /* Fall back to ordinary output action. */
2800 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2804 /* Figure out ODP output port. */
2805 ofp_port
= ntohs(oae
->port
);
2806 if (ofp_port
!= OFPP_IN_PORT
) {
2807 odp_port
= ofp_port_to_odp_port(ofp_port
);
2809 odp_port
= ctx
->flow
.in_port
;
2812 /* Add ODP actions. */
2813 remove_pop_action(ctx
);
2814 odp_actions_add(&ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2816 add_output_action(ctx
, odp_port
);
2817 odp_actions_add(&ctx
->out
, ODPAT_POP_PRIORITY
);
2819 /* Update NetFlow output port. */
2820 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2821 ctx
->nf_output_iface
= odp_port
;
2822 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2823 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2828 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2829 const struct nx_action_set_queue
*nasq
)
2834 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2837 /* Couldn't translate queue to a priority, so ignore. A warning
2838 * has already been logged. */
2842 remove_pop_action(ctx
);
2843 odp_actions_add(&ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2848 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2850 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2851 if (!(tci
& htons(VLAN_CFI
))) {
2852 odp_actions_add(&ctx
->out
, ODPAT_STRIP_VLAN
);
2854 union odp_action
*oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_DL_TCI
);
2855 oa
->dl_tci
.tci
= tci
& ~htons(VLAN_CFI
);
2860 xlate_reg_move_action(struct action_xlate_ctx
*ctx
,
2861 const struct nx_action_reg_move
*narm
)
2863 ovs_be16 old_tci
= ctx
->flow
.vlan_tci
;
2865 nxm_execute_reg_move(narm
, &ctx
->flow
);
2867 if (ctx
->flow
.vlan_tci
!= old_tci
) {
2868 xlate_set_dl_tci(ctx
);
2873 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2874 const struct nx_action_header
*nah
)
2876 const struct nx_action_resubmit
*nar
;
2877 const struct nx_action_set_tunnel
*nast
;
2878 const struct nx_action_set_queue
*nasq
;
2879 union odp_action
*oa
;
2880 enum nx_action_subtype subtype
= ntohs(nah
->subtype
);
2882 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2884 case NXAST_RESUBMIT
:
2885 nar
= (const struct nx_action_resubmit
*) nah
;
2886 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2889 case NXAST_SET_TUNNEL
:
2890 nast
= (const struct nx_action_set_tunnel
*) nah
;
2891 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_TUNNEL
);
2892 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2895 case NXAST_DROP_SPOOFED_ARP
:
2896 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2897 odp_actions_add(&ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2901 case NXAST_SET_QUEUE
:
2902 nasq
= (const struct nx_action_set_queue
*) nah
;
2903 xlate_set_queue_action(ctx
, nasq
);
2906 case NXAST_POP_QUEUE
:
2907 odp_actions_add(&ctx
->out
, ODPAT_POP_PRIORITY
);
2910 case NXAST_REG_MOVE
:
2911 xlate_reg_move_action(ctx
, (const struct nx_action_reg_move
*) nah
);
2914 case NXAST_REG_LOAD
:
2915 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
2919 /* Nothing to do. */
2922 /* If you add a new action here that modifies flow data, don't forget to
2923 * update the flow key in ctx->flow at the same time. */
2925 case NXAST_SNAT__OBSOLETE
:
2927 VLOG_DBG_RL(&rl
, "unknown Nicira action type %d", (int) subtype
);
2933 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2934 struct action_xlate_ctx
*ctx
)
2936 struct actions_iterator iter
;
2937 const union ofp_action
*ia
;
2938 const struct ofport
*port
;
2940 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2941 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2942 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2943 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2944 /* Drop this flow. */
2948 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2949 enum ofp_action_type type
= ntohs(ia
->type
);
2950 union odp_action
*oa
;
2954 xlate_output_action(ctx
, &ia
->output
);
2957 case OFPAT_SET_VLAN_VID
:
2958 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
2959 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
2960 xlate_set_dl_tci(ctx
);
2963 case OFPAT_SET_VLAN_PCP
:
2964 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
2965 ctx
->flow
.vlan_tci
|= htons(
2966 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
2967 xlate_set_dl_tci(ctx
);
2970 case OFPAT_STRIP_VLAN
:
2971 ctx
->flow
.vlan_tci
= htons(0);
2972 xlate_set_dl_tci(ctx
);
2975 case OFPAT_SET_DL_SRC
:
2976 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_DL_SRC
);
2977 memcpy(oa
->dl_addr
.dl_addr
,
2978 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2979 memcpy(ctx
->flow
.dl_src
,
2980 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2983 case OFPAT_SET_DL_DST
:
2984 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_DL_DST
);
2985 memcpy(oa
->dl_addr
.dl_addr
,
2986 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2987 memcpy(ctx
->flow
.dl_dst
,
2988 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2991 case OFPAT_SET_NW_SRC
:
2992 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_NW_SRC
);
2993 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2996 case OFPAT_SET_NW_DST
:
2997 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_NW_DST
);
2998 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
3001 case OFPAT_SET_NW_TOS
:
3002 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_NW_TOS
);
3003 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
3006 case OFPAT_SET_TP_SRC
:
3007 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_TP_SRC
);
3008 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
3011 case OFPAT_SET_TP_DST
:
3012 oa
= odp_actions_add(&ctx
->out
, ODPAT_SET_TP_DST
);
3013 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
3017 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
3021 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
3025 VLOG_DBG_RL(&rl
, "unknown action type %d", (int) type
);
3032 action_xlate_ctx_init(struct action_xlate_ctx
*ctx
,
3033 struct ofproto
*ofproto
, const struct flow
*flow
,
3034 const struct ofpbuf
*packet
)
3036 ctx
->ofproto
= ofproto
;
3038 ctx
->packet
= packet
;
3039 ctx
->resubmit_hook
= NULL
;
3043 xlate_actions(struct action_xlate_ctx
*ctx
,
3044 const union ofp_action
*in
, size_t n_in
)
3046 COVERAGE_INC(ofproto_ofp2odp
);
3047 odp_actions_init(&ctx
->out
);
3049 ctx
->may_set_up_flow
= true;
3050 ctx
->nf_output_iface
= NF_OUT_DROP
;
3052 do_xlate_actions(in
, n_in
, ctx
);
3053 remove_pop_action(ctx
);
3055 /* Check with in-band control to see if we're allowed to set up this
3057 if (!in_band_rule_check(ctx
->ofproto
->in_band
, &ctx
->flow
, &ctx
->out
)) {
3058 ctx
->may_set_up_flow
= false;
3061 if (odp_actions_overflow(&ctx
->out
)) {
3062 COVERAGE_INC(odp_overflow
);
3063 odp_actions_init(&ctx
->out
);
3064 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
3069 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3070 * error message code (composed with ofp_mkerr()) for the caller to propagate
3071 * upward. Otherwise, returns 0.
3073 * The log message mentions 'msg_type'. */
3075 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3077 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3078 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3079 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3082 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3089 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3091 struct ofproto
*p
= ofconn
->ofproto
;
3092 struct ofp_packet_out
*opo
;
3093 struct ofpbuf payload
, *buffer
;
3094 union ofp_action
*ofp_actions
;
3095 struct action_xlate_ctx ctx
;
3096 struct ofpbuf request
;
3098 size_t n_ofp_actions
;
3102 COVERAGE_INC(ofproto_packet_out
);
3104 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3109 /* Get ofp_packet_out. */
3110 ofpbuf_use_const(&request
, oh
, ntohs(oh
->length
));
3111 opo
= ofpbuf_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3114 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3115 &ofp_actions
, &n_ofp_actions
);
3121 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3122 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3124 if (error
|| !buffer
) {
3133 /* Extract flow, check actions. */
3134 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3136 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3142 action_xlate_ctx_init(&ctx
, p
, &flow
, &payload
);
3143 error
= xlate_actions(&ctx
, ofp_actions
, n_ofp_actions
);
3145 dpif_execute(p
->dpif
, ctx
.out
.actions
, ctx
.out
.n_actions
, &payload
);
3149 ofpbuf_delete(buffer
);
3154 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3155 uint32_t config
, uint32_t mask
)
3157 mask
&= config
^ port
->opp
.config
;
3158 if (mask
& OFPPC_PORT_DOWN
) {
3159 if (config
& OFPPC_PORT_DOWN
) {
3160 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3162 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3165 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3166 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3167 if (mask
& REVALIDATE_BITS
) {
3168 COVERAGE_INC(ofproto_costly_flags
);
3169 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3170 p
->need_revalidate
= true;
3172 #undef REVALIDATE_BITS
3173 if (mask
& OFPPC_NO_PACKET_IN
) {
3174 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3179 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3181 struct ofproto
*p
= ofconn
->ofproto
;
3182 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
3183 struct ofport
*port
;
3186 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3191 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3193 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3194 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3195 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3197 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3198 if (opm
->advertise
) {
3199 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3205 static struct ofpbuf
*
3206 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3208 struct ofp_stats_reply
*osr
;
3211 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3212 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3214 osr
->flags
= htons(0);
3218 static struct ofpbuf
*
3219 start_ofp_stats_reply(const struct ofp_header
*request
, size_t body_len
)
3221 const struct ofp_stats_request
*osr
3222 = (const struct ofp_stats_request
*) request
;
3223 return make_ofp_stats_reply(osr
->header
.xid
, osr
->type
, body_len
);
3227 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3228 struct ofpbuf
**msgp
)
3230 struct ofpbuf
*msg
= *msgp
;
3231 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3232 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3233 struct ofp_stats_reply
*reply
= msg
->data
;
3234 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3235 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3236 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3238 return ofpbuf_put_uninit(*msgp
, nbytes
);
3241 static struct ofpbuf
*
3242 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3244 struct nicira_stats_msg
*nsm
;
3247 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3248 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3249 nsm
->type
= htons(OFPST_VENDOR
);
3250 nsm
->flags
= htons(0);
3251 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3252 nsm
->subtype
= htonl(subtype
);
3256 static struct ofpbuf
*
3257 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3259 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3263 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3264 struct ofpbuf
**msgp
)
3266 struct ofpbuf
*msg
= *msgp
;
3267 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3268 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3269 struct nicira_stats_msg
*reply
= msg
->data
;
3270 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3271 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3272 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3274 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3278 handle_desc_stats_request(struct ofconn
*ofconn
,
3279 const struct ofp_header
*request
)
3281 struct ofproto
*p
= ofconn
->ofproto
;
3282 struct ofp_desc_stats
*ods
;
3285 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3286 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3287 memset(ods
, 0, sizeof *ods
);
3288 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3289 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3290 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3291 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3292 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3293 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3299 handle_table_stats_request(struct ofconn
*ofconn
,
3300 const struct ofp_header
*request
)
3302 struct ofproto
*p
= ofconn
->ofproto
;
3303 struct ofp_table_stats
*ots
;
3306 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3308 /* Classifier table. */
3309 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3310 memset(ots
, 0, sizeof *ots
);
3311 strcpy(ots
->name
, "classifier");
3312 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3313 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3314 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3315 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3316 ots
->lookup_count
= htonll(0); /* XXX */
3317 ots
->matched_count
= htonll(0); /* XXX */
3319 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3324 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3325 struct ofpbuf
**msgp
)
3327 struct netdev_stats stats
;
3328 struct ofp_port_stats
*ops
;
3330 /* Intentionally ignore return value, since errors will set
3331 * 'stats' to all-1s, which is correct for OpenFlow, and
3332 * netdev_get_stats() will log errors. */
3333 netdev_get_stats(port
->netdev
, &stats
);
3335 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3336 ops
->port_no
= htons(port
->opp
.port_no
);
3337 memset(ops
->pad
, 0, sizeof ops
->pad
);
3338 ops
->rx_packets
= htonll(stats
.rx_packets
);
3339 ops
->tx_packets
= htonll(stats
.tx_packets
);
3340 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3341 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3342 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3343 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3344 ops
->rx_errors
= htonll(stats
.rx_errors
);
3345 ops
->tx_errors
= htonll(stats
.tx_errors
);
3346 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3347 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3348 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3349 ops
->collisions
= htonll(stats
.collisions
);
3353 handle_port_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3355 struct ofproto
*p
= ofconn
->ofproto
;
3356 const struct ofp_port_stats_request
*psr
= ofputil_stats_body(oh
);
3357 struct ofp_port_stats
*ops
;
3359 struct ofport
*port
;
3361 msg
= start_ofp_stats_reply(oh
, sizeof *ops
* 16);
3362 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3363 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3365 append_port_stat(port
, ofconn
, &msg
);
3368 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3369 append_port_stat(port
, ofconn
, &msg
);
3373 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3377 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3378 * '*packet_countp' and '*byte_countp'. The returned statistics include
3379 * statistics for all of 'rule''s facets. */
3381 query_stats(struct ofproto
*p
, struct rule
*rule
,
3382 uint64_t *packet_countp
, uint64_t *byte_countp
)
3384 uint64_t packet_count
, byte_count
;
3385 struct facet
*facet
;
3386 struct odp_flow
*odp_flows
;
3389 /* Start from historical data for 'rule' itself that are no longer tracked
3390 * by the datapath. This counts, for example, facets that have expired. */
3391 packet_count
= rule
->packet_count
;
3392 byte_count
= rule
->byte_count
;
3394 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3396 * Also, add any statistics that are not tracked by the datapath for each
3397 * facet. This includes, for example, statistics for packets that were
3398 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3400 odp_flows
= xzalloc(list_size(&rule
->facets
) * sizeof *odp_flows
);
3402 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
3403 struct odp_flow
*odp_flow
= &odp_flows
[n_odp_flows
++];
3404 odp_flow_key_from_flow(&odp_flow
->key
, &facet
->flow
);
3405 packet_count
+= facet
->packet_count
;
3406 byte_count
+= facet
->byte_count
;
3409 /* Fetch up-to-date statistics from the datapath and add them in. */
3410 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3413 for (i
= 0; i
< n_odp_flows
; i
++) {
3414 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3415 packet_count
+= odp_flow
->stats
.n_packets
;
3416 byte_count
+= odp_flow
->stats
.n_bytes
;
3421 /* Return the stats to the caller. */
3422 *packet_countp
= packet_count
;
3423 *byte_countp
= byte_count
;
3427 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3429 long long int msecs
= time_msec() - start
;
3430 *sec
= htonl(msecs
/ 1000);
3431 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3435 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3436 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3438 struct ofp_flow_stats
*ofs
;
3439 uint64_t packet_count
, byte_count
;
3440 size_t act_len
, len
;
3442 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3446 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3447 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3449 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3451 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3452 ofs
->length
= htons(len
);
3455 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
);
3456 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3457 ofs
->cookie
= rule
->flow_cookie
;
3458 ofs
->priority
= htons(rule
->cr
.priority
);
3459 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3460 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3461 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3462 ofs
->packet_count
= htonll(packet_count
);
3463 ofs
->byte_count
= htonll(byte_count
);
3464 if (rule
->n_actions
> 0) {
3465 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3470 is_valid_table(uint8_t table_id
)
3472 return table_id
== 0 || table_id
== 0xff;
3476 handle_flow_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3478 const struct ofp_flow_stats_request
*fsr
= ofputil_stats_body(oh
);
3479 struct ofpbuf
*reply
;
3481 COVERAGE_INC(ofproto_flows_req
);
3482 reply
= start_ofp_stats_reply(oh
, 1024);
3483 if (is_valid_table(fsr
->table_id
)) {
3484 struct cls_cursor cursor
;
3485 struct cls_rule target
;
3488 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3490 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3491 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3492 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3495 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3501 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3502 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3504 struct nx_flow_stats
*nfs
;
3505 uint64_t packet_count
, byte_count
;
3506 size_t act_len
, start_len
;
3507 struct ofpbuf
*reply
;
3509 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3513 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3515 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3517 start_len
= (*replyp
)->size
;
3518 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3521 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3524 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3525 nfs
->cookie
= rule
->flow_cookie
;
3526 nfs
->priority
= htons(rule
->cr
.priority
);
3527 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3528 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3529 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3530 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3531 nfs
->packet_count
= htonll(packet_count
);
3532 nfs
->byte_count
= htonll(byte_count
);
3533 if (rule
->n_actions
> 0) {
3534 ofpbuf_put(reply
, rule
->actions
, act_len
);
3536 nfs
->length
= htons(reply
->size
- start_len
);
3540 handle_nxst_flow(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3542 struct nx_flow_stats_request
*nfsr
;
3543 struct cls_rule target
;
3544 struct ofpbuf
*reply
;
3548 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3550 /* Dissect the message. */
3551 nfsr
= ofpbuf_pull(&b
, sizeof *nfsr
);
3552 error
= nx_pull_match(&b
, ntohs(nfsr
->match_len
), 0, &target
);
3557 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3560 COVERAGE_INC(ofproto_flows_req
);
3561 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3562 if (is_valid_table(nfsr
->table_id
)) {
3563 struct cls_cursor cursor
;
3566 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3567 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3568 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3571 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3577 flow_stats_ds(struct ofproto
*ofproto
, struct rule
*rule
, struct ds
*results
)
3579 struct ofp_match match
;
3580 uint64_t packet_count
, byte_count
;
3581 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3583 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3584 ofputil_cls_rule_to_match(&rule
->cr
, NXFF_OPENFLOW10
, &match
);
3586 ds_put_format(results
, "duration=%llds, ",
3587 (time_msec() - rule
->created
) / 1000);
3588 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3589 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3590 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3591 ofp_print_match(results
, &match
, true);
3593 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3595 ds_put_cstr(results
, "drop");
3597 ds_put_cstr(results
, "\n");
3600 /* Adds a pretty-printed description of all flows to 'results', including
3601 * those marked hidden by secchan (e.g., by in-band control). */
3603 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3605 struct cls_cursor cursor
;
3608 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3609 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3610 flow_stats_ds(p
, rule
, results
);
3615 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3616 ovs_be16 out_port
, uint8_t table_id
,
3617 struct ofp_aggregate_stats_reply
*oasr
)
3619 uint64_t total_packets
= 0;
3620 uint64_t total_bytes
= 0;
3623 COVERAGE_INC(ofproto_agg_request
);
3625 if (is_valid_table(table_id
)) {
3626 struct cls_cursor cursor
;
3629 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3630 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3631 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3632 uint64_t packet_count
;
3633 uint64_t byte_count
;
3635 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3637 total_packets
+= packet_count
;
3638 total_bytes
+= byte_count
;
3644 oasr
->flow_count
= htonl(n_flows
);
3645 oasr
->packet_count
= htonll(total_packets
);
3646 oasr
->byte_count
= htonll(total_bytes
);
3647 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3651 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3652 const struct ofp_header
*oh
)
3654 const struct ofp_aggregate_stats_request
*request
= ofputil_stats_body(oh
);
3655 struct ofp_aggregate_stats_reply
*reply
;
3656 struct cls_rule target
;
3659 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3662 msg
= start_ofp_stats_reply(oh
, sizeof *reply
);
3663 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3664 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3665 request
->table_id
, reply
);
3666 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3671 handle_nxst_aggregate(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3673 struct nx_aggregate_stats_request
*request
;
3674 struct ofp_aggregate_stats_reply
*reply
;
3675 struct cls_rule target
;
3680 ofpbuf_use_const(&b
, oh
, ntohs(oh
->length
));
3682 /* Dissect the message. */
3683 request
= ofpbuf_pull(&b
, sizeof *request
);
3684 error
= nx_pull_match(&b
, ntohs(request
->match_len
), 0, &target
);
3689 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3693 COVERAGE_INC(ofproto_flows_req
);
3694 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3695 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3696 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3697 request
->table_id
, reply
);
3698 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3703 struct queue_stats_cbdata
{
3704 struct ofconn
*ofconn
;
3705 struct ofport
*ofport
;
3710 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3711 const struct netdev_queue_stats
*stats
)
3713 struct ofp_queue_stats
*reply
;
3715 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3716 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3717 memset(reply
->pad
, 0, sizeof reply
->pad
);
3718 reply
->queue_id
= htonl(queue_id
);
3719 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3720 reply
->tx_packets
= htonll(stats
->tx_packets
);
3721 reply
->tx_errors
= htonll(stats
->tx_errors
);
3725 handle_queue_stats_dump_cb(uint32_t queue_id
,
3726 struct netdev_queue_stats
*stats
,
3729 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3731 put_queue_stats(cbdata
, queue_id
, stats
);
3735 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3736 struct queue_stats_cbdata
*cbdata
)
3738 cbdata
->ofport
= port
;
3739 if (queue_id
== OFPQ_ALL
) {
3740 netdev_dump_queue_stats(port
->netdev
,
3741 handle_queue_stats_dump_cb
, cbdata
);
3743 struct netdev_queue_stats stats
;
3745 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3746 put_queue_stats(cbdata
, queue_id
, &stats
);
3752 handle_queue_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3754 struct ofproto
*ofproto
= ofconn
->ofproto
;
3755 const struct ofp_queue_stats_request
*qsr
;
3756 struct queue_stats_cbdata cbdata
;
3757 struct ofport
*port
;
3758 unsigned int port_no
;
3761 qsr
= ofputil_stats_body(oh
);
3763 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3766 COVERAGE_INC(ofproto_queue_req
);
3768 cbdata
.ofconn
= ofconn
;
3769 cbdata
.msg
= start_ofp_stats_reply(oh
, 128);
3771 port_no
= ntohs(qsr
->port_no
);
3772 queue_id
= ntohl(qsr
->queue_id
);
3773 if (port_no
== OFPP_ALL
) {
3774 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3775 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3777 } else if (port_no
< ofproto
->max_ports
) {
3778 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3780 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3783 ofpbuf_delete(cbdata
.msg
);
3784 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3786 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3791 static long long int
3792 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3794 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3798 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3799 const struct odp_flow_stats
*stats
)
3801 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3802 if (used
> facet
->used
) {
3804 if (used
> facet
->rule
->used
) {
3805 facet
->rule
->used
= used
;
3807 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3811 /* Folds the statistics from 'stats' into the counters in 'facet'.
3813 * Because of the meaning of a facet's counters, it only makes sense to do this
3814 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3815 * packet that was sent by hand or if it represents statistics that have been
3816 * cleared out of the datapath. */
3818 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3819 const struct odp_flow_stats
*stats
)
3821 if (stats
->n_packets
) {
3822 facet_update_time(ofproto
, facet
, stats
);
3823 facet
->packet_count
+= stats
->n_packets
;
3824 facet
->byte_count
+= stats
->n_bytes
;
3825 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3829 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3830 * in which no matching flow already exists in the flow table.
3832 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3833 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3834 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3836 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3839 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3841 struct ofproto
*p
= ofconn
->ofproto
;
3842 struct ofpbuf
*packet
;
3847 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3848 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3849 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3853 if (fm
->buffer_id
!= UINT32_MAX
) {
3854 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3858 in_port
= UINT16_MAX
;
3861 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
3862 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3863 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3864 rule_insert(p
, rule
);
3866 rule_execute(p
, rule
, in_port
, packet
);
3871 static struct rule
*
3872 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3874 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
3878 send_buffered_packet(struct ofconn
*ofconn
,
3879 struct rule
*rule
, uint32_t buffer_id
)
3881 struct ofpbuf
*packet
;
3885 if (buffer_id
== UINT32_MAX
) {
3889 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
3894 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
3899 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3901 struct modify_flows_cbdata
{
3902 struct ofproto
*ofproto
;
3903 const struct flow_mod
*fm
;
3907 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
3910 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3911 * encoded by ofp_mkerr() on failure.
3913 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3916 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3918 struct ofproto
*p
= ofconn
->ofproto
;
3919 struct rule
*match
= NULL
;
3920 struct cls_cursor cursor
;
3923 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
3924 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3925 if (!rule_is_hidden(rule
)) {
3927 modify_flow(p
, fm
, rule
);
3932 /* This credits the packet to whichever flow happened to match last.
3933 * That's weird. Maybe we should do a lookup for the flow that
3934 * actually matches the packet? Who knows. */
3935 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
3938 return add_flow(ofconn
, fm
);
3942 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3943 * code as encoded by ofp_mkerr() on failure.
3945 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3948 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3950 struct ofproto
*p
= ofconn
->ofproto
;
3951 struct rule
*rule
= find_flow_strict(p
, fm
);
3952 if (rule
&& !rule_is_hidden(rule
)) {
3953 modify_flow(p
, fm
, rule
);
3954 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
3956 return add_flow(ofconn
, fm
);
3960 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3961 * been identified as a flow in 'p''s flow table to be modified, by changing
3962 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3963 * ofp_action[] structures). */
3965 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
3967 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
3969 rule
->flow_cookie
= fm
->cookie
;
3971 /* If the actions are the same, do nothing. */
3972 if (fm
->n_actions
== rule
->n_actions
3974 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
3978 /* Replace actions. */
3979 free(rule
->actions
);
3980 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
3981 rule
->n_actions
= fm
->n_actions
;
3983 p
->need_revalidate
= true;
3988 /* OFPFC_DELETE implementation. */
3990 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
3992 /* Implements OFPFC_DELETE. */
3994 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
3996 struct rule
*rule
, *next_rule
;
3997 struct cls_cursor cursor
;
3999 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4000 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4001 delete_flow(p
, rule
, htons(fm
->out_port
));
4005 /* Implements OFPFC_DELETE_STRICT. */
4007 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4009 struct rule
*rule
= find_flow_strict(p
, fm
);
4011 delete_flow(p
, rule
, htons(fm
->out_port
));
4015 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4016 * been identified as a flow to delete from 'p''s flow table, by deleting the
4017 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4020 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4021 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4022 * specified 'out_port'. */
4024 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4026 if (rule_is_hidden(rule
)) {
4030 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4034 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4035 rule_remove(p
, rule
);
4039 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4041 struct ofproto
*p
= ofconn
->ofproto
;
4045 error
= reject_slave_controller(ofconn
, "flow_mod");
4050 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn
->flow_format
);
4055 /* We do not support the emergency flow cache. It will hopefully get
4056 * dropped from OpenFlow in the near future. */
4057 if (fm
.flags
& OFPFF_EMERG
) {
4058 /* There isn't a good fit for an error code, so just state that the
4059 * flow table is full. */
4060 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4063 error
= validate_actions(fm
.actions
, fm
.n_actions
,
4064 &fm
.cr
.flow
, p
->max_ports
);
4069 switch (fm
.command
) {
4071 return add_flow(ofconn
, &fm
);
4074 return modify_flows_loose(ofconn
, &fm
);
4076 case OFPFC_MODIFY_STRICT
:
4077 return modify_flow_strict(ofconn
, &fm
);
4080 delete_flows_loose(p
, &fm
);
4083 case OFPFC_DELETE_STRICT
:
4084 delete_flow_strict(p
, &fm
);
4088 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4093 handle_tun_id_from_cookie(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4095 const struct nxt_tun_id_cookie
*msg
4096 = (const struct nxt_tun_id_cookie
*) oh
;
4098 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4103 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4105 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
4106 struct nx_role_request
*reply
;
4110 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4111 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4113 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4116 role
= ntohl(nrr
->role
);
4117 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4118 && role
!= NX_ROLE_SLAVE
) {
4119 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4121 /* There's no good error code for this. */
4122 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4125 if (role
== NX_ROLE_MASTER
) {
4126 struct ofconn
*other
;
4128 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4129 if (other
->role
== NX_ROLE_MASTER
) {
4130 other
->role
= NX_ROLE_SLAVE
;
4134 ofconn
->role
= role
;
4136 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
4137 reply
->role
= htonl(role
);
4138 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4144 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4146 const struct nxt_set_flow_format
*msg
4147 = (const struct nxt_set_flow_format
*) oh
;
4150 format
= ntohl(msg
->format
);
4151 if (format
== NXFF_OPENFLOW10
4152 || format
== NXFF_TUN_ID_FROM_COOKIE
4153 || format
== NXFF_NXM
) {
4154 ofconn
->flow_format
= format
;
4157 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4162 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4164 struct ofp_header
*ob
;
4167 /* Currently, everything executes synchronously, so we can just
4168 * immediately send the barrier reply. */
4169 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4170 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4175 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4177 const struct ofp_header
*oh
= msg
->data
;
4178 const struct ofputil_msg_type
*type
;
4181 error
= ofputil_decode_msg_type(oh
, &type
);
4186 switch (ofputil_msg_type_code(type
)) {
4187 /* OpenFlow requests. */
4188 case OFPUTIL_OFPT_ECHO_REQUEST
:
4189 return handle_echo_request(ofconn
, oh
);
4191 case OFPUTIL_OFPT_FEATURES_REQUEST
:
4192 return handle_features_request(ofconn
, oh
);
4194 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
4195 return handle_get_config_request(ofconn
, oh
);
4197 case OFPUTIL_OFPT_SET_CONFIG
:
4198 return handle_set_config(ofconn
, msg
->data
);
4200 case OFPUTIL_OFPT_PACKET_OUT
:
4201 return handle_packet_out(ofconn
, oh
);
4203 case OFPUTIL_OFPT_PORT_MOD
:
4204 return handle_port_mod(ofconn
, oh
);
4206 case OFPUTIL_OFPT_FLOW_MOD
:
4207 return handle_flow_mod(ofconn
, oh
);
4209 case OFPUTIL_OFPT_BARRIER_REQUEST
:
4210 return handle_barrier_request(ofconn
, oh
);
4212 /* OpenFlow replies. */
4213 case OFPUTIL_OFPT_ECHO_REPLY
:
4216 /* Nicira extension requests. */
4217 case OFPUTIL_NXT_STATUS_REQUEST
:
4218 return switch_status_handle_request(
4219 ofconn
->ofproto
->switch_status
, ofconn
->rconn
, oh
);
4221 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE
:
4222 return handle_tun_id_from_cookie(ofconn
, oh
);
4224 case OFPUTIL_NXT_ROLE_REQUEST
:
4225 return handle_role_request(ofconn
, oh
);
4227 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
4228 return handle_nxt_set_flow_format(ofconn
, oh
);
4230 case OFPUTIL_NXT_FLOW_MOD
:
4231 return handle_flow_mod(ofconn
, oh
);
4233 /* OpenFlow statistics requests. */
4234 case OFPUTIL_OFPST_DESC_REQUEST
:
4235 return handle_desc_stats_request(ofconn
, oh
);
4237 case OFPUTIL_OFPST_FLOW_REQUEST
:
4238 return handle_flow_stats_request(ofconn
, oh
);
4240 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
4241 return handle_aggregate_stats_request(ofconn
, oh
);
4243 case OFPUTIL_OFPST_TABLE_REQUEST
:
4244 return handle_table_stats_request(ofconn
, oh
);
4246 case OFPUTIL_OFPST_PORT_REQUEST
:
4247 return handle_port_stats_request(ofconn
, oh
);
4249 case OFPUTIL_OFPST_QUEUE_REQUEST
:
4250 return handle_queue_stats_request(ofconn
, oh
);
4252 /* Nicira extension statistics requests. */
4253 case OFPUTIL_NXST_FLOW_REQUEST
:
4254 return handle_nxst_flow(ofconn
, oh
);
4256 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
4257 return handle_nxst_aggregate(ofconn
, oh
);
4259 case OFPUTIL_INVALID
:
4260 case OFPUTIL_OFPT_HELLO
:
4261 case OFPUTIL_OFPT_ERROR
:
4262 case OFPUTIL_OFPT_FEATURES_REPLY
:
4263 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
4264 case OFPUTIL_OFPT_PACKET_IN
:
4265 case OFPUTIL_OFPT_FLOW_REMOVED
:
4266 case OFPUTIL_OFPT_PORT_STATUS
:
4267 case OFPUTIL_OFPT_BARRIER_REPLY
:
4268 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
4269 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
4270 case OFPUTIL_OFPST_DESC_REPLY
:
4271 case OFPUTIL_OFPST_FLOW_REPLY
:
4272 case OFPUTIL_OFPST_QUEUE_REPLY
:
4273 case OFPUTIL_OFPST_PORT_REPLY
:
4274 case OFPUTIL_OFPST_TABLE_REPLY
:
4275 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
4276 case OFPUTIL_NXT_STATUS_REPLY
:
4277 case OFPUTIL_NXT_ROLE_REPLY
:
4278 case OFPUTIL_NXT_FLOW_REMOVED
:
4279 case OFPUTIL_NXST_FLOW_REPLY
:
4280 case OFPUTIL_NXST_AGGREGATE_REPLY
:
4282 if (VLOG_IS_WARN_ENABLED()) {
4283 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4284 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4287 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
4288 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
4290 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4296 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4298 int error
= handle_openflow__(ofconn
, ofp_msg
);
4300 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4302 COVERAGE_INC(ofproto_recv_openflow
);
4306 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4308 struct odp_msg
*msg
= packet
->data
;
4309 struct ofpbuf payload
;
4310 struct facet
*facet
;
4313 ofpbuf_use_const(&payload
, msg
+ 1, msg
->length
- sizeof *msg
);
4314 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4316 packet
->l2
= payload
.l2
;
4317 packet
->l3
= payload
.l3
;
4318 packet
->l4
= payload
.l4
;
4319 packet
->l7
= payload
.l7
;
4321 /* Check with in-band control to see if this packet should be sent
4322 * to the local port regardless of the flow table. */
4323 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4324 union odp_action action
;
4326 memset(&action
, 0, sizeof(action
));
4327 action
.output
.type
= ODPAT_OUTPUT
;
4328 action
.output
.port
= ODPP_LOCAL
;
4329 dpif_execute(p
->dpif
, &action
, 1, &payload
);
4332 facet
= facet_lookup_valid(p
, &flow
);
4334 struct rule
*rule
= rule_lookup(p
, &flow
);
4336 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4337 struct ofport
*port
= get_port(p
, msg
->port
);
4339 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4340 COVERAGE_INC(ofproto_no_packet_in
);
4341 /* XXX install 'drop' flow entry */
4342 ofpbuf_delete(packet
);
4346 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4350 COVERAGE_INC(ofproto_packet_in
);
4351 send_packet_in(p
, packet
);
4355 facet
= facet_create(p
, rule
, &flow
, packet
);
4356 } else if (!facet
->may_install
) {
4357 /* The facet is not installable, that is, we need to process every
4358 * packet, so process the current packet's actions into 'facet'. */
4359 facet_make_actions(p
, facet
, packet
);
4362 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4364 * Extra-special case for fail-open mode.
4366 * We are in fail-open mode and the packet matched the fail-open rule,
4367 * but we are connected to a controller too. We should send the packet
4368 * up to the controller in the hope that it will try to set up a flow
4369 * and thereby allow us to exit fail-open.
4371 * See the top-level comment in fail-open.c for more information.
4373 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4374 DPIF_RECV_MSG_PADDING
));
4377 ofpbuf_pull(packet
, sizeof *msg
);
4378 facet_execute(p
, facet
, packet
);
4379 facet_install(p
, facet
, false);
4383 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4385 struct odp_msg
*msg
= packet
->data
;
4387 switch (msg
->type
) {
4388 case _ODPL_ACTION_NR
:
4389 COVERAGE_INC(ofproto_ctlr_action
);
4390 send_packet_in(p
, packet
);
4393 case _ODPL_SFLOW_NR
:
4395 ofproto_sflow_received(p
->sflow
, msg
);
4397 ofpbuf_delete(packet
);
4401 handle_odp_miss_msg(p
, packet
);
4405 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4411 /* Flow expiration. */
4413 static int ofproto_dp_max_idle(const struct ofproto
*);
4414 static void ofproto_update_used(struct ofproto
*);
4415 static void rule_expire(struct ofproto
*, struct rule
*);
4416 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4418 /* This function is called periodically by ofproto_run(). Its job is to
4419 * collect updates for the flows that have been installed into the datapath,
4420 * most importantly when they last were used, and then use that information to
4421 * expire flows that have not been used recently.
4423 * Returns the number of milliseconds after which it should be called again. */
4425 ofproto_expire(struct ofproto
*ofproto
)
4427 struct rule
*rule
, *next_rule
;
4428 struct cls_cursor cursor
;
4431 /* Update 'used' for each flow in the datapath. */
4432 ofproto_update_used(ofproto
);
4434 /* Expire facets that have been idle too long. */
4435 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4436 ofproto_expire_facets(ofproto
, dp_max_idle
);
4438 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4439 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4440 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4441 rule_expire(ofproto
, rule
);
4444 /* Let the hook know that we're at a stable point: all outstanding data
4445 * in existing flows has been accounted to the account_cb. Thus, the
4446 * hook can now reasonably do operations that depend on having accurate
4447 * flow volume accounting (currently, that's just bond rebalancing). */
4448 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4449 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4452 return MIN(dp_max_idle
, 1000);
4455 /* Update 'used' member of installed facets. */
4457 ofproto_update_used(struct ofproto
*p
)
4459 struct odp_flow
*flows
;
4464 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4469 for (i
= 0; i
< n_flows
; i
++) {
4470 struct odp_flow
*f
= &flows
[i
];
4471 struct facet
*facet
;
4474 odp_flow_key_to_flow(&f
->key
, &flow
);
4475 facet
= facet_find(p
, &flow
);
4477 if (facet
&& facet
->installed
) {
4478 facet_update_time(p
, facet
, &f
->stats
);
4479 facet_account(p
, facet
, f
->stats
.n_bytes
);
4481 /* There's a flow in the datapath that we know nothing about.
4483 COVERAGE_INC(ofproto_unexpected_rule
);
4484 dpif_flow_del(p
->dpif
, f
);
4491 /* Calculates and returns the number of milliseconds of idle time after which
4492 * facets should expire from the datapath and we should fold their statistics
4493 * into their parent rules in userspace. */
4495 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4498 * Idle time histogram.
4500 * Most of the time a switch has a relatively small number of facets. When
4501 * this is the case we might as well keep statistics for all of them in
4502 * userspace and to cache them in the kernel datapath for performance as
4505 * As the number of facets increases, the memory required to maintain
4506 * statistics about them in userspace and in the kernel becomes
4507 * significant. However, with a large number of facets it is likely that
4508 * only a few of them are "heavy hitters" that consume a large amount of
4509 * bandwidth. At this point, only heavy hitters are worth caching in the
4510 * kernel and maintaining in userspaces; other facets we can discard.
4512 * The technique used to compute the idle time is to build a histogram with
4513 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4514 * that is installed in the kernel gets dropped in the appropriate bucket.
4515 * After the histogram has been built, we compute the cutoff so that only
4516 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4517 * cached. At least the most-recently-used bucket of facets is kept, so
4518 * actually an arbitrary number of facets can be kept in any given
4519 * expiration run (though the next run will delete most of those unless
4520 * they receive additional data).
4522 * This requires a second pass through the facets, in addition to the pass
4523 * made by ofproto_update_used(), because the former function never looks
4524 * at uninstallable facets.
4526 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4527 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4528 int buckets
[N_BUCKETS
] = { 0 };
4529 struct facet
*facet
;
4534 total
= hmap_count(&ofproto
->facets
);
4535 if (total
<= 1000) {
4536 return N_BUCKETS
* BUCKET_WIDTH
;
4539 /* Build histogram. */
4541 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4542 long long int idle
= now
- facet
->used
;
4543 int bucket
= (idle
<= 0 ? 0
4544 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4545 : (unsigned int) idle
/ BUCKET_WIDTH
);
4549 /* Find the first bucket whose flows should be expired. */
4550 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4551 if (buckets
[bucket
]) {
4554 subtotal
+= buckets
[bucket
++];
4555 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4560 if (VLOG_IS_DBG_ENABLED()) {
4564 ds_put_cstr(&s
, "keep");
4565 for (i
= 0; i
< N_BUCKETS
; i
++) {
4567 ds_put_cstr(&s
, ", drop");
4570 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4573 VLOG_INFO("%s: %s (msec:count)",
4574 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4578 return bucket
* BUCKET_WIDTH
;
4582 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4584 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4585 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4586 struct ofexpired expired
;
4587 struct odp_flow odp_flow
;
4589 /* Get updated flow stats.
4591 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4592 * updated TCP flags and (2) the dpif_flow_list_all() in
4593 * ofproto_update_used() zeroed TCP flags. */
4594 memset(&odp_flow
, 0, sizeof odp_flow
);
4595 if (facet
->installed
) {
4596 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
4597 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4598 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4600 if (odp_flow
.stats
.n_packets
) {
4601 facet_update_time(ofproto
, facet
, &odp_flow
.stats
);
4602 netflow_flow_update_flags(&facet
->nf_flow
,
4603 odp_flow
.stats
.tcp_flags
);
4607 expired
.flow
= facet
->flow
;
4608 expired
.packet_count
= facet
->packet_count
+
4609 odp_flow
.stats
.n_packets
;
4610 expired
.byte_count
= facet
->byte_count
+ odp_flow
.stats
.n_bytes
;
4611 expired
.used
= facet
->used
;
4613 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4618 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4620 long long int cutoff
= time_msec() - dp_max_idle
;
4621 struct facet
*facet
, *next_facet
;
4623 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4624 facet_active_timeout(ofproto
, facet
);
4625 if (facet
->used
< cutoff
) {
4626 facet_remove(ofproto
, facet
);
4631 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4632 * then delete it entirely. */
4634 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4636 struct facet
*facet
, *next_facet
;
4640 /* Has 'rule' expired? */
4642 if (rule
->hard_timeout
4643 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4644 reason
= OFPRR_HARD_TIMEOUT
;
4645 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4646 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4647 reason
= OFPRR_IDLE_TIMEOUT
;
4652 COVERAGE_INC(ofproto_expired
);
4654 /* Update stats. (This is a no-op if the rule expired due to an idle
4655 * timeout, because that only happens when the rule has no facets left.) */
4656 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4657 facet_remove(ofproto
, facet
);
4660 /* Get rid of the rule. */
4661 if (!rule_is_hidden(rule
)) {
4662 rule_send_removed(ofproto
, rule
, reason
);
4664 rule_remove(ofproto
, rule
);
4667 static struct ofpbuf
*
4668 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4671 struct ofp_flow_removed
*ofr
;
4674 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4675 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofr
->match
);
4676 ofr
->cookie
= rule
->flow_cookie
;
4677 ofr
->priority
= htons(rule
->cr
.priority
);
4678 ofr
->reason
= reason
;
4679 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4680 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4681 ofr
->packet_count
= htonll(rule
->packet_count
);
4682 ofr
->byte_count
= htonll(rule
->byte_count
);
4687 static struct ofpbuf
*
4688 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4690 struct nx_flow_removed
*nfr
;
4694 nfr
= make_nxmsg(sizeof *nfr
, NXT_FLOW_REMOVED
, &buf
);
4696 match_len
= nx_put_match(buf
, &rule
->cr
);
4698 nfr
->cookie
= rule
->flow_cookie
;
4699 nfr
->priority
= htons(rule
->cr
.priority
);
4700 nfr
->reason
= reason
;
4701 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4702 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4703 nfr
->match_len
= htons(match_len
);
4704 nfr
->packet_count
= htonll(rule
->packet_count
);
4705 nfr
->byte_count
= htonll(rule
->byte_count
);
4711 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4713 struct ofconn
*ofconn
;
4715 if (!rule
->send_flow_removed
) {
4719 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4722 if (!rconn_is_connected(ofconn
->rconn
)
4723 || !ofconn_receives_async_msgs(ofconn
)) {
4727 msg
= (ofconn
->flow_format
== NXFF_NXM
4728 ? compose_nx_flow_removed(rule
, reason
)
4729 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4731 /* Account flow expirations under ofconn->reply_counter, the counter
4732 * for replies to OpenFlow requests. That works because preventing
4733 * OpenFlow requests from being processed also prevents new flows from
4734 * being added (and expiring). (It also prevents processing OpenFlow
4735 * requests that would not add new flows, so it is imperfect.) */
4736 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4740 /* pinsched callback for sending 'packet' on 'ofconn'. */
4742 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4744 struct ofconn
*ofconn
= ofconn_
;
4746 rconn_send_with_limit(ofconn
->rconn
, packet
,
4747 ofconn
->packet_in_counter
, 100);
4750 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4751 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4752 * packet scheduler for sending.
4754 * 'max_len' specifies the maximum number of bytes of the packet to send on
4755 * 'ofconn' (INT_MAX specifies no limit).
4757 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4758 * ownership is transferred to this function. */
4760 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4763 struct ofproto
*ofproto
= ofconn
->ofproto
;
4764 struct ofp_packet_in
*opi
= packet
->data
;
4765 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4766 int send_len
, trim_size
;
4770 if (opi
->reason
== OFPR_ACTION
) {
4771 buffer_id
= UINT32_MAX
;
4772 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4773 buffer_id
= pktbuf_get_null();
4774 } else if (!ofconn
->pktbuf
) {
4775 buffer_id
= UINT32_MAX
;
4777 struct ofpbuf payload
;
4779 ofpbuf_use_const(&payload
, opi
->data
,
4780 packet
->size
- offsetof(struct ofp_packet_in
, data
));
4781 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4784 /* Figure out how much of the packet to send. */
4785 send_len
= ntohs(opi
->total_len
);
4786 if (buffer_id
!= UINT32_MAX
) {
4787 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4789 send_len
= MIN(send_len
, max_len
);
4791 /* Adjust packet length and clone if necessary. */
4792 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4794 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4797 packet
->size
= trim_size
;
4800 /* Update packet headers. */
4801 opi
->buffer_id
= htonl(buffer_id
);
4802 update_openflow_length(packet
);
4804 /* Hand over to packet scheduler. It might immediately call into
4805 * do_send_packet_in() or it might buffer it for a while (until a later
4806 * call to pinsched_run()). */
4807 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4808 packet
, do_send_packet_in
, ofconn
);
4811 /* Replace struct odp_msg header in 'packet' by equivalent struct
4812 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4813 * returned by dpif_recv()).
4815 * The conversion is not complete: the caller still needs to trim any unneeded
4816 * payload off the end of the buffer, set the length in the OpenFlow header,
4817 * and set buffer_id. Those require us to know the controller settings and so
4818 * must be done on a per-controller basis.
4820 * Returns the maximum number of bytes of the packet that should be sent to
4821 * the controller (INT_MAX if no limit). */
4823 do_convert_to_packet_in(struct ofpbuf
*packet
)
4825 struct odp_msg
*msg
= packet
->data
;
4826 struct ofp_packet_in
*opi
;
4832 /* Extract relevant header fields */
4833 if (msg
->type
== _ODPL_ACTION_NR
) {
4834 reason
= OFPR_ACTION
;
4837 reason
= OFPR_NO_MATCH
;
4840 total_len
= msg
->length
- sizeof *msg
;
4841 in_port
= odp_port_to_ofp_port(msg
->port
);
4843 /* Repurpose packet buffer by overwriting header. */
4844 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4845 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4846 opi
->header
.version
= OFP_VERSION
;
4847 opi
->header
.type
= OFPT_PACKET_IN
;
4848 opi
->total_len
= htons(total_len
);
4849 opi
->in_port
= htons(in_port
);
4850 opi
->reason
= reason
;
4855 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4856 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4857 * as necessary according to their individual configurations.
4859 * 'packet' must have sufficient headroom to convert it into a struct
4860 * ofp_packet_in (e.g. as returned by dpif_recv()).
4862 * Takes ownership of 'packet'. */
4864 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4866 struct ofconn
*ofconn
, *prev
;
4869 max_len
= do_convert_to_packet_in(packet
);
4872 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4873 if (ofconn_receives_async_msgs(ofconn
)) {
4875 schedule_packet_in(prev
, packet
, max_len
, true);
4881 schedule_packet_in(prev
, packet
, max_len
, false);
4883 ofpbuf_delete(packet
);
4888 pick_datapath_id(const struct ofproto
*ofproto
)
4890 const struct ofport
*port
;
4892 port
= get_port(ofproto
, ODPP_LOCAL
);
4894 uint8_t ea
[ETH_ADDR_LEN
];
4897 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4899 return eth_addr_to_uint64(ea
);
4901 VLOG_WARN("could not get MAC address for %s (%s)",
4902 netdev_get_name(port
->netdev
), strerror(error
));
4904 return ofproto
->fallback_dpid
;
4908 pick_fallback_dpid(void)
4910 uint8_t ea
[ETH_ADDR_LEN
];
4911 eth_addr_nicira_random(ea
);
4912 return eth_addr_to_uint64(ea
);
4916 ofproto_unixctl_list(struct unixctl_conn
*conn
, const char *arg OVS_UNUSED
,
4917 void *aux OVS_UNUSED
)
4919 const struct shash_node
*node
;
4923 SHASH_FOR_EACH (node
, &all_ofprotos
) {
4924 ds_put_format(&results
, "%s\n", node
->name
);
4926 unixctl_command_reply(conn
, 200, ds_cstr(&results
));
4927 ds_destroy(&results
);
4930 struct ofproto_trace
{
4931 struct action_xlate_ctx ctx
;
4937 trace_format_rule(struct ds
*result
, int level
, const struct rule
*rule
)
4939 ds_put_char_multiple(result
, '\t', level
);
4941 ds_put_cstr(result
, "No match\n");
4945 ds_put_format(result
, "Rule: cookie=%#"PRIx64
" ",
4946 ntohll(rule
->flow_cookie
));
4947 cls_rule_format(&rule
->cr
, result
);
4948 ds_put_char(result
, '\n');
4950 ds_put_char_multiple(result
, '\t', level
);
4951 ds_put_cstr(result
, "OpenFlow ");
4952 ofp_print_actions(result
, (const struct ofp_action_header
*) rule
->actions
,
4953 rule
->n_actions
* sizeof *rule
->actions
);
4954 ds_put_char(result
, '\n');
4958 trace_format_flow(struct ds
*result
, int level
, const char *title
,
4959 struct ofproto_trace
*trace
)
4961 ds_put_char_multiple(result
, '\t', level
);
4962 ds_put_format(result
, "%s: ", title
);
4963 if (flow_equal(&trace
->ctx
.flow
, &trace
->flow
)) {
4964 ds_put_cstr(result
, "unchanged");
4966 flow_format(result
, &trace
->ctx
.flow
);
4967 trace
->flow
= trace
->ctx
.flow
;
4969 ds_put_char(result
, '\n');
4973 trace_resubmit(struct action_xlate_ctx
*ctx
, const struct rule
*rule
)
4975 struct ofproto_trace
*trace
= CONTAINER_OF(ctx
, struct ofproto_trace
, ctx
);
4976 struct ds
*result
= trace
->result
;
4978 ds_put_char(result
, '\n');
4979 trace_format_flow(result
, ctx
->recurse
+ 1, "Resubmitted flow", trace
);
4980 trace_format_rule(result
, ctx
->recurse
+ 1, rule
);
4984 ofproto_unixctl_trace(struct unixctl_conn
*conn
, const char *args_
,
4985 void *aux OVS_UNUSED
)
4987 char *dpname
, *in_port_s
, *tun_id_s
, *packet_s
;
4988 char *args
= xstrdup(args_
);
4989 char *save_ptr
= NULL
;
4990 struct ofproto
*ofproto
;
4991 struct ofpbuf packet
;
4999 ofpbuf_init(&packet
, strlen(args
) / 2);
5002 dpname
= strtok_r(args
, " ", &save_ptr
);
5003 tun_id_s
= strtok_r(NULL
, " ", &save_ptr
);
5004 in_port_s
= strtok_r(NULL
, " ", &save_ptr
);
5005 packet_s
= strtok_r(NULL
, "", &save_ptr
); /* Get entire rest of line. */
5006 if (!dpname
|| !in_port_s
|| !packet_s
) {
5007 unixctl_command_reply(conn
, 501, "Bad command syntax");
5011 ofproto
= shash_find_data(&all_ofprotos
, dpname
);
5013 unixctl_command_reply(conn
, 501, "Unknown ofproto (use ofproto/list "
5018 tun_id
= ntohl(strtoul(tun_id_s
, NULL
, 10));
5019 in_port
= ofp_port_to_odp_port(atoi(in_port_s
));
5021 packet_s
= ofpbuf_put_hex(&packet
, packet_s
, NULL
);
5022 packet_s
+= strspn(packet_s
, " ");
5023 if (*packet_s
!= '\0') {
5024 unixctl_command_reply(conn
, 501, "Trailing garbage in command");
5027 if (packet
.size
< ETH_HEADER_LEN
) {
5028 unixctl_command_reply(conn
, 501, "Packet data too short for Ethernet");
5032 ds_put_cstr(&result
, "Packet: ");
5033 s
= ofp_packet_to_string(packet
.data
, packet
.size
, packet
.size
);
5034 ds_put_cstr(&result
, s
);
5037 flow_extract(&packet
, tun_id
, in_port
, &flow
);
5038 ds_put_cstr(&result
, "Flow: ");
5039 flow_format(&result
, &flow
);
5040 ds_put_char(&result
, '\n');
5042 rule
= rule_lookup(ofproto
, &flow
);
5043 trace_format_rule(&result
, 0, rule
);
5045 struct ofproto_trace trace
;
5047 trace
.result
= &result
;
5049 action_xlate_ctx_init(&trace
.ctx
, ofproto
, &flow
, &packet
);
5050 trace
.ctx
.resubmit_hook
= trace_resubmit
;
5051 xlate_actions(&trace
.ctx
, rule
->actions
, rule
->n_actions
);
5053 ds_put_char(&result
, '\n');
5054 trace_format_flow(&result
, 0, "Final flow", &trace
);
5055 ds_put_cstr(&result
, "Datapath actions: ");
5056 format_odp_actions(&result
,
5057 trace
.ctx
.out
.actions
, trace
.ctx
.out
.n_actions
);
5060 unixctl_command_reply(conn
, 200, ds_cstr(&result
));
5063 ds_destroy(&result
);
5064 ofpbuf_uninit(&packet
);
5069 ofproto_unixctl_init(void)
5071 static bool registered
;
5077 unixctl_command_register("ofproto/list", ofproto_unixctl_list
, NULL
);
5078 unixctl_command_register("ofproto/trace", ofproto_unixctl_trace
, NULL
);
5082 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5083 struct odp_actions
*actions
, tag_type
*tags
,
5084 uint16_t *nf_output_iface
, void *ofproto_
)
5086 struct ofproto
*ofproto
= ofproto_
;
5089 /* Drop frames for reserved multicast addresses. */
5090 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5094 /* Learn source MAC (but don't try to learn from revalidation). */
5095 if (packet
!= NULL
) {
5096 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
5098 GRAT_ARP_LOCK_NONE
);
5100 /* The log messages here could actually be useful in debugging,
5101 * so keep the rate limit relatively high. */
5102 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5103 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5104 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5105 ofproto_revalidate(ofproto
, rev_tag
);
5109 /* Determine output port. */
5110 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
5113 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5114 nf_output_iface
, actions
);
5115 } else if (out_port
!= flow
->in_port
) {
5116 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
5117 *nf_output_iface
= out_port
;
5125 static const struct ofhooks default_ofhooks
= {
5126 default_normal_ofhook_cb
,