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"
94 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
95 struct netdev
*netdev
;
96 struct ofp_phy_port opp
; /* In host byte order. */
100 static void ofport_free(struct ofport
*);
101 static void hton_ofp_phy_port(struct ofp_phy_port
*);
103 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
104 const struct flow
*, struct ofproto
*,
105 const struct ofpbuf
*packet
,
106 struct odp_actions
*out
, tag_type
*tags
,
107 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
109 /* An OpenFlow flow. */
111 long long int used
; /* Time last used; time created if not used. */
112 long long int created
; /* Creation time. */
116 * - Do include packets and bytes from facets that have been deleted or
117 * whose own statistics have been folded into the rule.
119 * - Do include packets and bytes sent "by hand" that were accounted to
120 * the rule without any facet being involved (this is a rare corner
121 * case in rule_execute()).
123 * - Do not include packet or bytes that can be obtained from any facet's
124 * packet_count or byte_count member or that can be obtained from the
125 * datapath by, e.g., dpif_flow_get() for any facet.
127 uint64_t packet_count
; /* Number of packets received. */
128 uint64_t byte_count
; /* Number of bytes received. */
130 ovs_be64 flow_cookie
; /* Controller-issued identifier. */
132 struct cls_rule cr
; /* In owning ofproto's classifier. */
133 uint16_t idle_timeout
; /* In seconds from time of last use. */
134 uint16_t hard_timeout
; /* In seconds from time of creation. */
135 bool send_flow_removed
; /* Send a flow removed message? */
136 int n_actions
; /* Number of elements in actions[]. */
137 union ofp_action
*actions
; /* OpenFlow actions. */
138 struct list facets
; /* List of "struct facet"s. */
141 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
142 static bool rule_is_hidden(const struct rule
*);
144 static struct rule
*rule_create(const struct cls_rule
*,
145 const union ofp_action
*, size_t n_actions
,
146 uint16_t idle_timeout
, uint16_t hard_timeout
,
147 ovs_be64 flow_cookie
, bool send_flow_removed
);
148 static void rule_destroy(struct ofproto
*, struct rule
*);
149 static void rule_free(struct rule
*);
151 static struct rule
*rule_lookup(struct ofproto
*, const struct flow
*);
152 static void rule_insert(struct ofproto
*, struct rule
*);
153 static void rule_remove(struct ofproto
*, struct rule
*);
155 static void rule_send_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
157 /* An exact-match instantiation of an OpenFlow flow. */
159 long long int used
; /* Time last used; time created if not used. */
163 * - Do include packets and bytes sent "by hand", e.g. with
166 * - Do include packets and bytes that were obtained from the datapath
167 * when a flow was deleted (e.g. dpif_flow_del()) or when its
168 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
170 * - Do not include any packets or bytes that can currently be obtained
171 * from the datapath by, e.g., dpif_flow_get().
173 uint64_t packet_count
; /* Number of packets received. */
174 uint64_t byte_count
; /* Number of bytes received. */
176 /* Number of bytes passed to account_cb. This may include bytes that can
177 * currently obtained from the datapath (thus, it can be greater than
179 uint64_t accounted_bytes
;
181 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
182 struct list list_node
; /* In owning rule's 'facets' list. */
183 struct rule
*rule
; /* Owning rule. */
184 struct flow flow
; /* Exact-match flow. */
185 bool installed
; /* Installed in datapath? */
186 bool may_install
; /* True ordinarily; false if actions must
187 * be reassessed for every packet. */
188 int n_actions
; /* Number of elements in actions[]. */
189 union odp_action
*actions
; /* Datapath actions. */
190 tag_type tags
; /* Tags (set only by hooks). */
191 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
194 static struct facet
*facet_create(struct ofproto
*, struct rule
*,
196 const struct ofpbuf
*packet
);
197 static void facet_remove(struct ofproto
*, struct facet
*);
198 static void facet_free(struct facet
*);
200 static struct facet
*facet_lookup_valid(struct ofproto
*, const struct flow
*);
201 static bool facet_revalidate(struct ofproto
*, struct facet
*);
203 static void facet_install(struct ofproto
*, struct facet
*, bool zero_stats
);
204 static void facet_uninstall(struct ofproto
*, struct facet
*);
205 static void facet_flush_stats(struct ofproto
*, struct facet
*);
207 static void facet_make_actions(struct ofproto
*, struct facet
*,
208 const struct ofpbuf
*packet
);
209 static void facet_update_stats(struct ofproto
*, struct facet
*,
210 const struct odp_flow_stats
*);
212 /* ofproto supports two kinds of OpenFlow connections:
214 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
215 * maintains persistent connections to these controllers and by default
216 * sends them asynchronous messages such as packet-ins.
218 * - "Service" connections, e.g. from ovs-ofctl. When these connections
219 * drop, it is the other side's responsibility to reconnect them if
220 * necessary. ofproto does not send them asynchronous messages by default.
222 * Currently, active (tcp, ssl, unix) connections are always "primary"
223 * connections and passive (ptcp, pssl, punix) connections are always "service"
224 * connections. There is no inherent reason for this, but it reflects the
228 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
229 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
232 /* A listener for incoming OpenFlow "service" connections. */
234 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
235 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
237 /* These are not used by ofservice directly. They are settings for
238 * accepted "struct ofconn"s from the pvconn. */
239 int probe_interval
; /* Max idle time before probing, in seconds. */
240 int rate_limit
; /* Max packet-in rate in packets per second. */
241 int burst_limit
; /* Limit on accumulating packet credits. */
244 static struct ofservice
*ofservice_lookup(struct ofproto
*,
246 static int ofservice_create(struct ofproto
*,
247 const struct ofproto_controller
*);
248 static void ofservice_reconfigure(struct ofservice
*,
249 const struct ofproto_controller
*);
250 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
252 /* An OpenFlow connection. */
254 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
255 struct list node
; /* In struct ofproto's "all_conns" list. */
256 struct rconn
*rconn
; /* OpenFlow connection. */
257 enum ofconn_type type
; /* Type. */
258 enum nx_flow_format flow_format
; /* Currently selected flow format. */
260 /* OFPT_PACKET_IN related data. */
261 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
262 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
263 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
264 int miss_send_len
; /* Bytes to send of buffered packets. */
266 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
267 * requests, and the maximum number before we stop reading OpenFlow
269 #define OFCONN_REPLY_MAX 100
270 struct rconn_packet_counter
*reply_counter
;
272 /* type == OFCONN_PRIMARY only. */
273 enum nx_role role
; /* Role. */
274 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
275 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
276 struct status_category
*ss
; /* Switch status category. */
277 enum ofproto_band band
; /* In-band or out-of-band? */
280 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
281 * "schedulers" array. Their values are 0 and 1, and their meanings and values
282 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
283 * case anything ever changes, check their values here. */
284 #define N_SCHEDULERS 2
285 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
286 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
287 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
288 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
290 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
292 static void ofconn_destroy(struct ofconn
*);
293 static void ofconn_run(struct ofconn
*);
294 static void ofconn_wait(struct ofconn
*);
295 static bool ofconn_receives_async_msgs(const struct ofconn
*);
296 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
297 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
299 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
300 struct rconn_packet_counter
*counter
);
302 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
303 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
307 uint64_t datapath_id
; /* Datapath ID. */
308 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
309 char *mfr_desc
; /* Manufacturer. */
310 char *hw_desc
; /* Hardware. */
311 char *sw_desc
; /* Software version. */
312 char *serial_desc
; /* Serial number. */
313 char *dp_desc
; /* Datapath description. */
317 struct netdev_monitor
*netdev_monitor
;
318 struct hmap ports
; /* Contains "struct ofport"s. */
319 struct shash port_by_name
;
323 struct switch_status
*switch_status
;
324 struct fail_open
*fail_open
;
325 struct netflow
*netflow
;
326 struct ofproto_sflow
*sflow
;
328 /* In-band control. */
329 struct in_band
*in_band
;
330 long long int next_in_band_update
;
331 struct sockaddr_in
*extra_in_band_remotes
;
332 size_t n_extra_remotes
;
336 struct classifier cls
;
337 long long int next_expiration
;
341 bool need_revalidate
;
342 struct tag_set revalidate_set
;
344 /* OpenFlow connections. */
345 struct hmap controllers
; /* Controller "struct ofconn"s. */
346 struct list all_conns
; /* Contains "struct ofconn"s. */
347 enum ofproto_fail_mode fail_mode
;
349 /* OpenFlow listeners. */
350 struct hmap services
; /* Contains "struct ofservice"s. */
351 struct pvconn
**snoops
;
354 /* Hooks for ovs-vswitchd. */
355 const struct ofhooks
*ofhooks
;
358 /* Used by default ofhooks. */
359 struct mac_learning
*ml
;
362 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
364 static const struct ofhooks default_ofhooks
;
366 static uint64_t pick_datapath_id(const struct ofproto
*);
367 static uint64_t pick_fallback_dpid(void);
369 static int ofproto_expire(struct ofproto
*);
371 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
373 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
375 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
376 static void update_port(struct ofproto
*, const char *devname
);
377 static int init_ports(struct ofproto
*);
378 static void reinit_ports(struct ofproto
*);
381 ofproto_create(const char *datapath
, const char *datapath_type
,
382 const struct ofhooks
*ofhooks
, void *aux
,
383 struct ofproto
**ofprotop
)
385 struct odp_stats stats
;
392 /* Connect to datapath and start listening for messages. */
393 error
= dpif_open(datapath
, datapath_type
, &dpif
);
395 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
398 error
= dpif_get_dp_stats(dpif
, &stats
);
400 VLOG_ERR("failed to obtain stats for datapath %s: %s",
401 datapath
, strerror(error
));
405 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
407 VLOG_ERR("failed to listen on datapath %s: %s",
408 datapath
, strerror(error
));
412 dpif_flow_flush(dpif
);
413 dpif_recv_purge(dpif
);
415 /* Initialize settings. */
416 p
= xzalloc(sizeof *p
);
417 p
->fallback_dpid
= pick_fallback_dpid();
418 p
->datapath_id
= p
->fallback_dpid
;
419 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
420 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
421 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
422 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
423 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
425 /* Initialize datapath. */
427 p
->netdev_monitor
= netdev_monitor_create();
428 hmap_init(&p
->ports
);
429 shash_init(&p
->port_by_name
);
430 p
->max_ports
= stats
.max_ports
;
432 /* Initialize submodules. */
433 p
->switch_status
= switch_status_create(p
);
438 /* Initialize in-band control. */
440 p
->in_band_queue
= -1;
442 /* Initialize flow table. */
443 classifier_init(&p
->cls
);
444 p
->next_expiration
= time_msec() + 1000;
446 /* Initialize facet table. */
447 hmap_init(&p
->facets
);
448 p
->need_revalidate
= false;
449 tag_set_init(&p
->revalidate_set
);
451 /* Initialize OpenFlow connections. */
452 list_init(&p
->all_conns
);
453 hmap_init(&p
->controllers
);
454 hmap_init(&p
->services
);
458 /* Initialize hooks. */
460 p
->ofhooks
= ofhooks
;
464 p
->ofhooks
= &default_ofhooks
;
466 p
->ml
= mac_learning_create();
469 /* Pick final datapath ID. */
470 p
->datapath_id
= pick_datapath_id(p
);
471 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
478 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
480 uint64_t old_dpid
= p
->datapath_id
;
481 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
482 if (p
->datapath_id
!= old_dpid
) {
483 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
485 /* Force all active connections to reconnect, since there is no way to
486 * notify a controller that the datapath ID has changed. */
487 ofproto_reconnect_controllers(p
);
492 is_discovery_controller(const struct ofproto_controller
*c
)
494 return !strcmp(c
->target
, "discover");
498 is_in_band_controller(const struct ofproto_controller
*c
)
500 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
503 /* Creates a new controller in 'ofproto'. Some of the settings are initially
504 * drawn from 'c', but update_controller() needs to be called later to finish
505 * the new ofconn's configuration. */
507 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
509 struct discovery
*discovery
;
510 struct ofconn
*ofconn
;
512 if (is_discovery_controller(c
)) {
513 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
514 ofproto
->dpif
, ofproto
->switch_status
,
523 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
524 ofconn
->pktbuf
= pktbuf_create();
525 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
527 ofconn
->discovery
= discovery
;
529 char *name
= ofconn_make_name(ofproto
, c
->target
);
530 rconn_connect(ofconn
->rconn
, c
->target
, name
);
533 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
534 hash_string(c
->target
, 0));
537 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
538 * target or turn discovery on or off (these are done by creating new ofconns
539 * and deleting old ones), but it can update the rest of an ofconn's
542 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
546 ofconn
->band
= (is_in_band_controller(c
)
547 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
549 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
551 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
552 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
554 if (ofconn
->discovery
) {
555 discovery_set_update_resolv_conf(ofconn
->discovery
,
556 c
->update_resolv_conf
);
557 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
560 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
564 ofconn_get_target(const struct ofconn
*ofconn
)
566 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
569 static struct ofconn
*
570 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
572 struct ofconn
*ofconn
;
574 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
575 hash_string(target
, 0), &ofproto
->controllers
) {
576 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
584 update_in_band_remotes(struct ofproto
*ofproto
)
586 const struct ofconn
*ofconn
;
587 struct sockaddr_in
*addrs
;
588 size_t max_addrs
, n_addrs
;
592 /* Allocate enough memory for as many remotes as we could possibly have. */
593 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
594 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
597 /* Add all the remotes. */
599 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
600 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
602 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
606 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
607 if (sin
->sin_addr
.s_addr
) {
608 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
611 if (ofconn
->discovery
) {
615 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
616 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
619 /* Create or update or destroy in-band.
621 * Ordinarily we only enable in-band if there's at least one remote
622 * address, but discovery needs the in-band rules for DHCP to be installed
623 * even before we know any remote addresses. */
624 if (n_addrs
|| discovery
) {
625 if (!ofproto
->in_band
) {
626 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
629 if (ofproto
->in_band
) {
630 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
632 in_band_set_queue(ofproto
->in_band
, ofproto
->in_band_queue
);
633 ofproto
->next_in_band_update
= time_msec() + 1000;
635 in_band_destroy(ofproto
->in_band
);
636 ofproto
->in_band
= NULL
;
644 update_fail_open(struct ofproto
*p
)
646 struct ofconn
*ofconn
;
648 if (!hmap_is_empty(&p
->controllers
)
649 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
650 struct rconn
**rconns
;
654 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
658 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
659 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
660 rconns
[n
++] = ofconn
->rconn
;
663 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
664 /* p->fail_open takes ownership of 'rconns'. */
666 fail_open_destroy(p
->fail_open
);
672 ofproto_set_controllers(struct ofproto
*p
,
673 const struct ofproto_controller
*controllers
,
674 size_t n_controllers
)
676 struct shash new_controllers
;
677 struct ofconn
*ofconn
, *next_ofconn
;
678 struct ofservice
*ofservice
, *next_ofservice
;
682 /* Create newly configured controllers and services.
683 * Create a name to ofproto_controller mapping in 'new_controllers'. */
684 shash_init(&new_controllers
);
685 for (i
= 0; i
< n_controllers
; i
++) {
686 const struct ofproto_controller
*c
= &controllers
[i
];
688 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
689 if (!find_controller_by_target(p
, c
->target
)) {
690 add_controller(p
, c
);
692 } else if (!pvconn_verify_name(c
->target
)) {
693 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
697 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
698 dpif_name(p
->dpif
), c
->target
);
702 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
705 /* Delete controllers that are no longer configured.
706 * Update configuration of all now-existing controllers. */
708 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
709 struct ofproto_controller
*c
;
711 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
713 ofconn_destroy(ofconn
);
715 update_controller(ofconn
, c
);
722 /* Delete services that are no longer configured.
723 * Update configuration of all now-existing services. */
724 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
725 struct ofproto_controller
*c
;
727 c
= shash_find_data(&new_controllers
,
728 pvconn_get_name(ofservice
->pvconn
));
730 ofservice_destroy(p
, ofservice
);
732 ofservice_reconfigure(ofservice
, c
);
736 shash_destroy(&new_controllers
);
738 update_in_band_remotes(p
);
741 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
742 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
743 struct ofconn
, hmap_node
);
744 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
745 rconn_status_cb
, ofconn
->rconn
);
750 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
752 p
->fail_mode
= fail_mode
;
756 /* Drops the connections between 'ofproto' and all of its controllers, forcing
757 * them to reconnect. */
759 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
761 struct ofconn
*ofconn
;
763 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
764 rconn_reconnect(ofconn
->rconn
);
769 any_extras_changed(const struct ofproto
*ofproto
,
770 const struct sockaddr_in
*extras
, size_t n
)
774 if (n
!= ofproto
->n_extra_remotes
) {
778 for (i
= 0; i
< n
; i
++) {
779 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
780 const struct sockaddr_in
*new = &extras
[i
];
782 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
783 old
->sin_port
!= new->sin_port
) {
791 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
792 * in-band control should guarantee access, in the same way that in-band
793 * control guarantees access to OpenFlow controllers. */
795 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
796 const struct sockaddr_in
*extras
, size_t n
)
798 if (!any_extras_changed(ofproto
, extras
, n
)) {
802 free(ofproto
->extra_in_band_remotes
);
803 ofproto
->n_extra_remotes
= n
;
804 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
806 update_in_band_remotes(ofproto
);
809 /* Sets the OpenFlow queue used by flows set up by in-band control on
810 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
811 * flows will use the default queue. */
813 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
815 if (queue_id
!= ofproto
->in_band_queue
) {
816 ofproto
->in_band_queue
= queue_id
;
817 update_in_band_remotes(ofproto
);
822 ofproto_set_desc(struct ofproto
*p
,
823 const char *mfr_desc
, const char *hw_desc
,
824 const char *sw_desc
, const char *serial_desc
,
827 struct ofp_desc_stats
*ods
;
830 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
831 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
832 sizeof ods
->mfr_desc
);
835 p
->mfr_desc
= xstrdup(mfr_desc
);
838 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
839 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
840 sizeof ods
->hw_desc
);
843 p
->hw_desc
= xstrdup(hw_desc
);
846 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
847 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
848 sizeof ods
->sw_desc
);
851 p
->sw_desc
= xstrdup(sw_desc
);
854 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
855 VLOG_WARN("truncating serial_desc, must be less than %zu "
857 sizeof ods
->serial_num
);
859 free(p
->serial_desc
);
860 p
->serial_desc
= xstrdup(serial_desc
);
863 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
864 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
865 sizeof ods
->dp_desc
);
868 p
->dp_desc
= xstrdup(dp_desc
);
873 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
874 const struct svec
*svec
)
876 struct pvconn
**pvconns
= *pvconnsp
;
877 size_t n_pvconns
= *n_pvconnsp
;
881 for (i
= 0; i
< n_pvconns
; i
++) {
882 pvconn_close(pvconns
[i
]);
886 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
888 for (i
= 0; i
< svec
->n
; i
++) {
889 const char *name
= svec
->names
[i
];
890 struct pvconn
*pvconn
;
893 error
= pvconn_open(name
, &pvconn
);
895 pvconns
[n_pvconns
++] = pvconn
;
897 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
905 *n_pvconnsp
= n_pvconns
;
911 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
913 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
917 ofproto_set_netflow(struct ofproto
*ofproto
,
918 const struct netflow_options
*nf_options
)
920 if (nf_options
&& nf_options
->collectors
.n
) {
921 if (!ofproto
->netflow
) {
922 ofproto
->netflow
= netflow_create();
924 return netflow_set_options(ofproto
->netflow
, nf_options
);
926 netflow_destroy(ofproto
->netflow
);
927 ofproto
->netflow
= NULL
;
933 ofproto_set_sflow(struct ofproto
*ofproto
,
934 const struct ofproto_sflow_options
*oso
)
936 struct ofproto_sflow
*os
= ofproto
->sflow
;
939 struct ofport
*ofport
;
941 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
942 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
943 ofproto_sflow_add_port(os
, ofport
->odp_port
,
944 netdev_get_name(ofport
->netdev
));
947 ofproto_sflow_set_options(os
, oso
);
949 ofproto_sflow_destroy(os
);
950 ofproto
->sflow
= NULL
;
955 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
957 return ofproto
->datapath_id
;
961 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
963 return !hmap_is_empty(&ofproto
->controllers
);
966 enum ofproto_fail_mode
967 ofproto_get_fail_mode(const struct ofproto
*p
)
973 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
977 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
978 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
983 ofproto_destroy(struct ofproto
*p
)
985 struct ofservice
*ofservice
, *next_ofservice
;
986 struct ofconn
*ofconn
, *next_ofconn
;
987 struct ofport
*ofport
, *next_ofport
;
994 /* Destroy fail-open and in-band early, since they touch the classifier. */
995 fail_open_destroy(p
->fail_open
);
998 in_band_destroy(p
->in_band
);
1000 free(p
->extra_in_band_remotes
);
1002 ofproto_flush_flows(p
);
1003 classifier_destroy(&p
->cls
);
1004 hmap_destroy(&p
->facets
);
1006 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1007 ofconn_destroy(ofconn
);
1009 hmap_destroy(&p
->controllers
);
1011 dpif_close(p
->dpif
);
1012 netdev_monitor_destroy(p
->netdev_monitor
);
1013 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
1014 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1015 ofport_free(ofport
);
1017 shash_destroy(&p
->port_by_name
);
1019 switch_status_destroy(p
->switch_status
);
1020 netflow_destroy(p
->netflow
);
1021 ofproto_sflow_destroy(p
->sflow
);
1023 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
1024 ofservice_destroy(p
, ofservice
);
1026 hmap_destroy(&p
->services
);
1028 for (i
= 0; i
< p
->n_snoops
; i
++) {
1029 pvconn_close(p
->snoops
[i
]);
1033 mac_learning_destroy(p
->ml
);
1038 free(p
->serial_desc
);
1041 hmap_destroy(&p
->ports
);
1047 ofproto_run(struct ofproto
*p
)
1049 int error
= ofproto_run1(p
);
1051 error
= ofproto_run2(p
, false);
1057 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1059 if (error
== ENOBUFS
) {
1060 reinit_ports(ofproto
);
1061 } else if (!error
) {
1062 update_port(ofproto
, devname
);
1067 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1068 * means that 'ofconn' is more interesting for monitoring than a lower return
1071 snoop_preference(const struct ofconn
*ofconn
)
1073 switch (ofconn
->role
) {
1074 case NX_ROLE_MASTER
:
1081 /* Shouldn't happen. */
1086 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1087 * Connects this vconn to a controller. */
1089 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1091 struct ofconn
*ofconn
, *best
;
1093 /* Pick a controller for monitoring. */
1095 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1096 if (ofconn
->type
== OFCONN_PRIMARY
1097 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1103 rconn_add_monitor(best
->rconn
, vconn
);
1105 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1111 ofproto_run1(struct ofproto
*p
)
1113 struct ofconn
*ofconn
, *next_ofconn
;
1114 struct ofservice
*ofservice
;
1119 if (shash_is_empty(&p
->port_by_name
)) {
1123 for (i
= 0; i
< 50; i
++) {
1126 error
= dpif_recv(p
->dpif
, &buf
);
1128 if (error
== ENODEV
) {
1129 /* Someone destroyed the datapath behind our back. The caller
1130 * better destroy us and give up, because we're just going to
1131 * spin from here on out. */
1132 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1133 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1134 dpif_name(p
->dpif
));
1140 handle_odp_msg(p
, buf
);
1143 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1144 process_port_change(p
, error
, devname
);
1146 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1147 &devname
)) != EAGAIN
) {
1148 process_port_change(p
, error
, devname
);
1152 if (time_msec() >= p
->next_in_band_update
) {
1153 update_in_band_remotes(p
);
1155 in_band_run(p
->in_band
);
1158 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1162 /* Fail-open maintenance. Do this after processing the ofconns since
1163 * fail-open checks the status of the controller rconn. */
1165 fail_open_run(p
->fail_open
);
1168 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1169 struct vconn
*vconn
;
1172 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1174 struct rconn
*rconn
;
1177 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1178 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1179 rconn_connect_unreliably(rconn
, vconn
, name
);
1182 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1183 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1184 ofservice
->burst_limit
);
1185 } else if (retval
!= EAGAIN
) {
1186 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1190 for (i
= 0; i
< p
->n_snoops
; i
++) {
1191 struct vconn
*vconn
;
1194 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1196 add_snooper(p
, vconn
);
1197 } else if (retval
!= EAGAIN
) {
1198 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1202 if (time_msec() >= p
->next_expiration
) {
1203 int delay
= ofproto_expire(p
);
1204 p
->next_expiration
= time_msec() + delay
;
1205 COVERAGE_INC(ofproto_expiration
);
1209 netflow_run(p
->netflow
);
1212 ofproto_sflow_run(p
->sflow
);
1219 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1221 /* Figure out what we need to revalidate now, if anything. */
1222 struct tag_set revalidate_set
= p
->revalidate_set
;
1223 if (p
->need_revalidate
) {
1224 revalidate_all
= true;
1227 /* Clear the revalidation flags. */
1228 tag_set_init(&p
->revalidate_set
);
1229 p
->need_revalidate
= false;
1231 /* Now revalidate if there's anything to do. */
1232 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1233 struct facet
*facet
, *next
;
1235 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1237 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1238 facet_revalidate(p
, facet
);
1247 ofproto_wait(struct ofproto
*p
)
1249 struct ofservice
*ofservice
;
1250 struct ofconn
*ofconn
;
1253 dpif_recv_wait(p
->dpif
);
1254 dpif_port_poll_wait(p
->dpif
);
1255 netdev_monitor_poll_wait(p
->netdev_monitor
);
1256 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1257 ofconn_wait(ofconn
);
1260 poll_timer_wait_until(p
->next_in_band_update
);
1261 in_band_wait(p
->in_band
);
1264 fail_open_wait(p
->fail_open
);
1267 ofproto_sflow_wait(p
->sflow
);
1269 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1270 poll_immediate_wake();
1272 if (p
->need_revalidate
) {
1273 /* Shouldn't happen, but if it does just go around again. */
1274 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1275 poll_immediate_wake();
1276 } else if (p
->next_expiration
!= LLONG_MAX
) {
1277 poll_timer_wait_until(p
->next_expiration
);
1279 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1280 pvconn_wait(ofservice
->pvconn
);
1282 for (i
= 0; i
< p
->n_snoops
; i
++) {
1283 pvconn_wait(p
->snoops
[i
]);
1288 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1290 tag_set_add(&ofproto
->revalidate_set
, tag
);
1294 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1296 return &ofproto
->revalidate_set
;
1300 ofproto_is_alive(const struct ofproto
*p
)
1302 return !hmap_is_empty(&p
->controllers
);
1305 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1307 * This is almost the same as calling dpif_port_del() directly on the
1308 * datapath, but it also makes 'ofproto' close its open netdev for the port
1309 * (if any). This makes it possible to create a new netdev of a different
1310 * type under the same name, which otherwise the netdev library would refuse
1311 * to do because of the conflict. (The netdev would eventually get closed on
1312 * the next trip through ofproto_run(), but this interface is more direct.)
1314 * Returns 0 if successful, otherwise a positive errno. */
1316 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1318 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1319 const char *name
= ofport
? ofport
->opp
.name
: "<unknown>";
1322 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1324 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1325 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1326 } else if (ofport
) {
1327 /* 'name' is ofport->opp.name and update_port() is going to destroy
1328 * 'ofport'. Just in case update_port() refers to 'name' after it
1329 * destroys 'ofport', make a copy of it around the update_port()
1331 char *devname
= xstrdup(name
);
1332 update_port(ofproto
, devname
);
1338 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1339 * true if 'odp_port' exists and should be included, false otherwise. */
1341 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1343 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1344 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1348 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1349 const union ofp_action
*actions
, size_t n_actions
,
1350 const struct ofpbuf
*packet
)
1352 struct odp_actions odp_actions
;
1355 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1361 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1363 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
, packet
);
1367 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1368 * performs the 'n_actions' actions in 'actions'. The new flow will not
1371 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1372 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1373 * controllers; otherwise, it will be hidden.
1375 * The caller retains ownership of 'cls_rule' and 'actions'. */
1377 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1378 const union ofp_action
*actions
, size_t n_actions
)
1381 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1382 rule_insert(p
, rule
);
1386 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1390 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1393 rule_remove(ofproto
, rule
);
1398 ofproto_flush_flows(struct ofproto
*ofproto
)
1400 struct facet
*facet
, *next_facet
;
1401 struct rule
*rule
, *next_rule
;
1402 struct cls_cursor cursor
;
1404 COVERAGE_INC(ofproto_flush
);
1406 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1407 /* Mark the facet as not installed so that facet_remove() doesn't
1408 * bother trying to uninstall it. There is no point in uninstalling it
1409 * individually since we are about to blow away all the facets with
1410 * dpif_flow_flush(). */
1411 facet
->installed
= false;
1412 facet_remove(ofproto
, facet
);
1415 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1416 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1417 rule_remove(ofproto
, rule
);
1420 dpif_flow_flush(ofproto
->dpif
);
1421 if (ofproto
->in_band
) {
1422 in_band_flushed(ofproto
->in_band
);
1424 if (ofproto
->fail_open
) {
1425 fail_open_flushed(ofproto
->fail_open
);
1430 reinit_ports(struct ofproto
*p
)
1432 struct svec devnames
;
1433 struct ofport
*ofport
;
1434 struct odp_port
*odp_ports
;
1438 COVERAGE_INC(ofproto_reinit_ports
);
1440 svec_init(&devnames
);
1441 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1442 svec_add (&devnames
, ofport
->opp
.name
);
1444 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1445 for (i
= 0; i
< n_odp_ports
; i
++) {
1446 svec_add (&devnames
, odp_ports
[i
].devname
);
1450 svec_sort_unique(&devnames
);
1451 for (i
= 0; i
< devnames
.n
; i
++) {
1452 update_port(p
, devnames
.names
[i
]);
1454 svec_destroy(&devnames
);
1457 static struct ofport
*
1458 make_ofport(const struct odp_port
*odp_port
)
1460 struct netdev_options netdev_options
;
1461 enum netdev_flags flags
;
1462 struct ofport
*ofport
;
1463 struct netdev
*netdev
;
1466 memset(&netdev_options
, 0, sizeof netdev_options
);
1467 netdev_options
.name
= odp_port
->devname
;
1468 netdev_options
.type
= odp_port
->type
;
1469 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1471 error
= netdev_open(&netdev_options
, &netdev
);
1473 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1474 "cannot be opened (%s)",
1475 odp_port
->devname
, odp_port
->port
,
1476 odp_port
->devname
, strerror(error
));
1480 ofport
= xmalloc(sizeof *ofport
);
1481 ofport
->netdev
= netdev
;
1482 ofport
->odp_port
= odp_port
->port
;
1483 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1484 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1485 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1486 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1487 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1489 netdev_get_flags(netdev
, &flags
);
1490 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1492 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1494 netdev_get_features(netdev
,
1495 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1496 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1501 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1503 if (get_port(p
, odp_port
->port
)) {
1504 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1507 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1508 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1517 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1519 const struct ofp_phy_port
*a
= &a_
->opp
;
1520 const struct ofp_phy_port
*b
= &b_
->opp
;
1522 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1523 return (a
->port_no
== b
->port_no
1524 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1525 && !strcmp(a
->name
, b
->name
)
1526 && a
->state
== b
->state
1527 && a
->config
== b
->config
1528 && a
->curr
== b
->curr
1529 && a
->advertised
== b
->advertised
1530 && a
->supported
== b
->supported
1531 && a
->peer
== b
->peer
);
1535 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1538 /* XXX Should limit the number of queued port status change messages. */
1539 struct ofconn
*ofconn
;
1540 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1541 struct ofp_port_status
*ops
;
1544 /* Primary controllers, even slaves, should always get port status
1545 updates. Otherwise obey ofconn_receives_async_msgs(). */
1546 if (ofconn
->type
!= OFCONN_PRIMARY
1547 && !ofconn_receives_async_msgs(ofconn
)) {
1551 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1552 ops
->reason
= reason
;
1553 ops
->desc
= ofport
->opp
;
1554 hton_ofp_phy_port(&ops
->desc
);
1555 queue_tx(b
, ofconn
, NULL
);
1560 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1562 const char *netdev_name
= ofport
->opp
.name
;
1564 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1565 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1566 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1568 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1573 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1575 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1576 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1577 shash_delete(&p
->port_by_name
,
1578 shash_find(&p
->port_by_name
, ofport
->opp
.name
));
1580 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1585 ofport_free(struct ofport
*ofport
)
1588 netdev_close(ofport
->netdev
);
1593 static struct ofport
*
1594 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1596 struct ofport
*port
;
1598 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1599 hash_int(odp_port
, 0), &ofproto
->ports
) {
1600 if (port
->odp_port
== odp_port
) {
1608 update_port(struct ofproto
*p
, const char *devname
)
1610 struct odp_port odp_port
;
1611 struct ofport
*old_ofport
;
1612 struct ofport
*new_ofport
;
1615 COVERAGE_INC(ofproto_update_port
);
1617 /* Query the datapath for port information. */
1618 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1620 /* Find the old ofport. */
1621 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1624 /* There's no port named 'devname' but there might be a port with
1625 * the same port number. This could happen if a port is deleted
1626 * and then a new one added in its place very quickly, or if a port
1627 * is renamed. In the former case we want to send an OFPPR_DELETE
1628 * and an OFPPR_ADD, and in the latter case we want to send a
1629 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1630 * the old port's ifindex against the new port, or perhaps less
1631 * reliably but more portably by comparing the old port's MAC
1632 * against the new port's MAC. However, this code isn't that smart
1633 * and always sends an OFPPR_MODIFY (XXX). */
1634 old_ofport
= get_port(p
, odp_port
.port
);
1636 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1637 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1638 "%s", strerror(error
));
1642 /* Create a new ofport. */
1643 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1645 /* Eliminate a few pathological cases. */
1646 if (!old_ofport
&& !new_ofport
) {
1648 } else if (old_ofport
&& new_ofport
) {
1649 /* Most of the 'config' bits are OpenFlow soft state, but
1650 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1651 * OpenFlow bits from old_ofport. (make_ofport() only sets
1652 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1653 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1655 if (ofport_equal(old_ofport
, new_ofport
)) {
1656 /* False alarm--no change. */
1657 ofport_free(new_ofport
);
1662 /* Now deal with the normal cases. */
1664 ofport_remove(p
, old_ofport
);
1667 ofport_install(p
, new_ofport
);
1669 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1670 (!old_ofport
? OFPPR_ADD
1671 : !new_ofport
? OFPPR_DELETE
1673 ofport_free(old_ofport
);
1677 init_ports(struct ofproto
*p
)
1679 struct odp_port
*ports
;
1684 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1689 for (i
= 0; i
< n_ports
; i
++) {
1690 const struct odp_port
*odp_port
= &ports
[i
];
1691 if (!ofport_conflicts(p
, odp_port
)) {
1692 struct ofport
*ofport
= make_ofport(odp_port
);
1694 ofport_install(p
, ofport
);
1702 static struct ofconn
*
1703 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1705 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1706 ofconn
->ofproto
= p
;
1707 list_push_back(&p
->all_conns
, &ofconn
->node
);
1708 ofconn
->rconn
= rconn
;
1709 ofconn
->type
= type
;
1710 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1711 ofconn
->role
= NX_ROLE_OTHER
;
1712 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1713 ofconn
->pktbuf
= NULL
;
1714 ofconn
->miss_send_len
= 0;
1715 ofconn
->reply_counter
= rconn_packet_counter_create ();
1720 ofconn_destroy(struct ofconn
*ofconn
)
1722 if (ofconn
->type
== OFCONN_PRIMARY
) {
1723 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1725 discovery_destroy(ofconn
->discovery
);
1727 list_remove(&ofconn
->node
);
1728 switch_status_unregister(ofconn
->ss
);
1729 rconn_destroy(ofconn
->rconn
);
1730 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1731 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1732 pktbuf_destroy(ofconn
->pktbuf
);
1737 ofconn_run(struct ofconn
*ofconn
)
1739 struct ofproto
*p
= ofconn
->ofproto
;
1743 if (ofconn
->discovery
) {
1744 char *controller_name
;
1745 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1746 discovery_question_connectivity(ofconn
->discovery
);
1748 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1749 if (controller_name
) {
1750 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1751 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1754 rconn_disconnect(ofconn
->rconn
);
1759 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1760 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1763 rconn_run(ofconn
->rconn
);
1765 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1766 /* Limit the number of iterations to prevent other tasks from
1768 for (iteration
= 0; iteration
< 50; iteration
++) {
1769 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1774 fail_open_maybe_recover(p
->fail_open
);
1776 handle_openflow(ofconn
, of_msg
);
1777 ofpbuf_delete(of_msg
);
1781 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1782 ofconn_destroy(ofconn
);
1787 ofconn_wait(struct ofconn
*ofconn
)
1791 if (ofconn
->discovery
) {
1792 discovery_wait(ofconn
->discovery
);
1794 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1795 pinsched_wait(ofconn
->schedulers
[i
]);
1797 rconn_run_wait(ofconn
->rconn
);
1798 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1799 rconn_recv_wait(ofconn
->rconn
);
1801 COVERAGE_INC(ofproto_ofconn_stuck
);
1805 /* Returns true if 'ofconn' should receive asynchronous messages. */
1807 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1809 if (ofconn
->type
== OFCONN_PRIMARY
) {
1810 /* Primary controllers always get asynchronous messages unless they
1811 * have configured themselves as "slaves". */
1812 return ofconn
->role
!= NX_ROLE_SLAVE
;
1814 /* Service connections don't get asynchronous messages unless they have
1815 * explicitly asked for them by setting a nonzero miss send length. */
1816 return ofconn
->miss_send_len
> 0;
1820 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1821 * and 'target', suitable for use in log messages for identifying the
1824 * The name is dynamically allocated. The caller should free it (with free())
1825 * when it is no longer needed. */
1827 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1829 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1833 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1837 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1838 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1842 *s
= pinsched_create(rate
, burst
,
1843 ofconn
->ofproto
->switch_status
);
1845 pinsched_set_limits(*s
, rate
, burst
);
1848 pinsched_destroy(*s
);
1855 ofservice_reconfigure(struct ofservice
*ofservice
,
1856 const struct ofproto_controller
*c
)
1858 ofservice
->probe_interval
= c
->probe_interval
;
1859 ofservice
->rate_limit
= c
->rate_limit
;
1860 ofservice
->burst_limit
= c
->burst_limit
;
1863 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1864 * positive errno value. */
1866 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1868 struct ofservice
*ofservice
;
1869 struct pvconn
*pvconn
;
1872 error
= pvconn_open(c
->target
, &pvconn
);
1877 ofservice
= xzalloc(sizeof *ofservice
);
1878 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1879 hash_string(c
->target
, 0));
1880 ofservice
->pvconn
= pvconn
;
1882 ofservice_reconfigure(ofservice
, c
);
1888 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1890 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1891 pvconn_close(ofservice
->pvconn
);
1895 /* Finds and returns the ofservice within 'ofproto' that has the given
1896 * 'target', or a null pointer if none exists. */
1897 static struct ofservice
*
1898 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1900 struct ofservice
*ofservice
;
1902 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1903 &ofproto
->services
) {
1904 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1911 /* Returns true if 'rule' should be hidden from the controller.
1913 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1914 * (e.g. by in-band control) and are intentionally hidden from the
1917 rule_is_hidden(const struct rule
*rule
)
1919 return rule
->cr
.priority
> UINT16_MAX
;
1922 /* Creates and returns a new rule initialized as specified.
1924 * The caller is responsible for inserting the rule into the classifier (with
1925 * rule_insert()). */
1926 static struct rule
*
1927 rule_create(const struct cls_rule
*cls_rule
,
1928 const union ofp_action
*actions
, size_t n_actions
,
1929 uint16_t idle_timeout
, uint16_t hard_timeout
,
1930 ovs_be64 flow_cookie
, bool send_flow_removed
)
1932 struct rule
*rule
= xzalloc(sizeof *rule
);
1933 rule
->cr
= *cls_rule
;
1934 rule
->idle_timeout
= idle_timeout
;
1935 rule
->hard_timeout
= hard_timeout
;
1936 rule
->flow_cookie
= flow_cookie
;
1937 rule
->used
= rule
->created
= time_msec();
1938 rule
->send_flow_removed
= send_flow_removed
;
1939 list_init(&rule
->facets
);
1940 if (n_actions
> 0) {
1941 rule
->n_actions
= n_actions
;
1942 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1948 static struct rule
*
1949 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1951 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1955 rule_free(struct rule
*rule
)
1957 free(rule
->actions
);
1961 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1962 * destroying any that no longer has a rule (which is probably all of them).
1964 * The caller must have already removed 'rule' from the classifier. */
1966 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1968 struct facet
*facet
, *next_facet
;
1969 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
1970 facet_revalidate(ofproto
, facet
);
1975 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1976 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1979 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
1981 const union ofp_action
*oa
;
1982 struct actions_iterator i
;
1984 if (out_port
== htons(OFPP_NONE
)) {
1987 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1988 oa
= actions_next(&i
)) {
1989 if (action_outputs_to_port(oa
, out_port
)) {
1996 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1997 * 'packet', which arrived on 'in_port'.
1999 * Takes ownership of 'packet'. */
2001 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
2002 const union odp_action
*actions
, size_t n_actions
,
2003 struct ofpbuf
*packet
)
2005 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
2006 /* As an optimization, avoid a round-trip from userspace to kernel to
2007 * userspace. This also avoids possibly filling up kernel packet
2008 * buffers along the way. */
2009 struct odp_msg
*msg
;
2011 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
2012 msg
->type
= _ODPL_ACTION_NR
;
2013 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
2014 msg
->port
= in_port
;
2016 msg
->arg
= actions
[0].controller
.arg
;
2018 send_packet_in(ofproto
, packet
);
2024 error
= dpif_execute(ofproto
->dpif
, actions
, n_actions
, packet
);
2025 ofpbuf_delete(packet
);
2030 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2031 * statistics appropriately. 'packet' must have at least sizeof(struct
2032 * ofp_packet_in) bytes of headroom.
2034 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2035 * applying flow_extract() to 'packet' would yield the same flow as
2038 * 'facet' must have accurately composed ODP actions; that is, it must not be
2039 * in need of revalidation.
2041 * Takes ownership of 'packet'. */
2043 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2044 struct ofpbuf
*packet
)
2046 struct odp_flow_stats stats
;
2048 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2050 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2051 if (execute_odp_actions(ofproto
, facet
->flow
.in_port
,
2052 facet
->actions
, facet
->n_actions
, packet
)) {
2053 facet_update_stats(ofproto
, facet
, &stats
);
2054 facet
->used
= time_msec();
2055 netflow_flow_update_time(ofproto
->netflow
,
2056 &facet
->nf_flow
, facet
->used
);
2060 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2061 * statistics (or the statistics for one of its facets) appropriately.
2062 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2064 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2065 * with statistics for 'packet' either way.
2067 * Takes ownership of 'packet'. */
2069 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2070 struct ofpbuf
*packet
)
2072 struct facet
*facet
;
2073 struct odp_actions a
;
2077 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2079 flow_extract(packet
, 0, in_port
, &flow
);
2081 /* First look for a related facet. If we find one, account it to that. */
2082 facet
= facet_lookup_valid(ofproto
, &flow
);
2083 if (facet
&& facet
->rule
== rule
) {
2084 facet_execute(ofproto
, facet
, packet
);
2088 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2089 * create a new facet for it and use that. */
2090 if (rule_lookup(ofproto
, &flow
) == rule
) {
2091 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2092 facet_execute(ofproto
, facet
, packet
);
2093 facet_install(ofproto
, facet
, true);
2097 /* We can't account anything to a facet. If we were to try, then that
2098 * facet would have a non-matching rule, busting our invariants. */
2099 if (xlate_actions(rule
->actions
, rule
->n_actions
, &flow
, ofproto
,
2100 packet
, &a
, NULL
, 0, NULL
)) {
2101 ofpbuf_delete(packet
);
2104 size
= packet
->size
;
2105 if (execute_odp_actions(ofproto
, in_port
,
2106 a
.actions
, a
.n_actions
, packet
)) {
2107 rule
->used
= time_msec();
2108 rule
->packet_count
++;
2109 rule
->byte_count
+= size
;
2113 /* Inserts 'rule' into 'p''s flow table. */
2115 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2117 struct rule
*displaced_rule
;
2119 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2120 if (displaced_rule
) {
2121 rule_destroy(p
, displaced_rule
);
2123 p
->need_revalidate
= true;
2126 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2127 * 'flow' and an example 'packet' within that flow.
2129 * The caller must already have determined that no facet with an identical
2130 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2131 * 'ofproto''s classifier table. */
2132 static struct facet
*
2133 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2134 const struct flow
*flow
, const struct ofpbuf
*packet
)
2136 struct facet
*facet
;
2138 facet
= xzalloc(sizeof *facet
);
2139 facet
->used
= time_msec();
2140 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2141 list_push_back(&rule
->facets
, &facet
->list_node
);
2143 facet
->flow
= *flow
;
2144 netflow_flow_init(&facet
->nf_flow
);
2145 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2147 facet_make_actions(ofproto
, facet
, packet
);
2153 facet_free(struct facet
*facet
)
2155 free(facet
->actions
);
2159 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2161 * - Removes 'rule' from the classifier.
2163 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2164 * destroys them), via rule_destroy().
2167 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2169 COVERAGE_INC(ofproto_del_rule
);
2170 ofproto
->need_revalidate
= true;
2171 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2172 rule_destroy(ofproto
, rule
);
2175 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2177 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2178 * rule's statistics, via facet_uninstall().
2180 * - Removes 'facet' from its rule and from ofproto->facets.
2183 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2185 facet_uninstall(ofproto
, facet
);
2186 facet_flush_stats(ofproto
, facet
);
2187 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2188 list_remove(&facet
->list_node
);
2192 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2194 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2195 const struct ofpbuf
*packet
)
2197 const struct rule
*rule
= facet
->rule
;
2198 struct odp_actions a
;
2201 xlate_actions(rule
->actions
, rule
->n_actions
, &facet
->flow
, p
,
2202 packet
, &a
, &facet
->tags
, &facet
->may_install
,
2203 &facet
->nf_flow
.output_iface
);
2205 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2206 if (facet
->n_actions
!= a
.n_actions
2207 || memcmp(facet
->actions
, a
.actions
, actions_len
)) {
2208 free(facet
->actions
);
2209 facet
->n_actions
= a
.n_actions
;
2210 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2215 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
, int flags
,
2216 struct odp_flow_put
*put
)
2218 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2219 odp_flow_key_from_flow(&put
->flow
.key
, &facet
->flow
);
2220 put
->flow
.actions
= facet
->actions
;
2221 put
->flow
.n_actions
= facet
->n_actions
;
2222 put
->flow
.flags
= 0;
2224 return dpif_flow_put(ofproto
->dpif
, put
);
2227 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2228 * 'zero_stats' is true, clears any existing statistics from the datapath for
2231 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2233 if (facet
->may_install
) {
2234 struct odp_flow_put put
;
2237 flags
= ODPPF_CREATE
| ODPPF_MODIFY
;
2239 flags
|= ODPPF_ZERO_STATS
;
2241 if (!facet_put__(p
, facet
, flags
, &put
)) {
2242 facet
->installed
= true;
2247 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2248 * to the accounting hook function in the ofhooks structure. */
2250 facet_account(struct ofproto
*ofproto
,
2251 struct facet
*facet
, uint64_t extra_bytes
)
2253 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2255 if (ofproto
->ofhooks
->account_flow_cb
2256 && total_bytes
> facet
->accounted_bytes
)
2258 ofproto
->ofhooks
->account_flow_cb(
2259 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->n_actions
,
2260 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2261 facet
->accounted_bytes
= total_bytes
;
2265 /* If 'rule' is installed in the datapath, uninstalls it. */
2267 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2269 if (facet
->installed
) {
2270 struct odp_flow odp_flow
;
2272 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
2273 odp_flow
.actions
= NULL
;
2274 odp_flow
.n_actions
= 0;
2276 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2277 facet_update_stats(p
, facet
, &odp_flow
.stats
);
2279 facet
->installed
= false;
2283 /* Returns true if the only action for 'facet' is to send to the controller.
2284 * (We don't report NetFlow expiration messages for such facets because they
2285 * are just part of the control logic for the network, not real traffic). */
2287 facet_is_controller_flow(struct facet
*facet
)
2290 && facet
->rule
->n_actions
== 1
2291 && action_outputs_to_port(&facet
->rule
->actions
[0],
2292 htons(OFPP_CONTROLLER
)));
2295 /* Folds all of 'facet''s statistics into its rule. Also updates the
2296 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2298 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2300 facet_account(ofproto
, facet
, 0);
2302 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2303 struct ofexpired expired
;
2304 expired
.flow
= facet
->flow
;
2305 expired
.packet_count
= facet
->packet_count
;
2306 expired
.byte_count
= facet
->byte_count
;
2307 expired
.used
= facet
->used
;
2308 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2311 facet
->rule
->packet_count
+= facet
->packet_count
;
2312 facet
->rule
->byte_count
+= facet
->byte_count
;
2314 /* Reset counters to prevent double counting if 'facet' ever gets
2316 facet
->packet_count
= 0;
2317 facet
->byte_count
= 0;
2318 facet
->accounted_bytes
= 0;
2320 netflow_flow_clear(&facet
->nf_flow
);
2323 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2324 * Returns it if found, otherwise a null pointer.
2326 * The returned facet might need revalidation; use facet_lookup_valid()
2327 * instead if that is important. */
2328 static struct facet
*
2329 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2331 struct facet
*facet
;
2333 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2335 if (flow_equal(flow
, &facet
->flow
)) {
2343 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2344 * Returns it if found, otherwise a null pointer.
2346 * The returned facet is guaranteed to be valid. */
2347 static struct facet
*
2348 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2350 struct facet
*facet
= facet_find(ofproto
, flow
);
2352 /* The facet we found might not be valid, since we could be in need of
2353 * revalidation. If it is not valid, don't return it. */
2355 && ofproto
->need_revalidate
2356 && !facet_revalidate(ofproto
, facet
)) {
2357 COVERAGE_INC(ofproto_invalidated
);
2364 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2366 * - If the rule found is different from 'facet''s current rule, moves
2367 * 'facet' to the new rule and recompiles its actions.
2369 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2370 * where it is and recompiles its actions anyway.
2372 * - If there is none, destroys 'facet'.
2374 * Returns true if 'facet' still exists, false if it has been destroyed. */
2376 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2378 struct rule
*new_rule
;
2379 struct odp_actions a
;
2381 uint16_t new_nf_output_iface
;
2382 bool actions_changed
;
2384 COVERAGE_INC(facet_revalidate
);
2386 /* Determine the new rule. */
2387 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2389 /* No new rule, so delete the facet. */
2390 facet_remove(ofproto
, facet
);
2394 /* Calculate new ODP actions.
2396 * We are very cautious about actually modifying 'facet' state at this
2397 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2398 * so, we need to have the old state around to properly compose it. */
2399 xlate_actions(new_rule
->actions
, new_rule
->n_actions
, &facet
->flow
,
2400 ofproto
, NULL
, &a
, &facet
->tags
, &facet
->may_install
,
2401 &new_nf_output_iface
);
2402 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2403 actions_changed
= (facet
->n_actions
!= a
.n_actions
2404 || memcmp(facet
->actions
, a
.actions
, actions_len
));
2406 /* If the ODP actions changed or the installability changed, then we need
2407 * to talk to the datapath. */
2408 if (actions_changed
|| facet
->may_install
!= facet
->installed
) {
2409 if (facet
->may_install
) {
2410 struct odp_flow_put put
;
2412 memset(&put
.flow
.stats
, 0, sizeof put
.flow
.stats
);
2413 odp_flow_key_from_flow(&put
.flow
.key
, &facet
->flow
);
2414 put
.flow
.actions
= a
.actions
;
2415 put
.flow
.n_actions
= a
.n_actions
;
2417 put
.flags
= ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
;
2418 dpif_flow_put(ofproto
->dpif
, &put
);
2420 facet_update_stats(ofproto
, facet
, &put
.flow
.stats
);
2422 facet_uninstall(ofproto
, facet
);
2425 /* The datapath flow is gone or has zeroed stats, so push stats out of
2426 * 'facet' into 'rule'. */
2427 facet_flush_stats(ofproto
, facet
);
2430 /* Update 'facet' now that we've taken care of all the old state. */
2431 facet
->nf_flow
.output_iface
= new_nf_output_iface
;
2432 if (actions_changed
) {
2433 free(facet
->actions
);
2434 facet
->n_actions
= a
.n_actions
;
2435 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2437 if (facet
->rule
!= new_rule
) {
2438 COVERAGE_INC(facet_changed_rule
);
2439 list_remove(&facet
->list_node
);
2440 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2441 facet
->rule
= new_rule
;
2442 facet
->used
= new_rule
->created
;
2449 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2450 struct rconn_packet_counter
*counter
)
2452 update_openflow_length(msg
);
2453 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2459 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2462 struct ofpbuf
*buf
= make_ofp_error_msg(error
, oh
);
2464 COVERAGE_INC(ofproto_error
);
2465 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2470 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2472 opp
->port_no
= htons(opp
->port_no
);
2473 opp
->config
= htonl(opp
->config
);
2474 opp
->state
= htonl(opp
->state
);
2475 opp
->curr
= htonl(opp
->curr
);
2476 opp
->advertised
= htonl(opp
->advertised
);
2477 opp
->supported
= htonl(opp
->supported
);
2478 opp
->peer
= htonl(opp
->peer
);
2482 handle_echo_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2484 queue_tx(make_echo_reply(oh
), ofconn
, ofconn
->reply_counter
);
2489 handle_features_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2491 struct ofp_switch_features
*osf
;
2493 struct ofport
*port
;
2495 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2496 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2497 osf
->n_buffers
= htonl(pktbuf_capacity());
2499 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2500 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2501 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2502 (1u << OFPAT_SET_VLAN_VID
) |
2503 (1u << OFPAT_SET_VLAN_PCP
) |
2504 (1u << OFPAT_STRIP_VLAN
) |
2505 (1u << OFPAT_SET_DL_SRC
) |
2506 (1u << OFPAT_SET_DL_DST
) |
2507 (1u << OFPAT_SET_NW_SRC
) |
2508 (1u << OFPAT_SET_NW_DST
) |
2509 (1u << OFPAT_SET_NW_TOS
) |
2510 (1u << OFPAT_SET_TP_SRC
) |
2511 (1u << OFPAT_SET_TP_DST
) |
2512 (1u << OFPAT_ENQUEUE
));
2514 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2515 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2518 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2523 handle_get_config_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2526 struct ofp_switch_config
*osc
;
2530 /* Figure out flags. */
2531 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2532 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2535 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2536 osc
->flags
= htons(flags
);
2537 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2538 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2544 handle_set_config(struct ofconn
*ofconn
, const struct ofp_switch_config
*osc
)
2546 uint16_t flags
= ntohs(osc
->flags
);
2548 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2549 switch (flags
& OFPC_FRAG_MASK
) {
2550 case OFPC_FRAG_NORMAL
:
2551 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2553 case OFPC_FRAG_DROP
:
2554 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2557 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2563 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2569 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2571 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2572 a
->controller
.arg
= max_len
;
2575 struct action_xlate_ctx
{
2577 struct flow flow
; /* Flow to which these actions correspond. */
2578 int recurse
; /* Recursion level, via xlate_table_action. */
2579 struct ofproto
*ofproto
;
2580 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2581 * null pointer if we are revalidating
2582 * without a packet to refer to. */
2585 struct odp_actions
*out
; /* Datapath actions. */
2586 tag_type tags
; /* Tags associated with OFPP_NORMAL actions. */
2587 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2588 * be reassessed for every packet. */
2589 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2592 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2593 * flow translation. */
2594 #define MAX_RESUBMIT_RECURSION 8
2596 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2597 struct action_xlate_ctx
*ctx
);
2600 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2602 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2605 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2606 /* Forwarding disabled on port. */
2611 * We don't have an ofport record for this port, but it doesn't hurt to
2612 * allow forwarding to it anyhow. Maybe such a port will appear later
2613 * and we're pre-populating the flow table.
2617 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2618 ctx
->nf_output_iface
= port
;
2621 static struct rule
*
2622 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2624 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2628 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2630 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2631 uint16_t old_in_port
;
2634 /* Look up a flow with 'in_port' as the input port. Then restore the
2635 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2636 * have surprising behavior). */
2637 old_in_port
= ctx
->flow
.in_port
;
2638 ctx
->flow
.in_port
= in_port
;
2639 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2640 ctx
->flow
.in_port
= old_in_port
;
2644 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2648 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2650 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2651 MAX_RESUBMIT_RECURSION
);
2656 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2657 uint16_t *nf_output_iface
, struct odp_actions
*actions
)
2659 struct ofport
*ofport
;
2661 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2662 uint16_t odp_port
= ofport
->odp_port
;
2663 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2664 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= odp_port
;
2667 *nf_output_iface
= NF_OUT_FLOOD
;
2671 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2672 uint16_t port
, uint16_t max_len
)
2675 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2677 ctx
->nf_output_iface
= NF_OUT_DROP
;
2681 add_output_action(ctx
, ctx
->flow
.in_port
);
2684 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2687 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2688 ctx
->out
, &ctx
->tags
,
2689 &ctx
->nf_output_iface
,
2690 ctx
->ofproto
->aux
)) {
2691 COVERAGE_INC(ofproto_uninstallable
);
2692 ctx
->may_set_up_flow
= false;
2696 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2697 &ctx
->nf_output_iface
, ctx
->out
);
2700 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2701 &ctx
->nf_output_iface
, ctx
->out
);
2703 case OFPP_CONTROLLER
:
2704 add_controller_action(ctx
->out
, max_len
);
2707 add_output_action(ctx
, ODPP_LOCAL
);
2710 odp_port
= ofp_port_to_odp_port(port
);
2711 if (odp_port
!= ctx
->flow
.in_port
) {
2712 add_output_action(ctx
, odp_port
);
2717 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2718 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2719 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2720 ctx
->nf_output_iface
= prev_nf_output_iface
;
2721 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2722 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2723 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2728 xlate_output_action(struct action_xlate_ctx
*ctx
,
2729 const struct ofp_action_output
*oao
)
2731 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2734 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2735 * optimization, because we're going to add another action that sets the
2736 * priority immediately after, or because there are no actions following the
2739 remove_pop_action(struct action_xlate_ctx
*ctx
)
2741 size_t n
= ctx
->out
->n_actions
;
2742 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2743 ctx
->out
->n_actions
--;
2748 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2749 const struct ofp_action_enqueue
*oae
)
2751 uint16_t ofp_port
, odp_port
;
2755 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2758 /* Fall back to ordinary output action. */
2759 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2763 /* Figure out ODP output port. */
2764 ofp_port
= ntohs(oae
->port
);
2765 if (ofp_port
!= OFPP_IN_PORT
) {
2766 odp_port
= ofp_port_to_odp_port(ofp_port
);
2768 odp_port
= ctx
->flow
.in_port
;
2771 /* Add ODP actions. */
2772 remove_pop_action(ctx
);
2773 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2775 add_output_action(ctx
, odp_port
);
2776 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2778 /* Update NetFlow output port. */
2779 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2780 ctx
->nf_output_iface
= odp_port
;
2781 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2782 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2787 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2788 const struct nx_action_set_queue
*nasq
)
2793 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2796 /* Couldn't translate queue to a priority, so ignore. A warning
2797 * has already been logged. */
2801 remove_pop_action(ctx
);
2802 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2807 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2809 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2810 if (!(tci
& htons(VLAN_CFI
))) {
2811 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2813 union odp_action
*oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_TCI
);
2814 oa
->dl_tci
.tci
= tci
& ~htons(VLAN_CFI
);
2819 xlate_reg_move_action(struct action_xlate_ctx
*ctx
,
2820 const struct nx_action_reg_move
*narm
)
2822 ovs_be16 old_tci
= ctx
->flow
.vlan_tci
;
2824 nxm_execute_reg_move(narm
, &ctx
->flow
);
2826 if (ctx
->flow
.vlan_tci
!= old_tci
) {
2827 xlate_set_dl_tci(ctx
);
2832 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2833 const struct nx_action_header
*nah
)
2835 const struct nx_action_resubmit
*nar
;
2836 const struct nx_action_set_tunnel
*nast
;
2837 const struct nx_action_set_queue
*nasq
;
2838 union odp_action
*oa
;
2839 int subtype
= ntohs(nah
->subtype
);
2841 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2843 case NXAST_RESUBMIT
:
2844 nar
= (const struct nx_action_resubmit
*) nah
;
2845 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2848 case NXAST_SET_TUNNEL
:
2849 nast
= (const struct nx_action_set_tunnel
*) nah
;
2850 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2851 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2854 case NXAST_DROP_SPOOFED_ARP
:
2855 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2856 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2860 case NXAST_SET_QUEUE
:
2861 nasq
= (const struct nx_action_set_queue
*) nah
;
2862 xlate_set_queue_action(ctx
, nasq
);
2865 case NXAST_POP_QUEUE
:
2866 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2869 case NXAST_REG_MOVE
:
2870 xlate_reg_move_action(ctx
, (const struct nx_action_reg_move
*) nah
);
2873 case NXAST_REG_LOAD
:
2874 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
2878 /* Nothing to do. */
2881 /* If you add a new action here that modifies flow data, don't forget to
2882 * update the flow key in ctx->flow at the same time. */
2885 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2891 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2892 struct action_xlate_ctx
*ctx
)
2894 struct actions_iterator iter
;
2895 const union ofp_action
*ia
;
2896 const struct ofport
*port
;
2898 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2899 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2900 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2901 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2902 /* Drop this flow. */
2906 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2907 uint16_t type
= ntohs(ia
->type
);
2908 union odp_action
*oa
;
2912 xlate_output_action(ctx
, &ia
->output
);
2915 case OFPAT_SET_VLAN_VID
:
2916 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
2917 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
2918 xlate_set_dl_tci(ctx
);
2921 case OFPAT_SET_VLAN_PCP
:
2922 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
2923 ctx
->flow
.vlan_tci
|= htons(
2924 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
2925 xlate_set_dl_tci(ctx
);
2928 case OFPAT_STRIP_VLAN
:
2929 ctx
->flow
.vlan_tci
= htons(0);
2930 xlate_set_dl_tci(ctx
);
2933 case OFPAT_SET_DL_SRC
:
2934 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2935 memcpy(oa
->dl_addr
.dl_addr
,
2936 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2937 memcpy(ctx
->flow
.dl_src
,
2938 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2941 case OFPAT_SET_DL_DST
:
2942 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2943 memcpy(oa
->dl_addr
.dl_addr
,
2944 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2945 memcpy(ctx
->flow
.dl_dst
,
2946 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2949 case OFPAT_SET_NW_SRC
:
2950 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2951 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2954 case OFPAT_SET_NW_DST
:
2955 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2956 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2959 case OFPAT_SET_NW_TOS
:
2960 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2961 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2964 case OFPAT_SET_TP_SRC
:
2965 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2966 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2969 case OFPAT_SET_TP_DST
:
2970 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2971 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2975 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2979 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2983 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2990 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2991 const struct flow
*flow
, struct ofproto
*ofproto
,
2992 const struct ofpbuf
*packet
,
2993 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2994 uint16_t *nf_output_iface
)
2996 struct action_xlate_ctx ctx
;
2998 COVERAGE_INC(ofproto_ofp2odp
);
2999 odp_actions_init(out
);
3002 ctx
.ofproto
= ofproto
;
3003 ctx
.packet
= packet
;
3006 ctx
.may_set_up_flow
= true;
3007 ctx
.nf_output_iface
= NF_OUT_DROP
;
3008 do_xlate_actions(in
, n_in
, &ctx
);
3009 remove_pop_action(&ctx
);
3011 /* Check with in-band control to see if we're allowed to set up this
3013 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
3014 ctx
.may_set_up_flow
= false;
3020 if (may_set_up_flow
) {
3021 *may_set_up_flow
= ctx
.may_set_up_flow
;
3023 if (nf_output_iface
) {
3024 *nf_output_iface
= ctx
.nf_output_iface
;
3026 if (odp_actions_overflow(out
)) {
3027 COVERAGE_INC(odp_overflow
);
3028 odp_actions_init(out
);
3029 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
3034 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3035 * error message code (composed with ofp_mkerr()) for the caller to propagate
3036 * upward. Otherwise, returns 0.
3038 * The log message mentions 'msg_type'. */
3040 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3042 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3043 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3044 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3047 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3054 handle_packet_out(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3056 struct ofproto
*p
= ofconn
->ofproto
;
3057 struct ofp_packet_out
*opo
;
3058 struct ofpbuf payload
, *buffer
;
3059 union ofp_action
*ofp_actions
;
3060 struct odp_actions odp_actions
;
3061 struct ofpbuf request
;
3063 size_t n_ofp_actions
;
3067 COVERAGE_INC(ofproto_packet_out
);
3069 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3074 /* Get ofp_packet_out. */
3075 request
.data
= (void *) oh
;
3076 request
.size
= ntohs(oh
->length
);
3077 opo
= ofpbuf_try_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3079 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3083 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3084 &ofp_actions
, &n_ofp_actions
);
3090 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3091 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3093 if (error
|| !buffer
) {
3102 /* Extract flow, check actions. */
3103 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3105 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3111 error
= xlate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
, &payload
,
3112 &odp_actions
, NULL
, NULL
, NULL
);
3114 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
,
3119 ofpbuf_delete(buffer
);
3124 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3125 uint32_t config
, uint32_t mask
)
3127 mask
&= config
^ port
->opp
.config
;
3128 if (mask
& OFPPC_PORT_DOWN
) {
3129 if (config
& OFPPC_PORT_DOWN
) {
3130 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3132 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3135 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3136 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3137 if (mask
& REVALIDATE_BITS
) {
3138 COVERAGE_INC(ofproto_costly_flags
);
3139 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3140 p
->need_revalidate
= true;
3142 #undef REVALIDATE_BITS
3143 if (mask
& OFPPC_NO_PACKET_IN
) {
3144 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3149 handle_port_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3151 struct ofproto
*p
= ofconn
->ofproto
;
3152 const struct ofp_port_mod
*opm
= (const struct ofp_port_mod
*) oh
;
3153 struct ofport
*port
;
3156 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3161 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3163 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3164 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3165 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3167 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3168 if (opm
->advertise
) {
3169 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3175 static struct ofpbuf
*
3176 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3178 struct ofp_stats_reply
*osr
;
3181 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3182 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3184 osr
->flags
= htons(0);
3188 static struct ofpbuf
*
3189 start_ofp_stats_reply(const struct ofp_header
*request
, size_t body_len
)
3191 const struct ofp_stats_request
*osr
3192 = (const struct ofp_stats_request
*) request
;
3193 return make_ofp_stats_reply(osr
->header
.xid
, osr
->type
, body_len
);
3197 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3198 struct ofpbuf
**msgp
)
3200 struct ofpbuf
*msg
= *msgp
;
3201 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3202 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3203 struct ofp_stats_reply
*reply
= msg
->data
;
3204 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3205 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3206 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3208 return ofpbuf_put_uninit(*msgp
, nbytes
);
3211 static struct ofpbuf
*
3212 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3214 struct nicira_stats_msg
*nsm
;
3217 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3218 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3219 nsm
->type
= htons(OFPST_VENDOR
);
3220 nsm
->flags
= htons(0);
3221 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3222 nsm
->subtype
= htonl(subtype
);
3226 static struct ofpbuf
*
3227 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3229 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3233 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3234 struct ofpbuf
**msgp
)
3236 struct ofpbuf
*msg
= *msgp
;
3237 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3238 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3239 struct nicira_stats_msg
*reply
= msg
->data
;
3240 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3241 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3242 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3244 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3248 handle_desc_stats_request(struct ofconn
*ofconn
,
3249 const struct ofp_header
*request
)
3251 struct ofproto
*p
= ofconn
->ofproto
;
3252 struct ofp_desc_stats
*ods
;
3255 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3256 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3257 memset(ods
, 0, sizeof *ods
);
3258 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3259 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3260 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3261 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3262 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3263 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3269 handle_table_stats_request(struct ofconn
*ofconn
,
3270 const struct ofp_header
*request
)
3272 struct ofproto
*p
= ofconn
->ofproto
;
3273 struct ofp_table_stats
*ots
;
3276 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3278 /* Classifier table. */
3279 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3280 memset(ots
, 0, sizeof *ots
);
3281 strcpy(ots
->name
, "classifier");
3282 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3283 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3284 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3285 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3286 ots
->lookup_count
= htonll(0); /* XXX */
3287 ots
->matched_count
= htonll(0); /* XXX */
3289 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3294 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3295 struct ofpbuf
**msgp
)
3297 struct netdev_stats stats
;
3298 struct ofp_port_stats
*ops
;
3300 /* Intentionally ignore return value, since errors will set
3301 * 'stats' to all-1s, which is correct for OpenFlow, and
3302 * netdev_get_stats() will log errors. */
3303 netdev_get_stats(port
->netdev
, &stats
);
3305 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3306 ops
->port_no
= htons(port
->opp
.port_no
);
3307 memset(ops
->pad
, 0, sizeof ops
->pad
);
3308 ops
->rx_packets
= htonll(stats
.rx_packets
);
3309 ops
->tx_packets
= htonll(stats
.tx_packets
);
3310 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3311 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3312 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3313 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3314 ops
->rx_errors
= htonll(stats
.rx_errors
);
3315 ops
->tx_errors
= htonll(stats
.tx_errors
);
3316 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3317 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3318 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3319 ops
->collisions
= htonll(stats
.collisions
);
3323 handle_port_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3325 struct ofproto
*p
= ofconn
->ofproto
;
3326 const struct ofp_port_stats_request
*psr
= ofputil_stats_body(oh
);
3327 struct ofp_port_stats
*ops
;
3329 struct ofport
*port
;
3331 msg
= start_ofp_stats_reply(oh
, sizeof *ops
* 16);
3332 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3333 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3335 append_port_stat(port
, ofconn
, &msg
);
3338 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3339 append_port_stat(port
, ofconn
, &msg
);
3343 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3347 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3348 * '*packet_countp' and '*byte_countp'. The returned statistics include
3349 * statistics for all of 'rule''s facets. */
3351 query_stats(struct ofproto
*p
, struct rule
*rule
,
3352 uint64_t *packet_countp
, uint64_t *byte_countp
)
3354 uint64_t packet_count
, byte_count
;
3355 struct facet
*facet
;
3356 struct odp_flow
*odp_flows
;
3359 /* Start from historical data for 'rule' itself that are no longer tracked
3360 * by the datapath. This counts, for example, facets that have expired. */
3361 packet_count
= rule
->packet_count
;
3362 byte_count
= rule
->byte_count
;
3364 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3366 * Also, add any statistics that are not tracked by the datapath for each
3367 * facet. This includes, for example, statistics for packets that were
3368 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3370 odp_flows
= xzalloc(list_size(&rule
->facets
) * sizeof *odp_flows
);
3372 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
3373 struct odp_flow
*odp_flow
= &odp_flows
[n_odp_flows
++];
3374 odp_flow_key_from_flow(&odp_flow
->key
, &facet
->flow
);
3375 packet_count
+= facet
->packet_count
;
3376 byte_count
+= facet
->byte_count
;
3379 /* Fetch up-to-date statistics from the datapath and add them in. */
3380 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3383 for (i
= 0; i
< n_odp_flows
; i
++) {
3384 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3385 packet_count
+= odp_flow
->stats
.n_packets
;
3386 byte_count
+= odp_flow
->stats
.n_bytes
;
3391 /* Return the stats to the caller. */
3392 *packet_countp
= packet_count
;
3393 *byte_countp
= byte_count
;
3397 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3399 long long int msecs
= time_msec() - start
;
3400 *sec
= htonl(msecs
/ 1000);
3401 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3405 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3406 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3408 struct ofp_flow_stats
*ofs
;
3409 uint64_t packet_count
, byte_count
;
3410 size_t act_len
, len
;
3412 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3416 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3417 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3419 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3421 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3422 ofs
->length
= htons(len
);
3425 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
);
3426 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3427 ofs
->cookie
= rule
->flow_cookie
;
3428 ofs
->priority
= htons(rule
->cr
.priority
);
3429 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3430 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3431 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3432 ofs
->packet_count
= htonll(packet_count
);
3433 ofs
->byte_count
= htonll(byte_count
);
3434 if (rule
->n_actions
> 0) {
3435 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3440 is_valid_table(uint8_t table_id
)
3442 return table_id
== 0 || table_id
== 0xff;
3446 handle_flow_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3448 const struct ofp_flow_stats_request
*fsr
= ofputil_stats_body(oh
);
3449 struct ofpbuf
*reply
;
3451 COVERAGE_INC(ofproto_flows_req
);
3452 reply
= start_ofp_stats_reply(oh
, 1024);
3453 if (is_valid_table(fsr
->table_id
)) {
3454 struct cls_cursor cursor
;
3455 struct cls_rule target
;
3458 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3460 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3461 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3462 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3465 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3471 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3472 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3474 struct nx_flow_stats
*nfs
;
3475 uint64_t packet_count
, byte_count
;
3476 size_t act_len
, start_len
;
3477 struct ofpbuf
*reply
;
3479 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3483 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3485 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3487 start_len
= (*replyp
)->size
;
3488 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3491 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3494 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3495 nfs
->cookie
= rule
->flow_cookie
;
3496 nfs
->priority
= htons(rule
->cr
.priority
);
3497 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3498 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3499 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3500 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3501 nfs
->packet_count
= htonll(packet_count
);
3502 nfs
->byte_count
= htonll(byte_count
);
3503 if (rule
->n_actions
> 0) {
3504 ofpbuf_put(reply
, rule
->actions
, act_len
);
3506 nfs
->length
= htons(reply
->size
- start_len
);
3510 handle_nxst_flow(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3512 struct nx_flow_stats_request
*nfsr
;
3513 struct cls_rule target
;
3514 struct ofpbuf
*reply
;
3518 b
.data
= (void *) oh
;
3519 b
.size
= ntohs(oh
->length
);
3521 /* Dissect the message. */
3522 nfsr
= ofpbuf_try_pull(&b
, sizeof *nfsr
);
3524 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3526 error
= nx_pull_match(&b
, ntohs(nfsr
->match_len
), 0, &target
);
3531 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3534 COVERAGE_INC(ofproto_flows_req
);
3535 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3536 if (is_valid_table(nfsr
->table_id
)) {
3537 struct cls_cursor cursor
;
3540 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3541 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3542 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3545 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3551 flow_stats_ds(struct ofproto
*ofproto
, struct rule
*rule
, struct ds
*results
)
3553 struct ofp_match match
;
3554 uint64_t packet_count
, byte_count
;
3555 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3557 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3558 ofputil_cls_rule_to_match(&rule
->cr
, NXFF_OPENFLOW10
, &match
);
3560 ds_put_format(results
, "duration=%llds, ",
3561 (time_msec() - rule
->created
) / 1000);
3562 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3563 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3564 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3565 ofp_print_match(results
, &match
, true);
3567 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3569 ds_put_cstr(results
, "drop");
3571 ds_put_cstr(results
, "\n");
3574 /* Adds a pretty-printed description of all flows to 'results', including
3575 * those marked hidden by secchan (e.g., by in-band control). */
3577 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3579 struct cls_cursor cursor
;
3582 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3583 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3584 flow_stats_ds(p
, rule
, results
);
3589 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3590 ovs_be16 out_port
, uint8_t table_id
,
3591 struct ofp_aggregate_stats_reply
*oasr
)
3593 uint64_t total_packets
= 0;
3594 uint64_t total_bytes
= 0;
3597 COVERAGE_INC(ofproto_agg_request
);
3599 if (is_valid_table(table_id
)) {
3600 struct cls_cursor cursor
;
3603 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3604 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3605 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3606 uint64_t packet_count
;
3607 uint64_t byte_count
;
3609 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3611 total_packets
+= packet_count
;
3612 total_bytes
+= byte_count
;
3618 oasr
->flow_count
= htonl(n_flows
);
3619 oasr
->packet_count
= htonll(total_packets
);
3620 oasr
->byte_count
= htonll(total_bytes
);
3621 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3625 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3626 const struct ofp_header
*oh
)
3628 const struct ofp_aggregate_stats_request
*request
= ofputil_stats_body(oh
);
3629 struct ofp_aggregate_stats_reply
*reply
;
3630 struct cls_rule target
;
3633 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3636 msg
= start_ofp_stats_reply(oh
, sizeof *reply
);
3637 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3638 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3639 request
->table_id
, reply
);
3640 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3645 handle_nxst_aggregate(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3647 struct nx_aggregate_stats_request
*request
;
3648 struct ofp_aggregate_stats_reply
*reply
;
3649 struct cls_rule target
;
3654 b
.data
= (void *) oh
;
3655 b
.size
= ntohs(oh
->length
);
3657 /* Dissect the message. */
3658 request
= ofpbuf_try_pull(&b
, sizeof *request
);
3660 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3662 error
= nx_pull_match(&b
, ntohs(request
->match_len
), 0, &target
);
3667 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3671 COVERAGE_INC(ofproto_flows_req
);
3672 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3673 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3674 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3675 request
->table_id
, reply
);
3676 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3681 struct queue_stats_cbdata
{
3682 struct ofconn
*ofconn
;
3683 struct ofport
*ofport
;
3688 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3689 const struct netdev_queue_stats
*stats
)
3691 struct ofp_queue_stats
*reply
;
3693 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3694 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3695 memset(reply
->pad
, 0, sizeof reply
->pad
);
3696 reply
->queue_id
= htonl(queue_id
);
3697 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3698 reply
->tx_packets
= htonll(stats
->tx_packets
);
3699 reply
->tx_errors
= htonll(stats
->tx_errors
);
3703 handle_queue_stats_dump_cb(uint32_t queue_id
,
3704 struct netdev_queue_stats
*stats
,
3707 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3709 put_queue_stats(cbdata
, queue_id
, stats
);
3713 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3714 struct queue_stats_cbdata
*cbdata
)
3716 cbdata
->ofport
= port
;
3717 if (queue_id
== OFPQ_ALL
) {
3718 netdev_dump_queue_stats(port
->netdev
,
3719 handle_queue_stats_dump_cb
, cbdata
);
3721 struct netdev_queue_stats stats
;
3723 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3724 put_queue_stats(cbdata
, queue_id
, &stats
);
3730 handle_queue_stats_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
3732 struct ofproto
*ofproto
= ofconn
->ofproto
;
3733 const struct ofp_queue_stats_request
*qsr
;
3734 struct queue_stats_cbdata cbdata
;
3735 struct ofport
*port
;
3736 unsigned int port_no
;
3739 qsr
= ofputil_stats_body(oh
);
3741 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3744 COVERAGE_INC(ofproto_queue_req
);
3746 cbdata
.ofconn
= ofconn
;
3747 cbdata
.msg
= start_ofp_stats_reply(oh
, 128);
3749 port_no
= ntohs(qsr
->port_no
);
3750 queue_id
= ntohl(qsr
->queue_id
);
3751 if (port_no
== OFPP_ALL
) {
3752 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3753 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3755 } else if (port_no
< ofproto
->max_ports
) {
3756 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3758 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3761 ofpbuf_delete(cbdata
.msg
);
3762 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3764 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3769 static long long int
3770 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3772 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3776 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3777 const struct odp_flow_stats
*stats
)
3779 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3780 if (used
> facet
->used
) {
3782 if (used
> facet
->rule
->used
) {
3783 facet
->rule
->used
= used
;
3785 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3789 /* Folds the statistics from 'stats' into the counters in 'facet'.
3791 * Because of the meaning of a facet's counters, it only makes sense to do this
3792 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3793 * packet that was sent by hand or if it represents statistics that have been
3794 * cleared out of the datapath. */
3796 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3797 const struct odp_flow_stats
*stats
)
3799 if (stats
->n_packets
) {
3800 facet_update_time(ofproto
, facet
, stats
);
3801 facet
->packet_count
+= stats
->n_packets
;
3802 facet
->byte_count
+= stats
->n_bytes
;
3803 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3807 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3808 * in which no matching flow already exists in the flow table.
3810 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3811 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3812 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3814 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3817 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3819 struct ofproto
*p
= ofconn
->ofproto
;
3820 struct ofpbuf
*packet
;
3825 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3826 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3827 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3831 if (fm
->buffer_id
!= UINT32_MAX
) {
3832 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3836 in_port
= UINT16_MAX
;
3839 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
3840 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3841 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3842 rule_insert(p
, rule
);
3844 rule_execute(p
, rule
, in_port
, packet
);
3849 static struct rule
*
3850 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3852 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
3856 send_buffered_packet(struct ofconn
*ofconn
,
3857 struct rule
*rule
, uint32_t buffer_id
)
3859 struct ofpbuf
*packet
;
3863 if (buffer_id
== UINT32_MAX
) {
3867 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
3872 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
3877 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3879 struct modify_flows_cbdata
{
3880 struct ofproto
*ofproto
;
3881 const struct flow_mod
*fm
;
3885 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
3888 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3889 * encoded by ofp_mkerr() on failure.
3891 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3894 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3896 struct ofproto
*p
= ofconn
->ofproto
;
3897 struct rule
*match
= NULL
;
3898 struct cls_cursor cursor
;
3901 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
3902 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3903 if (!rule_is_hidden(rule
)) {
3905 modify_flow(p
, fm
, rule
);
3910 /* This credits the packet to whichever flow happened to match last.
3911 * That's weird. Maybe we should do a lookup for the flow that
3912 * actually matches the packet? Who knows. */
3913 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
3916 return add_flow(ofconn
, fm
);
3920 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3921 * code as encoded by ofp_mkerr() on failure.
3923 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3926 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3928 struct ofproto
*p
= ofconn
->ofproto
;
3929 struct rule
*rule
= find_flow_strict(p
, fm
);
3930 if (rule
&& !rule_is_hidden(rule
)) {
3931 modify_flow(p
, fm
, rule
);
3932 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
3934 return add_flow(ofconn
, fm
);
3938 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3939 * been identified as a flow in 'p''s flow table to be modified, by changing
3940 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3941 * ofp_action[] structures). */
3943 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
3945 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
3947 rule
->flow_cookie
= fm
->cookie
;
3949 /* If the actions are the same, do nothing. */
3950 if (fm
->n_actions
== rule
->n_actions
3952 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
3956 /* Replace actions. */
3957 free(rule
->actions
);
3958 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
3959 rule
->n_actions
= fm
->n_actions
;
3961 p
->need_revalidate
= true;
3966 /* OFPFC_DELETE implementation. */
3968 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
3970 /* Implements OFPFC_DELETE. */
3972 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
3974 struct rule
*rule
, *next_rule
;
3975 struct cls_cursor cursor
;
3977 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
3978 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
3979 delete_flow(p
, rule
, htons(fm
->out_port
));
3983 /* Implements OFPFC_DELETE_STRICT. */
3985 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
3987 struct rule
*rule
= find_flow_strict(p
, fm
);
3989 delete_flow(p
, rule
, htons(fm
->out_port
));
3993 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3994 * been identified as a flow to delete from 'p''s flow table, by deleting the
3995 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3998 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3999 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4000 * specified 'out_port'. */
4002 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4004 if (rule_is_hidden(rule
)) {
4008 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4012 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4013 rule_remove(p
, rule
);
4017 handle_flow_mod(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4019 struct ofproto
*p
= ofconn
->ofproto
;
4023 error
= reject_slave_controller(ofconn
, "flow_mod");
4028 error
= ofputil_decode_flow_mod(&fm
, oh
, ofconn
->flow_format
);
4033 /* We do not support the emergency flow cache. It will hopefully get
4034 * dropped from OpenFlow in the near future. */
4035 if (fm
.flags
& OFPFF_EMERG
) {
4036 /* There isn't a good fit for an error code, so just state that the
4037 * flow table is full. */
4038 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4041 error
= validate_actions(fm
.actions
, fm
.n_actions
,
4042 &fm
.cr
.flow
, p
->max_ports
);
4047 switch (fm
.command
) {
4049 return add_flow(ofconn
, &fm
);
4052 return modify_flows_loose(ofconn
, &fm
);
4054 case OFPFC_MODIFY_STRICT
:
4055 return modify_flow_strict(ofconn
, &fm
);
4058 delete_flows_loose(p
, &fm
);
4061 case OFPFC_DELETE_STRICT
:
4062 delete_flow_strict(p
, &fm
);
4066 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4071 handle_tun_id_from_cookie(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4073 const struct nxt_tun_id_cookie
*msg
4074 = (const struct nxt_tun_id_cookie
*) oh
;
4076 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4081 handle_role_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4083 struct nx_role_request
*nrr
= (struct nx_role_request
*) oh
;
4084 struct nx_role_request
*reply
;
4088 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4089 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4091 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4094 role
= ntohl(nrr
->role
);
4095 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4096 && role
!= NX_ROLE_SLAVE
) {
4097 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4099 /* There's no good error code for this. */
4100 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4103 if (role
== NX_ROLE_MASTER
) {
4104 struct ofconn
*other
;
4106 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4107 if (other
->role
== NX_ROLE_MASTER
) {
4108 other
->role
= NX_ROLE_SLAVE
;
4112 ofconn
->role
= role
;
4114 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, oh
->xid
, &buf
);
4115 reply
->role
= htonl(role
);
4116 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4122 handle_nxt_set_flow_format(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4124 const struct nxt_set_flow_format
*msg
4125 = (const struct nxt_set_flow_format
*) oh
;
4128 format
= ntohl(msg
->format
);
4129 if (format
== NXFF_OPENFLOW10
4130 || format
== NXFF_TUN_ID_FROM_COOKIE
4131 || format
== NXFF_NXM
) {
4132 ofconn
->flow_format
= format
;
4135 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4140 handle_barrier_request(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
4142 struct ofp_header
*ob
;
4145 /* Currently, everything executes synchronously, so we can just
4146 * immediately send the barrier reply. */
4147 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4148 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4153 handle_openflow__(struct ofconn
*ofconn
, const struct ofpbuf
*msg
)
4155 const struct ofp_header
*oh
= msg
->data
;
4156 const struct ofputil_msg_type
*type
;
4159 error
= ofputil_decode_msg_type(oh
, &type
);
4164 switch (ofputil_msg_type_code(type
)) {
4165 /* OpenFlow requests. */
4166 case OFPUTIL_OFPT_ECHO_REQUEST
:
4167 return handle_echo_request(ofconn
, oh
);
4169 case OFPUTIL_OFPT_FEATURES_REQUEST
:
4170 return handle_features_request(ofconn
, oh
);
4172 case OFPUTIL_OFPT_GET_CONFIG_REQUEST
:
4173 return handle_get_config_request(ofconn
, oh
);
4175 case OFPUTIL_OFPT_SET_CONFIG
:
4176 return handle_set_config(ofconn
, msg
->data
);
4178 case OFPUTIL_OFPT_PACKET_OUT
:
4179 return handle_packet_out(ofconn
, oh
);
4181 case OFPUTIL_OFPT_PORT_MOD
:
4182 return handle_port_mod(ofconn
, oh
);
4184 case OFPUTIL_OFPT_FLOW_MOD
:
4185 return handle_flow_mod(ofconn
, oh
);
4187 case OFPUTIL_OFPT_BARRIER_REQUEST
:
4188 return handle_barrier_request(ofconn
, oh
);
4190 /* OpenFlow replies. */
4191 case OFPUTIL_OFPT_ECHO_REPLY
:
4194 /* Nicira extension requests. */
4195 case OFPUTIL_NXT_STATUS_REQUEST
:
4196 return switch_status_handle_request(
4197 ofconn
->ofproto
->switch_status
, ofconn
->rconn
, oh
);
4199 case OFPUTIL_NXT_TUN_ID_FROM_COOKIE
:
4200 return handle_tun_id_from_cookie(ofconn
, oh
);
4202 case OFPUTIL_NXT_ROLE_REQUEST
:
4203 return handle_role_request(ofconn
, oh
);
4205 case OFPUTIL_NXT_SET_FLOW_FORMAT
:
4206 return handle_nxt_set_flow_format(ofconn
, oh
);
4208 case OFPUTIL_NXT_FLOW_MOD
:
4209 return handle_flow_mod(ofconn
, oh
);
4211 /* OpenFlow statistics requests. */
4212 case OFPUTIL_OFPST_DESC_REQUEST
:
4213 return handle_desc_stats_request(ofconn
, oh
);
4215 case OFPUTIL_OFPST_FLOW_REQUEST
:
4216 return handle_flow_stats_request(ofconn
, oh
);
4218 case OFPUTIL_OFPST_AGGREGATE_REQUEST
:
4219 return handle_aggregate_stats_request(ofconn
, oh
);
4221 case OFPUTIL_OFPST_TABLE_REQUEST
:
4222 return handle_table_stats_request(ofconn
, oh
);
4224 case OFPUTIL_OFPST_PORT_REQUEST
:
4225 return handle_port_stats_request(ofconn
, oh
);
4227 case OFPUTIL_OFPST_QUEUE_REQUEST
:
4228 return handle_queue_stats_request(ofconn
, oh
);
4230 /* Nicira extension statistics requests. */
4231 case OFPUTIL_NXST_FLOW_REQUEST
:
4232 return handle_nxst_flow(ofconn
, oh
);
4234 case OFPUTIL_NXST_AGGREGATE_REQUEST
:
4235 return handle_nxst_aggregate(ofconn
, oh
);
4237 case OFPUTIL_INVALID
:
4238 case OFPUTIL_OFPT_HELLO
:
4239 case OFPUTIL_OFPT_ERROR
:
4240 case OFPUTIL_OFPT_FEATURES_REPLY
:
4241 case OFPUTIL_OFPT_GET_CONFIG_REPLY
:
4242 case OFPUTIL_OFPT_PACKET_IN
:
4243 case OFPUTIL_OFPT_FLOW_REMOVED
:
4244 case OFPUTIL_OFPT_PORT_STATUS
:
4245 case OFPUTIL_OFPT_BARRIER_REPLY
:
4246 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REQUEST
:
4247 case OFPUTIL_OFPT_QUEUE_GET_CONFIG_REPLY
:
4248 case OFPUTIL_OFPST_DESC_REPLY
:
4249 case OFPUTIL_OFPST_FLOW_REPLY
:
4250 case OFPUTIL_OFPST_QUEUE_REPLY
:
4251 case OFPUTIL_OFPST_PORT_REPLY
:
4252 case OFPUTIL_OFPST_TABLE_REPLY
:
4253 case OFPUTIL_OFPST_AGGREGATE_REPLY
:
4254 case OFPUTIL_NXT_STATUS_REPLY
:
4255 case OFPUTIL_NXT_ROLE_REPLY
:
4256 case OFPUTIL_NXT_FLOW_REMOVED
:
4257 case OFPUTIL_NXST_FLOW_REPLY
:
4258 case OFPUTIL_NXST_AGGREGATE_REPLY
:
4260 if (VLOG_IS_WARN_ENABLED()) {
4261 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4262 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4265 if (oh
->type
== OFPT_STATS_REQUEST
|| oh
->type
== OFPT_STATS_REPLY
) {
4266 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
4268 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4274 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4276 int error
= handle_openflow__(ofconn
, ofp_msg
);
4278 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4280 COVERAGE_INC(ofproto_recv_openflow
);
4284 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4286 struct odp_msg
*msg
= packet
->data
;
4287 struct ofpbuf payload
;
4288 struct facet
*facet
;
4291 payload
.data
= msg
+ 1;
4292 payload
.size
= msg
->length
- sizeof *msg
;
4293 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4295 packet
->l2
= payload
.l2
;
4296 packet
->l3
= payload
.l3
;
4297 packet
->l4
= payload
.l4
;
4298 packet
->l7
= payload
.l7
;
4300 /* Check with in-band control to see if this packet should be sent
4301 * to the local port regardless of the flow table. */
4302 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4303 union odp_action action
;
4305 memset(&action
, 0, sizeof(action
));
4306 action
.output
.type
= ODPAT_OUTPUT
;
4307 action
.output
.port
= ODPP_LOCAL
;
4308 dpif_execute(p
->dpif
, &action
, 1, &payload
);
4311 facet
= facet_lookup_valid(p
, &flow
);
4313 struct rule
*rule
= rule_lookup(p
, &flow
);
4315 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4316 struct ofport
*port
= get_port(p
, msg
->port
);
4318 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4319 COVERAGE_INC(ofproto_no_packet_in
);
4320 /* XXX install 'drop' flow entry */
4321 ofpbuf_delete(packet
);
4325 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4329 COVERAGE_INC(ofproto_packet_in
);
4330 send_packet_in(p
, packet
);
4334 facet
= facet_create(p
, rule
, &flow
, packet
);
4335 } else if (!facet
->may_install
) {
4336 /* The facet is not installable, that is, we need to process every
4337 * packet, so process the current packet's actions into 'facet'. */
4338 facet_make_actions(p
, facet
, packet
);
4341 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4343 * Extra-special case for fail-open mode.
4345 * We are in fail-open mode and the packet matched the fail-open rule,
4346 * but we are connected to a controller too. We should send the packet
4347 * up to the controller in the hope that it will try to set up a flow
4348 * and thereby allow us to exit fail-open.
4350 * See the top-level comment in fail-open.c for more information.
4352 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4353 DPIF_RECV_MSG_PADDING
));
4356 ofpbuf_pull(packet
, sizeof *msg
);
4357 facet_execute(p
, facet
, packet
);
4358 facet_install(p
, facet
, false);
4362 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4364 struct odp_msg
*msg
= packet
->data
;
4366 switch (msg
->type
) {
4367 case _ODPL_ACTION_NR
:
4368 COVERAGE_INC(ofproto_ctlr_action
);
4369 send_packet_in(p
, packet
);
4372 case _ODPL_SFLOW_NR
:
4374 ofproto_sflow_received(p
->sflow
, msg
);
4376 ofpbuf_delete(packet
);
4380 handle_odp_miss_msg(p
, packet
);
4384 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4390 /* Flow expiration. */
4392 static int ofproto_dp_max_idle(const struct ofproto
*);
4393 static void ofproto_update_used(struct ofproto
*);
4394 static void rule_expire(struct ofproto
*, struct rule
*);
4395 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4397 /* This function is called periodically by ofproto_run(). Its job is to
4398 * collect updates for the flows that have been installed into the datapath,
4399 * most importantly when they last were used, and then use that information to
4400 * expire flows that have not been used recently.
4402 * Returns the number of milliseconds after which it should be called again. */
4404 ofproto_expire(struct ofproto
*ofproto
)
4406 struct rule
*rule
, *next_rule
;
4407 struct cls_cursor cursor
;
4410 /* Update 'used' for each flow in the datapath. */
4411 ofproto_update_used(ofproto
);
4413 /* Expire facets that have been idle too long. */
4414 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4415 ofproto_expire_facets(ofproto
, dp_max_idle
);
4417 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4418 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4419 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4420 rule_expire(ofproto
, rule
);
4423 /* Let the hook know that we're at a stable point: all outstanding data
4424 * in existing flows has been accounted to the account_cb. Thus, the
4425 * hook can now reasonably do operations that depend on having accurate
4426 * flow volume accounting (currently, that's just bond rebalancing). */
4427 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4428 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4431 return MIN(dp_max_idle
, 1000);
4434 /* Update 'used' member of installed facets. */
4436 ofproto_update_used(struct ofproto
*p
)
4438 struct odp_flow
*flows
;
4443 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4448 for (i
= 0; i
< n_flows
; i
++) {
4449 struct odp_flow
*f
= &flows
[i
];
4450 struct facet
*facet
;
4453 odp_flow_key_to_flow(&f
->key
, &flow
);
4454 facet
= facet_find(p
, &flow
);
4456 if (facet
&& facet
->installed
) {
4457 facet_update_time(p
, facet
, &f
->stats
);
4458 facet_account(p
, facet
, f
->stats
.n_bytes
);
4460 /* There's a flow in the datapath that we know nothing about.
4462 COVERAGE_INC(ofproto_unexpected_rule
);
4463 dpif_flow_del(p
->dpif
, f
);
4470 /* Calculates and returns the number of milliseconds of idle time after which
4471 * facets should expire from the datapath and we should fold their statistics
4472 * into their parent rules in userspace. */
4474 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4477 * Idle time histogram.
4479 * Most of the time a switch has a relatively small number of facets. When
4480 * this is the case we might as well keep statistics for all of them in
4481 * userspace and to cache them in the kernel datapath for performance as
4484 * As the number of facets increases, the memory required to maintain
4485 * statistics about them in userspace and in the kernel becomes
4486 * significant. However, with a large number of facets it is likely that
4487 * only a few of them are "heavy hitters" that consume a large amount of
4488 * bandwidth. At this point, only heavy hitters are worth caching in the
4489 * kernel and maintaining in userspaces; other facets we can discard.
4491 * The technique used to compute the idle time is to build a histogram with
4492 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4493 * that is installed in the kernel gets dropped in the appropriate bucket.
4494 * After the histogram has been built, we compute the cutoff so that only
4495 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4496 * cached. At least the most-recently-used bucket of facets is kept, so
4497 * actually an arbitrary number of facets can be kept in any given
4498 * expiration run (though the next run will delete most of those unless
4499 * they receive additional data).
4501 * This requires a second pass through the facets, in addition to the pass
4502 * made by ofproto_update_used(), because the former function never looks
4503 * at uninstallable facets.
4505 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4506 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4507 int buckets
[N_BUCKETS
] = { 0 };
4508 struct facet
*facet
;
4513 total
= hmap_count(&ofproto
->facets
);
4514 if (total
<= 1000) {
4515 return N_BUCKETS
* BUCKET_WIDTH
;
4518 /* Build histogram. */
4520 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4521 long long int idle
= now
- facet
->used
;
4522 int bucket
= (idle
<= 0 ? 0
4523 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4524 : (unsigned int) idle
/ BUCKET_WIDTH
);
4528 /* Find the first bucket whose flows should be expired. */
4529 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4530 if (buckets
[bucket
]) {
4533 subtotal
+= buckets
[bucket
++];
4534 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4539 if (VLOG_IS_DBG_ENABLED()) {
4543 ds_put_cstr(&s
, "keep");
4544 for (i
= 0; i
< N_BUCKETS
; i
++) {
4546 ds_put_cstr(&s
, ", drop");
4549 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4552 VLOG_INFO("%s: %s (msec:count)",
4553 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4557 return bucket
* BUCKET_WIDTH
;
4561 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4563 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4564 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4565 struct ofexpired expired
;
4566 struct odp_flow odp_flow
;
4568 /* Get updated flow stats.
4570 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4571 * updated TCP flags and (2) the dpif_flow_list_all() in
4572 * ofproto_update_used() zeroed TCP flags. */
4573 memset(&odp_flow
, 0, sizeof odp_flow
);
4574 if (facet
->installed
) {
4575 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
4576 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4577 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4579 if (odp_flow
.stats
.n_packets
) {
4580 facet_update_time(ofproto
, facet
, &odp_flow
.stats
);
4581 netflow_flow_update_flags(&facet
->nf_flow
,
4582 odp_flow
.stats
.tcp_flags
);
4586 expired
.flow
= facet
->flow
;
4587 expired
.packet_count
= facet
->packet_count
+
4588 odp_flow
.stats
.n_packets
;
4589 expired
.byte_count
= facet
->byte_count
+ odp_flow
.stats
.n_bytes
;
4590 expired
.used
= facet
->used
;
4592 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4597 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4599 long long int cutoff
= time_msec() - dp_max_idle
;
4600 struct facet
*facet
, *next_facet
;
4602 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4603 facet_active_timeout(ofproto
, facet
);
4604 if (facet
->used
< cutoff
) {
4605 facet_remove(ofproto
, facet
);
4610 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4611 * then delete it entirely. */
4613 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4615 struct facet
*facet
, *next_facet
;
4619 /* Has 'rule' expired? */
4621 if (rule
->hard_timeout
4622 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4623 reason
= OFPRR_HARD_TIMEOUT
;
4624 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4625 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4626 reason
= OFPRR_IDLE_TIMEOUT
;
4631 COVERAGE_INC(ofproto_expired
);
4633 /* Update stats. (This is a no-op if the rule expired due to an idle
4634 * timeout, because that only happens when the rule has no facets left.) */
4635 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4636 facet_remove(ofproto
, facet
);
4639 /* Get rid of the rule. */
4640 if (!rule_is_hidden(rule
)) {
4641 rule_send_removed(ofproto
, rule
, reason
);
4643 rule_remove(ofproto
, rule
);
4646 static struct ofpbuf
*
4647 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4650 struct ofp_flow_removed
*ofr
;
4653 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4654 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofr
->match
);
4655 ofr
->cookie
= rule
->flow_cookie
;
4656 ofr
->priority
= htons(rule
->cr
.priority
);
4657 ofr
->reason
= reason
;
4658 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4659 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4660 ofr
->packet_count
= htonll(rule
->packet_count
);
4661 ofr
->byte_count
= htonll(rule
->byte_count
);
4666 static struct ofpbuf
*
4667 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4669 struct nx_flow_removed
*nfr
;
4673 nfr
= make_nxmsg(sizeof *nfr
, NXT_FLOW_REMOVED
, &buf
);
4675 match_len
= nx_put_match(buf
, &rule
->cr
);
4677 nfr
->cookie
= rule
->flow_cookie
;
4678 nfr
->priority
= htons(rule
->cr
.priority
);
4679 nfr
->reason
= reason
;
4680 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4681 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4682 nfr
->match_len
= htons(match_len
);
4683 nfr
->packet_count
= htonll(rule
->packet_count
);
4684 nfr
->byte_count
= htonll(rule
->byte_count
);
4690 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4692 struct ofconn
*ofconn
;
4694 if (!rule
->send_flow_removed
) {
4698 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4701 if (!rconn_is_connected(ofconn
->rconn
)
4702 || !ofconn_receives_async_msgs(ofconn
)) {
4706 msg
= (ofconn
->flow_format
== NXFF_NXM
4707 ? compose_nx_flow_removed(rule
, reason
)
4708 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4710 /* Account flow expirations under ofconn->reply_counter, the counter
4711 * for replies to OpenFlow requests. That works because preventing
4712 * OpenFlow requests from being processed also prevents new flows from
4713 * being added (and expiring). (It also prevents processing OpenFlow
4714 * requests that would not add new flows, so it is imperfect.) */
4715 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4719 /* pinsched callback for sending 'packet' on 'ofconn'. */
4721 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4723 struct ofconn
*ofconn
= ofconn_
;
4725 rconn_send_with_limit(ofconn
->rconn
, packet
,
4726 ofconn
->packet_in_counter
, 100);
4729 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4730 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4731 * packet scheduler for sending.
4733 * 'max_len' specifies the maximum number of bytes of the packet to send on
4734 * 'ofconn' (INT_MAX specifies no limit).
4736 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4737 * ownership is transferred to this function. */
4739 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4742 struct ofproto
*ofproto
= ofconn
->ofproto
;
4743 struct ofp_packet_in
*opi
= packet
->data
;
4744 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4745 int send_len
, trim_size
;
4749 if (opi
->reason
== OFPR_ACTION
) {
4750 buffer_id
= UINT32_MAX
;
4751 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4752 buffer_id
= pktbuf_get_null();
4753 } else if (!ofconn
->pktbuf
) {
4754 buffer_id
= UINT32_MAX
;
4756 struct ofpbuf payload
;
4757 payload
.data
= opi
->data
;
4758 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4759 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4762 /* Figure out how much of the packet to send. */
4763 send_len
= ntohs(opi
->total_len
);
4764 if (buffer_id
!= UINT32_MAX
) {
4765 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4767 send_len
= MIN(send_len
, max_len
);
4769 /* Adjust packet length and clone if necessary. */
4770 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4772 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4775 packet
->size
= trim_size
;
4778 /* Update packet headers. */
4779 opi
->buffer_id
= htonl(buffer_id
);
4780 update_openflow_length(packet
);
4782 /* Hand over to packet scheduler. It might immediately call into
4783 * do_send_packet_in() or it might buffer it for a while (until a later
4784 * call to pinsched_run()). */
4785 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4786 packet
, do_send_packet_in
, ofconn
);
4789 /* Replace struct odp_msg header in 'packet' by equivalent struct
4790 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4791 * returned by dpif_recv()).
4793 * The conversion is not complete: the caller still needs to trim any unneeded
4794 * payload off the end of the buffer, set the length in the OpenFlow header,
4795 * and set buffer_id. Those require us to know the controller settings and so
4796 * must be done on a per-controller basis.
4798 * Returns the maximum number of bytes of the packet that should be sent to
4799 * the controller (INT_MAX if no limit). */
4801 do_convert_to_packet_in(struct ofpbuf
*packet
)
4803 struct odp_msg
*msg
= packet
->data
;
4804 struct ofp_packet_in
*opi
;
4810 /* Extract relevant header fields */
4811 if (msg
->type
== _ODPL_ACTION_NR
) {
4812 reason
= OFPR_ACTION
;
4815 reason
= OFPR_NO_MATCH
;
4818 total_len
= msg
->length
- sizeof *msg
;
4819 in_port
= odp_port_to_ofp_port(msg
->port
);
4821 /* Repurpose packet buffer by overwriting header. */
4822 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4823 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4824 opi
->header
.version
= OFP_VERSION
;
4825 opi
->header
.type
= OFPT_PACKET_IN
;
4826 opi
->total_len
= htons(total_len
);
4827 opi
->in_port
= htons(in_port
);
4828 opi
->reason
= reason
;
4833 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4834 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4835 * as necessary according to their individual configurations.
4837 * 'packet' must have sufficient headroom to convert it into a struct
4838 * ofp_packet_in (e.g. as returned by dpif_recv()).
4840 * Takes ownership of 'packet'. */
4842 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4844 struct ofconn
*ofconn
, *prev
;
4847 max_len
= do_convert_to_packet_in(packet
);
4850 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4851 if (ofconn_receives_async_msgs(ofconn
)) {
4853 schedule_packet_in(prev
, packet
, max_len
, true);
4859 schedule_packet_in(prev
, packet
, max_len
, false);
4861 ofpbuf_delete(packet
);
4866 pick_datapath_id(const struct ofproto
*ofproto
)
4868 const struct ofport
*port
;
4870 port
= get_port(ofproto
, ODPP_LOCAL
);
4872 uint8_t ea
[ETH_ADDR_LEN
];
4875 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4877 return eth_addr_to_uint64(ea
);
4879 VLOG_WARN("could not get MAC address for %s (%s)",
4880 netdev_get_name(port
->netdev
), strerror(error
));
4882 return ofproto
->fallback_dpid
;
4886 pick_fallback_dpid(void)
4888 uint8_t ea
[ETH_ADDR_LEN
];
4889 eth_addr_nicira_random(ea
);
4890 return eth_addr_to_uint64(ea
);
4894 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
4895 struct odp_actions
*actions
, tag_type
*tags
,
4896 uint16_t *nf_output_iface
, void *ofproto_
)
4898 struct ofproto
*ofproto
= ofproto_
;
4901 /* Drop frames for reserved multicast addresses. */
4902 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4906 /* Learn source MAC (but don't try to learn from revalidation). */
4907 if (packet
!= NULL
) {
4908 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4910 GRAT_ARP_LOCK_NONE
);
4912 /* The log messages here could actually be useful in debugging,
4913 * so keep the rate limit relatively high. */
4914 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4915 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4916 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4917 ofproto_revalidate(ofproto
, rev_tag
);
4921 /* Determine output port. */
4922 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
4925 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
4926 nf_output_iface
, actions
);
4927 } else if (out_port
!= flow
->in_port
) {
4928 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4929 *nf_output_iface
= out_port
;
4937 static const struct ofhooks default_ofhooks
= {
4938 default_normal_ofhook_cb
,