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 update_used(struct ofproto
*);
328 static void update_stats(struct ofproto
*, struct rule
*,
329 const struct odp_flow_stats
*);
330 static void expire_rule(struct cls_rule
*, void *ofproto
);
331 static void active_timeout(struct ofproto
*ofproto
, struct rule
*rule
);
332 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
333 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
335 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
337 static void handle_openflow(struct ofconn
*, struct ofproto
*,
340 static void refresh_port_groups(struct ofproto
*);
342 static struct ofport
*get_port(const struct ofproto
*, uint16_t odp_port
);
343 static void update_port(struct ofproto
*, const char *devname
);
344 static int init_ports(struct ofproto
*);
345 static void reinit_ports(struct ofproto
*);
348 ofproto_create(const char *datapath
, const char *datapath_type
,
349 const struct ofhooks
*ofhooks
, void *aux
,
350 struct ofproto
**ofprotop
)
352 struct odp_stats stats
;
359 /* Connect to datapath and start listening for messages. */
360 error
= dpif_open(datapath
, datapath_type
, &dpif
);
362 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
365 error
= dpif_get_dp_stats(dpif
, &stats
);
367 VLOG_ERR("failed to obtain stats for datapath %s: %s",
368 datapath
, strerror(error
));
372 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
374 VLOG_ERR("failed to listen on datapath %s: %s",
375 datapath
, strerror(error
));
379 dpif_flow_flush(dpif
);
380 dpif_recv_purge(dpif
);
382 /* Initialize settings. */
383 p
= xzalloc(sizeof *p
);
384 p
->fallback_dpid
= pick_fallback_dpid();
385 p
->datapath_id
= p
->fallback_dpid
;
386 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
387 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
388 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
389 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
390 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
392 /* Initialize datapath. */
394 p
->netdev_monitor
= netdev_monitor_create();
395 hmap_init(&p
->ports
);
396 shash_init(&p
->port_by_name
);
397 p
->max_ports
= stats
.max_ports
;
399 /* Initialize submodules. */
400 p
->switch_status
= switch_status_create(p
);
406 /* Initialize flow table. */
407 classifier_init(&p
->cls
);
408 p
->need_revalidate
= false;
409 p
->next_expiration
= time_msec() + 1000;
410 tag_set_init(&p
->revalidate_set
);
412 /* Initialize OpenFlow connections. */
413 list_init(&p
->all_conns
);
414 hmap_init(&p
->controllers
);
415 hmap_init(&p
->services
);
419 /* Initialize hooks. */
421 p
->ofhooks
= ofhooks
;
425 p
->ofhooks
= &default_ofhooks
;
427 p
->ml
= mac_learning_create();
430 /* Pick final datapath ID. */
431 p
->datapath_id
= pick_datapath_id(p
);
432 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
439 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
441 uint64_t old_dpid
= p
->datapath_id
;
442 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
443 if (p
->datapath_id
!= old_dpid
) {
444 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
446 /* Force all active connections to reconnect, since there is no way to
447 * notify a controller that the datapath ID has changed. */
448 ofproto_reconnect_controllers(p
);
453 is_discovery_controller(const struct ofproto_controller
*c
)
455 return !strcmp(c
->target
, "discover");
459 is_in_band_controller(const struct ofproto_controller
*c
)
461 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
464 /* Creates a new controller in 'ofproto'. Some of the settings are initially
465 * drawn from 'c', but update_controller() needs to be called later to finish
466 * the new ofconn's configuration. */
468 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
470 struct discovery
*discovery
;
471 struct ofconn
*ofconn
;
473 if (is_discovery_controller(c
)) {
474 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
475 ofproto
->dpif
, ofproto
->switch_status
,
484 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_PRIMARY
);
485 ofconn
->pktbuf
= pktbuf_create();
486 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
488 ofconn
->discovery
= discovery
;
490 char *name
= ofconn_make_name(ofproto
, c
->target
);
491 rconn_connect(ofconn
->rconn
, c
->target
, name
);
494 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
495 hash_string(c
->target
, 0));
498 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
499 * target or turn discovery on or off (these are done by creating new ofconns
500 * and deleting old ones), but it can update the rest of an ofconn's
503 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
507 ofconn
->band
= (is_in_band_controller(c
)
508 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
510 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
512 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
513 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
515 if (ofconn
->discovery
) {
516 discovery_set_update_resolv_conf(ofconn
->discovery
,
517 c
->update_resolv_conf
);
518 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
521 ofconn_set_rate_limit(ofconn
, c
->rate_limit
, c
->burst_limit
);
525 ofconn_get_target(const struct ofconn
*ofconn
)
527 return ofconn
->discovery
? "discover" : rconn_get_target(ofconn
->rconn
);
530 static struct ofconn
*
531 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
533 struct ofconn
*ofconn
;
535 HMAP_FOR_EACH_WITH_HASH (ofconn
, hmap_node
,
536 hash_string(target
, 0), &ofproto
->controllers
) {
537 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
545 update_in_band_remotes(struct ofproto
*ofproto
)
547 const struct ofconn
*ofconn
;
548 struct sockaddr_in
*addrs
;
549 size_t max_addrs
, n_addrs
;
553 /* Allocate enough memory for as many remotes as we could possibly have. */
554 max_addrs
= ofproto
->n_extra_remotes
+ hmap_count(&ofproto
->controllers
);
555 addrs
= xmalloc(max_addrs
* sizeof *addrs
);
558 /* Add all the remotes. */
560 HMAP_FOR_EACH (ofconn
, hmap_node
, &ofproto
->controllers
) {
561 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
563 if (ofconn
->band
== OFPROTO_OUT_OF_BAND
) {
567 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
568 if (sin
->sin_addr
.s_addr
) {
569 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
572 if (ofconn
->discovery
) {
576 for (i
= 0; i
< ofproto
->n_extra_remotes
; i
++) {
577 addrs
[n_addrs
++] = ofproto
->extra_in_band_remotes
[i
];
580 /* Create or update or destroy in-band.
582 * Ordinarily we only enable in-band if there's at least one remote
583 * address, but discovery needs the in-band rules for DHCP to be installed
584 * even before we know any remote addresses. */
585 if (n_addrs
|| discovery
) {
586 if (!ofproto
->in_band
) {
587 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
590 if (ofproto
->in_band
) {
591 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
593 ofproto
->next_in_band_update
= time_msec() + 1000;
595 in_band_destroy(ofproto
->in_band
);
596 ofproto
->in_band
= NULL
;
604 update_fail_open(struct ofproto
*p
)
606 struct ofconn
*ofconn
;
608 if (!hmap_is_empty(&p
->controllers
)
609 && p
->fail_mode
== OFPROTO_FAIL_STANDALONE
) {
610 struct rconn
**rconns
;
614 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
618 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
619 HMAP_FOR_EACH (ofconn
, hmap_node
, &p
->controllers
) {
620 rconns
[n
++] = ofconn
->rconn
;
623 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
624 /* p->fail_open takes ownership of 'rconns'. */
626 fail_open_destroy(p
->fail_open
);
632 ofproto_set_controllers(struct ofproto
*p
,
633 const struct ofproto_controller
*controllers
,
634 size_t n_controllers
)
636 struct shash new_controllers
;
637 struct ofconn
*ofconn
, *next_ofconn
;
638 struct ofservice
*ofservice
, *next_ofservice
;
642 /* Create newly configured controllers and services.
643 * Create a name to ofproto_controller mapping in 'new_controllers'. */
644 shash_init(&new_controllers
);
645 for (i
= 0; i
< n_controllers
; i
++) {
646 const struct ofproto_controller
*c
= &controllers
[i
];
648 if (!vconn_verify_name(c
->target
) || !strcmp(c
->target
, "discover")) {
649 if (!find_controller_by_target(p
, c
->target
)) {
650 add_controller(p
, c
);
652 } else if (!pvconn_verify_name(c
->target
)) {
653 if (!ofservice_lookup(p
, c
->target
) && ofservice_create(p
, c
)) {
657 VLOG_WARN_RL(&rl
, "%s: unsupported controller \"%s\"",
658 dpif_name(p
->dpif
), c
->target
);
662 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
665 /* Delete controllers that are no longer configured.
666 * Update configuration of all now-existing controllers. */
668 HMAP_FOR_EACH_SAFE (ofconn
, next_ofconn
, hmap_node
, &p
->controllers
) {
669 struct ofproto_controller
*c
;
671 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
673 ofconn_destroy(ofconn
);
675 update_controller(ofconn
, c
);
682 /* Delete services that are no longer configured.
683 * Update configuration of all now-existing services. */
684 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
685 struct ofproto_controller
*c
;
687 c
= shash_find_data(&new_controllers
,
688 pvconn_get_name(ofservice
->pvconn
));
690 ofservice_destroy(p
, ofservice
);
692 ofservice_reconfigure(ofservice
, c
);
696 shash_destroy(&new_controllers
);
698 update_in_band_remotes(p
);
701 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
702 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
703 struct ofconn
, hmap_node
);
704 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
705 rconn_status_cb
, ofconn
->rconn
);
710 ofproto_set_fail_mode(struct ofproto
*p
, enum ofproto_fail_mode fail_mode
)
712 p
->fail_mode
= fail_mode
;
716 /* Drops the connections between 'ofproto' and all of its controllers, forcing
717 * them to reconnect. */
719 ofproto_reconnect_controllers(struct ofproto
*ofproto
)
721 struct ofconn
*ofconn
;
723 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
724 rconn_reconnect(ofconn
->rconn
);
729 any_extras_changed(const struct ofproto
*ofproto
,
730 const struct sockaddr_in
*extras
, size_t n
)
734 if (n
!= ofproto
->n_extra_remotes
) {
738 for (i
= 0; i
< n
; i
++) {
739 const struct sockaddr_in
*old
= &ofproto
->extra_in_band_remotes
[i
];
740 const struct sockaddr_in
*new = &extras
[i
];
742 if (old
->sin_addr
.s_addr
!= new->sin_addr
.s_addr
||
743 old
->sin_port
!= new->sin_port
) {
751 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
752 * in-band control should guarantee access, in the same way that in-band
753 * control guarantees access to OpenFlow controllers. */
755 ofproto_set_extra_in_band_remotes(struct ofproto
*ofproto
,
756 const struct sockaddr_in
*extras
, size_t n
)
758 if (!any_extras_changed(ofproto
, extras
, n
)) {
762 free(ofproto
->extra_in_band_remotes
);
763 ofproto
->n_extra_remotes
= n
;
764 ofproto
->extra_in_band_remotes
= xmemdup(extras
, n
* sizeof *extras
);
766 update_in_band_remotes(ofproto
);
770 ofproto_set_desc(struct ofproto
*p
,
771 const char *mfr_desc
, const char *hw_desc
,
772 const char *sw_desc
, const char *serial_desc
,
775 struct ofp_desc_stats
*ods
;
778 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
779 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
780 sizeof ods
->mfr_desc
);
783 p
->mfr_desc
= xstrdup(mfr_desc
);
786 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
787 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
788 sizeof ods
->hw_desc
);
791 p
->hw_desc
= xstrdup(hw_desc
);
794 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
795 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
796 sizeof ods
->sw_desc
);
799 p
->sw_desc
= xstrdup(sw_desc
);
802 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
803 VLOG_WARN("truncating serial_desc, must be less than %zu "
805 sizeof ods
->serial_num
);
807 free(p
->serial_desc
);
808 p
->serial_desc
= xstrdup(serial_desc
);
811 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
812 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
813 sizeof ods
->dp_desc
);
816 p
->dp_desc
= xstrdup(dp_desc
);
821 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
822 const struct svec
*svec
)
824 struct pvconn
**pvconns
= *pvconnsp
;
825 size_t n_pvconns
= *n_pvconnsp
;
829 for (i
= 0; i
< n_pvconns
; i
++) {
830 pvconn_close(pvconns
[i
]);
834 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
836 for (i
= 0; i
< svec
->n
; i
++) {
837 const char *name
= svec
->names
[i
];
838 struct pvconn
*pvconn
;
841 error
= pvconn_open(name
, &pvconn
);
843 pvconns
[n_pvconns
++] = pvconn
;
845 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
853 *n_pvconnsp
= n_pvconns
;
859 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
861 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
865 ofproto_set_netflow(struct ofproto
*ofproto
,
866 const struct netflow_options
*nf_options
)
868 if (nf_options
&& nf_options
->collectors
.n
) {
869 if (!ofproto
->netflow
) {
870 ofproto
->netflow
= netflow_create();
872 return netflow_set_options(ofproto
->netflow
, nf_options
);
874 netflow_destroy(ofproto
->netflow
);
875 ofproto
->netflow
= NULL
;
881 ofproto_set_sflow(struct ofproto
*ofproto
,
882 const struct ofproto_sflow_options
*oso
)
884 struct ofproto_sflow
*os
= ofproto
->sflow
;
887 struct ofport
*ofport
;
889 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
890 refresh_port_groups(ofproto
);
891 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->ports
) {
892 ofproto_sflow_add_port(os
, ofport
->odp_port
,
893 netdev_get_name(ofport
->netdev
));
896 ofproto_sflow_set_options(os
, oso
);
898 ofproto_sflow_destroy(os
);
899 ofproto
->sflow
= NULL
;
904 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
906 return ofproto
->datapath_id
;
910 ofproto_has_primary_controller(const struct ofproto
*ofproto
)
912 return !hmap_is_empty(&ofproto
->controllers
);
915 enum ofproto_fail_mode
916 ofproto_get_fail_mode(const struct ofproto
*p
)
922 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
926 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
927 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
932 ofproto_destroy(struct ofproto
*p
)
934 struct ofservice
*ofservice
, *next_ofservice
;
935 struct ofconn
*ofconn
, *next_ofconn
;
936 struct ofport
*ofport
, *next_ofport
;
943 /* Destroy fail-open and in-band early, since they touch the classifier. */
944 fail_open_destroy(p
->fail_open
);
947 in_band_destroy(p
->in_band
);
949 free(p
->extra_in_band_remotes
);
951 ofproto_flush_flows(p
);
952 classifier_destroy(&p
->cls
);
954 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
955 ofconn_destroy(ofconn
);
957 hmap_destroy(&p
->controllers
);
960 netdev_monitor_destroy(p
->netdev_monitor
);
961 HMAP_FOR_EACH_SAFE (ofport
, next_ofport
, hmap_node
, &p
->ports
) {
962 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
965 shash_destroy(&p
->port_by_name
);
967 switch_status_destroy(p
->switch_status
);
968 netflow_destroy(p
->netflow
);
969 ofproto_sflow_destroy(p
->sflow
);
971 HMAP_FOR_EACH_SAFE (ofservice
, next_ofservice
, node
, &p
->services
) {
972 ofservice_destroy(p
, ofservice
);
974 hmap_destroy(&p
->services
);
976 for (i
= 0; i
< p
->n_snoops
; i
++) {
977 pvconn_close(p
->snoops
[i
]);
981 mac_learning_destroy(p
->ml
);
986 free(p
->serial_desc
);
989 hmap_destroy(&p
->ports
);
995 ofproto_run(struct ofproto
*p
)
997 int error
= ofproto_run1(p
);
999 error
= ofproto_run2(p
, false);
1005 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
1007 if (error
== ENOBUFS
) {
1008 reinit_ports(ofproto
);
1009 } else if (!error
) {
1010 update_port(ofproto
, devname
);
1015 /* Returns a "preference level" for snooping 'ofconn'. A higher return value
1016 * means that 'ofconn' is more interesting for monitoring than a lower return
1019 snoop_preference(const struct ofconn
*ofconn
)
1021 switch (ofconn
->role
) {
1022 case NX_ROLE_MASTER
:
1029 /* Shouldn't happen. */
1034 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
1035 * Connects this vconn to a controller. */
1037 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
1039 struct ofconn
*ofconn
, *best
;
1041 /* Pick a controller for monitoring. */
1043 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
1044 if (ofconn
->type
== OFCONN_PRIMARY
1045 && (!best
|| snoop_preference(ofconn
) > snoop_preference(best
))) {
1051 rconn_add_monitor(best
->rconn
, vconn
);
1053 VLOG_INFO_RL(&rl
, "no controller connection to snoop");
1059 ofproto_run1(struct ofproto
*p
)
1061 struct ofconn
*ofconn
, *next_ofconn
;
1062 struct ofservice
*ofservice
;
1067 if (shash_is_empty(&p
->port_by_name
)) {
1071 for (i
= 0; i
< 50; i
++) {
1074 error
= dpif_recv(p
->dpif
, &buf
);
1076 if (error
== ENODEV
) {
1077 /* Someone destroyed the datapath behind our back. The caller
1078 * better destroy us and give up, because we're just going to
1079 * spin from here on out. */
1080 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
1081 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
1082 dpif_name(p
->dpif
));
1088 handle_odp_msg(p
, buf
);
1091 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
1092 process_port_change(p
, error
, devname
);
1094 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
1095 &devname
)) != EAGAIN
) {
1096 process_port_change(p
, error
, devname
);
1100 if (time_msec() >= p
->next_in_band_update
) {
1101 update_in_band_remotes(p
);
1103 in_band_run(p
->in_band
);
1106 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, node
, &p
->all_conns
) {
1107 ofconn_run(ofconn
, p
);
1110 /* Fail-open maintenance. Do this after processing the ofconns since
1111 * fail-open checks the status of the controller rconn. */
1113 fail_open_run(p
->fail_open
);
1116 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1117 struct vconn
*vconn
;
1120 retval
= pvconn_accept(ofservice
->pvconn
, OFP_VERSION
, &vconn
);
1122 struct rconn
*rconn
;
1125 rconn
= rconn_create(ofservice
->probe_interval
, 0);
1126 name
= ofconn_make_name(p
, vconn_get_name(vconn
));
1127 rconn_connect_unreliably(rconn
, vconn
, name
);
1130 ofconn
= ofconn_create(p
, rconn
, OFCONN_SERVICE
);
1131 ofconn_set_rate_limit(ofconn
, ofservice
->rate_limit
,
1132 ofservice
->burst_limit
);
1133 } else if (retval
!= EAGAIN
) {
1134 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1138 for (i
= 0; i
< p
->n_snoops
; i
++) {
1139 struct vconn
*vconn
;
1142 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1144 add_snooper(p
, vconn
);
1145 } else if (retval
!= EAGAIN
) {
1146 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1150 if (time_msec() >= p
->next_expiration
) {
1151 COVERAGE_INC(ofproto_expiration
);
1152 p
->next_expiration
= time_msec() + 1000;
1155 classifier_for_each(&p
->cls
, CLS_INC_ALL
, expire_rule
, p
);
1157 /* Let the hook know that we're at a stable point: all outstanding data
1158 * in existing flows has been accounted to the account_cb. Thus, the
1159 * hook can now reasonably do operations that depend on having accurate
1160 * flow volume accounting (currently, that's just bond rebalancing). */
1161 if (p
->ofhooks
->account_checkpoint_cb
) {
1162 p
->ofhooks
->account_checkpoint_cb(p
->aux
);
1167 netflow_run(p
->netflow
);
1170 ofproto_sflow_run(p
->sflow
);
1176 struct revalidate_cbdata
{
1177 struct ofproto
*ofproto
;
1178 bool revalidate_all
; /* Revalidate all exact-match rules? */
1179 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
1180 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
1184 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1186 if (p
->need_revalidate
|| revalidate_all
1187 || !tag_set_is_empty(&p
->revalidate_set
)) {
1188 struct revalidate_cbdata cbdata
;
1190 cbdata
.revalidate_all
= revalidate_all
;
1191 cbdata
.revalidate_subrules
= p
->need_revalidate
;
1192 cbdata
.revalidate_set
= p
->revalidate_set
;
1193 tag_set_init(&p
->revalidate_set
);
1194 COVERAGE_INC(ofproto_revalidate
);
1195 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
1196 p
->need_revalidate
= false;
1203 ofproto_wait(struct ofproto
*p
)
1205 struct ofservice
*ofservice
;
1206 struct ofconn
*ofconn
;
1209 dpif_recv_wait(p
->dpif
);
1210 dpif_port_poll_wait(p
->dpif
);
1211 netdev_monitor_poll_wait(p
->netdev_monitor
);
1212 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1213 ofconn_wait(ofconn
);
1216 poll_timer_wait_until(p
->next_in_band_update
);
1217 in_band_wait(p
->in_band
);
1220 fail_open_wait(p
->fail_open
);
1223 ofproto_sflow_wait(p
->sflow
);
1225 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1226 poll_immediate_wake();
1228 if (p
->need_revalidate
) {
1229 /* Shouldn't happen, but if it does just go around again. */
1230 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1231 poll_immediate_wake();
1232 } else if (p
->next_expiration
!= LLONG_MAX
) {
1233 poll_timer_wait_until(p
->next_expiration
);
1235 HMAP_FOR_EACH (ofservice
, node
, &p
->services
) {
1236 pvconn_wait(ofservice
->pvconn
);
1238 for (i
= 0; i
< p
->n_snoops
; i
++) {
1239 pvconn_wait(p
->snoops
[i
]);
1244 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1246 tag_set_add(&ofproto
->revalidate_set
, tag
);
1250 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1252 return &ofproto
->revalidate_set
;
1256 ofproto_is_alive(const struct ofproto
*p
)
1258 return !hmap_is_empty(&p
->controllers
);
1262 ofproto_send_packet(struct ofproto
*p
, const flow_t
*flow
,
1263 const union ofp_action
*actions
, size_t n_actions
,
1264 const struct ofpbuf
*packet
)
1266 struct odp_actions odp_actions
;
1269 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1275 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1277 dpif_execute(p
->dpif
, flow
->in_port
, odp_actions
.actions
,
1278 odp_actions
.n_actions
, packet
);
1283 ofproto_add_flow(struct ofproto
*p
,
1284 const flow_t
*flow
, uint32_t wildcards
, unsigned int priority
,
1285 const union ofp_action
*actions
, size_t n_actions
,
1289 rule
= rule_create(p
, NULL
, actions
, n_actions
,
1290 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */,
1292 cls_rule_from_flow(flow
, wildcards
, priority
, &rule
->cr
);
1293 rule_insert(p
, rule
, NULL
, 0);
1297 ofproto_delete_flow(struct ofproto
*ofproto
, const flow_t
*flow
,
1298 uint32_t wildcards
, unsigned int priority
)
1302 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1306 rule_remove(ofproto
, rule
);
1311 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1313 struct rule
*rule
= rule_from_cls_rule(rule_
);
1314 struct ofproto
*ofproto
= ofproto_
;
1316 /* Mark the flow as not installed, even though it might really be
1317 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1318 * There is no point in uninstalling it individually since we are about to
1319 * blow away all the flows with dpif_flow_flush(). */
1320 rule
->installed
= false;
1322 rule_remove(ofproto
, rule
);
1326 ofproto_flush_flows(struct ofproto
*ofproto
)
1328 COVERAGE_INC(ofproto_flush
);
1329 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1330 dpif_flow_flush(ofproto
->dpif
);
1331 if (ofproto
->in_band
) {
1332 in_band_flushed(ofproto
->in_band
);
1334 if (ofproto
->fail_open
) {
1335 fail_open_flushed(ofproto
->fail_open
);
1340 reinit_ports(struct ofproto
*p
)
1342 struct svec devnames
;
1343 struct ofport
*ofport
;
1344 struct odp_port
*odp_ports
;
1348 svec_init(&devnames
);
1349 HMAP_FOR_EACH (ofport
, hmap_node
, &p
->ports
) {
1350 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1352 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1353 for (i
= 0; i
< n_odp_ports
; i
++) {
1354 svec_add (&devnames
, odp_ports
[i
].devname
);
1358 svec_sort_unique(&devnames
);
1359 for (i
= 0; i
< devnames
.n
; i
++) {
1360 update_port(p
, devnames
.names
[i
]);
1362 svec_destroy(&devnames
);
1366 refresh_port_group(struct ofproto
*p
, unsigned int group
)
1370 struct ofport
*port
;
1372 assert(group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
);
1374 ports
= xmalloc(hmap_count(&p
->ports
) * sizeof *ports
);
1376 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
1377 if (group
== DP_GROUP_ALL
|| !(port
->opp
.config
& OFPPC_NO_FLOOD
)) {
1378 ports
[n_ports
++] = port
->odp_port
;
1381 dpif_port_group_set(p
->dpif
, group
, ports
, n_ports
);
1388 refresh_port_groups(struct ofproto
*p
)
1390 size_t n_flood
= refresh_port_group(p
, DP_GROUP_FLOOD
);
1391 size_t n_all
= refresh_port_group(p
, DP_GROUP_ALL
);
1393 ofproto_sflow_set_group_sizes(p
->sflow
, n_flood
, n_all
);
1397 static struct ofport
*
1398 make_ofport(const struct odp_port
*odp_port
)
1400 struct netdev_options netdev_options
;
1401 enum netdev_flags flags
;
1402 struct ofport
*ofport
;
1403 struct netdev
*netdev
;
1407 memset(&netdev_options
, 0, sizeof netdev_options
);
1408 netdev_options
.name
= odp_port
->devname
;
1409 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1411 error
= netdev_open(&netdev_options
, &netdev
);
1413 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1414 "cannot be opened (%s)",
1415 odp_port
->devname
, odp_port
->port
,
1416 odp_port
->devname
, strerror(error
));
1420 ofport
= xmalloc(sizeof *ofport
);
1421 ofport
->netdev
= netdev
;
1422 ofport
->odp_port
= odp_port
->port
;
1423 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1424 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1425 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1426 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1427 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1429 netdev_get_flags(netdev
, &flags
);
1430 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1432 netdev_get_carrier(netdev
, &carrier
);
1433 ofport
->opp
.state
= carrier
? 0 : OFPPS_LINK_DOWN
;
1435 netdev_get_features(netdev
,
1436 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1437 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1442 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1444 if (get_port(p
, odp_port
->port
)) {
1445 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1448 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1449 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1458 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1460 const struct ofp_phy_port
*a
= &a_
->opp
;
1461 const struct ofp_phy_port
*b
= &b_
->opp
;
1463 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1464 return (a
->port_no
== b
->port_no
1465 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1466 && !strcmp((char *) a
->name
, (char *) b
->name
)
1467 && a
->state
== b
->state
1468 && a
->config
== b
->config
1469 && a
->curr
== b
->curr
1470 && a
->advertised
== b
->advertised
1471 && a
->supported
== b
->supported
1472 && a
->peer
== b
->peer
);
1476 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1479 /* XXX Should limit the number of queued port status change messages. */
1480 struct ofconn
*ofconn
;
1481 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
1482 struct ofp_port_status
*ops
;
1485 if (!ofconn_receives_async_msgs(ofconn
)) {
1489 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1490 ops
->reason
= reason
;
1491 ops
->desc
= ofport
->opp
;
1492 hton_ofp_phy_port(&ops
->desc
);
1493 queue_tx(b
, ofconn
, NULL
);
1495 if (p
->ofhooks
->port_changed_cb
) {
1496 p
->ofhooks
->port_changed_cb(reason
, &ofport
->opp
, p
->aux
);
1501 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1503 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1505 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1506 hmap_insert(&p
->ports
, &ofport
->hmap_node
, hash_int(ofport
->odp_port
, 0));
1507 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1509 ofproto_sflow_add_port(p
->sflow
, ofport
->odp_port
, netdev_name
);
1514 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1516 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1517 hmap_remove(&p
->ports
, &ofport
->hmap_node
);
1518 shash_delete(&p
->port_by_name
,
1519 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1521 ofproto_sflow_del_port(p
->sflow
, ofport
->odp_port
);
1526 ofport_free(struct ofport
*ofport
)
1529 netdev_close(ofport
->netdev
);
1534 static struct ofport
*
1535 get_port(const struct ofproto
*ofproto
, uint16_t odp_port
)
1537 struct ofport
*port
;
1539 HMAP_FOR_EACH_IN_BUCKET (port
, hmap_node
,
1540 hash_int(odp_port
, 0), &ofproto
->ports
) {
1541 if (port
->odp_port
== odp_port
) {
1549 update_port(struct ofproto
*p
, const char *devname
)
1551 struct odp_port odp_port
;
1552 struct ofport
*old_ofport
;
1553 struct ofport
*new_ofport
;
1556 COVERAGE_INC(ofproto_update_port
);
1558 /* Query the datapath for port information. */
1559 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1561 /* Find the old ofport. */
1562 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1565 /* There's no port named 'devname' but there might be a port with
1566 * the same port number. This could happen if a port is deleted
1567 * and then a new one added in its place very quickly, or if a port
1568 * is renamed. In the former case we want to send an OFPPR_DELETE
1569 * and an OFPPR_ADD, and in the latter case we want to send a
1570 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1571 * the old port's ifindex against the new port, or perhaps less
1572 * reliably but more portably by comparing the old port's MAC
1573 * against the new port's MAC. However, this code isn't that smart
1574 * and always sends an OFPPR_MODIFY (XXX). */
1575 old_ofport
= get_port(p
, odp_port
.port
);
1577 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1578 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1579 "%s", strerror(error
));
1583 /* Create a new ofport. */
1584 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1586 /* Eliminate a few pathological cases. */
1587 if (!old_ofport
&& !new_ofport
) {
1589 } else if (old_ofport
&& new_ofport
) {
1590 /* Most of the 'config' bits are OpenFlow soft state, but
1591 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1592 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1593 * leaves the other bits 0.) */
1594 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1596 if (ofport_equal(old_ofport
, new_ofport
)) {
1597 /* False alarm--no change. */
1598 ofport_free(new_ofport
);
1603 /* Now deal with the normal cases. */
1605 ofport_remove(p
, old_ofport
);
1608 ofport_install(p
, new_ofport
);
1610 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1611 (!old_ofport
? OFPPR_ADD
1612 : !new_ofport
? OFPPR_DELETE
1614 ofport_free(old_ofport
);
1616 /* Update port groups. */
1617 refresh_port_groups(p
);
1621 init_ports(struct ofproto
*p
)
1623 struct odp_port
*ports
;
1628 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1633 for (i
= 0; i
< n_ports
; i
++) {
1634 const struct odp_port
*odp_port
= &ports
[i
];
1635 if (!ofport_conflicts(p
, odp_port
)) {
1636 struct ofport
*ofport
= make_ofport(odp_port
);
1638 ofport_install(p
, ofport
);
1643 refresh_port_groups(p
);
1647 static struct ofconn
*
1648 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1650 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1651 ofconn
->ofproto
= p
;
1652 list_push_back(&p
->all_conns
, &ofconn
->node
);
1653 ofconn
->rconn
= rconn
;
1654 ofconn
->type
= type
;
1655 ofconn
->role
= NX_ROLE_OTHER
;
1656 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1657 ofconn
->pktbuf
= NULL
;
1658 ofconn
->miss_send_len
= 0;
1659 ofconn
->reply_counter
= rconn_packet_counter_create ();
1664 ofconn_destroy(struct ofconn
*ofconn
)
1666 if (ofconn
->type
== OFCONN_PRIMARY
) {
1667 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1669 discovery_destroy(ofconn
->discovery
);
1671 list_remove(&ofconn
->node
);
1672 switch_status_unregister(ofconn
->ss
);
1673 rconn_destroy(ofconn
->rconn
);
1674 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1675 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1676 pktbuf_destroy(ofconn
->pktbuf
);
1681 ofconn_run(struct ofconn
*ofconn
, struct ofproto
*p
)
1686 if (ofconn
->discovery
) {
1687 char *controller_name
;
1688 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1689 discovery_question_connectivity(ofconn
->discovery
);
1691 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1692 if (controller_name
) {
1693 char *ofconn_name
= ofconn_make_name(p
, controller_name
);
1694 rconn_connect(ofconn
->rconn
, controller_name
, ofconn_name
);
1697 rconn_disconnect(ofconn
->rconn
);
1702 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1703 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1706 rconn_run(ofconn
->rconn
);
1708 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1709 /* Limit the number of iterations to prevent other tasks from
1711 for (iteration
= 0; iteration
< 50; iteration
++) {
1712 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1717 fail_open_maybe_recover(p
->fail_open
);
1719 handle_openflow(ofconn
, p
, of_msg
);
1720 ofpbuf_delete(of_msg
);
1724 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1725 ofconn_destroy(ofconn
);
1730 ofconn_wait(struct ofconn
*ofconn
)
1734 if (ofconn
->discovery
) {
1735 discovery_wait(ofconn
->discovery
);
1737 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1738 pinsched_wait(ofconn
->schedulers
[i
]);
1740 rconn_run_wait(ofconn
->rconn
);
1741 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1742 rconn_recv_wait(ofconn
->rconn
);
1744 COVERAGE_INC(ofproto_ofconn_stuck
);
1748 /* Returns true if 'ofconn' should receive asynchronous messages. */
1750 ofconn_receives_async_msgs(const struct ofconn
*ofconn
)
1752 if (ofconn
->type
== OFCONN_PRIMARY
) {
1753 /* Primary controllers always get asynchronous messages unless they
1754 * have configured themselves as "slaves". */
1755 return ofconn
->role
!= NX_ROLE_SLAVE
;
1757 /* Service connections don't get asynchronous messages unless they have
1758 * explicitly asked for them by setting a nonzero miss send length. */
1759 return ofconn
->miss_send_len
> 0;
1763 /* Returns a human-readable name for an OpenFlow connection between 'ofproto'
1764 * and 'target', suitable for use in log messages for identifying the
1767 * The name is dynamically allocated. The caller should free it (with free())
1768 * when it is no longer needed. */
1770 ofconn_make_name(const struct ofproto
*ofproto
, const char *target
)
1772 return xasprintf("%s<->%s", dpif_base_name(ofproto
->dpif
), target
);
1776 ofconn_set_rate_limit(struct ofconn
*ofconn
, int rate
, int burst
)
1780 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1781 struct pinsched
**s
= &ofconn
->schedulers
[i
];
1785 *s
= pinsched_create(rate
, burst
,
1786 ofconn
->ofproto
->switch_status
);
1788 pinsched_set_limits(*s
, rate
, burst
);
1791 pinsched_destroy(*s
);
1798 ofservice_reconfigure(struct ofservice
*ofservice
,
1799 const struct ofproto_controller
*c
)
1801 ofservice
->probe_interval
= c
->probe_interval
;
1802 ofservice
->rate_limit
= c
->rate_limit
;
1803 ofservice
->burst_limit
= c
->burst_limit
;
1806 /* Creates a new ofservice in 'ofproto'. Returns 0 if successful, otherwise a
1807 * positive errno value. */
1809 ofservice_create(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
1811 struct ofservice
*ofservice
;
1812 struct pvconn
*pvconn
;
1815 error
= pvconn_open(c
->target
, &pvconn
);
1820 ofservice
= xzalloc(sizeof *ofservice
);
1821 hmap_insert(&ofproto
->services
, &ofservice
->node
,
1822 hash_string(c
->target
, 0));
1823 ofservice
->pvconn
= pvconn
;
1825 ofservice_reconfigure(ofservice
, c
);
1831 ofservice_destroy(struct ofproto
*ofproto
, struct ofservice
*ofservice
)
1833 hmap_remove(&ofproto
->services
, &ofservice
->node
);
1834 pvconn_close(ofservice
->pvconn
);
1838 /* Finds and returns the ofservice within 'ofproto' that has the given
1839 * 'target', or a null pointer if none exists. */
1840 static struct ofservice
*
1841 ofservice_lookup(struct ofproto
*ofproto
, const char *target
)
1843 struct ofservice
*ofservice
;
1845 HMAP_FOR_EACH_WITH_HASH (ofservice
, node
, hash_string(target
, 0),
1846 &ofproto
->services
) {
1847 if (!strcmp(pvconn_get_name(ofservice
->pvconn
), target
)) {
1854 /* Caller is responsible for initializing the 'cr' member of the returned
1856 static struct rule
*
1857 rule_create(struct ofproto
*ofproto
, struct rule
*super
,
1858 const union ofp_action
*actions
, size_t n_actions
,
1859 uint16_t idle_timeout
, uint16_t hard_timeout
,
1860 uint64_t flow_cookie
, bool send_flow_removed
)
1862 struct rule
*rule
= xzalloc(sizeof *rule
);
1863 rule
->idle_timeout
= idle_timeout
;
1864 rule
->hard_timeout
= hard_timeout
;
1865 rule
->flow_cookie
= flow_cookie
;
1866 rule
->used
= rule
->created
= time_msec();
1867 rule
->send_flow_removed
= send_flow_removed
;
1868 rule
->super
= super
;
1870 list_push_back(&super
->list
, &rule
->list
);
1872 list_init(&rule
->list
);
1874 rule
->n_actions
= n_actions
;
1875 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1876 netflow_flow_clear(&rule
->nf_flow
);
1877 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->created
);
1882 static struct rule
*
1883 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1885 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1889 rule_free(struct rule
*rule
)
1891 free(rule
->actions
);
1892 free(rule
->odp_actions
);
1896 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1897 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1898 * through all of its subrules and revalidates them, destroying any that no
1899 * longer has a super-rule (which is probably all of them).
1901 * Before calling this function, the caller must make have removed 'rule' from
1902 * the classifier. If 'rule' is an exact-match rule, the caller is also
1903 * responsible for ensuring that it has been uninstalled from the datapath. */
1905 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1908 struct rule
*subrule
, *next
;
1909 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
1910 revalidate_rule(ofproto
, subrule
);
1913 list_remove(&rule
->list
);
1919 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1921 const union ofp_action
*oa
;
1922 struct actions_iterator i
;
1924 if (out_port
== htons(OFPP_NONE
)) {
1927 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1928 oa
= actions_next(&i
)) {
1929 if (action_outputs_to_port(oa
, out_port
)) {
1936 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
1937 * 'packet', which arrived on 'in_port'.
1939 * Takes ownership of 'packet'. */
1941 execute_odp_actions(struct ofproto
*ofproto
, uint16_t in_port
,
1942 const union odp_action
*actions
, size_t n_actions
,
1943 struct ofpbuf
*packet
)
1945 if (n_actions
== 1 && actions
[0].type
== ODPAT_CONTROLLER
) {
1946 /* As an optimization, avoid a round-trip from userspace to kernel to
1947 * userspace. This also avoids possibly filling up kernel packet
1948 * buffers along the way. */
1949 struct odp_msg
*msg
;
1951 msg
= ofpbuf_push_uninit(packet
, sizeof *msg
);
1952 msg
->type
= _ODPL_ACTION_NR
;
1953 msg
->length
= sizeof(struct odp_msg
) + packet
->size
;
1954 msg
->port
= in_port
;
1956 msg
->arg
= actions
[0].controller
.arg
;
1958 send_packet_in(ofproto
, packet
);
1964 error
= dpif_execute(ofproto
->dpif
, in_port
,
1965 actions
, n_actions
, packet
);
1966 ofpbuf_delete(packet
);
1971 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1972 * 'flow' and is considered to have arrived on ODP port 'in_port'. 'packet'
1973 * must have at least sizeof(struct ofp_packet_in) bytes of headroom.
1975 * The flow that 'packet' actually contains does not need to actually match
1976 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1977 * the packet and byte counters for 'rule' will be credited for the packet sent
1978 * out whether or not the packet actually matches 'rule'.
1980 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1981 * the caller must already have accurately composed ODP actions for it given
1982 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1983 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1984 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1985 * actions and apply them to 'packet'.
1987 * Takes ownership of 'packet'. */
1989 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1990 struct ofpbuf
*packet
, const flow_t
*flow
)
1992 const union odp_action
*actions
;
1993 struct odp_flow_stats stats
;
1995 struct odp_actions a
;
1997 assert(ofpbuf_headroom(packet
) >= sizeof(struct ofp_packet_in
));
1999 /* Grab or compose the ODP actions.
2001 * The special case for an exact-match 'rule' where 'flow' is not the
2002 * rule's flow is important to avoid, e.g., sending a packet out its input
2003 * port simply because the ODP actions were composed for the wrong
2005 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
2006 struct rule
*super
= rule
->super
? rule
->super
: rule
;
2007 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
2008 packet
, &a
, NULL
, 0, NULL
)) {
2009 ofpbuf_delete(packet
);
2012 actions
= a
.actions
;
2013 n_actions
= a
.n_actions
;
2015 actions
= rule
->odp_actions
;
2016 n_actions
= rule
->n_odp_actions
;
2019 /* Execute the ODP actions. */
2020 flow_extract_stats(flow
, packet
, &stats
);
2021 if (execute_odp_actions(ofproto
, flow
->in_port
,
2022 actions
, n_actions
, packet
)) {
2023 update_stats(ofproto
, rule
, &stats
);
2024 rule
->used
= time_msec();
2025 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->used
);
2029 /* Inserts 'rule' into 'p''s flow table.
2031 * If 'packet' is nonnull, takes ownership of 'packet', executes 'rule''s
2032 * actions on it and credits the statistics for sending the packet to 'rule'.
2033 * 'packet' must have at least sizeof(struct ofp_packet_in) bytes of
2036 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
2039 struct rule
*displaced_rule
;
2041 /* Insert the rule in the classifier. */
2042 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
2043 if (!rule
->cr
.wc
.wildcards
) {
2044 rule_make_actions(p
, rule
, packet
);
2047 /* Send the packet and credit it to the rule. */
2050 flow_extract(packet
, 0, in_port
, &flow
);
2051 rule_execute(p
, rule
, packet
, &flow
);
2054 /* Install the rule in the datapath only after sending the packet, to
2055 * avoid packet reordering. */
2056 if (rule
->cr
.wc
.wildcards
) {
2057 COVERAGE_INC(ofproto_add_wc_flow
);
2058 p
->need_revalidate
= true;
2060 rule_install(p
, rule
, displaced_rule
);
2063 /* Free the rule that was displaced, if any. */
2064 if (displaced_rule
) {
2065 rule_destroy(p
, displaced_rule
);
2069 static struct rule
*
2070 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
2073 struct rule
*subrule
= rule_create(ofproto
, rule
, NULL
, 0,
2074 rule
->idle_timeout
, rule
->hard_timeout
,
2076 COVERAGE_INC(ofproto_subrule_create
);
2077 cls_rule_from_flow(flow
, 0, (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
2078 : rule
->cr
.priority
), &subrule
->cr
);
2079 classifier_insert_exact(&ofproto
->cls
, &subrule
->cr
);
2085 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
2087 if (rule
->cr
.wc
.wildcards
) {
2088 COVERAGE_INC(ofproto_del_wc_flow
);
2089 ofproto
->need_revalidate
= true;
2091 rule_uninstall(ofproto
, rule
);
2093 classifier_remove(&ofproto
->cls
, &rule
->cr
);
2094 rule_destroy(ofproto
, rule
);
2097 /* Returns true if the actions changed, false otherwise. */
2099 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
2100 const struct ofpbuf
*packet
)
2102 const struct rule
*super
;
2103 struct odp_actions a
;
2106 assert(!rule
->cr
.wc
.wildcards
);
2108 super
= rule
->super
? rule
->super
: rule
;
2110 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
2111 packet
, &a
, &rule
->tags
, &rule
->may_install
,
2112 &rule
->nf_flow
.output_iface
);
2114 actions_len
= a
.n_actions
* sizeof *a
.actions
;
2115 if (rule
->n_odp_actions
!= a
.n_actions
2116 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
2117 COVERAGE_INC(ofproto_odp_unchanged
);
2118 free(rule
->odp_actions
);
2119 rule
->n_odp_actions
= a
.n_actions
;
2120 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
2128 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
2129 struct odp_flow_put
*put
)
2131 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
2132 put
->flow
.key
= rule
->cr
.flow
;
2133 put
->flow
.actions
= rule
->odp_actions
;
2134 put
->flow
.n_actions
= rule
->n_odp_actions
;
2135 put
->flow
.flags
= 0;
2137 return dpif_flow_put(ofproto
->dpif
, put
);
2141 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
2143 assert(!rule
->cr
.wc
.wildcards
);
2145 if (rule
->may_install
) {
2146 struct odp_flow_put put
;
2147 if (!do_put_flow(p
, rule
,
2148 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
2150 rule
->installed
= true;
2151 if (displaced_rule
) {
2152 update_stats(p
, displaced_rule
, &put
.flow
.stats
);
2153 rule_post_uninstall(p
, displaced_rule
);
2156 } else if (displaced_rule
) {
2157 rule_uninstall(p
, displaced_rule
);
2162 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
2164 if (rule
->installed
) {
2165 struct odp_flow_put put
;
2166 COVERAGE_INC(ofproto_dp_missed
);
2167 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
2169 rule_install(ofproto
, rule
, NULL
);
2174 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
2176 bool actions_changed
;
2177 uint16_t new_out_iface
, old_out_iface
;
2179 old_out_iface
= rule
->nf_flow
.output_iface
;
2180 actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
2182 if (rule
->may_install
) {
2183 if (rule
->installed
) {
2184 if (actions_changed
) {
2185 struct odp_flow_put put
;
2186 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
2187 | ODPPF_ZERO_STATS
, &put
);
2188 update_stats(ofproto
, rule
, &put
.flow
.stats
);
2190 /* Temporarily set the old output iface so that NetFlow
2191 * messages have the correct output interface for the old
2193 new_out_iface
= rule
->nf_flow
.output_iface
;
2194 rule
->nf_flow
.output_iface
= old_out_iface
;
2195 rule_post_uninstall(ofproto
, rule
);
2196 rule
->nf_flow
.output_iface
= new_out_iface
;
2199 rule_install(ofproto
, rule
, NULL
);
2202 rule_uninstall(ofproto
, rule
);
2207 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
2209 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
2211 if (ofproto
->ofhooks
->account_flow_cb
2212 && total_bytes
> rule
->accounted_bytes
)
2214 ofproto
->ofhooks
->account_flow_cb(
2215 &rule
->cr
.flow
, rule
->tags
, rule
->odp_actions
, rule
->n_odp_actions
,
2216 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
2217 rule
->accounted_bytes
= total_bytes
;
2222 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
2224 assert(!rule
->cr
.wc
.wildcards
);
2225 if (rule
->installed
) {
2226 struct odp_flow odp_flow
;
2228 odp_flow
.key
= rule
->cr
.flow
;
2229 odp_flow
.actions
= NULL
;
2230 odp_flow
.n_actions
= 0;
2232 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
2233 update_stats(p
, rule
, &odp_flow
.stats
);
2235 rule
->installed
= false;
2237 rule_post_uninstall(p
, rule
);
2242 is_controller_rule(struct rule
*rule
)
2244 /* If the only action is send to the controller then don't report
2245 * NetFlow expiration messages since it is just part of the control
2246 * logic for the network and not real traffic. */
2250 && rule
->super
->n_actions
== 1
2251 && action_outputs_to_port(&rule
->super
->actions
[0],
2252 htons(OFPP_CONTROLLER
)));
2256 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
2258 struct rule
*super
= rule
->super
;
2260 rule_account(ofproto
, rule
, 0);
2262 if (ofproto
->netflow
&& !is_controller_rule(rule
)) {
2263 struct ofexpired expired
;
2264 expired
.flow
= rule
->cr
.flow
;
2265 expired
.packet_count
= rule
->packet_count
;
2266 expired
.byte_count
= rule
->byte_count
;
2267 expired
.used
= rule
->used
;
2268 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
2271 super
->packet_count
+= rule
->packet_count
;
2272 super
->byte_count
+= rule
->byte_count
;
2274 /* Reset counters to prevent double counting if the rule ever gets
2276 rule
->packet_count
= 0;
2277 rule
->byte_count
= 0;
2278 rule
->accounted_bytes
= 0;
2280 netflow_flow_clear(&rule
->nf_flow
);
2285 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
2286 struct rconn_packet_counter
*counter
)
2288 update_openflow_length(msg
);
2289 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2295 send_error(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2296 int error
, const void *data
, size_t len
)
2299 struct ofp_error_msg
*oem
;
2301 if (!(error
>> 16)) {
2302 VLOG_WARN_RL(&rl
, "not sending bad error code %d to controller",
2307 COVERAGE_INC(ofproto_error
);
2308 oem
= make_openflow_xid(len
+ sizeof *oem
, OFPT_ERROR
,
2309 oh
? oh
->xid
: 0, &buf
);
2310 oem
->type
= htons((unsigned int) error
>> 16);
2311 oem
->code
= htons(error
& 0xffff);
2312 memcpy(oem
->data
, data
, len
);
2313 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2317 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2320 size_t oh_length
= ntohs(oh
->length
);
2321 send_error(ofconn
, oh
, error
, oh
, MIN(oh_length
, 64));
2325 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2327 opp
->port_no
= htons(opp
->port_no
);
2328 opp
->config
= htonl(opp
->config
);
2329 opp
->state
= htonl(opp
->state
);
2330 opp
->curr
= htonl(opp
->curr
);
2331 opp
->advertised
= htonl(opp
->advertised
);
2332 opp
->supported
= htonl(opp
->supported
);
2333 opp
->peer
= htonl(opp
->peer
);
2337 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2339 struct ofp_header
*rq
= oh
;
2340 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2345 handle_features_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2346 struct ofp_header
*oh
)
2348 struct ofp_switch_features
*osf
;
2350 struct ofport
*port
;
2352 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2353 osf
->datapath_id
= htonll(p
->datapath_id
);
2354 osf
->n_buffers
= htonl(pktbuf_capacity());
2356 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2357 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2358 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2359 (1u << OFPAT_SET_VLAN_VID
) |
2360 (1u << OFPAT_SET_VLAN_PCP
) |
2361 (1u << OFPAT_STRIP_VLAN
) |
2362 (1u << OFPAT_SET_DL_SRC
) |
2363 (1u << OFPAT_SET_DL_DST
) |
2364 (1u << OFPAT_SET_NW_SRC
) |
2365 (1u << OFPAT_SET_NW_DST
) |
2366 (1u << OFPAT_SET_NW_TOS
) |
2367 (1u << OFPAT_SET_TP_SRC
) |
2368 (1u << OFPAT_SET_TP_DST
) |
2369 (1u << OFPAT_ENQUEUE
));
2371 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
2372 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2375 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2380 handle_get_config_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2381 struct ofp_header
*oh
)
2384 struct ofp_switch_config
*osc
;
2388 /* Figure out flags. */
2389 dpif_get_drop_frags(p
->dpif
, &drop_frags
);
2390 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2393 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2394 osc
->flags
= htons(flags
);
2395 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2396 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2402 handle_set_config(struct ofproto
*p
, struct ofconn
*ofconn
,
2403 struct ofp_switch_config
*osc
)
2408 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2412 flags
= ntohs(osc
->flags
);
2414 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2415 switch (flags
& OFPC_FRAG_MASK
) {
2416 case OFPC_FRAG_NORMAL
:
2417 dpif_set_drop_frags(p
->dpif
, false);
2419 case OFPC_FRAG_DROP
:
2420 dpif_set_drop_frags(p
->dpif
, true);
2423 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2429 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2435 add_output_group_action(struct odp_actions
*actions
, uint16_t group
,
2436 uint16_t *nf_output_iface
)
2438 odp_actions_add(actions
, ODPAT_OUTPUT_GROUP
)->output_group
.group
= group
;
2440 if (group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
) {
2441 *nf_output_iface
= NF_OUT_FLOOD
;
2446 add_controller_action(struct odp_actions
*actions
, uint16_t max_len
)
2448 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2449 a
->controller
.arg
= max_len
;
2452 struct action_xlate_ctx
{
2454 flow_t flow
; /* Flow to which these actions correspond. */
2455 int recurse
; /* Recursion level, via xlate_table_action. */
2456 struct ofproto
*ofproto
;
2457 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2458 * null pointer if we are revalidating
2459 * without a packet to refer to. */
2462 struct odp_actions
*out
; /* Datapath actions. */
2463 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
2464 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2465 * be reassessed for every packet. */
2466 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2469 /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a
2470 * flow translation. */
2471 #define MAX_RESUBMIT_RECURSION 8
2473 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2474 struct action_xlate_ctx
*ctx
);
2477 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2479 const struct ofport
*ofport
= get_port(ctx
->ofproto
, port
);
2482 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2483 /* Forwarding disabled on port. */
2488 * We don't have an ofport record for this port, but it doesn't hurt to
2489 * allow forwarding to it anyhow. Maybe such a port will appear later
2490 * and we're pre-populating the flow table.
2494 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2495 ctx
->nf_output_iface
= port
;
2498 static struct rule
*
2499 lookup_valid_rule(struct ofproto
*ofproto
, const flow_t
*flow
)
2502 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2504 /* The rule we found might not be valid, since we could be in need of
2505 * revalidation. If it is not valid, don't return it. */
2508 && ofproto
->need_revalidate
2509 && !revalidate_rule(ofproto
, rule
)) {
2510 COVERAGE_INC(ofproto_invalidated
);
2518 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2520 if (ctx
->recurse
< MAX_RESUBMIT_RECURSION
) {
2521 uint16_t old_in_port
;
2524 /* Look up a flow with 'in_port' as the input port. Then restore the
2525 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2526 * have surprising behavior). */
2527 old_in_port
= ctx
->flow
.in_port
;
2528 ctx
->flow
.in_port
= in_port
;
2529 rule
= lookup_valid_rule(ctx
->ofproto
, &ctx
->flow
);
2530 ctx
->flow
.in_port
= old_in_port
;
2538 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2542 struct vlog_rate_limit recurse_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
2544 VLOG_ERR_RL(&recurse_rl
, "NXAST_RESUBMIT recursed over %d times",
2545 MAX_RESUBMIT_RECURSION
);
2550 xlate_output_action__(struct action_xlate_ctx
*ctx
,
2551 uint16_t port
, uint16_t max_len
)
2554 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2556 ctx
->nf_output_iface
= NF_OUT_DROP
;
2560 add_output_action(ctx
, ctx
->flow
.in_port
);
2563 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2566 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2567 ctx
->out
, ctx
->tags
,
2568 &ctx
->nf_output_iface
,
2569 ctx
->ofproto
->aux
)) {
2570 COVERAGE_INC(ofproto_uninstallable
);
2571 ctx
->may_set_up_flow
= false;
2575 add_output_group_action(ctx
->out
, DP_GROUP_FLOOD
,
2576 &ctx
->nf_output_iface
);
2579 add_output_group_action(ctx
->out
, DP_GROUP_ALL
, &ctx
->nf_output_iface
);
2581 case OFPP_CONTROLLER
:
2582 add_controller_action(ctx
->out
, max_len
);
2585 add_output_action(ctx
, ODPP_LOCAL
);
2588 odp_port
= ofp_port_to_odp_port(port
);
2589 if (odp_port
!= ctx
->flow
.in_port
) {
2590 add_output_action(ctx
, odp_port
);
2595 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2596 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2597 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2598 ctx
->nf_output_iface
= prev_nf_output_iface
;
2599 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2600 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2601 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2606 xlate_output_action(struct action_xlate_ctx
*ctx
,
2607 const struct ofp_action_output
*oao
)
2609 xlate_output_action__(ctx
, ntohs(oao
->port
), ntohs(oao
->max_len
));
2612 /* If the final ODP action in 'ctx' is "pop priority", drop it, as an
2613 * optimization, because we're going to add another action that sets the
2614 * priority immediately after, or because there are no actions following the
2617 remove_pop_action(struct action_xlate_ctx
*ctx
)
2619 size_t n
= ctx
->out
->n_actions
;
2620 if (n
> 0 && ctx
->out
->actions
[n
- 1].type
== ODPAT_POP_PRIORITY
) {
2621 ctx
->out
->n_actions
--;
2626 xlate_enqueue_action(struct action_xlate_ctx
*ctx
,
2627 const struct ofp_action_enqueue
*oae
)
2629 uint16_t ofp_port
, odp_port
;
2633 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(oae
->queue_id
),
2636 /* Fall back to ordinary output action. */
2637 xlate_output_action__(ctx
, ntohs(oae
->port
), 0);
2641 /* Figure out ODP output port. */
2642 ofp_port
= ntohs(oae
->port
);
2643 if (ofp_port
!= OFPP_IN_PORT
) {
2644 odp_port
= ofp_port_to_odp_port(ofp_port
);
2646 odp_port
= ctx
->flow
.in_port
;
2649 /* Add ODP actions. */
2650 remove_pop_action(ctx
);
2651 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2653 add_output_action(ctx
, odp_port
);
2654 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2656 /* Update NetFlow output port. */
2657 if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2658 ctx
->nf_output_iface
= odp_port
;
2659 } else if (ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2660 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2665 xlate_set_queue_action(struct action_xlate_ctx
*ctx
,
2666 const struct nx_action_set_queue
*nasq
)
2671 error
= dpif_queue_to_priority(ctx
->ofproto
->dpif
, ntohl(nasq
->queue_id
),
2674 /* Couldn't translate queue to a priority, so ignore. A warning
2675 * has already been logged. */
2679 remove_pop_action(ctx
);
2680 odp_actions_add(ctx
->out
, ODPAT_SET_PRIORITY
)->priority
.priority
2685 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2686 const struct nx_action_header
*nah
)
2688 const struct nx_action_resubmit
*nar
;
2689 const struct nx_action_set_tunnel
*nast
;
2690 const struct nx_action_set_queue
*nasq
;
2691 union odp_action
*oa
;
2692 int subtype
= ntohs(nah
->subtype
);
2694 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2696 case NXAST_RESUBMIT
:
2697 nar
= (const struct nx_action_resubmit
*) nah
;
2698 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2701 case NXAST_SET_TUNNEL
:
2702 nast
= (const struct nx_action_set_tunnel
*) nah
;
2703 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2704 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2707 case NXAST_DROP_SPOOFED_ARP
:
2708 if (ctx
->flow
.dl_type
== htons(ETH_TYPE_ARP
)) {
2709 odp_actions_add(ctx
->out
, ODPAT_DROP_SPOOFED_ARP
);
2713 case NXAST_SET_QUEUE
:
2714 nasq
= (const struct nx_action_set_queue
*) nah
;
2715 xlate_set_queue_action(ctx
, nasq
);
2718 case NXAST_POP_QUEUE
:
2719 odp_actions_add(ctx
->out
, ODPAT_POP_PRIORITY
);
2722 /* If you add a new action here that modifies flow data, don't forget to
2723 * update the flow key in ctx->flow at the same time. */
2726 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2732 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2733 struct action_xlate_ctx
*ctx
)
2735 struct actions_iterator iter
;
2736 const union ofp_action
*ia
;
2737 const struct ofport
*port
;
2739 port
= get_port(ctx
->ofproto
, ctx
->flow
.in_port
);
2740 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2741 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, eth_addr_stp
)
2742 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2743 /* Drop this flow. */
2747 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2748 uint16_t type
= ntohs(ia
->type
);
2749 union odp_action
*oa
;
2753 xlate_output_action(ctx
, &ia
->output
);
2756 case OFPAT_SET_VLAN_VID
:
2757 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_VID
);
2758 ctx
->flow
.dl_vlan
= oa
->vlan_vid
.vlan_vid
= ia
->vlan_vid
.vlan_vid
;
2761 case OFPAT_SET_VLAN_PCP
:
2762 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_PCP
);
2763 ctx
->flow
.dl_vlan_pcp
= oa
->vlan_pcp
.vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2766 case OFPAT_STRIP_VLAN
:
2767 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2768 ctx
->flow
.dl_vlan
= htons(OFP_VLAN_NONE
);
2769 ctx
->flow
.dl_vlan_pcp
= 0;
2772 case OFPAT_SET_DL_SRC
:
2773 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2774 memcpy(oa
->dl_addr
.dl_addr
,
2775 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2776 memcpy(ctx
->flow
.dl_src
,
2777 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2780 case OFPAT_SET_DL_DST
:
2781 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2782 memcpy(oa
->dl_addr
.dl_addr
,
2783 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2784 memcpy(ctx
->flow
.dl_dst
,
2785 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2788 case OFPAT_SET_NW_SRC
:
2789 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2790 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2793 case OFPAT_SET_NW_DST
:
2794 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2795 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2798 case OFPAT_SET_NW_TOS
:
2799 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2800 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2803 case OFPAT_SET_TP_SRC
:
2804 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2805 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2808 case OFPAT_SET_TP_DST
:
2809 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2810 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2814 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2818 xlate_enqueue_action(ctx
, (const struct ofp_action_enqueue
*) ia
);
2822 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2829 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2830 const flow_t
*flow
, struct ofproto
*ofproto
,
2831 const struct ofpbuf
*packet
,
2832 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2833 uint16_t *nf_output_iface
)
2835 tag_type no_tags
= 0;
2836 struct action_xlate_ctx ctx
;
2837 COVERAGE_INC(ofproto_ofp2odp
);
2838 odp_actions_init(out
);
2841 ctx
.ofproto
= ofproto
;
2842 ctx
.packet
= packet
;
2844 ctx
.tags
= tags
? tags
: &no_tags
;
2845 ctx
.may_set_up_flow
= true;
2846 ctx
.nf_output_iface
= NF_OUT_DROP
;
2847 do_xlate_actions(in
, n_in
, &ctx
);
2848 remove_pop_action(&ctx
);
2850 /* Check with in-band control to see if we're allowed to set up this
2852 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2853 ctx
.may_set_up_flow
= false;
2856 if (may_set_up_flow
) {
2857 *may_set_up_flow
= ctx
.may_set_up_flow
;
2859 if (nf_output_iface
) {
2860 *nf_output_iface
= ctx
.nf_output_iface
;
2862 if (odp_actions_overflow(out
)) {
2863 COVERAGE_INC(odp_overflow
);
2864 odp_actions_init(out
);
2865 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
2870 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2871 * error message code (composed with ofp_mkerr()) for the caller to propagate
2872 * upward. Otherwise, returns 0.
2874 * 'oh' is used to make log messages more informative. */
2876 reject_slave_controller(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2878 if (ofconn
->type
== OFCONN_PRIMARY
&& ofconn
->role
== NX_ROLE_SLAVE
) {
2879 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2882 type_name
= ofp_message_type_to_string(oh
->type
);
2883 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
2887 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
2894 handle_packet_out(struct ofproto
*p
, struct ofconn
*ofconn
,
2895 struct ofp_header
*oh
)
2897 struct ofp_packet_out
*opo
;
2898 struct ofpbuf payload
, *buffer
;
2899 struct odp_actions actions
;
2905 error
= reject_slave_controller(ofconn
, oh
);
2910 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2914 opo
= (struct ofp_packet_out
*) oh
;
2916 COVERAGE_INC(ofproto_packet_out
);
2917 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2918 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2920 if (error
|| !buffer
) {
2928 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2929 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2930 &flow
, p
, &payload
, &actions
, NULL
, NULL
, NULL
);
2935 dpif_execute(p
->dpif
, flow
.in_port
, actions
.actions
, actions
.n_actions
,
2937 ofpbuf_delete(buffer
);
2943 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2944 uint32_t config
, uint32_t mask
)
2946 mask
&= config
^ port
->opp
.config
;
2947 if (mask
& OFPPC_PORT_DOWN
) {
2948 if (config
& OFPPC_PORT_DOWN
) {
2949 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2951 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2954 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2955 if (mask
& REVALIDATE_BITS
) {
2956 COVERAGE_INC(ofproto_costly_flags
);
2957 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2958 p
->need_revalidate
= true;
2960 #undef REVALIDATE_BITS
2961 if (mask
& OFPPC_NO_FLOOD
) {
2962 port
->opp
.config
^= OFPPC_NO_FLOOD
;
2963 refresh_port_groups(p
);
2965 if (mask
& OFPPC_NO_PACKET_IN
) {
2966 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2971 handle_port_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
2972 struct ofp_header
*oh
)
2974 const struct ofp_port_mod
*opm
;
2975 struct ofport
*port
;
2978 error
= reject_slave_controller(ofconn
, oh
);
2982 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2986 opm
= (struct ofp_port_mod
*) oh
;
2988 port
= get_port(p
, ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2990 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2991 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2992 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2994 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2995 if (opm
->advertise
) {
2996 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
3002 static struct ofpbuf
*
3003 make_stats_reply(uint32_t xid
, uint16_t type
, size_t body_len
)
3005 struct ofp_stats_reply
*osr
;
3008 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
3009 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
3011 osr
->flags
= htons(0);
3015 static struct ofpbuf
*
3016 start_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
3018 return make_stats_reply(request
->header
.xid
, request
->type
, body_len
);
3022 append_stats_reply(size_t nbytes
, struct ofconn
*ofconn
, struct ofpbuf
**msgp
)
3024 struct ofpbuf
*msg
= *msgp
;
3025 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
3026 if (nbytes
+ msg
->size
> UINT16_MAX
) {
3027 struct ofp_stats_reply
*reply
= msg
->data
;
3028 reply
->flags
= htons(OFPSF_REPLY_MORE
);
3029 *msgp
= make_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
3030 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3032 return ofpbuf_put_uninit(*msgp
, nbytes
);
3036 handle_desc_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3037 struct ofp_stats_request
*request
)
3039 struct ofp_desc_stats
*ods
;
3042 msg
= start_stats_reply(request
, sizeof *ods
);
3043 ods
= append_stats_reply(sizeof *ods
, ofconn
, &msg
);
3044 memset(ods
, 0, sizeof *ods
);
3045 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
3046 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
3047 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
3048 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
3049 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
3050 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3056 count_subrules(struct cls_rule
*cls_rule
, void *n_subrules_
)
3058 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
3059 int *n_subrules
= n_subrules_
;
3067 handle_table_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3068 struct ofp_stats_request
*request
)
3070 struct ofp_table_stats
*ots
;
3072 struct odp_stats dpstats
;
3073 int n_exact
, n_subrules
, n_wild
;
3075 msg
= start_stats_reply(request
, sizeof *ots
* 2);
3077 /* Count rules of various kinds. */
3079 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, count_subrules
, &n_subrules
);
3080 n_exact
= classifier_count_exact(&p
->cls
) - n_subrules
;
3081 n_wild
= classifier_count(&p
->cls
) - classifier_count_exact(&p
->cls
);
3084 dpif_get_dp_stats(p
->dpif
, &dpstats
);
3085 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
3086 memset(ots
, 0, sizeof *ots
);
3087 ots
->table_id
= TABLEID_HASH
;
3088 strcpy(ots
->name
, "hash");
3089 ots
->wildcards
= htonl(0);
3090 ots
->max_entries
= htonl(dpstats
.max_capacity
);
3091 ots
->active_count
= htonl(n_exact
);
3092 ots
->lookup_count
= htonll(dpstats
.n_frags
+ dpstats
.n_hit
+
3094 ots
->matched_count
= htonll(dpstats
.n_hit
); /* XXX */
3096 /* Classifier table. */
3097 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
3098 memset(ots
, 0, sizeof *ots
);
3099 ots
->table_id
= TABLEID_CLASSIFIER
;
3100 strcpy(ots
->name
, "classifier");
3101 ots
->wildcards
= p
->tun_id_from_cookie
? htonl(OVSFW_ALL
)
3103 ots
->max_entries
= htonl(65536);
3104 ots
->active_count
= htonl(n_wild
);
3105 ots
->lookup_count
= htonll(0); /* XXX */
3106 ots
->matched_count
= htonll(0); /* XXX */
3108 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3113 append_port_stat(struct ofport
*port
, struct ofconn
*ofconn
,
3114 struct ofpbuf
**msgp
)
3116 struct netdev_stats stats
;
3117 struct ofp_port_stats
*ops
;
3119 /* Intentionally ignore return value, since errors will set
3120 * 'stats' to all-1s, which is correct for OpenFlow, and
3121 * netdev_get_stats() will log errors. */
3122 netdev_get_stats(port
->netdev
, &stats
);
3124 ops
= append_stats_reply(sizeof *ops
, ofconn
, msgp
);
3125 ops
->port_no
= htons(port
->opp
.port_no
);
3126 memset(ops
->pad
, 0, sizeof ops
->pad
);
3127 ops
->rx_packets
= htonll(stats
.rx_packets
);
3128 ops
->tx_packets
= htonll(stats
.tx_packets
);
3129 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
3130 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
3131 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
3132 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
3133 ops
->rx_errors
= htonll(stats
.rx_errors
);
3134 ops
->tx_errors
= htonll(stats
.tx_errors
);
3135 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
3136 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
3137 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
3138 ops
->collisions
= htonll(stats
.collisions
);
3142 handle_port_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3143 struct ofp_stats_request
*osr
,
3146 struct ofp_port_stats_request
*psr
;
3147 struct ofp_port_stats
*ops
;
3149 struct ofport
*port
;
3151 if (arg_size
!= sizeof *psr
) {
3152 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3154 psr
= (struct ofp_port_stats_request
*) osr
->body
;
3156 msg
= start_stats_reply(osr
, sizeof *ops
* 16);
3157 if (psr
->port_no
!= htons(OFPP_NONE
)) {
3158 port
= get_port(p
, ofp_port_to_odp_port(ntohs(psr
->port_no
)));
3160 append_port_stat(port
, ofconn
, &msg
);
3163 HMAP_FOR_EACH (port
, hmap_node
, &p
->ports
) {
3164 append_port_stat(port
, ofconn
, &msg
);
3168 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3172 struct flow_stats_cbdata
{
3173 struct ofproto
*ofproto
;
3174 struct ofconn
*ofconn
;
3179 /* Obtains statistic counters for 'rule' within 'p' and stores them into
3180 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
3181 * returned statistic include statistics for all of 'rule''s subrules. */
3183 query_stats(struct ofproto
*p
, struct rule
*rule
,
3184 uint64_t *packet_countp
, uint64_t *byte_countp
)
3186 uint64_t packet_count
, byte_count
;
3187 struct rule
*subrule
;
3188 struct odp_flow
*odp_flows
;
3191 /* Start from historical data for 'rule' itself that are no longer tracked
3192 * by the datapath. This counts, for example, subrules that have
3194 packet_count
= rule
->packet_count
;
3195 byte_count
= rule
->byte_count
;
3197 /* Prepare to ask the datapath for statistics on 'rule', or if it is
3198 * wildcarded then on all of its subrules.
3200 * Also, add any statistics that are not tracked by the datapath for each
3201 * subrule. This includes, for example, statistics for packets that were
3202 * executed "by hand" by ofproto via dpif_execute() but must be accounted
3204 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
3205 odp_flows
= xzalloc(n_odp_flows
* sizeof *odp_flows
);
3206 if (rule
->cr
.wc
.wildcards
) {
3208 LIST_FOR_EACH (subrule
, list
, &rule
->list
) {
3209 odp_flows
[i
++].key
= subrule
->cr
.flow
;
3210 packet_count
+= subrule
->packet_count
;
3211 byte_count
+= subrule
->byte_count
;
3214 odp_flows
[0].key
= rule
->cr
.flow
;
3217 /* Fetch up-to-date statistics from the datapath and add them in. */
3218 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
3220 for (i
= 0; i
< n_odp_flows
; i
++) {
3221 struct odp_flow
*odp_flow
= &odp_flows
[i
];
3222 packet_count
+= odp_flow
->stats
.n_packets
;
3223 byte_count
+= odp_flow
->stats
.n_bytes
;
3228 /* Return the stats to the caller. */
3229 *packet_countp
= packet_count
;
3230 *byte_countp
= byte_count
;
3234 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3236 struct rule
*rule
= rule_from_cls_rule(rule_
);
3237 struct flow_stats_cbdata
*cbdata
= cbdata_
;
3238 struct ofp_flow_stats
*ofs
;
3239 uint64_t packet_count
, byte_count
;
3240 size_t act_len
, len
;
3241 long long int tdiff
= time_msec() - rule
->created
;
3242 uint32_t sec
= tdiff
/ 1000;
3243 uint32_t msec
= tdiff
- (sec
* 1000);
3245 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3249 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3250 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
3252 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3254 ofs
= append_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
3255 ofs
->length
= htons(len
);
3256 ofs
->table_id
= rule
->cr
.wc
.wildcards
? TABLEID_CLASSIFIER
: TABLEID_HASH
;
3258 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3259 cbdata
->ofproto
->tun_id_from_cookie
, &ofs
->match
);
3260 ofs
->duration_sec
= htonl(sec
);
3261 ofs
->duration_nsec
= htonl(msec
* 1000000);
3262 ofs
->cookie
= rule
->flow_cookie
;
3263 ofs
->priority
= htons(rule
->cr
.priority
);
3264 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
3265 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
3266 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
3267 ofs
->packet_count
= htonll(packet_count
);
3268 ofs
->byte_count
= htonll(byte_count
);
3269 memcpy(ofs
->actions
, rule
->actions
, act_len
);
3273 table_id_to_include(uint8_t table_id
)
3275 return (table_id
== TABLEID_HASH
? CLS_INC_EXACT
3276 : table_id
== TABLEID_CLASSIFIER
? CLS_INC_WILD
3277 : table_id
== 0xff ? CLS_INC_ALL
3282 handle_flow_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3283 const struct ofp_stats_request
*osr
,
3286 struct ofp_flow_stats_request
*fsr
;
3287 struct flow_stats_cbdata cbdata
;
3288 struct cls_rule target
;
3290 if (arg_size
!= sizeof *fsr
) {
3291 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3293 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
3295 COVERAGE_INC(ofproto_flows_req
);
3297 cbdata
.ofconn
= ofconn
;
3298 cbdata
.out_port
= fsr
->out_port
;
3299 cbdata
.msg
= start_stats_reply(osr
, 1024);
3300 cls_rule_from_match(&fsr
->match
, 0, false, 0, &target
);
3301 classifier_for_each_match(&p
->cls
, &target
,
3302 table_id_to_include(fsr
->table_id
),
3303 flow_stats_cb
, &cbdata
);
3304 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3308 struct flow_stats_ds_cbdata
{
3309 struct ofproto
*ofproto
;
3314 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
3316 struct rule
*rule
= rule_from_cls_rule(rule_
);
3317 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
3318 struct ds
*results
= cbdata
->results
;
3319 struct ofp_match match
;
3320 uint64_t packet_count
, byte_count
;
3321 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
3323 /* Don't report on subrules. */
3324 if (rule
->super
!= NULL
) {
3328 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3329 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
3330 cbdata
->ofproto
->tun_id_from_cookie
, &match
);
3332 ds_put_format(results
, "duration=%llds, ",
3333 (time_msec() - rule
->created
) / 1000);
3334 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
3335 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
3336 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
3337 ofp_print_match(results
, &match
, true);
3338 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
3339 ds_put_cstr(results
, "\n");
3342 /* Adds a pretty-printed description of all flows to 'results', including
3343 * those marked hidden by secchan (e.g., by in-band control). */
3345 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
3347 struct ofp_match match
;
3348 struct cls_rule target
;
3349 struct flow_stats_ds_cbdata cbdata
;
3351 memset(&match
, 0, sizeof match
);
3352 match
.wildcards
= htonl(OVSFW_ALL
);
3355 cbdata
.results
= results
;
3357 cls_rule_from_match(&match
, 0, false, 0, &target
);
3358 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3359 flow_stats_ds_cb
, &cbdata
);
3362 struct aggregate_stats_cbdata
{
3363 struct ofproto
*ofproto
;
3365 uint64_t packet_count
;
3366 uint64_t byte_count
;
3371 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
3373 struct rule
*rule
= rule_from_cls_rule(rule_
);
3374 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
3375 uint64_t packet_count
, byte_count
;
3377 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
3381 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
3383 cbdata
->packet_count
+= packet_count
;
3384 cbdata
->byte_count
+= byte_count
;
3389 handle_aggregate_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3390 const struct ofp_stats_request
*osr
,
3393 struct ofp_aggregate_stats_request
*asr
;
3394 struct ofp_aggregate_stats_reply
*reply
;
3395 struct aggregate_stats_cbdata cbdata
;
3396 struct cls_rule target
;
3399 if (arg_size
!= sizeof *asr
) {
3400 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3402 asr
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3404 COVERAGE_INC(ofproto_agg_request
);
3406 cbdata
.out_port
= asr
->out_port
;
3407 cbdata
.packet_count
= 0;
3408 cbdata
.byte_count
= 0;
3410 cls_rule_from_match(&asr
->match
, 0, false, 0, &target
);
3411 classifier_for_each_match(&p
->cls
, &target
,
3412 table_id_to_include(asr
->table_id
),
3413 aggregate_stats_cb
, &cbdata
);
3415 msg
= start_stats_reply(osr
, sizeof *reply
);
3416 reply
= append_stats_reply(sizeof *reply
, ofconn
, &msg
);
3417 reply
->flow_count
= htonl(cbdata
.n_flows
);
3418 reply
->packet_count
= htonll(cbdata
.packet_count
);
3419 reply
->byte_count
= htonll(cbdata
.byte_count
);
3420 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3424 struct queue_stats_cbdata
{
3425 struct ofconn
*ofconn
;
3426 struct ofport
*ofport
;
3431 put_queue_stats(struct queue_stats_cbdata
*cbdata
, uint32_t queue_id
,
3432 const struct netdev_queue_stats
*stats
)
3434 struct ofp_queue_stats
*reply
;
3436 reply
= append_stats_reply(sizeof *reply
, cbdata
->ofconn
, &cbdata
->msg
);
3437 reply
->port_no
= htons(cbdata
->ofport
->opp
.port_no
);
3438 memset(reply
->pad
, 0, sizeof reply
->pad
);
3439 reply
->queue_id
= htonl(queue_id
);
3440 reply
->tx_bytes
= htonll(stats
->tx_bytes
);
3441 reply
->tx_packets
= htonll(stats
->tx_packets
);
3442 reply
->tx_errors
= htonll(stats
->tx_errors
);
3446 handle_queue_stats_dump_cb(uint32_t queue_id
,
3447 struct netdev_queue_stats
*stats
,
3450 struct queue_stats_cbdata
*cbdata
= cbdata_
;
3452 put_queue_stats(cbdata
, queue_id
, stats
);
3456 handle_queue_stats_for_port(struct ofport
*port
, uint32_t queue_id
,
3457 struct queue_stats_cbdata
*cbdata
)
3459 cbdata
->ofport
= port
;
3460 if (queue_id
== OFPQ_ALL
) {
3461 netdev_dump_queue_stats(port
->netdev
,
3462 handle_queue_stats_dump_cb
, cbdata
);
3464 struct netdev_queue_stats stats
;
3466 if (!netdev_get_queue_stats(port
->netdev
, queue_id
, &stats
)) {
3467 put_queue_stats(cbdata
, queue_id
, &stats
);
3473 handle_queue_stats_request(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3474 const struct ofp_stats_request
*osr
,
3477 struct ofp_queue_stats_request
*qsr
;
3478 struct queue_stats_cbdata cbdata
;
3479 struct ofport
*port
;
3480 unsigned int port_no
;
3483 if (arg_size
!= sizeof *qsr
) {
3484 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3486 qsr
= (struct ofp_queue_stats_request
*) osr
->body
;
3488 COVERAGE_INC(ofproto_queue_req
);
3490 cbdata
.ofconn
= ofconn
;
3491 cbdata
.msg
= start_stats_reply(osr
, 128);
3493 port_no
= ntohs(qsr
->port_no
);
3494 queue_id
= ntohl(qsr
->queue_id
);
3495 if (port_no
== OFPP_ALL
) {
3496 HMAP_FOR_EACH (port
, hmap_node
, &ofproto
->ports
) {
3497 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3499 } else if (port_no
< ofproto
->max_ports
) {
3500 port
= get_port(ofproto
, ofp_port_to_odp_port(port_no
));
3502 handle_queue_stats_for_port(port
, queue_id
, &cbdata
);
3505 ofpbuf_delete(cbdata
.msg
);
3506 return ofp_mkerr(OFPET_QUEUE_OP_FAILED
, OFPQOFC_BAD_PORT
);
3508 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
3514 handle_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3515 struct ofp_header
*oh
)
3517 struct ofp_stats_request
*osr
;
3521 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3526 osr
= (struct ofp_stats_request
*) oh
;
3528 switch (ntohs(osr
->type
)) {
3530 return handle_desc_stats_request(p
, ofconn
, osr
);
3533 return handle_flow_stats_request(p
, ofconn
, osr
, arg_size
);
3535 case OFPST_AGGREGATE
:
3536 return handle_aggregate_stats_request(p
, ofconn
, osr
, arg_size
);
3539 return handle_table_stats_request(p
, ofconn
, osr
);
3542 return handle_port_stats_request(p
, ofconn
, osr
, arg_size
);
3545 return handle_queue_stats_request(p
, ofconn
, osr
, arg_size
);
3548 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3551 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3555 static long long int
3556 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3558 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3562 update_time(struct ofproto
*ofproto
, struct rule
*rule
,
3563 const struct odp_flow_stats
*stats
)
3565 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3566 if (used
> rule
->used
) {
3568 if (rule
->super
&& used
> rule
->super
->used
) {
3569 rule
->super
->used
= used
;
3571 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, used
);
3576 update_stats(struct ofproto
*ofproto
, struct rule
*rule
,
3577 const struct odp_flow_stats
*stats
)
3579 if (stats
->n_packets
) {
3580 update_time(ofproto
, rule
, stats
);
3581 rule
->packet_count
+= stats
->n_packets
;
3582 rule
->byte_count
+= stats
->n_bytes
;
3583 netflow_flow_update_flags(&rule
->nf_flow
, stats
->tcp_flags
);
3587 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3588 * in which no matching flow already exists in the flow table.
3590 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3591 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3592 * code as encoded by ofp_mkerr() on failure.
3594 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3597 add_flow(struct ofproto
*p
, struct ofconn
*ofconn
,
3598 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3600 struct ofpbuf
*packet
;
3605 if (ofm
->flags
& htons(OFPFF_CHECK_OVERLAP
)) {
3609 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3611 if (classifier_rule_overlaps(&p
->cls
, &flow
, wildcards
,
3612 ntohs(ofm
->priority
))) {
3613 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3617 rule
= rule_create(p
, NULL
, (const union ofp_action
*) ofm
->actions
,
3618 n_actions
, ntohs(ofm
->idle_timeout
),
3619 ntohs(ofm
->hard_timeout
), ofm
->cookie
,
3620 ofm
->flags
& htons(OFPFF_SEND_FLOW_REM
));
3621 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
),
3622 p
->tun_id_from_cookie
, ofm
->cookie
, &rule
->cr
);
3625 if (ofm
->buffer_id
!= htonl(UINT32_MAX
)) {
3626 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3630 in_port
= UINT16_MAX
;
3633 rule_insert(p
, rule
, packet
, in_port
);
3637 static struct rule
*
3638 find_flow_strict(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3643 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3645 return rule_from_cls_rule(classifier_find_rule_exactly(
3646 &p
->cls
, &flow
, wildcards
,
3647 ntohs(ofm
->priority
)));
3651 send_buffered_packet(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3652 struct rule
*rule
, const struct ofp_flow_mod
*ofm
)
3654 struct ofpbuf
*packet
;
3659 if (ofm
->buffer_id
== htonl(UINT32_MAX
)) {
3663 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3669 flow_extract(packet
, 0, in_port
, &flow
);
3670 rule_execute(ofproto
, rule
, packet
, &flow
);
3675 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3677 struct modify_flows_cbdata
{
3678 struct ofproto
*ofproto
;
3679 const struct ofp_flow_mod
*ofm
;
3684 static int modify_flow(struct ofproto
*, const struct ofp_flow_mod
*,
3685 size_t n_actions
, struct rule
*);
3686 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3688 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3689 * encoded by ofp_mkerr() on failure.
3691 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3694 modify_flows_loose(struct ofproto
*p
, struct ofconn
*ofconn
,
3695 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3697 struct modify_flows_cbdata cbdata
;
3698 struct cls_rule target
;
3702 cbdata
.n_actions
= n_actions
;
3703 cbdata
.match
= NULL
;
3705 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3708 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3709 modify_flows_cb
, &cbdata
);
3711 /* This credits the packet to whichever flow happened to happened to
3712 * match last. That's weird. Maybe we should do a lookup for the
3713 * flow that actually matches the packet? Who knows. */
3714 send_buffered_packet(p
, ofconn
, cbdata
.match
, ofm
);
3717 return add_flow(p
, ofconn
, ofm
, n_actions
);
3721 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3722 * code as encoded by ofp_mkerr() on failure.
3724 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3727 modify_flow_strict(struct ofproto
*p
, struct ofconn
*ofconn
,
3728 struct ofp_flow_mod
*ofm
, size_t n_actions
)
3730 struct rule
*rule
= find_flow_strict(p
, ofm
);
3731 if (rule
&& !rule_is_hidden(rule
)) {
3732 modify_flow(p
, ofm
, n_actions
, rule
);
3733 return send_buffered_packet(p
, ofconn
, rule
, ofm
);
3735 return add_flow(p
, ofconn
, ofm
, n_actions
);
3739 /* Callback for modify_flows_loose(). */
3741 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3743 struct rule
*rule
= rule_from_cls_rule(rule_
);
3744 struct modify_flows_cbdata
*cbdata
= cbdata_
;
3746 if (!rule_is_hidden(rule
)) {
3747 cbdata
->match
= rule
;
3748 modify_flow(cbdata
->ofproto
, cbdata
->ofm
, cbdata
->n_actions
, rule
);
3752 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3753 * been identified as a flow in 'p''s flow table to be modified, by changing
3754 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3755 * ofp_action[] structures). */
3757 modify_flow(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
3758 size_t n_actions
, struct rule
*rule
)
3760 size_t actions_len
= n_actions
* sizeof *rule
->actions
;
3762 rule
->flow_cookie
= ofm
->cookie
;
3764 /* If the actions are the same, do nothing. */
3765 if (n_actions
== rule
->n_actions
3766 && !memcmp(ofm
->actions
, rule
->actions
, actions_len
))
3771 /* Replace actions. */
3772 free(rule
->actions
);
3773 rule
->actions
= xmemdup(ofm
->actions
, actions_len
);
3774 rule
->n_actions
= n_actions
;
3776 /* Make sure that the datapath gets updated properly. */
3777 if (rule
->cr
.wc
.wildcards
) {
3778 COVERAGE_INC(ofproto_mod_wc_flow
);
3779 p
->need_revalidate
= true;
3781 rule_update_actions(p
, rule
);
3787 /* OFPFC_DELETE implementation. */
3789 struct delete_flows_cbdata
{
3790 struct ofproto
*ofproto
;
3794 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
3795 static void delete_flow(struct ofproto
*, struct rule
*, uint16_t out_port
);
3797 /* Implements OFPFC_DELETE. */
3799 delete_flows_loose(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3801 struct delete_flows_cbdata cbdata
;
3802 struct cls_rule target
;
3805 cbdata
.out_port
= ofm
->out_port
;
3807 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3810 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3811 delete_flows_cb
, &cbdata
);
3814 /* Implements OFPFC_DELETE_STRICT. */
3816 delete_flow_strict(struct ofproto
*p
, struct ofp_flow_mod
*ofm
)
3818 struct rule
*rule
= find_flow_strict(p
, ofm
);
3820 delete_flow(p
, rule
, ofm
->out_port
);
3824 /* Callback for delete_flows_loose(). */
3826 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3828 struct rule
*rule
= rule_from_cls_rule(rule_
);
3829 struct delete_flows_cbdata
*cbdata
= cbdata_
;
3831 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
3834 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3835 * been identified as a flow to delete from 'p''s flow table, by deleting the
3836 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3839 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3840 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3841 * specified 'out_port'. */
3843 delete_flow(struct ofproto
*p
, struct rule
*rule
, uint16_t out_port
)
3845 if (rule_is_hidden(rule
)) {
3849 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
3853 send_flow_removed(p
, rule
, time_msec(), OFPRR_DELETE
);
3854 rule_remove(p
, rule
);
3858 handle_flow_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
3859 struct ofp_flow_mod
*ofm
)
3861 struct ofp_match orig_match
;
3865 error
= reject_slave_controller(ofconn
, &ofm
->header
);
3869 error
= check_ofp_message_array(&ofm
->header
, OFPT_FLOW_MOD
, sizeof *ofm
,
3870 sizeof *ofm
->actions
, &n_actions
);
3875 /* We do not support the emergency flow cache. It will hopefully
3876 * get dropped from OpenFlow in the near future. */
3877 if (ofm
->flags
& htons(OFPFF_EMERG
)) {
3878 /* There isn't a good fit for an error code, so just state that the
3879 * flow table is full. */
3880 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
3883 /* Normalize ofp->match. If normalization actually changes anything, then
3884 * log the differences. */
3885 ofm
->match
.pad1
[0] = ofm
->match
.pad2
[0] = 0;
3886 orig_match
= ofm
->match
;
3887 normalize_match(&ofm
->match
);
3888 if (memcmp(&ofm
->match
, &orig_match
, sizeof orig_match
)) {
3889 static struct vlog_rate_limit normal_rl
= VLOG_RATE_LIMIT_INIT(1, 1);
3890 if (!VLOG_DROP_INFO(&normal_rl
)) {
3891 char *old
= ofp_match_to_literal_string(&orig_match
);
3892 char *new = ofp_match_to_literal_string(&ofm
->match
);
3893 VLOG_INFO("%s: normalization changed ofp_match, details:",
3894 rconn_get_name(ofconn
->rconn
));
3895 VLOG_INFO(" pre: %s", old
);
3896 VLOG_INFO("post: %s", new);
3902 if (!ofm
->match
.wildcards
) {
3903 ofm
->priority
= htons(UINT16_MAX
);
3906 error
= validate_actions((const union ofp_action
*) ofm
->actions
,
3907 n_actions
, p
->max_ports
);
3912 switch (ntohs(ofm
->command
)) {
3914 return add_flow(p
, ofconn
, ofm
, n_actions
);
3917 return modify_flows_loose(p
, ofconn
, ofm
, n_actions
);
3919 case OFPFC_MODIFY_STRICT
:
3920 return modify_flow_strict(p
, ofconn
, ofm
, n_actions
);
3923 delete_flows_loose(p
, ofm
);
3926 case OFPFC_DELETE_STRICT
:
3927 delete_flow_strict(p
, ofm
);
3931 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
3936 handle_tun_id_from_cookie(struct ofproto
*p
, struct nxt_tun_id_cookie
*msg
)
3940 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
3945 p
->tun_id_from_cookie
= !!msg
->set
;
3950 handle_role_request(struct ofproto
*ofproto
,
3951 struct ofconn
*ofconn
, struct nicira_header
*msg
)
3953 struct nx_role_request
*nrr
;
3954 struct nx_role_request
*reply
;
3958 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
3959 VLOG_WARN_RL(&rl
, "received role request of length %u (expected %zu)",
3960 ntohs(msg
->header
.length
), sizeof *nrr
);
3961 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3963 nrr
= (struct nx_role_request
*) msg
;
3965 if (ofconn
->type
!= OFCONN_PRIMARY
) {
3966 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
3968 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3971 role
= ntohl(nrr
->role
);
3972 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3973 && role
!= NX_ROLE_SLAVE
) {
3974 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
3976 /* There's no good error code for this. */
3977 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
3980 if (role
== NX_ROLE_MASTER
) {
3981 struct ofconn
*other
;
3983 HMAP_FOR_EACH (other
, hmap_node
, &ofproto
->controllers
) {
3984 if (other
->role
== NX_ROLE_MASTER
) {
3985 other
->role
= NX_ROLE_SLAVE
;
3989 ofconn
->role
= role
;
3991 reply
= make_openflow_xid(sizeof *reply
, OFPT_VENDOR
, msg
->header
.xid
,
3993 reply
->nxh
.vendor
= htonl(NX_VENDOR_ID
);
3994 reply
->nxh
.subtype
= htonl(NXT_ROLE_REPLY
);
3995 reply
->role
= htonl(role
);
3996 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4002 handle_vendor(struct ofproto
*p
, struct ofconn
*ofconn
, void *msg
)
4004 struct ofp_vendor_header
*ovh
= msg
;
4005 struct nicira_header
*nh
;
4007 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
4008 VLOG_WARN_RL(&rl
, "received vendor message of length %u "
4009 "(expected at least %zu)",
4010 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
4011 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4013 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
4014 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
4016 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
4017 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %u "
4018 "(expected at least %zu)",
4019 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
4020 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
4024 switch (ntohl(nh
->subtype
)) {
4025 case NXT_STATUS_REQUEST
:
4026 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
4029 case NXT_TUN_ID_FROM_COOKIE
:
4030 return handle_tun_id_from_cookie(p
, msg
);
4032 case NXT_ROLE_REQUEST
:
4033 return handle_role_request(p
, ofconn
, msg
);
4036 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
4040 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
4042 struct ofp_header
*ob
;
4045 /* Currently, everything executes synchronously, so we can just
4046 * immediately send the barrier reply. */
4047 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
4048 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
4053 handle_openflow(struct ofconn
*ofconn
, struct ofproto
*p
,
4054 struct ofpbuf
*ofp_msg
)
4056 struct ofp_header
*oh
= ofp_msg
->data
;
4059 COVERAGE_INC(ofproto_recv_openflow
);
4061 case OFPT_ECHO_REQUEST
:
4062 error
= handle_echo_request(ofconn
, oh
);
4065 case OFPT_ECHO_REPLY
:
4069 case OFPT_FEATURES_REQUEST
:
4070 error
= handle_features_request(p
, ofconn
, oh
);
4073 case OFPT_GET_CONFIG_REQUEST
:
4074 error
= handle_get_config_request(p
, ofconn
, oh
);
4077 case OFPT_SET_CONFIG
:
4078 error
= handle_set_config(p
, ofconn
, ofp_msg
->data
);
4081 case OFPT_PACKET_OUT
:
4082 error
= handle_packet_out(p
, ofconn
, ofp_msg
->data
);
4086 error
= handle_port_mod(p
, ofconn
, oh
);
4090 error
= handle_flow_mod(p
, ofconn
, ofp_msg
->data
);
4093 case OFPT_STATS_REQUEST
:
4094 error
= handle_stats_request(p
, ofconn
, oh
);
4098 error
= handle_vendor(p
, ofconn
, ofp_msg
->data
);
4101 case OFPT_BARRIER_REQUEST
:
4102 error
= handle_barrier_request(ofconn
, oh
);
4106 if (VLOG_IS_WARN_ENABLED()) {
4107 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
4108 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
4111 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
4116 send_error_oh(ofconn
, ofp_msg
->data
, error
);
4121 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4123 struct odp_msg
*msg
= packet
->data
;
4125 struct ofpbuf payload
;
4128 payload
.data
= msg
+ 1;
4129 payload
.size
= msg
->length
- sizeof *msg
;
4130 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
4132 /* Check with in-band control to see if this packet should be sent
4133 * to the local port regardless of the flow table. */
4134 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
4135 union odp_action action
;
4137 memset(&action
, 0, sizeof(action
));
4138 action
.output
.type
= ODPAT_OUTPUT
;
4139 action
.output
.port
= ODPP_LOCAL
;
4140 dpif_execute(p
->dpif
, flow
.in_port
, &action
, 1, &payload
);
4143 rule
= lookup_valid_rule(p
, &flow
);
4145 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
4146 struct ofport
*port
= get_port(p
, msg
->port
);
4148 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
4149 COVERAGE_INC(ofproto_no_packet_in
);
4150 /* XXX install 'drop' flow entry */
4151 ofpbuf_delete(packet
);
4155 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
4158 COVERAGE_INC(ofproto_packet_in
);
4159 send_packet_in(p
, packet
);
4163 if (rule
->cr
.wc
.wildcards
) {
4164 rule
= rule_create_subrule(p
, rule
, &flow
);
4165 rule_make_actions(p
, rule
, packet
);
4167 if (!rule
->may_install
) {
4168 /* The rule is not installable, that is, we need to process every
4169 * packet, so process the current packet and set its actions into
4171 rule_make_actions(p
, rule
, packet
);
4173 /* XXX revalidate rule if it needs it */
4177 if (rule
->super
&& rule
->super
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
4179 * Extra-special case for fail-open mode.
4181 * We are in fail-open mode and the packet matched the fail-open rule,
4182 * but we are connected to a controller too. We should send the packet
4183 * up to the controller in the hope that it will try to set up a flow
4184 * and thereby allow us to exit fail-open.
4186 * See the top-level comment in fail-open.c for more information.
4188 send_packet_in(p
, ofpbuf_clone_with_headroom(packet
,
4189 DPIF_RECV_MSG_PADDING
));
4192 ofpbuf_pull(packet
, sizeof *msg
);
4193 rule_execute(p
, rule
, packet
, &flow
);
4194 rule_reinstall(p
, rule
);
4198 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
4200 struct odp_msg
*msg
= packet
->data
;
4202 switch (msg
->type
) {
4203 case _ODPL_ACTION_NR
:
4204 COVERAGE_INC(ofproto_ctlr_action
);
4205 send_packet_in(p
, packet
);
4208 case _ODPL_SFLOW_NR
:
4210 ofproto_sflow_received(p
->sflow
, msg
);
4212 ofpbuf_delete(packet
);
4216 handle_odp_miss_msg(p
, packet
);
4220 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
4227 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
4229 struct rule
*sub
= rule_from_cls_rule(sub_
);
4230 struct revalidate_cbdata
*cbdata
= cbdata_
;
4232 if (cbdata
->revalidate_all
4233 || (cbdata
->revalidate_subrules
&& sub
->super
)
4234 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
4235 revalidate_rule(cbdata
->ofproto
, sub
);
4240 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
4242 const flow_t
*flow
= &rule
->cr
.flow
;
4244 COVERAGE_INC(ofproto_revalidate_rule
);
4247 super
= rule_from_cls_rule(classifier_lookup_wild(&p
->cls
, flow
));
4249 rule_remove(p
, rule
);
4251 } else if (super
!= rule
->super
) {
4252 COVERAGE_INC(ofproto_revalidate_moved
);
4253 list_remove(&rule
->list
);
4254 list_push_back(&super
->list
, &rule
->list
);
4255 rule
->super
= super
;
4256 rule
->hard_timeout
= super
->hard_timeout
;
4257 rule
->idle_timeout
= super
->idle_timeout
;
4258 rule
->created
= super
->created
;
4263 rule_update_actions(p
, rule
);
4267 static struct ofpbuf
*
4268 compose_flow_removed(struct ofproto
*p
, const struct rule
*rule
,
4269 long long int now
, uint8_t reason
)
4271 struct ofp_flow_removed
*ofr
;
4273 long long int tdiff
= now
- rule
->created
;
4274 uint32_t sec
= tdiff
/ 1000;
4275 uint32_t msec
= tdiff
- (sec
* 1000);
4277 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
4278 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, p
->tun_id_from_cookie
,
4280 ofr
->cookie
= rule
->flow_cookie
;
4281 ofr
->priority
= htons(rule
->cr
.priority
);
4282 ofr
->reason
= reason
;
4283 ofr
->duration_sec
= htonl(sec
);
4284 ofr
->duration_nsec
= htonl(msec
* 1000000);
4285 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
4286 ofr
->packet_count
= htonll(rule
->packet_count
);
4287 ofr
->byte_count
= htonll(rule
->byte_count
);
4293 uninstall_idle_flow(struct ofproto
*ofproto
, struct rule
*rule
)
4295 assert(rule
->installed
);
4296 assert(!rule
->cr
.wc
.wildcards
);
4299 rule_remove(ofproto
, rule
);
4301 rule_uninstall(ofproto
, rule
);
4306 send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
4307 long long int now
, uint8_t reason
)
4309 struct ofconn
*ofconn
;
4310 struct ofconn
*prev
;
4311 struct ofpbuf
*buf
= NULL
;
4313 /* We limit the maximum number of queued flow expirations it by accounting
4314 * them under the counter for replies. That works because preventing
4315 * OpenFlow requests from being processed also prevents new flows from
4316 * being added (and expiring). (It also prevents processing OpenFlow
4317 * requests that would not add new flows, so it is imperfect.) */
4320 LIST_FOR_EACH (ofconn
, node
, &p
->all_conns
) {
4321 if (rule
->send_flow_removed
&& rconn_is_connected(ofconn
->rconn
)
4322 && ofconn_receives_async_msgs(ofconn
)) {
4324 queue_tx(ofpbuf_clone(buf
), prev
, prev
->reply_counter
);
4326 buf
= compose_flow_removed(p
, rule
, now
, reason
);
4332 queue_tx(buf
, prev
, prev
->reply_counter
);
4338 expire_rule(struct cls_rule
*cls_rule
, void *p_
)
4340 struct ofproto
*p
= p_
;
4341 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
4342 long long int hard_expire
, idle_expire
, expire
, now
;
4344 hard_expire
= (rule
->hard_timeout
4345 ? rule
->created
+ rule
->hard_timeout
* 1000
4347 idle_expire
= (rule
->idle_timeout
4348 && (rule
->super
|| list_is_empty(&rule
->list
))
4349 ? rule
->used
+ rule
->idle_timeout
* 1000
4351 expire
= MIN(hard_expire
, idle_expire
);
4355 if (rule
->installed
&& now
>= rule
->used
+ 5000) {
4356 uninstall_idle_flow(p
, rule
);
4357 } else if (!rule
->cr
.wc
.wildcards
) {
4358 active_timeout(p
, rule
);
4364 COVERAGE_INC(ofproto_expired
);
4366 /* Update stats. This code will be a no-op if the rule expired
4367 * due to an idle timeout. */
4368 if (rule
->cr
.wc
.wildcards
) {
4369 struct rule
*subrule
, *next
;
4370 LIST_FOR_EACH_SAFE (subrule
, next
, list
, &rule
->list
) {
4371 rule_remove(p
, subrule
);
4374 rule_uninstall(p
, rule
);
4377 if (!rule_is_hidden(rule
)) {
4378 send_flow_removed(p
, rule
, now
,
4380 ? OFPRR_HARD_TIMEOUT
: OFPRR_IDLE_TIMEOUT
));
4382 rule_remove(p
, rule
);
4386 active_timeout(struct ofproto
*ofproto
, struct rule
*rule
)
4388 if (ofproto
->netflow
&& !is_controller_rule(rule
) &&
4389 netflow_active_timeout_expired(ofproto
->netflow
, &rule
->nf_flow
)) {
4390 struct ofexpired expired
;
4391 struct odp_flow odp_flow
;
4393 /* Get updated flow stats. */
4394 memset(&odp_flow
, 0, sizeof odp_flow
);
4395 if (rule
->installed
) {
4396 odp_flow
.key
= rule
->cr
.flow
;
4397 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
4398 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
4400 if (odp_flow
.stats
.n_packets
) {
4401 update_time(ofproto
, rule
, &odp_flow
.stats
);
4402 netflow_flow_update_flags(&rule
->nf_flow
,
4403 odp_flow
.stats
.tcp_flags
);
4407 expired
.flow
= rule
->cr
.flow
;
4408 expired
.packet_count
= rule
->packet_count
+
4409 odp_flow
.stats
.n_packets
;
4410 expired
.byte_count
= rule
->byte_count
+ odp_flow
.stats
.n_bytes
;
4411 expired
.used
= rule
->used
;
4413 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
4415 /* Schedule us to send the accumulated records once we have
4416 * collected all of them. */
4417 poll_immediate_wake();
4422 update_used(struct ofproto
*p
)
4424 struct odp_flow
*flows
;
4429 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
4434 for (i
= 0; i
< n_flows
; i
++) {
4435 struct odp_flow
*f
= &flows
[i
];
4438 rule
= rule_from_cls_rule(
4439 classifier_find_rule_exactly(&p
->cls
, &f
->key
, 0, UINT16_MAX
));
4440 if (!rule
|| !rule
->installed
) {
4441 COVERAGE_INC(ofproto_unexpected_rule
);
4442 dpif_flow_del(p
->dpif
, f
);
4446 update_time(p
, rule
, &f
->stats
);
4447 rule_account(p
, rule
, f
->stats
.n_bytes
);
4452 /* pinsched callback for sending 'packet' on 'ofconn'. */
4454 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
4456 struct ofconn
*ofconn
= ofconn_
;
4458 rconn_send_with_limit(ofconn
->rconn
, packet
,
4459 ofconn
->packet_in_counter
, 100);
4462 /* Takes 'packet', which has been converted with do_convert_to_packet_in(), and
4463 * finalizes its content for sending on 'ofconn', and passes it to 'ofconn''s
4464 * packet scheduler for sending.
4466 * 'max_len' specifies the maximum number of bytes of the packet to send on
4467 * 'ofconn' (INT_MAX specifies no limit).
4469 * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise,
4470 * ownership is transferred to this function. */
4472 schedule_packet_in(struct ofconn
*ofconn
, struct ofpbuf
*packet
, int max_len
,
4475 struct ofproto
*ofproto
= ofconn
->ofproto
;
4476 struct ofp_packet_in
*opi
= packet
->data
;
4477 uint16_t in_port
= ofp_port_to_odp_port(ntohs(opi
->in_port
));
4478 int send_len
, trim_size
;
4482 if (opi
->reason
== OFPR_ACTION
) {
4483 buffer_id
= UINT32_MAX
;
4484 } else if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
4485 buffer_id
= pktbuf_get_null();
4486 } else if (!ofconn
->pktbuf
) {
4487 buffer_id
= UINT32_MAX
;
4489 struct ofpbuf payload
;
4490 payload
.data
= opi
->data
;
4491 payload
.size
= packet
->size
- offsetof(struct ofp_packet_in
, data
);
4492 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, in_port
);
4495 /* Figure out how much of the packet to send. */
4496 send_len
= ntohs(opi
->total_len
);
4497 if (buffer_id
!= UINT32_MAX
) {
4498 send_len
= MIN(send_len
, ofconn
->miss_send_len
);
4500 send_len
= MIN(send_len
, max_len
);
4502 /* Adjust packet length and clone if necessary. */
4503 trim_size
= offsetof(struct ofp_packet_in
, data
) + send_len
;
4505 packet
= ofpbuf_clone_data(packet
->data
, trim_size
);
4508 packet
->size
= trim_size
;
4511 /* Update packet headers. */
4512 opi
->buffer_id
= htonl(buffer_id
);
4513 update_openflow_length(packet
);
4515 /* Hand over to packet scheduler. It might immediately call into
4516 * do_send_packet_in() or it might buffer it for a while (until a later
4517 * call to pinsched_run()). */
4518 pinsched_send(ofconn
->schedulers
[opi
->reason
], in_port
,
4519 packet
, do_send_packet_in
, ofconn
);
4522 /* Replace struct odp_msg header in 'packet' by equivalent struct
4523 * ofp_packet_in. The odp_msg must have sufficient headroom to do so (e.g. as
4524 * returned by dpif_recv()).
4526 * The conversion is not complete: the caller still needs to trim any unneeded
4527 * payload off the end of the buffer, set the length in the OpenFlow header,
4528 * and set buffer_id. Those require us to know the controller settings and so
4529 * must be done on a per-controller basis.
4531 * Returns the maximum number of bytes of the packet that should be sent to
4532 * the controller (INT_MAX if no limit). */
4534 do_convert_to_packet_in(struct ofpbuf
*packet
)
4536 struct odp_msg
*msg
= packet
->data
;
4537 struct ofp_packet_in
*opi
;
4543 /* Extract relevant header fields */
4544 if (msg
->type
== _ODPL_ACTION_NR
) {
4545 reason
= OFPR_ACTION
;
4548 reason
= OFPR_NO_MATCH
;
4551 total_len
= msg
->length
- sizeof *msg
;
4552 in_port
= odp_port_to_ofp_port(msg
->port
);
4554 /* Repurpose packet buffer by overwriting header. */
4555 ofpbuf_pull(packet
, sizeof(struct odp_msg
));
4556 opi
= ofpbuf_push_zeros(packet
, offsetof(struct ofp_packet_in
, data
));
4557 opi
->header
.version
= OFP_VERSION
;
4558 opi
->header
.type
= OFPT_PACKET_IN
;
4559 opi
->total_len
= htons(total_len
);
4560 opi
->in_port
= htons(in_port
);
4561 opi
->reason
= reason
;
4566 /* Given 'packet' containing an odp_msg of type _ODPL_ACTION_NR or
4567 * _ODPL_MISS_NR, sends an OFPT_PACKET_IN message to each OpenFlow controller
4568 * as necessary according to their individual configurations.
4570 * 'packet' must have sufficient headroom to convert it into a struct
4571 * ofp_packet_in (e.g. as returned by dpif_recv()).
4573 * Takes ownership of 'packet'. */
4575 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
4577 struct ofconn
*ofconn
, *prev
;
4580 max_len
= do_convert_to_packet_in(packet
);
4583 LIST_FOR_EACH (ofconn
, node
, &ofproto
->all_conns
) {
4584 if (ofconn_receives_async_msgs(ofconn
)) {
4586 schedule_packet_in(prev
, packet
, max_len
, true);
4592 schedule_packet_in(prev
, packet
, max_len
, false);
4594 ofpbuf_delete(packet
);
4599 pick_datapath_id(const struct ofproto
*ofproto
)
4601 const struct ofport
*port
;
4603 port
= get_port(ofproto
, ODPP_LOCAL
);
4605 uint8_t ea
[ETH_ADDR_LEN
];
4608 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4610 return eth_addr_to_uint64(ea
);
4612 VLOG_WARN("could not get MAC address for %s (%s)",
4613 netdev_get_name(port
->netdev
), strerror(error
));
4615 return ofproto
->fallback_dpid
;
4619 pick_fallback_dpid(void)
4621 uint8_t ea
[ETH_ADDR_LEN
];
4622 eth_addr_nicira_random(ea
);
4623 return eth_addr_to_uint64(ea
);
4627 default_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
4628 struct odp_actions
*actions
, tag_type
*tags
,
4629 uint16_t *nf_output_iface
, void *ofproto_
)
4631 struct ofproto
*ofproto
= ofproto_
;
4634 /* Drop frames for reserved multicast addresses. */
4635 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4639 /* Learn source MAC (but don't try to learn from revalidation). */
4640 if (packet
!= NULL
) {
4641 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4643 GRAT_ARP_LOCK_NONE
);
4645 /* The log messages here could actually be useful in debugging,
4646 * so keep the rate limit relatively high. */
4647 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4648 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4649 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4650 ofproto_revalidate(ofproto
, rev_tag
);
4654 /* Determine output port. */
4655 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
,
4658 add_output_group_action(actions
, DP_GROUP_FLOOD
, nf_output_iface
);
4659 } else if (out_port
!= flow
->in_port
) {
4660 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4661 *nf_output_iface
= out_port
;
4669 static const struct ofhooks default_ofhooks
= {
4671 default_normal_ofhook_cb
,