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 "classifier.h"
29 #include "discovery.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
40 #include "ofp-print.h"
42 #include "ofproto-sflow.h"
44 #include "openflow/nicira-ext.h"
45 #include "openflow/openflow.h"
46 #include "openvswitch/datapath-protocol.h"
50 #include "poll-loop.h"
54 #include "stream-ssl.h"
63 VLOG_DEFINE_THIS_MODULE(ofproto
)
65 #include "sflow_api.h"
69 TABLEID_CLASSIFIER
= 1
74 struct hmap_node hmap_node
; /* In struct ofproto's "ports" hmap. */
75 struct netdev
*netdev
;
76 struct ofp_phy_port opp
; /* In host byte order. */
80 static void ofport_free(struct ofport
*);
81 static void hton_ofp_phy_port(struct ofp_phy_port
*);
83 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
84 const flow_t
*flow
, struct ofproto
*ofproto
,
85 const struct ofpbuf
*packet
,
86 struct odp_actions
*out
, tag_type
*tags
,
87 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
92 uint64_t flow_cookie
; /* Controller-issued identifier.
93 (Kept in network-byte order.) */
94 uint16_t idle_timeout
; /* In seconds from time of last use. */
95 uint16_t hard_timeout
; /* In seconds from time of creation. */
96 bool send_flow_removed
; /* Send a flow removed message? */
97 long long int used
; /* Last-used time (0 if never used). */
98 long long int created
; /* Creation time. */
99 uint64_t packet_count
; /* Number of packets received. */
100 uint64_t byte_count
; /* Number of bytes received. */
101 uint64_t accounted_bytes
; /* Number of bytes passed to account_cb. */
102 tag_type tags
; /* Tags (set only by hooks). */
103 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
105 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
106 * exact-match rule (having cr.wc.wildcards of 0) generated from the
107 * wildcard rule 'super'. In this case, 'list' is an element of the
110 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
111 * a list of subrules. A super-rule with no wildcards (where
112 * cr.wc.wildcards is 0) will never have any subrules. */
118 * 'n_actions' is the number of elements in the 'actions' array. A single
119 * action may take up more more than one element's worth of space.
121 * A subrule has no actions (it uses the super-rule's actions). */
123 union ofp_action
*actions
;
127 * A super-rule with wildcard fields never has ODP actions (since the
128 * datapath only supports exact-match flows). */
129 bool installed
; /* Installed in datapath? */
130 bool may_install
; /* True ordinarily; false if actions must
131 * be reassessed for every packet. */
133 union odp_action
*odp_actions
;
137 rule_is_hidden(const struct rule
*rule
)
139 /* Subrules are merely an implementation detail, so hide them from the
141 if (rule
->super
!= NULL
) {
145 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
146 * (e.g. by in-band control) and are intentionally hidden from the
148 if (rule
->cr
.priority
> UINT16_MAX
) {
155 static struct rule
*rule_create(struct ofproto
*, struct rule
*super
,
156 const union ofp_action
*, size_t n_actions
,
157 uint16_t idle_timeout
, uint16_t hard_timeout
,
158 uint64_t flow_cookie
, bool send_flow_removed
);
159 static void rule_free(struct rule
*);
160 static void rule_destroy(struct ofproto
*, struct rule
*);
161 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
162 static void rule_insert(struct ofproto
*, struct rule
*,
163 struct ofpbuf
*packet
, uint16_t in_port
);
164 static void rule_remove(struct ofproto
*, struct rule
*);
165 static bool rule_make_actions(struct ofproto
*, struct rule
*,
166 const struct ofpbuf
*packet
);
167 static void rule_install(struct ofproto
*, struct rule
*,
168 struct rule
*displaced_rule
);
169 static void rule_uninstall(struct ofproto
*, struct rule
*);
170 static void rule_post_uninstall(struct ofproto
*, struct rule
*);
171 static void send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
172 long long int now
, uint8_t reason
);
174 /* ofproto supports two kinds of OpenFlow connections:
176 * - "Primary" connections to ordinary OpenFlow controllers. ofproto
177 * maintains persistent connections to these controllers and by default
178 * sends them asynchronous messages such as packet-ins.
180 * - "Service" connections, e.g. from ovs-ofctl. When these connections
181 * drop, it is the other side's responsibility to reconnect them if
182 * necessary. ofproto does not send them asynchronous messages by default.
184 * Currently, active (tcp, ssl, unix) connections are always "primary"
185 * connections and passive (ptcp, pssl, punix) connections are always "service"
186 * connections. There is no inherent reason for this, but it reflects the
190 OFCONN_PRIMARY
, /* An ordinary OpenFlow controller. */
191 OFCONN_SERVICE
/* A service connection, e.g. "ovs-ofctl". */
194 /* A listener for incoming OpenFlow "service" connections. */
196 struct hmap_node node
; /* In struct ofproto's "services" hmap. */
197 struct pvconn
*pvconn
; /* OpenFlow connection listener. */
199 /* These are not used by ofservice directly. They are settings for
200 * accepted "struct ofconn"s from the pvconn. */
201 int probe_interval
; /* Max idle time before probing, in seconds. */
202 int rate_limit
; /* Max packet-in rate in packets per second. */
203 int burst_limit
; /* Limit on accumulating packet credits. */
206 static struct ofservice
*ofservice_lookup(struct ofproto
*,
208 static int ofservice_create(struct ofproto
*,
209 const struct ofproto_controller
*);
210 static void ofservice_reconfigure(struct ofservice
*,
211 const struct ofproto_controller
*);
212 static void ofservice_destroy(struct ofproto
*, struct ofservice
*);
214 /* An OpenFlow connection. */
216 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
217 struct list node
; /* In struct ofproto's "all_conns" list. */
218 struct rconn
*rconn
; /* OpenFlow connection. */
219 enum ofconn_type type
; /* Type. */
221 /* OFPT_PACKET_IN related data. */
222 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
223 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
224 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
225 int miss_send_len
; /* Bytes to send of buffered packets. */
227 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
228 * requests, and the maximum number before we stop reading OpenFlow
230 #define OFCONN_REPLY_MAX 100
231 struct rconn_packet_counter
*reply_counter
;
233 /* type == OFCONN_PRIMARY only. */
234 enum nx_role role
; /* Role. */
235 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
236 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
237 struct status_category
*ss
; /* Switch status category. */
238 enum ofproto_band band
; /* In-band or out-of-band? */
241 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
242 * "schedulers" array. Their values are 0 and 1, and their meanings and values
243 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
244 * case anything ever changes, check their values here. */
245 #define N_SCHEDULERS 2
246 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
247 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
248 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
249 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
251 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
253 static void ofconn_destroy(struct ofconn
*);
254 static void ofconn_run(struct ofconn
*, struct ofproto
*);
255 static void ofconn_wait(struct ofconn
*);
256 static bool ofconn_receives_async_msgs(const struct ofconn
*);
257 static char *ofconn_make_name(const struct ofproto
*, const char *target
);
258 static void ofconn_set_rate_limit(struct ofconn
*, int rate
, int burst
);
260 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
261 struct rconn_packet_counter
*counter
);
263 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
264 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
268 uint64_t datapath_id
; /* Datapath ID. */
269 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
270 char *mfr_desc
; /* Manufacturer. */
271 char *hw_desc
; /* Hardware. */
272 char *sw_desc
; /* Software version. */
273 char *serial_desc
; /* Serial number. */
274 char *dp_desc
; /* Datapath description. */
278 struct netdev_monitor
*netdev_monitor
;
279 struct hmap ports
; /* Contains "struct ofport"s. */
280 struct shash port_by_name
;
284 struct switch_status
*switch_status
;
285 struct fail_open
*fail_open
;
286 struct netflow
*netflow
;
287 struct ofproto_sflow
*sflow
;
289 /* In-band control. */
290 struct in_band
*in_band
;
291 long long int next_in_band_update
;
292 struct sockaddr_in
*extra_in_band_remotes
;
293 size_t n_extra_remotes
;
296 struct classifier cls
;
297 bool need_revalidate
;
298 long long int next_expiration
;
299 struct tag_set revalidate_set
;
300 bool tun_id_from_cookie
;
302 /* OpenFlow connections. */
303 struct hmap controllers
; /* Controller "struct ofconn"s. */
304 struct list all_conns
; /* Contains "struct ofconn"s. */
305 enum ofproto_fail_mode fail_mode
;
307 /* OpenFlow listeners. */
308 struct hmap services
; /* Contains "struct ofservice"s. */
309 struct pvconn
**snoops
;
312 /* Hooks for ovs-vswitchd. */
313 const struct ofhooks
*ofhooks
;
316 /* Used by default ofhooks. */
317 struct mac_learning
*ml
;
320 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
322 static const struct ofhooks default_ofhooks
;
324 static uint64_t pick_datapath_id(const struct ofproto
*);
325 static uint64_t pick_fallback_dpid(void);
327 static void ofproto_expire(struct ofproto
*);
329 static void update_stats(struct ofproto
*, struct rule
*,
330 const struct odp_flow_stats
*);
331 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
332 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
334 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
336 static void handle_openflow(struct ofconn
*, struct ofproto
*,
339 static void refresh_port_groups(struct ofproto
*);
341 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
342 static void update_port(struct ofproto
*, const char *devname
);
343 static int init_ports(struct ofproto
*);
344 static void reinit_ports(struct ofproto
*);
347 ofproto_create(const char *datapath
, const char *datapath_type
,
348 const struct ofhooks
*ofhooks
, void *aux
,
349 struct ofproto
**ofprotop
)
351 struct odp_stats stats
;
358 /* Connect to datapath and start listening for messages. */
359 error
= dpif_open(datapath
, datapath_type
, &dpif
);
361 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
364 error
= dpif_get_dp_stats(dpif
, &stats
);
366 VLOG_ERR("failed to obtain stats for datapath %s: %s",
367 datapath
, strerror(error
));
371 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
373 VLOG_ERR("failed to listen on datapath %s: %s",
374 datapath
, strerror(error
));
378 dpif_flow_flush(dpif
);
379 dpif_recv_purge(dpif
);
381 /* Initialize settings. */
382 p
= xzalloc(sizeof *p
);
383 p
->fallback_dpid
= pick_fallback_dpid();
384 p
->datapath_id
= p
->fallback_dpid
;
385 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
386 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
387 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
388 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
389 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
391 /* Initialize datapath. */
393 p
->netdev_monitor
= netdev_monitor_create();
394 hmap_init(&p
->ports
);
395 shash_init(&p
->port_by_name
);
396 p
->max_ports
= stats
.max_ports
;
398 /* Initialize submodules. */
399 p
->switch_status
= switch_status_create(p
);
405 /* Initialize flow table. */
406 classifier_init(&p
->cls
);
407 p
->need_revalidate
= false;
408 p
->next_expiration
= time_msec() + 1000;
409 tag_set_init(&p
->revalidate_set
);
411 /* Initialize OpenFlow connections. */
412 list_init(&p
->all_conns
);
413 hmap_init(&p
->controllers
);
414 hmap_init(&p
->services
);
418 /* Initialize hooks. */
420 p
->ofhooks
= ofhooks
;
424 p
->ofhooks
= &default_ofhooks
;
426 p
->ml
= mac_learning_create();
429 /* Pick final datapath ID. */
430 p
->datapath_id
= pick_datapath_id(p
);
431 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
438 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
440 uint64_t old_dpid
= p
->datapath_id
;
441 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
442 if (p
->datapath_id
!= old_dpid
) {
443 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
445 /* Force all active connections to reconnect, since there is no way to
446 * notify a controller that the datapath ID has changed. */
447 ofproto_reconnect_controllers(p
);
452 is_discovery_controller(const struct ofproto_controller
*c
)
454 return !strcmp(c
->target
, "discover");
458 is_in_band_controller(const struct ofproto_controller
*c
)
460 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
463 /* Creates a new controller in 'ofproto'. Some of the settings are initially
464 * drawn from 'c', but update_controller() needs to be called later to finish
465 * the new ofconn's configuration. */
467 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
469 struct discovery
*discovery
;
470 struct ofconn
*ofconn
;
472 if (is_discovery_controller(c
)) {
473 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
474 ofproto
->dpif
, ofproto
->switch_status
,
483 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
484 ofconn
->pktbuf
= pktbuf_create();
485 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
487 ofconn
->discovery
= discovery
;
489 char *name
= ofconn_make_name(ofproto
, c
->target
);
490 rconn_connect(ofconn
->rconn
, c
->target
, name
);
493 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
494 hash_string(c
->target
, 0));
497 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
498 * target or turn discovery on or off (these are done by creating new ofconns
499 * and deleting old ones), but it can update the rest of an ofconn's
502 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
506 ofconn
->band
= (is_in_band_controller(c
)
507 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
509 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
511 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
512 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
514 if (ofconn
->discovery
) {
515 discovery_set_update_resolv_conf(ofconn
->discovery
,
516 c
->update_resolv_conf
);
517 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
520 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
524 ofconn_get_target(const struct ofconn
*ofconn
)
526 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
529 static struct ofconn
*
530 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
532 struct ofconn
*ofconn
;
534 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
535 hash_string(target
, 0), &ofproto
->controllers
) {
536 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
544 update_in_band_remotes(struct ofproto
*ofproto
)
546 const struct ofconn
*ofconn
;
547 struct sockaddr_in
*addrs
;
548 size_t max_addrs
, n_addrs
;
552 /* Allocate enough memory for as many remotes as we could possibly have. */
553 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
554 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
557 /* Add all the remotes. */
559 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
560 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
562 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
566 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
567 if (sin
->sin_addr
.s_addr
) {
568 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
571 if (ofconn
->discovery
) {
575 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
576 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
579 /* Create or update or destroy in-band.
581 * Ordinarily we only enable in-band if there's at least one remote
582 * address, but discovery needs the in-band rules for DHCP to be installed
583 * even before we know any remote addresses. */
584 if (n_addrs
|| discovery
) {
585 if (!ofproto
->in_band
) {
586 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
589 if (ofproto
->in_band
) {
590 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
592 ofproto
->next_in_band_update
= time_msec() + 1000;
594 in_band_destroy(ofproto
->in_band
);
595 ofproto
->in_band
= NULL
;
603 update_fail_open(struct ofproto
*p
)
605 struct ofconn
*ofconn
;
607 if (!hmap_is_empty(&p
->controllers
)
608 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
609 struct rconn
**rconns
;
613 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
617 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
618 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
619 rconns
[n
++] = ofconn
->rconn
;
622 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
623 /* p->fail_open takes ownership of 'rconns'. */
625 fail_open_destroy(p
->fail_open
);
631 ofproto_set_controllers(struct ofproto
*p
,
632 const struct ofproto_controller
*controllers
,
633 size_t n_controllers
)
635 struct shash new_controllers
;
636 struct ofconn
*ofconn
, *next_ofconn
;
637 struct ofservice
*ofservice
, *next_ofservice
;
641 /* Create newly configured controllers and services.
642 * Create a name to ofproto_controller mapping in 'new_controllers'. */
643 shash_init(&new_controllers
);
644 for (i
= 0; i
< n_controllers
; i
++) {
645 const struct ofproto_controller
*c
= &controllers
[i
];
647 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
648 if (!find_controller_by_target(p
, c
->target
)) {
649 add_controller(p
, c
);
651 } else if (!pvconn_verify_name(c
->target
)) {
652 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
656 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
657 dpif_name(p
->dpif
), c
->target
);
661 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
664 /* Delete controllers that are no longer configured.
665 * Update configuration of all now-existing controllers. */
667 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
668 struct ofproto_controller
*c
;
670 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
672 ofconn_destroy(ofconn
);
674 update_controller(ofconn
, c
);
681 /* Delete services that are no longer configured.
682 * Update configuration of all now-existing services. */
683 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
684 struct ofproto_controller
*c
;
686 c
= shash_find_data(&new_controllers
,
687 pvconn_get_name(ofservice
->pvconn
));
689 ofservice_destroy(p
, ofservice
);
691 ofservice_reconfigure(ofservice
, c
);
695 shash_destroy(&new_controllers
);
697 update_in_band_remotes(p
);
700 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
701 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
702 struct ofconn
, hmap_node
);
703 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
704 rconn_status_cb
, ofconn
->rconn
);
709 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
711 p
->fail_mode
= fail_mode
;
715 /* Drops the connections between 'ofproto' and all of its controllers, forcing
716 * them to reconnect. */
718 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
720 struct ofconn
*ofconn
;
722 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
723 rconn_reconnect(ofconn
->rconn
);
728 any_extras_changed(const struct ofproto
*ofproto
,
729 const struct sockaddr_in
*extras
, size_t n
)
733 if (n
!= ofproto
->n_extra_remotes
) {
737 for (i
= 0; i
< n
; i
++) {
738 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
739 const struct sockaddr_in
*new = &extras
[i
];
741 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
742 old
->sin_port
!= new->sin_port
) {
750 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
751 * in-band control should guarantee access, in the same way that in-band
752 * control guarantees access to OpenFlow controllers. */
754 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
755 const struct sockaddr_in
*extras
, size_t n
)
757 if (!any_extras_changed(ofproto
, extras
, n
)) {
761 free(ofproto
->extra_in_band_remotes
);
762 ofproto
->n_extra_remotes
= n
;
763 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
765 update_in_band_remotes(ofproto
);
769 ofproto_set_desc(struct ofproto
*p
,
770 const char *mfr_desc
, const char *hw_desc
,
771 const char *sw_desc
, const char *serial_desc
,
774 struct ofp_desc_stats
*ods
;
777 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
778 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
779 sizeof ods
->mfr_desc
);
782 p
->mfr_desc
= xstrdup(mfr_desc
);
785 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
786 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
787 sizeof ods
->hw_desc
);
790 p
->hw_desc
= xstrdup(hw_desc
);
793 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
794 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
795 sizeof ods
->sw_desc
);
798 p
->sw_desc
= xstrdup(sw_desc
);
801 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
802 VLOG_WARN("truncating serial_desc, must be less than %zu "
804 sizeof ods
->serial_num
);
806 free(p
->serial_desc
);
807 p
->serial_desc
= xstrdup(serial_desc
);
810 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
811 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
812 sizeof ods
->dp_desc
);
815 p
->dp_desc
= xstrdup(dp_desc
);
820 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
821 const struct svec
*svec
)
823 struct pvconn
**pvconns
= *pvconnsp
;
824 size_t n_pvconns
= *n_pvconnsp
;
828 for (i
= 0; i
< n_pvconns
; i
++) {
829 pvconn_close(pvconns
[i
]);
833 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
835 for (i
= 0; i
< svec
->n
; i
++) {
836 const char *name
= svec
->names
[i
];
837 struct pvconn
*pvconn
;
840 error
= pvconn_open(name
, &pvconn
);
842 pvconns
[n_pvconns
++] = pvconn
;
844 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
852 *n_pvconnsp
= n_pvconns
;
858 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
860 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
864 ofproto_set_netflow(struct ofproto
*ofproto
,
865 const struct netflow_options
*nf_options
)
867 if (nf_options
&& nf_options
->collectors
.n
) {
868 if (!ofproto
->netflow
) {
869 ofproto
->netflow
= netflow_create();
871 return netflow_set_options(ofproto
->netflow
, nf_options
);
873 netflow_destroy(ofproto
->netflow
);
874 ofproto
->netflow
= NULL
;
880 ofproto_set_sflow(struct ofproto
*ofproto
,
881 const struct ofproto_sflow_options
*oso
)
883 struct ofproto_sflow
*os
= ofproto
->sflow
;
886 struct ofport
*ofport
;
888 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
889 refresh_port_groups(ofproto
);
890 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
891 ofproto_sflow_add_port(os
, ofport
->odp_port
,
892 netdev_get_name(ofport
->netdev
));
895 ofproto_sflow_set_options(os
, oso
);
897 ofproto_sflow_destroy(os
);
898 ofproto
->sflow
= NULL
;
903 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
905 return ofproto
->datapath_id
;
909 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
911 return !hmap_is_empty(&ofproto
->controllers
);
914 enum ofproto_fail_mode
915 ofproto_get_fail_mode(const struct ofproto
*p
)
921 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
925 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
926 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
931 ofproto_destroy(struct ofproto
*p
)
933 struct ofservice
*ofservice
, *next_ofservice
;
934 struct ofconn
*ofconn
, *next_ofconn
;
935 struct ofport
*ofport
, *next_ofport
;
942 /* Destroy fail-open and in-band early, since they touch the classifier. */
943 fail_open_destroy(p
->fail_open
);
946 in_band_destroy(p
->in_band
);
948 free(p
->extra_in_band_remotes
);
950 ofproto_flush_flows(p
);
951 classifier_destroy(&p
->cls
);
953 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
954 ofconn_destroy(ofconn
);
956 hmap_destroy(&p
->controllers
);
959 netdev_monitor_destroy(p
->netdev_monitor
);
960 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
961 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
964 shash_destroy(&p
->port_by_name
);
966 switch_status_destroy(p
->switch_status
);
967 netflow_destroy(p
->netflow
);
968 ofproto_sflow_destroy(p
->sflow
);
970 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
971 ofservice_destroy(p
, ofservice
);
973 hmap_destroy(&p
->services
);
975 for (i
= 0; i
< p
->n_snoops
; i
++) {
976 pvconn_close(p
->snoops
[i
]);
980 mac_learning_destroy(p
->ml
);
985 free(p
->serial_desc
);
988 hmap_destroy(&p
->ports
);
994 ofproto_run(struct ofproto
*p
)
996 int error
= ofproto_run1(p
);
998 error
= ofproto_run2(p
, false);
1004 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1006 if (error
== ENOBUFS
) {
1007 reinit_ports(ofproto
);
1008 } else if (!error
) {
1009 update_port(ofproto
, devname
);
1014 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1015 * means that 'ofconn' is more interesting for monitoring than a lower return
1018 snoop_preference(const struct ofconn
*ofconn
)
1020 switch (ofconn
->role
) {
1021 case NX_ROLE_MASTER
:
1028 /* Shouldn't happen. */
1033 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1034 * Connects this vconn to a controller. */
1036 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1038 struct ofconn
*ofconn
, *best
;
1040 /* Pick a controller for monitoring. */
1042 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1043 if (ofconn
->type
== OFCONN_PRIMARY
1044 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1050 rconn_add_monitor(best
->rconn
, vconn
);
1052 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1058 ofproto_run1(struct ofproto
*p
)
1060 struct ofconn
*ofconn
, *next_ofconn
;
1061 struct ofservice
*ofservice
;
1066 if (shash_is_empty(&p
->port_by_name
)) {
1070 for (i
= 0; i
< 50; i
++) {
1073 error
= dpif_recv(p
->dpif
, &buf
);
1075 if (error
== ENODEV
) {
1076 /* Someone destroyed the datapath behind our back. The caller
1077 * better destroy us and give up, because we're just going to
1078 * spin from here on out. */
1079 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1080 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1081 dpif_name(p
->dpif
));
1087 handle_odp_msg(p
, buf
);
1090 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1091 process_port_change(p
, error
, devname
);
1093 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1094 &devname
)) != EAGAIN
) {
1095 process_port_change(p
, error
, devname
);
1099 if (time_msec() >= p
->next_in_band_update
) {
1100 update_in_band_remotes(p
);
1102 in_band_run(p
->in_band
);
1105 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1106 ofconn_run(ofconn
, p
);
1109 /* Fail-open maintenance. Do this after processing the ofconns since
1110 * fail-open checks the status of the controller rconn. */
1112 fail_open_run(p
->fail_open
);
1115 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1116 struct vconn
*vconn
;
1119 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1121 struct rconn
*rconn
;
1124 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1125 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1126 rconn_connect_unreliably(rconn
, vconn
, name
);
1129 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1130 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1131 ofservice
->burst_limit
);
1132 } else if (retval
!= EAGAIN
) {
1133 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1137 for (i
= 0; i
< p
->n_snoops
; i
++) {
1138 struct vconn
*vconn
;
1141 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1143 add_snooper(p
, vconn
);
1144 } else if (retval
!= EAGAIN
) {
1145 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1149 if (time_msec() >= p
->next_expiration
) {
1150 COVERAGE_INC(ofproto_expiration
);
1152 p
->next_expiration
= time_msec() + 1000;
1156 netflow_run(p
->netflow
);
1159 ofproto_sflow_run(p
->sflow
);
1165 struct revalidate_cbdata
{
1166 struct ofproto
*ofproto
;
1167 bool revalidate_all
; /* Revalidate all exact-match rules? */
1168 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
1169 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
1173 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1175 if (p
->need_revalidate
|| revalidate_all
1176 || !tag_set_is_empty(&p
->revalidate_set
)) {
1177 struct revalidate_cbdata cbdata
;
1179 cbdata
.revalidate_all
= revalidate_all
;
1180 cbdata
.revalidate_subrules
= p
->need_revalidate
;
1181 cbdata
.revalidate_set
= p
->revalidate_set
;
1182 tag_set_init(&p
->revalidate_set
);
1183 COVERAGE_INC(ofproto_revalidate
);
1184 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
1185 p
->need_revalidate
= false;
1192 ofproto_wait(struct ofproto
*p
)
1194 struct ofservice
*ofservice
;
1195 struct ofconn
*ofconn
;
1198 dpif_recv_wait(p
->dpif
);
1199 dpif_port_poll_wait(p
->dpif
);
1200 netdev_monitor_poll_wait(p
->netdev_monitor
);
1201 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1202 ofconn_wait(ofconn
);
1205 poll_timer_wait_until(p
->next_in_band_update
);
1206 in_band_wait(p
->in_band
);
1209 fail_open_wait(p
->fail_open
);
1212 ofproto_sflow_wait(p
->sflow
);
1214 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1215 poll_immediate_wake();
1217 if (p
->need_revalidate
) {
1218 /* Shouldn't happen, but if it does just go around again. */
1219 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1220 poll_immediate_wake();
1221 } else if (p
->next_expiration
!= LLONG_MAX
) {
1222 poll_timer_wait_until(p
->next_expiration
);
1224 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1225 pvconn_wait(ofservice
->pvconn
);
1227 for (i
= 0; i
< p
->n_snoops
; i
++) {
1228 pvconn_wait(p
->snoops
[i
]);
1233 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1235 tag_set_add(&ofproto
->revalidate_set
, tag
);
1239 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1241 return &ofproto
->revalidate_set
;
1245 ofproto_is_alive(const struct ofproto
*p
)
1247 return !hmap_is_empty(&p
->controllers
);
1251 ofproto_send_packet(struct ofproto
*p
, const flow_t
*flow
,
1252 const union ofp_action
*actions
, size_t n_actions
,
1253 const struct ofpbuf
*packet
)
1255 struct odp_actions odp_actions
;
1258 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1264 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1266 dpif_execute(p
->dpif
, flow
->in_port
, odp_actions
.actions
,
1267 odp_actions
.n_actions
, packet
);
1272 ofproto_add_flow(struct ofproto
*p
,
1273 const flow_t
*flow
, uint32_t wildcards
, unsigned int priority
,
1274 const union ofp_action
*actions
, size_t n_actions
,
1278 rule
= rule_create(p
, NULL
, actions
, n_actions
,
1279 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */,
1281 cls_rule_from_flow(flow
, wildcards
, priority
, &rule
->cr
);
1282 rule_insert(p
, rule
, NULL
, 0);
1286 ofproto_delete_flow(struct ofproto
*ofproto
, const flow_t
*flow
,
1287 uint32_t wildcards
, unsigned int priority
)
1291 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1295 rule_remove(ofproto
, rule
);
1300 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1302 struct rule
*rule
= rule_from_cls_rule(rule_
);
1303 struct ofproto
*ofproto
= ofproto_
;
1305 /* Mark the flow as not installed, even though it might really be
1306 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1307 * There is no point in uninstalling it individually since we are about to
1308 * blow away all the flows with dpif_flow_flush(). */
1309 rule
->installed
= false;
1311 rule_remove(ofproto
, rule
);
1315 ofproto_flush_flows(struct ofproto
*ofproto
)
1317 COVERAGE_INC(ofproto_flush
);
1318 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1319 dpif_flow_flush(ofproto
->dpif
);
1320 if (ofproto
->in_band
) {
1321 in_band_flushed(ofproto
->in_band
);
1323 if (ofproto
->fail_open
) {
1324 fail_open_flushed(ofproto
->fail_open
);
1329 reinit_ports(struct ofproto
*p
)
1331 struct svec devnames
;
1332 struct ofport
*ofport
;
1333 struct odp_port
*odp_ports
;
1337 svec_init(&devnames
);
1338 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1339 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1341 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1342 for (i
= 0; i
< n_odp_ports
; i
++) {
1343 svec_add (&devnames
, odp_ports
[i
].devname
);
1347 svec_sort_unique(&devnames
);
1348 for (i
= 0; i
< devnames
.n
; i
++) {
1349 update_port(p
, devnames
.names
[i
]);
1351 svec_destroy(&devnames
);
1355 refresh_port_group(struct ofproto
*p
, unsigned int group
)
1359 struct ofport
*port
;
1361 assert(group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
);
1363 ports
= xmalloc(hmap_count(&p
->ports
) * sizeof *ports
);
1365 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
1366 if (group
== DP_GROUP_ALL
|| !(port
->opp
.config
& OFPPC_NO_FLOOD
)) {
1367 ports
[n_ports
++] = port
->odp_port
;
1370 dpif_port_group_set(p
->dpif
, group
, ports
, n_ports
);
1377 refresh_port_groups(struct ofproto
*p
)
1379 size_t n_flood
= refresh_port_group(p
, DP_GROUP_FLOOD
);
1380 size_t n_all
= refresh_port_group(p
, DP_GROUP_ALL
);
1382 ofproto_sflow_set_group_sizes(p
->sflow
, n_flood
, n_all
);
1386 static struct ofport
*
1387 make_ofport(const struct odp_port
*odp_port
)
1389 struct netdev_options netdev_options
;
1390 enum netdev_flags flags
;
1391 struct ofport
*ofport
;
1392 struct netdev
*netdev
;
1396 memset(&netdev_options
, 0, sizeof netdev_options
);
1397 netdev_options
.name
= odp_port
->devname
;
1398 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1400 error
= netdev_open(&netdev_options
, &netdev
);
1402 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1403 "cannot be opened (%s)",
1404 odp_port
->devname
, odp_port
->port
,
1405 odp_port
->devname
, strerror(error
));
1409 ofport
= xmalloc(sizeof *ofport
);
1410 ofport
->netdev
= netdev
;
1411 ofport
->odp_port
= odp_port
->port
;
1412 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1413 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1414 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1415 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1416 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1418 netdev_get_flags(netdev
, &flags
);
1419 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1421 netdev_get_carrier(netdev
, &carrier
);
1422 ofport
->opp
.state
= carrier
? 0 : OFPPS_LINK_DOWN
;
1424 netdev_get_features(netdev
,
1425 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1426 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1431 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1433 if (get_port(p
, odp_port
->port
)) {
1434 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1437 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1438 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1447 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1449 const struct ofp_phy_port
*a
= &a_
->opp
;
1450 const struct ofp_phy_port
*b
= &b_
->opp
;
1452 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1453 return (a
->port_no
== b
->port_no
1454 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1455 && !strcmp((char *) a
->name
, (char *) b
->name
)
1456 && a
->state
== b
->state
1457 && a
->config
== b
->config
1458 && a
->curr
== b
->curr
1459 && a
->advertised
== b
->advertised
1460 && a
->supported
== b
->supported
1461 && a
->peer
== b
->peer
);
1465 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1468 /* XXX Should limit the number of queued port status change messages. */
1469 struct ofconn
*ofconn
;
1470 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1471 struct ofp_port_status
*ops
;
1474 if (!ofconn_receives_async_msgs(ofconn
)) {
1478 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1479 ops
->reason
= reason
;
1480 ops
->desc
= ofport
->opp
;
1481 hton_ofp_phy_port(&ops
->desc
);
1482 queue_tx(b
, ofconn
, NULL
);
1484 if (p
->ofhooks
->port_changed_cb
) {
1485 p
->ofhooks
->port_changed_cb(reason
, &ofport
->opp
, p
->aux
);
1490 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1492 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1494 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1495 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1496 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1498 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1503 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1505 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1506 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1507 shash_delete(&p
->port_by_name
,
1508 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1510 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1515 ofport_free(struct ofport
*ofport
)
1518 netdev_close(ofport
->netdev
);
1523 static struct ofport
*
1524 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1526 struct ofport
*port
;
1528 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1529 hash_int(odp_port
, 0), &ofproto
->ports
) {
1530 if (port
->odp_port
== odp_port
) {
1538 update_port(struct ofproto
*p
, const char *devname
)
1540 struct odp_port odp_port
;
1541 struct ofport
*old_ofport
;
1542 struct ofport
*new_ofport
;
1545 COVERAGE_INC(ofproto_update_port
);
1547 /* Query the datapath for port information. */
1548 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1550 /* Find the old ofport. */
1551 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1554 /* There's no port named 'devname' but there might be a port with
1555 * the same port number. This could happen if a port is deleted
1556 * and then a new one added in its place very quickly, or if a port
1557 * is renamed. In the former case we want to send an OFPPR_DELETE
1558 * and an OFPPR_ADD, and in the latter case we want to send a
1559 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1560 * the old port's ifindex against the new port, or perhaps less
1561 * reliably but more portably by comparing the old port's MAC
1562 * against the new port's MAC. However, this code isn't that smart
1563 * and always sends an OFPPR_MODIFY (XXX). */
1564 old_ofport
= get_port(p
, odp_port
.port
);
1566 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1567 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1568 "%s", strerror(error
));
1572 /* Create a new ofport. */
1573 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1575 /* Eliminate a few pathological cases. */
1576 if (!old_ofport
&& !new_ofport
) {
1578 } else if (old_ofport
&& new_ofport
) {
1579 /* Most of the 'config' bits are OpenFlow soft state, but
1580 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1581 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1582 * leaves the other bits 0.) */
1583 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1585 if (ofport_equal(old_ofport
, new_ofport
)) {
1586 /* False alarm--no change. */
1587 ofport_free(new_ofport
);
1592 /* Now deal with the normal cases. */
1594 ofport_remove(p
, old_ofport
);
1597 ofport_install(p
, new_ofport
);
1599 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1600 (!old_ofport
? OFPPR_ADD
1601 : !new_ofport
? OFPPR_DELETE
1603 ofport_free(old_ofport
);
1605 /* Update port groups. */
1606 refresh_port_groups(p
);
1610 init_ports(struct ofproto
*p
)
1612 struct odp_port
*ports
;
1617 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1622 for (i
= 0; i
< n_ports
; i
++) {
1623 const struct odp_port
*odp_port
= &ports
[i
];
1624 if (!ofport_conflicts(p
, odp_port
)) {
1625 struct ofport
*ofport
= make_ofport(odp_port
);
1627 ofport_install(p
, ofport
);
1632 refresh_port_groups(p
);
1636 static struct ofconn
*
1637 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1639 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1640 ofconn
->ofproto
= p
;
1641 list_push_back(&p
->all_conns
, &ofconn
->node
);
1642 ofconn
->rconn
= rconn
;
1643 ofconn
->type
= type
;
1644 ofconn
->role
= NX_ROLE_OTHER
;
1645 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1646 ofconn
->pktbuf
= NULL
;
1647 ofconn
->miss_send_len
= 0;
1648 ofconn
->reply_counter
= rconn_packet_counter_create ();
1653 ofconn_destroy(struct ofconn
*ofconn
)
1655 if (ofconn
->type
== OFCONN_PRIMARY
) {
1656 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1658 discovery_destroy(ofconn
->discovery
);
1660 list_remove(&ofconn
->node
);
1661 switch_status_unregister(ofconn
->ss
);
1662 rconn_destroy(ofconn
->rconn
);
1663 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1664 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1665 pktbuf_destroy(ofconn
->pktbuf
);
1670 ofconn_run(struct ofconn
*ofconn
, struct ofproto
*p
)
1675 if (ofconn
->discovery
) {
1676 char *controller_name
;
1677 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1678 discovery_question_connectivity(ofconn
->discovery
);
1680 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1681 if (controller_name
) {
1682 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1683 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1686 rconn_disconnect(ofconn
->rconn
);
1691 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1692 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1695 rconn_run(ofconn
->rconn
);
1697 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1698 /* Limit the number of iterations to prevent other tasks from
1700 for (iteration
= 0; iteration
< 50; iteration
++) {
1701 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1706 fail_open_maybe_recover(p
->fail_open
);
1708 handle_openflow(ofconn
, p
, of_msg
);
1709 ofpbuf_delete(of_msg
);
1713 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1714 ofconn_destroy(ofconn
);
1719 ofconn_wait(struct ofconn
*ofconn
)
1723 if (ofconn
->discovery
) {
1724 discovery_wait(ofconn
->discovery
);
1726 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1727 pinsched_wait(ofconn
->schedulers
[i
]);
1729 rconn_run_wait(ofconn
->rconn
);
1730 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1731 rconn_recv_wait(ofconn
->rconn
);
1733 COVERAGE_INC(ofproto_ofconn_stuck
);
1737 /* Returns true if 'ofconn' should receive asynchronous messages. */
1739 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1741 if (ofconn
->type
== OFCONN_PRIMARY
) {
1742 /* Primary controllers always get asynchronous messages unless they
1743 * have configured themselves as "slaves". */
1744 return ofconn
->role
!= NX_ROLE_SLAVE
;
1746 /* Service connections don't get asynchronous messages unless they have
1747 * explicitly asked for them by setting a nonzero miss send length. */
1748 return ofconn
->miss_send_len
> 0;
1752 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1753 * and 'target', suitable for use in log messages for identifying the
1756 * The name is dynamically allocated. The caller should free it (with free())
1757 * when it is no longer needed. */
1759 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1761 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1765 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1769 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1770 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1774 *s
= pinsched_create(rate
, burst
,
1775 ofconn
->ofproto
->switch_status
);
1777 pinsched_set_limits(*s
, rate
, burst
);
1780 pinsched_destroy(*s
);
1787 ofservice_reconfigure(struct ofservice
*ofservice
,
1788 const struct ofproto_controller
*c
)
1790 ofservice
->probe_interval
= c
->probe_interval
;
1791 ofservice
->rate_limit
= c
->rate_limit
;
1792 ofservice
->burst_limit
= c
->burst_limit
;
1795 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1796 * positive errno value. */
1798 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1800 struct ofservice
*ofservice
;
1801 struct pvconn
*pvconn
;
1804 error
= pvconn_open(c
->target
, &pvconn
);
1809 ofservice
= xzalloc(sizeof *ofservice
);
1810 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1811 hash_string(c
->target
, 0));
1812 ofservice
->pvconn
= pvconn
;
1814 ofservice_reconfigure(ofservice
, c
);
1820 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1822 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1823 pvconn_close(ofservice
->pvconn
);
1827 /* Finds and returns the ofservice within 'ofproto' that has the given
1828 * 'target', or a null pointer if none exists. */
1829 static struct ofservice
*
1830 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1832 struct ofservice
*ofservice
;
1834 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1835 &ofproto
->services
) {
1836 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1843 /* Caller is responsible for initializing the 'cr' member of the returned
1845 static struct rule
*
1846 rule_create(struct ofproto
*ofproto
, struct rule
*super
,
1847 const union ofp_action
*actions
, size_t n_actions
,
1848 uint16_t idle_timeout
, uint16_t hard_timeout
,
1849 uint64_t flow_cookie
, bool send_flow_removed
)
1851 struct rule
*rule
= xzalloc(sizeof *rule
);
1852 rule
->idle_timeout
= idle_timeout
;
1853 rule
->hard_timeout
= hard_timeout
;
1854 rule
->flow_cookie
= flow_cookie
;
1855 rule
->used
= rule
->created
= time_msec();
1856 rule
->send_flow_removed
= send_flow_removed
;
1857 rule
->super
= super
;
1859 list_push_back(&super
->list
, &rule
->list
);
1861 list_init(&rule
->list
);
1863 if (n_actions
> 0) {
1864 rule
->n_actions
= n_actions
;
1865 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1867 netflow_flow_clear(&rule
->nf_flow
);
1868 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->created
);
1873 static struct rule
*
1874 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1876 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1880 rule_free(struct rule
*rule
)
1882 free(rule
->actions
);
1883 free(rule
->odp_actions
);
1887 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1888 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1889 * through all of its subrules and revalidates them, destroying any that no
1890 * longer has a super-rule (which is probably all of them).
1892 * Before calling this function, the caller must make have removed 'rule' from
1893 * the classifier. If 'rule' is an exact-match rule, the caller is also
1894 * responsible for ensuring that it has been uninstalled from the datapath. */
1896 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1899 struct rule
*subrule
, *next
;
1900 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
1901 revalidate_rule(ofproto
, subrule
);
1904 list_remove(&rule
->list
);
1910 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1912 const union ofp_action
*oa
;
1913 struct actions_iterator i
;
1915 if (out_port
== htons(OFPP_NONE
)) {
1918 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1919 oa
= actions_next(&i
)) {
1920 if (action_outputs_to_port(oa
, out_port
)) {
1927 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1928 * 'packet', which arrived on 'in_port'.
1930 * Takes ownership of 'packet'. */
1932 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
1933 const union odp_action
*actions
, size_t n_actions
,
1934 struct ofpbuf
*packet
)
1936 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
1937 /* As an optimization, avoid a round-trip from userspace to kernel to
1938 * userspace. This also avoids possibly filling up kernel packet
1939 * buffers along the way. */
1940 struct odp_msg
*msg
;
1942 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
1943 msg
->type
= _ODPL_ACTION_NR
;
1944 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
1945 msg
->port
= in_port
;
1947 msg
->arg
= actions
[0].controller
.arg
;
1949 send_packet_in(ofproto
, packet
);
1955 error
= dpif_execute(ofproto
->dpif
, in_port
,
1956 actions
, n_actions
, packet
);
1957 ofpbuf_delete(packet
);
1962 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1963 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1964 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1966 * The flow that 'packet' actually contains does not need to actually match
1967 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1968 * the packet and byte counters for 'rule' will be credited for the packet sent
1969 * out whether or not the packet actually matches 'rule'.
1971 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1972 * the caller must already have accurately composed ODP actions for it given
1973 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1974 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1975 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1976 * actions and apply them to 'packet'.
1978 * Takes ownership of 'packet'. */
1980 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1981 struct ofpbuf
*packet
, const flow_t
*flow
)
1983 const union odp_action
*actions
;
1984 struct odp_flow_stats stats
;
1986 struct odp_actions a
;
1988 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1990 /* Grab or compose the ODP actions.
1992 * The special case for an exact-match 'rule' where 'flow' is not the
1993 * rule's flow is important to avoid, e.g., sending a packet out its input
1994 * port simply because the ODP actions were composed for the wrong
1996 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
1997 struct rule
*super
= rule
->super
? rule
->super
: rule
;
1998 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
1999 packet
, &a
, NULL
, 0, NULL
)) {
2000 ofpbuf_delete(packet
);
2003 actions
= a
.actions
;
2004 n_actions
= a
.n_actions
;
2006 actions
= rule
->odp_actions
;
2007 n_actions
= rule
->n_odp_actions
;
2010 /* Execute the ODP actions. */
2011 flow_extract_stats(flow
, packet
, &stats
);
2012 if (execute_odp_actions(ofproto
, flow
->in_port
,
2013 actions
, n_actions
, packet
)) {
2014 update_stats(ofproto
, rule
, &stats
);
2015 rule
->used
= time_msec();
2016 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->used
);
2020 /* Inserts 'rule' into 'p''s flow table.
2022 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2023 * actions on it and credits the statistics for sending the packet to 'rule'.
2024 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2027 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
2030 struct rule
*displaced_rule
;
2032 /* Insert the rule in the classifier. */
2033 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2034 if (!rule
->cr
.wc
.wildcards
) {
2035 rule_make_actions(p
, rule
, packet
);
2038 /* Send the packet and credit it to the rule. */
2041 flow_extract(packet
, 0, in_port
, &flow
);
2042 rule_execute(p
, rule
, packet
, &flow
);
2045 /* Install the rule in the datapath only after sending the packet, to
2046 * avoid packet reordering. */
2047 if (rule
->cr
.wc
.wildcards
) {
2048 COVERAGE_INC(ofproto_add_wc_flow
);
2049 p
->need_revalidate
= true;
2051 rule_install(p
, rule
, displaced_rule
);
2054 /* Free the rule that was displaced, if any. */
2055 if (displaced_rule
) {
2056 rule_destroy(p
, displaced_rule
);
2060 static struct rule
*
2061 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
2064 struct rule
*subrule
= rule_create(ofproto
, rule
, NULL
, 0,
2065 rule
->idle_timeout
, rule
->hard_timeout
,
2067 COVERAGE_INC(ofproto_subrule_create
);
2068 cls_rule_from_flow(flow
, 0, (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
2069 : rule
->cr
.priority
), &subrule
->cr
);
2070 classifier_insert_exact(&ofproto
->cls
, &subrule
->cr
);
2076 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2078 if (rule
->cr
.wc
.wildcards
) {
2079 COVERAGE_INC(ofproto_del_wc_flow
);
2080 ofproto
->need_revalidate
= true;
2082 rule_uninstall(ofproto
, rule
);
2084 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2085 rule_destroy(ofproto
, rule
);
2088 /* Returns true if the actions changed, false otherwise. */
2090 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
2091 const struct ofpbuf
*packet
)
2093 const struct rule
*super
;
2094 struct odp_actions a
;
2097 assert(!rule
->cr
.wc
.wildcards
);
2099 super
= rule
->super
? rule
->super
: rule
;
2101 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
2102 packet
, &a
, &rule
->tags
, &rule
->may_install
,
2103 &rule
->nf_flow
.output_iface
);
2105 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2106 if (rule
->n_odp_actions
!= a
.n_actions
2107 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
2108 COVERAGE_INC(ofproto_odp_unchanged
);
2109 free(rule
->odp_actions
);
2110 rule
->n_odp_actions
= a
.n_actions
;
2111 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
2119 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
2120 struct odp_flow_put
*put
)
2122 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2123 put
->flow
.key
= rule
->cr
.flow
;
2124 put
->flow
.actions
= rule
->odp_actions
;
2125 put
->flow
.n_actions
= rule
->n_odp_actions
;
2126 put
->flow
.flags
= 0;
2128 return dpif_flow_put(ofproto
->dpif
, put
);
2132 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
2134 assert(!rule
->cr
.wc
.wildcards
);
2136 if (rule
->may_install
) {
2137 struct odp_flow_put put
;
2138 if (!do_put_flow(p
, rule
,
2139 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
2141 rule
->installed
= true;
2142 if (displaced_rule
) {
2143 update_stats(p
, displaced_rule
, &put
.flow
.stats
);
2144 rule_post_uninstall(p
, displaced_rule
);
2147 } else if (displaced_rule
) {
2148 rule_uninstall(p
, displaced_rule
);
2153 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
2155 if (rule
->installed
) {
2156 struct odp_flow_put put
;
2157 COVERAGE_INC(ofproto_dp_missed
);
2158 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
2160 rule_install(ofproto
, rule
, NULL
);
2165 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
2167 bool actions_changed
;
2168 uint16_t new_out_iface
, old_out_iface
;
2170 old_out_iface
= rule
->nf_flow
.output_iface
;
2171 actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
2173 if (rule
->may_install
) {
2174 if (rule
->installed
) {
2175 if (actions_changed
) {
2176 struct odp_flow_put put
;
2177 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
2178 | ODPPF_ZERO_STATS
, &put
);
2179 update_stats(ofproto
, rule
, &put
.flow
.stats
);
2181 /* Temporarily set the old output iface so that NetFlow
2182 * messages have the correct output interface for the old
2184 new_out_iface
= rule
->nf_flow
.output_iface
;
2185 rule
->nf_flow
.output_iface
= old_out_iface
;
2186 rule_post_uninstall(ofproto
, rule
);
2187 rule
->nf_flow
.output_iface
= new_out_iface
;
2190 rule_install(ofproto
, rule
, NULL
);
2193 rule_uninstall(ofproto
, rule
);
2198 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
2200 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
2202 if (ofproto
->ofhooks
->account_flow_cb
2203 && total_bytes
> rule
->accounted_bytes
)
2205 ofproto
->ofhooks
->account_flow_cb(
2206 &rule
->cr
.flow
, rule
->tags
, rule
->odp_actions
, rule
->n_odp_actions
,
2207 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
2208 rule
->accounted_bytes
= total_bytes
;
2213 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
2215 assert(!rule
->cr
.wc
.wildcards
);
2216 if (rule
->installed
) {
2217 struct odp_flow odp_flow
;
2219 odp_flow
.key
= rule
->cr
.flow
;
2220 odp_flow
.actions
= NULL
;
2221 odp_flow
.n_actions
= 0;
2223 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2224 update_stats(p
, rule
, &odp_flow
.stats
);
2226 rule
->installed
= false;
2228 rule_post_uninstall(p
, rule
);
2233 is_controller_rule(struct rule
*rule
)
2235 /* If the only action is send to the controller then don't report
2236 * NetFlow expiration messages since it is just part of the control
2237 * logic for the network and not real traffic. */
2241 && rule
->super
->n_actions
== 1
2242 && action_outputs_to_port(&rule
->super
->actions
[0],
2243 htons(OFPP_CONTROLLER
)));
2247 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
2249 struct rule
*super
= rule
->super
;
2251 rule_account(ofproto
, rule
, 0);
2253 if (ofproto
->netflow
&& !is_controller_rule(rule
)) {
2254 struct ofexpired expired
;
2255 expired
.flow
= rule
->cr
.flow
;
2256 expired
.packet_count
= rule
->packet_count
;
2257 expired
.byte_count
= rule
->byte_count
;
2258 expired
.used
= rule
->used
;
2259 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
2262 super
->packet_count
+= rule
->packet_count
;
2263 super
->byte_count
+= rule
->byte_count
;
2265 /* Reset counters to prevent double counting if the rule ever gets
2267 rule
->packet_count
= 0;
2268 rule
->byte_count
= 0;
2269 rule
->accounted_bytes
= 0;
2271 netflow_flow_clear(&rule
->nf_flow
);
2276 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2277 struct rconn_packet_counter
*counter
)
2279 update_openflow_length(msg
);
2280 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2286 send_error(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2287 int error
, const void *data
, size_t len
)
2290 struct ofp_error_msg
*oem
;
2292 if (!(error
>> 16)) {
2293 VLOG_WARN_RL(&rl
, "not sending bad error code %d to controller",
2298 COVERAGE_INC(ofproto_error
);
2299 oem
= make_openflow_xid(len
+ sizeof *oem
, OFPT_ERROR
,
2300 oh
? oh
->xid
: 0, &buf
);
2301 oem
->type
= htons((unsigned int) error
>> 16);
2302 oem
->code
= htons(error
& 0xffff);
2303 memcpy(oem
->data
, data
, len
);
2304 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2308 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2311 size_t oh_length
= ntohs(oh
->length
);
2312 send_error(ofconn
, oh
, error
, oh
, MIN(oh_length
, 64));
2316 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2318 opp
->port_no
= htons(opp
->port_no
);
2319 opp
->config
= htonl(opp
->config
);
2320 opp
->state
= htonl(opp
->state
);
2321 opp
->curr
= htonl(opp
->curr
);
2322 opp
->advertised
= htonl(opp
->advertised
);
2323 opp
->supported
= htonl(opp
->supported
);
2324 opp
->peer
= htonl(opp
->peer
);
2328 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2330 struct ofp_header
*rq
= oh
;
2331 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2336 handle_features_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2337 struct ofp_header
*oh
)
2339 struct ofp_switch_features
*osf
;
2341 struct ofport
*port
;
2343 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2344 osf
->datapath_id
= htonll(p
->datapath_id
);
2345 osf
->n_buffers
= htonl(pktbuf_capacity());
2347 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2348 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2349 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2350 (1u << OFPAT_SET_VLAN_VID
) |
2351 (1u << OFPAT_SET_VLAN_PCP
) |
2352 (1u << OFPAT_STRIP_VLAN
) |
2353 (1u << OFPAT_SET_DL_SRC
) |
2354 (1u << OFPAT_SET_DL_DST
) |
2355 (1u << OFPAT_SET_NW_SRC
) |
2356 (1u << OFPAT_SET_NW_DST
) |
2357 (1u << OFPAT_SET_NW_TOS
) |
2358 (1u << OFPAT_SET_TP_SRC
) |
2359 (1u << OFPAT_SET_TP_DST
) |
2360 (1u << OFPAT_ENQUEUE
));
2362 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2363 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2366 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2371 handle_get_config_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2372 struct ofp_header
*oh
)
2375 struct ofp_switch_config
*osc
;
2379 /* Figure out flags. */
2380 dpif_get_drop_frags(p
->dpif
, &drop_frags
);
2381 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2384 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2385 osc
->flags
= htons(flags
);
2386 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2387 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2393 handle_set_config(struct ofproto
*p
, struct ofconn
*ofconn
,
2394 struct ofp_switch_config
*osc
)
2399 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2403 flags
= ntohs(osc
->flags
);
2405 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2406 switch (flags
& OFPC_FRAG_MASK
) {
2407 case OFPC_FRAG_NORMAL
:
2408 dpif_set_drop_frags(p
->dpif
, false);
2410 case OFPC_FRAG_DROP
:
2411 dpif_set_drop_frags(p
->dpif
, true);
2414 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2420 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2426 add_output_group_action(struct odp_actions
*actions
, uint16_t group
,
2427 uint16_t *nf_output_iface
)
2429 odp_actions_add(actions
, ODPAT_OUTPUT_GROUP
)->output_group
.group
= group
;
2431 if (group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
) {
2432 *nf_output_iface
= NF_OUT_FLOOD
;
2437 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2439 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2440 a
->controller
.arg
= max_len
;
2443 struct action_xlate_ctx
{
2445 flow_t flow
; /* Flow to which these actions correspond. */
2446 int recurse
; /* Recursion level, via xlate_table_action. */
2447 struct ofproto
*ofproto
;
2448 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2449 * null pointer if we are revalidating
2450 * without a packet to refer to. */
2453 struct odp_actions
*out
; /* Datapath actions. */
2454 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
2455 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2456 * be reassessed for every packet. */
2457 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2460 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2461 * flow translation. */
2462 #define MAX_RESUBMIT_RECURSION 8
2464 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2465 struct action_xlate_ctx
*ctx
);
2468 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2470 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2473 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2474 /* Forwarding disabled on port. */
2479 * We don't have an ofport record for this port, but it doesn't hurt to
2480 * allow forwarding to it anyhow. Maybe such a port will appear later
2481 * and we're pre-populating the flow table.
2485 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2486 ctx
->nf_output_iface
= port
;
2489 static struct rule
*
2490 lookup_valid_rule(struct ofproto
*ofproto
, const flow_t
*flow
)
2493 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2495 /* The rule we found might not be valid, since we could be in need of
2496 * revalidation. If it is not valid, don't return it. */
2499 && ofproto
->need_revalidate
2500 && !revalidate_rule(ofproto
, rule
)) {
2501 COVERAGE_INC(ofproto_invalidated
);
2509 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2511 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2512 uint16_t old_in_port
;
2515 /* Look up a flow with 'in_port' as the input port. Then restore the
2516 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2517 * have surprising behavior). */
2518 old_in_port
= ctx
->flow
.in_port
;
2519 ctx
->flow
.in_port
= in_port
;
2520 rule
= lookup_valid_rule(ctx
->ofproto
, &ctx
->flow
);
2521 ctx
->flow
.in_port
= old_in_port
;
2529 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2533 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2535 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2536 MAX_RESUBMIT_RECURSION
);
2541 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2542 uint16_t port
, uint16_t max_len
)
2545 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2547 ctx
->nf_output_iface
= NF_OUT_DROP
;
2551 add_output_action(ctx
, ctx
->flow
.in_port
);
2554 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2557 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2558 ctx
->out
, ctx
->tags
,
2559 &ctx
->nf_output_iface
,
2560 ctx
->ofproto
->aux
)) {
2561 COVERAGE_INC(ofproto_uninstallable
);
2562 ctx
->may_set_up_flow
= false;
2566 add_output_group_action(ctx
->out
, DP_GROUP_FLOOD
,
2567 &ctx
->nf_output_iface
);
2570 add_output_group_action(ctx
->out
, DP_GROUP_ALL
, &ctx
->nf_output_iface
);
2572 case OFPP_CONTROLLER
:
2573 add_controller_action(ctx
->out
, max_len
);
2576 add_output_action(ctx
, ODPP_LOCAL
);
2579 odp_port
= ofp_port_to_odp_port(port
);
2580 if (odp_port
!= ctx
->flow
.in_port
) {
2581 add_output_action(ctx
, odp_port
);
2586 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2587 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2588 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2589 ctx
->nf_output_iface
= prev_nf_output_iface
;
2590 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2591 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2592 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2597 xlate_output_action(struct action_xlate_ctx
*ctx
,
2598 const struct ofp_action_output
*oao
)
2600 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2603 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2604 * optimization, because we're going to add another action that sets the
2605 * priority immediately after, or because there are no actions following the
2608 remove_pop_action(struct action_xlate_ctx
*ctx
)
2610 size_t n
= ctx
->out
->n_actions
;
2611 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2612 ctx
->out
->n_actions
--;
2617 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2618 const struct ofp_action_enqueue
*oae
)
2620 uint16_t ofp_port
, odp_port
;
2624 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2627 /* Fall back to ordinary output action. */
2628 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2632 /* Figure out ODP output port. */
2633 ofp_port
= ntohs(oae
->port
);
2634 if (ofp_port
!= OFPP_IN_PORT
) {
2635 odp_port
= ofp_port_to_odp_port(ofp_port
);
2637 odp_port
= ctx
->flow
.in_port
;
2640 /* Add ODP actions. */
2641 remove_pop_action(ctx
);
2642 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2644 add_output_action(ctx
, odp_port
);
2645 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2647 /* Update NetFlow output port. */
2648 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2649 ctx
->nf_output_iface
= odp_port
;
2650 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2651 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2656 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2657 const struct nx_action_set_queue
*nasq
)
2662 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2665 /* Couldn't translate queue to a priority, so ignore. A warning
2666 * has already been logged. */
2670 remove_pop_action(ctx
);
2671 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2676 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2677 const struct nx_action_header
*nah
)
2679 const struct nx_action_resubmit
*nar
;
2680 const struct nx_action_set_tunnel
*nast
;
2681 const struct nx_action_set_queue
*nasq
;
2682 union odp_action
*oa
;
2683 int subtype
= ntohs(nah
->subtype
);
2685 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2687 case NXAST_RESUBMIT
:
2688 nar
= (const struct nx_action_resubmit
*) nah
;
2689 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2692 case NXAST_SET_TUNNEL
:
2693 nast
= (const struct nx_action_set_tunnel
*) nah
;
2694 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2695 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2698 case NXAST_DROP_SPOOFED_ARP
:
2699 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2700 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2704 case NXAST_SET_QUEUE
:
2705 nasq
= (const struct nx_action_set_queue
*) nah
;
2706 xlate_set_queue_action(ctx
, nasq
);
2709 case NXAST_POP_QUEUE
:
2710 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2713 /* If you add a new action here that modifies flow data, don't forget to
2714 * update the flow key in ctx->flow at the same time. */
2717 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2723 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2724 struct action_xlate_ctx
*ctx
)
2726 struct actions_iterator iter
;
2727 const union ofp_action
*ia
;
2728 const struct ofport
*port
;
2730 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2731 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2732 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2733 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2734 /* Drop this flow. */
2738 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2739 uint16_t type
= ntohs(ia
->type
);
2740 union odp_action
*oa
;
2744 xlate_output_action(ctx
, &ia
->output
);
2747 case OFPAT_SET_VLAN_VID
:
2748 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_VID
);
2749 ctx
->flow
.dl_vlan
= oa
->vlan_vid
.vlan_vid
= ia
->vlan_vid
.vlan_vid
;
2752 case OFPAT_SET_VLAN_PCP
:
2753 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_PCP
);
2754 ctx
->flow
.dl_vlan_pcp
= oa
->vlan_pcp
.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 flow_t
*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 * 'oh' is used to make log messages more informative. */
2867 reject_slave_controller(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
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);
2873 type_name
= ofp_message_type_to_string(oh
->type
);
2874 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
2878 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
2885 handle_packet_out(struct ofproto
*p
, struct ofconn
*ofconn
,
2886 struct ofp_header
*oh
)
2888 struct ofp_packet_out
*opo
;
2889 struct ofpbuf payload
, *buffer
;
2890 struct odp_actions actions
;
2896 error
= reject_slave_controller(ofconn
, oh
);
2901 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2905 opo
= (struct ofp_packet_out
*) oh
;
2907 COVERAGE_INC(ofproto_packet_out
);
2908 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2909 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2911 if (error
|| !buffer
) {
2919 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2920 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2921 &flow
, p
, &payload
, &actions
, NULL
, NULL
, NULL
);
2926 dpif_execute(p
->dpif
, flow
.in_port
, actions
.actions
, actions
.n_actions
,
2928 ofpbuf_delete(buffer
);
2934 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2935 uint32_t config
, uint32_t mask
)
2937 mask
&= config
^ port
->opp
.config
;
2938 if (mask
& OFPPC_PORT_DOWN
) {
2939 if (config
& OFPPC_PORT_DOWN
) {
2940 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2942 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2945 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2946 if (mask
& REVALIDATE_BITS
) {
2947 COVERAGE_INC(ofproto_costly_flags
);
2948 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2949 p
->need_revalidate
= true;
2951 #undef REVALIDATE_BITS
2952 if (mask
& OFPPC_NO_FLOOD
) {
2953 port
->opp
.config
^= OFPPC_NO_FLOOD
;
2954 refresh_port_groups(p
);
2956 if (mask
& OFPPC_NO_PACKET_IN
) {
2957 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2962 handle_port_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
2963 struct ofp_header
*oh
)
2965 const struct ofp_port_mod
*opm
;
2966 struct ofport
*port
;
2969 error
= reject_slave_controller(ofconn
, oh
);
2973 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2977 opm
= (struct ofp_port_mod
*) oh
;
2979 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2981 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2982 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2983 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2985 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2986 if (opm
->advertise
) {
2987 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2993 static struct ofpbuf
*
2994 make_stats_reply(uint32_t xid
, uint16_t type
, size_t body_len
)
2996 struct ofp_stats_reply
*osr
;
2999 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3000 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3002 osr
->flags
= htons(0);
3006 static struct ofpbuf
*
3007 start_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
3009 return make_stats_reply(request
->header
.xid
, request
->type
, body_len
);
3013 append_stats_reply(size_t nbytes
, struct ofconn
*ofconn
, struct ofpbuf
**msgp
)
3015 struct ofpbuf
*msg
= *msgp
;
3016 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3017 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3018 struct ofp_stats_reply
*reply
= msg
->data
;
3019 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3020 *msgp
= make_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3021 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3023 return ofpbuf_put_uninit(*msgp
, nbytes
);
3027 handle_desc_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3028 struct ofp_stats_request
*request
)
3030 struct ofp_desc_stats
*ods
;
3033 msg
= start_stats_reply(request
, sizeof *ods
);
3034 ods
= append_stats_reply(sizeof *ods
, ofconn
, &msg
);
3035 memset(ods
, 0, sizeof *ods
);
3036 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3037 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3038 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3039 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3040 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3041 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3047 count_subrules(struct cls_rule
*cls_rule
, void *n_subrules_
)
3049 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
3050 int *n_subrules
= n_subrules_
;
3058 handle_table_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3059 struct ofp_stats_request
*request
)
3061 struct ofp_table_stats
*ots
;
3063 struct odp_stats dpstats
;
3064 int n_exact
, n_subrules
, n_wild
;
3066 msg
= start_stats_reply(request
, sizeof *ots
* 2);
3068 /* Count rules of various kinds. */
3070 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, count_subrules
, &n_subrules
);
3071 n_exact
= classifier_count_exact(&p
->cls
) - n_subrules
;
3072 n_wild
= classifier_count(&p
->cls
) - classifier_count_exact(&p
->cls
);
3075 dpif_get_dp_stats(p
->dpif
, &dpstats
);
3076 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
3077 memset(ots
, 0, sizeof *ots
);
3078 ots
->table_id
= TABLEID_HASH
;
3079 strcpy(ots
->name
, "hash");
3080 ots
->wildcards
= htonl(0);
3081 ots
->max_entries
= htonl(dpstats
.max_capacity
);
3082 ots
->active_count
= htonl(n_exact
);
3083 ots
->lookup_count
= htonll(dpstats
.n_frags
+ dpstats
.n_hit
+
3085 ots
->matched_count
= htonll(dpstats
.n_hit
); /* XXX */
3087 /* Classifier table. */
3088 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
3089 memset(ots
, 0, sizeof *ots
);
3090 ots
->table_id
= TABLEID_CLASSIFIER
;
3091 strcpy(ots
->name
, "classifier");
3092 ots
->wildcards
= p
->tun_id_from_cookie
? htonl(OVSFW_ALL
)
3094 ots
->max_entries
= htonl(65536);
3095 ots
->active_count
= htonl(n_wild
);
3096 ots
->lookup_count
= htonll(0); /* XXX */
3097 ots
->matched_count
= htonll(0); /* XXX */
3099 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3104 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3105 struct ofpbuf
**msgp
)
3107 struct netdev_stats stats
;
3108 struct ofp_port_stats
*ops
;
3110 /* Intentionally ignore return value, since errors will set
3111 * 'stats' to all-1s, which is correct for OpenFlow, and
3112 * netdev_get_stats() will log errors. */
3113 netdev_get_stats(port
->netdev
, &stats
);
3115 ops
= append_stats_reply(sizeof *ops
, ofconn
, msgp
);
3116 ops
->port_no
= htons(port
->opp
.port_no
);
3117 memset(ops
->pad
, 0, sizeof ops
->pad
);
3118 ops
->rx_packets
= htonll(stats
.rx_packets
);
3119 ops
->tx_packets
= htonll(stats
.tx_packets
);
3120 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3121 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3122 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3123 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3124 ops
->rx_errors
= htonll(stats
.rx_errors
);
3125 ops
->tx_errors
= htonll(stats
.tx_errors
);
3126 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3127 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3128 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3129 ops
->collisions
= htonll(stats
.collisions
);
3133 handle_port_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3134 struct ofp_stats_request
*osr
,
3137 struct ofp_port_stats_request
*psr
;
3138 struct ofp_port_stats
*ops
;
3140 struct ofport
*port
;
3142 if (arg_size
!= sizeof *psr
) {
3143 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3145 psr
= (struct ofp_port_stats_request
*) osr
->body
;
3147 msg
= start_stats_reply(osr
, sizeof *ops
* 16);
3148 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3149 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3151 append_port_stat(port
, ofconn
, &msg
);
3154 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3155 append_port_stat(port
, ofconn
, &msg
);
3159 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3163 struct flow_stats_cbdata
{
3164 struct ofproto
*ofproto
;
3165 struct ofconn
*ofconn
;
3170 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3171 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3172 * returned statistic include statistics for all of 'rule''s subrules. */
3174 query_stats(struct ofproto
*p
, struct rule
*rule
,
3175 uint64_t *packet_countp
, uint64_t *byte_countp
)
3177 uint64_t packet_count
, byte_count
;
3178 struct rule
*subrule
;
3179 struct odp_flow
*odp_flows
;
3182 /* Start from historical data for 'rule' itself that are no longer tracked
3183 * by the datapath. This counts, for example, subrules that have
3185 packet_count
= rule
->packet_count
;
3186 byte_count
= rule
->byte_count
;
3188 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3189 * wildcarded then on all of its subrules.
3191 * Also, add any statistics that are not tracked by the datapath for each
3192 * subrule. This includes, for example, statistics for packets that were
3193 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3195 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
3196 odp_flows
= xzalloc(n_odp_flows
* sizeof *odp_flows
);
3197 if (rule
->cr
.wc
.wildcards
) {
3199 LIST_FOR_EACH (subrule
, list
, &rule
->list
) {
3200 odp_flows
[i
++].key
= subrule
->cr
.flow
;
3201 packet_count
+= subrule
->packet_count
;
3202 byte_count
+= subrule
->byte_count
;
3205 odp_flows
[0].key
= rule
->cr
.flow
;
3208 /* Fetch up-to-date statistics from the datapath and add them in. */
3209 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3211 for (i
= 0; i
< n_odp_flows
; i
++) {
3212 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3213 packet_count
+= odp_flow
->stats
.n_packets
;
3214 byte_count
+= odp_flow
->stats
.n_bytes
;
3219 /* Return the stats to the caller. */
3220 *packet_countp
= packet_count
;
3221 *byte_countp
= byte_count
;
3225 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3227 struct rule
*rule
= rule_from_cls_rule(rule_
);
3228 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3229 struct ofp_flow_stats
*ofs
;
3230 uint64_t packet_count
, byte_count
;
3231 size_t act_len
, len
;
3232 long long int tdiff
= time_msec() - rule
->created
;
3233 uint32_t sec
= tdiff
/ 1000;
3234 uint32_t msec
= tdiff
- (sec
* 1000);
3236 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3240 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3241 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3243 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3245 ofs
= append_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
3246 ofs
->length
= htons(len
);
3247 ofs
->table_id
= rule
->cr
.wc
.wildcards
? TABLEID_CLASSIFIER
: TABLEID_HASH
;
3249 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3250 cbdata
->ofproto
->tun_id_from_cookie
, &ofs
->match
);
3251 ofs
->duration_sec
= htonl(sec
);
3252 ofs
->duration_nsec
= htonl(msec
* 1000000);
3253 ofs
->cookie
= rule
->flow_cookie
;
3254 ofs
->priority
= htons(rule
->cr
.priority
);
3255 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3256 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3257 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3258 ofs
->packet_count
= htonll(packet_count
);
3259 ofs
->byte_count
= htonll(byte_count
);
3260 if (rule
->n_actions
> 0) {
3261 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3266 table_id_to_include(uint8_t table_id
)
3268 return (table_id
== TABLEID_HASH
? CLS_INC_EXACT
3269 : table_id
== TABLEID_CLASSIFIER
? CLS_INC_WILD
3270 : table_id
== 0xff ? CLS_INC_ALL
3275 handle_flow_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3276 const struct ofp_stats_request
*osr
,
3279 struct ofp_flow_stats_request
*fsr
;
3280 struct flow_stats_cbdata cbdata
;
3281 struct cls_rule target
;
3283 if (arg_size
!= sizeof *fsr
) {
3284 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3286 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3288 COVERAGE_INC(ofproto_flows_req
);
3290 cbdata
.ofconn
= ofconn
;
3291 cbdata
.out_port
= fsr
->out_port
;
3292 cbdata
.msg
= start_stats_reply(osr
, 1024);
3293 cls_rule_from_match(&fsr
->match
, 0, false, 0, &target
);
3294 classifier_for_each_match(&p
->cls
, &target
,
3295 table_id_to_include(fsr
->table_id
),
3296 flow_stats_cb
, &cbdata
);
3297 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3301 struct flow_stats_ds_cbdata
{
3302 struct ofproto
*ofproto
;
3307 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
3309 struct rule
*rule
= rule_from_cls_rule(rule_
);
3310 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
3311 struct ds
*results
= cbdata
->results
;
3312 struct ofp_match match
;
3313 uint64_t packet_count
, byte_count
;
3314 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3316 /* Don't report on subrules. */
3317 if (rule
->super
!= NULL
) {
3321 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3322 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3323 cbdata
->ofproto
->tun_id_from_cookie
, &match
);
3325 ds_put_format(results
, "duration=%llds, ",
3326 (time_msec() - rule
->created
) / 1000);
3327 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3328 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3329 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3330 ofp_print_match(results
, &match
, true);
3332 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3334 ds_put_cstr(results
, "\n");
3337 /* Adds a pretty-printed description of all flows to 'results', including
3338 * those marked hidden by secchan (e.g., by in-band control). */
3340 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3342 struct ofp_match match
;
3343 struct cls_rule target
;
3344 struct flow_stats_ds_cbdata cbdata
;
3346 memset(&match
, 0, sizeof match
);
3347 match
.wildcards
= htonl(OVSFW_ALL
);
3350 cbdata
.results
= results
;
3352 cls_rule_from_match(&match
, 0, false, 0, &target
);
3353 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3354 flow_stats_ds_cb
, &cbdata
);
3357 struct aggregate_stats_cbdata
{
3358 struct ofproto
*ofproto
;
3360 uint64_t packet_count
;
3361 uint64_t byte_count
;
3366 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3368 struct rule
*rule
= rule_from_cls_rule(rule_
);
3369 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
3370 uint64_t packet_count
, byte_count
;
3372 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3376 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3378 cbdata
->packet_count
+= packet_count
;
3379 cbdata
->byte_count
+= byte_count
;
3384 handle_aggregate_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3385 const struct ofp_stats_request
*osr
,
3388 struct ofp_aggregate_stats_request
*asr
;
3389 struct ofp_aggregate_stats_reply
*reply
;
3390 struct aggregate_stats_cbdata cbdata
;
3391 struct cls_rule target
;
3394 if (arg_size
!= sizeof *asr
) {
3395 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3397 asr
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3399 COVERAGE_INC(ofproto_agg_request
);
3401 cbdata
.out_port
= asr
->out_port
;
3402 cbdata
.packet_count
= 0;
3403 cbdata
.byte_count
= 0;
3405 cls_rule_from_match(&asr
->match
, 0, false, 0, &target
);
3406 classifier_for_each_match(&p
->cls
, &target
,
3407 table_id_to_include(asr
->table_id
),
3408 aggregate_stats_cb
, &cbdata
);
3410 msg
= start_stats_reply(osr
, sizeof *reply
);
3411 reply
= append_stats_reply(sizeof *reply
, ofconn
, &msg
);
3412 reply
->flow_count
= htonl(cbdata
.n_flows
);
3413 reply
->packet_count
= htonll(cbdata
.packet_count
);
3414 reply
->byte_count
= htonll(cbdata
.byte_count
);
3415 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3419 struct queue_stats_cbdata
{
3420 struct ofconn
*ofconn
;
3421 struct ofport
*ofport
;
3426 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3427 const struct netdev_queue_stats
*stats
)
3429 struct ofp_queue_stats
*reply
;
3431 reply
= append_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3432 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3433 memset(reply
->pad
, 0, sizeof reply
->pad
);
3434 reply
->queue_id
= htonl(queue_id
);
3435 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3436 reply
->tx_packets
= htonll(stats
->tx_packets
);
3437 reply
->tx_errors
= htonll(stats
->tx_errors
);
3441 handle_queue_stats_dump_cb(uint32_t queue_id
,
3442 struct netdev_queue_stats
*stats
,
3445 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3447 put_queue_stats(cbdata
, queue_id
, stats
);
3451 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3452 struct queue_stats_cbdata
*cbdata
)
3454 cbdata
->ofport
= port
;
3455 if (queue_id
== OFPQ_ALL
) {
3456 netdev_dump_queue_stats(port
->netdev
,
3457 handle_queue_stats_dump_cb
, cbdata
);
3459 struct netdev_queue_stats stats
;
3461 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3462 put_queue_stats(cbdata
, queue_id
, &stats
);
3468 handle_queue_stats_request(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3469 const struct ofp_stats_request
*osr
,
3472 struct ofp_queue_stats_request
*qsr
;
3473 struct queue_stats_cbdata cbdata
;
3474 struct ofport
*port
;
3475 unsigned int port_no
;
3478 if (arg_size
!= sizeof *qsr
) {
3479 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3481 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3483 COVERAGE_INC(ofproto_queue_req
);
3485 cbdata
.ofconn
= ofconn
;
3486 cbdata
.msg
= start_stats_reply(osr
, 128);
3488 port_no
= ntohs(qsr
->port_no
);
3489 queue_id
= ntohl(qsr
->queue_id
);
3490 if (port_no
== OFPP_ALL
) {
3491 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3492 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3494 } else if (port_no
< ofproto
->max_ports
) {
3495 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3497 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3500 ofpbuf_delete(cbdata
.msg
);
3501 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3503 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3509 handle_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3510 struct ofp_header
*oh
)
3512 struct ofp_stats_request
*osr
;
3516 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3521 osr
= (struct ofp_stats_request
*) oh
;
3523 switch (ntohs(osr
->type
)) {
3525 return handle_desc_stats_request(p
, ofconn
, osr
);
3528 return handle_flow_stats_request(p
, ofconn
, osr
, arg_size
);
3530 case OFPST_AGGREGATE
:
3531 return handle_aggregate_stats_request(p
, ofconn
, osr
, arg_size
);
3534 return handle_table_stats_request(p
, ofconn
, osr
);
3537 return handle_port_stats_request(p
, ofconn
, osr
, arg_size
);
3540 return handle_queue_stats_request(p
, ofconn
, osr
, arg_size
);
3543 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3546 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3550 static long long int
3551 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3553 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3557 update_time(struct ofproto
*ofproto
, struct rule
*rule
,
3558 const struct odp_flow_stats
*stats
)
3560 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3561 if (used
> rule
->used
) {
3563 if (rule
->super
&& used
> rule
->super
->used
) {
3564 rule
->super
->used
= used
;
3566 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, used
);
3571 update_stats(struct ofproto
*ofproto
, struct rule
*rule
,
3572 const struct odp_flow_stats
*stats
)
3574 if (stats
->n_packets
) {
3575 update_time(ofproto
, rule
, stats
);
3576 rule
->packet_count
+= stats
->n_packets
;
3577 rule
->byte_count
+= stats
->n_bytes
;
3578 netflow_flow_update_flags(&rule
->nf_flow
, stats
->tcp_flags
);
3582 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3583 * in which no matching flow already exists in the flow table.
3585 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3586 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3587 * code as encoded by ofp_mkerr() on failure.
3589 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3592 add_flow(struct ofproto
*p
, struct ofconn
*ofconn
,
3593 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3595 struct ofpbuf
*packet
;
3600 if (ofm
->flags
& htons(OFPFF_CHECK_OVERLAP
)) {
3604 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3606 if (classifier_rule_overlaps(&p
->cls
, &flow
, wildcards
,
3607 ntohs(ofm
->priority
))) {
3608 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3612 rule
= rule_create(p
, NULL
, (const union ofp_action
*) ofm
->actions
,
3613 n_actions
, ntohs(ofm
->idle_timeout
),
3614 ntohs(ofm
->hard_timeout
), ofm
->cookie
,
3615 ofm
->flags
& htons(OFPFF_SEND_FLOW_REM
));
3616 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
),
3617 p
->tun_id_from_cookie
, ofm
->cookie
, &rule
->cr
);
3620 if (ofm
->buffer_id
!= htonl(UINT32_MAX
)) {
3621 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3625 in_port
= UINT16_MAX
;
3628 rule_insert(p
, rule
, packet
, in_port
);
3632 static struct rule
*
3633 find_flow_strict(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3638 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3640 return rule_from_cls_rule(classifier_find_rule_exactly(
3641 &p
->cls
, &flow
, wildcards
,
3642 ntohs(ofm
->priority
)));
3646 send_buffered_packet(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3647 struct rule
*rule
, const struct ofp_flow_mod
*ofm
)
3649 struct ofpbuf
*packet
;
3654 if (ofm
->buffer_id
== htonl(UINT32_MAX
)) {
3658 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3664 flow_extract(packet
, 0, in_port
, &flow
);
3665 rule_execute(ofproto
, rule
, packet
, &flow
);
3670 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3672 struct modify_flows_cbdata
{
3673 struct ofproto
*ofproto
;
3674 const struct ofp_flow_mod
*ofm
;
3679 static int modify_flow(struct ofproto
*, const struct ofp_flow_mod
*,
3680 size_t n_actions
, struct rule
*);
3681 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3683 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3684 * encoded by ofp_mkerr() on failure.
3686 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3689 modify_flows_loose(struct ofproto
*p
, struct ofconn
*ofconn
,
3690 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3692 struct modify_flows_cbdata cbdata
;
3693 struct cls_rule target
;
3697 cbdata
.n_actions
= n_actions
;
3698 cbdata
.match
= NULL
;
3700 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3703 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3704 modify_flows_cb
, &cbdata
);
3706 /* This credits the packet to whichever flow happened to happened to
3707 * match last. That's weird. Maybe we should do a lookup for the
3708 * flow that actually matches the packet? Who knows. */
3709 send_buffered_packet(p
, ofconn
, cbdata
.match
, ofm
);
3712 return add_flow(p
, ofconn
, ofm
, n_actions
);
3716 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3717 * code as encoded by ofp_mkerr() on failure.
3719 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3722 modify_flow_strict(struct ofproto
*p
, struct ofconn
*ofconn
,
3723 struct ofp_flow_mod
*ofm
, size_t n_actions
)
3725 struct rule
*rule
= find_flow_strict(p
, ofm
);
3726 if (rule
&& !rule_is_hidden(rule
)) {
3727 modify_flow(p
, ofm
, n_actions
, rule
);
3728 return send_buffered_packet(p
, ofconn
, rule
, ofm
);
3730 return add_flow(p
, ofconn
, ofm
, n_actions
);
3734 /* Callback for modify_flows_loose(). */
3736 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3738 struct rule
*rule
= rule_from_cls_rule(rule_
);
3739 struct modify_flows_cbdata
*cbdata
= cbdata_
;
3741 if (!rule_is_hidden(rule
)) {
3742 cbdata
->match
= rule
;
3743 modify_flow(cbdata
->ofproto
, cbdata
->ofm
, cbdata
->n_actions
, rule
);
3747 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3748 * been identified as a flow in 'p''s flow table to be modified, by changing
3749 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3750 * ofp_action[] structures). */
3752 modify_flow(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
3753 size_t n_actions
, struct rule
*rule
)
3755 size_t actions_len
= n_actions
* sizeof *rule
->actions
;
3757 rule
->flow_cookie
= ofm
->cookie
;
3759 /* If the actions are the same, do nothing. */
3760 if (n_actions
== rule
->n_actions
3761 && (!n_actions
|| !memcmp(ofm
->actions
, rule
->actions
, actions_len
)))
3766 /* Replace actions. */
3767 free(rule
->actions
);
3768 rule
->actions
= n_actions
? xmemdup(ofm
->actions
, actions_len
) : NULL
;
3769 rule
->n_actions
= n_actions
;
3771 /* Make sure that the datapath gets updated properly. */
3772 if (rule
->cr
.wc
.wildcards
) {
3773 COVERAGE_INC(ofproto_mod_wc_flow
);
3774 p
->need_revalidate
= true;
3776 rule_update_actions(p
, rule
);
3782 /* OFPFC_DELETE implementation. */
3784 struct delete_flows_cbdata
{
3785 struct ofproto
*ofproto
;
3789 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
3790 static void delete_flow(struct ofproto
*, struct rule
*, uint16_t out_port
);
3792 /* Implements OFPFC_DELETE. */
3794 delete_flows_loose(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3796 struct delete_flows_cbdata cbdata
;
3797 struct cls_rule target
;
3800 cbdata
.out_port
= ofm
->out_port
;
3802 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3805 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3806 delete_flows_cb
, &cbdata
);
3809 /* Implements OFPFC_DELETE_STRICT. */
3811 delete_flow_strict(struct ofproto
*p
, struct ofp_flow_mod
*ofm
)
3813 struct rule
*rule
= find_flow_strict(p
, ofm
);
3815 delete_flow(p
, rule
, ofm
->out_port
);
3819 /* Callback for delete_flows_loose(). */
3821 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3823 struct rule
*rule
= rule_from_cls_rule(rule_
);
3824 struct delete_flows_cbdata
*cbdata
= cbdata_
;
3826 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
3829 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3830 * been identified as a flow to delete from 'p''s flow table, by deleting the
3831 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3834 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3835 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3836 * specified 'out_port'. */
3838 delete_flow(struct ofproto
*p
, struct rule
*rule
, uint16_t out_port
)
3840 if (rule_is_hidden(rule
)) {
3844 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
3848 send_flow_removed(p
, rule
, time_msec(), OFPRR_DELETE
);
3849 rule_remove(p
, rule
);
3853 handle_flow_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
3854 struct ofp_flow_mod
*ofm
)
3856 struct ofp_match orig_match
;
3860 error
= reject_slave_controller(ofconn
, &ofm
->header
);
3864 error
= check_ofp_message_array(&ofm
->header
, OFPT_FLOW_MOD
, sizeof *ofm
,
3865 sizeof *ofm
->actions
, &n_actions
);
3870 /* We do not support the emergency flow cache. It will hopefully
3871 * get dropped from OpenFlow in the near future. */
3872 if (ofm
->flags
& htons(OFPFF_EMERG
)) {
3873 /* There isn't a good fit for an error code, so just state that the
3874 * flow table is full. */
3875 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
3878 /* Normalize ofp->match. If normalization actually changes anything, then
3879 * log the differences. */
3880 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
3881 orig_match
= ofm
->match
;
3882 normalize_match(&ofm
->match
);
3883 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
3884 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
3885 if (!VLOG_DROP_INFO(&normal_rl
)) {
3886 char *old
= ofp_match_to_literal_string(&orig_match
);
3887 char *new = ofp_match_to_literal_string(&ofm
->match
);
3888 VLOG_INFO("%s: normalization changed ofp_match, details:",
3889 rconn_get_name(ofconn
->rconn
));
3890 VLOG_INFO(" pre: %s", old
);
3891 VLOG_INFO("post: %s", new);
3897 if (!ofm
->match
.wildcards
) {
3898 ofm
->priority
= htons(UINT16_MAX
);
3901 error
= validate_actions((const union ofp_action
*) ofm
->actions
,
3902 n_actions
, p
->max_ports
);
3907 switch (ntohs(ofm
->command
)) {
3909 return add_flow(p
, ofconn
, ofm
, n_actions
);
3912 return modify_flows_loose(p
, ofconn
, ofm
, n_actions
);
3914 case OFPFC_MODIFY_STRICT
:
3915 return modify_flow_strict(p
, ofconn
, ofm
, n_actions
);
3918 delete_flows_loose(p
, ofm
);
3921 case OFPFC_DELETE_STRICT
:
3922 delete_flow_strict(p
, ofm
);
3926 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
3931 handle_tun_id_from_cookie(struct ofproto
*p
, struct nxt_tun_id_cookie
*msg
)
3935 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
3940 p
->tun_id_from_cookie
= !!msg
->set
;
3945 handle_role_request(struct ofproto
*ofproto
,
3946 struct ofconn
*ofconn
, struct nicira_header
*msg
)
3948 struct nx_role_request
*nrr
;
3949 struct nx_role_request
*reply
;
3953 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
3954 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
3955 ntohs(msg
->header
.length
), sizeof *nrr
);
3956 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3958 nrr
= (struct nx_role_request
*) msg
;
3960 if (ofconn
->type
!= OFCONN_PRIMARY
) {
3961 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
3963 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3966 role
= ntohl(nrr
->role
);
3967 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3968 && role
!= NX_ROLE_SLAVE
) {
3969 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
3971 /* There's no good error code for this. */
3972 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
3975 if (role
== NX_ROLE_MASTER
) {
3976 struct ofconn
*other
;
3978 HMAP_FOR_EACH (other
, hmap_node
, &ofproto
->controllers
) {
3979 if (other
->role
== NX_ROLE_MASTER
) {
3980 other
->role
= NX_ROLE_SLAVE
;
3984 ofconn
->role
= role
;
3986 reply
= make_openflow_xid(sizeof *reply
, OFPT_VENDOR
, msg
->header
.xid
,
3988 reply
->nxh
.vendor
= htonl(NX_VENDOR_ID
);
3989 reply
->nxh
.subtype
= htonl(NXT_ROLE_REPLY
);
3990 reply
->role
= htonl(role
);
3991 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3997 handle_vendor(struct ofproto
*p
, struct ofconn
*ofconn
, void *msg
)
3999 struct ofp_vendor_header
*ovh
= msg
;
4000 struct nicira_header
*nh
;
4002 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
4003 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
4004 "(expected at least %zu)",
4005 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
4006 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4008 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
4009 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
4011 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
4012 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
4013 "(expected at least %zu)",
4014 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
4015 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4019 switch (ntohl(nh
->subtype
)) {
4020 case NXT_STATUS_REQUEST
:
4021 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
4024 case NXT_TUN_ID_FROM_COOKIE
:
4025 return handle_tun_id_from_cookie(p
, msg
);
4027 case NXT_ROLE_REQUEST
:
4028 return handle_role_request(p
, ofconn
, msg
);
4031 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
4035 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4037 struct ofp_header
*ob
;
4040 /* Currently, everything executes synchronously, so we can just
4041 * immediately send the barrier reply. */
4042 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4043 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4048 handle_openflow(struct ofconn
*ofconn
, struct ofproto
*p
,
4049 struct ofpbuf
*ofp_msg
)
4051 struct ofp_header
*oh
= ofp_msg
->data
;
4054 COVERAGE_INC(ofproto_recv_openflow
);
4056 case OFPT_ECHO_REQUEST
:
4057 error
= handle_echo_request(ofconn
, oh
);
4060 case OFPT_ECHO_REPLY
:
4064 case OFPT_FEATURES_REQUEST
:
4065 error
= handle_features_request(p
, ofconn
, oh
);
4068 case OFPT_GET_CONFIG_REQUEST
:
4069 error
= handle_get_config_request(p
, ofconn
, oh
);
4072 case OFPT_SET_CONFIG
:
4073 error
= handle_set_config(p
, ofconn
, ofp_msg
->data
);
4076 case OFPT_PACKET_OUT
:
4077 error
= handle_packet_out(p
, ofconn
, ofp_msg
->data
);
4081 error
= handle_port_mod(p
, ofconn
, oh
);
4085 error
= handle_flow_mod(p
, ofconn
, ofp_msg
->data
);
4088 case OFPT_STATS_REQUEST
:
4089 error
= handle_stats_request(p
, ofconn
, oh
);
4093 error
= handle_vendor(p
, ofconn
, ofp_msg
->data
);
4096 case OFPT_BARRIER_REQUEST
:
4097 error
= handle_barrier_request(ofconn
, oh
);
4101 if (VLOG_IS_WARN_ENABLED()) {
4102 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4103 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4106 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4111 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4116 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4118 struct odp_msg
*msg
= packet
->data
;
4120 struct ofpbuf payload
;
4123 payload
.data
= msg
+ 1;
4124 payload
.size
= msg
->length
- sizeof *msg
;
4125 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4127 /* Check with in-band control to see if this packet should be sent
4128 * to the local port regardless of the flow table. */
4129 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4130 union odp_action action
;
4132 memset(&action
, 0, sizeof(action
));
4133 action
.output
.type
= ODPAT_OUTPUT
;
4134 action
.output
.port
= ODPP_LOCAL
;
4135 dpif_execute(p
->dpif
, flow
.in_port
, &action
, 1, &payload
);
4138 rule
= lookup_valid_rule(p
, &flow
);
4140 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4141 struct ofport
*port
= get_port(p
, msg
->port
);
4143 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4144 COVERAGE_INC(ofproto_no_packet_in
);
4145 /* XXX install 'drop' flow entry */
4146 ofpbuf_delete(packet
);
4150 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
4153 COVERAGE_INC(ofproto_packet_in
);
4154 send_packet_in(p
, packet
);
4158 if (rule
->cr
.wc
.wildcards
) {
4159 rule
= rule_create_subrule(p
, rule
, &flow
);
4160 rule_make_actions(p
, rule
, packet
);
4162 if (!rule
->may_install
) {
4163 /* The rule is not installable, that is, we need to process every
4164 * packet, so process the current packet and set its actions into
4166 rule_make_actions(p
, rule
, packet
);
4168 /* XXX revalidate rule if it needs it */
4172 if (rule
->super
&& rule
->super
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4174 * Extra-special case for fail-open mode.
4176 * We are in fail-open mode and the packet matched the fail-open rule,
4177 * but we are connected to a controller too. We should send the packet
4178 * up to the controller in the hope that it will try to set up a flow
4179 * and thereby allow us to exit fail-open.
4181 * See the top-level comment in fail-open.c for more information.
4183 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4184 DPIF_RECV_MSG_PADDING
));
4187 ofpbuf_pull(packet
, sizeof *msg
);
4188 rule_execute(p
, rule
, packet
, &flow
);
4189 rule_reinstall(p
, rule
);
4193 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4195 struct odp_msg
*msg
= packet
->data
;
4197 switch (msg
->type
) {
4198 case _ODPL_ACTION_NR
:
4199 COVERAGE_INC(ofproto_ctlr_action
);
4200 send_packet_in(p
, packet
);
4203 case _ODPL_SFLOW_NR
:
4205 ofproto_sflow_received(p
->sflow
, msg
);
4207 ofpbuf_delete(packet
);
4211 handle_odp_miss_msg(p
, packet
);
4215 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4221 /* Flow expiration. */
4223 struct expire_cbdata
{
4224 struct ofproto
*ofproto
;
4227 static void ofproto_update_used(struct ofproto
*);
4228 static void rule_expire(struct cls_rule
*, void *cbdata
);
4230 /* This function is called periodically by ofproto_run(). Its job is to
4231 * collect updates for the flows that have been installed into the datapath,
4232 * most importantly when they last were used, and then use that information to
4233 * expire flows that have not been used recently. */
4235 ofproto_expire(struct ofproto
*ofproto
)
4237 struct expire_cbdata cbdata
;
4239 /* Update 'used' for each flow in the datapath. */
4240 ofproto_update_used(ofproto
);
4242 /* Expire idle flows. */
4243 cbdata
.ofproto
= ofproto
;
4244 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, rule_expire
, &cbdata
);
4246 /* Let the hook know that we're at a stable point: all outstanding data
4247 * in existing flows has been accounted to the account_cb. Thus, the
4248 * hook can now reasonably do operations that depend on having accurate
4249 * flow volume accounting (currently, that's just bond rebalancing). */
4250 if (ofproto
->ofhooks
->account_checkpoint_cb
) {
4251 ofproto
->ofhooks
->account_checkpoint_cb(ofproto
->aux
);
4255 /* Update 'used' member of each flow currently installed into the datapath. */
4257 ofproto_update_used(struct ofproto
*p
)
4259 struct odp_flow
*flows
;
4264 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4269 for (i
= 0; i
< n_flows
; i
++) {
4270 struct odp_flow
*f
= &flows
[i
];
4273 rule
= rule_from_cls_rule(
4274 classifier_find_rule_exactly(&p
->cls
, &f
->key
, 0, UINT16_MAX
));
4276 if (rule
&& rule
->installed
) {
4277 update_time(p
, rule
, &f
->stats
);
4278 rule_account(p
, rule
, f
->stats
.n_bytes
);
4280 /* There's a flow in the datapath that we know nothing about.
4282 COVERAGE_INC(ofproto_unexpected_rule
);
4283 dpif_flow_del(p
->dpif
, f
);
4291 rule_active_timeout(struct ofproto
*ofproto
, struct rule
*rule
)
4293 if (ofproto
->netflow
&& !is_controller_rule(rule
) &&
4294 netflow_active_timeout_expired(ofproto
->netflow
, &rule
->nf_flow
)) {
4295 struct ofexpired expired
;
4296 struct odp_flow odp_flow
;
4298 /* Get updated flow stats.
4300 * XXX We could avoid this call entirely if (1) ofproto_update_used()
4301 * updated TCP flags and (2) the dpif_flow_list_all() in
4302 * ofproto_update_used() zeroed TCP flags. */
4303 memset(&odp_flow
, 0, sizeof odp_flow
);
4304 if (rule
->installed
) {
4305 odp_flow
.key
= rule
->cr
.flow
;
4306 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4307 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4309 if (odp_flow
.stats
.n_packets
) {
4310 update_time(ofproto
, rule
, &odp_flow
.stats
);
4311 netflow_flow_update_flags(&rule
->nf_flow
,
4312 odp_flow
.stats
.tcp_flags
);
4316 expired
.flow
= rule
->cr
.flow
;
4317 expired
.packet_count
= rule
->packet_count
+
4318 odp_flow
.stats
.n_packets
;
4319 expired
.byte_count
= rule
->byte_count
+ odp_flow
.stats
.n_bytes
;
4320 expired
.used
= rule
->used
;
4322 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
4324 /* Schedule us to send the accumulated records once we have
4325 * collected all of them. */
4326 poll_immediate_wake();
4330 /* If 'cls_rule' is an OpenFlow rule, that has expired according to OpenFlow
4331 * rules, then delete it entirely.
4333 * If 'cls_rule' is a subrule, that has not been used recently, remove it from
4334 * the datapath and fold its statistics back into its super-rule.
4336 * (This is a callback function for classifier_for_each().) */
4338 rule_expire(struct cls_rule
*cls_rule
, void *cbdata_
)
4340 struct expire_cbdata
*cbdata
= cbdata_
;
4341 struct ofproto
*ofproto
= cbdata
->ofproto
;
4342 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
4343 long long int hard_expire
, idle_expire
, expire
, now
;
4345 /* Calculate OpenFlow expiration times for 'rule'. */
4346 hard_expire
= (rule
->hard_timeout
4347 ? rule
->created
+ rule
->hard_timeout
* 1000
4349 idle_expire
= (rule
->idle_timeout
4350 && (rule
->super
|| list_is_empty(&rule
->list
))
4351 ? rule
->used
+ rule
->idle_timeout
* 1000
4353 expire
= MIN(hard_expire
, idle_expire
);
4357 /* 'rule' has not expired according to OpenFlow rules. */
4358 if (rule
->installed
&& now
>= rule
->used
+ 5000) {
4359 /* This rule is idle, so uninstall it from the datapath. */
4361 rule_remove(ofproto
, rule
);
4363 rule_uninstall(ofproto
, rule
);
4365 } else if (!rule
->cr
.wc
.wildcards
) {
4366 /* Send NetFlow active timeout if appropriate. */
4367 rule_active_timeout(cbdata
->ofproto
, rule
);
4370 /* 'rule' has expired according to OpenFlow rules. */
4371 COVERAGE_INC(ofproto_expired
);
4373 /* Update stats. (This is a no-op if the rule expired due to an idle
4374 * timeout, because that only happens when the rule has no subrules
4376 if (rule
->cr
.wc
.wildcards
) {
4377 struct rule
*subrule
, *next
;
4378 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
4379 rule_remove(cbdata
->ofproto
, subrule
);
4382 rule_uninstall(cbdata
->ofproto
, rule
);
4385 /* Get rid of the rule. */
4386 if (!rule_is_hidden(rule
)) {
4387 send_flow_removed(cbdata
->ofproto
, rule
, now
,
4389 ? OFPRR_HARD_TIMEOUT
: OFPRR_IDLE_TIMEOUT
));
4391 rule_remove(cbdata
->ofproto
, rule
);
4396 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
4398 struct rule
*sub
= rule_from_cls_rule(sub_
);
4399 struct revalidate_cbdata
*cbdata
= cbdata_
;
4401 if (cbdata
->revalidate_all
4402 || (cbdata
->revalidate_subrules
&& sub
->super
)
4403 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
4404 revalidate_rule(cbdata
->ofproto
, sub
);
4409 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
4411 const flow_t
*flow
= &rule
->cr
.flow
;
4413 COVERAGE_INC(ofproto_revalidate_rule
);
4416 super
= rule_from_cls_rule(classifier_lookup_wild(&p
->cls
, flow
));
4418 rule_remove(p
, rule
);
4420 } else if (super
!= rule
->super
) {
4421 COVERAGE_INC(ofproto_revalidate_moved
);
4422 list_remove(&rule
->list
);
4423 list_push_back(&super
->list
, &rule
->list
);
4424 rule
->super
= super
;
4425 rule
->hard_timeout
= super
->hard_timeout
;
4426 rule
->idle_timeout
= super
->idle_timeout
;
4427 rule
->created
= super
->created
;
4432 rule_update_actions(p
, rule
);
4436 static struct ofpbuf
*
4437 compose_flow_removed(struct ofproto
*p
, const struct rule
*rule
,
4438 long long int now
, uint8_t reason
)
4440 struct ofp_flow_removed
*ofr
;
4442 long long int tdiff
= now
- rule
->created
;
4443 uint32_t sec
= tdiff
/ 1000;
4444 uint32_t msec
= tdiff
- (sec
* 1000);
4446 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4447 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, p
->tun_id_from_cookie
,
4449 ofr
->cookie
= rule
->flow_cookie
;
4450 ofr
->priority
= htons(rule
->cr
.priority
);
4451 ofr
->reason
= reason
;
4452 ofr
->duration_sec
= htonl(sec
);
4453 ofr
->duration_nsec
= htonl(msec
* 1000000);
4454 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4455 ofr
->packet_count
= htonll(rule
->packet_count
);
4456 ofr
->byte_count
= htonll(rule
->byte_count
);
4462 send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
4463 long long int now
, uint8_t reason
)
4465 struct ofconn
*ofconn
;
4466 struct ofconn
*prev
;
4467 struct ofpbuf
*buf
= NULL
;
4469 if (!rule
->send_flow_removed
) {
4473 /* We limit the maximum number of queued flow expirations it by accounting
4474 * them under the counter for replies. That works because preventing
4475 * OpenFlow requests from being processed also prevents new flows from
4476 * being added (and expiring). (It also prevents processing OpenFlow
4477 * requests that would not add new flows, so it is imperfect.) */
4480 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4481 if (rconn_is_connected(ofconn
->rconn
)
4482 && ofconn_receives_async_msgs(ofconn
)) {
4484 queue_tx(ofpbuf_clone(buf
), prev
, prev
->reply_counter
);
4486 buf
= compose_flow_removed(p
, rule
, now
, reason
);
4492 queue_tx(buf
, prev
, prev
->reply_counter
);
4496 /* pinsched callback for sending 'packet' on 'ofconn'. */
4498 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4500 struct ofconn
*ofconn
= ofconn_
;
4502 rconn_send_with_limit(ofconn
->rconn
, packet
,
4503 ofconn
->packet_in_counter
, 100);
4506 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4507 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4508 * packet scheduler for sending.
4510 * 'max_len' specifies the maximum number of bytes of the packet to send on
4511 * 'ofconn' (INT_MAX specifies no limit).
4513 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4514 * ownership is transferred to this function. */
4516 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4519 struct ofproto
*ofproto
= ofconn
->ofproto
;
4520 struct ofp_packet_in
*opi
= packet
->data
;
4521 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4522 int send_len
, trim_size
;
4526 if (opi
->reason
== OFPR_ACTION
) {
4527 buffer_id
= UINT32_MAX
;
4528 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4529 buffer_id
= pktbuf_get_null();
4530 } else if (!ofconn
->pktbuf
) {
4531 buffer_id
= UINT32_MAX
;
4533 struct ofpbuf payload
;
4534 payload
.data
= opi
->data
;
4535 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4536 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4539 /* Figure out how much of the packet to send. */
4540 send_len
= ntohs(opi
->total_len
);
4541 if (buffer_id
!= UINT32_MAX
) {
4542 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4544 send_len
= MIN(send_len
, max_len
);
4546 /* Adjust packet length and clone if necessary. */
4547 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4549 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4552 packet
->size
= trim_size
;
4555 /* Update packet headers. */
4556 opi
->buffer_id
= htonl(buffer_id
);
4557 update_openflow_length(packet
);
4559 /* Hand over to packet scheduler. It might immediately call into
4560 * do_send_packet_in() or it might buffer it for a while (until a later
4561 * call to pinsched_run()). */
4562 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4563 packet
, do_send_packet_in
, ofconn
);
4566 /* Replace struct odp_msg header in 'packet' by equivalent struct
4567 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4568 * returned by dpif_recv()).
4570 * The conversion is not complete: the caller still needs to trim any unneeded
4571 * payload off the end of the buffer, set the length in the OpenFlow header,
4572 * and set buffer_id. Those require us to know the controller settings and so
4573 * must be done on a per-controller basis.
4575 * Returns the maximum number of bytes of the packet that should be sent to
4576 * the controller (INT_MAX if no limit). */
4578 do_convert_to_packet_in(struct ofpbuf
*packet
)
4580 struct odp_msg
*msg
= packet
->data
;
4581 struct ofp_packet_in
*opi
;
4587 /* Extract relevant header fields */
4588 if (msg
->type
== _ODPL_ACTION_NR
) {
4589 reason
= OFPR_ACTION
;
4592 reason
= OFPR_NO_MATCH
;
4595 total_len
= msg
->length
- sizeof *msg
;
4596 in_port
= odp_port_to_ofp_port(msg
->port
);
4598 /* Repurpose packet buffer by overwriting header. */
4599 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4600 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4601 opi
->header
.version
= OFP_VERSION
;
4602 opi
->header
.type
= OFPT_PACKET_IN
;
4603 opi
->total_len
= htons(total_len
);
4604 opi
->in_port
= htons(in_port
);
4605 opi
->reason
= reason
;
4610 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4611 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4612 * as necessary according to their individual configurations.
4614 * 'packet' must have sufficient headroom to convert it into a struct
4615 * ofp_packet_in (e.g. as returned by dpif_recv()).
4617 * Takes ownership of 'packet'. */
4619 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4621 struct ofconn
*ofconn
, *prev
;
4624 max_len
= do_convert_to_packet_in(packet
);
4627 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4628 if (ofconn_receives_async_msgs(ofconn
)) {
4630 schedule_packet_in(prev
, packet
, max_len
, true);
4636 schedule_packet_in(prev
, packet
, max_len
, false);
4638 ofpbuf_delete(packet
);
4643 pick_datapath_id(const struct ofproto
*ofproto
)
4645 const struct ofport
*port
;
4647 port
= get_port(ofproto
, ODPP_LOCAL
);
4649 uint8_t ea
[ETH_ADDR_LEN
];
4652 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4654 return eth_addr_to_uint64(ea
);
4656 VLOG_WARN("could not get MAC address for %s (%s)",
4657 netdev_get_name(port
->netdev
), strerror(error
));
4659 return ofproto
->fallback_dpid
;
4663 pick_fallback_dpid(void)
4665 uint8_t ea
[ETH_ADDR_LEN
];
4666 eth_addr_nicira_random(ea
);
4667 return eth_addr_to_uint64(ea
);
4671 default_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
4672 struct odp_actions
*actions
, tag_type
*tags
,
4673 uint16_t *nf_output_iface
, void *ofproto_
)
4675 struct ofproto
*ofproto
= ofproto_
;
4678 /* Drop frames for reserved multicast addresses. */
4679 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4683 /* Learn source MAC (but don't try to learn from revalidation). */
4684 if (packet
!= NULL
) {
4685 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4687 GRAT_ARP_LOCK_NONE
);
4689 /* The log messages here could actually be useful in debugging,
4690 * so keep the rate limit relatively high. */
4691 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4692 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4693 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4694 ofproto_revalidate(ofproto
, rev_tag
);
4698 /* Determine output port. */
4699 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
4702 add_output_group_action(actions
, DP_GROUP_FLOOD
, nf_output_iface
);
4703 } else if (out_port
!= flow
->in_port
) {
4704 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4705 *nf_output_iface
= out_port
;
4713 static const struct ofhooks default_ofhooks
= {
4715 default_normal_ofhook_cb
,