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 struct ofpbuf
*packet
, uint16_t in_port
);
129 static void rule_remove(struct ofproto
*, struct rule
*);
131 static void rule_send_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
133 /* An exact-match instantiation of an OpenFlow flow. */
135 long long int used
; /* Time last used; time created if not used. */
139 * - Do include packets and bytes sent "by hand", e.g. with
142 * - Do include packets and bytes that were obtained from the datapath
143 * when a flow was deleted (e.g. dpif_flow_del()) or when its
144 * statistics were reset (e.g. dpif_flow_put() with ODPPF_ZERO_STATS).
146 * - Do not include any packets or bytes that can currently be obtained
147 * from the datapath by, e.g., dpif_flow_get().
149 uint64_t packet_count
; /* Number of packets received. */
150 uint64_t byte_count
; /* Number of bytes received. */
152 /* Number of bytes passed to account_cb. This may include bytes that can
153 * currently obtained from the datapath (thus, it can be greater than
155 uint64_t accounted_bytes
;
157 struct hmap_node hmap_node
; /* In owning ofproto's 'facets' hmap. */
158 struct list list_node
; /* In owning rule's 'facets' list. */
159 struct rule
*rule
; /* Owning rule. */
160 struct flow flow
; /* Exact-match flow. */
161 bool installed
; /* Installed in datapath? */
162 bool may_install
; /* True ordinarily; false if actions must
163 * be reassessed for every packet. */
164 int n_actions
; /* Number of elements in actions[]. */
165 union odp_action
*actions
; /* Datapath actions. */
166 tag_type tags
; /* Tags (set only by hooks). */
167 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
170 static struct facet
*facet_create(struct ofproto
*, struct rule
*,
172 const struct ofpbuf
*packet
);
173 static void facet_remove(struct ofproto
*, struct facet
*);
174 static void facet_free(struct facet
*);
176 static struct facet
*facet_lookup_valid(struct ofproto
*, const struct flow
*);
177 static bool facet_revalidate(struct ofproto
*, struct facet
*);
179 static void facet_install(struct ofproto
*, struct facet
*, bool zero_stats
);
180 static void facet_uninstall(struct ofproto
*, struct facet
*);
181 static void facet_flush_stats(struct ofproto
*, struct facet
*);
183 static void facet_make_actions(struct ofproto
*, struct facet
*,
184 const struct ofpbuf
*packet
);
185 static void facet_update_stats(struct ofproto
*, struct facet
*,
186 const struct odp_flow_stats
*);
188 /* ofproto supports two kinds of OpenFlow connections:
190 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
191 * maintains persistent connections to these controllers and by default
192 * sends them asynchronous messages such as packet-ins.
194 * - "Service" connections, e.g. from ovs-ofctl. When these connections
195 * drop, it is the other side's responsibility to reconnect them if
196 * necessary. ofproto does not send them asynchronous messages by default.
198 * Currently, active (tcp, ssl, unix) connections are always "primary"
199 * connections and passive (ptcp, pssl, punix) connections are always "service"
200 * connections. There is no inherent reason for this, but it reflects the
204 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
205 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
208 /* A listener for incoming OpenFlow "service" connections. */
210 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
211 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
213 /* These are not used by ofservice directly. They are settings for
214 * accepted "struct ofconn"s from the pvconn. */
215 int probe_interval
; /* Max idle time before probing, in seconds. */
216 int rate_limit
; /* Max packet-in rate in packets per second. */
217 int burst_limit
; /* Limit on accumulating packet credits. */
220 static struct ofservice
*ofservice_lookup(struct ofproto
*,
222 static int ofservice_create(struct ofproto
*,
223 const struct ofproto_controller
*);
224 static void ofservice_reconfigure(struct ofservice
*,
225 const struct ofproto_controller
*);
226 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
228 /* An OpenFlow connection. */
230 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
231 struct list node
; /* In struct ofproto's "all_conns" list. */
232 struct rconn
*rconn
; /* OpenFlow connection. */
233 enum ofconn_type type
; /* Type. */
234 int flow_format
; /* One of NXFF_*. */
236 /* OFPT_PACKET_IN related data. */
237 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
238 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
239 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
240 int miss_send_len
; /* Bytes to send of buffered packets. */
242 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
243 * requests, and the maximum number before we stop reading OpenFlow
245 #define OFCONN_REPLY_MAX 100
246 struct rconn_packet_counter
*reply_counter
;
248 /* type == OFCONN_PRIMARY only. */
249 enum nx_role role
; /* Role. */
250 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
251 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
252 struct status_category
*ss
; /* Switch status category. */
253 enum ofproto_band band
; /* In-band or out-of-band? */
256 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
257 * "schedulers" array. Their values are 0 and 1, and their meanings and values
258 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
259 * case anything ever changes, check their values here. */
260 #define N_SCHEDULERS 2
261 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
262 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
263 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
264 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
266 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
268 static void ofconn_destroy(struct ofconn
*);
269 static void ofconn_run(struct ofconn
*);
270 static void ofconn_wait(struct ofconn
*);
271 static bool ofconn_receives_async_msgs(const struct ofconn
*);
272 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
273 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
275 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
276 struct rconn_packet_counter
*counter
);
278 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
279 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
283 uint64_t datapath_id
; /* Datapath ID. */
284 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
285 char *mfr_desc
; /* Manufacturer. */
286 char *hw_desc
; /* Hardware. */
287 char *sw_desc
; /* Software version. */
288 char *serial_desc
; /* Serial number. */
289 char *dp_desc
; /* Datapath description. */
293 struct netdev_monitor
*netdev_monitor
;
294 struct hmap ports
; /* Contains "struct ofport"s. */
295 struct shash port_by_name
;
299 struct switch_status
*switch_status
;
300 struct fail_open
*fail_open
;
301 struct netflow
*netflow
;
302 struct ofproto_sflow
*sflow
;
304 /* In-band control. */
305 struct in_band
*in_band
;
306 long long int next_in_band_update
;
307 struct sockaddr_in
*extra_in_band_remotes
;
308 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
);
414 /* Initialize flow table. */
415 classifier_init(&p
->cls
);
416 p
->next_expiration
= time_msec() + 1000;
418 /* Initialize facet table. */
419 hmap_init(&p
->facets
);
420 p
->need_revalidate
= false;
421 tag_set_init(&p
->revalidate_set
);
423 /* Initialize OpenFlow connections. */
424 list_init(&p
->all_conns
);
425 hmap_init(&p
->controllers
);
426 hmap_init(&p
->services
);
430 /* Initialize hooks. */
432 p
->ofhooks
= ofhooks
;
436 p
->ofhooks
= &default_ofhooks
;
438 p
->ml
= mac_learning_create();
441 /* Pick final datapath ID. */
442 p
->datapath_id
= pick_datapath_id(p
);
443 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
450 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
452 uint64_t old_dpid
= p
->datapath_id
;
453 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
454 if (p
->datapath_id
!= old_dpid
) {
455 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
457 /* Force all active connections to reconnect, since there is no way to
458 * notify a controller that the datapath ID has changed. */
459 ofproto_reconnect_controllers(p
);
464 is_discovery_controller(const struct ofproto_controller
*c
)
466 return !strcmp(c
->target
, "discover");
470 is_in_band_controller(const struct ofproto_controller
*c
)
472 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
475 /* Creates a new controller in 'ofproto'. Some of the settings are initially
476 * drawn from 'c', but update_controller() needs to be called later to finish
477 * the new ofconn's configuration. */
479 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
481 struct discovery
*discovery
;
482 struct ofconn
*ofconn
;
484 if (is_discovery_controller(c
)) {
485 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
486 ofproto
->dpif
, ofproto
->switch_status
,
495 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
496 ofconn
->pktbuf
= pktbuf_create();
497 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
499 ofconn
->discovery
= discovery
;
501 char *name
= ofconn_make_name(ofproto
, c
->target
);
502 rconn_connect(ofconn
->rconn
, c
->target
, name
);
505 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
506 hash_string(c
->target
, 0));
509 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
510 * target or turn discovery on or off (these are done by creating new ofconns
511 * and deleting old ones), but it can update the rest of an ofconn's
514 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
518 ofconn
->band
= (is_in_band_controller(c
)
519 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
521 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
523 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
524 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
526 if (ofconn
->discovery
) {
527 discovery_set_update_resolv_conf(ofconn
->discovery
,
528 c
->update_resolv_conf
);
529 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
532 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
536 ofconn_get_target(const struct ofconn
*ofconn
)
538 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
541 static struct ofconn
*
542 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
544 struct ofconn
*ofconn
;
546 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
547 hash_string(target
, 0), &ofproto
->controllers
) {
548 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
556 update_in_band_remotes(struct ofproto
*ofproto
)
558 const struct ofconn
*ofconn
;
559 struct sockaddr_in
*addrs
;
560 size_t max_addrs
, n_addrs
;
564 /* Allocate enough memory for as many remotes as we could possibly have. */
565 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
566 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
569 /* Add all the remotes. */
571 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
572 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
574 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
578 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
579 if (sin
->sin_addr
.s_addr
) {
580 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
583 if (ofconn
->discovery
) {
587 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
588 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
591 /* Create or update or destroy in-band.
593 * Ordinarily we only enable in-band if there's at least one remote
594 * address, but discovery needs the in-band rules for DHCP to be installed
595 * even before we know any remote addresses. */
596 if (n_addrs
|| discovery
) {
597 if (!ofproto
->in_band
) {
598 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
601 if (ofproto
->in_band
) {
602 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
604 ofproto
->next_in_band_update
= time_msec() + 1000;
606 in_band_destroy(ofproto
->in_band
);
607 ofproto
->in_band
= NULL
;
615 update_fail_open(struct ofproto
*p
)
617 struct ofconn
*ofconn
;
619 if (!hmap_is_empty(&p
->controllers
)
620 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
621 struct rconn
**rconns
;
625 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
629 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
630 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
631 rconns
[n
++] = ofconn
->rconn
;
634 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
635 /* p->fail_open takes ownership of 'rconns'. */
637 fail_open_destroy(p
->fail_open
);
643 ofproto_set_controllers(struct ofproto
*p
,
644 const struct ofproto_controller
*controllers
,
645 size_t n_controllers
)
647 struct shash new_controllers
;
648 struct ofconn
*ofconn
, *next_ofconn
;
649 struct ofservice
*ofservice
, *next_ofservice
;
653 /* Create newly configured controllers and services.
654 * Create a name to ofproto_controller mapping in 'new_controllers'. */
655 shash_init(&new_controllers
);
656 for (i
= 0; i
< n_controllers
; i
++) {
657 const struct ofproto_controller
*c
= &controllers
[i
];
659 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
660 if (!find_controller_by_target(p
, c
->target
)) {
661 add_controller(p
, c
);
663 } else if (!pvconn_verify_name(c
->target
)) {
664 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
668 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
669 dpif_name(p
->dpif
), c
->target
);
673 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
676 /* Delete controllers that are no longer configured.
677 * Update configuration of all now-existing controllers. */
679 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
680 struct ofproto_controller
*c
;
682 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
684 ofconn_destroy(ofconn
);
686 update_controller(ofconn
, c
);
693 /* Delete services that are no longer configured.
694 * Update configuration of all now-existing services. */
695 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
696 struct ofproto_controller
*c
;
698 c
= shash_find_data(&new_controllers
,
699 pvconn_get_name(ofservice
->pvconn
));
701 ofservice_destroy(p
, ofservice
);
703 ofservice_reconfigure(ofservice
, c
);
707 shash_destroy(&new_controllers
);
709 update_in_band_remotes(p
);
712 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
713 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
714 struct ofconn
, hmap_node
);
715 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
716 rconn_status_cb
, ofconn
->rconn
);
721 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
723 p
->fail_mode
= fail_mode
;
727 /* Drops the connections between 'ofproto' and all of its controllers, forcing
728 * them to reconnect. */
730 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
732 struct ofconn
*ofconn
;
734 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
735 rconn_reconnect(ofconn
->rconn
);
740 any_extras_changed(const struct ofproto
*ofproto
,
741 const struct sockaddr_in
*extras
, size_t n
)
745 if (n
!= ofproto
->n_extra_remotes
) {
749 for (i
= 0; i
< n
; i
++) {
750 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
751 const struct sockaddr_in
*new = &extras
[i
];
753 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
754 old
->sin_port
!= new->sin_port
) {
762 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
763 * in-band control should guarantee access, in the same way that in-band
764 * control guarantees access to OpenFlow controllers. */
766 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
767 const struct sockaddr_in
*extras
, size_t n
)
769 if (!any_extras_changed(ofproto
, extras
, n
)) {
773 free(ofproto
->extra_in_band_remotes
);
774 ofproto
->n_extra_remotes
= n
;
775 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
777 update_in_band_remotes(ofproto
);
781 ofproto_set_desc(struct ofproto
*p
,
782 const char *mfr_desc
, const char *hw_desc
,
783 const char *sw_desc
, const char *serial_desc
,
786 struct ofp_desc_stats
*ods
;
789 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
790 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
791 sizeof ods
->mfr_desc
);
794 p
->mfr_desc
= xstrdup(mfr_desc
);
797 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
798 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
799 sizeof ods
->hw_desc
);
802 p
->hw_desc
= xstrdup(hw_desc
);
805 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
806 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
807 sizeof ods
->sw_desc
);
810 p
->sw_desc
= xstrdup(sw_desc
);
813 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
814 VLOG_WARN("truncating serial_desc, must be less than %zu "
816 sizeof ods
->serial_num
);
818 free(p
->serial_desc
);
819 p
->serial_desc
= xstrdup(serial_desc
);
822 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
823 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
824 sizeof ods
->dp_desc
);
827 p
->dp_desc
= xstrdup(dp_desc
);
832 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
833 const struct svec
*svec
)
835 struct pvconn
**pvconns
= *pvconnsp
;
836 size_t n_pvconns
= *n_pvconnsp
;
840 for (i
= 0; i
< n_pvconns
; i
++) {
841 pvconn_close(pvconns
[i
]);
845 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
847 for (i
= 0; i
< svec
->n
; i
++) {
848 const char *name
= svec
->names
[i
];
849 struct pvconn
*pvconn
;
852 error
= pvconn_open(name
, &pvconn
);
854 pvconns
[n_pvconns
++] = pvconn
;
856 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
864 *n_pvconnsp
= n_pvconns
;
870 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
872 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
876 ofproto_set_netflow(struct ofproto
*ofproto
,
877 const struct netflow_options
*nf_options
)
879 if (nf_options
&& nf_options
->collectors
.n
) {
880 if (!ofproto
->netflow
) {
881 ofproto
->netflow
= netflow_create();
883 return netflow_set_options(ofproto
->netflow
, nf_options
);
885 netflow_destroy(ofproto
->netflow
);
886 ofproto
->netflow
= NULL
;
892 ofproto_set_sflow(struct ofproto
*ofproto
,
893 const struct ofproto_sflow_options
*oso
)
895 struct ofproto_sflow
*os
= ofproto
->sflow
;
898 struct ofport
*ofport
;
900 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
901 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
902 ofproto_sflow_add_port(os
, ofport
->odp_port
,
903 netdev_get_name(ofport
->netdev
));
906 ofproto_sflow_set_options(os
, oso
);
908 ofproto_sflow_destroy(os
);
909 ofproto
->sflow
= NULL
;
914 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
916 return ofproto
->datapath_id
;
920 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
922 return !hmap_is_empty(&ofproto
->controllers
);
925 enum ofproto_fail_mode
926 ofproto_get_fail_mode(const struct ofproto
*p
)
932 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
936 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
937 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
942 ofproto_destroy(struct ofproto
*p
)
944 struct ofservice
*ofservice
, *next_ofservice
;
945 struct ofconn
*ofconn
, *next_ofconn
;
946 struct ofport
*ofport
, *next_ofport
;
953 /* Destroy fail-open and in-band early, since they touch the classifier. */
954 fail_open_destroy(p
->fail_open
);
957 in_band_destroy(p
->in_band
);
959 free(p
->extra_in_band_remotes
);
961 ofproto_flush_flows(p
);
962 classifier_destroy(&p
->cls
);
963 hmap_destroy(&p
->facets
);
965 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
966 ofconn_destroy(ofconn
);
968 hmap_destroy(&p
->controllers
);
971 netdev_monitor_destroy(p
->netdev_monitor
);
972 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
973 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
976 shash_destroy(&p
->port_by_name
);
978 switch_status_destroy(p
->switch_status
);
979 netflow_destroy(p
->netflow
);
980 ofproto_sflow_destroy(p
->sflow
);
982 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
983 ofservice_destroy(p
, ofservice
);
985 hmap_destroy(&p
->services
);
987 for (i
= 0; i
< p
->n_snoops
; i
++) {
988 pvconn_close(p
->snoops
[i
]);
992 mac_learning_destroy(p
->ml
);
997 free(p
->serial_desc
);
1000 hmap_destroy(&p
->ports
);
1006 ofproto_run(struct ofproto
*p
)
1008 int error
= ofproto_run1(p
);
1010 error
= ofproto_run2(p
, false);
1016 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1018 if (error
== ENOBUFS
) {
1019 reinit_ports(ofproto
);
1020 } else if (!error
) {
1021 update_port(ofproto
, devname
);
1026 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1027 * means that 'ofconn' is more interesting for monitoring than a lower return
1030 snoop_preference(const struct ofconn
*ofconn
)
1032 switch (ofconn
->role
) {
1033 case NX_ROLE_MASTER
:
1040 /* Shouldn't happen. */
1045 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1046 * Connects this vconn to a controller. */
1048 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1050 struct ofconn
*ofconn
, *best
;
1052 /* Pick a controller for monitoring. */
1054 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1055 if (ofconn
->type
== OFCONN_PRIMARY
1056 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1062 rconn_add_monitor(best
->rconn
, vconn
);
1064 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1070 ofproto_run1(struct ofproto
*p
)
1072 struct ofconn
*ofconn
, *next_ofconn
;
1073 struct ofservice
*ofservice
;
1078 if (shash_is_empty(&p
->port_by_name
)) {
1082 for (i
= 0; i
< 50; i
++) {
1085 error
= dpif_recv(p
->dpif
, &buf
);
1087 if (error
== ENODEV
) {
1088 /* Someone destroyed the datapath behind our back. The caller
1089 * better destroy us and give up, because we're just going to
1090 * spin from here on out. */
1091 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1092 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1093 dpif_name(p
->dpif
));
1099 handle_odp_msg(p
, buf
);
1102 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1103 process_port_change(p
, error
, devname
);
1105 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1106 &devname
)) != EAGAIN
) {
1107 process_port_change(p
, error
, devname
);
1111 if (time_msec() >= p
->next_in_band_update
) {
1112 update_in_band_remotes(p
);
1114 in_band_run(p
->in_band
);
1117 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1121 /* Fail-open maintenance. Do this after processing the ofconns since
1122 * fail-open checks the status of the controller rconn. */
1124 fail_open_run(p
->fail_open
);
1127 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1128 struct vconn
*vconn
;
1131 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1133 struct rconn
*rconn
;
1136 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1137 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1138 rconn_connect_unreliably(rconn
, vconn
, name
);
1141 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1142 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1143 ofservice
->burst_limit
);
1144 } else if (retval
!= EAGAIN
) {
1145 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1149 for (i
= 0; i
< p
->n_snoops
; i
++) {
1150 struct vconn
*vconn
;
1153 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1155 add_snooper(p
, vconn
);
1156 } else if (retval
!= EAGAIN
) {
1157 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1161 if (time_msec() >= p
->next_expiration
) {
1162 int delay
= ofproto_expire(p
);
1163 p
->next_expiration
= time_msec() + delay
;
1164 COVERAGE_INC(ofproto_expiration
);
1168 netflow_run(p
->netflow
);
1171 ofproto_sflow_run(p
->sflow
);
1178 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1180 /* Figure out what we need to revalidate now, if anything. */
1181 struct tag_set revalidate_set
= p
->revalidate_set
;
1182 if (p
->need_revalidate
) {
1183 revalidate_all
= true;
1186 /* Clear the revalidation flags. */
1187 tag_set_init(&p
->revalidate_set
);
1188 p
->need_revalidate
= false;
1190 /* Now revalidate if there's anything to do. */
1191 if (revalidate_all
|| !tag_set_is_empty(&revalidate_set
)) {
1192 struct facet
*facet
, *next
;
1194 HMAP_FOR_EACH_SAFE (facet
, next
, hmap_node
, &p
->facets
) {
1196 || tag_set_intersects(&revalidate_set
, facet
->tags
)) {
1197 facet_revalidate(p
, facet
);
1206 ofproto_wait(struct ofproto
*p
)
1208 struct ofservice
*ofservice
;
1209 struct ofconn
*ofconn
;
1212 dpif_recv_wait(p
->dpif
);
1213 dpif_port_poll_wait(p
->dpif
);
1214 netdev_monitor_poll_wait(p
->netdev_monitor
);
1215 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1216 ofconn_wait(ofconn
);
1219 poll_timer_wait_until(p
->next_in_band_update
);
1220 in_band_wait(p
->in_band
);
1223 fail_open_wait(p
->fail_open
);
1226 ofproto_sflow_wait(p
->sflow
);
1228 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1229 poll_immediate_wake();
1231 if (p
->need_revalidate
) {
1232 /* Shouldn't happen, but if it does just go around again. */
1233 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1234 poll_immediate_wake();
1235 } else if (p
->next_expiration
!= LLONG_MAX
) {
1236 poll_timer_wait_until(p
->next_expiration
);
1238 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1239 pvconn_wait(ofservice
->pvconn
);
1241 for (i
= 0; i
< p
->n_snoops
; i
++) {
1242 pvconn_wait(p
->snoops
[i
]);
1247 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1249 tag_set_add(&ofproto
->revalidate_set
, tag
);
1253 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1255 return &ofproto
->revalidate_set
;
1259 ofproto_is_alive(const struct ofproto
*p
)
1261 return !hmap_is_empty(&p
->controllers
);
1264 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1266 * This is almost the same as calling dpif_port_del() directly on the
1267 * datapath, but it also makes 'ofproto' close its open netdev for the port
1268 * (if any). This makes it possible to create a new netdev of a different
1269 * type under the same name, which otherwise the netdev library would refuse
1270 * to do because of the conflict. (The netdev would eventually get closed on
1271 * the next trip through ofproto_run(), but this interface is more direct.)
1273 * Returns 0 if successful, otherwise a positive errno. */
1275 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1277 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1278 const char *name
= ofport
? (char *) ofport
->opp
.name
: "<unknown>";
1281 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1283 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1284 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1285 } else if (ofport
) {
1286 /* 'name' is ofport->opp.name and update_port() is going to destroy
1287 * 'ofport'. Just in case update_port() refers to 'name' after it
1288 * destroys 'ofport', make a copy of it around the update_port()
1290 char *devname
= xstrdup(name
);
1291 update_port(ofproto
, devname
);
1297 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1298 * true if 'odp_port' exists and should be included, false otherwise. */
1300 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1302 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1303 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1307 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1308 const union ofp_action
*actions
, size_t n_actions
,
1309 const struct ofpbuf
*packet
)
1311 struct odp_actions odp_actions
;
1314 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1320 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1322 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
, packet
);
1326 /* Adds a flow to the OpenFlow flow table in 'p' that matches 'cls_rule' and
1327 * performs the 'n_actions' actions in 'actions'. The new flow will not
1330 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1331 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1332 * controllers; otherwise, it will be hidden.
1334 * The caller retains ownership of 'cls_rule' and 'actions'. */
1336 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1337 const union ofp_action
*actions
, size_t n_actions
)
1340 rule
= rule_create(cls_rule
, actions
, n_actions
, 0, 0, 0, false);
1341 rule_insert(p
, rule
, NULL
, 0);
1345 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1349 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1352 rule_remove(ofproto
, rule
);
1357 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1359 struct rule
*rule
= rule_from_cls_rule(rule_
);
1360 struct ofproto
*ofproto
= ofproto_
;
1362 rule_remove(ofproto
, rule
);
1366 ofproto_flush_flows(struct ofproto
*ofproto
)
1368 struct facet
*facet
, *next_facet
;
1370 COVERAGE_INC(ofproto_flush
);
1372 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
1373 /* Mark the facet as not installed so that facet_remove() doesn't
1374 * bother trying to uninstall it. There is no point in uninstalling it
1375 * individually since we are about to blow away all the facets with
1376 * dpif_flow_flush(). */
1377 facet
->installed
= false;
1378 facet_remove(ofproto
, facet
);
1380 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1381 dpif_flow_flush(ofproto
->dpif
);
1382 if (ofproto
->in_band
) {
1383 in_band_flushed(ofproto
->in_band
);
1385 if (ofproto
->fail_open
) {
1386 fail_open_flushed(ofproto
->fail_open
);
1391 reinit_ports(struct ofproto
*p
)
1393 struct svec devnames
;
1394 struct ofport
*ofport
;
1395 struct odp_port
*odp_ports
;
1399 COVERAGE_INC(ofproto_reinit_ports
);
1401 svec_init(&devnames
);
1402 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1403 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1405 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1406 for (i
= 0; i
< n_odp_ports
; i
++) {
1407 svec_add (&devnames
, odp_ports
[i
].devname
);
1411 svec_sort_unique(&devnames
);
1412 for (i
= 0; i
< devnames
.n
; i
++) {
1413 update_port(p
, devnames
.names
[i
]);
1415 svec_destroy(&devnames
);
1418 static struct ofport
*
1419 make_ofport(const struct odp_port
*odp_port
)
1421 struct netdev_options netdev_options
;
1422 enum netdev_flags flags
;
1423 struct ofport
*ofport
;
1424 struct netdev
*netdev
;
1427 memset(&netdev_options
, 0, sizeof netdev_options
);
1428 netdev_options
.name
= odp_port
->devname
;
1429 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1431 error
= netdev_open(&netdev_options
, &netdev
);
1433 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1434 "cannot be opened (%s)",
1435 odp_port
->devname
, odp_port
->port
,
1436 odp_port
->devname
, strerror(error
));
1440 ofport
= xmalloc(sizeof *ofport
);
1441 ofport
->netdev
= netdev
;
1442 ofport
->odp_port
= odp_port
->port
;
1443 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1444 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1445 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1446 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1447 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1449 netdev_get_flags(netdev
, &flags
);
1450 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1452 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1454 netdev_get_features(netdev
,
1455 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1456 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1461 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1463 if (get_port(p
, odp_port
->port
)) {
1464 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1467 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1468 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1477 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1479 const struct ofp_phy_port
*a
= &a_
->opp
;
1480 const struct ofp_phy_port
*b
= &b_
->opp
;
1482 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1483 return (a
->port_no
== b
->port_no
1484 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1485 && !strcmp((char *) a
->name
, (char *) b
->name
)
1486 && a
->state
== b
->state
1487 && a
->config
== b
->config
1488 && a
->curr
== b
->curr
1489 && a
->advertised
== b
->advertised
1490 && a
->supported
== b
->supported
1491 && a
->peer
== b
->peer
);
1495 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1498 /* XXX Should limit the number of queued port status change messages. */
1499 struct ofconn
*ofconn
;
1500 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1501 struct ofp_port_status
*ops
;
1504 if (!ofconn_receives_async_msgs(ofconn
)) {
1508 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1509 ops
->reason
= reason
;
1510 ops
->desc
= ofport
->opp
;
1511 hton_ofp_phy_port(&ops
->desc
);
1512 queue_tx(b
, ofconn
, NULL
);
1517 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1519 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1521 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1522 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1523 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1525 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1530 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1532 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1533 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1534 shash_delete(&p
->port_by_name
,
1535 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1537 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1542 ofport_free(struct ofport
*ofport
)
1545 netdev_close(ofport
->netdev
);
1550 static struct ofport
*
1551 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1553 struct ofport
*port
;
1555 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1556 hash_int(odp_port
, 0), &ofproto
->ports
) {
1557 if (port
->odp_port
== odp_port
) {
1565 update_port(struct ofproto
*p
, const char *devname
)
1567 struct odp_port odp_port
;
1568 struct ofport
*old_ofport
;
1569 struct ofport
*new_ofport
;
1572 COVERAGE_INC(ofproto_update_port
);
1574 /* Query the datapath for port information. */
1575 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1577 /* Find the old ofport. */
1578 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1581 /* There's no port named 'devname' but there might be a port with
1582 * the same port number. This could happen if a port is deleted
1583 * and then a new one added in its place very quickly, or if a port
1584 * is renamed. In the former case we want to send an OFPPR_DELETE
1585 * and an OFPPR_ADD, and in the latter case we want to send a
1586 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1587 * the old port's ifindex against the new port, or perhaps less
1588 * reliably but more portably by comparing the old port's MAC
1589 * against the new port's MAC. However, this code isn't that smart
1590 * and always sends an OFPPR_MODIFY (XXX). */
1591 old_ofport
= get_port(p
, odp_port
.port
);
1593 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1594 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1595 "%s", strerror(error
));
1599 /* Create a new ofport. */
1600 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1602 /* Eliminate a few pathological cases. */
1603 if (!old_ofport
&& !new_ofport
) {
1605 } else if (old_ofport
&& new_ofport
) {
1606 /* Most of the 'config' bits are OpenFlow soft state, but
1607 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1608 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1609 * leaves the other bits 0.) */
1610 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1612 if (ofport_equal(old_ofport
, new_ofport
)) {
1613 /* False alarm--no change. */
1614 ofport_free(new_ofport
);
1619 /* Now deal with the normal cases. */
1621 ofport_remove(p
, old_ofport
);
1624 ofport_install(p
, new_ofport
);
1626 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1627 (!old_ofport
? OFPPR_ADD
1628 : !new_ofport
? OFPPR_DELETE
1630 ofport_free(old_ofport
);
1634 init_ports(struct ofproto
*p
)
1636 struct odp_port
*ports
;
1641 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1646 for (i
= 0; i
< n_ports
; i
++) {
1647 const struct odp_port
*odp_port
= &ports
[i
];
1648 if (!ofport_conflicts(p
, odp_port
)) {
1649 struct ofport
*ofport
= make_ofport(odp_port
);
1651 ofport_install(p
, ofport
);
1659 static struct ofconn
*
1660 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1662 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1663 ofconn
->ofproto
= p
;
1664 list_push_back(&p
->all_conns
, &ofconn
->node
);
1665 ofconn
->rconn
= rconn
;
1666 ofconn
->type
= type
;
1667 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1668 ofconn
->role
= NX_ROLE_OTHER
;
1669 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1670 ofconn
->pktbuf
= NULL
;
1671 ofconn
->miss_send_len
= 0;
1672 ofconn
->reply_counter
= rconn_packet_counter_create ();
1677 ofconn_destroy(struct ofconn
*ofconn
)
1679 if (ofconn
->type
== OFCONN_PRIMARY
) {
1680 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1682 discovery_destroy(ofconn
->discovery
);
1684 list_remove(&ofconn
->node
);
1685 switch_status_unregister(ofconn
->ss
);
1686 rconn_destroy(ofconn
->rconn
);
1687 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1688 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1689 pktbuf_destroy(ofconn
->pktbuf
);
1694 ofconn_run(struct ofconn
*ofconn
)
1696 struct ofproto
*p
= ofconn
->ofproto
;
1700 if (ofconn
->discovery
) {
1701 char *controller_name
;
1702 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1703 discovery_question_connectivity(ofconn
->discovery
);
1705 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1706 if (controller_name
) {
1707 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1708 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1711 rconn_disconnect(ofconn
->rconn
);
1716 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1717 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1720 rconn_run(ofconn
->rconn
);
1722 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1723 /* Limit the number of iterations to prevent other tasks from
1725 for (iteration
= 0; iteration
< 50; iteration
++) {
1726 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1731 fail_open_maybe_recover(p
->fail_open
);
1733 handle_openflow(ofconn
, of_msg
);
1734 ofpbuf_delete(of_msg
);
1738 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1739 ofconn_destroy(ofconn
);
1744 ofconn_wait(struct ofconn
*ofconn
)
1748 if (ofconn
->discovery
) {
1749 discovery_wait(ofconn
->discovery
);
1751 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1752 pinsched_wait(ofconn
->schedulers
[i
]);
1754 rconn_run_wait(ofconn
->rconn
);
1755 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1756 rconn_recv_wait(ofconn
->rconn
);
1758 COVERAGE_INC(ofproto_ofconn_stuck
);
1762 /* Returns true if 'ofconn' should receive asynchronous messages. */
1764 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1766 if (ofconn
->type
== OFCONN_PRIMARY
) {
1767 /* Primary controllers always get asynchronous messages unless they
1768 * have configured themselves as "slaves". */
1769 return ofconn
->role
!= NX_ROLE_SLAVE
;
1771 /* Service connections don't get asynchronous messages unless they have
1772 * explicitly asked for them by setting a nonzero miss send length. */
1773 return ofconn
->miss_send_len
> 0;
1777 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1778 * and 'target', suitable for use in log messages for identifying the
1781 * The name is dynamically allocated. The caller should free it (with free())
1782 * when it is no longer needed. */
1784 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1786 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1790 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1794 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1795 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1799 *s
= pinsched_create(rate
, burst
,
1800 ofconn
->ofproto
->switch_status
);
1802 pinsched_set_limits(*s
, rate
, burst
);
1805 pinsched_destroy(*s
);
1812 ofservice_reconfigure(struct ofservice
*ofservice
,
1813 const struct ofproto_controller
*c
)
1815 ofservice
->probe_interval
= c
->probe_interval
;
1816 ofservice
->rate_limit
= c
->rate_limit
;
1817 ofservice
->burst_limit
= c
->burst_limit
;
1820 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1821 * positive errno value. */
1823 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1825 struct ofservice
*ofservice
;
1826 struct pvconn
*pvconn
;
1829 error
= pvconn_open(c
->target
, &pvconn
);
1834 ofservice
= xzalloc(sizeof *ofservice
);
1835 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1836 hash_string(c
->target
, 0));
1837 ofservice
->pvconn
= pvconn
;
1839 ofservice_reconfigure(ofservice
, c
);
1845 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1847 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1848 pvconn_close(ofservice
->pvconn
);
1852 /* Finds and returns the ofservice within 'ofproto' that has the given
1853 * 'target', or a null pointer if none exists. */
1854 static struct ofservice
*
1855 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1857 struct ofservice
*ofservice
;
1859 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1860 &ofproto
->services
) {
1861 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1868 /* Returns true if 'rule' should be hidden from the controller.
1870 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
1871 * (e.g. by in-band control) and are intentionally hidden from the
1874 rule_is_hidden(const struct rule
*rule
)
1876 return rule
->cr
.priority
> UINT16_MAX
;
1879 /* Creates and returns a new rule initialized as specified.
1881 * The caller is responsible for inserting the rule into the classifier (with
1882 * rule_insert()). */
1883 static struct rule
*
1884 rule_create(const struct cls_rule
*cls_rule
,
1885 const union ofp_action
*actions
, size_t n_actions
,
1886 uint16_t idle_timeout
, uint16_t hard_timeout
,
1887 ovs_be64 flow_cookie
, bool send_flow_removed
)
1889 struct rule
*rule
= xzalloc(sizeof *rule
);
1890 rule
->cr
= *cls_rule
;
1891 rule
->idle_timeout
= idle_timeout
;
1892 rule
->hard_timeout
= hard_timeout
;
1893 rule
->flow_cookie
= flow_cookie
;
1894 rule
->used
= rule
->created
= time_msec();
1895 rule
->send_flow_removed
= send_flow_removed
;
1896 list_init(&rule
->facets
);
1897 if (n_actions
> 0) {
1898 rule
->n_actions
= n_actions
;
1899 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1905 static struct rule
*
1906 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1908 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1912 rule_free(struct rule
*rule
)
1914 free(rule
->actions
);
1918 /* Destroys 'rule' and iterates through all of its facets and revalidates them,
1919 * destroying any that no longer has a rule (which is probably all of them).
1921 * The caller must have already removed 'rule' from the classifier. */
1923 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1925 struct facet
*facet
, *next_facet
;
1926 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
1927 facet_revalidate(ofproto
, facet
);
1932 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
1933 * that outputs to 'out_port' (output to OFPP_FLOOD and OFPP_ALL doesn't
1936 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
1938 const union ofp_action
*oa
;
1939 struct actions_iterator i
;
1941 if (out_port
== htons(OFPP_NONE
)) {
1944 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1945 oa
= actions_next(&i
)) {
1946 if (action_outputs_to_port(oa
, out_port
)) {
1953 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1954 * 'packet', which arrived on 'in_port'.
1956 * Takes ownership of 'packet'. */
1958 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
1959 const union odp_action
*actions
, size_t n_actions
,
1960 struct ofpbuf
*packet
)
1962 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
1963 /* As an optimization, avoid a round-trip from userspace to kernel to
1964 * userspace. This also avoids possibly filling up kernel packet
1965 * buffers along the way. */
1966 struct odp_msg
*msg
;
1968 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
1969 msg
->type
= _ODPL_ACTION_NR
;
1970 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
1971 msg
->port
= in_port
;
1973 msg
->arg
= actions
[0].controller
.arg
;
1975 send_packet_in(ofproto
, packet
);
1981 error
= dpif_execute(ofproto
->dpif
, actions
, n_actions
, packet
);
1982 ofpbuf_delete(packet
);
1987 /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s
1988 * statistics appropriately. 'packet' must have at least sizeof(struct
1989 * ofp_packet_in) bytes of headroom.
1991 * For correct results, 'packet' must actually be in 'facet''s flow; that is,
1992 * applying flow_extract() to 'packet' would yield the same flow as
1995 * 'facet' must have accurately composed ODP actions; that is, it must not be
1996 * in need of revalidation.
1998 * Takes ownership of 'packet'. */
2000 facet_execute(struct ofproto
*ofproto
, struct facet
*facet
,
2001 struct ofpbuf
*packet
)
2003 struct odp_flow_stats stats
;
2005 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2007 flow_extract_stats(&facet
->flow
, packet
, &stats
);
2008 if (execute_odp_actions(ofproto
, facet
->flow
.in_port
,
2009 facet
->actions
, facet
->n_actions
, packet
)) {
2010 facet_update_stats(ofproto
, facet
, &stats
);
2011 facet
->used
= time_msec();
2012 netflow_flow_update_time(ofproto
->netflow
,
2013 &facet
->nf_flow
, facet
->used
);
2017 /* Executes the actions indicated by 'rule' on 'packet' and credits 'rule''s
2018 * statistics (or the statistics for one of its facets) appropriately.
2019 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of headroom.
2021 * 'packet' doesn't necessarily have to match 'rule'. 'rule' will be credited
2022 * with statistics for 'packet' either way.
2024 * Takes ownership of 'packet'. */
2026 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
, uint16_t in_port
,
2027 struct ofpbuf
*packet
)
2029 struct facet
*facet
;
2030 struct odp_actions a
;
2034 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
2036 flow_extract(packet
, 0, in_port
, &flow
);
2038 /* First look for a related facet. If we find one, account it to that. */
2039 facet
= facet_lookup_valid(ofproto
, &flow
);
2040 if (facet
&& facet
->rule
== rule
) {
2041 facet_execute(ofproto
, facet
, packet
);
2045 /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then
2046 * create a new facet for it and use that. */
2047 if (rule_lookup(ofproto
, &flow
) == rule
) {
2048 facet
= facet_create(ofproto
, rule
, &flow
, packet
);
2049 facet_execute(ofproto
, facet
, packet
);
2050 facet_install(ofproto
, facet
, true);
2054 /* We can't account anything to a facet. If we were to try, then that
2055 * facet would have a non-matching rule, busting our invariants. */
2056 if (xlate_actions(rule
->actions
, rule
->n_actions
, &flow
, ofproto
,
2057 packet
, &a
, NULL
, 0, NULL
)) {
2058 ofpbuf_delete(packet
);
2061 size
= packet
->size
;
2062 if (execute_odp_actions(ofproto
, in_port
,
2063 a
.actions
, a
.n_actions
, packet
)) {
2064 rule
->used
= time_msec();
2065 rule
->packet_count
++;
2066 rule
->byte_count
+= size
;
2070 /* Inserts 'rule' into 'p''s flow table.
2072 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2073 * actions on it and credits the statistics for sending the packet to 'rule'.
2074 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2077 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
2080 struct rule
*displaced_rule
;
2082 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2083 if (displaced_rule
) {
2084 rule_destroy(p
, displaced_rule
);
2086 p
->need_revalidate
= true;
2089 rule_execute(p
, rule
, in_port
, packet
);
2093 /* Creates and returns a new facet within 'ofproto' owned by 'rule', given a
2094 * 'flow' and an example 'packet' within that flow.
2096 * The caller must already have determined that no facet with an identical
2097 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
2098 * 'ofproto''s classifier table. */
2099 static struct facet
*
2100 facet_create(struct ofproto
*ofproto
, struct rule
*rule
,
2101 const struct flow
*flow
, const struct ofpbuf
*packet
)
2103 struct facet
*facet
;
2105 facet
= xzalloc(sizeof *facet
);
2106 facet
->used
= time_msec();
2107 hmap_insert(&ofproto
->facets
, &facet
->hmap_node
, flow_hash(flow
, 0));
2108 list_push_back(&rule
->facets
, &facet
->list_node
);
2110 facet
->flow
= *flow
;
2111 netflow_flow_init(&facet
->nf_flow
);
2112 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, facet
->used
);
2114 facet_make_actions(ofproto
, facet
, packet
);
2120 facet_free(struct facet
*facet
)
2122 free(facet
->actions
);
2126 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2128 * - Removes 'rule' from the classifier.
2130 * - If 'rule' has facets, revalidates them (and possibly uninstalls and
2131 * destroys them), via rule_destroy().
2134 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2136 COVERAGE_INC(ofproto_del_rule
);
2137 ofproto
->need_revalidate
= true;
2138 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2139 rule_destroy(ofproto
, rule
);
2142 /* Remove 'facet' from 'ofproto' and free up the associated memory:
2144 * - If 'facet' was installed in the datapath, uninstalls it and updates its
2145 * rule's statistics, via facet_uninstall().
2147 * - Removes 'facet' from its rule and from ofproto->facets.
2150 facet_remove(struct ofproto
*ofproto
, struct facet
*facet
)
2152 facet_uninstall(ofproto
, facet
);
2153 facet_flush_stats(ofproto
, facet
);
2154 hmap_remove(&ofproto
->facets
, &facet
->hmap_node
);
2155 list_remove(&facet
->list_node
);
2159 /* Composes the ODP actions for 'facet' based on its rule's actions. */
2161 facet_make_actions(struct ofproto
*p
, struct facet
*facet
,
2162 const struct ofpbuf
*packet
)
2164 const struct rule
*rule
= facet
->rule
;
2165 struct odp_actions a
;
2168 xlate_actions(rule
->actions
, rule
->n_actions
, &facet
->flow
, p
,
2169 packet
, &a
, &facet
->tags
, &facet
->may_install
,
2170 &facet
->nf_flow
.output_iface
);
2172 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2173 if (facet
->n_actions
!= a
.n_actions
2174 || memcmp(facet
->actions
, a
.actions
, actions_len
)) {
2175 free(facet
->actions
);
2176 facet
->n_actions
= a
.n_actions
;
2177 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2182 facet_put__(struct ofproto
*ofproto
, struct facet
*facet
, int flags
,
2183 struct odp_flow_put
*put
)
2185 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2186 odp_flow_key_from_flow(&put
->flow
.key
, &facet
->flow
);
2187 put
->flow
.actions
= facet
->actions
;
2188 put
->flow
.n_actions
= facet
->n_actions
;
2189 put
->flow
.flags
= 0;
2191 return dpif_flow_put(ofproto
->dpif
, put
);
2194 /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If
2195 * 'zero_stats' is true, clears any existing statistics from the datapath for
2198 facet_install(struct ofproto
*p
, struct facet
*facet
, bool zero_stats
)
2200 if (facet
->may_install
) {
2201 struct odp_flow_put put
;
2204 flags
= ODPPF_CREATE
| ODPPF_MODIFY
;
2206 flags
|= ODPPF_ZERO_STATS
;
2208 if (!facet_put__(p
, facet
, flags
, &put
)) {
2209 facet
->installed
= true;
2214 /* Ensures that the bytes in 'facet', plus 'extra_bytes', have been passed up
2215 * to the accounting hook function in the ofhooks structure. */
2217 facet_account(struct ofproto
*ofproto
,
2218 struct facet
*facet
, uint64_t extra_bytes
)
2220 uint64_t total_bytes
= facet
->byte_count
+ extra_bytes
;
2222 if (ofproto
->ofhooks
->account_flow_cb
2223 && total_bytes
> facet
->accounted_bytes
)
2225 ofproto
->ofhooks
->account_flow_cb(
2226 &facet
->flow
, facet
->tags
, facet
->actions
, facet
->n_actions
,
2227 total_bytes
- facet
->accounted_bytes
, ofproto
->aux
);
2228 facet
->accounted_bytes
= total_bytes
;
2232 /* If 'rule' is installed in the datapath, uninstalls it. */
2234 facet_uninstall(struct ofproto
*p
, struct facet
*facet
)
2236 if (facet
->installed
) {
2237 struct odp_flow odp_flow
;
2239 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
2240 odp_flow
.actions
= NULL
;
2241 odp_flow
.n_actions
= 0;
2243 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2244 facet_update_stats(p
, facet
, &odp_flow
.stats
);
2246 facet
->installed
= false;
2250 /* Returns true if the only action for 'facet' is to send to the controller.
2251 * (We don't report NetFlow expiration messages for such facets because they
2252 * are just part of the control logic for the network, not real traffic). */
2254 facet_is_controller_flow(struct facet
*facet
)
2257 && facet
->rule
->n_actions
== 1
2258 && action_outputs_to_port(&facet
->rule
->actions
[0],
2259 htons(OFPP_CONTROLLER
)));
2262 /* Folds all of 'facet''s statistics into its rule. Also updates the
2263 * accounting ofhook and emits a NetFlow expiration if appropriate. */
2265 facet_flush_stats(struct ofproto
*ofproto
, struct facet
*facet
)
2267 facet_account(ofproto
, facet
, 0);
2269 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
)) {
2270 struct ofexpired expired
;
2271 expired
.flow
= facet
->flow
;
2272 expired
.packet_count
= facet
->packet_count
;
2273 expired
.byte_count
= facet
->byte_count
;
2274 expired
.used
= facet
->used
;
2275 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
2278 facet
->rule
->packet_count
+= facet
->packet_count
;
2279 facet
->rule
->byte_count
+= facet
->byte_count
;
2281 /* Reset counters to prevent double counting if 'facet' ever gets
2283 facet
->packet_count
= 0;
2284 facet
->byte_count
= 0;
2285 facet
->accounted_bytes
= 0;
2287 netflow_flow_clear(&facet
->nf_flow
);
2290 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2291 * Returns it if found, otherwise a null pointer.
2293 * The returned facet might need revalidation; use facet_lookup_valid()
2294 * instead if that is important. */
2295 static struct facet
*
2296 facet_find(struct ofproto
*ofproto
, const struct flow
*flow
)
2298 struct facet
*facet
;
2300 HMAP_FOR_EACH_WITH_HASH (facet
, hmap_node
, flow_hash(flow
, 0),
2302 if (flow_equal(flow
, &facet
->flow
)) {
2310 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
2311 * Returns it if found, otherwise a null pointer.
2313 * The returned facet is guaranteed to be valid. */
2314 static struct facet
*
2315 facet_lookup_valid(struct ofproto
*ofproto
, const struct flow
*flow
)
2317 struct facet
*facet
= facet_find(ofproto
, flow
);
2319 /* The facet we found might not be valid, since we could be in need of
2320 * revalidation. If it is not valid, don't return it. */
2322 && ofproto
->need_revalidate
2323 && !facet_revalidate(ofproto
, facet
)) {
2324 COVERAGE_INC(ofproto_invalidated
);
2331 /* Re-searches 'ofproto''s classifier for a rule matching 'facet':
2333 * - If the rule found is different from 'facet''s current rule, moves
2334 * 'facet' to the new rule and recompiles its actions.
2336 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
2337 * where it is and recompiles its actions anyway.
2339 * - If there is none, destroys 'facet'.
2341 * Returns true if 'facet' still exists, false if it has been destroyed. */
2343 facet_revalidate(struct ofproto
*ofproto
, struct facet
*facet
)
2345 struct rule
*new_rule
;
2346 struct odp_actions a
;
2348 uint16_t new_nf_output_iface
;
2349 bool actions_changed
;
2351 COVERAGE_INC(facet_revalidate
);
2353 /* Determine the new rule. */
2354 new_rule
= rule_lookup(ofproto
, &facet
->flow
);
2356 /* No new rule, so delete the facet. */
2357 facet_remove(ofproto
, facet
);
2361 /* Calculate new ODP actions.
2363 * We are very cautious about actually modifying 'facet' state at this
2364 * point, because we might need to, e.g., emit a NetFlow expiration and, if
2365 * so, we need to have the old state around to properly compose it. */
2366 xlate_actions(new_rule
->actions
, new_rule
->n_actions
, &facet
->flow
,
2367 ofproto
, NULL
, &a
, &facet
->tags
, &facet
->may_install
,
2368 &new_nf_output_iface
);
2369 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2370 actions_changed
= (facet
->n_actions
!= a
.n_actions
2371 || memcmp(facet
->actions
, a
.actions
, actions_len
));
2373 /* If the ODP actions changed or the installability changed, then we need
2374 * to talk to the datapath. */
2375 if (actions_changed
|| facet
->may_install
!= facet
->installed
) {
2376 if (facet
->may_install
) {
2377 struct odp_flow_put put
;
2379 memset(&put
.flow
.stats
, 0, sizeof put
.flow
.stats
);
2380 odp_flow_key_from_flow(&put
.flow
.key
, &facet
->flow
);
2381 put
.flow
.actions
= a
.actions
;
2382 put
.flow
.n_actions
= a
.n_actions
;
2384 put
.flags
= ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
;
2385 dpif_flow_put(ofproto
->dpif
, &put
);
2387 facet_update_stats(ofproto
, facet
, &put
.flow
.stats
);
2389 facet_uninstall(ofproto
, facet
);
2392 /* The datapath flow is gone or has zeroed stats, so push stats out of
2393 * 'facet' into 'rule'. */
2394 facet_flush_stats(ofproto
, facet
);
2397 /* Update 'facet' now that we've taken care of all the old state. */
2398 facet
->nf_flow
.output_iface
= new_nf_output_iface
;
2399 if (actions_changed
) {
2400 free(facet
->actions
);
2401 facet
->n_actions
= a
.n_actions
;
2402 facet
->actions
= xmemdup(a
.actions
, actions_len
);
2404 if (facet
->rule
!= new_rule
) {
2405 COVERAGE_INC(facet_changed_rule
);
2406 list_remove(&facet
->list_node
);
2407 list_push_back(&new_rule
->facets
, &facet
->list_node
);
2408 facet
->rule
= new_rule
;
2409 facet
->used
= new_rule
->created
;
2416 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2417 struct rconn_packet_counter
*counter
)
2419 update_openflow_length(msg
);
2420 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2426 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2429 struct ofpbuf
*buf
= make_ofp_error_msg(error
, oh
);
2431 COVERAGE_INC(ofproto_error
);
2432 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2437 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2439 opp
->port_no
= htons(opp
->port_no
);
2440 opp
->config
= htonl(opp
->config
);
2441 opp
->state
= htonl(opp
->state
);
2442 opp
->curr
= htonl(opp
->curr
);
2443 opp
->advertised
= htonl(opp
->advertised
);
2444 opp
->supported
= htonl(opp
->supported
);
2445 opp
->peer
= htonl(opp
->peer
);
2449 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2451 struct ofp_header
*rq
= oh
;
2452 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2457 handle_features_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2459 struct ofp_switch_features
*osf
;
2461 struct ofport
*port
;
2463 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2464 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2465 osf
->n_buffers
= htonl(pktbuf_capacity());
2467 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2468 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2469 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2470 (1u << OFPAT_SET_VLAN_VID
) |
2471 (1u << OFPAT_SET_VLAN_PCP
) |
2472 (1u << OFPAT_STRIP_VLAN
) |
2473 (1u << OFPAT_SET_DL_SRC
) |
2474 (1u << OFPAT_SET_DL_DST
) |
2475 (1u << OFPAT_SET_NW_SRC
) |
2476 (1u << OFPAT_SET_NW_DST
) |
2477 (1u << OFPAT_SET_NW_TOS
) |
2478 (1u << OFPAT_SET_TP_SRC
) |
2479 (1u << OFPAT_SET_TP_DST
) |
2480 (1u << OFPAT_ENQUEUE
));
2482 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2483 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2486 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2491 handle_get_config_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2494 struct ofp_switch_config
*osc
;
2498 /* Figure out flags. */
2499 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2500 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2503 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2504 osc
->flags
= htons(flags
);
2505 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2506 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2512 handle_set_config(struct ofconn
*ofconn
, struct ofp_switch_config
*osc
)
2517 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2521 flags
= ntohs(osc
->flags
);
2523 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2524 switch (flags
& OFPC_FRAG_MASK
) {
2525 case OFPC_FRAG_NORMAL
:
2526 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2528 case OFPC_FRAG_DROP
:
2529 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2532 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2538 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2544 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2546 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2547 a
->controller
.arg
= max_len
;
2550 struct action_xlate_ctx
{
2552 struct flow flow
; /* Flow to which these actions correspond. */
2553 int recurse
; /* Recursion level, via xlate_table_action. */
2554 struct ofproto
*ofproto
;
2555 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2556 * null pointer if we are revalidating
2557 * without a packet to refer to. */
2560 struct odp_actions
*out
; /* Datapath actions. */
2561 tag_type tags
; /* Tags associated with OFPP_NORMAL actions. */
2562 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2563 * be reassessed for every packet. */
2564 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2567 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2568 * flow translation. */
2569 #define MAX_RESUBMIT_RECURSION 8
2571 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2572 struct action_xlate_ctx
*ctx
);
2575 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2577 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2580 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2581 /* Forwarding disabled on port. */
2586 * We don't have an ofport record for this port, but it doesn't hurt to
2587 * allow forwarding to it anyhow. Maybe such a port will appear later
2588 * and we're pre-populating the flow table.
2592 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2593 ctx
->nf_output_iface
= port
;
2596 static struct rule
*
2597 rule_lookup(struct ofproto
*ofproto
, const struct flow
*flow
)
2599 return rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
,
2604 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2606 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2607 uint16_t old_in_port
;
2610 /* Look up a flow with 'in_port' as the input port. Then restore the
2611 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2612 * have surprising behavior). */
2613 old_in_port
= ctx
->flow
.in_port
;
2614 ctx
->flow
.in_port
= in_port
;
2615 rule
= rule_lookup(ctx
->ofproto
, &ctx
->flow
);
2616 ctx
->flow
.in_port
= old_in_port
;
2620 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2624 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2626 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2627 MAX_RESUBMIT_RECURSION
);
2632 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2633 uint16_t *nf_output_iface
, struct odp_actions
*actions
)
2635 struct ofport
*ofport
;
2637 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2638 uint16_t odp_port
= ofport
->odp_port
;
2639 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2640 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= odp_port
;
2643 *nf_output_iface
= NF_OUT_FLOOD
;
2647 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2648 uint16_t port
, uint16_t max_len
)
2651 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2653 ctx
->nf_output_iface
= NF_OUT_DROP
;
2657 add_output_action(ctx
, ctx
->flow
.in_port
);
2660 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2663 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2664 ctx
->out
, &ctx
->tags
,
2665 &ctx
->nf_output_iface
,
2666 ctx
->ofproto
->aux
)) {
2667 COVERAGE_INC(ofproto_uninstallable
);
2668 ctx
->may_set_up_flow
= false;
2672 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2673 &ctx
->nf_output_iface
, ctx
->out
);
2676 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2677 &ctx
->nf_output_iface
, ctx
->out
);
2679 case OFPP_CONTROLLER
:
2680 add_controller_action(ctx
->out
, max_len
);
2683 add_output_action(ctx
, ODPP_LOCAL
);
2686 odp_port
= ofp_port_to_odp_port(port
);
2687 if (odp_port
!= ctx
->flow
.in_port
) {
2688 add_output_action(ctx
, odp_port
);
2693 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2694 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2695 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2696 ctx
->nf_output_iface
= prev_nf_output_iface
;
2697 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2698 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2699 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2704 xlate_output_action(struct action_xlate_ctx
*ctx
,
2705 const struct ofp_action_output
*oao
)
2707 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2710 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2711 * optimization, because we're going to add another action that sets the
2712 * priority immediately after, or because there are no actions following the
2715 remove_pop_action(struct action_xlate_ctx
*ctx
)
2717 size_t n
= ctx
->out
->n_actions
;
2718 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2719 ctx
->out
->n_actions
--;
2724 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2725 const struct ofp_action_enqueue
*oae
)
2727 uint16_t ofp_port
, odp_port
;
2731 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2734 /* Fall back to ordinary output action. */
2735 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2739 /* Figure out ODP output port. */
2740 ofp_port
= ntohs(oae
->port
);
2741 if (ofp_port
!= OFPP_IN_PORT
) {
2742 odp_port
= ofp_port_to_odp_port(ofp_port
);
2744 odp_port
= ctx
->flow
.in_port
;
2747 /* Add ODP actions. */
2748 remove_pop_action(ctx
);
2749 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2751 add_output_action(ctx
, odp_port
);
2752 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2754 /* Update NetFlow output port. */
2755 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2756 ctx
->nf_output_iface
= odp_port
;
2757 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2758 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2763 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2764 const struct nx_action_set_queue
*nasq
)
2769 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2772 /* Couldn't translate queue to a priority, so ignore. A warning
2773 * has already been logged. */
2777 remove_pop_action(ctx
);
2778 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2783 xlate_set_dl_tci(struct action_xlate_ctx
*ctx
)
2785 ovs_be16 dl_vlan
= ctx
->flow
.dl_vlan
;
2786 uint8_t dl_vlan_pcp
= ctx
->flow
.dl_vlan_pcp
;
2788 if (dl_vlan
== htons(OFP_VLAN_NONE
)) {
2789 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2791 union odp_action
*oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_TCI
);
2792 oa
->dl_tci
.tci
= htons(ntohs(dl_vlan
& htons(VLAN_VID_MASK
))
2793 | (dl_vlan_pcp
<< VLAN_PCP_SHIFT
)
2799 xlate_reg_move_action(struct action_xlate_ctx
*ctx
,
2800 const struct nx_action_reg_move
*narm
)
2802 ovs_be16 old_vlan
= ctx
->flow
.dl_vlan
;
2803 uint8_t old_pcp
= ctx
->flow
.dl_vlan_pcp
;
2805 nxm_execute_reg_move(narm
, &ctx
->flow
);
2807 if (ctx
->flow
.dl_vlan
!= old_vlan
|| ctx
->flow
.dl_vlan_pcp
!= old_pcp
) {
2808 xlate_set_dl_tci(ctx
);
2813 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2814 const struct nx_action_header
*nah
)
2816 const struct nx_action_resubmit
*nar
;
2817 const struct nx_action_set_tunnel
*nast
;
2818 const struct nx_action_set_queue
*nasq
;
2819 union odp_action
*oa
;
2820 int subtype
= ntohs(nah
->subtype
);
2822 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2824 case NXAST_RESUBMIT
:
2825 nar
= (const struct nx_action_resubmit
*) nah
;
2826 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2829 case NXAST_SET_TUNNEL
:
2830 nast
= (const struct nx_action_set_tunnel
*) nah
;
2831 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2832 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2835 case NXAST_DROP_SPOOFED_ARP
:
2836 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2837 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2841 case NXAST_SET_QUEUE
:
2842 nasq
= (const struct nx_action_set_queue
*) nah
;
2843 xlate_set_queue_action(ctx
, nasq
);
2846 case NXAST_POP_QUEUE
:
2847 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2850 case NXAST_REG_MOVE
:
2851 xlate_reg_move_action(ctx
, (const struct nx_action_reg_move
*) nah
);
2854 case NXAST_REG_LOAD
:
2855 nxm_execute_reg_load((const struct nx_action_reg_load
*) nah
,
2859 /* If you add a new action here that modifies flow data, don't forget to
2860 * update the flow key in ctx->flow at the same time. */
2863 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2869 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2870 struct action_xlate_ctx
*ctx
)
2872 struct actions_iterator iter
;
2873 const union ofp_action
*ia
;
2874 const struct ofport
*port
;
2876 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2877 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2878 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2879 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2880 /* Drop this flow. */
2884 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2885 uint16_t type
= ntohs(ia
->type
);
2886 union odp_action
*oa
;
2890 xlate_output_action(ctx
, &ia
->output
);
2893 case OFPAT_SET_VLAN_VID
:
2894 ctx
->flow
.dl_vlan
= ia
->vlan_vid
.vlan_vid
;
2895 xlate_set_dl_tci(ctx
);
2898 case OFPAT_SET_VLAN_PCP
:
2899 ctx
->flow
.dl_vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2900 xlate_set_dl_tci(ctx
);
2903 case OFPAT_STRIP_VLAN
:
2904 ctx
->flow
.dl_vlan
= htons(OFP_VLAN_NONE
);
2905 ctx
->flow
.dl_vlan_pcp
= 0;
2906 xlate_set_dl_tci(ctx
);
2909 case OFPAT_SET_DL_SRC
:
2910 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2911 memcpy(oa
->dl_addr
.dl_addr
,
2912 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2913 memcpy(ctx
->flow
.dl_src
,
2914 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2917 case OFPAT_SET_DL_DST
:
2918 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2919 memcpy(oa
->dl_addr
.dl_addr
,
2920 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2921 memcpy(ctx
->flow
.dl_dst
,
2922 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2925 case OFPAT_SET_NW_SRC
:
2926 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2927 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2930 case OFPAT_SET_NW_DST
:
2931 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2932 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2935 case OFPAT_SET_NW_TOS
:
2936 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2937 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2940 case OFPAT_SET_TP_SRC
:
2941 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2942 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2945 case OFPAT_SET_TP_DST
:
2946 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2947 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2951 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2955 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2959 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2966 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2967 const struct flow
*flow
, struct ofproto
*ofproto
,
2968 const struct ofpbuf
*packet
,
2969 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2970 uint16_t *nf_output_iface
)
2972 struct action_xlate_ctx ctx
;
2974 COVERAGE_INC(ofproto_ofp2odp
);
2975 odp_actions_init(out
);
2978 ctx
.ofproto
= ofproto
;
2979 ctx
.packet
= packet
;
2982 ctx
.may_set_up_flow
= true;
2983 ctx
.nf_output_iface
= NF_OUT_DROP
;
2984 do_xlate_actions(in
, n_in
, &ctx
);
2985 remove_pop_action(&ctx
);
2987 /* Check with in-band control to see if we're allowed to set up this
2989 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2990 ctx
.may_set_up_flow
= false;
2996 if (may_set_up_flow
) {
2997 *may_set_up_flow
= ctx
.may_set_up_flow
;
2999 if (nf_output_iface
) {
3000 *nf_output_iface
= ctx
.nf_output_iface
;
3002 if (odp_actions_overflow(out
)) {
3003 COVERAGE_INC(odp_overflow
);
3004 odp_actions_init(out
);
3005 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
3010 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3011 * error message code (composed with ofp_mkerr()) for the caller to propagate
3012 * upward. Otherwise, returns 0.
3014 * The log message mentions 'msg_type'. */
3016 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
3018 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
3019 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
3020 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
3023 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3030 handle_packet_out(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3032 struct ofproto
*p
= ofconn
->ofproto
;
3033 struct ofp_packet_out
*opo
;
3034 struct ofpbuf payload
, *buffer
;
3035 union ofp_action
*ofp_actions
;
3036 struct odp_actions odp_actions
;
3037 struct ofpbuf request
;
3039 size_t n_ofp_actions
;
3043 COVERAGE_INC(ofproto_packet_out
);
3045 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
3050 /* Get ofp_packet_out. */
3052 request
.size
= ntohs(oh
->length
);
3053 opo
= ofpbuf_try_pull(&request
, offsetof(struct ofp_packet_out
, actions
));
3055 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3059 error
= ofputil_pull_actions(&request
, ntohs(opo
->actions_len
),
3060 &ofp_actions
, &n_ofp_actions
);
3066 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
3067 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
3069 if (error
|| !buffer
) {
3078 /* Extract flow, check actions. */
3079 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)),
3081 error
= validate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
->max_ports
);
3087 error
= xlate_actions(ofp_actions
, n_ofp_actions
, &flow
, p
, &payload
,
3088 &odp_actions
, NULL
, NULL
, NULL
);
3090 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
,
3095 ofpbuf_delete(buffer
);
3100 update_port_config(struct ofproto
*p
, struct ofport
*port
,
3101 uint32_t config
, uint32_t mask
)
3103 mask
&= config
^ port
->opp
.config
;
3104 if (mask
& OFPPC_PORT_DOWN
) {
3105 if (config
& OFPPC_PORT_DOWN
) {
3106 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
3108 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
3111 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
3112 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
3113 if (mask
& REVALIDATE_BITS
) {
3114 COVERAGE_INC(ofproto_costly_flags
);
3115 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
3116 p
->need_revalidate
= true;
3118 #undef REVALIDATE_BITS
3119 if (mask
& OFPPC_NO_PACKET_IN
) {
3120 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
3125 handle_port_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3127 struct ofproto
*p
= ofconn
->ofproto
;
3128 const struct ofp_port_mod
*opm
;
3129 struct ofport
*port
;
3132 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
3136 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
3140 opm
= (struct ofp_port_mod
*) oh
;
3142 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
3144 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
3145 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
3146 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
3148 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
3149 if (opm
->advertise
) {
3150 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3156 static struct ofpbuf
*
3157 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
3159 struct ofp_stats_reply
*osr
;
3162 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3163 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3165 osr
->flags
= htons(0);
3169 static struct ofpbuf
*
3170 start_ofp_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
3172 return make_ofp_stats_reply(request
->header
.xid
, request
->type
, body_len
);
3176 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3177 struct ofpbuf
**msgp
)
3179 struct ofpbuf
*msg
= *msgp
;
3180 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3181 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3182 struct ofp_stats_reply
*reply
= msg
->data
;
3183 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3184 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3185 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3187 return ofpbuf_put_uninit(*msgp
, nbytes
);
3190 static struct ofpbuf
*
3191 make_nxstats_reply(ovs_be32 xid
, ovs_be32 subtype
, size_t body_len
)
3193 struct nicira_stats_msg
*nsm
;
3196 msg
= ofpbuf_new(MIN(sizeof *nsm
+ body_len
, UINT16_MAX
));
3197 nsm
= put_openflow_xid(sizeof *nsm
, OFPT_STATS_REPLY
, xid
, msg
);
3198 nsm
->type
= htons(OFPST_VENDOR
);
3199 nsm
->flags
= htons(0);
3200 nsm
->vendor
= htonl(NX_VENDOR_ID
);
3201 nsm
->subtype
= htonl(subtype
);
3205 static struct ofpbuf
*
3206 start_nxstats_reply(const struct nicira_stats_msg
*request
, size_t body_len
)
3208 return make_nxstats_reply(request
->header
.xid
, request
->subtype
, body_len
);
3212 append_nxstats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3213 struct ofpbuf
**msgp
)
3215 struct ofpbuf
*msg
= *msgp
;
3216 assert(nbytes
<= UINT16_MAX
- sizeof(struct nicira_stats_msg
));
3217 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3218 struct nicira_stats_msg
*reply
= msg
->data
;
3219 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3220 *msgp
= make_nxstats_reply(reply
->header
.xid
, reply
->subtype
, nbytes
);
3221 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3223 ofpbuf_prealloc_tailroom(*msgp
, nbytes
);
3227 handle_desc_stats_request(struct ofconn
*ofconn
,
3228 struct ofp_stats_request
*request
)
3230 struct ofproto
*p
= ofconn
->ofproto
;
3231 struct ofp_desc_stats
*ods
;
3234 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3235 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3236 memset(ods
, 0, sizeof *ods
);
3237 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3238 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3239 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3240 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3241 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3242 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3248 handle_table_stats_request(struct ofconn
*ofconn
,
3249 struct ofp_stats_request
*request
)
3251 struct ofproto
*p
= ofconn
->ofproto
;
3252 struct ofp_table_stats
*ots
;
3255 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3257 /* Classifier table. */
3258 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3259 memset(ots
, 0, sizeof *ots
);
3260 strcpy(ots
->name
, "classifier");
3261 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3262 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3263 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3264 ots
->active_count
= htonl(classifier_count(&p
->cls
));
3265 ots
->lookup_count
= htonll(0); /* XXX */
3266 ots
->matched_count
= htonll(0); /* XXX */
3268 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3273 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3274 struct ofpbuf
**msgp
)
3276 struct netdev_stats stats
;
3277 struct ofp_port_stats
*ops
;
3279 /* Intentionally ignore return value, since errors will set
3280 * 'stats' to all-1s, which is correct for OpenFlow, and
3281 * netdev_get_stats() will log errors. */
3282 netdev_get_stats(port
->netdev
, &stats
);
3284 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3285 ops
->port_no
= htons(port
->opp
.port_no
);
3286 memset(ops
->pad
, 0, sizeof ops
->pad
);
3287 ops
->rx_packets
= htonll(stats
.rx_packets
);
3288 ops
->tx_packets
= htonll(stats
.tx_packets
);
3289 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3290 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3291 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3292 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3293 ops
->rx_errors
= htonll(stats
.rx_errors
);
3294 ops
->tx_errors
= htonll(stats
.tx_errors
);
3295 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3296 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3297 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3298 ops
->collisions
= htonll(stats
.collisions
);
3302 handle_port_stats_request(struct ofconn
*ofconn
, struct ofp_stats_request
*osr
,
3305 struct ofproto
*p
= ofconn
->ofproto
;
3306 struct ofp_port_stats_request
*psr
;
3307 struct ofp_port_stats
*ops
;
3309 struct ofport
*port
;
3311 if (arg_size
!= sizeof *psr
) {
3312 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3314 psr
= (struct ofp_port_stats_request
*) osr
->body
;
3316 msg
= start_ofp_stats_reply(osr
, sizeof *ops
* 16);
3317 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3318 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3320 append_port_stat(port
, ofconn
, &msg
);
3323 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3324 append_port_stat(port
, ofconn
, &msg
);
3328 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3332 struct flow_stats_cbdata
{
3333 struct ofconn
*ofconn
;
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 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3398 struct rule
*rule
= rule_from_cls_rule(rule_
);
3399 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3400 struct ofp_flow_stats
*ofs
;
3401 uint64_t packet_count
, byte_count
;
3402 size_t act_len
, len
;
3404 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3408 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3409 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3411 query_stats(cbdata
->ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3413 ofs
= append_ofp_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
3414 ofs
->length
= htons(len
);
3417 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3418 cbdata
->ofconn
->flow_format
, &ofs
->match
);
3419 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3420 ofs
->cookie
= rule
->flow_cookie
;
3421 ofs
->priority
= htons(rule
->cr
.priority
);
3422 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3423 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3424 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3425 ofs
->packet_count
= htonll(packet_count
);
3426 ofs
->byte_count
= htonll(byte_count
);
3427 if (rule
->n_actions
> 0) {
3428 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3433 table_id_to_include(uint8_t table_id
)
3435 return table_id
== 0 || table_id
== 0xff ? CLS_INC_ALL
: 0;
3439 handle_flow_stats_request(struct ofconn
*ofconn
,
3440 const struct ofp_stats_request
*osr
, size_t arg_size
)
3442 struct ofp_flow_stats_request
*fsr
;
3443 struct flow_stats_cbdata cbdata
;
3444 struct cls_rule target
;
3446 if (arg_size
!= sizeof *fsr
) {
3447 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3449 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3451 COVERAGE_INC(ofproto_flows_req
);
3452 cbdata
.ofconn
= ofconn
;
3453 cbdata
.out_port
= fsr
->out_port
;
3454 cbdata
.msg
= start_ofp_stats_reply(osr
, 1024);
3455 cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3456 classifier_for_each_match(&ofconn
->ofproto
->cls
, &target
,
3457 table_id_to_include(fsr
->table_id
),
3458 flow_stats_cb
, &cbdata
);
3459 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3464 nx_flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3466 struct rule
*rule
= rule_from_cls_rule(rule_
);
3467 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3468 struct nx_flow_stats
*nfs
;
3469 uint64_t packet_count
, byte_count
;
3470 size_t act_len
, start_len
;
3472 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3476 query_stats(cbdata
->ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3478 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3480 start_len
= cbdata
->msg
->size
;
3481 append_nxstats_reply(sizeof *nfs
+ NXM_MAX_LEN
+ act_len
,
3482 cbdata
->ofconn
, &cbdata
->msg
);
3483 nfs
= ofpbuf_put_uninit(cbdata
->msg
, sizeof *nfs
);
3486 calc_flow_duration(rule
->created
, &nfs
->duration_sec
, &nfs
->duration_nsec
);
3487 nfs
->cookie
= rule
->flow_cookie
;
3488 nfs
->priority
= htons(rule
->cr
.priority
);
3489 nfs
->idle_timeout
= htons(rule
->idle_timeout
);
3490 nfs
->hard_timeout
= htons(rule
->hard_timeout
);
3491 nfs
->match_len
= htons(nx_put_match(cbdata
->msg
, &rule
->cr
));
3492 memset(nfs
->pad2
, 0, sizeof nfs
->pad2
);
3493 nfs
->packet_count
= htonll(packet_count
);
3494 nfs
->byte_count
= htonll(byte_count
);
3495 if (rule
->n_actions
> 0) {
3496 ofpbuf_put(cbdata
->msg
, rule
->actions
, act_len
);
3498 nfs
->length
= htons(cbdata
->msg
->size
- start_len
);
3502 handle_nxst_flow(struct ofconn
*ofconn
, struct ofpbuf
*b
)
3504 struct nx_flow_stats_request
*nfsr
;
3505 struct flow_stats_cbdata cbdata
;
3506 struct cls_rule target
;
3509 /* Dissect the message. */
3510 nfsr
= ofpbuf_try_pull(b
, sizeof *nfsr
);
3512 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3514 error
= nx_pull_match(b
, ntohs(nfsr
->match_len
), 0, &target
);
3519 COVERAGE_INC(ofproto_flows_req
);
3520 cbdata
.ofconn
= ofconn
;
3521 cbdata
.out_port
= nfsr
->out_port
;
3522 cbdata
.msg
= start_nxstats_reply(&nfsr
->nsm
, 1024);
3523 classifier_for_each_match(&ofconn
->ofproto
->cls
, &target
,
3524 table_id_to_include(nfsr
->table_id
),
3525 nx_flow_stats_cb
, &cbdata
);
3526 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3530 struct flow_stats_ds_cbdata
{
3531 struct ofproto
*ofproto
;
3536 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
3538 struct rule
*rule
= rule_from_cls_rule(rule_
);
3539 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
3540 struct ds
*results
= cbdata
->results
;
3541 struct ofp_match match
;
3542 uint64_t packet_count
, byte_count
;
3543 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3545 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3546 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3547 NXFF_OPENFLOW10
, &match
);
3549 ds_put_format(results
, "duration=%llds, ",
3550 (time_msec() - rule
->created
) / 1000);
3551 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3552 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3553 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3554 ofp_print_match(results
, &match
, true);
3556 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3558 ds_put_cstr(results
, "drop");
3560 ds_put_cstr(results
, "\n");
3563 /* Adds a pretty-printed description of all flows to 'results', including
3564 * those marked hidden by secchan (e.g., by in-band control). */
3566 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3568 struct ofp_match match
;
3569 struct cls_rule target
;
3570 struct flow_stats_ds_cbdata cbdata
;
3572 memset(&match
, 0, sizeof match
);
3573 match
.wildcards
= htonl(OVSFW_ALL
);
3576 cbdata
.results
= results
;
3578 cls_rule_from_match(&match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3579 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3580 flow_stats_ds_cb
, &cbdata
);
3583 struct aggregate_stats_cbdata
{
3584 struct ofproto
*ofproto
;
3586 uint64_t packet_count
;
3587 uint64_t byte_count
;
3592 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3594 struct rule
*rule
= rule_from_cls_rule(rule_
);
3595 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
3596 uint64_t packet_count
, byte_count
;
3598 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3602 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3604 cbdata
->packet_count
+= packet_count
;
3605 cbdata
->byte_count
+= byte_count
;
3610 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3611 ovs_be16 out_port
, uint8_t table_id
,
3612 struct ofp_aggregate_stats_reply
*oasr
)
3614 struct aggregate_stats_cbdata cbdata
;
3616 COVERAGE_INC(ofproto_agg_request
);
3617 cbdata
.ofproto
= ofproto
;
3618 cbdata
.out_port
= out_port
;
3619 cbdata
.packet_count
= 0;
3620 cbdata
.byte_count
= 0;
3622 classifier_for_each_match(&ofproto
->cls
, target
,
3623 table_id_to_include(table_id
),
3624 aggregate_stats_cb
, &cbdata
);
3626 oasr
->flow_count
= htonl(cbdata
.n_flows
);
3627 oasr
->packet_count
= htonll(cbdata
.packet_count
);
3628 oasr
->byte_count
= htonll(cbdata
.byte_count
);
3629 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3633 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3634 const struct ofp_stats_request
*osr
,
3637 struct ofp_aggregate_stats_request
*request
;
3638 struct ofp_aggregate_stats_reply
*reply
;
3639 struct cls_rule target
;
3642 if (arg_size
!= sizeof *request
) {
3643 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3645 request
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3647 cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3649 msg
= start_ofp_stats_reply(osr
, sizeof *reply
);
3650 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3651 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3652 request
->table_id
, reply
);
3653 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3658 handle_nxst_aggregate(struct ofconn
*ofconn
, struct ofpbuf
*b
)
3660 struct nx_aggregate_stats_request
*request
;
3661 struct ofp_aggregate_stats_reply
*reply
;
3662 struct cls_rule target
;
3666 /* Dissect the message. */
3667 request
= ofpbuf_try_pull(b
, sizeof *request
);
3669 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3671 error
= nx_pull_match(b
, ntohs(request
->match_len
), 0, &target
);
3677 COVERAGE_INC(ofproto_flows_req
);
3678 buf
= start_nxstats_reply(&request
->nsm
, sizeof *reply
);
3679 reply
= ofpbuf_put_uninit(buf
, sizeof *reply
);
3680 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3681 request
->table_id
, reply
);
3682 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3687 struct queue_stats_cbdata
{
3688 struct ofconn
*ofconn
;
3689 struct ofport
*ofport
;
3694 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3695 const struct netdev_queue_stats
*stats
)
3697 struct ofp_queue_stats
*reply
;
3699 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3700 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3701 memset(reply
->pad
, 0, sizeof reply
->pad
);
3702 reply
->queue_id
= htonl(queue_id
);
3703 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3704 reply
->tx_packets
= htonll(stats
->tx_packets
);
3705 reply
->tx_errors
= htonll(stats
->tx_errors
);
3709 handle_queue_stats_dump_cb(uint32_t queue_id
,
3710 struct netdev_queue_stats
*stats
,
3713 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3715 put_queue_stats(cbdata
, queue_id
, stats
);
3719 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3720 struct queue_stats_cbdata
*cbdata
)
3722 cbdata
->ofport
= port
;
3723 if (queue_id
== OFPQ_ALL
) {
3724 netdev_dump_queue_stats(port
->netdev
,
3725 handle_queue_stats_dump_cb
, cbdata
);
3727 struct netdev_queue_stats stats
;
3729 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3730 put_queue_stats(cbdata
, queue_id
, &stats
);
3736 handle_queue_stats_request(struct ofconn
*ofconn
,
3737 const struct ofp_stats_request
*osr
,
3740 struct ofproto
*ofproto
= ofconn
->ofproto
;
3741 struct ofp_queue_stats_request
*qsr
;
3742 struct queue_stats_cbdata cbdata
;
3743 struct ofport
*port
;
3744 unsigned int port_no
;
3747 if (arg_size
!= sizeof *qsr
) {
3748 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3750 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3752 COVERAGE_INC(ofproto_queue_req
);
3754 cbdata
.ofconn
= ofconn
;
3755 cbdata
.msg
= start_ofp_stats_reply(osr
, 128);
3757 port_no
= ntohs(qsr
->port_no
);
3758 queue_id
= ntohl(qsr
->queue_id
);
3759 if (port_no
== OFPP_ALL
) {
3760 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3761 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3763 } else if (port_no
< ofproto
->max_ports
) {
3764 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3766 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3769 ofpbuf_delete(cbdata
.msg
);
3770 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3772 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3778 handle_vendor_stats_request(struct ofconn
*ofconn
,
3779 struct ofp_stats_request
*osr
, size_t arg_size
)
3781 struct nicira_stats_msg
*nsm
;
3786 VLOG_WARN_RL(&rl
, "truncated vendor stats request body");
3787 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3790 memcpy(&vendor
, osr
->body
, sizeof vendor
);
3791 if (vendor
!= htonl(NX_VENDOR_ID
)) {
3792 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3795 if (ntohs(osr
->header
.length
) < sizeof(struct nicira_stats_msg
)) {
3796 VLOG_WARN_RL(&rl
, "truncated Nicira stats request");
3797 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3800 nsm
= (struct nicira_stats_msg
*) osr
;
3802 b
.size
= ntohs(nsm
->header
.length
);
3803 switch (ntohl(nsm
->subtype
)) {
3805 return handle_nxst_flow(ofconn
, &b
);
3807 case NXST_AGGREGATE
:
3808 return handle_nxst_aggregate(ofconn
, &b
);
3811 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
3816 handle_stats_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3818 struct ofp_stats_request
*osr
;
3822 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3827 osr
= (struct ofp_stats_request
*) oh
;
3829 switch (ntohs(osr
->type
)) {
3831 return handle_desc_stats_request(ofconn
, osr
);
3834 return handle_flow_stats_request(ofconn
, osr
, arg_size
);
3836 case OFPST_AGGREGATE
:
3837 return handle_aggregate_stats_request(ofconn
, osr
, arg_size
);
3840 return handle_table_stats_request(ofconn
, osr
);
3843 return handle_port_stats_request(ofconn
, osr
, arg_size
);
3846 return handle_queue_stats_request(ofconn
, osr
, arg_size
);
3849 return handle_vendor_stats_request(ofconn
, osr
, arg_size
);
3852 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3856 static long long int
3857 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3859 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3863 facet_update_time(struct ofproto
*ofproto
, struct facet
*facet
,
3864 const struct odp_flow_stats
*stats
)
3866 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3867 if (used
> facet
->used
) {
3869 if (used
> facet
->rule
->used
) {
3870 facet
->rule
->used
= used
;
3872 netflow_flow_update_time(ofproto
->netflow
, &facet
->nf_flow
, used
);
3876 /* Folds the statistics from 'stats' into the counters in 'facet'.
3878 * Because of the meaning of a facet's counters, it only makes sense to do this
3879 * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a
3880 * packet that was sent by hand or if it represents statistics that have been
3881 * cleared out of the datapath. */
3883 facet_update_stats(struct ofproto
*ofproto
, struct facet
*facet
,
3884 const struct odp_flow_stats
*stats
)
3886 if (stats
->n_packets
) {
3887 facet_update_time(ofproto
, facet
, stats
);
3888 facet
->packet_count
+= stats
->n_packets
;
3889 facet
->byte_count
+= stats
->n_bytes
;
3890 netflow_flow_update_flags(&facet
->nf_flow
, stats
->tcp_flags
);
3898 uint16_t idle_timeout
;
3899 uint16_t hard_timeout
;
3903 union ofp_action
*actions
;
3907 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3908 * in which no matching flow already exists in the flow table.
3910 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3911 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3912 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3914 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3917 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3919 struct ofproto
*p
= ofconn
->ofproto
;
3920 struct ofpbuf
*packet
;
3925 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3926 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3927 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3931 if (fm
->buffer_id
!= UINT32_MAX
) {
3932 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3936 in_port
= UINT16_MAX
;
3939 rule
= rule_create(&fm
->cr
, fm
->actions
, fm
->n_actions
,
3940 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3941 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3942 rule_insert(p
, rule
, packet
, in_port
);
3946 static struct rule
*
3947 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3949 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
3953 send_buffered_packet(struct ofconn
*ofconn
,
3954 struct rule
*rule
, uint32_t buffer_id
)
3956 struct ofpbuf
*packet
;
3960 if (buffer_id
== UINT32_MAX
) {
3964 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
3969 rule_execute(ofconn
->ofproto
, rule
, in_port
, packet
);
3974 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3976 struct modify_flows_cbdata
{
3977 struct ofproto
*ofproto
;
3978 const struct flow_mod
*fm
;
3982 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
3984 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3986 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3987 * encoded by ofp_mkerr() on failure.
3989 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3992 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3994 struct modify_flows_cbdata cbdata
;
3996 cbdata
.ofproto
= ofconn
->ofproto
;
3998 cbdata
.match
= NULL
;
4000 classifier_for_each_match(&ofconn
->ofproto
->cls
, &fm
->cr
, CLS_INC_ALL
,
4001 modify_flows_cb
, &cbdata
);
4003 /* This credits the packet to whichever flow happened to happened to
4004 * match last. That's weird. Maybe we should do a lookup for the
4005 * flow that actually matches the packet? Who knows. */
4006 send_buffered_packet(ofconn
, cbdata
.match
, fm
->buffer_id
);
4009 return add_flow(ofconn
, fm
);
4013 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4014 * code as encoded by ofp_mkerr() on failure.
4016 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
4019 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4021 struct ofproto
*p
= ofconn
->ofproto
;
4022 struct rule
*rule
= find_flow_strict(p
, fm
);
4023 if (rule
&& !rule_is_hidden(rule
)) {
4024 modify_flow(p
, fm
, rule
);
4025 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
4027 return add_flow(ofconn
, fm
);
4031 /* Callback for modify_flows_loose(). */
4033 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
4035 struct rule
*rule
= rule_from_cls_rule(rule_
);
4036 struct modify_flows_cbdata
*cbdata
= cbdata_
;
4038 if (!rule_is_hidden(rule
)) {
4039 cbdata
->match
= rule
;
4040 modify_flow(cbdata
->ofproto
, cbdata
->fm
, rule
);
4044 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
4045 * been identified as a flow in 'p''s flow table to be modified, by changing
4046 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
4047 * ofp_action[] structures). */
4049 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
4051 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
4053 rule
->flow_cookie
= fm
->cookie
;
4055 /* If the actions are the same, do nothing. */
4056 if (fm
->n_actions
== rule
->n_actions
4058 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
4062 /* Replace actions. */
4063 free(rule
->actions
);
4064 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
4065 rule
->n_actions
= fm
->n_actions
;
4067 p
->need_revalidate
= true;
4072 /* OFPFC_DELETE implementation. */
4074 struct delete_flows_cbdata
{
4075 struct ofproto
*ofproto
;
4079 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
4080 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
4082 /* Implements OFPFC_DELETE. */
4084 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
4086 struct delete_flows_cbdata cbdata
;
4089 cbdata
.out_port
= htons(fm
->out_port
);
4091 classifier_for_each_match(&p
->cls
, &fm
->cr
, CLS_INC_ALL
,
4092 delete_flows_cb
, &cbdata
);
4095 /* Implements OFPFC_DELETE_STRICT. */
4097 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
4099 struct rule
*rule
= find_flow_strict(p
, fm
);
4101 delete_flow(p
, rule
, htons(fm
->out_port
));
4105 /* Callback for delete_flows_loose(). */
4107 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
4109 struct rule
*rule
= rule_from_cls_rule(rule_
);
4110 struct delete_flows_cbdata
*cbdata
= cbdata_
;
4112 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
4115 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
4116 * been identified as a flow to delete from 'p''s flow table, by deleting the
4117 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
4120 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
4121 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
4122 * specified 'out_port'. */
4124 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
4126 if (rule_is_hidden(rule
)) {
4130 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
4134 rule_send_removed(p
, rule
, OFPRR_DELETE
);
4135 rule_remove(p
, rule
);
4139 flow_mod_core(struct ofconn
*ofconn
, struct flow_mod
*fm
)
4141 struct ofproto
*p
= ofconn
->ofproto
;
4144 error
= reject_slave_controller(ofconn
, "flow_mod");
4149 error
= validate_actions(fm
->actions
, fm
->n_actions
,
4150 &fm
->cr
.flow
, p
->max_ports
);
4155 /* We do not support the emergency flow cache. It will hopefully
4156 * get dropped from OpenFlow in the near future. */
4157 if (fm
->flags
& OFPFF_EMERG
) {
4158 /* There isn't a good fit for an error code, so just state that the
4159 * flow table is full. */
4160 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
4163 switch (fm
->command
) {
4165 return add_flow(ofconn
, fm
);
4168 return modify_flows_loose(ofconn
, fm
);
4170 case OFPFC_MODIFY_STRICT
:
4171 return modify_flow_strict(ofconn
, fm
);
4174 delete_flows_loose(p
, fm
);
4177 case OFPFC_DELETE_STRICT
:
4178 delete_flow_strict(p
, fm
);
4182 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
4187 handle_ofpt_flow_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4189 struct ofp_match orig_match
;
4190 struct ofp_flow_mod
*ofm
;
4196 b
.size
= ntohs(oh
->length
);
4198 /* Dissect the message. */
4199 ofm
= ofpbuf_try_pull(&b
, sizeof *ofm
);
4201 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4203 error
= ofputil_pull_actions(&b
, b
.size
, &fm
.actions
, &fm
.n_actions
);
4208 /* Normalize ofm->match. If normalization actually changes anything, then
4209 * log the differences. */
4210 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
4211 orig_match
= ofm
->match
;
4212 normalize_match(&ofm
->match
);
4213 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
4214 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
4215 if (!VLOG_DROP_INFO(&normal_rl
)) {
4216 char *old
= ofp_match_to_literal_string(&orig_match
);
4217 char *new = ofp_match_to_literal_string(&ofm
->match
);
4218 VLOG_INFO("%s: normalization changed ofp_match, details:",
4219 rconn_get_name(ofconn
->rconn
));
4220 VLOG_INFO(" pre: %s", old
);
4221 VLOG_INFO("post: %s", new);
4227 /* Translate the message. */
4228 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
), ofconn
->flow_format
,
4229 ofm
->cookie
, &fm
.cr
);
4230 fm
.cookie
= ofm
->cookie
;
4231 fm
.command
= ntohs(ofm
->command
);
4232 fm
.idle_timeout
= ntohs(ofm
->idle_timeout
);
4233 fm
.hard_timeout
= ntohs(ofm
->hard_timeout
);
4234 fm
.buffer_id
= ntohl(ofm
->buffer_id
);
4235 fm
.out_port
= ntohs(ofm
->out_port
);
4236 fm
.flags
= ntohs(ofm
->flags
);
4238 /* Execute the command. */
4239 return flow_mod_core(ofconn
, &fm
);
4243 handle_nxt_flow_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4245 struct nx_flow_mod
*nfm
;
4251 b
.size
= ntohs(oh
->length
);
4253 /* Dissect the message. */
4254 nfm
= ofpbuf_try_pull(&b
, sizeof *nfm
);
4256 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4258 error
= nx_pull_match(&b
, ntohs(nfm
->match_len
), ntohs(nfm
->priority
),
4263 error
= ofputil_pull_actions(&b
, b
.size
, &fm
.actions
, &fm
.n_actions
);
4268 /* Translate the message. */
4269 fm
.cookie
= nfm
->cookie
;
4270 fm
.command
= ntohs(nfm
->command
);
4271 fm
.idle_timeout
= ntohs(nfm
->idle_timeout
);
4272 fm
.hard_timeout
= ntohs(nfm
->hard_timeout
);
4273 fm
.buffer_id
= ntohl(nfm
->buffer_id
);
4274 fm
.out_port
= ntohs(nfm
->out_port
);
4275 fm
.flags
= ntohs(nfm
->flags
);
4277 /* Execute the command. */
4278 return flow_mod_core(ofconn
, &fm
);
4282 handle_tun_id_from_cookie(struct ofconn
*ofconn
, struct nxt_tun_id_cookie
*msg
)
4286 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
4291 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
4296 handle_role_request(struct ofconn
*ofconn
, struct nicira_header
*msg
)
4298 struct nx_role_request
*nrr
;
4299 struct nx_role_request
*reply
;
4303 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
4304 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
4305 ntohs(msg
->header
.length
), sizeof *nrr
);
4306 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4308 nrr
= (struct nx_role_request
*) msg
;
4310 if (ofconn
->type
!= OFCONN_PRIMARY
) {
4311 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
4313 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4316 role
= ntohl(nrr
->role
);
4317 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
4318 && role
!= NX_ROLE_SLAVE
) {
4319 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
4321 /* There's no good error code for this. */
4322 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
4325 if (role
== NX_ROLE_MASTER
) {
4326 struct ofconn
*other
;
4328 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
4329 if (other
->role
== NX_ROLE_MASTER
) {
4330 other
->role
= NX_ROLE_SLAVE
;
4334 ofconn
->role
= role
;
4336 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, msg
->header
.xid
,
4338 reply
->role
= htonl(role
);
4339 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4345 handle_nxt_set_flow_format(struct ofconn
*ofconn
,
4346 struct nxt_set_flow_format
*msg
)
4351 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
4356 format
= ntohl(msg
->format
);
4357 if (format
== NXFF_OPENFLOW10
4358 || format
== NXFF_TUN_ID_FROM_COOKIE
4359 || format
== NXFF_NXM
) {
4360 ofconn
->flow_format
= format
;
4363 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
4368 handle_vendor(struct ofconn
*ofconn
, void *msg
)
4370 struct ofproto
*p
= ofconn
->ofproto
;
4371 struct ofp_vendor_header
*ovh
= msg
;
4372 struct nicira_header
*nh
;
4374 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
4375 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
4376 "(expected at least %zu)",
4377 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
4378 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4380 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
4381 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
4383 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
4384 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
4385 "(expected at least %zu)",
4386 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
4387 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4391 switch (ntohl(nh
->subtype
)) {
4392 case NXT_STATUS_REQUEST
:
4393 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
4396 case NXT_TUN_ID_FROM_COOKIE
:
4397 return handle_tun_id_from_cookie(ofconn
, msg
);
4399 case NXT_ROLE_REQUEST
:
4400 return handle_role_request(ofconn
, msg
);
4402 case NXT_SET_FLOW_FORMAT
:
4403 return handle_nxt_set_flow_format(ofconn
, msg
);
4406 return handle_nxt_flow_mod(ofconn
, &ovh
->header
);
4409 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
4413 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4415 struct ofp_header
*ob
;
4418 /* Currently, everything executes synchronously, so we can just
4419 * immediately send the barrier reply. */
4420 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4421 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4426 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4428 struct ofp_header
*oh
= ofp_msg
->data
;
4431 COVERAGE_INC(ofproto_recv_openflow
);
4433 case OFPT_ECHO_REQUEST
:
4434 error
= handle_echo_request(ofconn
, oh
);
4437 case OFPT_ECHO_REPLY
:
4441 case OFPT_FEATURES_REQUEST
:
4442 error
= handle_features_request(ofconn
, oh
);
4445 case OFPT_GET_CONFIG_REQUEST
:
4446 error
= handle_get_config_request(ofconn
, oh
);
4449 case OFPT_SET_CONFIG
:
4450 error
= handle_set_config(ofconn
, ofp_msg
->data
);
4453 case OFPT_PACKET_OUT
:
4454 error
= handle_packet_out(ofconn
, ofp_msg
->data
);
4458 error
= handle_port_mod(ofconn
, oh
);
4462 error
= handle_ofpt_flow_mod(ofconn
, ofp_msg
->data
);
4465 case OFPT_STATS_REQUEST
:
4466 error
= handle_stats_request(ofconn
, oh
);
4470 error
= handle_vendor(ofconn
, ofp_msg
->data
);
4473 case OFPT_BARRIER_REQUEST
:
4474 error
= handle_barrier_request(ofconn
, oh
);
4478 if (VLOG_IS_WARN_ENABLED()) {
4479 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4480 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4483 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4488 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4493 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4495 struct odp_msg
*msg
= packet
->data
;
4496 struct ofpbuf payload
;
4497 struct facet
*facet
;
4500 payload
.data
= msg
+ 1;
4501 payload
.size
= msg
->length
- sizeof *msg
;
4502 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4504 /* Check with in-band control to see if this packet should be sent
4505 * to the local port regardless of the flow table. */
4506 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4507 union odp_action action
;
4509 memset(&action
, 0, sizeof(action
));
4510 action
.output
.type
= ODPAT_OUTPUT
;
4511 action
.output
.port
= ODPP_LOCAL
;
4512 dpif_execute(p
->dpif
, &action
, 1, &payload
);
4515 facet
= facet_lookup_valid(p
, &flow
);
4517 struct rule
*rule
= rule_lookup(p
, &flow
);
4519 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4520 struct ofport
*port
= get_port(p
, msg
->port
);
4522 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4523 COVERAGE_INC(ofproto_no_packet_in
);
4524 /* XXX install 'drop' flow entry */
4525 ofpbuf_delete(packet
);
4529 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
,
4533 COVERAGE_INC(ofproto_packet_in
);
4534 send_packet_in(p
, packet
);
4538 facet
= facet_create(p
, rule
, &flow
, packet
);
4539 } else if (!facet
->may_install
) {
4540 /* The facet is not installable, that is, we need to process every
4541 * packet, so process the current packet's actions into 'facet'. */
4542 facet_make_actions(p
, facet
, packet
);
4545 if (facet
->rule
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4547 * Extra-special case for fail-open mode.
4549 * We are in fail-open mode and the packet matched the fail-open rule,
4550 * but we are connected to a controller too. We should send the packet
4551 * up to the controller in the hope that it will try to set up a flow
4552 * and thereby allow us to exit fail-open.
4554 * See the top-level comment in fail-open.c for more information.
4556 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4557 DPIF_RECV_MSG_PADDING
));
4560 ofpbuf_pull(packet
, sizeof *msg
);
4561 facet_execute(p
, facet
, packet
);
4562 facet_install(p
, facet
, false);
4566 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4568 struct odp_msg
*msg
= packet
->data
;
4570 switch (msg
->type
) {
4571 case _ODPL_ACTION_NR
:
4572 COVERAGE_INC(ofproto_ctlr_action
);
4573 send_packet_in(p
, packet
);
4576 case _ODPL_SFLOW_NR
:
4578 ofproto_sflow_received(p
->sflow
, msg
);
4580 ofpbuf_delete(packet
);
4584 handle_odp_miss_msg(p
, packet
);
4588 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4594 /* Flow expiration. */
4596 struct expire_cbdata
{
4597 struct ofproto
*ofproto
;
4601 static int ofproto_dp_max_idle(const struct ofproto
*);
4602 static void ofproto_update_used(struct ofproto
*);
4603 static void rule_expire(struct cls_rule
*, void *cbdata
);
4604 static void ofproto_expire_facets(struct ofproto
*, int dp_max_idle
);
4606 /* This function is called periodically by ofproto_run(). Its job is to
4607 * collect updates for the flows that have been installed into the datapath,
4608 * most importantly when they last were used, and then use that information to
4609 * expire flows that have not been used recently.
4611 * Returns the number of milliseconds after which it should be called again. */
4613 ofproto_expire(struct ofproto
*ofproto
)
4615 struct expire_cbdata cbdata
;
4617 /* Update 'used' for each flow in the datapath. */
4618 ofproto_update_used(ofproto
);
4620 /* Expire facets that have been idle too long. */
4621 cbdata
.dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4622 ofproto_expire_facets(ofproto
, cbdata
.dp_max_idle
);
4624 /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */
4625 cbdata
.ofproto
= ofproto
;
4626 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, rule_expire
, &cbdata
);
4628 /* Let the hook know that we're at a stable point: all outstanding data
4629 * in existing flows has been accounted to the account_cb. Thus, the
4630 * hook can now reasonably do operations that depend on having accurate
4631 * flow volume accounting (currently, that's just bond rebalancing). */
4632 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4633 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4636 return MIN(cbdata
.dp_max_idle
, 1000);
4639 /* Update 'used' member of installed facets. */
4641 ofproto_update_used(struct ofproto
*p
)
4643 struct odp_flow
*flows
;
4648 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4653 for (i
= 0; i
< n_flows
; i
++) {
4654 struct odp_flow
*f
= &flows
[i
];
4655 struct facet
*facet
;
4658 odp_flow_key_to_flow(&f
->key
, &flow
);
4659 facet
= facet_find(p
, &flow
);
4661 if (facet
&& facet
->installed
) {
4662 facet_update_time(p
, facet
, &f
->stats
);
4663 facet_account(p
, facet
, f
->stats
.n_bytes
);
4665 /* There's a flow in the datapath that we know nothing about.
4667 COVERAGE_INC(ofproto_unexpected_rule
);
4668 dpif_flow_del(p
->dpif
, f
);
4675 /* Calculates and returns the number of milliseconds of idle time after which
4676 * facets should expire from the datapath and we should fold their statistics
4677 * into their parent rules in userspace. */
4679 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4682 * Idle time histogram.
4684 * Most of the time a switch has a relatively small number of facets. When
4685 * this is the case we might as well keep statistics for all of them in
4686 * userspace and to cache them in the kernel datapath for performance as
4689 * As the number of facets increases, the memory required to maintain
4690 * statistics about them in userspace and in the kernel becomes
4691 * significant. However, with a large number of facets it is likely that
4692 * only a few of them are "heavy hitters" that consume a large amount of
4693 * bandwidth. At this point, only heavy hitters are worth caching in the
4694 * kernel and maintaining in userspaces; other facets we can discard.
4696 * The technique used to compute the idle time is to build a histogram with
4697 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet
4698 * that is installed in the kernel gets dropped in the appropriate bucket.
4699 * After the histogram has been built, we compute the cutoff so that only
4700 * the most-recently-used 1% of facets (but at least 1000 flows) are kept
4701 * cached. At least the most-recently-used bucket of facets is kept, so
4702 * actually an arbitrary number of facets can be kept in any given
4703 * expiration run (though the next run will delete most of those unless
4704 * they receive additional data).
4706 * This requires a second pass through the facets, in addition to the pass
4707 * made by ofproto_update_used(), because the former function never looks
4708 * at uninstallable facets.
4710 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4711 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4712 int buckets
[N_BUCKETS
] = { 0 };
4713 struct facet
*facet
;
4718 total
= hmap_count(&ofproto
->facets
);
4719 if (total
<= 1000) {
4720 return N_BUCKETS
* BUCKET_WIDTH
;
4723 /* Build histogram. */
4725 HMAP_FOR_EACH (facet
, hmap_node
, &ofproto
->facets
) {
4726 long long int idle
= now
- facet
->used
;
4727 int bucket
= (idle
<= 0 ? 0
4728 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4729 : (unsigned int) idle
/ BUCKET_WIDTH
);
4733 /* Find the first bucket whose flows should be expired. */
4734 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4735 if (buckets
[bucket
]) {
4738 subtotal
+= buckets
[bucket
++];
4739 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4744 if (VLOG_IS_DBG_ENABLED()) {
4748 ds_put_cstr(&s
, "keep");
4749 for (i
= 0; i
< N_BUCKETS
; i
++) {
4751 ds_put_cstr(&s
, ", drop");
4754 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4757 VLOG_INFO("%s: %s (msec:count)",
4758 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4762 return bucket
* BUCKET_WIDTH
;
4766 facet_active_timeout(struct ofproto
*ofproto
, struct facet
*facet
)
4768 if (ofproto
->netflow
&& !facet_is_controller_flow(facet
) &&
4769 netflow_active_timeout_expired(ofproto
->netflow
, &facet
->nf_flow
)) {
4770 struct ofexpired expired
;
4771 struct odp_flow odp_flow
;
4773 /* Get updated flow stats.
4775 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4776 * updated TCP flags and (2) the dpif_flow_list_all() in
4777 * ofproto_update_used() zeroed TCP flags. */
4778 memset(&odp_flow
, 0, sizeof odp_flow
);
4779 if (facet
->installed
) {
4780 odp_flow_key_from_flow(&odp_flow
.key
, &facet
->flow
);
4781 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4782 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4784 if (odp_flow
.stats
.n_packets
) {
4785 facet_update_time(ofproto
, facet
, &odp_flow
.stats
);
4786 netflow_flow_update_flags(&facet
->nf_flow
,
4787 odp_flow
.stats
.tcp_flags
);
4791 expired
.flow
= facet
->flow
;
4792 expired
.packet_count
= facet
->packet_count
+
4793 odp_flow
.stats
.n_packets
;
4794 expired
.byte_count
= facet
->byte_count
+ odp_flow
.stats
.n_bytes
;
4795 expired
.used
= facet
->used
;
4797 netflow_expire(ofproto
->netflow
, &facet
->nf_flow
, &expired
);
4802 ofproto_expire_facets(struct ofproto
*ofproto
, int dp_max_idle
)
4804 long long int cutoff
= time_msec() - dp_max_idle
;
4805 struct facet
*facet
, *next_facet
;
4807 HMAP_FOR_EACH_SAFE (facet
, next_facet
, hmap_node
, &ofproto
->facets
) {
4808 facet_active_timeout(ofproto
, facet
);
4809 if (facet
->used
< cutoff
) {
4810 facet_remove(ofproto
, facet
);
4815 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4816 * rules, then delete it entirely.
4818 * (This is a callback function for classifier_for_each().) */
4820 rule_expire(struct cls_rule
*cls_rule
, void *cbdata_
)
4822 struct expire_cbdata
*cbdata
= cbdata_
;
4823 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
4824 struct facet
*facet
, *next_facet
;
4828 /* Has 'rule' expired? */
4830 if (rule
->hard_timeout
4831 && now
> rule
->created
+ rule
->hard_timeout
* 1000) {
4832 reason
= OFPRR_HARD_TIMEOUT
;
4833 } else if (rule
->idle_timeout
&& list_is_empty(&rule
->facets
)
4834 && now
>rule
->used
+ rule
->idle_timeout
* 1000) {
4835 reason
= OFPRR_IDLE_TIMEOUT
;
4840 COVERAGE_INC(ofproto_expired
);
4842 /* Update stats. (This is a no-op if the rule expired due to an idle
4843 * timeout, because that only happens when the rule has no facets left.) */
4844 LIST_FOR_EACH_SAFE (facet
, next_facet
, list_node
, &rule
->facets
) {
4845 facet_remove(cbdata
->ofproto
, facet
);
4848 /* Get rid of the rule. */
4849 if (!rule_is_hidden(rule
)) {
4850 rule_send_removed(cbdata
->ofproto
, rule
, reason
);
4852 rule_remove(cbdata
->ofproto
, rule
);
4855 static struct ofpbuf
*
4856 compose_ofp_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4859 struct ofp_flow_removed
*ofr
;
4862 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4863 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, ofconn
->flow_format
,
4865 ofr
->cookie
= rule
->flow_cookie
;
4866 ofr
->priority
= htons(rule
->cr
.priority
);
4867 ofr
->reason
= reason
;
4868 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4869 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4870 ofr
->packet_count
= htonll(rule
->packet_count
);
4871 ofr
->byte_count
= htonll(rule
->byte_count
);
4876 static struct ofpbuf
*
4877 compose_nx_flow_removed(const struct rule
*rule
, uint8_t reason
)
4879 struct nx_flow_removed
*nfr
;
4883 nfr
= make_nxmsg(sizeof *nfr
, NXT_FLOW_REMOVED
, &buf
);
4885 match_len
= nx_put_match(buf
, &rule
->cr
);
4887 nfr
->cookie
= rule
->flow_cookie
;
4888 nfr
->priority
= htons(rule
->cr
.priority
);
4889 nfr
->reason
= reason
;
4890 calc_flow_duration(rule
->created
, &nfr
->duration_sec
, &nfr
->duration_nsec
);
4891 nfr
->idle_timeout
= htons(rule
->idle_timeout
);
4892 nfr
->match_len
= htons(match_len
);
4893 nfr
->packet_count
= htonll(rule
->packet_count
);
4894 nfr
->byte_count
= htonll(rule
->byte_count
);
4900 rule_send_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4902 struct ofconn
*ofconn
;
4904 if (!rule
->send_flow_removed
) {
4908 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4911 if (!rconn_is_connected(ofconn
->rconn
)
4912 || !ofconn_receives_async_msgs(ofconn
)) {
4916 msg
= (ofconn
->flow_format
== NXFF_NXM
4917 ? compose_nx_flow_removed(rule
, reason
)
4918 : compose_ofp_flow_removed(ofconn
, rule
, reason
));
4920 /* Account flow expirations under ofconn->reply_counter, the counter
4921 * for replies to OpenFlow requests. That works because preventing
4922 * OpenFlow requests from being processed also prevents new flows from
4923 * being added (and expiring). (It also prevents processing OpenFlow
4924 * requests that would not add new flows, so it is imperfect.) */
4925 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4929 /* pinsched callback for sending 'packet' on 'ofconn'. */
4931 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4933 struct ofconn
*ofconn
= ofconn_
;
4935 rconn_send_with_limit(ofconn
->rconn
, packet
,
4936 ofconn
->packet_in_counter
, 100);
4939 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4940 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4941 * packet scheduler for sending.
4943 * 'max_len' specifies the maximum number of bytes of the packet to send on
4944 * 'ofconn' (INT_MAX specifies no limit).
4946 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4947 * ownership is transferred to this function. */
4949 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4952 struct ofproto
*ofproto
= ofconn
->ofproto
;
4953 struct ofp_packet_in
*opi
= packet
->data
;
4954 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4955 int send_len
, trim_size
;
4959 if (opi
->reason
== OFPR_ACTION
) {
4960 buffer_id
= UINT32_MAX
;
4961 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4962 buffer_id
= pktbuf_get_null();
4963 } else if (!ofconn
->pktbuf
) {
4964 buffer_id
= UINT32_MAX
;
4966 struct ofpbuf payload
;
4967 payload
.data
= opi
->data
;
4968 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4969 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4972 /* Figure out how much of the packet to send. */
4973 send_len
= ntohs(opi
->total_len
);
4974 if (buffer_id
!= UINT32_MAX
) {
4975 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4977 send_len
= MIN(send_len
, max_len
);
4979 /* Adjust packet length and clone if necessary. */
4980 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4982 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4985 packet
->size
= trim_size
;
4988 /* Update packet headers. */
4989 opi
->buffer_id
= htonl(buffer_id
);
4990 update_openflow_length(packet
);
4992 /* Hand over to packet scheduler. It might immediately call into
4993 * do_send_packet_in() or it might buffer it for a while (until a later
4994 * call to pinsched_run()). */
4995 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4996 packet
, do_send_packet_in
, ofconn
);
4999 /* Replace struct odp_msg header in 'packet' by equivalent struct
5000 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
5001 * returned by dpif_recv()).
5003 * The conversion is not complete: the caller still needs to trim any unneeded
5004 * payload off the end of the buffer, set the length in the OpenFlow header,
5005 * and set buffer_id. Those require us to know the controller settings and so
5006 * must be done on a per-controller basis.
5008 * Returns the maximum number of bytes of the packet that should be sent to
5009 * the controller (INT_MAX if no limit). */
5011 do_convert_to_packet_in(struct ofpbuf
*packet
)
5013 struct odp_msg
*msg
= packet
->data
;
5014 struct ofp_packet_in
*opi
;
5020 /* Extract relevant header fields */
5021 if (msg
->type
== _ODPL_ACTION_NR
) {
5022 reason
= OFPR_ACTION
;
5025 reason
= OFPR_NO_MATCH
;
5028 total_len
= msg
->length
- sizeof *msg
;
5029 in_port
= odp_port_to_ofp_port(msg
->port
);
5031 /* Repurpose packet buffer by overwriting header. */
5032 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
5033 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
5034 opi
->header
.version
= OFP_VERSION
;
5035 opi
->header
.type
= OFPT_PACKET_IN
;
5036 opi
->total_len
= htons(total_len
);
5037 opi
->in_port
= htons(in_port
);
5038 opi
->reason
= reason
;
5043 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
5044 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
5045 * as necessary according to their individual configurations.
5047 * 'packet' must have sufficient headroom to convert it into a struct
5048 * ofp_packet_in (e.g. as returned by dpif_recv()).
5050 * Takes ownership of 'packet'. */
5052 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
5054 struct ofconn
*ofconn
, *prev
;
5057 max_len
= do_convert_to_packet_in(packet
);
5060 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
5061 if (ofconn_receives_async_msgs(ofconn
)) {
5063 schedule_packet_in(prev
, packet
, max_len
, true);
5069 schedule_packet_in(prev
, packet
, max_len
, false);
5071 ofpbuf_delete(packet
);
5076 pick_datapath_id(const struct ofproto
*ofproto
)
5078 const struct ofport
*port
;
5080 port
= get_port(ofproto
, ODPP_LOCAL
);
5082 uint8_t ea
[ETH_ADDR_LEN
];
5085 error
= netdev_get_etheraddr(port
->netdev
, ea
);
5087 return eth_addr_to_uint64(ea
);
5089 VLOG_WARN("could not get MAC address for %s (%s)",
5090 netdev_get_name(port
->netdev
), strerror(error
));
5092 return ofproto
->fallback_dpid
;
5096 pick_fallback_dpid(void)
5098 uint8_t ea
[ETH_ADDR_LEN
];
5099 eth_addr_nicira_random(ea
);
5100 return eth_addr_to_uint64(ea
);
5104 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
5105 struct odp_actions
*actions
, tag_type
*tags
,
5106 uint16_t *nf_output_iface
, void *ofproto_
)
5108 struct ofproto
*ofproto
= ofproto_
;
5111 /* Drop frames for reserved multicast addresses. */
5112 if (eth_addr_is_reserved(flow
->dl_dst
)) {
5116 /* Learn source MAC (but don't try to learn from revalidation). */
5117 if (packet
!= NULL
) {
5118 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
5120 GRAT_ARP_LOCK_NONE
);
5122 /* The log messages here could actually be useful in debugging,
5123 * so keep the rate limit relatively high. */
5124 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
5125 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
5126 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
5127 ofproto_revalidate(ofproto
, rev_tag
);
5131 /* Determine output port. */
5132 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
5135 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
5136 nf_output_iface
, actions
);
5137 } else if (out_port
!= flow
->in_port
) {
5138 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
5139 *nf_output_iface
= out_port
;
5147 static const struct ofhooks default_ofhooks
= {
5148 default_normal_ofhook_cb
,