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"
41 #include "ofp-print.h"
43 #include "ofproto-sflow.h"
45 #include "openflow/nicira-ext.h"
46 #include "openflow/openflow.h"
47 #include "openvswitch/datapath-protocol.h"
51 #include "poll-loop.h"
55 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto
);
65 #include "sflow_api.h"
68 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
69 struct netdev
*netdev
;
70 struct ofp_phy_port opp
; /* In host byte order. */
74 static void ofport_free(struct ofport
*);
75 static void hton_ofp_phy_port(struct ofp_phy_port
*);
77 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
78 const struct flow
*, struct ofproto
*,
79 const struct ofpbuf
*packet
,
80 struct odp_actions
*out
, tag_type
*tags
,
81 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
86 ovs_be64 flow_cookie
; /* Controller-issued identifier. */
87 uint16_t idle_timeout
; /* In seconds from time of last use. */
88 uint16_t hard_timeout
; /* In seconds from time of creation. */
89 bool send_flow_removed
; /* Send a flow removed message? */
90 long long int used
; /* Last-used time (0 if never used). */
91 long long int created
; /* Creation time. */
92 uint64_t packet_count
; /* Number of packets received. */
93 uint64_t byte_count
; /* Number of bytes received. */
94 uint64_t accounted_bytes
; /* Number of bytes passed to account_cb. */
95 tag_type tags
; /* Tags (set only by hooks). */
96 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
98 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
99 * exact-match rule (having cr.wc.wildcards of 0) generated from the
100 * wildcard rule 'super'. In this case, 'list' is an element of the
103 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
104 * a list of subrules. A super-rule with no wildcards (where
105 * cr.wc.wildcards is 0) will never have any subrules. */
111 * 'n_actions' is the number of elements in the 'actions' array. A single
112 * action may take up more more than one element's worth of space.
114 * A subrule has no actions (it uses the super-rule's actions). */
116 union ofp_action
*actions
;
120 * A super-rule with wildcard fields never has ODP actions (since the
121 * datapath only supports exact-match flows). */
122 bool installed
; /* Installed in datapath? */
123 bool may_install
; /* True ordinarily; false if actions must
124 * be reassessed for every packet. */
126 union odp_action
*odp_actions
;
130 rule_is_hidden(const struct rule
*rule
)
132 /* Subrules are merely an implementation detail, so hide them from the
134 if (rule
->super
!= NULL
) {
138 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
139 * (e.g. by in-band control) and are intentionally hidden from the
141 if (rule
->cr
.priority
> UINT16_MAX
) {
148 static struct rule
*rule_create(struct ofproto
*, struct rule
*super
,
149 const union ofp_action
*, size_t n_actions
,
150 uint16_t idle_timeout
, uint16_t hard_timeout
,
151 ovs_be64 flow_cookie
, bool send_flow_removed
);
152 static void rule_free(struct rule
*);
153 static void rule_destroy(struct ofproto
*, struct rule
*);
154 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
155 static void rule_insert(struct ofproto
*, struct rule
*,
156 struct ofpbuf
*packet
, uint16_t in_port
);
157 static void rule_remove(struct ofproto
*, struct rule
*);
158 static bool rule_make_actions(struct ofproto
*, struct rule
*,
159 const struct ofpbuf
*packet
);
160 static void rule_install(struct ofproto
*, struct rule
*,
161 struct rule
*displaced_rule
);
162 static void rule_uninstall(struct ofproto
*, struct rule
*);
163 static void rule_post_uninstall(struct ofproto
*, struct rule
*);
164 static void send_flow_removed(struct ofproto
*, struct rule
*, uint8_t reason
);
166 /* ofproto supports two kinds of OpenFlow connections:
168 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
169 * maintains persistent connections to these controllers and by default
170 * sends them asynchronous messages such as packet-ins.
172 * - "Service" connections, e.g. from ovs-ofctl. When these connections
173 * drop, it is the other side's responsibility to reconnect them if
174 * necessary. ofproto does not send them asynchronous messages by default.
176 * Currently, active (tcp, ssl, unix) connections are always "primary"
177 * connections and passive (ptcp, pssl, punix) connections are always "service"
178 * connections. There is no inherent reason for this, but it reflects the
182 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
183 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
186 /* A listener for incoming OpenFlow "service" connections. */
188 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
189 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
191 /* These are not used by ofservice directly. They are settings for
192 * accepted "struct ofconn"s from the pvconn. */
193 int probe_interval
; /* Max idle time before probing, in seconds. */
194 int rate_limit
; /* Max packet-in rate in packets per second. */
195 int burst_limit
; /* Limit on accumulating packet credits. */
198 static struct ofservice
*ofservice_lookup(struct ofproto
*,
200 static int ofservice_create(struct ofproto
*,
201 const struct ofproto_controller
*);
202 static void ofservice_reconfigure(struct ofservice
*,
203 const struct ofproto_controller
*);
204 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
206 /* An OpenFlow connection. */
208 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
209 struct list node
; /* In struct ofproto's "all_conns" list. */
210 struct rconn
*rconn
; /* OpenFlow connection. */
211 enum ofconn_type type
; /* Type. */
212 int flow_format
; /* One of NXFF_*. */
214 /* OFPT_PACKET_IN related data. */
215 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
216 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
217 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
218 int miss_send_len
; /* Bytes to send of buffered packets. */
220 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
221 * requests, and the maximum number before we stop reading OpenFlow
223 #define OFCONN_REPLY_MAX 100
224 struct rconn_packet_counter
*reply_counter
;
226 /* type == OFCONN_PRIMARY only. */
227 enum nx_role role
; /* Role. */
228 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
229 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
230 struct status_category
*ss
; /* Switch status category. */
231 enum ofproto_band band
; /* In-band or out-of-band? */
234 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
235 * "schedulers" array. Their values are 0 and 1, and their meanings and values
236 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
237 * case anything ever changes, check their values here. */
238 #define N_SCHEDULERS 2
239 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
240 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
241 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
242 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
244 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
246 static void ofconn_destroy(struct ofconn
*);
247 static void ofconn_run(struct ofconn
*);
248 static void ofconn_wait(struct ofconn
*);
249 static bool ofconn_receives_async_msgs(const struct ofconn
*);
250 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
251 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
253 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
254 struct rconn_packet_counter
*counter
);
256 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
257 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
261 uint64_t datapath_id
; /* Datapath ID. */
262 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
263 char *mfr_desc
; /* Manufacturer. */
264 char *hw_desc
; /* Hardware. */
265 char *sw_desc
; /* Software version. */
266 char *serial_desc
; /* Serial number. */
267 char *dp_desc
; /* Datapath description. */
271 struct netdev_monitor
*netdev_monitor
;
272 struct hmap ports
; /* Contains "struct ofport"s. */
273 struct shash port_by_name
;
277 struct switch_status
*switch_status
;
278 struct fail_open
*fail_open
;
279 struct netflow
*netflow
;
280 struct ofproto_sflow
*sflow
;
282 /* In-band control. */
283 struct in_band
*in_band
;
284 long long int next_in_band_update
;
285 struct sockaddr_in
*extra_in_band_remotes
;
286 size_t n_extra_remotes
;
289 struct classifier cls
;
290 bool need_revalidate
;
291 long long int next_expiration
;
292 struct tag_set revalidate_set
;
294 /* OpenFlow connections. */
295 struct hmap controllers
; /* Controller "struct ofconn"s. */
296 struct list all_conns
; /* Contains "struct ofconn"s. */
297 enum ofproto_fail_mode fail_mode
;
299 /* OpenFlow listeners. */
300 struct hmap services
; /* Contains "struct ofservice"s. */
301 struct pvconn
**snoops
;
304 /* Hooks for ovs-vswitchd. */
305 const struct ofhooks
*ofhooks
;
308 /* Used by default ofhooks. */
309 struct mac_learning
*ml
;
312 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
314 static const struct ofhooks default_ofhooks
;
316 static uint64_t pick_datapath_id(const struct ofproto
*);
317 static uint64_t pick_fallback_dpid(void);
319 static int ofproto_expire(struct ofproto
*);
321 static void update_stats(struct ofproto
*, struct rule
*,
322 const struct odp_flow_stats
*);
323 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
324 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
326 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
328 static void handle_openflow(struct ofconn
*, struct ofpbuf
*);
330 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
331 static void update_port(struct ofproto
*, const char *devname
);
332 static int init_ports(struct ofproto
*);
333 static void reinit_ports(struct ofproto
*);
336 ofproto_create(const char *datapath
, const char *datapath_type
,
337 const struct ofhooks
*ofhooks
, void *aux
,
338 struct ofproto
**ofprotop
)
340 struct odp_stats stats
;
347 /* Connect to datapath and start listening for messages. */
348 error
= dpif_open(datapath
, datapath_type
, &dpif
);
350 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
353 error
= dpif_get_dp_stats(dpif
, &stats
);
355 VLOG_ERR("failed to obtain stats for datapath %s: %s",
356 datapath
, strerror(error
));
360 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
362 VLOG_ERR("failed to listen on datapath %s: %s",
363 datapath
, strerror(error
));
367 dpif_flow_flush(dpif
);
368 dpif_recv_purge(dpif
);
370 /* Initialize settings. */
371 p
= xzalloc(sizeof *p
);
372 p
->fallback_dpid
= pick_fallback_dpid();
373 p
->datapath_id
= p
->fallback_dpid
;
374 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
375 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
376 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
377 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
378 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
380 /* Initialize datapath. */
382 p
->netdev_monitor
= netdev_monitor_create();
383 hmap_init(&p
->ports
);
384 shash_init(&p
->port_by_name
);
385 p
->max_ports
= stats
.max_ports
;
387 /* Initialize submodules. */
388 p
->switch_status
= switch_status_create(p
);
394 /* Initialize flow table. */
395 classifier_init(&p
->cls
);
396 p
->need_revalidate
= false;
397 p
->next_expiration
= time_msec() + 1000;
398 tag_set_init(&p
->revalidate_set
);
400 /* Initialize OpenFlow connections. */
401 list_init(&p
->all_conns
);
402 hmap_init(&p
->controllers
);
403 hmap_init(&p
->services
);
407 /* Initialize hooks. */
409 p
->ofhooks
= ofhooks
;
413 p
->ofhooks
= &default_ofhooks
;
415 p
->ml
= mac_learning_create();
418 /* Pick final datapath ID. */
419 p
->datapath_id
= pick_datapath_id(p
);
420 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
427 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
429 uint64_t old_dpid
= p
->datapath_id
;
430 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
431 if (p
->datapath_id
!= old_dpid
) {
432 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
434 /* Force all active connections to reconnect, since there is no way to
435 * notify a controller that the datapath ID has changed. */
436 ofproto_reconnect_controllers(p
);
441 is_discovery_controller(const struct ofproto_controller
*c
)
443 return !strcmp(c
->target
, "discover");
447 is_in_band_controller(const struct ofproto_controller
*c
)
449 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
452 /* Creates a new controller in 'ofproto'. Some of the settings are initially
453 * drawn from 'c', but update_controller() needs to be called later to finish
454 * the new ofconn's configuration. */
456 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
458 struct discovery
*discovery
;
459 struct ofconn
*ofconn
;
461 if (is_discovery_controller(c
)) {
462 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
463 ofproto
->dpif
, ofproto
->switch_status
,
472 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
473 ofconn
->pktbuf
= pktbuf_create();
474 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
476 ofconn
->discovery
= discovery
;
478 char *name
= ofconn_make_name(ofproto
, c
->target
);
479 rconn_connect(ofconn
->rconn
, c
->target
, name
);
482 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
483 hash_string(c
->target
, 0));
486 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
487 * target or turn discovery on or off (these are done by creating new ofconns
488 * and deleting old ones), but it can update the rest of an ofconn's
491 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
495 ofconn
->band
= (is_in_band_controller(c
)
496 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
498 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
500 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
501 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
503 if (ofconn
->discovery
) {
504 discovery_set_update_resolv_conf(ofconn
->discovery
,
505 c
->update_resolv_conf
);
506 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
509 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
513 ofconn_get_target(const struct ofconn
*ofconn
)
515 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
518 static struct ofconn
*
519 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
521 struct ofconn
*ofconn
;
523 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
524 hash_string(target
, 0), &ofproto
->controllers
) {
525 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
533 update_in_band_remotes(struct ofproto
*ofproto
)
535 const struct ofconn
*ofconn
;
536 struct sockaddr_in
*addrs
;
537 size_t max_addrs
, n_addrs
;
541 /* Allocate enough memory for as many remotes as we could possibly have. */
542 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
543 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
546 /* Add all the remotes. */
548 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
549 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
551 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
555 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
556 if (sin
->sin_addr
.s_addr
) {
557 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
560 if (ofconn
->discovery
) {
564 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
565 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
568 /* Create or update or destroy in-band.
570 * Ordinarily we only enable in-band if there's at least one remote
571 * address, but discovery needs the in-band rules for DHCP to be installed
572 * even before we know any remote addresses. */
573 if (n_addrs
|| discovery
) {
574 if (!ofproto
->in_band
) {
575 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
578 if (ofproto
->in_band
) {
579 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
581 ofproto
->next_in_band_update
= time_msec() + 1000;
583 in_band_destroy(ofproto
->in_band
);
584 ofproto
->in_band
= NULL
;
592 update_fail_open(struct ofproto
*p
)
594 struct ofconn
*ofconn
;
596 if (!hmap_is_empty(&p
->controllers
)
597 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
598 struct rconn
**rconns
;
602 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
606 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
607 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
608 rconns
[n
++] = ofconn
->rconn
;
611 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
612 /* p->fail_open takes ownership of 'rconns'. */
614 fail_open_destroy(p
->fail_open
);
620 ofproto_set_controllers(struct ofproto
*p
,
621 const struct ofproto_controller
*controllers
,
622 size_t n_controllers
)
624 struct shash new_controllers
;
625 struct ofconn
*ofconn
, *next_ofconn
;
626 struct ofservice
*ofservice
, *next_ofservice
;
630 /* Create newly configured controllers and services.
631 * Create a name to ofproto_controller mapping in 'new_controllers'. */
632 shash_init(&new_controllers
);
633 for (i
= 0; i
< n_controllers
; i
++) {
634 const struct ofproto_controller
*c
= &controllers
[i
];
636 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
637 if (!find_controller_by_target(p
, c
->target
)) {
638 add_controller(p
, c
);
640 } else if (!pvconn_verify_name(c
->target
)) {
641 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
645 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
646 dpif_name(p
->dpif
), c
->target
);
650 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
653 /* Delete controllers that are no longer configured.
654 * Update configuration of all now-existing controllers. */
656 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
657 struct ofproto_controller
*c
;
659 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
661 ofconn_destroy(ofconn
);
663 update_controller(ofconn
, c
);
670 /* Delete services that are no longer configured.
671 * Update configuration of all now-existing services. */
672 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
673 struct ofproto_controller
*c
;
675 c
= shash_find_data(&new_controllers
,
676 pvconn_get_name(ofservice
->pvconn
));
678 ofservice_destroy(p
, ofservice
);
680 ofservice_reconfigure(ofservice
, c
);
684 shash_destroy(&new_controllers
);
686 update_in_band_remotes(p
);
689 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
690 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
691 struct ofconn
, hmap_node
);
692 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
693 rconn_status_cb
, ofconn
->rconn
);
698 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
700 p
->fail_mode
= fail_mode
;
704 /* Drops the connections between 'ofproto' and all of its controllers, forcing
705 * them to reconnect. */
707 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
709 struct ofconn
*ofconn
;
711 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
712 rconn_reconnect(ofconn
->rconn
);
717 any_extras_changed(const struct ofproto
*ofproto
,
718 const struct sockaddr_in
*extras
, size_t n
)
722 if (n
!= ofproto
->n_extra_remotes
) {
726 for (i
= 0; i
< n
; i
++) {
727 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
728 const struct sockaddr_in
*new = &extras
[i
];
730 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
731 old
->sin_port
!= new->sin_port
) {
739 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
740 * in-band control should guarantee access, in the same way that in-band
741 * control guarantees access to OpenFlow controllers. */
743 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
744 const struct sockaddr_in
*extras
, size_t n
)
746 if (!any_extras_changed(ofproto
, extras
, n
)) {
750 free(ofproto
->extra_in_band_remotes
);
751 ofproto
->n_extra_remotes
= n
;
752 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
754 update_in_band_remotes(ofproto
);
758 ofproto_set_desc(struct ofproto
*p
,
759 const char *mfr_desc
, const char *hw_desc
,
760 const char *sw_desc
, const char *serial_desc
,
763 struct ofp_desc_stats
*ods
;
766 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
767 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
768 sizeof ods
->mfr_desc
);
771 p
->mfr_desc
= xstrdup(mfr_desc
);
774 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
775 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
776 sizeof ods
->hw_desc
);
779 p
->hw_desc
= xstrdup(hw_desc
);
782 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
783 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
784 sizeof ods
->sw_desc
);
787 p
->sw_desc
= xstrdup(sw_desc
);
790 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
791 VLOG_WARN("truncating serial_desc, must be less than %zu "
793 sizeof ods
->serial_num
);
795 free(p
->serial_desc
);
796 p
->serial_desc
= xstrdup(serial_desc
);
799 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
800 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
801 sizeof ods
->dp_desc
);
804 p
->dp_desc
= xstrdup(dp_desc
);
809 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
810 const struct svec
*svec
)
812 struct pvconn
**pvconns
= *pvconnsp
;
813 size_t n_pvconns
= *n_pvconnsp
;
817 for (i
= 0; i
< n_pvconns
; i
++) {
818 pvconn_close(pvconns
[i
]);
822 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
824 for (i
= 0; i
< svec
->n
; i
++) {
825 const char *name
= svec
->names
[i
];
826 struct pvconn
*pvconn
;
829 error
= pvconn_open(name
, &pvconn
);
831 pvconns
[n_pvconns
++] = pvconn
;
833 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
841 *n_pvconnsp
= n_pvconns
;
847 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
849 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
853 ofproto_set_netflow(struct ofproto
*ofproto
,
854 const struct netflow_options
*nf_options
)
856 if (nf_options
&& nf_options
->collectors
.n
) {
857 if (!ofproto
->netflow
) {
858 ofproto
->netflow
= netflow_create();
860 return netflow_set_options(ofproto
->netflow
, nf_options
);
862 netflow_destroy(ofproto
->netflow
);
863 ofproto
->netflow
= NULL
;
869 ofproto_set_sflow(struct ofproto
*ofproto
,
870 const struct ofproto_sflow_options
*oso
)
872 struct ofproto_sflow
*os
= ofproto
->sflow
;
875 struct ofport
*ofport
;
877 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
878 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
879 ofproto_sflow_add_port(os
, ofport
->odp_port
,
880 netdev_get_name(ofport
->netdev
));
883 ofproto_sflow_set_options(os
, oso
);
885 ofproto_sflow_destroy(os
);
886 ofproto
->sflow
= NULL
;
891 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
893 return ofproto
->datapath_id
;
897 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
899 return !hmap_is_empty(&ofproto
->controllers
);
902 enum ofproto_fail_mode
903 ofproto_get_fail_mode(const struct ofproto
*p
)
909 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
913 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
914 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
919 ofproto_destroy(struct ofproto
*p
)
921 struct ofservice
*ofservice
, *next_ofservice
;
922 struct ofconn
*ofconn
, *next_ofconn
;
923 struct ofport
*ofport
, *next_ofport
;
930 /* Destroy fail-open and in-band early, since they touch the classifier. */
931 fail_open_destroy(p
->fail_open
);
934 in_band_destroy(p
->in_band
);
936 free(p
->extra_in_band_remotes
);
938 ofproto_flush_flows(p
);
939 classifier_destroy(&p
->cls
);
941 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
942 ofconn_destroy(ofconn
);
944 hmap_destroy(&p
->controllers
);
947 netdev_monitor_destroy(p
->netdev_monitor
);
948 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
949 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
952 shash_destroy(&p
->port_by_name
);
954 switch_status_destroy(p
->switch_status
);
955 netflow_destroy(p
->netflow
);
956 ofproto_sflow_destroy(p
->sflow
);
958 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
959 ofservice_destroy(p
, ofservice
);
961 hmap_destroy(&p
->services
);
963 for (i
= 0; i
< p
->n_snoops
; i
++) {
964 pvconn_close(p
->snoops
[i
]);
968 mac_learning_destroy(p
->ml
);
973 free(p
->serial_desc
);
976 hmap_destroy(&p
->ports
);
982 ofproto_run(struct ofproto
*p
)
984 int error
= ofproto_run1(p
);
986 error
= ofproto_run2(p
, false);
992 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
994 if (error
== ENOBUFS
) {
995 reinit_ports(ofproto
);
997 update_port(ofproto
, devname
);
1002 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1003 * means that 'ofconn' is more interesting for monitoring than a lower return
1006 snoop_preference(const struct ofconn
*ofconn
)
1008 switch (ofconn
->role
) {
1009 case NX_ROLE_MASTER
:
1016 /* Shouldn't happen. */
1021 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1022 * Connects this vconn to a controller. */
1024 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1026 struct ofconn
*ofconn
, *best
;
1028 /* Pick a controller for monitoring. */
1030 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1031 if (ofconn
->type
== OFCONN_PRIMARY
1032 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1038 rconn_add_monitor(best
->rconn
, vconn
);
1040 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1046 ofproto_run1(struct ofproto
*p
)
1048 struct ofconn
*ofconn
, *next_ofconn
;
1049 struct ofservice
*ofservice
;
1054 if (shash_is_empty(&p
->port_by_name
)) {
1058 for (i
= 0; i
< 50; i
++) {
1061 error
= dpif_recv(p
->dpif
, &buf
);
1063 if (error
== ENODEV
) {
1064 /* Someone destroyed the datapath behind our back. The caller
1065 * better destroy us and give up, because we're just going to
1066 * spin from here on out. */
1067 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1068 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1069 dpif_name(p
->dpif
));
1075 handle_odp_msg(p
, buf
);
1078 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1079 process_port_change(p
, error
, devname
);
1081 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1082 &devname
)) != EAGAIN
) {
1083 process_port_change(p
, error
, devname
);
1087 if (time_msec() >= p
->next_in_band_update
) {
1088 update_in_band_remotes(p
);
1090 in_band_run(p
->in_band
);
1093 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1097 /* Fail-open maintenance. Do this after processing the ofconns since
1098 * fail-open checks the status of the controller rconn. */
1100 fail_open_run(p
->fail_open
);
1103 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1104 struct vconn
*vconn
;
1107 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1109 struct rconn
*rconn
;
1112 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1113 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1114 rconn_connect_unreliably(rconn
, vconn
, name
);
1117 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1118 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1119 ofservice
->burst_limit
);
1120 } else if (retval
!= EAGAIN
) {
1121 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1125 for (i
= 0; i
< p
->n_snoops
; i
++) {
1126 struct vconn
*vconn
;
1129 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1131 add_snooper(p
, vconn
);
1132 } else if (retval
!= EAGAIN
) {
1133 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1137 if (time_msec() >= p
->next_expiration
) {
1138 int delay
= ofproto_expire(p
);
1139 p
->next_expiration
= time_msec() + delay
;
1140 COVERAGE_INC(ofproto_expiration
);
1144 netflow_run(p
->netflow
);
1147 ofproto_sflow_run(p
->sflow
);
1153 struct revalidate_cbdata
{
1154 struct ofproto
*ofproto
;
1155 bool revalidate_all
; /* Revalidate all exact-match rules? */
1156 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
1157 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
1161 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1163 if (p
->need_revalidate
|| revalidate_all
1164 || !tag_set_is_empty(&p
->revalidate_set
)) {
1165 struct revalidate_cbdata cbdata
;
1167 cbdata
.revalidate_all
= revalidate_all
;
1168 cbdata
.revalidate_subrules
= p
->need_revalidate
;
1169 cbdata
.revalidate_set
= p
->revalidate_set
;
1170 tag_set_init(&p
->revalidate_set
);
1171 COVERAGE_INC(ofproto_revalidate
);
1172 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
1173 p
->need_revalidate
= false;
1180 ofproto_wait(struct ofproto
*p
)
1182 struct ofservice
*ofservice
;
1183 struct ofconn
*ofconn
;
1186 dpif_recv_wait(p
->dpif
);
1187 dpif_port_poll_wait(p
->dpif
);
1188 netdev_monitor_poll_wait(p
->netdev_monitor
);
1189 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1190 ofconn_wait(ofconn
);
1193 poll_timer_wait_until(p
->next_in_band_update
);
1194 in_band_wait(p
->in_band
);
1197 fail_open_wait(p
->fail_open
);
1200 ofproto_sflow_wait(p
->sflow
);
1202 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1203 poll_immediate_wake();
1205 if (p
->need_revalidate
) {
1206 /* Shouldn't happen, but if it does just go around again. */
1207 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1208 poll_immediate_wake();
1209 } else if (p
->next_expiration
!= LLONG_MAX
) {
1210 poll_timer_wait_until(p
->next_expiration
);
1212 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1213 pvconn_wait(ofservice
->pvconn
);
1215 for (i
= 0; i
< p
->n_snoops
; i
++) {
1216 pvconn_wait(p
->snoops
[i
]);
1221 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1223 tag_set_add(&ofproto
->revalidate_set
, tag
);
1227 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1229 return &ofproto
->revalidate_set
;
1233 ofproto_is_alive(const struct ofproto
*p
)
1235 return !hmap_is_empty(&p
->controllers
);
1238 /* Deletes port number 'odp_port' from the datapath for 'ofproto'.
1240 * This is almost the same as calling dpif_port_del() directly on the
1241 * datapath, but it also makes 'ofproto' close its open netdev for the port
1242 * (if any). This makes it possible to create a new netdev of a different
1243 * type under the same name, which otherwise the netdev library would refuse
1244 * to do because of the conflict. (The netdev would eventually get closed on
1245 * the next trip through ofproto_run(), but this interface is more direct.)
1247 * Returns 0 if successful, otherwise a positive errno. */
1249 ofproto_port_del(struct ofproto
*ofproto
, uint16_t odp_port
)
1251 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1252 const char *name
= ofport
? (char *) ofport
->opp
.name
: "<unknown>";
1255 error
= dpif_port_del(ofproto
->dpif
, odp_port
);
1257 VLOG_ERR("%s: failed to remove port %"PRIu16
" (%s) interface (%s)",
1258 dpif_name(ofproto
->dpif
), odp_port
, name
, strerror(error
));
1259 } else if (ofport
) {
1260 /* 'name' is ofport->opp.name and update_port() is going to destroy
1261 * 'ofport'. Just in case update_port() refers to 'name' after it
1262 * destroys 'ofport', make a copy of it around the update_port()
1264 char *devname
= xstrdup(name
);
1265 update_port(ofproto
, devname
);
1271 /* Checks if 'ofproto' thinks 'odp_port' should be included in floods. Returns
1272 * true if 'odp_port' exists and should be included, false otherwise. */
1274 ofproto_port_is_floodable(struct ofproto
*ofproto
, uint16_t odp_port
)
1276 struct ofport
*ofport
= get_port(ofproto
, odp_port
);
1277 return ofport
&& !(ofport
->opp
.config
& OFPPC_NO_FLOOD
);
1281 ofproto_send_packet(struct ofproto
*p
, const struct flow
*flow
,
1282 const union ofp_action
*actions
, size_t n_actions
,
1283 const struct ofpbuf
*packet
)
1285 struct odp_actions odp_actions
;
1288 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1294 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1296 dpif_execute(p
->dpif
, odp_actions
.actions
, odp_actions
.n_actions
, packet
);
1301 ofproto_add_flow(struct ofproto
*p
, const struct cls_rule
*cls_rule
,
1302 const union ofp_action
*actions
, size_t n_actions
,
1306 rule
= rule_create(p
, NULL
, actions
, n_actions
,
1307 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */,
1309 rule
->cr
= *cls_rule
;
1310 rule_insert(p
, rule
, NULL
, 0);
1314 ofproto_delete_flow(struct ofproto
*ofproto
, const struct cls_rule
*target
)
1318 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1321 rule_remove(ofproto
, rule
);
1326 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1328 struct rule
*rule
= rule_from_cls_rule(rule_
);
1329 struct ofproto
*ofproto
= ofproto_
;
1331 /* Mark the flow as not installed, even though it might really be
1332 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1333 * There is no point in uninstalling it individually since we are about to
1334 * blow away all the flows with dpif_flow_flush(). */
1335 rule
->installed
= false;
1337 rule_remove(ofproto
, rule
);
1341 ofproto_flush_flows(struct ofproto
*ofproto
)
1343 COVERAGE_INC(ofproto_flush
);
1344 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1345 dpif_flow_flush(ofproto
->dpif
);
1346 if (ofproto
->in_band
) {
1347 in_band_flushed(ofproto
->in_band
);
1349 if (ofproto
->fail_open
) {
1350 fail_open_flushed(ofproto
->fail_open
);
1355 reinit_ports(struct ofproto
*p
)
1357 struct svec devnames
;
1358 struct ofport
*ofport
;
1359 struct odp_port
*odp_ports
;
1363 COVERAGE_INC(ofproto_reinit_ports
);
1365 svec_init(&devnames
);
1366 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1367 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1369 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1370 for (i
= 0; i
< n_odp_ports
; i
++) {
1371 svec_add (&devnames
, odp_ports
[i
].devname
);
1375 svec_sort_unique(&devnames
);
1376 for (i
= 0; i
< devnames
.n
; i
++) {
1377 update_port(p
, devnames
.names
[i
]);
1379 svec_destroy(&devnames
);
1382 static struct ofport
*
1383 make_ofport(const struct odp_port
*odp_port
)
1385 struct netdev_options netdev_options
;
1386 enum netdev_flags flags
;
1387 struct ofport
*ofport
;
1388 struct netdev
*netdev
;
1391 memset(&netdev_options
, 0, sizeof netdev_options
);
1392 netdev_options
.name
= odp_port
->devname
;
1393 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1395 error
= netdev_open(&netdev_options
, &netdev
);
1397 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1398 "cannot be opened (%s)",
1399 odp_port
->devname
, odp_port
->port
,
1400 odp_port
->devname
, strerror(error
));
1404 ofport
= xmalloc(sizeof *ofport
);
1405 ofport
->netdev
= netdev
;
1406 ofport
->odp_port
= odp_port
->port
;
1407 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1408 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1409 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1410 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1411 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1413 netdev_get_flags(netdev
, &flags
);
1414 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1416 ofport
->opp
.state
= netdev_get_carrier(netdev
) ? 0 : OFPPS_LINK_DOWN
;
1418 netdev_get_features(netdev
,
1419 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1420 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1425 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1427 if (get_port(p
, odp_port
->port
)) {
1428 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1431 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1432 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1441 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1443 const struct ofp_phy_port
*a
= &a_
->opp
;
1444 const struct ofp_phy_port
*b
= &b_
->opp
;
1446 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1447 return (a
->port_no
== b
->port_no
1448 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1449 && !strcmp((char *) a
->name
, (char *) b
->name
)
1450 && a
->state
== b
->state
1451 && a
->config
== b
->config
1452 && a
->curr
== b
->curr
1453 && a
->advertised
== b
->advertised
1454 && a
->supported
== b
->supported
1455 && a
->peer
== b
->peer
);
1459 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1462 /* XXX Should limit the number of queued port status change messages. */
1463 struct ofconn
*ofconn
;
1464 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1465 struct ofp_port_status
*ops
;
1468 if (!ofconn_receives_async_msgs(ofconn
)) {
1472 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1473 ops
->reason
= reason
;
1474 ops
->desc
= ofport
->opp
;
1475 hton_ofp_phy_port(&ops
->desc
);
1476 queue_tx(b
, ofconn
, NULL
);
1481 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1483 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1485 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1486 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1487 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1489 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1494 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1496 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1497 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1498 shash_delete(&p
->port_by_name
,
1499 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1501 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1506 ofport_free(struct ofport
*ofport
)
1509 netdev_close(ofport
->netdev
);
1514 static struct ofport
*
1515 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1517 struct ofport
*port
;
1519 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1520 hash_int(odp_port
, 0), &ofproto
->ports
) {
1521 if (port
->odp_port
== odp_port
) {
1529 update_port(struct ofproto
*p
, const char *devname
)
1531 struct odp_port odp_port
;
1532 struct ofport
*old_ofport
;
1533 struct ofport
*new_ofport
;
1536 COVERAGE_INC(ofproto_update_port
);
1538 /* Query the datapath for port information. */
1539 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1541 /* Find the old ofport. */
1542 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1545 /* There's no port named 'devname' but there might be a port with
1546 * the same port number. This could happen if a port is deleted
1547 * and then a new one added in its place very quickly, or if a port
1548 * is renamed. In the former case we want to send an OFPPR_DELETE
1549 * and an OFPPR_ADD, and in the latter case we want to send a
1550 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1551 * the old port's ifindex against the new port, or perhaps less
1552 * reliably but more portably by comparing the old port's MAC
1553 * against the new port's MAC. However, this code isn't that smart
1554 * and always sends an OFPPR_MODIFY (XXX). */
1555 old_ofport
= get_port(p
, odp_port
.port
);
1557 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1558 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1559 "%s", strerror(error
));
1563 /* Create a new ofport. */
1564 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1566 /* Eliminate a few pathological cases. */
1567 if (!old_ofport
&& !new_ofport
) {
1569 } else if (old_ofport
&& new_ofport
) {
1570 /* Most of the 'config' bits are OpenFlow soft state, but
1571 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1572 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1573 * leaves the other bits 0.) */
1574 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1576 if (ofport_equal(old_ofport
, new_ofport
)) {
1577 /* False alarm--no change. */
1578 ofport_free(new_ofport
);
1583 /* Now deal with the normal cases. */
1585 ofport_remove(p
, old_ofport
);
1588 ofport_install(p
, new_ofport
);
1590 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1591 (!old_ofport
? OFPPR_ADD
1592 : !new_ofport
? OFPPR_DELETE
1594 ofport_free(old_ofport
);
1598 init_ports(struct ofproto
*p
)
1600 struct odp_port
*ports
;
1605 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1610 for (i
= 0; i
< n_ports
; i
++) {
1611 const struct odp_port
*odp_port
= &ports
[i
];
1612 if (!ofport_conflicts(p
, odp_port
)) {
1613 struct ofport
*ofport
= make_ofport(odp_port
);
1615 ofport_install(p
, ofport
);
1623 static struct ofconn
*
1624 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1626 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1627 ofconn
->ofproto
= p
;
1628 list_push_back(&p
->all_conns
, &ofconn
->node
);
1629 ofconn
->rconn
= rconn
;
1630 ofconn
->type
= type
;
1631 ofconn
->flow_format
= NXFF_OPENFLOW10
;
1632 ofconn
->role
= NX_ROLE_OTHER
;
1633 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1634 ofconn
->pktbuf
= NULL
;
1635 ofconn
->miss_send_len
= 0;
1636 ofconn
->reply_counter
= rconn_packet_counter_create ();
1641 ofconn_destroy(struct ofconn
*ofconn
)
1643 if (ofconn
->type
== OFCONN_PRIMARY
) {
1644 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1646 discovery_destroy(ofconn
->discovery
);
1648 list_remove(&ofconn
->node
);
1649 switch_status_unregister(ofconn
->ss
);
1650 rconn_destroy(ofconn
->rconn
);
1651 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1652 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1653 pktbuf_destroy(ofconn
->pktbuf
);
1658 ofconn_run(struct ofconn
*ofconn
)
1660 struct ofproto
*p
= ofconn
->ofproto
;
1664 if (ofconn
->discovery
) {
1665 char *controller_name
;
1666 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1667 discovery_question_connectivity(ofconn
->discovery
);
1669 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1670 if (controller_name
) {
1671 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1672 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1675 rconn_disconnect(ofconn
->rconn
);
1680 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1681 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1684 rconn_run(ofconn
->rconn
);
1686 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1687 /* Limit the number of iterations to prevent other tasks from
1689 for (iteration
= 0; iteration
< 50; iteration
++) {
1690 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1695 fail_open_maybe_recover(p
->fail_open
);
1697 handle_openflow(ofconn
, of_msg
);
1698 ofpbuf_delete(of_msg
);
1702 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1703 ofconn_destroy(ofconn
);
1708 ofconn_wait(struct ofconn
*ofconn
)
1712 if (ofconn
->discovery
) {
1713 discovery_wait(ofconn
->discovery
);
1715 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1716 pinsched_wait(ofconn
->schedulers
[i
]);
1718 rconn_run_wait(ofconn
->rconn
);
1719 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1720 rconn_recv_wait(ofconn
->rconn
);
1722 COVERAGE_INC(ofproto_ofconn_stuck
);
1726 /* Returns true if 'ofconn' should receive asynchronous messages. */
1728 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1730 if (ofconn
->type
== OFCONN_PRIMARY
) {
1731 /* Primary controllers always get asynchronous messages unless they
1732 * have configured themselves as "slaves". */
1733 return ofconn
->role
!= NX_ROLE_SLAVE
;
1735 /* Service connections don't get asynchronous messages unless they have
1736 * explicitly asked for them by setting a nonzero miss send length. */
1737 return ofconn
->miss_send_len
> 0;
1741 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1742 * and 'target', suitable for use in log messages for identifying the
1745 * The name is dynamically allocated. The caller should free it (with free())
1746 * when it is no longer needed. */
1748 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1750 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1754 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1758 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1759 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1763 *s
= pinsched_create(rate
, burst
,
1764 ofconn
->ofproto
->switch_status
);
1766 pinsched_set_limits(*s
, rate
, burst
);
1769 pinsched_destroy(*s
);
1776 ofservice_reconfigure(struct ofservice
*ofservice
,
1777 const struct ofproto_controller
*c
)
1779 ofservice
->probe_interval
= c
->probe_interval
;
1780 ofservice
->rate_limit
= c
->rate_limit
;
1781 ofservice
->burst_limit
= c
->burst_limit
;
1784 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1785 * positive errno value. */
1787 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1789 struct ofservice
*ofservice
;
1790 struct pvconn
*pvconn
;
1793 error
= pvconn_open(c
->target
, &pvconn
);
1798 ofservice
= xzalloc(sizeof *ofservice
);
1799 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1800 hash_string(c
->target
, 0));
1801 ofservice
->pvconn
= pvconn
;
1803 ofservice_reconfigure(ofservice
, c
);
1809 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1811 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1812 pvconn_close(ofservice
->pvconn
);
1816 /* Finds and returns the ofservice within 'ofproto' that has the given
1817 * 'target', or a null pointer if none exists. */
1818 static struct ofservice
*
1819 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1821 struct ofservice
*ofservice
;
1823 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1824 &ofproto
->services
) {
1825 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1832 /* Caller is responsible for initializing the 'cr' member of the returned
1834 static struct rule
*
1835 rule_create(struct ofproto
*ofproto
, struct rule
*super
,
1836 const union ofp_action
*actions
, size_t n_actions
,
1837 uint16_t idle_timeout
, uint16_t hard_timeout
,
1838 ovs_be64 flow_cookie
, bool send_flow_removed
)
1840 struct rule
*rule
= xzalloc(sizeof *rule
);
1841 rule
->idle_timeout
= idle_timeout
;
1842 rule
->hard_timeout
= hard_timeout
;
1843 rule
->flow_cookie
= flow_cookie
;
1844 rule
->used
= rule
->created
= time_msec();
1845 rule
->send_flow_removed
= send_flow_removed
;
1846 rule
->super
= super
;
1848 list_push_back(&super
->list
, &rule
->list
);
1850 list_init(&rule
->list
);
1852 if (n_actions
> 0) {
1853 rule
->n_actions
= n_actions
;
1854 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1856 netflow_flow_clear(&rule
->nf_flow
);
1857 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->created
);
1862 static struct rule
*
1863 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1865 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1869 rule_free(struct rule
*rule
)
1871 free(rule
->actions
);
1872 free(rule
->odp_actions
);
1876 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1877 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1878 * through all of its subrules and revalidates them, destroying any that no
1879 * longer has a super-rule (which is probably all of them).
1881 * Before calling this function, the caller must make have removed 'rule' from
1882 * the classifier. If 'rule' is an exact-match rule, the caller is also
1883 * responsible for ensuring that it has been uninstalled from the datapath. */
1885 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1888 struct rule
*subrule
, *next
;
1889 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
1890 revalidate_rule(ofproto
, subrule
);
1893 list_remove(&rule
->list
);
1899 rule_has_out_port(const struct rule
*rule
, ovs_be16 out_port
)
1901 const union ofp_action
*oa
;
1902 struct actions_iterator i
;
1904 if (out_port
== htons(OFPP_NONE
)) {
1907 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1908 oa
= actions_next(&i
)) {
1909 if (action_outputs_to_port(oa
, out_port
)) {
1916 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1917 * 'packet', which arrived on 'in_port'.
1919 * Takes ownership of 'packet'. */
1921 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
1922 const union odp_action
*actions
, size_t n_actions
,
1923 struct ofpbuf
*packet
)
1925 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
1926 /* As an optimization, avoid a round-trip from userspace to kernel to
1927 * userspace. This also avoids possibly filling up kernel packet
1928 * buffers along the way. */
1929 struct odp_msg
*msg
;
1931 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
1932 msg
->type
= _ODPL_ACTION_NR
;
1933 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
1934 msg
->port
= in_port
;
1936 msg
->arg
= actions
[0].controller
.arg
;
1938 send_packet_in(ofproto
, packet
);
1944 error
= dpif_execute(ofproto
->dpif
, actions
, n_actions
, packet
);
1945 ofpbuf_delete(packet
);
1950 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1951 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1952 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1954 * The flow that 'packet' actually contains does not need to actually match
1955 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1956 * the packet and byte counters for 'rule' will be credited for the packet sent
1957 * out whether or not the packet actually matches 'rule'.
1959 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1960 * the caller must already have accurately composed ODP actions for it given
1961 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1962 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1963 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1964 * actions and apply them to 'packet'.
1966 * Takes ownership of 'packet'. */
1968 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1969 struct ofpbuf
*packet
, const struct flow
*flow
)
1971 const union odp_action
*actions
;
1972 struct odp_flow_stats stats
;
1974 struct odp_actions a
;
1976 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1978 /* Grab or compose the ODP actions.
1980 * The special case for an exact-match 'rule' where 'flow' is not the
1981 * rule's flow is important to avoid, e.g., sending a packet out its input
1982 * port simply because the ODP actions were composed for the wrong
1984 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
1985 struct rule
*super
= rule
->super
? rule
->super
: rule
;
1986 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
1987 packet
, &a
, NULL
, 0, NULL
)) {
1988 ofpbuf_delete(packet
);
1991 actions
= a
.actions
;
1992 n_actions
= a
.n_actions
;
1994 actions
= rule
->odp_actions
;
1995 n_actions
= rule
->n_odp_actions
;
1998 /* Execute the ODP actions. */
1999 flow_extract_stats(flow
, packet
, &stats
);
2000 if (execute_odp_actions(ofproto
, flow
->in_port
,
2001 actions
, n_actions
, packet
)) {
2002 update_stats(ofproto
, rule
, &stats
);
2003 rule
->used
= time_msec();
2004 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->used
);
2008 /* Inserts 'rule' into 'p''s flow table.
2010 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2011 * actions on it and credits the statistics for sending the packet to 'rule'.
2012 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2015 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
2018 struct rule
*displaced_rule
;
2020 /* Insert the rule in the classifier. */
2021 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2022 if (!rule
->cr
.wc
.wildcards
) {
2023 rule_make_actions(p
, rule
, packet
);
2026 /* Send the packet and credit it to the rule. */
2029 flow_extract(packet
, 0, in_port
, &flow
);
2030 rule_execute(p
, rule
, packet
, &flow
);
2033 /* Install the rule in the datapath only after sending the packet, to
2034 * avoid packet reordering. */
2035 if (rule
->cr
.wc
.wildcards
) {
2036 COVERAGE_INC(ofproto_add_wc_flow
);
2037 p
->need_revalidate
= true;
2039 rule_install(p
, rule
, displaced_rule
);
2042 /* Free the rule that was displaced, if any. */
2043 if (displaced_rule
) {
2044 rule_destroy(p
, displaced_rule
);
2048 static struct rule
*
2049 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
2050 const struct flow
*flow
)
2052 struct rule
*subrule
= rule_create(ofproto
, rule
, NULL
, 0,
2053 rule
->idle_timeout
, rule
->hard_timeout
,
2055 COVERAGE_INC(ofproto_subrule_create
);
2056 cls_rule_init_exact(flow
, (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
2057 : rule
->cr
.priority
),
2060 if (classifier_insert(&ofproto
->cls
, &subrule
->cr
)) {
2068 /* Remove 'rule' from 'ofproto' and free up the associated memory:
2070 * - If 'rule' was installed in the datapath, uninstalls it and updates
2071 * 'rule''s statistics (or its super-rule's statistics, if it is a
2072 * subrule), via rule_uninstall().
2074 * - Removes 'rule' from the classifier.
2076 * - If 'rule' is a super-rule that has subrules, revalidates (and possibly
2077 * uninstalls and destroys) its subrules, via rule_destroy().
2080 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2082 if (rule
->cr
.wc
.wildcards
) {
2083 COVERAGE_INC(ofproto_del_wc_flow
);
2084 ofproto
->need_revalidate
= true;
2086 rule_uninstall(ofproto
, rule
);
2088 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2089 rule_destroy(ofproto
, rule
);
2092 /* Returns true if the actions changed, false otherwise. */
2094 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
2095 const struct ofpbuf
*packet
)
2097 const struct rule
*super
;
2098 struct odp_actions a
;
2101 assert(!rule
->cr
.wc
.wildcards
);
2103 super
= rule
->super
? rule
->super
: rule
;
2105 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
2106 packet
, &a
, &rule
->tags
, &rule
->may_install
,
2107 &rule
->nf_flow
.output_iface
);
2109 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2110 if (rule
->n_odp_actions
!= a
.n_actions
2111 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
2112 COVERAGE_INC(ofproto_odp_unchanged
);
2113 free(rule
->odp_actions
);
2114 rule
->n_odp_actions
= a
.n_actions
;
2115 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
2123 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
2124 struct odp_flow_put
*put
)
2126 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2127 odp_flow_key_from_flow(&put
->flow
.key
, &rule
->cr
.flow
);
2128 put
->flow
.actions
= rule
->odp_actions
;
2129 put
->flow
.n_actions
= rule
->n_odp_actions
;
2130 put
->flow
.flags
= 0;
2132 return dpif_flow_put(ofproto
->dpif
, put
);
2136 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
2138 assert(!rule
->cr
.wc
.wildcards
);
2140 if (rule
->may_install
) {
2141 struct odp_flow_put put
;
2142 if (!do_put_flow(p
, rule
,
2143 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
2145 rule
->installed
= true;
2146 if (displaced_rule
) {
2147 update_stats(p
, displaced_rule
, &put
.flow
.stats
);
2148 rule_post_uninstall(p
, displaced_rule
);
2151 } else if (displaced_rule
) {
2152 rule_uninstall(p
, displaced_rule
);
2157 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
2159 if (rule
->installed
) {
2160 struct odp_flow_put put
;
2161 COVERAGE_INC(ofproto_dp_missed
);
2162 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
2164 rule_install(ofproto
, rule
, NULL
);
2169 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
2171 bool actions_changed
;
2172 uint16_t new_out_iface
, old_out_iface
;
2174 old_out_iface
= rule
->nf_flow
.output_iface
;
2175 actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
2177 if (rule
->may_install
) {
2178 if (rule
->installed
) {
2179 if (actions_changed
) {
2180 struct odp_flow_put put
;
2181 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
2182 | ODPPF_ZERO_STATS
, &put
);
2183 update_stats(ofproto
, rule
, &put
.flow
.stats
);
2185 /* Temporarily set the old output iface so that NetFlow
2186 * messages have the correct output interface for the old
2188 new_out_iface
= rule
->nf_flow
.output_iface
;
2189 rule
->nf_flow
.output_iface
= old_out_iface
;
2190 rule_post_uninstall(ofproto
, rule
);
2191 rule
->nf_flow
.output_iface
= new_out_iface
;
2194 rule_install(ofproto
, rule
, NULL
);
2197 rule_uninstall(ofproto
, rule
);
2202 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
2204 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
2206 if (ofproto
->ofhooks
->account_flow_cb
2207 && total_bytes
> rule
->accounted_bytes
)
2209 ofproto
->ofhooks
->account_flow_cb(
2210 &rule
->cr
.flow
, rule
->tags
, rule
->odp_actions
, rule
->n_odp_actions
,
2211 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
2212 rule
->accounted_bytes
= total_bytes
;
2216 /* 'rule' must be an exact-match rule in 'p'.
2218 * If 'rule' is installed in the datapath, uninstalls it and updates's
2219 * statistics. If 'rule' is a subrule, the statistics that are updated are
2220 * actually its super-rule's statistics; otherwise 'rule''s own statistics are
2223 * If 'rule' is not installed, this function has no effect. */
2225 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
2227 assert(!rule
->cr
.wc
.wildcards
);
2228 if (rule
->installed
) {
2229 struct odp_flow odp_flow
;
2231 odp_flow_key_from_flow(&odp_flow
.key
, &rule
->cr
.flow
);
2232 odp_flow
.actions
= NULL
;
2233 odp_flow
.n_actions
= 0;
2235 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2236 update_stats(p
, rule
, &odp_flow
.stats
);
2238 rule
->installed
= false;
2240 rule_post_uninstall(p
, rule
);
2245 is_controller_rule(struct rule
*rule
)
2247 /* If the only action is send to the controller then don't report
2248 * NetFlow expiration messages since it is just part of the control
2249 * logic for the network and not real traffic. */
2253 && rule
->super
->n_actions
== 1
2254 && action_outputs_to_port(&rule
->super
->actions
[0],
2255 htons(OFPP_CONTROLLER
)));
2259 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
2261 struct rule
*super
= rule
->super
;
2263 rule_account(ofproto
, rule
, 0);
2265 if (ofproto
->netflow
&& !is_controller_rule(rule
)) {
2266 struct ofexpired expired
;
2267 expired
.flow
= rule
->cr
.flow
;
2268 expired
.packet_count
= rule
->packet_count
;
2269 expired
.byte_count
= rule
->byte_count
;
2270 expired
.used
= rule
->used
;
2271 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
2274 super
->packet_count
+= rule
->packet_count
;
2275 super
->byte_count
+= rule
->byte_count
;
2277 /* Reset counters to prevent double counting if the rule ever gets
2279 rule
->packet_count
= 0;
2280 rule
->byte_count
= 0;
2281 rule
->accounted_bytes
= 0;
2283 netflow_flow_clear(&rule
->nf_flow
);
2288 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2289 struct rconn_packet_counter
*counter
)
2291 update_openflow_length(msg
);
2292 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2298 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2301 struct ofpbuf
*buf
= make_ofp_error_msg(error
, oh
);
2303 COVERAGE_INC(ofproto_error
);
2304 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2309 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2311 opp
->port_no
= htons(opp
->port_no
);
2312 opp
->config
= htonl(opp
->config
);
2313 opp
->state
= htonl(opp
->state
);
2314 opp
->curr
= htonl(opp
->curr
);
2315 opp
->advertised
= htonl(opp
->advertised
);
2316 opp
->supported
= htonl(opp
->supported
);
2317 opp
->peer
= htonl(opp
->peer
);
2321 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2323 struct ofp_header
*rq
= oh
;
2324 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2329 handle_features_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2331 struct ofp_switch_features
*osf
;
2333 struct ofport
*port
;
2335 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2336 osf
->datapath_id
= htonll(ofconn
->ofproto
->datapath_id
);
2337 osf
->n_buffers
= htonl(pktbuf_capacity());
2339 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2340 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2341 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2342 (1u << OFPAT_SET_VLAN_VID
) |
2343 (1u << OFPAT_SET_VLAN_PCP
) |
2344 (1u << OFPAT_STRIP_VLAN
) |
2345 (1u << OFPAT_SET_DL_SRC
) |
2346 (1u << OFPAT_SET_DL_DST
) |
2347 (1u << OFPAT_SET_NW_SRC
) |
2348 (1u << OFPAT_SET_NW_DST
) |
2349 (1u << OFPAT_SET_NW_TOS
) |
2350 (1u << OFPAT_SET_TP_SRC
) |
2351 (1u << OFPAT_SET_TP_DST
) |
2352 (1u << OFPAT_ENQUEUE
));
2354 HMAP_FOR_EACH (port
, hmap_node
, &ofconn
->ofproto
->ports
) {
2355 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2358 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2363 handle_get_config_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2366 struct ofp_switch_config
*osc
;
2370 /* Figure out flags. */
2371 dpif_get_drop_frags(ofconn
->ofproto
->dpif
, &drop_frags
);
2372 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2375 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2376 osc
->flags
= htons(flags
);
2377 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2378 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2384 handle_set_config(struct ofconn
*ofconn
, struct ofp_switch_config
*osc
)
2389 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2393 flags
= ntohs(osc
->flags
);
2395 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2396 switch (flags
& OFPC_FRAG_MASK
) {
2397 case OFPC_FRAG_NORMAL
:
2398 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, false);
2400 case OFPC_FRAG_DROP
:
2401 dpif_set_drop_frags(ofconn
->ofproto
->dpif
, true);
2404 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2410 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2416 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2418 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2419 a
->controller
.arg
= max_len
;
2422 struct action_xlate_ctx
{
2424 struct flow flow
; /* Flow to which these actions correspond. */
2425 int recurse
; /* Recursion level, via xlate_table_action. */
2426 struct ofproto
*ofproto
;
2427 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2428 * null pointer if we are revalidating
2429 * without a packet to refer to. */
2432 struct odp_actions
*out
; /* Datapath actions. */
2433 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
2434 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2435 * be reassessed for every packet. */
2436 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2439 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2440 * flow translation. */
2441 #define MAX_RESUBMIT_RECURSION 8
2443 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2444 struct action_xlate_ctx
*ctx
);
2447 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2449 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2452 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2453 /* Forwarding disabled on port. */
2458 * We don't have an ofport record for this port, but it doesn't hurt to
2459 * allow forwarding to it anyhow. Maybe such a port will appear later
2460 * and we're pre-populating the flow table.
2464 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2465 ctx
->nf_output_iface
= port
;
2468 static struct rule
*
2469 lookup_valid_rule(struct ofproto
*ofproto
, const struct flow
*flow
)
2472 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
,
2475 /* The rule we found might not be valid, since we could be in need of
2476 * revalidation. If it is not valid, don't return it. */
2479 && ofproto
->need_revalidate
2480 && !revalidate_rule(ofproto
, rule
)) {
2481 COVERAGE_INC(ofproto_invalidated
);
2489 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2491 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2492 uint16_t old_in_port
;
2495 /* Look up a flow with 'in_port' as the input port. Then restore the
2496 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2497 * have surprising behavior). */
2498 old_in_port
= ctx
->flow
.in_port
;
2499 ctx
->flow
.in_port
= in_port
;
2500 rule
= lookup_valid_rule(ctx
->ofproto
, &ctx
->flow
);
2501 ctx
->flow
.in_port
= old_in_port
;
2509 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2513 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2515 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2516 MAX_RESUBMIT_RECURSION
);
2521 flood_packets(struct ofproto
*ofproto
, uint16_t odp_in_port
, uint32_t mask
,
2522 uint16_t *nf_output_iface
, struct odp_actions
*actions
)
2524 struct ofport
*ofport
;
2526 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
2527 uint16_t odp_port
= ofport
->odp_port
;
2528 if (odp_port
!= odp_in_port
&& !(ofport
->opp
.config
& mask
)) {
2529 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= odp_port
;
2532 *nf_output_iface
= NF_OUT_FLOOD
;
2536 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2537 uint16_t port
, uint16_t max_len
)
2540 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2542 ctx
->nf_output_iface
= NF_OUT_DROP
;
2546 add_output_action(ctx
, ctx
->flow
.in_port
);
2549 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2552 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2553 ctx
->out
, ctx
->tags
,
2554 &ctx
->nf_output_iface
,
2555 ctx
->ofproto
->aux
)) {
2556 COVERAGE_INC(ofproto_uninstallable
);
2557 ctx
->may_set_up_flow
= false;
2561 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, OFPPC_NO_FLOOD
,
2562 &ctx
->nf_output_iface
, ctx
->out
);
2565 flood_packets(ctx
->ofproto
, ctx
->flow
.in_port
, 0,
2566 &ctx
->nf_output_iface
, ctx
->out
);
2568 case OFPP_CONTROLLER
:
2569 add_controller_action(ctx
->out
, max_len
);
2572 add_output_action(ctx
, ODPP_LOCAL
);
2575 odp_port
= ofp_port_to_odp_port(port
);
2576 if (odp_port
!= ctx
->flow
.in_port
) {
2577 add_output_action(ctx
, odp_port
);
2582 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2583 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2584 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2585 ctx
->nf_output_iface
= prev_nf_output_iface
;
2586 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2587 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2588 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2593 xlate_output_action(struct action_xlate_ctx
*ctx
,
2594 const struct ofp_action_output
*oao
)
2596 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2599 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2600 * optimization, because we're going to add another action that sets the
2601 * priority immediately after, or because there are no actions following the
2604 remove_pop_action(struct action_xlate_ctx
*ctx
)
2606 size_t n
= ctx
->out
->n_actions
;
2607 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2608 ctx
->out
->n_actions
--;
2613 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2614 const struct ofp_action_enqueue
*oae
)
2616 uint16_t ofp_port
, odp_port
;
2620 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2623 /* Fall back to ordinary output action. */
2624 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2628 /* Figure out ODP output port. */
2629 ofp_port
= ntohs(oae
->port
);
2630 if (ofp_port
!= OFPP_IN_PORT
) {
2631 odp_port
= ofp_port_to_odp_port(ofp_port
);
2633 odp_port
= ctx
->flow
.in_port
;
2636 /* Add ODP actions. */
2637 remove_pop_action(ctx
);
2638 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2640 add_output_action(ctx
, odp_port
);
2641 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2643 /* Update NetFlow output port. */
2644 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2645 ctx
->nf_output_iface
= odp_port
;
2646 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2647 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2652 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2653 const struct nx_action_set_queue
*nasq
)
2658 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2661 /* Couldn't translate queue to a priority, so ignore. A warning
2662 * has already been logged. */
2666 remove_pop_action(ctx
);
2667 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2672 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2673 const struct nx_action_header
*nah
)
2675 const struct nx_action_resubmit
*nar
;
2676 const struct nx_action_set_tunnel
*nast
;
2677 const struct nx_action_set_queue
*nasq
;
2678 union odp_action
*oa
;
2679 int subtype
= ntohs(nah
->subtype
);
2681 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2683 case NXAST_RESUBMIT
:
2684 nar
= (const struct nx_action_resubmit
*) nah
;
2685 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2688 case NXAST_SET_TUNNEL
:
2689 nast
= (const struct nx_action_set_tunnel
*) nah
;
2690 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2691 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2694 case NXAST_DROP_SPOOFED_ARP
:
2695 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2696 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2700 case NXAST_SET_QUEUE
:
2701 nasq
= (const struct nx_action_set_queue
*) nah
;
2702 xlate_set_queue_action(ctx
, nasq
);
2705 case NXAST_POP_QUEUE
:
2706 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2709 /* If you add a new action here that modifies flow data, don't forget to
2710 * update the flow key in ctx->flow at the same time. */
2713 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2719 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2720 struct action_xlate_ctx
*ctx
)
2722 struct actions_iterator iter
;
2723 const union ofp_action
*ia
;
2724 const struct ofport
*port
;
2726 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2727 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2728 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2729 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2730 /* Drop this flow. */
2734 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2735 uint16_t type
= ntohs(ia
->type
);
2736 union odp_action
*oa
;
2740 xlate_output_action(ctx
, &ia
->output
);
2743 case OFPAT_SET_VLAN_VID
:
2744 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_TCI
);
2745 oa
->dl_tci
.tci
= ia
->vlan_vid
.vlan_vid
;
2746 oa
->dl_tci
.tci
|= htons(ctx
->flow
.dl_vlan_pcp
<< VLAN_PCP_SHIFT
);
2747 ctx
->flow
.dl_vlan
= ia
->vlan_vid
.vlan_vid
;
2750 case OFPAT_SET_VLAN_PCP
:
2751 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_TCI
);
2752 oa
->dl_tci
.tci
= htons(ia
->vlan_pcp
.vlan_pcp
<< VLAN_PCP_SHIFT
);
2753 oa
->dl_tci
.tci
|= ctx
->flow
.dl_vlan
;
2754 ctx
->flow
.dl_vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2757 case OFPAT_STRIP_VLAN
:
2758 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2759 ctx
->flow
.dl_vlan
= htons(OFP_VLAN_NONE
);
2760 ctx
->flow
.dl_vlan_pcp
= 0;
2763 case OFPAT_SET_DL_SRC
:
2764 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2765 memcpy(oa
->dl_addr
.dl_addr
,
2766 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2767 memcpy(ctx
->flow
.dl_src
,
2768 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2771 case OFPAT_SET_DL_DST
:
2772 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2773 memcpy(oa
->dl_addr
.dl_addr
,
2774 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2775 memcpy(ctx
->flow
.dl_dst
,
2776 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2779 case OFPAT_SET_NW_SRC
:
2780 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2781 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2784 case OFPAT_SET_NW_DST
:
2785 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2786 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2789 case OFPAT_SET_NW_TOS
:
2790 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2791 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2794 case OFPAT_SET_TP_SRC
:
2795 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2796 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2799 case OFPAT_SET_TP_DST
:
2800 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2801 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2805 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2809 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2813 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2820 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2821 const struct flow
*flow
, struct ofproto
*ofproto
,
2822 const struct ofpbuf
*packet
,
2823 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2824 uint16_t *nf_output_iface
)
2826 tag_type no_tags
= 0;
2827 struct action_xlate_ctx ctx
;
2828 COVERAGE_INC(ofproto_ofp2odp
);
2829 odp_actions_init(out
);
2832 ctx
.ofproto
= ofproto
;
2833 ctx
.packet
= packet
;
2835 ctx
.tags
= tags
? tags
: &no_tags
;
2836 ctx
.may_set_up_flow
= true;
2837 ctx
.nf_output_iface
= NF_OUT_DROP
;
2838 do_xlate_actions(in
, n_in
, &ctx
);
2839 remove_pop_action(&ctx
);
2841 /* Check with in-band control to see if we're allowed to set up this
2843 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2844 ctx
.may_set_up_flow
= false;
2847 if (may_set_up_flow
) {
2848 *may_set_up_flow
= ctx
.may_set_up_flow
;
2850 if (nf_output_iface
) {
2851 *nf_output_iface
= ctx
.nf_output_iface
;
2853 if (odp_actions_overflow(out
)) {
2854 COVERAGE_INC(odp_overflow
);
2855 odp_actions_init(out
);
2856 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
2861 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2862 * error message code (composed with ofp_mkerr()) for the caller to propagate
2863 * upward. Otherwise, returns 0.
2865 * The log message mentions 'msg_type'. */
2867 reject_slave_controller(struct ofconn
*ofconn
, const const char *msg_type
)
2869 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
2870 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2871 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
2874 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
2881 handle_packet_out(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2883 struct ofproto
*p
= ofconn
->ofproto
;
2884 struct ofp_packet_out
*opo
;
2885 struct ofpbuf payload
, *buffer
;
2886 struct odp_actions actions
;
2892 error
= reject_slave_controller(ofconn
, "OFPT_PACKET_OUT");
2897 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2901 opo
= (struct ofp_packet_out
*) oh
;
2903 COVERAGE_INC(ofproto_packet_out
);
2904 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2905 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2907 if (error
|| !buffer
) {
2915 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2916 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2917 &flow
, p
, &payload
, &actions
, NULL
, NULL
, NULL
);
2919 dpif_execute(p
->dpif
, actions
.actions
, actions
.n_actions
, &payload
);
2921 ofpbuf_delete(buffer
);
2927 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2928 uint32_t config
, uint32_t mask
)
2930 mask
&= config
^ port
->opp
.config
;
2931 if (mask
& OFPPC_PORT_DOWN
) {
2932 if (config
& OFPPC_PORT_DOWN
) {
2933 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2935 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2938 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | \
2939 OFPPC_NO_FWD | OFPPC_NO_FLOOD)
2940 if (mask
& REVALIDATE_BITS
) {
2941 COVERAGE_INC(ofproto_costly_flags
);
2942 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2943 p
->need_revalidate
= true;
2945 #undef REVALIDATE_BITS
2946 if (mask
& OFPPC_NO_PACKET_IN
) {
2947 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2952 handle_port_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2954 struct ofproto
*p
= ofconn
->ofproto
;
2955 const struct ofp_port_mod
*opm
;
2956 struct ofport
*port
;
2959 error
= reject_slave_controller(ofconn
, "OFPT_PORT_MOD");
2963 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2967 opm
= (struct ofp_port_mod
*) oh
;
2969 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2971 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2972 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2973 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2975 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2976 if (opm
->advertise
) {
2977 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2983 static struct ofpbuf
*
2984 make_ofp_stats_reply(ovs_be32 xid
, ovs_be16 type
, size_t body_len
)
2986 struct ofp_stats_reply
*osr
;
2989 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
2990 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
2992 osr
->flags
= htons(0);
2996 static struct ofpbuf
*
2997 start_ofp_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
2999 return make_ofp_stats_reply(request
->header
.xid
, request
->type
, body_len
);
3003 append_ofp_stats_reply(size_t nbytes
, struct ofconn
*ofconn
,
3004 struct ofpbuf
**msgp
)
3006 struct ofpbuf
*msg
= *msgp
;
3007 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3008 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3009 struct ofp_stats_reply
*reply
= msg
->data
;
3010 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3011 *msgp
= make_ofp_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3012 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3014 return ofpbuf_put_uninit(*msgp
, nbytes
);
3018 handle_desc_stats_request(struct ofconn
*ofconn
,
3019 struct ofp_stats_request
*request
)
3021 struct ofproto
*p
= ofconn
->ofproto
;
3022 struct ofp_desc_stats
*ods
;
3025 msg
= start_ofp_stats_reply(request
, sizeof *ods
);
3026 ods
= append_ofp_stats_reply(sizeof *ods
, ofconn
, &msg
);
3027 memset(ods
, 0, sizeof *ods
);
3028 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3029 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3030 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3031 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3032 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3033 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3039 handle_table_stats_request(struct ofconn
*ofconn
,
3040 struct ofp_stats_request
*request
)
3042 struct ofproto
*p
= ofconn
->ofproto
;
3043 struct ofp_table_stats
*ots
;
3048 msg
= start_ofp_stats_reply(request
, sizeof *ots
* 2);
3050 /* Count rules other than subrules. */
3051 n_rules
= classifier_count(&p
->cls
);
3052 CLASSIFIER_FOR_EACH_EXACT_RULE (rule
, cr
, &p
->cls
) {
3058 /* Classifier table. */
3059 ots
= append_ofp_stats_reply(sizeof *ots
, ofconn
, &msg
);
3060 memset(ots
, 0, sizeof *ots
);
3061 strcpy(ots
->name
, "classifier");
3062 ots
->wildcards
= (ofconn
->flow_format
== NXFF_OPENFLOW10
3063 ? htonl(OFPFW_ALL
) : htonl(OVSFW_ALL
));
3064 ots
->max_entries
= htonl(1024 * 1024); /* An arbitrary big number. */
3065 ots
->active_count
= htonl(n_rules
);
3066 ots
->lookup_count
= htonll(0); /* XXX */
3067 ots
->matched_count
= htonll(0); /* XXX */
3069 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3074 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3075 struct ofpbuf
**msgp
)
3077 struct netdev_stats stats
;
3078 struct ofp_port_stats
*ops
;
3080 /* Intentionally ignore return value, since errors will set
3081 * 'stats' to all-1s, which is correct for OpenFlow, and
3082 * netdev_get_stats() will log errors. */
3083 netdev_get_stats(port
->netdev
, &stats
);
3085 ops
= append_ofp_stats_reply(sizeof *ops
, ofconn
, msgp
);
3086 ops
->port_no
= htons(port
->opp
.port_no
);
3087 memset(ops
->pad
, 0, sizeof ops
->pad
);
3088 ops
->rx_packets
= htonll(stats
.rx_packets
);
3089 ops
->tx_packets
= htonll(stats
.tx_packets
);
3090 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3091 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3092 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3093 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3094 ops
->rx_errors
= htonll(stats
.rx_errors
);
3095 ops
->tx_errors
= htonll(stats
.tx_errors
);
3096 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3097 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3098 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3099 ops
->collisions
= htonll(stats
.collisions
);
3103 handle_port_stats_request(struct ofconn
*ofconn
, struct ofp_stats_request
*osr
,
3106 struct ofproto
*p
= ofconn
->ofproto
;
3107 struct ofp_port_stats_request
*psr
;
3108 struct ofp_port_stats
*ops
;
3110 struct ofport
*port
;
3112 if (arg_size
!= sizeof *psr
) {
3113 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3115 psr
= (struct ofp_port_stats_request
*) osr
->body
;
3117 msg
= start_ofp_stats_reply(osr
, sizeof *ops
* 16);
3118 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3119 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3121 append_port_stat(port
, ofconn
, &msg
);
3124 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3125 append_port_stat(port
, ofconn
, &msg
);
3129 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3133 struct flow_stats_cbdata
{
3134 struct ofconn
*ofconn
;
3139 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3140 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3141 * returned statistic include statistics for all of 'rule''s subrules. */
3143 query_stats(struct ofproto
*p
, struct rule
*rule
,
3144 uint64_t *packet_countp
, uint64_t *byte_countp
)
3146 uint64_t packet_count
, byte_count
;
3147 struct rule
*subrule
;
3148 struct odp_flow
*odp_flows
;
3151 /* Start from historical data for 'rule' itself that are no longer tracked
3152 * by the datapath. This counts, for example, subrules that have
3154 packet_count
= rule
->packet_count
;
3155 byte_count
= rule
->byte_count
;
3157 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3158 * wildcarded then on all of its subrules.
3160 * Also, add any statistics that are not tracked by the datapath for each
3161 * subrule. This includes, for example, statistics for packets that were
3162 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3164 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
3165 odp_flows
= xzalloc(n_odp_flows
* sizeof *odp_flows
);
3166 if (rule
->cr
.wc
.wildcards
) {
3168 LIST_FOR_EACH (subrule
, list
, &rule
->list
) {
3169 odp_flow_key_from_flow(&odp_flows
[i
++].key
, &subrule
->cr
.flow
);
3170 packet_count
+= subrule
->packet_count
;
3171 byte_count
+= subrule
->byte_count
;
3174 odp_flow_key_from_flow(&odp_flows
[0].key
, &rule
->cr
.flow
);
3177 /* Fetch up-to-date statistics from the datapath and add them in. */
3178 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3180 for (i
= 0; i
< n_odp_flows
; i
++) {
3181 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3182 packet_count
+= odp_flow
->stats
.n_packets
;
3183 byte_count
+= odp_flow
->stats
.n_bytes
;
3188 /* Return the stats to the caller. */
3189 *packet_countp
= packet_count
;
3190 *byte_countp
= byte_count
;
3194 calc_flow_duration(long long int start
, ovs_be32
*sec
, ovs_be32
*nsec
)
3196 long long int msecs
= time_msec() - start
;
3197 *sec
= htonl(msecs
/ 1000);
3198 *nsec
= htonl((msecs
% 1000) * (1000 * 1000));
3202 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3204 struct rule
*rule
= rule_from_cls_rule(rule_
);
3205 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3206 struct ofp_flow_stats
*ofs
;
3207 uint64_t packet_count
, byte_count
;
3208 size_t act_len
, len
;
3210 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3214 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3215 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3217 query_stats(cbdata
->ofconn
->ofproto
, rule
, &packet_count
, &byte_count
);
3219 ofs
= append_ofp_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
3220 ofs
->length
= htons(len
);
3223 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3224 cbdata
->ofconn
->flow_format
, &ofs
->match
);
3225 calc_flow_duration(rule
->created
, &ofs
->duration_sec
, &ofs
->duration_nsec
);
3226 ofs
->cookie
= rule
->flow_cookie
;
3227 ofs
->priority
= htons(rule
->cr
.priority
);
3228 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3229 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3230 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3231 ofs
->packet_count
= htonll(packet_count
);
3232 ofs
->byte_count
= htonll(byte_count
);
3233 if (rule
->n_actions
> 0) {
3234 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3239 table_id_to_include(uint8_t table_id
)
3241 return table_id
== 0 || table_id
== 0xff ? CLS_INC_ALL
: 0;
3245 handle_flow_stats_request(struct ofconn
*ofconn
,
3246 const struct ofp_stats_request
*osr
, size_t arg_size
)
3248 struct ofp_flow_stats_request
*fsr
;
3249 struct flow_stats_cbdata cbdata
;
3250 struct cls_rule target
;
3252 if (arg_size
!= sizeof *fsr
) {
3253 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3255 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3257 COVERAGE_INC(ofproto_flows_req
);
3258 cbdata
.ofconn
= ofconn
;
3259 cbdata
.out_port
= fsr
->out_port
;
3260 cbdata
.msg
= start_ofp_stats_reply(osr
, 1024);
3261 cls_rule_from_match(&fsr
->match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3262 classifier_for_each_match(&ofconn
->ofproto
->cls
, &target
,
3263 table_id_to_include(fsr
->table_id
),
3264 flow_stats_cb
, &cbdata
);
3265 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3269 struct flow_stats_ds_cbdata
{
3270 struct ofproto
*ofproto
;
3275 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
3277 struct rule
*rule
= rule_from_cls_rule(rule_
);
3278 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
3279 struct ds
*results
= cbdata
->results
;
3280 struct ofp_match match
;
3281 uint64_t packet_count
, byte_count
;
3282 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3284 /* Don't report on subrules. */
3285 if (rule
->super
!= NULL
) {
3289 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3290 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3291 NXFF_OPENFLOW10
, &match
);
3293 ds_put_format(results
, "duration=%llds, ",
3294 (time_msec() - rule
->created
) / 1000);
3295 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3296 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3297 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3298 ofp_print_match(results
, &match
, true);
3300 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3302 ds_put_cstr(results
, "drop");
3304 ds_put_cstr(results
, "\n");
3307 /* Adds a pretty-printed description of all flows to 'results', including
3308 * those marked hidden by secchan (e.g., by in-band control). */
3310 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3312 struct ofp_match match
;
3313 struct cls_rule target
;
3314 struct flow_stats_ds_cbdata cbdata
;
3316 memset(&match
, 0, sizeof match
);
3317 match
.wildcards
= htonl(OVSFW_ALL
);
3320 cbdata
.results
= results
;
3322 cls_rule_from_match(&match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3323 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3324 flow_stats_ds_cb
, &cbdata
);
3327 struct aggregate_stats_cbdata
{
3328 struct ofproto
*ofproto
;
3330 uint64_t packet_count
;
3331 uint64_t byte_count
;
3336 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3338 struct rule
*rule
= rule_from_cls_rule(rule_
);
3339 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
3340 uint64_t packet_count
, byte_count
;
3342 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3346 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3348 cbdata
->packet_count
+= packet_count
;
3349 cbdata
->byte_count
+= byte_count
;
3354 query_aggregate_stats(struct ofproto
*ofproto
, struct cls_rule
*target
,
3355 ovs_be16 out_port
, uint8_t table_id
,
3356 struct ofp_aggregate_stats_reply
*oasr
)
3358 struct aggregate_stats_cbdata cbdata
;
3360 COVERAGE_INC(ofproto_agg_request
);
3361 cbdata
.ofproto
= ofproto
;
3362 cbdata
.out_port
= out_port
;
3363 cbdata
.packet_count
= 0;
3364 cbdata
.byte_count
= 0;
3366 classifier_for_each_match(&ofproto
->cls
, target
,
3367 table_id_to_include(table_id
),
3368 aggregate_stats_cb
, &cbdata
);
3370 oasr
->flow_count
= htonl(cbdata
.n_flows
);
3371 oasr
->packet_count
= htonll(cbdata
.packet_count
);
3372 oasr
->byte_count
= htonll(cbdata
.byte_count
);
3373 memset(oasr
->pad
, 0, sizeof oasr
->pad
);
3377 handle_aggregate_stats_request(struct ofconn
*ofconn
,
3378 const struct ofp_stats_request
*osr
,
3381 struct ofp_aggregate_stats_request
*request
;
3382 struct ofp_aggregate_stats_reply
*reply
;
3383 struct cls_rule target
;
3386 if (arg_size
!= sizeof *request
) {
3387 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3389 request
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3391 cls_rule_from_match(&request
->match
, 0, NXFF_OPENFLOW10
, 0, &target
);
3393 msg
= start_ofp_stats_reply(osr
, sizeof *reply
);
3394 reply
= append_ofp_stats_reply(sizeof *reply
, ofconn
, &msg
);
3395 query_aggregate_stats(ofconn
->ofproto
, &target
, request
->out_port
,
3396 request
->table_id
, reply
);
3397 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3401 struct queue_stats_cbdata
{
3402 struct ofconn
*ofconn
;
3403 struct ofport
*ofport
;
3408 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3409 const struct netdev_queue_stats
*stats
)
3411 struct ofp_queue_stats
*reply
;
3413 reply
= append_ofp_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3414 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3415 memset(reply
->pad
, 0, sizeof reply
->pad
);
3416 reply
->queue_id
= htonl(queue_id
);
3417 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3418 reply
->tx_packets
= htonll(stats
->tx_packets
);
3419 reply
->tx_errors
= htonll(stats
->tx_errors
);
3423 handle_queue_stats_dump_cb(uint32_t queue_id
,
3424 struct netdev_queue_stats
*stats
,
3427 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3429 put_queue_stats(cbdata
, queue_id
, stats
);
3433 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3434 struct queue_stats_cbdata
*cbdata
)
3436 cbdata
->ofport
= port
;
3437 if (queue_id
== OFPQ_ALL
) {
3438 netdev_dump_queue_stats(port
->netdev
,
3439 handle_queue_stats_dump_cb
, cbdata
);
3441 struct netdev_queue_stats stats
;
3443 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3444 put_queue_stats(cbdata
, queue_id
, &stats
);
3450 handle_queue_stats_request(struct ofconn
*ofconn
,
3451 const struct ofp_stats_request
*osr
,
3454 struct ofproto
*ofproto
= ofconn
->ofproto
;
3455 struct ofp_queue_stats_request
*qsr
;
3456 struct queue_stats_cbdata cbdata
;
3457 struct ofport
*port
;
3458 unsigned int port_no
;
3461 if (arg_size
!= sizeof *qsr
) {
3462 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3464 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3466 COVERAGE_INC(ofproto_queue_req
);
3468 cbdata
.ofconn
= ofconn
;
3469 cbdata
.msg
= start_ofp_stats_reply(osr
, 128);
3471 port_no
= ntohs(qsr
->port_no
);
3472 queue_id
= ntohl(qsr
->queue_id
);
3473 if (port_no
== OFPP_ALL
) {
3474 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3475 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3477 } else if (port_no
< ofproto
->max_ports
) {
3478 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3480 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3483 ofpbuf_delete(cbdata
.msg
);
3484 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3486 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3492 handle_stats_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3494 struct ofp_stats_request
*osr
;
3498 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3503 osr
= (struct ofp_stats_request
*) oh
;
3505 switch (ntohs(osr
->type
)) {
3507 return handle_desc_stats_request(ofconn
, osr
);
3510 return handle_flow_stats_request(ofconn
, osr
, arg_size
);
3512 case OFPST_AGGREGATE
:
3513 return handle_aggregate_stats_request(ofconn
, osr
, arg_size
);
3516 return handle_table_stats_request(ofconn
, osr
);
3519 return handle_port_stats_request(ofconn
, osr
, arg_size
);
3522 return handle_queue_stats_request(ofconn
, osr
, arg_size
);
3525 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3528 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3532 static long long int
3533 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3535 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3539 update_time(struct ofproto
*ofproto
, struct rule
*rule
,
3540 const struct odp_flow_stats
*stats
)
3542 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3543 if (used
> rule
->used
) {
3545 if (rule
->super
&& used
> rule
->super
->used
) {
3546 rule
->super
->used
= used
;
3548 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, used
);
3553 update_stats(struct ofproto
*ofproto
, struct rule
*rule
,
3554 const struct odp_flow_stats
*stats
)
3556 if (stats
->n_packets
) {
3557 update_time(ofproto
, rule
, stats
);
3558 rule
->packet_count
+= stats
->n_packets
;
3559 rule
->byte_count
+= stats
->n_bytes
;
3560 netflow_flow_update_flags(&rule
->nf_flow
, stats
->tcp_flags
);
3568 uint16_t idle_timeout
;
3569 uint16_t hard_timeout
;
3573 union ofp_action
*actions
;
3577 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3578 * in which no matching flow already exists in the flow table.
3580 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3581 * ofp_actions, to ofconn->ofproto's flow table. Returns 0 on success or an
3582 * OpenFlow error code as encoded by ofp_mkerr() on failure.
3584 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3587 add_flow(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3589 struct ofproto
*p
= ofconn
->ofproto
;
3590 struct ofpbuf
*packet
;
3595 if (fm
->flags
& OFPFF_CHECK_OVERLAP
3596 && classifier_rule_overlaps(&p
->cls
, &fm
->cr
)) {
3597 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3600 rule
= rule_create(p
, NULL
, fm
->actions
, fm
->n_actions
,
3601 fm
->idle_timeout
, fm
->hard_timeout
, fm
->cookie
,
3602 fm
->flags
& OFPFF_SEND_FLOW_REM
);
3606 if (fm
->buffer_id
!= UINT32_MAX
) {
3607 error
= pktbuf_retrieve(ofconn
->pktbuf
, fm
->buffer_id
,
3611 in_port
= UINT16_MAX
;
3614 rule_insert(p
, rule
, packet
, in_port
);
3618 static struct rule
*
3619 find_flow_strict(struct ofproto
*p
, const struct flow_mod
*fm
)
3621 return rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
, &fm
->cr
));
3625 send_buffered_packet(struct ofconn
*ofconn
,
3626 struct rule
*rule
, uint32_t buffer_id
)
3628 struct ofpbuf
*packet
;
3633 if (buffer_id
== UINT32_MAX
) {
3637 error
= pktbuf_retrieve(ofconn
->pktbuf
, buffer_id
, &packet
, &in_port
);
3642 flow_extract(packet
, 0, in_port
, &flow
);
3643 rule_execute(ofconn
->ofproto
, rule
, packet
, &flow
);
3648 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3650 struct modify_flows_cbdata
{
3651 struct ofproto
*ofproto
;
3652 const struct flow_mod
*fm
;
3656 static int modify_flow(struct ofproto
*, const struct flow_mod
*,
3658 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3660 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3661 * encoded by ofp_mkerr() on failure.
3663 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3666 modify_flows_loose(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3668 struct modify_flows_cbdata cbdata
;
3670 cbdata
.ofproto
= ofconn
->ofproto
;
3672 cbdata
.match
= NULL
;
3674 classifier_for_each_match(&ofconn
->ofproto
->cls
, &fm
->cr
, CLS_INC_ALL
,
3675 modify_flows_cb
, &cbdata
);
3677 /* This credits the packet to whichever flow happened to happened to
3678 * match last. That's weird. Maybe we should do a lookup for the
3679 * flow that actually matches the packet? Who knows. */
3680 send_buffered_packet(ofconn
, cbdata
.match
, fm
->buffer_id
);
3683 return add_flow(ofconn
, fm
);
3687 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3688 * code as encoded by ofp_mkerr() on failure.
3690 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3693 modify_flow_strict(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3695 struct ofproto
*p
= ofconn
->ofproto
;
3696 struct rule
*rule
= find_flow_strict(p
, fm
);
3697 if (rule
&& !rule_is_hidden(rule
)) {
3698 modify_flow(p
, fm
, rule
);
3699 return send_buffered_packet(ofconn
, rule
, fm
->buffer_id
);
3701 return add_flow(ofconn
, fm
);
3705 /* Callback for modify_flows_loose(). */
3707 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3709 struct rule
*rule
= rule_from_cls_rule(rule_
);
3710 struct modify_flows_cbdata
*cbdata
= cbdata_
;
3712 if (!rule_is_hidden(rule
)) {
3713 cbdata
->match
= rule
;
3714 modify_flow(cbdata
->ofproto
, cbdata
->fm
, rule
);
3718 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3719 * been identified as a flow in 'p''s flow table to be modified, by changing
3720 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3721 * ofp_action[] structures). */
3723 modify_flow(struct ofproto
*p
, const struct flow_mod
*fm
, struct rule
*rule
)
3725 size_t actions_len
= fm
->n_actions
* sizeof *rule
->actions
;
3727 rule
->flow_cookie
= fm
->cookie
;
3729 /* If the actions are the same, do nothing. */
3730 if (fm
->n_actions
== rule
->n_actions
3732 || !memcmp(fm
->actions
, rule
->actions
, actions_len
))) {
3736 /* Replace actions. */
3737 free(rule
->actions
);
3738 rule
->actions
= fm
->n_actions
? xmemdup(fm
->actions
, actions_len
) : NULL
;
3739 rule
->n_actions
= fm
->n_actions
;
3741 /* Make sure that the datapath gets updated properly. */
3742 if (rule
->cr
.wc
.wildcards
) {
3743 COVERAGE_INC(ofproto_mod_wc_flow
);
3744 p
->need_revalidate
= true;
3746 rule_update_actions(p
, rule
);
3752 /* OFPFC_DELETE implementation. */
3754 struct delete_flows_cbdata
{
3755 struct ofproto
*ofproto
;
3759 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
3760 static void delete_flow(struct ofproto
*, struct rule
*, ovs_be16 out_port
);
3762 /* Implements OFPFC_DELETE. */
3764 delete_flows_loose(struct ofproto
*p
, const struct flow_mod
*fm
)
3766 struct delete_flows_cbdata cbdata
;
3769 cbdata
.out_port
= htons(fm
->out_port
);
3771 classifier_for_each_match(&p
->cls
, &fm
->cr
, CLS_INC_ALL
,
3772 delete_flows_cb
, &cbdata
);
3775 /* Implements OFPFC_DELETE_STRICT. */
3777 delete_flow_strict(struct ofproto
*p
, struct flow_mod
*fm
)
3779 struct rule
*rule
= find_flow_strict(p
, fm
);
3781 delete_flow(p
, rule
, htons(fm
->out_port
));
3785 /* Callback for delete_flows_loose(). */
3787 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3789 struct rule
*rule
= rule_from_cls_rule(rule_
);
3790 struct delete_flows_cbdata
*cbdata
= cbdata_
;
3792 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
3795 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3796 * been identified as a flow to delete from 'p''s flow table, by deleting the
3797 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3800 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3801 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3802 * specified 'out_port'. */
3804 delete_flow(struct ofproto
*p
, struct rule
*rule
, ovs_be16 out_port
)
3806 if (rule_is_hidden(rule
)) {
3810 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
3814 send_flow_removed(p
, rule
, OFPRR_DELETE
);
3815 rule_remove(p
, rule
);
3819 flow_mod_core(struct ofconn
*ofconn
, struct flow_mod
*fm
)
3821 struct ofproto
*p
= ofconn
->ofproto
;
3824 error
= reject_slave_controller(ofconn
, "flow_mod");
3829 error
= validate_actions(fm
->actions
, fm
->n_actions
, p
->max_ports
);
3834 /* We do not support the emergency flow cache. It will hopefully
3835 * get dropped from OpenFlow in the near future. */
3836 if (fm
->flags
& OFPFF_EMERG
) {
3837 /* There isn't a good fit for an error code, so just state that the
3838 * flow table is full. */
3839 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
3842 switch (fm
->command
) {
3844 return add_flow(ofconn
, fm
);
3847 return modify_flows_loose(ofconn
, fm
);
3849 case OFPFC_MODIFY_STRICT
:
3850 return modify_flow_strict(ofconn
, fm
);
3853 delete_flows_loose(p
, fm
);
3856 case OFPFC_DELETE_STRICT
:
3857 delete_flow_strict(p
, fm
);
3861 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
3866 handle_flow_mod(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3868 struct ofp_match orig_match
;
3869 struct ofp_flow_mod
*ofm
;
3875 b
.size
= ntohs(oh
->length
);
3877 /* Dissect the message. */
3878 ofm
= ofpbuf_try_pull(&b
, sizeof *ofm
);
3880 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3882 error
= ofputil_pull_actions(&b
, b
.size
, &fm
.actions
, &fm
.n_actions
);
3887 /* Normalize ofm->match. If normalization actually changes anything, then
3888 * log the differences. */
3889 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
3890 orig_match
= ofm
->match
;
3891 normalize_match(&ofm
->match
);
3892 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
3893 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
3894 if (!VLOG_DROP_INFO(&normal_rl
)) {
3895 char *old
= ofp_match_to_literal_string(&orig_match
);
3896 char *new = ofp_match_to_literal_string(&ofm
->match
);
3897 VLOG_INFO("%s: normalization changed ofp_match, details:",
3898 rconn_get_name(ofconn
->rconn
));
3899 VLOG_INFO(" pre: %s", old
);
3900 VLOG_INFO("post: %s", new);
3906 /* Translate the message. */
3907 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
), ofconn
->flow_format
,
3908 ofm
->cookie
, &fm
.cr
);
3909 fm
.cookie
= ofm
->cookie
;
3910 fm
.command
= ntohs(ofm
->command
);
3911 fm
.idle_timeout
= ntohs(ofm
->idle_timeout
);
3912 fm
.hard_timeout
= ntohs(ofm
->hard_timeout
);
3913 fm
.buffer_id
= ntohl(ofm
->buffer_id
);
3914 fm
.out_port
= ntohs(ofm
->out_port
);
3915 fm
.flags
= ntohs(ofm
->flags
);
3917 /* Execute the command. */
3918 return flow_mod_core(ofconn
, &fm
);
3922 handle_tun_id_from_cookie(struct ofconn
*ofconn
, struct nxt_tun_id_cookie
*msg
)
3926 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
3931 ofconn
->flow_format
= msg
->set
? NXFF_TUN_ID_FROM_COOKIE
: NXFF_OPENFLOW10
;
3936 handle_role_request(struct ofconn
*ofconn
, struct nicira_header
*msg
)
3938 struct nx_role_request
*nrr
;
3939 struct nx_role_request
*reply
;
3943 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
3944 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
3945 ntohs(msg
->header
.length
), sizeof *nrr
);
3946 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3948 nrr
= (struct nx_role_request
*) msg
;
3950 if (ofconn
->type
!= OFCONN_PRIMARY
) {
3951 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
3953 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3956 role
= ntohl(nrr
->role
);
3957 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3958 && role
!= NX_ROLE_SLAVE
) {
3959 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
3961 /* There's no good error code for this. */
3962 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
3965 if (role
== NX_ROLE_MASTER
) {
3966 struct ofconn
*other
;
3968 HMAP_FOR_EACH (other
, hmap_node
, &ofconn
->ofproto
->controllers
) {
3969 if (other
->role
== NX_ROLE_MASTER
) {
3970 other
->role
= NX_ROLE_SLAVE
;
3974 ofconn
->role
= role
;
3976 reply
= make_nxmsg_xid(sizeof *reply
, NXT_ROLE_REPLY
, msg
->header
.xid
,
3978 reply
->role
= htonl(role
);
3979 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3985 handle_vendor(struct ofconn
*ofconn
, void *msg
)
3987 struct ofproto
*p
= ofconn
->ofproto
;
3988 struct ofp_vendor_header
*ovh
= msg
;
3989 struct nicira_header
*nh
;
3991 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
3992 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
3993 "(expected at least %zu)",
3994 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
3995 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3997 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
3998 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
4000 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
4001 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
4002 "(expected at least %zu)",
4003 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
4004 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4008 switch (ntohl(nh
->subtype
)) {
4009 case NXT_STATUS_REQUEST
:
4010 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
4013 case NXT_TUN_ID_FROM_COOKIE
:
4014 return handle_tun_id_from_cookie(ofconn
, msg
);
4016 case NXT_ROLE_REQUEST
:
4017 return handle_role_request(ofconn
, msg
);
4020 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
4024 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4026 struct ofp_header
*ob
;
4029 /* Currently, everything executes synchronously, so we can just
4030 * immediately send the barrier reply. */
4031 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4032 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4037 handle_openflow(struct ofconn
*ofconn
, struct ofpbuf
*ofp_msg
)
4039 struct ofp_header
*oh
= ofp_msg
->data
;
4042 COVERAGE_INC(ofproto_recv_openflow
);
4044 case OFPT_ECHO_REQUEST
:
4045 error
= handle_echo_request(ofconn
, oh
);
4048 case OFPT_ECHO_REPLY
:
4052 case OFPT_FEATURES_REQUEST
:
4053 error
= handle_features_request(ofconn
, oh
);
4056 case OFPT_GET_CONFIG_REQUEST
:
4057 error
= handle_get_config_request(ofconn
, oh
);
4060 case OFPT_SET_CONFIG
:
4061 error
= handle_set_config(ofconn
, ofp_msg
->data
);
4064 case OFPT_PACKET_OUT
:
4065 error
= handle_packet_out(ofconn
, ofp_msg
->data
);
4069 error
= handle_port_mod(ofconn
, oh
);
4073 error
= handle_flow_mod(ofconn
, ofp_msg
->data
);
4076 case OFPT_STATS_REQUEST
:
4077 error
= handle_stats_request(ofconn
, oh
);
4081 error
= handle_vendor(ofconn
, ofp_msg
->data
);
4084 case OFPT_BARRIER_REQUEST
:
4085 error
= handle_barrier_request(ofconn
, oh
);
4089 if (VLOG_IS_WARN_ENABLED()) {
4090 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4091 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4094 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4099 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4104 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4106 struct odp_msg
*msg
= packet
->data
;
4108 struct ofpbuf payload
;
4111 payload
.data
= msg
+ 1;
4112 payload
.size
= msg
->length
- sizeof *msg
;
4113 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4115 /* Check with in-band control to see if this packet should be sent
4116 * to the local port regardless of the flow table. */
4117 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4118 union odp_action action
;
4120 memset(&action
, 0, sizeof(action
));
4121 action
.output
.type
= ODPAT_OUTPUT
;
4122 action
.output
.port
= ODPP_LOCAL
;
4123 dpif_execute(p
->dpif
, &action
, 1, &payload
);
4126 rule
= lookup_valid_rule(p
, &flow
);
4128 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4129 struct ofport
*port
= get_port(p
, msg
->port
);
4131 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4132 COVERAGE_INC(ofproto_no_packet_in
);
4133 /* XXX install 'drop' flow entry */
4134 ofpbuf_delete(packet
);
4138 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
4141 COVERAGE_INC(ofproto_packet_in
);
4142 send_packet_in(p
, packet
);
4146 if (rule
->cr
.wc
.wildcards
) {
4147 rule
= rule_create_subrule(p
, rule
, &flow
);
4148 rule_make_actions(p
, rule
, packet
);
4150 if (!rule
->may_install
) {
4151 /* The rule is not installable, that is, we need to process every
4152 * packet, so process the current packet and set its actions into
4154 rule_make_actions(p
, rule
, packet
);
4156 /* XXX revalidate rule if it needs it */
4160 if (rule
->super
&& rule
->super
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4162 * Extra-special case for fail-open mode.
4164 * We are in fail-open mode and the packet matched the fail-open rule,
4165 * but we are connected to a controller too. We should send the packet
4166 * up to the controller in the hope that it will try to set up a flow
4167 * and thereby allow us to exit fail-open.
4169 * See the top-level comment in fail-open.c for more information.
4171 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4172 DPIF_RECV_MSG_PADDING
));
4175 ofpbuf_pull(packet
, sizeof *msg
);
4176 rule_execute(p
, rule
, packet
, &flow
);
4177 rule_reinstall(p
, rule
);
4181 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4183 struct odp_msg
*msg
= packet
->data
;
4185 switch (msg
->type
) {
4186 case _ODPL_ACTION_NR
:
4187 COVERAGE_INC(ofproto_ctlr_action
);
4188 send_packet_in(p
, packet
);
4191 case _ODPL_SFLOW_NR
:
4193 ofproto_sflow_received(p
->sflow
, msg
);
4195 ofpbuf_delete(packet
);
4199 handle_odp_miss_msg(p
, packet
);
4203 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4209 /* Flow expiration. */
4211 struct expire_cbdata
{
4212 struct ofproto
*ofproto
;
4216 static int ofproto_dp_max_idle(const struct ofproto
*);
4217 static void ofproto_update_used(struct ofproto
*);
4218 static void rule_expire(struct cls_rule
*, void *cbdata
);
4220 /* This function is called periodically by ofproto_run(). Its job is to
4221 * collect updates for the flows that have been installed into the datapath,
4222 * most importantly when they last were used, and then use that information to
4223 * expire flows that have not been used recently.
4225 * Returns the number of milliseconds after which it should be called again. */
4227 ofproto_expire(struct ofproto
*ofproto
)
4229 struct expire_cbdata cbdata
;
4231 /* Update 'used' for each flow in the datapath. */
4232 ofproto_update_used(ofproto
);
4234 /* Expire idle flows.
4236 * A wildcarded flow is idle only when all of its subrules have expired due
4237 * to becoming idle, so iterate through the exact-match flows first. */
4238 cbdata
.ofproto
= ofproto
;
4239 cbdata
.dp_max_idle
= ofproto_dp_max_idle(ofproto
);
4240 classifier_for_each(&ofproto
->cls
, CLS_INC_EXACT
, rule_expire
, &cbdata
);
4241 classifier_for_each(&ofproto
->cls
, CLS_INC_WILD
, rule_expire
, &cbdata
);
4243 /* Let the hook know that we're at a stable point: all outstanding data
4244 * in existing flows has been accounted to the account_cb. Thus, the
4245 * hook can now reasonably do operations that depend on having accurate
4246 * flow volume accounting (currently, that's just bond rebalancing). */
4247 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4248 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4251 return MIN(cbdata
.dp_max_idle
, 1000);
4254 /* Update 'used' member of each flow currently installed into the datapath. */
4256 ofproto_update_used(struct ofproto
*p
)
4258 struct odp_flow
*flows
;
4263 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4268 for (i
= 0; i
< n_flows
; i
++) {
4269 struct odp_flow
*f
= &flows
[i
];
4270 struct cls_rule target
;
4274 odp_flow_key_to_flow(&f
->key
, &flow
);
4275 cls_rule_init_exact(&flow
, UINT16_MAX
, &target
);
4277 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&p
->cls
,
4280 if (rule
&& rule
->installed
) {
4281 update_time(p
, rule
, &f
->stats
);
4282 rule_account(p
, rule
, f
->stats
.n_bytes
);
4284 /* There's a flow in the datapath that we know nothing about.
4286 COVERAGE_INC(ofproto_unexpected_rule
);
4287 dpif_flow_del(p
->dpif
, f
);
4294 /* Calculates and returns the number of milliseconds of idle time after which
4295 * flows should expire from the datapath and we should fold their statistics
4296 * into their parent rules in userspace. */
4298 ofproto_dp_max_idle(const struct ofproto
*ofproto
)
4301 * Idle time histogram.
4303 * Most of the time a switch has a relatively small number of flows. When
4304 * this is the case we might as well keep statistics for all of them in
4305 * userspace and to cache them in the kernel datapath for performance as
4308 * As the number of flows increases, the memory required to maintain
4309 * statistics about them in userspace and in the kernel becomes
4310 * significant. However, with a large number of flows it is likely that
4311 * only a few of them are "heavy hitters" that consume a large amount of
4312 * bandwidth. At this point, only heavy hitters are worth caching in the
4313 * kernel and maintaining in userspaces; other flows we can discard.
4315 * The technique used to compute the idle time is to build a histogram with
4316 * N_BUCKETS bucket whose width is BUCKET_WIDTH msecs each. Each flow that
4317 * is installed in the kernel gets dropped in the appropriate bucket.
4318 * After the histogram has been built, we compute the cutoff so that only
4319 * the most-recently-used 1% of flows (but at least 1000 flows) are kept
4320 * cached. At least the most-recently-used bucket of flows is kept, so
4321 * actually an arbitrary number of flows can be kept in any given
4322 * expiration run (though the next run will delete most of those unless
4323 * they receive additional data).
4325 * This requires a second pass through the exact-match flows, in addition
4326 * to the pass made by ofproto_update_used(), because the former function
4327 * never looks at uninstallable flows.
4329 enum { BUCKET_WIDTH
= ROUND_UP(100, TIME_UPDATE_INTERVAL
) };
4330 enum { N_BUCKETS
= 5000 / BUCKET_WIDTH
};
4331 int buckets
[N_BUCKETS
] = { 0 };
4337 total
= classifier_count_exact(&ofproto
->cls
);
4338 if (total
<= 1000) {
4339 return N_BUCKETS
* BUCKET_WIDTH
;
4342 /* Build histogram. */
4344 CLASSIFIER_FOR_EACH_EXACT_RULE (rule
, cr
, &ofproto
->cls
) {
4345 long long int idle
= now
- rule
->used
;
4346 int bucket
= (idle
<= 0 ? 0
4347 : idle
>= BUCKET_WIDTH
* N_BUCKETS
? N_BUCKETS
- 1
4348 : (unsigned int) idle
/ BUCKET_WIDTH
);
4352 /* Find the first bucket whose flows should be expired. */
4353 for (bucket
= 0; bucket
< N_BUCKETS
; bucket
++) {
4354 if (buckets
[bucket
]) {
4357 subtotal
+= buckets
[bucket
++];
4358 } while (bucket
< N_BUCKETS
&& subtotal
< MAX(1000, total
/ 100));
4363 if (VLOG_IS_DBG_ENABLED()) {
4367 ds_put_cstr(&s
, "keep");
4368 for (i
= 0; i
< N_BUCKETS
; i
++) {
4370 ds_put_cstr(&s
, ", drop");
4373 ds_put_format(&s
, " %d:%d", i
* BUCKET_WIDTH
, buckets
[i
]);
4376 VLOG_INFO("%s: %s (msec:count)",
4377 dpif_name(ofproto
->dpif
), ds_cstr(&s
));
4381 return bucket
* BUCKET_WIDTH
;
4385 rule_active_timeout(struct ofproto
*ofproto
, struct rule
*rule
)
4387 if (ofproto
->netflow
&& !is_controller_rule(rule
) &&
4388 netflow_active_timeout_expired(ofproto
->netflow
, &rule
->nf_flow
)) {
4389 struct ofexpired expired
;
4390 struct odp_flow odp_flow
;
4392 /* Get updated flow stats.
4394 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4395 * updated TCP flags and (2) the dpif_flow_list_all() in
4396 * ofproto_update_used() zeroed TCP flags. */
4397 memset(&odp_flow
, 0, sizeof odp_flow
);
4398 if (rule
->installed
) {
4399 odp_flow_key_from_flow(&odp_flow
.key
, &rule
->cr
.flow
);
4400 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4401 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4403 if (odp_flow
.stats
.n_packets
) {
4404 update_time(ofproto
, rule
, &odp_flow
.stats
);
4405 netflow_flow_update_flags(&rule
->nf_flow
,
4406 odp_flow
.stats
.tcp_flags
);
4410 expired
.flow
= rule
->cr
.flow
;
4411 expired
.packet_count
= rule
->packet_count
+
4412 odp_flow
.stats
.n_packets
;
4413 expired
.byte_count
= rule
->byte_count
+ odp_flow
.stats
.n_bytes
;
4414 expired
.used
= rule
->used
;
4416 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
4420 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4421 * rules, then delete it entirely.
4423 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4424 * the datapath and fold its statistics back into its super-rule.
4426 * (This is a callback function for classifier_for_each().) */
4428 rule_expire(struct cls_rule
*cls_rule
, void *cbdata_
)
4430 struct expire_cbdata
*cbdata
= cbdata_
;
4431 struct ofproto
*ofproto
= cbdata
->ofproto
;
4432 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
4433 long long int hard_expire
, idle_expire
, expire
, now
;
4435 /* Calculate OpenFlow expiration times for 'rule'. */
4436 hard_expire
= (rule
->hard_timeout
4437 ? rule
->created
+ rule
->hard_timeout
* 1000
4439 idle_expire
= (rule
->idle_timeout
4440 && (rule
->super
|| list_is_empty(&rule
->list
))
4441 ? rule
->used
+ rule
->idle_timeout
* 1000
4443 expire
= MIN(hard_expire
, idle_expire
);
4447 /* 'rule' has not expired according to OpenFlow rules. */
4448 if (!rule
->cr
.wc
.wildcards
) {
4449 if (now
>= rule
->used
+ cbdata
->dp_max_idle
) {
4450 /* This rule is idle, so drop it to free up resources. */
4452 /* It's not part of the OpenFlow flow table, so we can
4453 * delete it entirely and fold its statistics into its
4455 rule_remove(ofproto
, rule
);
4457 /* It is part of the OpenFlow flow table, so we have to
4458 * keep the rule but we can at least uninstall it from the
4460 rule_uninstall(ofproto
, rule
);
4463 /* Send NetFlow active timeout if appropriate. */
4464 rule_active_timeout(cbdata
->ofproto
, rule
);
4468 /* 'rule' has expired according to OpenFlow rules. */
4469 COVERAGE_INC(ofproto_expired
);
4471 /* Update stats. (This is a no-op if the rule expired due to an idle
4472 * timeout, because that only happens when the rule has no subrules
4474 if (rule
->cr
.wc
.wildcards
) {
4475 struct rule
*subrule
, *next
;
4476 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
4477 rule_remove(cbdata
->ofproto
, subrule
);
4480 rule_uninstall(cbdata
->ofproto
, rule
);
4483 /* Get rid of the rule. */
4484 if (!rule_is_hidden(rule
)) {
4485 send_flow_removed(cbdata
->ofproto
, rule
,
4487 ? OFPRR_HARD_TIMEOUT
: OFPRR_IDLE_TIMEOUT
));
4489 rule_remove(cbdata
->ofproto
, rule
);
4494 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
4496 struct rule
*sub
= rule_from_cls_rule(sub_
);
4497 struct revalidate_cbdata
*cbdata
= cbdata_
;
4499 if (cbdata
->revalidate_all
4500 || (cbdata
->revalidate_subrules
&& sub
->super
)
4501 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
4502 revalidate_rule(cbdata
->ofproto
, sub
);
4507 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
4509 const struct flow
*flow
= &rule
->cr
.flow
;
4511 COVERAGE_INC(ofproto_revalidate_rule
);
4514 super
= rule_from_cls_rule(classifier_lookup(&p
->cls
, flow
,
4517 rule_remove(p
, rule
);
4519 } else if (super
!= rule
->super
) {
4520 COVERAGE_INC(ofproto_revalidate_moved
);
4521 list_remove(&rule
->list
);
4522 list_push_back(&super
->list
, &rule
->list
);
4523 rule
->super
= super
;
4524 rule
->hard_timeout
= super
->hard_timeout
;
4525 rule
->idle_timeout
= super
->idle_timeout
;
4526 rule
->created
= super
->created
;
4531 rule_update_actions(p
, rule
);
4535 static struct ofpbuf
*
4536 compose_flow_removed(struct ofconn
*ofconn
, const struct rule
*rule
,
4539 struct ofp_flow_removed
*ofr
;
4542 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4543 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, ofconn
->flow_format
,
4545 ofr
->cookie
= rule
->flow_cookie
;
4546 ofr
->priority
= htons(rule
->cr
.priority
);
4547 ofr
->reason
= reason
;
4548 calc_flow_duration(rule
->created
, &ofr
->duration_sec
, &ofr
->duration_nsec
);
4549 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4550 ofr
->packet_count
= htonll(rule
->packet_count
);
4551 ofr
->byte_count
= htonll(rule
->byte_count
);
4557 send_flow_removed(struct ofproto
*p
, struct rule
*rule
, uint8_t reason
)
4559 struct ofconn
*ofconn
;
4561 if (!rule
->send_flow_removed
) {
4565 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4568 if (!rconn_is_connected(ofconn
->rconn
)
4569 || !ofconn_receives_async_msgs(ofconn
)) {
4573 msg
= compose_flow_removed(ofconn
, rule
, reason
);
4575 /* Account flow expirations under ofconn->reply_counter, the counter
4576 * for replies to OpenFlow requests. That works because preventing
4577 * OpenFlow requests from being processed also prevents new flows from
4578 * being added (and expiring). (It also prevents processing OpenFlow
4579 * requests that would not add new flows, so it is imperfect.) */
4580 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
4584 /* pinsched callback for sending 'packet' on 'ofconn'. */
4586 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4588 struct ofconn
*ofconn
= ofconn_
;
4590 rconn_send_with_limit(ofconn
->rconn
, packet
,
4591 ofconn
->packet_in_counter
, 100);
4594 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4595 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4596 * packet scheduler for sending.
4598 * 'max_len' specifies the maximum number of bytes of the packet to send on
4599 * 'ofconn' (INT_MAX specifies no limit).
4601 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4602 * ownership is transferred to this function. */
4604 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4607 struct ofproto
*ofproto
= ofconn
->ofproto
;
4608 struct ofp_packet_in
*opi
= packet
->data
;
4609 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4610 int send_len
, trim_size
;
4614 if (opi
->reason
== OFPR_ACTION
) {
4615 buffer_id
= UINT32_MAX
;
4616 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4617 buffer_id
= pktbuf_get_null();
4618 } else if (!ofconn
->pktbuf
) {
4619 buffer_id
= UINT32_MAX
;
4621 struct ofpbuf payload
;
4622 payload
.data
= opi
->data
;
4623 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4624 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4627 /* Figure out how much of the packet to send. */
4628 send_len
= ntohs(opi
->total_len
);
4629 if (buffer_id
!= UINT32_MAX
) {
4630 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4632 send_len
= MIN(send_len
, max_len
);
4634 /* Adjust packet length and clone if necessary. */
4635 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4637 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4640 packet
->size
= trim_size
;
4643 /* Update packet headers. */
4644 opi
->buffer_id
= htonl(buffer_id
);
4645 update_openflow_length(packet
);
4647 /* Hand over to packet scheduler. It might immediately call into
4648 * do_send_packet_in() or it might buffer it for a while (until a later
4649 * call to pinsched_run()). */
4650 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4651 packet
, do_send_packet_in
, ofconn
);
4654 /* Replace struct odp_msg header in 'packet' by equivalent struct
4655 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4656 * returned by dpif_recv()).
4658 * The conversion is not complete: the caller still needs to trim any unneeded
4659 * payload off the end of the buffer, set the length in the OpenFlow header,
4660 * and set buffer_id. Those require us to know the controller settings and so
4661 * must be done on a per-controller basis.
4663 * Returns the maximum number of bytes of the packet that should be sent to
4664 * the controller (INT_MAX if no limit). */
4666 do_convert_to_packet_in(struct ofpbuf
*packet
)
4668 struct odp_msg
*msg
= packet
->data
;
4669 struct ofp_packet_in
*opi
;
4675 /* Extract relevant header fields */
4676 if (msg
->type
== _ODPL_ACTION_NR
) {
4677 reason
= OFPR_ACTION
;
4680 reason
= OFPR_NO_MATCH
;
4683 total_len
= msg
->length
- sizeof *msg
;
4684 in_port
= odp_port_to_ofp_port(msg
->port
);
4686 /* Repurpose packet buffer by overwriting header. */
4687 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4688 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4689 opi
->header
.version
= OFP_VERSION
;
4690 opi
->header
.type
= OFPT_PACKET_IN
;
4691 opi
->total_len
= htons(total_len
);
4692 opi
->in_port
= htons(in_port
);
4693 opi
->reason
= reason
;
4698 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4699 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4700 * as necessary according to their individual configurations.
4702 * 'packet' must have sufficient headroom to convert it into a struct
4703 * ofp_packet_in (e.g. as returned by dpif_recv()).
4705 * Takes ownership of 'packet'. */
4707 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4709 struct ofconn
*ofconn
, *prev
;
4712 max_len
= do_convert_to_packet_in(packet
);
4715 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4716 if (ofconn_receives_async_msgs(ofconn
)) {
4718 schedule_packet_in(prev
, packet
, max_len
, true);
4724 schedule_packet_in(prev
, packet
, max_len
, false);
4726 ofpbuf_delete(packet
);
4731 pick_datapath_id(const struct ofproto
*ofproto
)
4733 const struct ofport
*port
;
4735 port
= get_port(ofproto
, ODPP_LOCAL
);
4737 uint8_t ea
[ETH_ADDR_LEN
];
4740 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4742 return eth_addr_to_uint64(ea
);
4744 VLOG_WARN("could not get MAC address for %s (%s)",
4745 netdev_get_name(port
->netdev
), strerror(error
));
4747 return ofproto
->fallback_dpid
;
4751 pick_fallback_dpid(void)
4753 uint8_t ea
[ETH_ADDR_LEN
];
4754 eth_addr_nicira_random(ea
);
4755 return eth_addr_to_uint64(ea
);
4759 default_normal_ofhook_cb(const struct flow
*flow
, const struct ofpbuf
*packet
,
4760 struct odp_actions
*actions
, tag_type
*tags
,
4761 uint16_t *nf_output_iface
, void *ofproto_
)
4763 struct ofproto
*ofproto
= ofproto_
;
4766 /* Drop frames for reserved multicast addresses. */
4767 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4771 /* Learn source MAC (but don't try to learn from revalidation). */
4772 if (packet
!= NULL
) {
4773 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4775 GRAT_ARP_LOCK_NONE
);
4777 /* The log messages here could actually be useful in debugging,
4778 * so keep the rate limit relatively high. */
4779 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4780 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4781 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4782 ofproto_revalidate(ofproto
, rev_tag
);
4786 /* Determine output port. */
4787 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
4790 flood_packets(ofproto
, flow
->in_port
, OFPPC_NO_FLOOD
,
4791 nf_output_iface
, actions
);
4792 } else if (out_port
!= flow
->in_port
) {
4793 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4794 *nf_output_iface
= out_port
;
4802 static const struct ofhooks default_ofhooks
= {
4803 default_normal_ofhook_cb
,