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 #include "sflow_api.h"
69 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
70 struct netdev
*netdev
;
71 struct ofp_phy_port opp
; /* In host byte order. */
75 static void ofport_free(struct ofport
*);
76 static void hton_ofp_phy_port(struct ofp_phy_port
*);
78 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
79 const struct flow
*, struct ofproto
*,
80 const struct ofpbuf
*packet
,
81 struct odp_actions
*out
, tag_type
*tags
,
82 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
84 /* An OpenFlow flow. */
86 long long int used
; /* Time last used; time created if not used. */
87 long long int created
; /* Creation time. */
91 * - Do include packets and bytes from facets that have been deleted or
92 * whose own statistics have been folded into the rule.
94 * - Do include packets and bytes sent "by hand" that were accounted to
95 * the rule without any facet being involved (this is a rare corner
96 * case in rule_execute()).
98 * - Do not include packet or bytes that can be obtained from any facet's
99 * packet_count or byte_count member or that can be obtained from the
100 * datapath by, e.g., dpif_flow_get() for any facet.
102 uint64_t packet_count
; /* Number of packets received. */
103 uint64_t byte_count
; /* Number of bytes received. */
105 ovs_be64 flow_cookie
; /* Controller-issued identifier. */
107 struct cls_rule cr
; /* In owning ofproto's classifier. */
108 uint16_t idle_timeout
; /* In seconds from time of last use. */
109 uint16_t hard_timeout
; /* In seconds from time of creation. */
110 bool send_flow_removed
; /* Send a flow removed message? */
111 int n_actions
; /* Number of elements in actions[]. */
112 union ofp_action
*actions
; /* OpenFlow actions. */
113 struct list facets
; /* List of "struct facet"s. */
116 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
117 static bool rule_is_hidden(const struct rule
*);
119 static struct rule
*rule_create(const struct cls_rule
*,
120 const union ofp_action
*, size_t n_actions
,
121 uint16_t idle_timeout
, uint16_t hard_timeout
,
122 ovs_be64 flow_cookie
, bool send_flow_removed
);
123 static void rule_destroy(struct ofproto
*, struct rule
*);
124 static void rule_free(struct rule
*);
126 static struct rule
*rule_lookup(struct ofproto
*, const struct flow
*);
127 static void rule_insert(struct ofproto
*, struct rule
*);
128 static void rule_remove(struct ofproto
*, struct rule
*);
130 static void rule_send_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
132 /* An exact-match instantiation of an OpenFlow flow. */
134 long long int used
; /* Time last used; time created if not used. */
138 * - Do include packets and bytes sent "by hand", e.g. with
141 * - Do include packets and bytes that were obtained from the datapath
142 * when a flow was deleted (e.g. dpif_flow_del()) or when its
143 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
145 * - Do not include any packets or bytes that can currently be obtained
146 * from the datapath by, e.g., dpif_flow_get().
148 uint64_t packet_count
; /* Number of packets received. */
149 uint64_t byte_count
; /* Number of bytes received. */
151 /* Number of bytes passed to account_cb. This may include bytes that can
152 * currently obtained from the datapath (thus, it can be greater than
154 uint64_t accounted_bytes
;
156 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
157 struct list list_node
; /* In owning rule's 'facets' list. */
158 struct rule
*rule
; /* Owning rule. */
159 struct flow flow
; /* Exact-match flow. */
160 bool installed
; /* Installed in datapath? */
161 bool may_install
; /* True ordinarily; false if actions must
162 * be reassessed for every packet. */
163 int n_actions
; /* Number of elements in actions[]. */
164 union odp_action
*actions
; /* Datapath actions. */
165 tag_type tags
; /* Tags (set only by hooks). */
166 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
169 static struct facet
*facet_create(struct ofproto
*, struct rule
*,
171 const struct ofpbuf
*packet
);
172 static void facet_remove(struct ofproto
*, struct facet
*);
173 static void facet_free(struct facet
*);
175 static struct facet
*facet_lookup_valid(struct ofproto
*, const struct flow
*);
176 static bool facet_revalidate(struct ofproto
*, struct facet
*);
178 static void facet_install(struct ofproto
*, struct facet
*, bool zero_stats
);
179 static void facet_uninstall(struct ofproto
*, struct facet
*);
180 static void facet_flush_stats(struct ofproto
*, struct facet
*);
182 static void facet_make_actions(struct ofproto
*, struct facet
*,
183 const struct ofpbuf
*packet
);
184 static void facet_update_stats(struct ofproto
*, struct facet
*,
185 const struct odp_flow_stats
*);
187 /* ofproto supports two kinds of OpenFlow connections:
189 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
190 * maintains persistent connections to these controllers and by default
191 * sends them asynchronous messages such as packet-ins.
193 * - "Service" connections, e.g. from ovs-ofctl. When these connections
194 * drop, it is the other side's responsibility to reconnect them if
195 * necessary. ofproto does not send them asynchronous messages by default.
197 * Currently, active (tcp, ssl, unix) connections are always "primary"
198 * connections and passive (ptcp, pssl, punix) connections are always "service"
199 * connections. There is no inherent reason for this, but it reflects the
203 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
204 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
207 /* A listener for incoming OpenFlow "service" connections. */
209 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
210 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
212 /* These are not used by ofservice directly. They are settings for
213 * accepted "struct ofconn"s from the pvconn. */
214 int probe_interval
; /* Max idle time before probing, in seconds. */
215 int rate_limit
; /* Max packet-in rate in packets per second. */
216 int burst_limit
; /* Limit on accumulating packet credits. */
219 static struct ofservice
*ofservice_lookup(struct ofproto
*,
221 static int ofservice_create(struct ofproto
*,
222 const struct ofproto_controller
*);
223 static void ofservice_reconfigure(struct ofservice
*,
224 const struct ofproto_controller
*);
225 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
227 /* An OpenFlow connection. */
229 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
230 struct list node
; /* In struct ofproto's "all_conns" list. */
231 struct rconn
*rconn
; /* OpenFlow connection. */
232 enum ofconn_type type
; /* Type. */
233 int flow_format
; /* One of NXFF_*. */
235 /* OFPT_PACKET_IN related data. */
236 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
237 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
238 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
239 int miss_send_len
; /* Bytes to send of buffered packets. */
241 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
242 * requests, and the maximum number before we stop reading OpenFlow
244 #define OFCONN_REPLY_MAX 100
245 struct rconn_packet_counter
*reply_counter
;
247 /* type == OFCONN_PRIMARY only. */
248 enum nx_role role
; /* Role. */
249 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
250 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
251 struct status_category
*ss
; /* Switch status category. */
252 enum ofproto_band band
; /* In-band or out-of-band? */
255 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
256 * "schedulers" array. Their values are 0 and 1, and their meanings and values
257 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
258 * case anything ever changes, check their values here. */
259 #define N_SCHEDULERS 2
260 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
261 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
262 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
263 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
265 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
267 static void ofconn_destroy(struct ofconn
*);
268 static void ofconn_run(struct ofconn
*);
269 static void ofconn_wait(struct ofconn
*);
270 static bool ofconn_receives_async_msgs(const struct ofconn
*);
271 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
272 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
274 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
275 struct rconn_packet_counter
*counter
);
277 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
278 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
282 uint64_t datapath_id
; /* Datapath ID. */
283 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
284 char *mfr_desc
; /* Manufacturer. */
285 char *hw_desc
; /* Hardware. */
286 char *sw_desc
; /* Software version. */
287 char *serial_desc
; /* Serial number. */
288 char *dp_desc
; /* Datapath description. */
292 struct netdev_monitor
*netdev_monitor
;
293 struct hmap ports
; /* Contains "struct ofport"s. */
294 struct shash port_by_name
;
298 struct switch_status
*switch_status
;
299 struct fail_open
*fail_open
;
300 struct netflow
*netflow
;
301 struct ofproto_sflow
*sflow
;
303 /* In-band control. */
304 struct in_band
*in_band
;
305 long long int next_in_band_update
;
306 struct sockaddr_in
*extra_in_band_remotes
;
307 size_t n_extra_remotes
;
311 struct classifier cls
;
312 long long int next_expiration
;
316 bool need_revalidate
;
317 struct tag_set revalidate_set
;
319 /* OpenFlow connections. */
320 struct hmap controllers
; /* Controller "struct ofconn"s. */
321 struct list all_conns
; /* Contains "struct ofconn"s. */
322 enum ofproto_fail_mode fail_mode
;
324 /* OpenFlow listeners. */
325 struct hmap services
; /* Contains "struct ofservice"s. */
326 struct pvconn
**snoops
;
329 /* Hooks for ovs-vswitchd. */
330 const struct ofhooks
*ofhooks
;
333 /* Used by default ofhooks. */
334 struct mac_learning
*ml
;
337 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
339 static const struct ofhooks default_ofhooks
;
341 static uint64_t pick_datapath_id(const struct ofproto
*);
342 static uint64_t pick_fallback_dpid(void);
344 static int ofproto_expire(struct ofproto
*);
346 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
348 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
350 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
351 static void update_port(struct ofproto
*, const char *devname
);
352 static int init_ports(struct ofproto
*);
353 static void reinit_ports(struct ofproto
*);
356 ofproto_create(const char *datapath
, const char *datapath_type
,
357 const struct ofhooks
*ofhooks
, void *aux
,
358 struct ofproto
**ofprotop
)
360 struct odp_stats stats
;
367 /* Connect to datapath and start listening for messages. */
368 error
= dpif_open(datapath
, datapath_type
, &dpif
);
370 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
373 error
= dpif_get_dp_stats(dpif
, &stats
);
375 VLOG_ERR("failed to obtain stats for datapath %s: %s",
376 datapath
, strerror(error
));
380 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
382 VLOG_ERR("failed to listen on datapath %s: %s",
383 datapath
, strerror(error
));
387 dpif_flow_flush(dpif
);
388 dpif_recv_purge(dpif
);
390 /* Initialize settings. */
391 p
= xzalloc(sizeof *p
);
392 p
->fallback_dpid
= pick_fallback_dpid();
393 p
->datapath_id
= p
->fallback_dpid
;
394 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
395 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
396 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
397 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
398 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
400 /* Initialize datapath. */
402 p
->netdev_monitor
= netdev_monitor_create();
403 hmap_init(&p
->ports
);
404 shash_init(&p
->port_by_name
);
405 p
->max_ports
= stats
.max_ports
;
407 /* Initialize submodules. */
408 p
->switch_status
= switch_status_create(p
);
413 /* Initialize in-band control. */
415 p
->in_band_queue
= -1;
417 /* Initialize flow table. */
418 classifier_init(&p
->cls
);
419 p
->next_expiration
= time_msec() + 1000;
421 /* Initialize facet table. */
422 hmap_init(&p
->facets
);
423 p
->need_revalidate
= false;
424 tag_set_init(&p
->revalidate_set
);
426 /* Initialize OpenFlow connections. */
427 list_init(&p
->all_conns
);
428 hmap_init(&p
->controllers
);
429 hmap_init(&p
->services
);
433 /* Initialize hooks. */
435 p
->ofhooks
= ofhooks
;
439 p
->ofhooks
= &default_ofhooks
;
441 p
->ml
= mac_learning_create();
444 /* Pick final datapath ID. */
445 p
->datapath_id
= pick_datapath_id(p
);
446 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
453 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
455 uint64_t old_dpid
= p
->datapath_id
;
456 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
457 if (p
->datapath_id
!= old_dpid
) {
458 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
460 /* Force all active connections to reconnect, since there is no way to
461 * notify a controller that the datapath ID has changed. */
462 ofproto_reconnect_controllers(p
);
467 is_discovery_controller(const struct ofproto_controller
*c
)
469 return !strcmp(c
->target
, "discover");
473 is_in_band_controller(const struct ofproto_controller
*c
)
475 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
478 /* Creates a new controller in 'ofproto'. Some of the settings are initially
479 * drawn from 'c', but update_controller() needs to be called later to finish
480 * the new ofconn's configuration. */
482 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
484 struct discovery
*discovery
;
485 struct ofconn
*ofconn
;
487 if (is_discovery_controller(c
)) {
488 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
489 ofproto
->dpif
, ofproto
->switch_status
,
498 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
499 ofconn
->pktbuf
= pktbuf_create();
500 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
502 ofconn
->discovery
= discovery
;
504 char *name
= ofconn_make_name(ofproto
, c
->target
);
505 rconn_connect(ofconn
->rconn
, c
->target
, name
);
508 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
509 hash_string(c
->target
, 0));
512 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
513 * target or turn discovery on or off (these are done by creating new ofconns
514 * and deleting old ones), but it can update the rest of an ofconn's
517 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
521 ofconn
->band
= (is_in_band_controller(c
)
522 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
524 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
526 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
527 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
529 if (ofconn
->discovery
) {
530 discovery_set_update_resolv_conf(ofconn
->discovery
,
531 c
->update_resolv_conf
);
532 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
535 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
539 ofconn_get_target(const struct ofconn
*ofconn
)
541 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
544 static struct ofconn
*
545 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
547 struct ofconn
*ofconn
;
549 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
550 hash_string(target
, 0), &ofproto
->controllers
) {
551 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
559 update_in_band_remotes(struct ofproto
*ofproto
)
561 const struct ofconn
*ofconn
;
562 struct sockaddr_in
*addrs
;
563 size_t max_addrs
, n_addrs
;
567 /* Allocate enough memory for as many remotes as we could possibly have. */
568 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
569 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
572 /* Add all the remotes. */
574 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
575 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
577 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
581 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
582 if (sin
->sin_addr
.s_addr
) {
583 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
586 if (ofconn
->discovery
) {
590 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
591 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
594 /* Create or update or destroy in-band.
596 * Ordinarily we only enable in-band if there's at least one remote
597 * address, but discovery needs the in-band rules for DHCP to be installed
598 * even before we know any remote addresses. */
599 if (n_addrs
|| discovery
) {
600 if (!ofproto
->in_band
) {
601 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
604 if (ofproto
->in_band
) {
605 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
607 in_band_set_queue(ofproto
->in_band
, ofproto
->in_band_queue
);
608 ofproto
->next_in_band_update
= time_msec() + 1000;
610 in_band_destroy(ofproto
->in_band
);
611 ofproto
->in_band
= NULL
;
619 update_fail_open(struct ofproto
*p
)
621 struct ofconn
*ofconn
;
623 if (!hmap_is_empty(&p
->controllers
)
624 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
625 struct rconn
**rconns
;
629 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
633 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
634 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
635 rconns
[n
++] = ofconn
->rconn
;
638 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
639 /* p->fail_open takes ownership of 'rconns'. */
641 fail_open_destroy(p
->fail_open
);
647 ofproto_set_controllers(struct ofproto
*p
,
648 const struct ofproto_controller
*controllers
,
649 size_t n_controllers
)
651 struct shash new_controllers
;
652 struct ofconn
*ofconn
, *next_ofconn
;
653 struct ofservice
*ofservice
, *next_ofservice
;
657 /* Create newly configured controllers and services.
658 * Create a name to ofproto_controller mapping in 'new_controllers'. */
659 shash_init(&new_controllers
);
660 for (i
= 0; i
< n_controllers
; i
++) {
661 const struct ofproto_controller
*c
= &controllers
[i
];
663 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
664 if (!find_controller_by_target(p
, c
->target
)) {
665 add_controller(p
, c
);
667 } else if (!pvconn_verify_name(c
->target
)) {
668 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
672 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
673 dpif_name(p
->dpif
), c
->target
);
677 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
680 /* Delete controllers that are no longer configured.
681 * Update configuration of all now-existing controllers. */
683 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
684 struct ofproto_controller
*c
;
686 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
688 ofconn_destroy(ofconn
);
690 update_controller(ofconn
, c
);
697 /* Delete services that are no longer configured.
698 * Update configuration of all now-existing services. */
699 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
700 struct ofproto_controller
*c
;
702 c
= shash_find_data(&new_controllers
,
703 pvconn_get_name(ofservice
->pvconn
));
705 ofservice_destroy(p
, ofservice
);
707 ofservice_reconfigure(ofservice
, c
);
711 shash_destroy(&new_controllers
);
713 update_in_band_remotes(p
);
716 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
717 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
718 struct ofconn
, hmap_node
);
719 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
720 rconn_status_cb
, ofconn
->rconn
);
725 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
727 p
->fail_mode
= fail_mode
;
731 /* Drops the connections between 'ofproto' and all of its controllers, forcing
732 * them to reconnect. */
734 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
736 struct ofconn
*ofconn
;
738 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
739 rconn_reconnect(ofconn
->rconn
);
744 any_extras_changed(const struct ofproto
*ofproto
,
745 const struct sockaddr_in
*extras
, size_t n
)
749 if (n
!= ofproto
->n_extra_remotes
) {
753 for (i
= 0; i
< n
; i
++) {
754 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
755 const struct sockaddr_in
*new = &extras
[i
];
757 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
758 old
->sin_port
!= new->sin_port
) {
766 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
767 * in-band control should guarantee access, in the same way that in-band
768 * control guarantees access to OpenFlow controllers. */
770 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
771 const struct sockaddr_in
*extras
, size_t n
)
773 if (!any_extras_changed(ofproto
, extras
, n
)) {
777 free(ofproto
->extra_in_band_remotes
);
778 ofproto
->n_extra_remotes
= n
;
779 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
781 update_in_band_remotes(ofproto
);
784 /* Sets the OpenFlow queue used by flows set up by in-band control on
785 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
786 * flows will use the default queue. */
788 ofproto_set_in_band_queue(struct ofproto
*ofproto
, int queue_id
)
790 if (queue_id
!= ofproto
->in_band_queue
) {
791 ofproto
->in_band_queue
= queue_id
;
792 update_in_band_remotes(ofproto
);
797 ofproto_set_desc(struct ofproto
*p
,
798 const char *mfr_desc
, const char *hw_desc
,
799 const char *sw_desc
, const char *serial_desc
,
802 struct ofp_desc_stats
*ods
;
805 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
806 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
807 sizeof ods
->mfr_desc
);
810 p
->mfr_desc
= xstrdup(mfr_desc
);
813 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
814 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
815 sizeof ods
->hw_desc
);
818 p
->hw_desc
= xstrdup(hw_desc
);
821 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
822 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
823 sizeof ods
->sw_desc
);
826 p
->sw_desc
= xstrdup(sw_desc
);
829 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
830 VLOG_WARN("truncating serial_desc, must be less than %zu "
832 sizeof ods
->serial_num
);
834 free(p
->serial_desc
);
835 p
->serial_desc
= xstrdup(serial_desc
);
838 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
839 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
840 sizeof ods
->dp_desc
);
843 p
->dp_desc
= xstrdup(dp_desc
);
848 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
849 const struct svec
*svec
)
851 struct pvconn
**pvconns
= *pvconnsp
;
852 size_t n_pvconns
= *n_pvconnsp
;
856 for (i
= 0; i
< n_pvconns
; i
++) {
857 pvconn_close(pvconns
[i
]);
861 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
863 for (i
= 0; i
< svec
->n
; i
++) {
864 const char *name
= svec
->names
[i
];
865 struct pvconn
*pvconn
;
868 error
= pvconn_open(name
, &pvconn
);
870 pvconns
[n_pvconns
++] = pvconn
;
872 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
880 *n_pvconnsp
= n_pvconns
;
886 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
888 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
892 ofproto_set_netflow(struct ofproto
*ofproto
,
893 const struct netflow_options
*nf_options
)
895 if (nf_options
&& nf_options
->collectors
.n
) {
896 if (!ofproto
->netflow
) {
897 ofproto
->netflow
= netflow_create();
899 return netflow_set_options(ofproto
->netflow
, nf_options
);
901 netflow_destroy(ofproto
->netflow
);
902 ofproto
->netflow
= NULL
;
908 ofproto_set_sflow(struct ofproto
*ofproto
,
909 const struct ofproto_sflow_options
*oso
)
911 struct ofproto_sflow
*os
= ofproto
->sflow
;
914 struct ofport
*ofport
;
916 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
917 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
918 ofproto_sflow_add_port(os
, ofport
->odp_port
,
919 netdev_get_name(ofport
->netdev
));
922 ofproto_sflow_set_options(os
, oso
);
924 ofproto_sflow_destroy(os
);
925 ofproto
->sflow
= NULL
;
930 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
932 return ofproto
->datapath_id
;
936 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
938 return !hmap_is_empty(&ofproto
->controllers
);
941 enum ofproto_fail_mode
942 ofproto_get_fail_mode(const struct ofproto
*p
)
948 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
952 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
953 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
958 ofproto_destroy(struct ofproto
*p
)
960 struct ofservice
*ofservice
, *next_ofservice
;
961 struct ofconn
*ofconn
, *next_ofconn
;
962 struct ofport
*ofport
, *next_ofport
;
969 /* Destroy fail-open and in-band early, since they touch the classifier. */
970 fail_open_destroy(p
->fail_open
);
973 in_band_destroy(p
->in_band
);
975 free(p
->extra_in_band_remotes
);
977 ofproto_flush_flows(p
);
978 classifier_destroy(&p
->cls
);
979 hmap_destroy(&p
->facets
);
981 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
982 ofconn_destroy(ofconn
);
984 hmap_destroy(&p
->controllers
);
987 netdev_monitor_destroy(p
->netdev_monitor
);
988 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
989 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
992 shash_destroy(&p
->port_by_name
);
994 switch_status_destroy(p
->switch_status
);
995 netflow_destroy(p
->netflow
);
996 ofproto_sflow_destroy(p
->sflow
);
998 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
999 ofservice_destroy(p
, ofservice
);
1001 hmap_destroy(&p
->services
);
1003 for (i
= 0; i
< p
->n_snoops
; i
++) {
1004 pvconn_close(p
->snoops
[i
]);
1008 mac_learning_destroy(p
->ml
);
1013 free(p
->serial_desc
);
1016 hmap_destroy(&p
->ports
);
1022 ofproto_run(struct ofproto
*p
)
1024 int error
= ofproto_run1(p
);
1026 error
= ofproto_run2(p
, false);
1032 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1034 if (error
== ENOBUFS
) {
1035 reinit_ports(ofproto
);
1036 } else if (!error
) {
1037 update_port(ofproto
, devname
);
1042 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1043 * means that 'ofconn' is more interesting for monitoring than a lower return
1046 snoop_preference(const struct ofconn
*ofconn
)
1048 switch (ofconn
->role
) {
1049 case NX_ROLE_MASTER
:
1056 /* Shouldn't happen. */
1061 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1062 * Connects this vconn to a controller. */
1064 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1066 struct ofconn
*ofconn
, *best
;
1068 /* Pick a controller for monitoring. */
1070 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1071 if (ofconn
->type
== OFCONN_PRIMARY
1072 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1078 rconn_add_monitor(best
->rconn
, vconn
);
1080 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1086 ofproto_run1(struct ofproto
*p
)
1088 struct ofconn
*ofconn
, *next_ofconn
;
1089 struct ofservice
*ofservice
;
1094 if (shash_is_empty(&p
->port_by_name
)) {
1098 for (i
= 0; i
< 50; i
++) {
1101 error
= dpif_recv(p
->dpif
, &buf
);
1103 if (error
== ENODEV
) {
1104 /* Someone destroyed the datapath behind our back. The caller
1105 * better destroy us and give up, because we're just going to
1106 * spin from here on out. */
1107 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1108 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1109 dpif_name(p
->dpif
));
1115 handle_odp_msg(p
, buf
);
1118 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1119 process_port_change(p
, error
, devname
);
1121 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1122 &devname
)) != EAGAIN
) {
1123 process_port_change(p
, error
, devname
);
1127 if (time_msec() >= p
->next_in_band_update
) {
1128 update_in_band_remotes(p
);
1130 in_band_run(p
->in_band
);
1133 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1137 /* Fail-open maintenance. Do this after processing the ofconns since
1138 * fail-open checks the status of the controller rconn. */
1140 fail_open_run(p
->fail_open
);
1143 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1144 struct vconn
*vconn
;
1147 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1149 struct rconn
*rconn
;
1152 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1153 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1154 rconn_connect_unreliably(rconn
, vconn
, name
);
1157 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1158 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1159 ofservice
->burst_limit
);
1160 } else if (retval
!= EAGAIN
) {
1161 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1165 for (i
= 0; i
< p
->n_snoops
; i
++) {
1166 struct vconn
*vconn
;
1169 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1171 add_snooper(p
, vconn
);
1172 } else if (retval
!= EAGAIN
) {
1173 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1177 if (time_msec() >= p
->next_expiration
) {
1178 int delay
= ofproto_expire(p
);
1179 p
->next_expiration
= time_msec() + delay
;
1180 COVERAGE_INC(ofproto_expiration
);
1184 netflow_run(p
->netflow
);
1187 ofproto_sflow_run(p
->sflow
);
1194 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1196 /* Figure out what we need to revalidate now, if anything. */
1197 struct tag_set revalidate_set
= p
->revalidate_set
;
1198 if (p
->need_revalidate
) {
1199 revalidate_all
= true;
1202 /* Clear the revalidation flags. */
1203 tag_set_init(&p
->revalidate_set
);
1204 p
->need_revalidate
= false;
1206 /* Now revalidate if there's anything to do. */
1207 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1208 struct facet
*facet
, *next
;
1210 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1212 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1213 facet_revalidate(p
, facet
);
1222 ofproto_wait(struct ofproto
*p
)
1224 struct ofservice
*ofservice
;
1225 struct ofconn
*ofconn
;
1228 dpif_recv_wait(p
->dpif
);
1229 dpif_port_poll_wait(p
->dpif
);
1230 netdev_monitor_poll_wait(p
->netdev_monitor
);
1231 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1232 ofconn_wait(ofconn
);
1235 poll_timer_wait_until(p
->next_in_band_update
);
1236 in_band_wait(p
->in_band
);
1239 fail_open_wait(p
->fail_open
);
1242 ofproto_sflow_wait(p
->sflow
);
1244 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1245 poll_immediate_wake();
1247 if (p
->need_revalidate
) {
1248 /* Shouldn't happen, but if it does just go around again. */
1249 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1250 poll_immediate_wake();
1251 } else if (p
->next_expiration
!= LLONG_MAX
) {
1252 poll_timer_wait_until(p
->next_expiration
);
1254 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1255 pvconn_wait(ofservice
->pvconn
);
1257 for (i
= 0; i
< p
->n_snoops
; i
++) {
1258 pvconn_wait(p
->snoops
[i
]);
1263 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1265 tag_set_add(&ofproto
->revalidate_set
, tag
);
1269 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1271 return &ofproto
->revalidate_set
;
1275 ofproto_is_alive(const struct ofproto
*p
)
1277 return !hmap_is_empty(&p
->controllers
);
1280 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1282 * This is almost the same as calling dpif_port_del() directly on the
1283 * datapath, but it also makes 'ofproto' close its open netdev for the port
1284 * (if any). This makes it possible to create a new netdev of a different
1285 * type under the same name, which otherwise the netdev library would refuse
1286 * to do because of the conflict. (The netdev would eventually get closed on
1287 * the next trip through ofproto_run(), but this interface is more direct.)
1289 * Returns 0 if successful, otherwise a positive errno. */
1291 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1293 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1294 const char *name
= ofport
? (char *) ofport
->opp
.name
: "<unknown>";
1297 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1299 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1300 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1301 } else if (ofport
) {
1302 /* 'name' is ofport->opp.name and update_port() is going to destroy
1303 * 'ofport'. Just in case update_port() refers to 'name' after it
1304 * destroys 'ofport', make a copy of it around the update_port()
1306 char *devname
= xstrdup(name
);
1307 update_port(ofproto
, devname
);
1313 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1314 * true if 'odp_port' exists and should be included, false otherwise. */
1316 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1318 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1319 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1323 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1324 const union ofp_action
*actions
, size_t n_actions
,
1325 const struct ofpbuf
*packet
)
1327 struct odp_actions odp_actions
;
1330 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1336 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1338 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
, packet
);
1342 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1343 * performs the 'n_actions' actions in 'actions'. The new flow will not
1346 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1347 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1348 * controllers; otherwise, it will be hidden.
1350 * The caller retains ownership of 'cls_rule' and 'actions'. */
1352 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1353 const union ofp_action
*actions
, size_t n_actions
)
1356 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1357 rule_insert(p
, rule
);
1361 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1365 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1368 rule_remove(ofproto
, rule
);
1373 ofproto_flush_flows(struct ofproto
*ofproto
)
1375 struct facet
*facet
, *next_facet
;
1376 struct rule
*rule
, *next_rule
;
1377 struct cls_cursor cursor
;
1379 COVERAGE_INC(ofproto_flush
);
1381 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1382 /* Mark the facet as not installed so that facet_remove() doesn't
1383 * bother trying to uninstall it. There is no point in uninstalling it
1384 * individually since we are about to blow away all the facets with
1385 * dpif_flow_flush(). */
1386 facet
->installed
= false;
1387 facet_remove(ofproto
, facet
);
1390 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
1391 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
1392 rule_remove(ofproto
, rule
);
1395 dpif_flow_flush(ofproto
->dpif
);
1396 if (ofproto
->in_band
) {
1397 in_band_flushed(ofproto
->in_band
);
1399 if (ofproto
->fail_open
) {
1400 fail_open_flushed(ofproto
->fail_open
);
1405 reinit_ports(struct ofproto
*p
)
1407 struct svec devnames
;
1408 struct ofport
*ofport
;
1409 struct odp_port
*odp_ports
;
1413 COVERAGE_INC(ofproto_reinit_ports
);
1415 svec_init(&devnames
);
1416 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1417 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1419 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1420 for (i
= 0; i
< n_odp_ports
; i
++) {
1421 svec_add (&devnames
, odp_ports
[i
].devname
);
1425 svec_sort_unique(&devnames
);
1426 for (i
= 0; i
< devnames
.n
; i
++) {
1427 update_port(p
, devnames
.names
[i
]);
1429 svec_destroy(&devnames
);
1432 static struct ofport
*
1433 make_ofport(const struct odp_port
*odp_port
)
1435 struct netdev_options netdev_options
;
1436 enum netdev_flags flags
;
1437 struct ofport
*ofport
;
1438 struct netdev
*netdev
;
1441 memset(&netdev_options
, 0, sizeof netdev_options
);
1442 netdev_options
.name
= odp_port
->devname
;
1443 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1445 error
= netdev_open(&netdev_options
, &netdev
);
1447 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1448 "cannot be opened (%s)",
1449 odp_port
->devname
, odp_port
->port
,
1450 odp_port
->devname
, strerror(error
));
1454 ofport
= xmalloc(sizeof *ofport
);
1455 ofport
->netdev
= netdev
;
1456 ofport
->odp_port
= odp_port
->port
;
1457 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1458 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1459 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1460 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1461 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1463 netdev_get_flags(netdev
, &flags
);
1464 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1466 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1468 netdev_get_features(netdev
,
1469 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1470 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1475 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1477 if (get_port(p
, odp_port
->port
)) {
1478 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1481 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1482 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1491 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1493 const struct ofp_phy_port
*a
= &a_
->opp
;
1494 const struct ofp_phy_port
*b
= &b_
->opp
;
1496 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1497 return (a
->port_no
== b
->port_no
1498 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1499 && !strcmp((char *) a
->name
, (char *) b
->name
)
1500 && a
->state
== b
->state
1501 && a
->config
== b
->config
1502 && a
->curr
== b
->curr
1503 && a
->advertised
== b
->advertised
1504 && a
->supported
== b
->supported
1505 && a
->peer
== b
->peer
);
1509 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1512 /* XXX Should limit the number of queued port status change messages. */
1513 struct ofconn
*ofconn
;
1514 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1515 struct ofp_port_status
*ops
;
1518 /* Primary controllers, even slaves, should always get port status
1519 updates. Otherwise obey ofconn_receives_async_msgs(). */
1520 if (ofconn
->type
!= OFCONN_PRIMARY
1521 && !ofconn_receives_async_msgs(ofconn
)) {
1525 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1526 ops
->reason
= reason
;
1527 ops
->desc
= ofport
->opp
;
1528 hton_ofp_phy_port(&ops
->desc
);
1529 queue_tx(b
, ofconn
, NULL
);
1534 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1536 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1538 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1539 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1540 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1542 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1547 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1549 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1550 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1551 shash_delete(&p
->port_by_name
,
1552 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1554 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1559 ofport_free(struct ofport
*ofport
)
1562 netdev_close(ofport
->netdev
);
1567 static struct ofport
*
1568 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1570 struct ofport
*port
;
1572 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1573 hash_int(odp_port
, 0), &ofproto
->ports
) {
1574 if (port
->odp_port
== odp_port
) {
1582 update_port(struct ofproto
*p
, const char *devname
)
1584 struct odp_port odp_port
;
1585 struct ofport
*old_ofport
;
1586 struct ofport
*new_ofport
;
1589 COVERAGE_INC(ofproto_update_port
);
1591 /* Query the datapath for port information. */
1592 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1594 /* Find the old ofport. */
1595 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1598 /* There's no port named 'devname' but there might be a port with
1599 * the same port number. This could happen if a port is deleted
1600 * and then a new one added in its place very quickly, or if a port
1601 * is renamed. In the former case we want to send an OFPPR_DELETE
1602 * and an OFPPR_ADD, and in the latter case we want to send a
1603 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1604 * the old port's ifindex against the new port, or perhaps less
1605 * reliably but more portably by comparing the old port's MAC
1606 * against the new port's MAC. However, this code isn't that smart
1607 * and always sends an OFPPR_MODIFY (XXX). */
1608 old_ofport
= get_port(p
, odp_port
.port
);
1610 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1611 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1612 "%s", strerror(error
));
1616 /* Create a new ofport. */
1617 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1619 /* Eliminate a few pathological cases. */
1620 if (!old_ofport
&& !new_ofport
) {
1622 } else if (old_ofport
&& new_ofport
) {
1623 /* Most of the 'config' bits are OpenFlow soft state, but
1624 * OFPPC_PORT_DOWN is maintained by the kernel. So transfer the
1625 * OpenFlow bits from old_ofport. (make_ofport() only sets
1626 * OFPPC_PORT_DOWN and leaves the other bits 0.) */
1627 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1629 if (ofport_equal(old_ofport
, new_ofport
)) {
1630 /* False alarm--no change. */
1631 ofport_free(new_ofport
);
1636 /* Now deal with the normal cases. */
1638 ofport_remove(p
, old_ofport
);
1641 ofport_install(p
, new_ofport
);
1643 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1644 (!old_ofport
? OFPPR_ADD
1645 : !new_ofport
? OFPPR_DELETE
1647 ofport_free(old_ofport
);
1651 init_ports(struct ofproto
*p
)
1653 struct odp_port
*ports
;
1658 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1663 for (i
= 0; i
< n_ports
; i
++) {
1664 const struct odp_port
*odp_port
= &ports
[i
];
1665 if (!ofport_conflicts(p
, odp_port
)) {
1666 struct ofport
*ofport
= make_ofport(odp_port
);
1668 ofport_install(p
, ofport
);
1676 static struct ofconn
*
1677 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1679 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1680 ofconn
->ofproto
= p
;
1681 list_push_back(&p
->all_conns
, &ofconn
->node
);
1682 ofconn
->rconn
= rconn
;
1683 ofconn
->type
= type
;
1684 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1685 ofconn
->role
= NX_ROLE_OTHER
;
1686 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1687 ofconn
->pktbuf
= NULL
;
1688 ofconn
->miss_send_len
= 0;
1689 ofconn
->reply_counter
= rconn_packet_counter_create ();
1694 ofconn_destroy(struct ofconn
*ofconn
)
1696 if (ofconn
->type
== OFCONN_PRIMARY
) {
1697 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1699 discovery_destroy(ofconn
->discovery
);
1701 list_remove(&ofconn
->node
);
1702 switch_status_unregister(ofconn
->ss
);
1703 rconn_destroy(ofconn
->rconn
);
1704 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1705 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1706 pktbuf_destroy(ofconn
->pktbuf
);
1711 ofconn_run(struct ofconn
*ofconn
)
1713 struct ofproto
*p
= ofconn
->ofproto
;
1717 if (ofconn
->discovery
) {
1718 char *controller_name
;
1719 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1720 discovery_question_connectivity(ofconn
->discovery
);
1722 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1723 if (controller_name
) {
1724 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1725 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1728 rconn_disconnect(ofconn
->rconn
);
1733 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1734 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1737 rconn_run(ofconn
->rconn
);
1739 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1740 /* Limit the number of iterations to prevent other tasks from
1742 for (iteration
= 0; iteration
< 50; iteration
++) {
1743 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1748 fail_open_maybe_recover(p
->fail_open
);
1750 handle_openflow(ofconn
, of_msg
);
1751 ofpbuf_delete(of_msg
);
1755 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1756 ofconn_destroy(ofconn
);
1761 ofconn_wait(struct ofconn
*ofconn
)
1765 if (ofconn
->discovery
) {
1766 discovery_wait(ofconn
->discovery
);
1768 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1769 pinsched_wait(ofconn
->schedulers
[i
]);
1771 rconn_run_wait(ofconn
->rconn
);
1772 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1773 rconn_recv_wait(ofconn
->rconn
);
1775 COVERAGE_INC(ofproto_ofconn_stuck
);
1779 /* Returns true if 'ofconn' should receive asynchronous messages. */
1781 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1783 if (ofconn
->type
== OFCONN_PRIMARY
) {
1784 /* Primary controllers always get asynchronous messages unless they
1785 * have configured themselves as "slaves". */
1786 return ofconn
->role
!= NX_ROLE_SLAVE
;
1788 /* Service connections don't get asynchronous messages unless they have
1789 * explicitly asked for them by setting a nonzero miss send length. */
1790 return ofconn
->miss_send_len
> 0;
1794 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1795 * and 'target', suitable for use in log messages for identifying the
1798 * The name is dynamically allocated. The caller should free it (with free())
1799 * when it is no longer needed. */
1801 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1803 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1807 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1811 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1812 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1816 *s
= pinsched_create(rate
, burst
,
1817 ofconn
->ofproto
->switch_status
);
1819 pinsched_set_limits(*s
, rate
, burst
);
1822 pinsched_destroy(*s
);
1829 ofservice_reconfigure(struct ofservice
*ofservice
,
1830 const struct ofproto_controller
*c
)
1832 ofservice
->probe_interval
= c
->probe_interval
;
1833 ofservice
->rate_limit
= c
->rate_limit
;
1834 ofservice
->burst_limit
= c
->burst_limit
;
1837 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1838 * positive errno value. */
1840 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1842 struct ofservice
*ofservice
;
1843 struct pvconn
*pvconn
;
1846 error
= pvconn_open(c
->target
, &pvconn
);
1851 ofservice
= xzalloc(sizeof *ofservice
);
1852 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1853 hash_string(c
->target
, 0));
1854 ofservice
->pvconn
= pvconn
;
1856 ofservice_reconfigure(ofservice
, c
);
1862 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1864 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1865 pvconn_close(ofservice
->pvconn
);
1869 /* Finds and returns the ofservice within 'ofproto' that has the given
1870 * 'target', or a null pointer if none exists. */
1871 static struct ofservice
*
1872 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1874 struct ofservice
*ofservice
;
1876 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1877 &ofproto
->services
) {
1878 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1885 /* Returns true if 'rule' should be hidden from the controller.
1887 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1888 * (e.g. by in-band control) and are intentionally hidden from the
1891 rule_is_hidden(const struct rule
*rule
)
1893 return rule
->cr
.priority
> UINT16_MAX
;
1896 /* Creates and returns a new rule initialized as specified.
1898 * The caller is responsible for inserting the rule into the classifier (with
1899 * rule_insert()). */
1900 static struct rule
*
1901 rule_create(const struct cls_rule
*cls_rule
,
1902 const union ofp_action
*actions
, size_t n_actions
,
1903 uint16_t idle_timeout
, uint16_t hard_timeout
,
1904 ovs_be64 flow_cookie
, bool send_flow_removed
)
1906 struct rule
*rule
= xzalloc(sizeof *rule
);
1907 rule
->cr
= *cls_rule
;
1908 rule
->idle_timeout
= idle_timeout
;
1909 rule
->hard_timeout
= hard_timeout
;
1910 rule
->flow_cookie
= flow_cookie
;
1911 rule
->used
= rule
->created
= time_msec();
1912 rule
->send_flow_removed
= send_flow_removed
;
1913 list_init(&rule
->facets
);
1914 if (n_actions
> 0) {
1915 rule
->n_actions
= n_actions
;
1916 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1922 static struct rule
*
1923 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1925 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1929 rule_free(struct rule
*rule
)
1931 free(rule
->actions
);
1935 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1936 * destroying any that no longer has a rule (which is probably all of them).
1938 * The caller must have already removed 'rule' from the classifier. */
1940 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1942 struct facet
*facet
, *next_facet
;
1943 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
1944 facet_revalidate(ofproto
, facet
);
1949 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1950 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1953 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
1955 const union ofp_action
*oa
;
1956 struct actions_iterator i
;
1958 if (out_port
== htons(OFPP_NONE
)) {
1961 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1962 oa
= actions_next(&i
)) {
1963 if (action_outputs_to_port(oa
, out_port
)) {
1970 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1971 * 'packet', which arrived on 'in_port'.
1973 * Takes ownership of 'packet'. */
1975 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
1976 const union odp_action
*actions
, size_t n_actions
,
1977 struct ofpbuf
*packet
)
1979 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
1980 /* As an optimization, avoid a round-trip from userspace to kernel to
1981 * userspace. This also avoids possibly filling up kernel packet
1982 * buffers along the way. */
1983 struct odp_msg
*msg
;
1985 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
1986 msg
->type
= _ODPL_ACTION_NR
;
1987 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
1988 msg
->port
= in_port
;
1990 msg
->arg
= actions
[0].controller
.arg
;
1992 send_packet_in(ofproto
, packet
);
1998 error
= dpif_execute(ofproto
->dpif
, actions
, n_actions
, packet
);
1999 ofpbuf_delete(packet
);
2004 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
2005 * statistics appropriately. 'packet' must have at least sizeof(struct
2006 * ofp_packet_in) bytes of headroom.
2008 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
2009 * applying flow_extract() to 'packet' would yield the same flow as
2012 * 'facet' must have accurately composed ODP actions; that is, it must not be
2013 * in need of revalidation.
2015 * Takes ownership of 'packet'. */
2017 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2018 struct ofpbuf
*packet
)
2020 struct odp_flow_stats stats
;
2022 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2024 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2025 if (execute_odp_actions(ofproto
, facet
->flow
.in_port
,
2026 facet
->actions
, facet
->n_actions
, packet
)) {
2027 facet_update_stats(ofproto
, facet
, &stats
);
2028 facet
->used
= time_msec();
2029 netflow_flow_update_time(ofproto
->netflow
,
2030 &facet
->nf_flow
, facet
->used
);
2034 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2035 * statistics (or the statistics for one of its facets) appropriately.
2036 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2038 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2039 * with statistics for 'packet' either way.
2041 * Takes ownership of 'packet'. */
2043 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2044 struct ofpbuf
*packet
)
2046 struct facet
*facet
;
2047 struct odp_actions a
;
2051 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2053 flow_extract(packet
, 0, in_port
, &flow
);
2055 /* First look for a related facet. If we find one, account it to that. */
2056 facet
= facet_lookup_valid(ofproto
, &flow
);
2057 if (facet
&& facet
->rule
== rule
) {
2058 facet_execute(ofproto
, facet
, packet
);
2062 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2063 * create a new facet for it and use that. */
2064 if (rule_lookup(ofproto
, &flow
) == rule
) {
2065 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2066 facet_execute(ofproto
, facet
, packet
);
2067 facet_install(ofproto
, facet
, true);
2071 /* We can't account anything to a facet. If we were to try, then that
2072 * facet would have a non-matching rule, busting our invariants. */
2073 if (xlate_actions(rule
->actions
, rule
->n_actions
, &flow
, ofproto
,
2074 packet
, &a
, NULL
, 0, NULL
)) {
2075 ofpbuf_delete(packet
);
2078 size
= packet
->size
;
2079 if (execute_odp_actions(ofproto
, in_port
,
2080 a
.actions
, a
.n_actions
, packet
)) {
2081 rule
->used
= time_msec();
2082 rule
->packet_count
++;
2083 rule
->byte_count
+= size
;
2087 /* Inserts 'rule' into 'p''s flow table. */
2089 rule_insert(struct ofproto
*p
, struct rule
*rule
)
2091 struct rule
*displaced_rule
;
2093 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2094 if (displaced_rule
) {
2095 rule_destroy(p
, displaced_rule
);
2097 p
->need_revalidate
= true;
2100 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2101 * 'flow' and an example 'packet' within that flow.
2103 * The caller must already have determined that no facet with an identical
2104 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2105 * 'ofproto''s classifier table. */
2106 static struct facet
*
2107 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2108 const struct flow
*flow
, const struct ofpbuf
*packet
)
2110 struct facet
*facet
;
2112 facet
= xzalloc(sizeof *facet
);
2113 facet
->used
= time_msec();
2114 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2115 list_push_back(&rule
->facets
, &facet
->list_node
);
2117 facet
->flow
= *flow
;
2118 netflow_flow_init(&facet
->nf_flow
);
2119 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2121 facet_make_actions(ofproto
, facet
, packet
);
2127 facet_free(struct facet
*facet
)
2129 free(facet
->actions
);
2133 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2135 * - Removes 'rule' from the classifier.
2137 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2138 * destroys them), via rule_destroy().
2141 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2143 COVERAGE_INC(ofproto_del_rule
);
2144 ofproto
->need_revalidate
= true;
2145 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2146 rule_destroy(ofproto
, rule
);
2149 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2151 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2152 * rule's statistics, via facet_uninstall().
2154 * - Removes 'facet' from its rule and from ofproto->facets.
2157 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2159 facet_uninstall(ofproto
, facet
);
2160 facet_flush_stats(ofproto
, facet
);
2161 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2162 list_remove(&facet
->list_node
);
2166 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2168 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2169 const struct ofpbuf
*packet
)
2171 const struct rule
*rule
= facet
->rule
;
2172 struct odp_actions a
;
2175 xlate_actions(rule
->actions
, rule
->n_actions
, &facet
->flow
, p
,
2176 packet
, &a
, &facet
->tags
, &facet
->may_install
,
2177 &facet
->nf_flow
.output_iface
);
2179 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2180 if (facet
->n_actions
!= a
.n_actions
2181 || memcmp(facet
->actions
, a
.actions
, actions_len
)) {
2182 free(facet
->actions
);
2183 facet
->n_actions
= a
.n_actions
;
2184 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2189 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
, int flags
,
2190 struct odp_flow_put
*put
)
2192 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2193 odp_flow_key_from_flow(&put
->flow
.key
, &facet
->flow
);
2194 put
->flow
.actions
= facet
->actions
;
2195 put
->flow
.n_actions
= facet
->n_actions
;
2196 put
->flow
.flags
= 0;
2198 return dpif_flow_put(ofproto
->dpif
, put
);
2201 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2202 * 'zero_stats' is true, clears any existing statistics from the datapath for
2205 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2207 if (facet
->may_install
) {
2208 struct odp_flow_put put
;
2211 flags
= ODPPF_CREATE
| ODPPF_MODIFY
;
2213 flags
|= ODPPF_ZERO_STATS
;
2215 if (!facet_put__(p
, facet
, flags
, &put
)) {
2216 facet
->installed
= true;
2221 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2222 * to the accounting hook function in the ofhooks structure. */
2224 facet_account(struct ofproto
*ofproto
,
2225 struct facet
*facet
, uint64_t extra_bytes
)
2227 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2229 if (ofproto
->ofhooks
->account_flow_cb
2230 && total_bytes
> facet
->accounted_bytes
)
2232 ofproto
->ofhooks
->account_flow_cb(
2233 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->n_actions
,
2234 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2235 facet
->accounted_bytes
= total_bytes
;
2239 /* If 'rule' is installed in the datapath, uninstalls it. */
2241 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2243 if (facet
->installed
) {
2244 struct odp_flow odp_flow
;
2246 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
2247 odp_flow
.actions
= NULL
;
2248 odp_flow
.n_actions
= 0;
2250 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2251 facet_update_stats(p
, facet
, &odp_flow
.stats
);
2253 facet
->installed
= false;
2257 /* Returns true if the only action for 'facet' is to send to the controller.
2258 * (We don't report NetFlow expiration messages for such facets because they
2259 * are just part of the control logic for the network, not real traffic). */
2261 facet_is_controller_flow(struct facet
*facet
)
2264 && facet
->rule
->n_actions
== 1
2265 && action_outputs_to_port(&facet
->rule
->actions
[0],
2266 htons(OFPP_CONTROLLER
)));
2269 /* Folds all of 'facet''s statistics into its rule. Also updates the
2270 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2272 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2274 facet_account(ofproto
, facet
, 0);
2276 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2277 struct ofexpired expired
;
2278 expired
.flow
= facet
->flow
;
2279 expired
.packet_count
= facet
->packet_count
;
2280 expired
.byte_count
= facet
->byte_count
;
2281 expired
.used
= facet
->used
;
2282 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2285 facet
->rule
->packet_count
+= facet
->packet_count
;
2286 facet
->rule
->byte_count
+= facet
->byte_count
;
2288 /* Reset counters to prevent double counting if 'facet' ever gets
2290 facet
->packet_count
= 0;
2291 facet
->byte_count
= 0;
2292 facet
->accounted_bytes
= 0;
2294 netflow_flow_clear(&facet
->nf_flow
);
2297 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2298 * Returns it if found, otherwise a null pointer.
2300 * The returned facet might need revalidation; use facet_lookup_valid()
2301 * instead if that is important. */
2302 static struct facet
*
2303 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2305 struct facet
*facet
;
2307 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2309 if (flow_equal(flow
, &facet
->flow
)) {
2317 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2318 * Returns it if found, otherwise a null pointer.
2320 * The returned facet is guaranteed to be valid. */
2321 static struct facet
*
2322 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2324 struct facet
*facet
= facet_find(ofproto
, flow
);
2326 /* The facet we found might not be valid, since we could be in need of
2327 * revalidation. If it is not valid, don't return it. */
2329 && ofproto
->need_revalidate
2330 && !facet_revalidate(ofproto
, facet
)) {
2331 COVERAGE_INC(ofproto_invalidated
);
2338 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2340 * - If the rule found is different from 'facet''s current rule, moves
2341 * 'facet' to the new rule and recompiles its actions.
2343 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2344 * where it is and recompiles its actions anyway.
2346 * - If there is none, destroys 'facet'.
2348 * Returns true if 'facet' still exists, false if it has been destroyed. */
2350 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2352 struct rule
*new_rule
;
2353 struct odp_actions a
;
2355 uint16_t new_nf_output_iface
;
2356 bool actions_changed
;
2358 COVERAGE_INC(facet_revalidate
);
2360 /* Determine the new rule. */
2361 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2363 /* No new rule, so delete the facet. */
2364 facet_remove(ofproto
, facet
);
2368 /* Calculate new ODP actions.
2370 * We are very cautious about actually modifying 'facet' state at this
2371 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2372 * so, we need to have the old state around to properly compose it. */
2373 xlate_actions(new_rule
->actions
, new_rule
->n_actions
, &facet
->flow
,
2374 ofproto
, NULL
, &a
, &facet
->tags
, &facet
->may_install
,
2375 &new_nf_output_iface
);
2376 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2377 actions_changed
= (facet
->n_actions
!= a
.n_actions
2378 || memcmp(facet
->actions
, a
.actions
, actions_len
));
2380 /* If the ODP actions changed or the installability changed, then we need
2381 * to talk to the datapath. */
2382 if (actions_changed
|| facet
->may_install
!= facet
->installed
) {
2383 if (facet
->may_install
) {
2384 struct odp_flow_put put
;
2386 memset(&put
.flow
.stats
, 0, sizeof put
.flow
.stats
);
2387 odp_flow_key_from_flow(&put
.flow
.key
, &facet
->flow
);
2388 put
.flow
.actions
= a
.actions
;
2389 put
.flow
.n_actions
= a
.n_actions
;
2391 put
.flags
= ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
;
2392 dpif_flow_put(ofproto
->dpif
, &put
);
2394 facet_update_stats(ofproto
, facet
, &put
.flow
.stats
);
2396 facet_uninstall(ofproto
, facet
);
2399 /* The datapath flow is gone or has zeroed stats, so push stats out of
2400 * 'facet' into 'rule'. */
2401 facet_flush_stats(ofproto
, facet
);
2404 /* Update 'facet' now that we've taken care of all the old state. */
2405 facet
->nf_flow
.output_iface
= new_nf_output_iface
;
2406 if (actions_changed
) {
2407 free(facet
->actions
);
2408 facet
->n_actions
= a
.n_actions
;
2409 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2411 if (facet
->rule
!= new_rule
) {
2412 COVERAGE_INC(facet_changed_rule
);
2413 list_remove(&facet
->list_node
);
2414 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2415 facet
->rule
= new_rule
;
2416 facet
->used
= new_rule
->created
;
2423 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2424 struct rconn_packet_counter
*counter
)
2426 update_openflow_length(msg
);
2427 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2433 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2436 struct ofpbuf
*buf
= make_ofp_error_msg(error
, oh
);
2438 COVERAGE_INC(ofproto_error
);
2439 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2444 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2446 opp
->port_no
= htons(opp
->port_no
);
2447 opp
->config
= htonl(opp
->config
);
2448 opp
->state
= htonl(opp
->state
);
2449 opp
->curr
= htonl(opp
->curr
);
2450 opp
->advertised
= htonl(opp
->advertised
);
2451 opp
->supported
= htonl(opp
->supported
);
2452 opp
->peer
= htonl(opp
->peer
);
2456 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2458 struct ofp_header
*rq
= oh
;
2459 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2464 handle_features_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2466 struct ofp_switch_features
*osf
;
2468 struct ofport
*port
;
2470 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2471 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2472 osf
->n_buffers
= htonl(pktbuf_capacity());
2474 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2475 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2476 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2477 (1u << OFPAT_SET_VLAN_VID
) |
2478 (1u << OFPAT_SET_VLAN_PCP
) |
2479 (1u << OFPAT_STRIP_VLAN
) |
2480 (1u << OFPAT_SET_DL_SRC
) |
2481 (1u << OFPAT_SET_DL_DST
) |
2482 (1u << OFPAT_SET_NW_SRC
) |
2483 (1u << OFPAT_SET_NW_DST
) |
2484 (1u << OFPAT_SET_NW_TOS
) |
2485 (1u << OFPAT_SET_TP_SRC
) |
2486 (1u << OFPAT_SET_TP_DST
) |
2487 (1u << OFPAT_ENQUEUE
));
2489 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2490 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2493 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2498 handle_get_config_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2501 struct ofp_switch_config
*osc
;
2505 /* Figure out flags. */
2506 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2507 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2510 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2511 osc
->flags
= htons(flags
);
2512 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2513 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2519 handle_set_config(struct ofconn
*ofconn
, struct ofp_switch_config
*osc
)
2524 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2528 flags
= ntohs(osc
->flags
);
2530 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2531 switch (flags
& OFPC_FRAG_MASK
) {
2532 case OFPC_FRAG_NORMAL
:
2533 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2535 case OFPC_FRAG_DROP
:
2536 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2539 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2545 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2551 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2553 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2554 a
->controller
.arg
= max_len
;
2557 struct action_xlate_ctx
{
2559 struct flow flow
; /* Flow to which these actions correspond. */
2560 int recurse
; /* Recursion level, via xlate_table_action. */
2561 struct ofproto
*ofproto
;
2562 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2563 * null pointer if we are revalidating
2564 * without a packet to refer to. */
2567 struct odp_actions
*out
; /* Datapath actions. */
2568 tag_type tags
; /* Tags associated with OFPP_NORMAL actions. */
2569 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2570 * be reassessed for every packet. */
2571 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2574 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2575 * flow translation. */
2576 #define MAX_RESUBMIT_RECURSION 8
2578 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2579 struct action_xlate_ctx
*ctx
);
2582 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2584 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2587 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2588 /* Forwarding disabled on port. */
2593 * We don't have an ofport record for this port, but it doesn't hurt to
2594 * allow forwarding to it anyhow. Maybe such a port will appear later
2595 * and we're pre-populating the flow table.
2599 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2600 ctx
->nf_output_iface
= port
;
2603 static struct rule
*
2604 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2606 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2610 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2612 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2613 uint16_t old_in_port
;
2616 /* Look up a flow with 'in_port' as the input port. Then restore the
2617 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2618 * have surprising behavior). */
2619 old_in_port
= ctx
->flow
.in_port
;
2620 ctx
->flow
.in_port
= in_port
;
2621 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2622 ctx
->flow
.in_port
= old_in_port
;
2626 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2630 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2632 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2633 MAX_RESUBMIT_RECURSION
);
2638 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2639 uint16_t *nf_output_iface
, struct odp_actions
*actions
)
2641 struct ofport
*ofport
;
2643 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2644 uint16_t odp_port
= ofport
->odp_port
;
2645 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2646 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= odp_port
;
2649 *nf_output_iface
= NF_OUT_FLOOD
;
2653 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2654 uint16_t port
, uint16_t max_len
)
2657 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2659 ctx
->nf_output_iface
= NF_OUT_DROP
;
2663 add_output_action(ctx
, ctx
->flow
.in_port
);
2666 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2669 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2670 ctx
->out
, &ctx
->tags
,
2671 &ctx
->nf_output_iface
,
2672 ctx
->ofproto
->aux
)) {
2673 COVERAGE_INC(ofproto_uninstallable
);
2674 ctx
->may_set_up_flow
= false;
2678 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2679 &ctx
->nf_output_iface
, ctx
->out
);
2682 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2683 &ctx
->nf_output_iface
, ctx
->out
);
2685 case OFPP_CONTROLLER
:
2686 add_controller_action(ctx
->out
, max_len
);
2689 add_output_action(ctx
, ODPP_LOCAL
);
2692 odp_port
= ofp_port_to_odp_port(port
);
2693 if (odp_port
!= ctx
->flow
.in_port
) {
2694 add_output_action(ctx
, odp_port
);
2699 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2700 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2701 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2702 ctx
->nf_output_iface
= prev_nf_output_iface
;
2703 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2704 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2705 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2710 xlate_output_action(struct action_xlate_ctx
*ctx
,
2711 const struct ofp_action_output
*oao
)
2713 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2716 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2717 * optimization, because we're going to add another action that sets the
2718 * priority immediately after, or because there are no actions following the
2721 remove_pop_action(struct action_xlate_ctx
*ctx
)
2723 size_t n
= ctx
->out
->n_actions
;
2724 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2725 ctx
->out
->n_actions
--;
2730 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2731 const struct ofp_action_enqueue
*oae
)
2733 uint16_t ofp_port
, odp_port
;
2737 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2740 /* Fall back to ordinary output action. */
2741 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2745 /* Figure out ODP output port. */
2746 ofp_port
= ntohs(oae
->port
);
2747 if (ofp_port
!= OFPP_IN_PORT
) {
2748 odp_port
= ofp_port_to_odp_port(ofp_port
);
2750 odp_port
= ctx
->flow
.in_port
;
2753 /* Add ODP actions. */
2754 remove_pop_action(ctx
);
2755 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2757 add_output_action(ctx
, odp_port
);
2758 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2760 /* Update NetFlow output port. */
2761 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2762 ctx
->nf_output_iface
= odp_port
;
2763 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2764 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2769 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2770 const struct nx_action_set_queue
*nasq
)
2775 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2778 /* Couldn't translate queue to a priority, so ignore. A warning
2779 * has already been logged. */
2783 remove_pop_action(ctx
);
2784 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2789 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2791 ovs_be16 tci
= ctx
->flow
.vlan_tci
;
2792 if (!(tci
& htons(VLAN_CFI
))) {
2793 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2795 union odp_action
*oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_TCI
);
2796 oa
->dl_tci
.tci
= tci
& ~htons(VLAN_CFI
);
2801 xlate_reg_move_action(struct action_xlate_ctx
*ctx
,
2802 const struct nx_action_reg_move
*narm
)
2804 ovs_be16 old_tci
= ctx
->flow
.vlan_tci
;
2806 nxm_execute_reg_move(narm
, &ctx
->flow
);
2808 if (ctx
->flow
.vlan_tci
!= old_tci
) {
2809 xlate_set_dl_tci(ctx
);
2814 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2815 const struct nx_action_header
*nah
)
2817 const struct nx_action_resubmit
*nar
;
2818 const struct nx_action_set_tunnel
*nast
;
2819 const struct nx_action_set_queue
*nasq
;
2820 union odp_action
*oa
;
2821 int subtype
= ntohs(nah
->subtype
);
2823 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2825 case NXAST_RESUBMIT
:
2826 nar
= (const struct nx_action_resubmit
*) nah
;
2827 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2830 case NXAST_SET_TUNNEL
:
2831 nast
= (const struct nx_action_set_tunnel
*) nah
;
2832 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2833 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2836 case NXAST_DROP_SPOOFED_ARP
:
2837 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2838 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2842 case NXAST_SET_QUEUE
:
2843 nasq
= (const struct nx_action_set_queue
*) nah
;
2844 xlate_set_queue_action(ctx
, nasq
);
2847 case NXAST_POP_QUEUE
:
2848 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2851 case NXAST_REG_MOVE
:
2852 xlate_reg_move_action(ctx
, (const struct nx_action_reg_move
*) nah
);
2855 case NXAST_REG_LOAD
:
2856 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
2860 /* Nothing to do. */
2863 /* If you add a new action here that modifies flow data, don't forget to
2864 * update the flow key in ctx->flow at the same time. */
2867 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2873 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2874 struct action_xlate_ctx
*ctx
)
2876 struct actions_iterator iter
;
2877 const union ofp_action
*ia
;
2878 const struct ofport
*port
;
2880 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2881 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2882 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2883 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2884 /* Drop this flow. */
2888 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2889 uint16_t type
= ntohs(ia
->type
);
2890 union odp_action
*oa
;
2894 xlate_output_action(ctx
, &ia
->output
);
2897 case OFPAT_SET_VLAN_VID
:
2898 ctx
->flow
.vlan_tci
&= ~htons(VLAN_VID_MASK
);
2899 ctx
->flow
.vlan_tci
|= ia
->vlan_vid
.vlan_vid
| htons(VLAN_CFI
);
2900 xlate_set_dl_tci(ctx
);
2903 case OFPAT_SET_VLAN_PCP
:
2904 ctx
->flow
.vlan_tci
&= ~htons(VLAN_PCP_MASK
);
2905 ctx
->flow
.vlan_tci
|= htons(
2906 (ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
) | VLAN_CFI
);
2907 xlate_set_dl_tci(ctx
);
2910 case OFPAT_STRIP_VLAN
:
2911 ctx
->flow
.vlan_tci
= htons(0);
2912 xlate_set_dl_tci(ctx
);
2915 case OFPAT_SET_DL_SRC
:
2916 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2917 memcpy(oa
->dl_addr
.dl_addr
,
2918 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2919 memcpy(ctx
->flow
.dl_src
,
2920 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2923 case OFPAT_SET_DL_DST
:
2924 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2925 memcpy(oa
->dl_addr
.dl_addr
,
2926 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2927 memcpy(ctx
->flow
.dl_dst
,
2928 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2931 case OFPAT_SET_NW_SRC
:
2932 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2933 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2936 case OFPAT_SET_NW_DST
:
2937 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2938 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2941 case OFPAT_SET_NW_TOS
:
2942 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2943 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2946 case OFPAT_SET_TP_SRC
:
2947 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2948 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2951 case OFPAT_SET_TP_DST
:
2952 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2953 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2957 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2961 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2965 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2972 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2973 const struct flow
*flow
, struct ofproto
*ofproto
,
2974 const struct ofpbuf
*packet
,
2975 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2976 uint16_t *nf_output_iface
)
2978 struct action_xlate_ctx ctx
;
2980 COVERAGE_INC(ofproto_ofp2odp
);
2981 odp_actions_init(out
);
2984 ctx
.ofproto
= ofproto
;
2985 ctx
.packet
= packet
;
2988 ctx
.may_set_up_flow
= true;
2989 ctx
.nf_output_iface
= NF_OUT_DROP
;
2990 do_xlate_actions(in
, n_in
, &ctx
);
2991 remove_pop_action(&ctx
);
2993 /* Check with in-band control to see if we're allowed to set up this
2995 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2996 ctx
.may_set_up_flow
= false;
3002 if (may_set_up_flow
) {
3003 *may_set_up_flow
= ctx
.may_set_up_flow
;
3005 if (nf_output_iface
) {
3006 *nf_output_iface
= ctx
.nf_output_iface
;
3008 if (odp_actions_overflow(out
)) {
3009 COVERAGE_INC(odp_overflow
);
3010 odp_actions_init(out
);
3011 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
3016 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3017 * error message code (composed with ofp_mkerr()) for the caller to propagate
3018 * upward. Otherwise, returns 0.
3020 * The log message mentions 'msg_type'. */
3022 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3024 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3025 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3026 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3029 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3036 handle_packet_out(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3038 struct ofproto
*p
= ofconn
->ofproto
;
3039 struct ofp_packet_out
*opo
;
3040 struct ofpbuf payload
, *buffer
;
3041 union ofp_action
*ofp_actions
;
3042 struct odp_actions odp_actions
;
3043 struct ofpbuf request
;
3045 size_t n_ofp_actions
;
3049 COVERAGE_INC(ofproto_packet_out
);
3051 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3056 /* Get ofp_packet_out. */
3058 request
.size
= ntohs(oh
->length
);
3059 opo
= ofpbuf_try_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3061 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3065 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3066 &ofp_actions
, &n_ofp_actions
);
3072 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3073 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3075 if (error
|| !buffer
) {
3084 /* Extract flow, check actions. */
3085 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3087 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3093 error
= xlate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
, &payload
,
3094 &odp_actions
, NULL
, NULL
, NULL
);
3096 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
,
3101 ofpbuf_delete(buffer
);
3106 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3107 uint32_t config
, uint32_t mask
)
3109 mask
&= config
^ port
->opp
.config
;
3110 if (mask
& OFPPC_PORT_DOWN
) {
3111 if (config
& OFPPC_PORT_DOWN
) {
3112 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3114 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3117 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3118 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3119 if (mask
& REVALIDATE_BITS
) {
3120 COVERAGE_INC(ofproto_costly_flags
);
3121 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3122 p
->need_revalidate
= true;
3124 #undef REVALIDATE_BITS
3125 if (mask
& OFPPC_NO_PACKET_IN
) {
3126 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3131 handle_port_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3133 struct ofproto
*p
= ofconn
->ofproto
;
3134 const struct ofp_port_mod
*opm
;
3135 struct ofport
*port
;
3138 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3142 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
3146 opm
= (struct ofp_port_mod
*) oh
;
3148 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3150 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3151 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3152 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3154 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3155 if (opm
->advertise
) {
3156 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3162 static struct ofpbuf
*
3163 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3165 struct ofp_stats_reply
*osr
;
3168 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3169 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3171 osr
->flags
= htons(0);
3175 static struct ofpbuf
*
3176 start_ofp_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
3178 return make_ofp_stats_reply(request
->header
.xid
, request
->type
, body_len
);
3182 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3183 struct ofpbuf
**msgp
)
3185 struct ofpbuf
*msg
= *msgp
;
3186 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3187 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3188 struct ofp_stats_reply
*reply
= msg
->data
;
3189 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3190 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3191 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3193 return ofpbuf_put_uninit(*msgp
, nbytes
);
3196 static struct ofpbuf
*
3197 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3199 struct nicira_stats_msg
*nsm
;
3202 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3203 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3204 nsm
->type
= htons(OFPST_VENDOR
);
3205 nsm
->flags
= htons(0);
3206 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3207 nsm
->subtype
= htonl(subtype
);
3211 static struct ofpbuf
*
3212 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3214 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3218 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3219 struct ofpbuf
**msgp
)
3221 struct ofpbuf
*msg
= *msgp
;
3222 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3223 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3224 struct nicira_stats_msg
*reply
= msg
->data
;
3225 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3226 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3227 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3229 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3233 handle_desc_stats_request(struct ofconn
*ofconn
,
3234 struct ofp_stats_request
*request
)
3236 struct ofproto
*p
= ofconn
->ofproto
;
3237 struct ofp_desc_stats
*ods
;
3240 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3241 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3242 memset(ods
, 0, sizeof *ods
);
3243 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3244 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3245 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3246 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3247 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3248 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3254 handle_table_stats_request(struct ofconn
*ofconn
,
3255 struct ofp_stats_request
*request
)
3257 struct ofproto
*p
= ofconn
->ofproto
;
3258 struct ofp_table_stats
*ots
;
3261 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3263 /* Classifier table. */
3264 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3265 memset(ots
, 0, sizeof *ots
);
3266 strcpy(ots
->name
, "classifier");
3267 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3268 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3269 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3270 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3271 ots
->lookup_count
= htonll(0); /* XXX */
3272 ots
->matched_count
= htonll(0); /* XXX */
3274 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3279 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3280 struct ofpbuf
**msgp
)
3282 struct netdev_stats stats
;
3283 struct ofp_port_stats
*ops
;
3285 /* Intentionally ignore return value, since errors will set
3286 * 'stats' to all-1s, which is correct for OpenFlow, and
3287 * netdev_get_stats() will log errors. */
3288 netdev_get_stats(port
->netdev
, &stats
);
3290 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3291 ops
->port_no
= htons(port
->opp
.port_no
);
3292 memset(ops
->pad
, 0, sizeof ops
->pad
);
3293 ops
->rx_packets
= htonll(stats
.rx_packets
);
3294 ops
->tx_packets
= htonll(stats
.tx_packets
);
3295 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3296 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3297 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3298 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3299 ops
->rx_errors
= htonll(stats
.rx_errors
);
3300 ops
->tx_errors
= htonll(stats
.tx_errors
);
3301 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3302 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3303 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3304 ops
->collisions
= htonll(stats
.collisions
);
3308 handle_port_stats_request(struct ofconn
*ofconn
, struct ofp_stats_request
*osr
,
3311 struct ofproto
*p
= ofconn
->ofproto
;
3312 struct ofp_port_stats_request
*psr
;
3313 struct ofp_port_stats
*ops
;
3315 struct ofport
*port
;
3317 if (arg_size
!= sizeof *psr
) {
3318 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3320 psr
= (struct ofp_port_stats_request
*) osr
->body
;
3322 msg
= start_ofp_stats_reply(osr
, sizeof *ops
* 16);
3323 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3324 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3326 append_port_stat(port
, ofconn
, &msg
);
3329 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3330 append_port_stat(port
, ofconn
, &msg
);
3334 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3338 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3339 * '*packet_countp' and '*byte_countp'. The returned statistics include
3340 * statistics for all of 'rule''s facets. */
3342 query_stats(struct ofproto
*p
, struct rule
*rule
,
3343 uint64_t *packet_countp
, uint64_t *byte_countp
)
3345 uint64_t packet_count
, byte_count
;
3346 struct facet
*facet
;
3347 struct odp_flow
*odp_flows
;
3350 /* Start from historical data for 'rule' itself that are no longer tracked
3351 * by the datapath. This counts, for example, facets that have expired. */
3352 packet_count
= rule
->packet_count
;
3353 byte_count
= rule
->byte_count
;
3355 /* Prepare to ask the datapath for statistics on all of the rule's facets.
3357 * Also, add any statistics that are not tracked by the datapath for each
3358 * facet. This includes, for example, statistics for packets that were
3359 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3361 odp_flows
= xzalloc(list_size(&rule
->facets
) * sizeof *odp_flows
);
3363 LIST_FOR_EACH (facet
, list_node
, &rule
->facets
) {
3364 struct odp_flow
*odp_flow
= &odp_flows
[n_odp_flows
++];
3365 odp_flow_key_from_flow(&odp_flow
->key
, &facet
->flow
);
3366 packet_count
+= facet
->packet_count
;
3367 byte_count
+= facet
->byte_count
;
3370 /* Fetch up-to-date statistics from the datapath and add them in. */
3371 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3374 for (i
= 0; i
< n_odp_flows
; i
++) {
3375 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3376 packet_count
+= odp_flow
->stats
.n_packets
;
3377 byte_count
+= odp_flow
->stats
.n_bytes
;
3382 /* Return the stats to the caller. */
3383 *packet_countp
= packet_count
;
3384 *byte_countp
= byte_count
;
3388 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3390 long long int msecs
= time_msec() - start
;
3391 *sec
= htonl(msecs
/ 1000);
3392 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3396 put_ofp_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3397 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3399 struct ofp_flow_stats
*ofs
;
3400 uint64_t packet_count
, byte_count
;
3401 size_t act_len
, len
;
3403 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3407 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3408 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3410 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3412 ofs
= append_ofp_stats_reply(len
, ofconn
, replyp
);
3413 ofs
->length
= htons(len
);
3416 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofs
->match
);
3417 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3418 ofs
->cookie
= rule
->flow_cookie
;
3419 ofs
->priority
= htons(rule
->cr
.priority
);
3420 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3421 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3422 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3423 ofs
->packet_count
= htonll(packet_count
);
3424 ofs
->byte_count
= htonll(byte_count
);
3425 if (rule
->n_actions
> 0) {
3426 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3431 is_valid_table(uint8_t table_id
)
3433 return table_id
== 0 || table_id
== 0xff;
3437 handle_flow_stats_request(struct ofconn
*ofconn
,
3438 const struct ofp_stats_request
*osr
, size_t arg_size
)
3440 struct ofp_flow_stats_request
*fsr
;
3441 struct ofpbuf
*reply
;
3443 if (arg_size
!= sizeof *fsr
) {
3444 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3446 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3448 COVERAGE_INC(ofproto_flows_req
);
3449 reply
= start_ofp_stats_reply(osr
, 1024);
3450 if (is_valid_table(fsr
->table_id
)) {
3451 struct cls_cursor cursor
;
3452 struct cls_rule target
;
3455 ofputil_cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0,
3457 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3458 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3459 put_ofp_flow_stats(ofconn
, rule
, fsr
->out_port
, &reply
);
3462 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3468 put_nx_flow_stats(struct ofconn
*ofconn
, struct rule
*rule
,
3469 ovs_be16 out_port
, struct ofpbuf
**replyp
)
3471 struct nx_flow_stats
*nfs
;
3472 uint64_t packet_count
, byte_count
;
3473 size_t act_len
, start_len
;
3474 struct ofpbuf
*reply
;
3476 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, out_port
)) {
3480 query_stats(ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3482 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3484 start_len
= (*replyp
)->size
;
3485 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
, ofconn
, replyp
);
3488 nfs
= ofpbuf_put_uninit(reply
, sizeof *nfs
);
3491 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3492 nfs
->cookie
= rule
->flow_cookie
;
3493 nfs
->priority
= htons(rule
->cr
.priority
);
3494 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3495 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3496 nfs
->match_len
= htons(nx_put_match(reply
, &rule
->cr
));
3497 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3498 nfs
->packet_count
= htonll(packet_count
);
3499 nfs
->byte_count
= htonll(byte_count
);
3500 if (rule
->n_actions
> 0) {
3501 ofpbuf_put(reply
, rule
->actions
, act_len
);
3503 nfs
->length
= htons(reply
->size
- start_len
);
3507 handle_nxst_flow(struct ofconn
*ofconn
, struct ofpbuf
*b
)
3509 struct nx_flow_stats_request
*nfsr
;
3510 struct cls_rule target
;
3511 struct ofpbuf
*reply
;
3514 /* Dissect the message. */
3515 nfsr
= ofpbuf_try_pull(b
, sizeof *nfsr
);
3517 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3519 error
= nx_pull_match(b
, ntohs(nfsr
->match_len
), 0, &target
);
3524 COVERAGE_INC(ofproto_flows_req
);
3525 reply
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3526 if (is_valid_table(nfsr
->table_id
)) {
3527 struct cls_cursor cursor
;
3530 cls_cursor_init(&cursor
, &ofconn
->ofproto
->cls
, &target
);
3531 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3532 put_nx_flow_stats(ofconn
, rule
, nfsr
->out_port
, &reply
);
3535 queue_tx(reply
, ofconn
, ofconn
->reply_counter
);
3541 flow_stats_ds(struct ofproto
*ofproto
, struct rule
*rule
, struct ds
*results
)
3543 struct ofp_match match
;
3544 uint64_t packet_count
, byte_count
;
3545 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3547 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3548 ofputil_cls_rule_to_match(&rule
->cr
, NXFF_OPENFLOW10
, &match
);
3550 ds_put_format(results
, "duration=%llds, ",
3551 (time_msec() - rule
->created
) / 1000);
3552 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3553 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3554 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3555 ofp_print_match(results
, &match
, true);
3557 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3559 ds_put_cstr(results
, "drop");
3561 ds_put_cstr(results
, "\n");
3564 /* Adds a pretty-printed description of all flows to 'results', including
3565 * those marked hidden by secchan (e.g., by in-band control). */
3567 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3569 struct cls_cursor cursor
;
3572 cls_cursor_init(&cursor
, &p
->cls
, NULL
);
3573 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3574 flow_stats_ds(p
, rule
, results
);
3579 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3580 ovs_be16 out_port
, uint8_t table_id
,
3581 struct ofp_aggregate_stats_reply
*oasr
)
3583 uint64_t total_packets
= 0;
3584 uint64_t total_bytes
= 0;
3587 COVERAGE_INC(ofproto_agg_request
);
3589 if (is_valid_table(table_id
)) {
3590 struct cls_cursor cursor
;
3593 cls_cursor_init(&cursor
, &ofproto
->cls
, target
);
3594 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3595 if (!rule_is_hidden(rule
) && rule_has_out_port(rule
, out_port
)) {
3596 uint64_t packet_count
;
3597 uint64_t byte_count
;
3599 query_stats(ofproto
, rule
, &packet_count
, &byte_count
);
3601 total_packets
+= packet_count
;
3602 total_bytes
+= byte_count
;
3608 oasr
->flow_count
= htonl(n_flows
);
3609 oasr
->packet_count
= htonll(total_packets
);
3610 oasr
->byte_count
= htonll(total_bytes
);
3611 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3615 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3616 const struct ofp_stats_request
*osr
,
3619 struct ofp_aggregate_stats_request
*request
;
3620 struct ofp_aggregate_stats_reply
*reply
;
3621 struct cls_rule target
;
3624 if (arg_size
!= sizeof *request
) {
3625 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3627 request
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3629 ofputil_cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0,
3632 msg
= start_ofp_stats_reply(osr
, sizeof *reply
);
3633 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3634 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3635 request
->table_id
, reply
);
3636 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3641 handle_nxst_aggregate(struct ofconn
*ofconn
, struct ofpbuf
*b
)
3643 struct nx_aggregate_stats_request
*request
;
3644 struct ofp_aggregate_stats_reply
*reply
;
3645 struct cls_rule target
;
3649 /* Dissect the message. */
3650 request
= ofpbuf_try_pull(b
, sizeof *request
);
3652 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3654 error
= nx_pull_match(b
, ntohs(request
->match_len
), 0, &target
);
3660 COVERAGE_INC(ofproto_flows_req
);
3661 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3662 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3663 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3664 request
->table_id
, reply
);
3665 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3670 struct queue_stats_cbdata
{
3671 struct ofconn
*ofconn
;
3672 struct ofport
*ofport
;
3677 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3678 const struct netdev_queue_stats
*stats
)
3680 struct ofp_queue_stats
*reply
;
3682 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3683 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3684 memset(reply
->pad
, 0, sizeof reply
->pad
);
3685 reply
->queue_id
= htonl(queue_id
);
3686 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3687 reply
->tx_packets
= htonll(stats
->tx_packets
);
3688 reply
->tx_errors
= htonll(stats
->tx_errors
);
3692 handle_queue_stats_dump_cb(uint32_t queue_id
,
3693 struct netdev_queue_stats
*stats
,
3696 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3698 put_queue_stats(cbdata
, queue_id
, stats
);
3702 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3703 struct queue_stats_cbdata
*cbdata
)
3705 cbdata
->ofport
= port
;
3706 if (queue_id
== OFPQ_ALL
) {
3707 netdev_dump_queue_stats(port
->netdev
,
3708 handle_queue_stats_dump_cb
, cbdata
);
3710 struct netdev_queue_stats stats
;
3712 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3713 put_queue_stats(cbdata
, queue_id
, &stats
);
3719 handle_queue_stats_request(struct ofconn
*ofconn
,
3720 const struct ofp_stats_request
*osr
,
3723 struct ofproto
*ofproto
= ofconn
->ofproto
;
3724 struct ofp_queue_stats_request
*qsr
;
3725 struct queue_stats_cbdata cbdata
;
3726 struct ofport
*port
;
3727 unsigned int port_no
;
3730 if (arg_size
!= sizeof *qsr
) {
3731 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3733 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3735 COVERAGE_INC(ofproto_queue_req
);
3737 cbdata
.ofconn
= ofconn
;
3738 cbdata
.msg
= start_ofp_stats_reply(osr
, 128);
3740 port_no
= ntohs(qsr
->port_no
);
3741 queue_id
= ntohl(qsr
->queue_id
);
3742 if (port_no
== OFPP_ALL
) {
3743 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3744 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3746 } else if (port_no
< ofproto
->max_ports
) {
3747 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3749 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3752 ofpbuf_delete(cbdata
.msg
);
3753 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3755 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3761 handle_vendor_stats_request(struct ofconn
*ofconn
,
3762 struct ofp_stats_request
*osr
, size_t arg_size
)
3764 struct nicira_stats_msg
*nsm
;
3769 VLOG_WARN_RL(&rl
, "truncated vendor stats request body");
3770 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3773 memcpy(&vendor
, osr
->body
, sizeof vendor
);
3774 if (vendor
!= htonl(NX_VENDOR_ID
)) {
3775 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3778 if (ntohs(osr
->header
.length
) < sizeof(struct nicira_stats_msg
)) {
3779 VLOG_WARN_RL(&rl
, "truncated Nicira stats request");
3780 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3783 nsm
= (struct nicira_stats_msg
*) osr
;
3785 b
.size
= ntohs(nsm
->header
.length
);
3786 switch (ntohl(nsm
->subtype
)) {
3788 return handle_nxst_flow(ofconn
, &b
);
3790 case NXST_AGGREGATE
:
3791 return handle_nxst_aggregate(ofconn
, &b
);
3794 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
3799 handle_stats_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3801 struct ofp_stats_request
*osr
;
3805 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3810 osr
= (struct ofp_stats_request
*) oh
;
3812 switch (ntohs(osr
->type
)) {
3814 return handle_desc_stats_request(ofconn
, osr
);
3817 return handle_flow_stats_request(ofconn
, osr
, arg_size
);
3819 case OFPST_AGGREGATE
:
3820 return handle_aggregate_stats_request(ofconn
, osr
, arg_size
);
3823 return handle_table_stats_request(ofconn
, osr
);
3826 return handle_port_stats_request(ofconn
, osr
, arg_size
);
3829 return handle_queue_stats_request(ofconn
, osr
, arg_size
);
3832 return handle_vendor_stats_request(ofconn
, osr
, arg_size
);
3835 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3839 static long long int
3840 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3842 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3846 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3847 const struct odp_flow_stats
*stats
)
3849 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3850 if (used
> facet
->used
) {
3852 if (used
> facet
->rule
->used
) {
3853 facet
->rule
->used
= used
;
3855 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3859 /* Folds the statistics from 'stats' into the counters in 'facet'.
3861 * Because of the meaning of a facet's counters, it only makes sense to do this
3862 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3863 * packet that was sent by hand or if it represents statistics that have been
3864 * cleared out of the datapath. */
3866 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3867 const struct odp_flow_stats
*stats
)
3869 if (stats
->n_packets
) {
3870 facet_update_time(ofproto
, facet
, stats
);
3871 facet
->packet_count
+= stats
->n_packets
;
3872 facet
->byte_count
+= stats
->n_bytes
;
3873 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3881 uint16_t idle_timeout
;
3882 uint16_t hard_timeout
;
3886 union ofp_action
*actions
;
3890 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3891 * in which no matching flow already exists in the flow table.
3893 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3894 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3895 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3897 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3900 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3902 struct ofproto
*p
= ofconn
->ofproto
;
3903 struct ofpbuf
*packet
;
3908 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3909 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3910 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3914 if (fm
->buffer_id
!= UINT32_MAX
) {
3915 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3919 in_port
= UINT16_MAX
;
3922 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
3923 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3924 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3925 rule_insert(p
, rule
);
3927 rule_execute(p
, rule
, in_port
, packet
);
3932 static struct rule
*
3933 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3935 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
3939 send_buffered_packet(struct ofconn
*ofconn
,
3940 struct rule
*rule
, uint32_t buffer_id
)
3942 struct ofpbuf
*packet
;
3946 if (buffer_id
== UINT32_MAX
) {
3950 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
3955 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
3960 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3962 struct modify_flows_cbdata
{
3963 struct ofproto
*ofproto
;
3964 const struct flow_mod
*fm
;
3968 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
3971 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3972 * encoded by ofp_mkerr() on failure.
3974 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3977 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3979 struct ofproto
*p
= ofconn
->ofproto
;
3980 struct rule
*match
= NULL
;
3981 struct cls_cursor cursor
;
3984 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
3985 CLS_CURSOR_FOR_EACH (rule
, cr
, &cursor
) {
3986 if (!rule_is_hidden(rule
)) {
3988 modify_flow(p
, fm
, rule
);
3993 /* This credits the packet to whichever flow happened to match last.
3994 * That's weird. Maybe we should do a lookup for the flow that
3995 * actually matches the packet? Who knows. */
3996 send_buffered_packet(ofconn
, match
, fm
->buffer_id
);
3999 return add_flow(ofconn
, fm
);
4003 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4004 * code as encoded by ofp_mkerr() on failure.
4006 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4009 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4011 struct ofproto
*p
= ofconn
->ofproto
;
4012 struct rule
*rule
= find_flow_strict(p
, fm
);
4013 if (rule
&& !rule_is_hidden(rule
)) {
4014 modify_flow(p
, fm
, rule
);
4015 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
4017 return add_flow(ofconn
, fm
);
4021 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4022 * been identified as a flow in 'p''s flow table to be modified, by changing
4023 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4024 * ofp_action[] structures). */
4026 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
4028 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
4030 rule
->flow_cookie
= fm
->cookie
;
4032 /* If the actions are the same, do nothing. */
4033 if (fm
->n_actions
== rule
->n_actions
4035 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
4039 /* Replace actions. */
4040 free(rule
->actions
);
4041 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
4042 rule
->n_actions
= fm
->n_actions
;
4044 p
->need_revalidate
= true;
4049 /* OFPFC_DELETE implementation. */
4051 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
4053 /* Implements OFPFC_DELETE. */
4055 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
4057 struct rule
*rule
, *next_rule
;
4058 struct cls_cursor cursor
;
4060 cls_cursor_init(&cursor
, &p
->cls
, &fm
->cr
);
4061 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4062 delete_flow(p
, rule
, htons(fm
->out_port
));
4066 /* Implements OFPFC_DELETE_STRICT. */
4068 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4070 struct rule
*rule
= find_flow_strict(p
, fm
);
4072 delete_flow(p
, rule
, htons(fm
->out_port
));
4076 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4077 * been identified as a flow to delete from 'p''s flow table, by deleting the
4078 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4081 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4082 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4083 * specified 'out_port'. */
4085 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4087 if (rule_is_hidden(rule
)) {
4091 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4095 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4096 rule_remove(p
, rule
);
4100 flow_mod_core(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4102 struct ofproto
*p
= ofconn
->ofproto
;
4105 error
= reject_slave_controller(ofconn
, "flow_mod");
4110 error
= validate_actions(fm
->actions
, fm
->n_actions
,
4111 &fm
->cr
.flow
, p
->max_ports
);
4116 /* We do not support the emergency flow cache. It will hopefully
4117 * get dropped from OpenFlow in the near future. */
4118 if (fm
->flags
& OFPFF_EMERG
) {
4119 /* There isn't a good fit for an error code, so just state that the
4120 * flow table is full. */
4121 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4124 switch (fm
->command
) {
4126 return add_flow(ofconn
, fm
);
4129 return modify_flows_loose(ofconn
, fm
);
4131 case OFPFC_MODIFY_STRICT
:
4132 return modify_flow_strict(ofconn
, fm
);
4135 delete_flows_loose(p
, fm
);
4138 case OFPFC_DELETE_STRICT
:
4139 delete_flow_strict(p
, fm
);
4143 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4148 handle_ofpt_flow_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4150 struct ofp_match orig_match
;
4151 struct ofp_flow_mod
*ofm
;
4157 b
.size
= ntohs(oh
->length
);
4159 /* Dissect the message. */
4160 ofm
= ofpbuf_try_pull(&b
, sizeof *ofm
);
4162 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4164 error
= ofputil_pull_actions(&b
, b
.size
, &fm
.actions
, &fm
.n_actions
);
4169 /* Normalize ofm->match. If normalization actually changes anything, then
4170 * log the differences. */
4171 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
4172 orig_match
= ofm
->match
;
4173 normalize_match(&ofm
->match
);
4174 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
4175 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
4176 if (!VLOG_DROP_INFO(&normal_rl
)) {
4177 char *old
= ofp_match_to_literal_string(&orig_match
);
4178 char *new = ofp_match_to_literal_string(&ofm
->match
);
4179 VLOG_INFO("%s: normalization changed ofp_match, details:",
4180 rconn_get_name(ofconn
->rconn
));
4181 VLOG_INFO(" pre: %s", old
);
4182 VLOG_INFO("post: %s", new);
4188 /* Translate the message. */
4189 ofputil_cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
),
4190 ofconn
->flow_format
, ofm
->cookie
, &fm
.cr
);
4191 fm
.cookie
= ofm
->cookie
;
4192 fm
.command
= ntohs(ofm
->command
);
4193 fm
.idle_timeout
= ntohs(ofm
->idle_timeout
);
4194 fm
.hard_timeout
= ntohs(ofm
->hard_timeout
);
4195 fm
.buffer_id
= ntohl(ofm
->buffer_id
);
4196 fm
.out_port
= ntohs(ofm
->out_port
);
4197 fm
.flags
= ntohs(ofm
->flags
);
4199 /* Execute the command. */
4200 return flow_mod_core(ofconn
, &fm
);
4204 handle_nxt_flow_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4206 struct nx_flow_mod
*nfm
;
4212 b
.size
= ntohs(oh
->length
);
4214 /* Dissect the message. */
4215 nfm
= ofpbuf_try_pull(&b
, sizeof *nfm
);
4217 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4219 error
= nx_pull_match(&b
, ntohs(nfm
->match_len
), ntohs(nfm
->priority
),
4224 error
= ofputil_pull_actions(&b
, b
.size
, &fm
.actions
, &fm
.n_actions
);
4229 /* Translate the message. */
4230 fm
.cookie
= nfm
->cookie
;
4231 fm
.command
= ntohs(nfm
->command
);
4232 fm
.idle_timeout
= ntohs(nfm
->idle_timeout
);
4233 fm
.hard_timeout
= ntohs(nfm
->hard_timeout
);
4234 fm
.buffer_id
= ntohl(nfm
->buffer_id
);
4235 fm
.out_port
= ntohs(nfm
->out_port
);
4236 fm
.flags
= ntohs(nfm
->flags
);
4238 /* Execute the command. */
4239 return flow_mod_core(ofconn
, &fm
);
4243 handle_tun_id_from_cookie(struct ofconn
*ofconn
, struct nxt_tun_id_cookie
*msg
)
4247 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
4252 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4257 handle_role_request(struct ofconn
*ofconn
, struct nicira_header
*msg
)
4259 struct nx_role_request
*nrr
;
4260 struct nx_role_request
*reply
;
4264 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
4265 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
4266 ntohs(msg
->header
.length
), sizeof *nrr
);
4267 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4269 nrr
= (struct nx_role_request
*) msg
;
4271 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4272 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4274 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4277 role
= ntohl(nrr
->role
);
4278 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4279 && role
!= NX_ROLE_SLAVE
) {
4280 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4282 /* There's no good error code for this. */
4283 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4286 if (role
== NX_ROLE_MASTER
) {
4287 struct ofconn
*other
;
4289 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4290 if (other
->role
== NX_ROLE_MASTER
) {
4291 other
->role
= NX_ROLE_SLAVE
;
4295 ofconn
->role
= role
;
4297 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, msg
->header
.xid
,
4299 reply
->role
= htonl(role
);
4300 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4306 handle_nxt_set_flow_format(struct ofconn
*ofconn
,
4307 struct nxt_set_flow_format
*msg
)
4312 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
4317 format
= ntohl(msg
->format
);
4318 if (format
== NXFF_OPENFLOW10
4319 || format
== NXFF_TUN_ID_FROM_COOKIE
4320 || format
== NXFF_NXM
) {
4321 ofconn
->flow_format
= format
;
4324 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4329 handle_vendor(struct ofconn
*ofconn
, void *msg
)
4331 struct ofproto
*p
= ofconn
->ofproto
;
4332 struct ofp_vendor_header
*ovh
= msg
;
4333 struct nicira_header
*nh
;
4335 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
4336 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
4337 "(expected at least %zu)",
4338 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
4339 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4341 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
4342 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
4344 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
4345 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
4346 "(expected at least %zu)",
4347 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
4348 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4352 switch (ntohl(nh
->subtype
)) {
4353 case NXT_STATUS_REQUEST
:
4354 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
4357 case NXT_TUN_ID_FROM_COOKIE
:
4358 return handle_tun_id_from_cookie(ofconn
, msg
);
4360 case NXT_ROLE_REQUEST
:
4361 return handle_role_request(ofconn
, msg
);
4363 case NXT_SET_FLOW_FORMAT
:
4364 return handle_nxt_set_flow_format(ofconn
, msg
);
4367 return handle_nxt_flow_mod(ofconn
, &ovh
->header
);
4370 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
4374 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4376 struct ofp_header
*ob
;
4379 /* Currently, everything executes synchronously, so we can just
4380 * immediately send the barrier reply. */
4381 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4382 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4387 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4389 struct ofp_header
*oh
= ofp_msg
->data
;
4392 COVERAGE_INC(ofproto_recv_openflow
);
4394 case OFPT_ECHO_REQUEST
:
4395 error
= handle_echo_request(ofconn
, oh
);
4398 case OFPT_ECHO_REPLY
:
4402 case OFPT_FEATURES_REQUEST
:
4403 error
= handle_features_request(ofconn
, oh
);
4406 case OFPT_GET_CONFIG_REQUEST
:
4407 error
= handle_get_config_request(ofconn
, oh
);
4410 case OFPT_SET_CONFIG
:
4411 error
= handle_set_config(ofconn
, ofp_msg
->data
);
4414 case OFPT_PACKET_OUT
:
4415 error
= handle_packet_out(ofconn
, ofp_msg
->data
);
4419 error
= handle_port_mod(ofconn
, oh
);
4423 error
= handle_ofpt_flow_mod(ofconn
, ofp_msg
->data
);
4426 case OFPT_STATS_REQUEST
:
4427 error
= handle_stats_request(ofconn
, oh
);
4431 error
= handle_vendor(ofconn
, ofp_msg
->data
);
4434 case OFPT_BARRIER_REQUEST
:
4435 error
= handle_barrier_request(ofconn
, oh
);
4439 if (VLOG_IS_WARN_ENABLED()) {
4440 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4441 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4444 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4449 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4454 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4456 struct odp_msg
*msg
= packet
->data
;
4457 struct ofpbuf payload
;
4458 struct facet
*facet
;
4461 payload
.data
= msg
+ 1;
4462 payload
.size
= msg
->length
- sizeof *msg
;
4463 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4465 packet
->l2
= payload
.l2
;
4466 packet
->l3
= payload
.l3
;
4467 packet
->l4
= payload
.l4
;
4468 packet
->l7
= payload
.l7
;
4470 /* Check with in-band control to see if this packet should be sent
4471 * to the local port regardless of the flow table. */
4472 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4473 union odp_action action
;
4475 memset(&action
, 0, sizeof(action
));
4476 action
.output
.type
= ODPAT_OUTPUT
;
4477 action
.output
.port
= ODPP_LOCAL
;
4478 dpif_execute(p
->dpif
, &action
, 1, &payload
);
4481 facet
= facet_lookup_valid(p
, &flow
);
4483 struct rule
*rule
= rule_lookup(p
, &flow
);
4485 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4486 struct ofport
*port
= get_port(p
, msg
->port
);
4488 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4489 COVERAGE_INC(ofproto_no_packet_in
);
4490 /* XXX install 'drop' flow entry */
4491 ofpbuf_delete(packet
);
4495 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4499 COVERAGE_INC(ofproto_packet_in
);
4500 send_packet_in(p
, packet
);
4504 facet
= facet_create(p
, rule
, &flow
, packet
);
4505 } else if (!facet
->may_install
) {
4506 /* The facet is not installable, that is, we need to process every
4507 * packet, so process the current packet's actions into 'facet'. */
4508 facet_make_actions(p
, facet
, packet
);
4511 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4513 * Extra-special case for fail-open mode.
4515 * We are in fail-open mode and the packet matched the fail-open rule,
4516 * but we are connected to a controller too. We should send the packet
4517 * up to the controller in the hope that it will try to set up a flow
4518 * and thereby allow us to exit fail-open.
4520 * See the top-level comment in fail-open.c for more information.
4522 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4523 DPIF_RECV_MSG_PADDING
));
4526 ofpbuf_pull(packet
, sizeof *msg
);
4527 facet_execute(p
, facet
, packet
);
4528 facet_install(p
, facet
, false);
4532 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4534 struct odp_msg
*msg
= packet
->data
;
4536 switch (msg
->type
) {
4537 case _ODPL_ACTION_NR
:
4538 COVERAGE_INC(ofproto_ctlr_action
);
4539 send_packet_in(p
, packet
);
4542 case _ODPL_SFLOW_NR
:
4544 ofproto_sflow_received(p
->sflow
, msg
);
4546 ofpbuf_delete(packet
);
4550 handle_odp_miss_msg(p
, packet
);
4554 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4560 /* Flow expiration. */
4562 static int ofproto_dp_max_idle(const struct ofproto
*);
4563 static void ofproto_update_used(struct ofproto
*);
4564 static void rule_expire(struct ofproto
*, struct rule
*);
4565 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4567 /* This function is called periodically by ofproto_run(). Its job is to
4568 * collect updates for the flows that have been installed into the datapath,
4569 * most importantly when they last were used, and then use that information to
4570 * expire flows that have not been used recently.
4572 * Returns the number of milliseconds after which it should be called again. */
4574 ofproto_expire(struct ofproto
*ofproto
)
4576 struct rule
*rule
, *next_rule
;
4577 struct cls_cursor cursor
;
4580 /* Update 'used' for each flow in the datapath. */
4581 ofproto_update_used(ofproto
);
4583 /* Expire facets that have been idle too long. */
4584 dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4585 ofproto_expire_facets(ofproto
, dp_max_idle
);
4587 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4588 cls_cursor_init(&cursor
, &ofproto
->cls
, NULL
);
4589 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, cr
, &cursor
) {
4590 rule_expire(ofproto
, rule
);
4593 /* Let the hook know that we're at a stable point: all outstanding data
4594 * in existing flows has been accounted to the account_cb. Thus, the
4595 * hook can now reasonably do operations that depend on having accurate
4596 * flow volume accounting (currently, that's just bond rebalancing). */
4597 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4598 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4601 return MIN(dp_max_idle
, 1000);
4604 /* Update 'used' member of installed facets. */
4606 ofproto_update_used(struct ofproto
*p
)
4608 struct odp_flow
*flows
;
4613 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4618 for (i
= 0; i
< n_flows
; i
++) {
4619 struct odp_flow
*f
= &flows
[i
];
4620 struct facet
*facet
;
4623 odp_flow_key_to_flow(&f
->key
, &flow
);
4624 facet
= facet_find(p
, &flow
);
4626 if (facet
&& facet
->installed
) {
4627 facet_update_time(p
, facet
, &f
->stats
);
4628 facet_account(p
, facet
, f
->stats
.n_bytes
);
4630 /* There's a flow in the datapath that we know nothing about.
4632 COVERAGE_INC(ofproto_unexpected_rule
);
4633 dpif_flow_del(p
->dpif
, f
);
4640 /* Calculates and returns the number of milliseconds of idle time after which
4641 * facets should expire from the datapath and we should fold their statistics
4642 * into their parent rules in userspace. */
4644 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4647 * Idle time histogram.
4649 * Most of the time a switch has a relatively small number of facets. When
4650 * this is the case we might as well keep statistics for all of them in
4651 * userspace and to cache them in the kernel datapath for performance as
4654 * As the number of facets increases, the memory required to maintain
4655 * statistics about them in userspace and in the kernel becomes
4656 * significant. However, with a large number of facets it is likely that
4657 * only a few of them are "heavy hitters" that consume a large amount of
4658 * bandwidth. At this point, only heavy hitters are worth caching in the
4659 * kernel and maintaining in userspaces; other facets we can discard.
4661 * The technique used to compute the idle time is to build a histogram with
4662 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4663 * that is installed in the kernel gets dropped in the appropriate bucket.
4664 * After the histogram has been built, we compute the cutoff so that only
4665 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4666 * cached. At least the most-recently-used bucket of facets is kept, so
4667 * actually an arbitrary number of facets can be kept in any given
4668 * expiration run (though the next run will delete most of those unless
4669 * they receive additional data).
4671 * This requires a second pass through the facets, in addition to the pass
4672 * made by ofproto_update_used(), because the former function never looks
4673 * at uninstallable facets.
4675 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4676 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4677 int buckets
[N_BUCKETS
] = { 0 };
4678 struct facet
*facet
;
4683 total
= hmap_count(&ofproto
->facets
);
4684 if (total
<= 1000) {
4685 return N_BUCKETS
* BUCKET_WIDTH
;
4688 /* Build histogram. */
4690 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4691 long long int idle
= now
- facet
->used
;
4692 int bucket
= (idle
<= 0 ? 0
4693 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4694 : (unsigned int) idle
/ BUCKET_WIDTH
);
4698 /* Find the first bucket whose flows should be expired. */
4699 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4700 if (buckets
[bucket
]) {
4703 subtotal
+= buckets
[bucket
++];
4704 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4709 if (VLOG_IS_DBG_ENABLED()) {
4713 ds_put_cstr(&s
, "keep");
4714 for (i
= 0; i
< N_BUCKETS
; i
++) {
4716 ds_put_cstr(&s
, ", drop");
4719 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4722 VLOG_INFO("%s: %s (msec:count)",
4723 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4727 return bucket
* BUCKET_WIDTH
;
4731 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4733 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4734 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4735 struct ofexpired expired
;
4736 struct odp_flow odp_flow
;
4738 /* Get updated flow stats.
4740 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4741 * updated TCP flags and (2) the dpif_flow_list_all() in
4742 * ofproto_update_used() zeroed TCP flags. */
4743 memset(&odp_flow
, 0, sizeof odp_flow
);
4744 if (facet
->installed
) {
4745 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
4746 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4747 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4749 if (odp_flow
.stats
.n_packets
) {
4750 facet_update_time(ofproto
, facet
, &odp_flow
.stats
);
4751 netflow_flow_update_flags(&facet
->nf_flow
,
4752 odp_flow
.stats
.tcp_flags
);
4756 expired
.flow
= facet
->flow
;
4757 expired
.packet_count
= facet
->packet_count
+
4758 odp_flow
.stats
.n_packets
;
4759 expired
.byte_count
= facet
->byte_count
+ odp_flow
.stats
.n_bytes
;
4760 expired
.used
= facet
->used
;
4762 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4767 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4769 long long int cutoff
= time_msec() - dp_max_idle
;
4770 struct facet
*facet
, *next_facet
;
4772 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4773 facet_active_timeout(ofproto
, facet
);
4774 if (facet
->used
< cutoff
) {
4775 facet_remove(ofproto
, facet
);
4780 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4781 * then delete it entirely. */
4783 rule_expire(struct ofproto
*ofproto
, struct rule
*rule
)
4785 struct facet
*facet
, *next_facet
;
4789 /* Has 'rule' expired? */
4791 if (rule
->hard_timeout
4792 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4793 reason
= OFPRR_HARD_TIMEOUT
;
4794 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4795 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4796 reason
= OFPRR_IDLE_TIMEOUT
;
4801 COVERAGE_INC(ofproto_expired
);
4803 /* Update stats. (This is a no-op if the rule expired due to an idle
4804 * timeout, because that only happens when the rule has no facets left.) */
4805 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4806 facet_remove(ofproto
, facet
);
4809 /* Get rid of the rule. */
4810 if (!rule_is_hidden(rule
)) {
4811 rule_send_removed(ofproto
, rule
, reason
);
4813 rule_remove(ofproto
, rule
);
4816 static struct ofpbuf
*
4817 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4820 struct ofp_flow_removed
*ofr
;
4823 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4824 ofputil_cls_rule_to_match(&rule
->cr
, ofconn
->flow_format
, &ofr
->match
);
4825 ofr
->cookie
= rule
->flow_cookie
;
4826 ofr
->priority
= htons(rule
->cr
.priority
);
4827 ofr
->reason
= reason
;
4828 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4829 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4830 ofr
->packet_count
= htonll(rule
->packet_count
);
4831 ofr
->byte_count
= htonll(rule
->byte_count
);
4836 static struct ofpbuf
*
4837 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4839 struct nx_flow_removed
*nfr
;
4843 nfr
= make_nxmsg(sizeof *nfr
, NXT_FLOW_REMOVED
, &buf
);
4845 match_len
= nx_put_match(buf
, &rule
->cr
);
4847 nfr
->cookie
= rule
->flow_cookie
;
4848 nfr
->priority
= htons(rule
->cr
.priority
);
4849 nfr
->reason
= reason
;
4850 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4851 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4852 nfr
->match_len
= htons(match_len
);
4853 nfr
->packet_count
= htonll(rule
->packet_count
);
4854 nfr
->byte_count
= htonll(rule
->byte_count
);
4860 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4862 struct ofconn
*ofconn
;
4864 if (!rule
->send_flow_removed
) {
4868 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4871 if (!rconn_is_connected(ofconn
->rconn
)
4872 || !ofconn_receives_async_msgs(ofconn
)) {
4876 msg
= (ofconn
->flow_format
== NXFF_NXM
4877 ? compose_nx_flow_removed(rule
, reason
)
4878 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4880 /* Account flow expirations under ofconn->reply_counter, the counter
4881 * for replies to OpenFlow requests. That works because preventing
4882 * OpenFlow requests from being processed also prevents new flows from
4883 * being added (and expiring). (It also prevents processing OpenFlow
4884 * requests that would not add new flows, so it is imperfect.) */
4885 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4889 /* pinsched callback for sending 'packet' on 'ofconn'. */
4891 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4893 struct ofconn
*ofconn
= ofconn_
;
4895 rconn_send_with_limit(ofconn
->rconn
, packet
,
4896 ofconn
->packet_in_counter
, 100);
4899 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4900 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4901 * packet scheduler for sending.
4903 * 'max_len' specifies the maximum number of bytes of the packet to send on
4904 * 'ofconn' (INT_MAX specifies no limit).
4906 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4907 * ownership is transferred to this function. */
4909 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4912 struct ofproto
*ofproto
= ofconn
->ofproto
;
4913 struct ofp_packet_in
*opi
= packet
->data
;
4914 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4915 int send_len
, trim_size
;
4919 if (opi
->reason
== OFPR_ACTION
) {
4920 buffer_id
= UINT32_MAX
;
4921 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4922 buffer_id
= pktbuf_get_null();
4923 } else if (!ofconn
->pktbuf
) {
4924 buffer_id
= UINT32_MAX
;
4926 struct ofpbuf payload
;
4927 payload
.data
= opi
->data
;
4928 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4929 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4932 /* Figure out how much of the packet to send. */
4933 send_len
= ntohs(opi
->total_len
);
4934 if (buffer_id
!= UINT32_MAX
) {
4935 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4937 send_len
= MIN(send_len
, max_len
);
4939 /* Adjust packet length and clone if necessary. */
4940 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4942 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4945 packet
->size
= trim_size
;
4948 /* Update packet headers. */
4949 opi
->buffer_id
= htonl(buffer_id
);
4950 update_openflow_length(packet
);
4952 /* Hand over to packet scheduler. It might immediately call into
4953 * do_send_packet_in() or it might buffer it for a while (until a later
4954 * call to pinsched_run()). */
4955 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4956 packet
, do_send_packet_in
, ofconn
);
4959 /* Replace struct odp_msg header in 'packet' by equivalent struct
4960 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4961 * returned by dpif_recv()).
4963 * The conversion is not complete: the caller still needs to trim any unneeded
4964 * payload off the end of the buffer, set the length in the OpenFlow header,
4965 * and set buffer_id. Those require us to know the controller settings and so
4966 * must be done on a per-controller basis.
4968 * Returns the maximum number of bytes of the packet that should be sent to
4969 * the controller (INT_MAX if no limit). */
4971 do_convert_to_packet_in(struct ofpbuf
*packet
)
4973 struct odp_msg
*msg
= packet
->data
;
4974 struct ofp_packet_in
*opi
;
4980 /* Extract relevant header fields */
4981 if (msg
->type
== _ODPL_ACTION_NR
) {
4982 reason
= OFPR_ACTION
;
4985 reason
= OFPR_NO_MATCH
;
4988 total_len
= msg
->length
- sizeof *msg
;
4989 in_port
= odp_port_to_ofp_port(msg
->port
);
4991 /* Repurpose packet buffer by overwriting header. */
4992 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4993 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4994 opi
->header
.version
= OFP_VERSION
;
4995 opi
->header
.type
= OFPT_PACKET_IN
;
4996 opi
->total_len
= htons(total_len
);
4997 opi
->in_port
= htons(in_port
);
4998 opi
->reason
= reason
;
5003 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
5004 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
5005 * as necessary according to their individual configurations.
5007 * 'packet' must have sufficient headroom to convert it into a struct
5008 * ofp_packet_in (e.g. as returned by dpif_recv()).
5010 * Takes ownership of 'packet'. */
5012 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
5014 struct ofconn
*ofconn
, *prev
;
5017 max_len
= do_convert_to_packet_in(packet
);
5020 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
5021 if (ofconn_receives_async_msgs(ofconn
)) {
5023 schedule_packet_in(prev
, packet
, max_len
, true);
5029 schedule_packet_in(prev
, packet
, max_len
, false);
5031 ofpbuf_delete(packet
);
5036 pick_datapath_id(const struct ofproto
*ofproto
)
5038 const struct ofport
*port
;
5040 port
= get_port(ofproto
, ODPP_LOCAL
);
5042 uint8_t ea
[ETH_ADDR_LEN
];
5045 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5047 return eth_addr_to_uint64(ea
);
5049 VLOG_WARN("could not get MAC address for %s (%s)",
5050 netdev_get_name(port
->netdev
), strerror(error
));
5052 return ofproto
->fallback_dpid
;
5056 pick_fallback_dpid(void)
5058 uint8_t ea
[ETH_ADDR_LEN
];
5059 eth_addr_nicira_random(ea
);
5060 return eth_addr_to_uint64(ea
);
5064 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5065 struct odp_actions
*actions
, tag_type
*tags
,
5066 uint16_t *nf_output_iface
, void *ofproto_
)
5068 struct ofproto
*ofproto
= ofproto_
;
5071 /* Drop frames for reserved multicast addresses. */
5072 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5076 /* Learn source MAC (but don't try to learn from revalidation). */
5077 if (packet
!= NULL
) {
5078 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
5080 GRAT_ARP_LOCK_NONE
);
5082 /* The log messages here could actually be useful in debugging,
5083 * so keep the rate limit relatively high. */
5084 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5085 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5086 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5087 ofproto_revalidate(ofproto
, rev_tag
);
5091 /* Determine output port. */
5092 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
5095 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5096 nf_output_iface
, actions
);
5097 } else if (out_port
!= flow
->in_port
) {
5098 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
5099 *nf_output_iface
= out_port
;
5107 static const struct ofhooks default_ofhooks
= {
5108 default_normal_ofhook_cb
,