2 * Copyright (c) 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
22 #include <netinet/in.h>
25 #include "classifier.h"
27 #include "discovery.h"
29 #include "dynamic-string.h"
30 #include "fail-open.h"
32 #include "mac-learning.h"
36 #include "ofp-print.h"
37 #include "ofproto-sflow.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openvswitch/datapath-protocol.h"
45 #include "poll-loop.h"
46 #include "port-array.h"
51 #include "stream-ssl.h"
59 #define THIS_MODULE VLM_ofproto
62 #include "sflow_api.h"
66 TABLEID_CLASSIFIER
= 1
70 struct netdev
*netdev
;
71 struct ofp_phy_port opp
; /* In host byte order. */
74 static void ofport_free(struct ofport
*);
75 static void hton_ofp_phy_port(struct ofp_phy_port
*);
77 static int xlate_actions(const union ofp_action
*in
, size_t n_in
,
78 const flow_t
*flow
, struct ofproto
*ofproto
,
79 const struct ofpbuf
*packet
,
80 struct odp_actions
*out
, tag_type
*tags
,
81 bool *may_set_up_flow
, uint16_t *nf_output_iface
);
86 uint64_t flow_cookie
; /* Controller-issued identifier.
87 (Kept in network-byte order.) */
88 uint16_t idle_timeout
; /* In seconds from time of last use. */
89 uint16_t hard_timeout
; /* In seconds from time of creation. */
90 bool send_flow_removed
; /* Send a flow removed message? */
91 long long int used
; /* Last-used time (0 if never used). */
92 long long int created
; /* Creation time. */
93 uint64_t packet_count
; /* Number of packets received. */
94 uint64_t byte_count
; /* Number of bytes received. */
95 uint64_t accounted_bytes
; /* Number of bytes passed to account_cb. */
96 tag_type tags
; /* Tags (set only by hooks). */
97 struct netflow_flow nf_flow
; /* Per-flow NetFlow tracking data. */
99 /* If 'super' is non-NULL, this rule is a subrule, that is, it is an
100 * exact-match rule (having cr.wc.wildcards of 0) generated from the
101 * wildcard rule 'super'. In this case, 'list' is an element of the
104 * If 'super' is NULL, this rule is a super-rule, and 'list' is the head of
105 * a list of subrules. A super-rule with no wildcards (where
106 * cr.wc.wildcards is 0) will never have any subrules. */
112 * 'n_actions' is the number of elements in the 'actions' array. A single
113 * action may take up more more than one element's worth of space.
115 * A subrule has no actions (it uses the super-rule's actions). */
117 union ofp_action
*actions
;
121 * A super-rule with wildcard fields never has ODP actions (since the
122 * datapath only supports exact-match flows). */
123 bool installed
; /* Installed in datapath? */
124 bool may_install
; /* True ordinarily; false if actions must
125 * be reassessed for every packet. */
127 union odp_action
*odp_actions
;
131 rule_is_hidden(const struct rule
*rule
)
133 /* Subrules are merely an implementation detail, so hide them from the
135 if (rule
->super
!= NULL
) {
139 /* Rules with priority higher than UINT16_MAX are set up by ofproto itself
140 * (e.g. by in-band control) and are intentionally hidden from the
142 if (rule
->cr
.priority
> UINT16_MAX
) {
149 static struct rule
*rule_create(struct ofproto
*, struct rule
*super
,
150 const union ofp_action
*, size_t n_actions
,
151 uint16_t idle_timeout
, uint16_t hard_timeout
,
152 uint64_t flow_cookie
, bool send_flow_removed
);
153 static void rule_free(struct rule
*);
154 static void rule_destroy(struct ofproto
*, struct rule
*);
155 static struct rule
*rule_from_cls_rule(const struct cls_rule
*);
156 static void rule_insert(struct ofproto
*, struct rule
*,
157 struct ofpbuf
*packet
, uint16_t in_port
);
158 static void rule_remove(struct ofproto
*, struct rule
*);
159 static bool rule_make_actions(struct ofproto
*, struct rule
*,
160 const struct ofpbuf
*packet
);
161 static void rule_install(struct ofproto
*, struct rule
*,
162 struct rule
*displaced_rule
);
163 static void rule_uninstall(struct ofproto
*, struct rule
*);
164 static void rule_post_uninstall(struct ofproto
*, struct rule
*);
165 static void send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
166 long long int now
, uint8_t reason
);
168 /* ofproto supports two kinds of OpenFlow connections:
170 * - "Controller connections": Connections to ordinary OpenFlow controllers.
171 * ofproto maintains persistent connections to these controllers and by
172 * default sends them asynchronous messages such as packet-ins.
174 * - "Transient connections", e.g. from ovs-ofctl. When these connections
175 * drop, it is the other side's responsibility to reconnect them if
176 * necessary. ofproto does not send them asynchronous messages by default.
179 OFCONN_CONTROLLER
, /* An OpenFlow controller. */
180 OFCONN_TRANSIENT
/* A transient connection. */
183 /* An OpenFlow connection. */
185 struct ofproto
*ofproto
; /* The ofproto that owns this connection. */
186 struct list node
; /* In struct ofproto's "all_conns" list. */
187 struct rconn
*rconn
; /* OpenFlow connection. */
188 enum ofconn_type type
; /* Type. */
190 /* OFPT_PACKET_IN related data. */
191 struct rconn_packet_counter
*packet_in_counter
; /* # queued on 'rconn'. */
192 struct pinsched
*schedulers
[2]; /* Indexed by reason code; see below. */
193 struct pktbuf
*pktbuf
; /* OpenFlow packet buffers. */
194 int miss_send_len
; /* Bytes to send of buffered packets. */
196 /* Number of OpenFlow messages queued on 'rconn' as replies to OpenFlow
197 * requests, and the maximum number before we stop reading OpenFlow
199 #define OFCONN_REPLY_MAX 100
200 struct rconn_packet_counter
*reply_counter
;
202 /* type == OFCONN_CONTROLLER only. */
203 enum nx_role role
; /* Role. */
204 struct hmap_node hmap_node
; /* In struct ofproto's "controllers" map. */
205 struct discovery
*discovery
; /* Controller discovery object, if enabled. */
206 struct status_category
*ss
; /* Switch status category. */
207 enum ofproto_band band
; /* In-band or out-of-band? */
210 /* We use OFPR_NO_MATCH and OFPR_ACTION as indexes into struct ofconn's
211 * "schedulers" array. Their values are 0 and 1, and their meanings and values
212 * coincide with _ODPL_MISS_NR and _ODPL_ACTION_NR, so this is convenient. In
213 * case anything ever changes, check their values here. */
214 #define N_SCHEDULERS 2
215 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== 0);
216 BUILD_ASSERT_DECL(OFPR_NO_MATCH
== _ODPL_MISS_NR
);
217 BUILD_ASSERT_DECL(OFPR_ACTION
== 1);
218 BUILD_ASSERT_DECL(OFPR_ACTION
== _ODPL_ACTION_NR
);
220 static struct ofconn
*ofconn_create(struct ofproto
*, struct rconn
*,
222 static void ofconn_destroy(struct ofconn
*);
223 static void ofconn_run(struct ofconn
*, struct ofproto
*);
224 static void ofconn_wait(struct ofconn
*);
225 static void queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
226 struct rconn_packet_counter
*counter
);
228 static void send_packet_in(struct ofproto
*, struct ofpbuf
*odp_msg
);
229 static void do_send_packet_in(struct ofpbuf
*odp_msg
, void *ofconn
);
233 uint64_t datapath_id
; /* Datapath ID. */
234 uint64_t fallback_dpid
; /* Datapath ID if no better choice found. */
235 char *mfr_desc
; /* Manufacturer. */
236 char *hw_desc
; /* Hardware. */
237 char *sw_desc
; /* Software version. */
238 char *serial_desc
; /* Serial number. */
239 char *dp_desc
; /* Datapath description. */
243 struct netdev_monitor
*netdev_monitor
;
244 struct port_array ports
; /* Index is ODP port nr; ofport->opp.port_no is
246 struct shash port_by_name
;
250 struct switch_status
*switch_status
;
251 struct fail_open
*fail_open
;
252 struct netflow
*netflow
;
253 struct ofproto_sflow
*sflow
;
255 /* In-band control. */
256 struct in_band
*in_band
;
257 long long int next_in_band_update
;
259 struct classifier cls
;
260 bool need_revalidate
;
261 long long int next_expiration
;
262 struct tag_set revalidate_set
;
263 bool tun_id_from_cookie
;
265 /* OpenFlow connections. */
266 struct hmap controllers
; /* Controller "struct ofconn"s. */
267 struct list all_conns
; /* Contains "struct ofconn"s. */
268 struct pvconn
**listeners
;
270 struct pvconn
**snoops
;
273 /* Hooks for ovs-vswitchd. */
274 const struct ofhooks
*ofhooks
;
277 /* Used by default ofhooks. */
278 struct mac_learning
*ml
;
281 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
283 static const struct ofhooks default_ofhooks
;
285 static uint64_t pick_datapath_id(const struct ofproto
*);
286 static uint64_t pick_fallback_dpid(void);
288 static void update_used(struct ofproto
*);
289 static void update_stats(struct ofproto
*, struct rule
*,
290 const struct odp_flow_stats
*);
291 static void expire_rule(struct cls_rule
*, void *ofproto
);
292 static void active_timeout(struct ofproto
*ofproto
, struct rule
*rule
);
293 static bool revalidate_rule(struct ofproto
*p
, struct rule
*rule
);
294 static void revalidate_cb(struct cls_rule
*rule_
, void *p_
);
296 static void handle_odp_msg(struct ofproto
*, struct ofpbuf
*);
298 static void handle_openflow(struct ofconn
*, struct ofproto
*,
301 static void refresh_port_groups(struct ofproto
*);
303 static void update_port(struct ofproto
*, const char *devname
);
304 static int init_ports(struct ofproto
*);
305 static void reinit_ports(struct ofproto
*);
308 ofproto_create(const char *datapath
, const char *datapath_type
,
309 const struct ofhooks
*ofhooks
, void *aux
,
310 struct ofproto
**ofprotop
)
312 struct odp_stats stats
;
319 /* Connect to datapath and start listening for messages. */
320 error
= dpif_open(datapath
, datapath_type
, &dpif
);
322 VLOG_ERR("failed to open datapath %s: %s", datapath
, strerror(error
));
325 error
= dpif_get_dp_stats(dpif
, &stats
);
327 VLOG_ERR("failed to obtain stats for datapath %s: %s",
328 datapath
, strerror(error
));
332 error
= dpif_recv_set_mask(dpif
, ODPL_MISS
| ODPL_ACTION
| ODPL_SFLOW
);
334 VLOG_ERR("failed to listen on datapath %s: %s",
335 datapath
, strerror(error
));
339 dpif_flow_flush(dpif
);
340 dpif_recv_purge(dpif
);
342 /* Initialize settings. */
343 p
= xzalloc(sizeof *p
);
344 p
->fallback_dpid
= pick_fallback_dpid();
345 p
->datapath_id
= p
->fallback_dpid
;
346 p
->mfr_desc
= xstrdup(DEFAULT_MFR_DESC
);
347 p
->hw_desc
= xstrdup(DEFAULT_HW_DESC
);
348 p
->sw_desc
= xstrdup(DEFAULT_SW_DESC
);
349 p
->serial_desc
= xstrdup(DEFAULT_SERIAL_DESC
);
350 p
->dp_desc
= xstrdup(DEFAULT_DP_DESC
);
352 /* Initialize datapath. */
354 p
->netdev_monitor
= netdev_monitor_create();
355 port_array_init(&p
->ports
);
356 shash_init(&p
->port_by_name
);
357 p
->max_ports
= stats
.max_ports
;
359 /* Initialize submodules. */
360 p
->switch_status
= switch_status_create(p
);
366 /* Initialize flow table. */
367 classifier_init(&p
->cls
);
368 p
->need_revalidate
= false;
369 p
->next_expiration
= time_msec() + 1000;
370 tag_set_init(&p
->revalidate_set
);
372 /* Initialize OpenFlow connections. */
373 list_init(&p
->all_conns
);
374 hmap_init(&p
->controllers
);
380 /* Initialize hooks. */
382 p
->ofhooks
= ofhooks
;
386 p
->ofhooks
= &default_ofhooks
;
388 p
->ml
= mac_learning_create();
391 /* Pick final datapath ID. */
392 p
->datapath_id
= pick_datapath_id(p
);
393 VLOG_INFO("using datapath ID %016"PRIx64
, p
->datapath_id
);
400 ofproto_set_datapath_id(struct ofproto
*p
, uint64_t datapath_id
)
402 uint64_t old_dpid
= p
->datapath_id
;
403 p
->datapath_id
= datapath_id
? datapath_id
: pick_datapath_id(p
);
404 if (p
->datapath_id
!= old_dpid
) {
405 struct ofconn
*ofconn
;
407 VLOG_INFO("datapath ID changed to %016"PRIx64
, p
->datapath_id
);
409 /* Force all active connections to reconnect, since there is no way to
410 * notify a controller that the datapath ID has changed. */
411 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
412 rconn_reconnect(ofconn
->rconn
);
418 is_discovery_controller(const struct ofproto_controller
*c
)
420 return !strcmp(c
->target
, "discover");
424 is_in_band_controller(const struct ofproto_controller
*c
)
426 return is_discovery_controller(c
) || c
->band
== OFPROTO_IN_BAND
;
429 /* Creates a new controller in 'ofproto'. Some of the settings are initially
430 * drawn from 'c', but update_controller() needs to be called later to finish
431 * the new ofconn's configuration. */
433 add_controller(struct ofproto
*ofproto
, const struct ofproto_controller
*c
)
435 struct discovery
*discovery
;
436 struct ofconn
*ofconn
;
438 if (is_discovery_controller(c
)) {
439 int error
= discovery_create(c
->accept_re
, c
->update_resolv_conf
,
440 ofproto
->dpif
, ofproto
->switch_status
,
449 ofconn
= ofconn_create(ofproto
, rconn_create(5, 8), OFCONN_CONTROLLER
);
450 ofconn
->pktbuf
= pktbuf_create();
451 ofconn
->miss_send_len
= OFP_DEFAULT_MISS_SEND_LEN
;
453 ofconn
->discovery
= discovery
;
455 rconn_connect(ofconn
->rconn
, c
->target
);
457 hmap_insert(&ofproto
->controllers
, &ofconn
->hmap_node
,
458 hash_string(c
->target
, 0));
461 /* Reconfigures 'ofconn' to match 'c'. This function cannot update an ofconn's
462 * target or turn discovery on or off (these are done by creating new ofconns
463 * and deleting old ones), but it can update the rest of an ofconn's
466 update_controller(struct ofconn
*ofconn
, const struct ofproto_controller
*c
)
468 struct ofproto
*ofproto
= ofconn
->ofproto
;
472 ofconn
->band
= (is_in_band_controller(c
)
473 ? OFPROTO_IN_BAND
: OFPROTO_OUT_OF_BAND
);
475 rconn_set_max_backoff(ofconn
->rconn
, c
->max_backoff
);
477 probe_interval
= c
->probe_interval
? MAX(c
->probe_interval
, 5) : 0;
478 rconn_set_probe_interval(ofconn
->rconn
, probe_interval
);
480 if (ofconn
->discovery
) {
481 discovery_set_update_resolv_conf(ofconn
->discovery
,
482 c
->update_resolv_conf
);
483 discovery_set_accept_controller_re(ofconn
->discovery
, c
->accept_re
);
486 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
487 struct pinsched
**s
= &ofconn
->schedulers
[i
];
489 if (c
->rate_limit
> 0) {
491 *s
= pinsched_create(c
->rate_limit
, c
->burst_limit
,
492 ofproto
->switch_status
);
494 pinsched_set_limits(*s
, c
->rate_limit
, c
->burst_limit
);
497 pinsched_destroy(*s
);
504 ofconn_get_target(const struct ofconn
*ofconn
)
506 return ofconn
->discovery
? "discover" : rconn_get_name(ofconn
->rconn
);
509 static struct ofconn
*
510 find_controller_by_target(struct ofproto
*ofproto
, const char *target
)
512 struct ofconn
*ofconn
;
514 HMAP_FOR_EACH_WITH_HASH (ofconn
, struct ofconn
, hmap_node
,
515 hash_string(target
, 0), &ofproto
->controllers
) {
516 if (!strcmp(ofconn_get_target(ofconn
), target
)) {
524 update_in_band_remotes(struct ofproto
*ofproto
)
526 const struct ofconn
*ofconn
;
527 struct sockaddr_in
*addrs
;
532 addrs
= xmalloc(hmap_count(&ofproto
->controllers
) * sizeof *addrs
);
535 /* Add all the remotes. */
537 HMAP_FOR_EACH (ofconn
, struct ofconn
, hmap_node
, &ofproto
->controllers
) {
538 struct sockaddr_in
*sin
= &addrs
[n_addrs
];
540 sin
->sin_addr
.s_addr
= rconn_get_remote_ip(ofconn
->rconn
);
541 if (sin
->sin_addr
.s_addr
) {
542 sin
->sin_port
= rconn_get_remote_port(ofconn
->rconn
);
545 if (ofconn
->discovery
) {
550 /* Create or update or destroy in-band.
552 * Ordinarily we only enable in-band if there's at least one remote
553 * address, but discovery needs the in-band rules for DHCP to be installed
554 * even before we know any remote addresses. */
555 if (n_addrs
|| discovery
) {
556 if (!ofproto
->in_band
) {
557 in_band_create(ofproto
, ofproto
->dpif
, ofproto
->switch_status
,
560 in_band_set_remotes(ofproto
->in_band
, addrs
, n_addrs
);
561 ofproto
->next_in_band_update
= time_msec() + 1000;
563 in_band_destroy(ofproto
->in_band
);
564 ofproto
->in_band
= NULL
;
572 ofproto_set_controllers(struct ofproto
*p
,
573 const struct ofproto_controller
*controllers
,
574 size_t n_controllers
)
576 struct shash new_controllers
;
577 enum ofproto_fail_mode fail_mode
;
578 struct ofconn
*ofconn
, *next
;
582 shash_init(&new_controllers
);
583 for (i
= 0; i
< n_controllers
; i
++) {
584 const struct ofproto_controller
*c
= &controllers
[i
];
586 shash_add_once(&new_controllers
, c
->target
, &controllers
[i
]);
587 if (!find_controller_by_target(p
, c
->target
)) {
588 add_controller(p
, c
);
592 fail_mode
= OFPROTO_FAIL_STANDALONE
;
594 HMAP_FOR_EACH_SAFE (ofconn
, next
, struct ofconn
, hmap_node
,
596 struct ofproto_controller
*c
;
598 c
= shash_find_data(&new_controllers
, ofconn_get_target(ofconn
));
600 ofconn_destroy(ofconn
);
602 update_controller(ofconn
, c
);
606 if (c
->fail
== OFPROTO_FAIL_SECURE
) {
607 fail_mode
= OFPROTO_FAIL_SECURE
;
611 shash_destroy(&new_controllers
);
613 update_in_band_remotes(p
);
615 if (!hmap_is_empty(&p
->controllers
)
616 && fail_mode
== OFPROTO_FAIL_STANDALONE
) {
617 struct rconn
**rconns
;
621 p
->fail_open
= fail_open_create(p
, p
->switch_status
);
625 rconns
= xmalloc(hmap_count(&p
->controllers
) * sizeof *rconns
);
626 HMAP_FOR_EACH (ofconn
, struct ofconn
, hmap_node
, &p
->controllers
) {
627 rconns
[n
++] = ofconn
->rconn
;
630 fail_open_set_controllers(p
->fail_open
, rconns
, n
);
631 /* p->fail_open takes ownership of 'rconns'. */
633 fail_open_destroy(p
->fail_open
);
637 if (!hmap_is_empty(&p
->controllers
) && !ss_exists
) {
638 ofconn
= CONTAINER_OF(hmap_first(&p
->controllers
),
639 struct ofconn
, hmap_node
);
640 ofconn
->ss
= switch_status_register(p
->switch_status
, "remote",
641 rconn_status_cb
, ofconn
->rconn
);
646 ofproto_set_desc(struct ofproto
*p
,
647 const char *mfr_desc
, const char *hw_desc
,
648 const char *sw_desc
, const char *serial_desc
,
651 struct ofp_desc_stats
*ods
;
654 if (strlen(mfr_desc
) >= sizeof ods
->mfr_desc
) {
655 VLOG_WARN("truncating mfr_desc, must be less than %zu characters",
656 sizeof ods
->mfr_desc
);
659 p
->mfr_desc
= xstrdup(mfr_desc
);
662 if (strlen(hw_desc
) >= sizeof ods
->hw_desc
) {
663 VLOG_WARN("truncating hw_desc, must be less than %zu characters",
664 sizeof ods
->hw_desc
);
667 p
->hw_desc
= xstrdup(hw_desc
);
670 if (strlen(sw_desc
) >= sizeof ods
->sw_desc
) {
671 VLOG_WARN("truncating sw_desc, must be less than %zu characters",
672 sizeof ods
->sw_desc
);
675 p
->sw_desc
= xstrdup(sw_desc
);
678 if (strlen(serial_desc
) >= sizeof ods
->serial_num
) {
679 VLOG_WARN("truncating serial_desc, must be less than %zu "
681 sizeof ods
->serial_num
);
683 free(p
->serial_desc
);
684 p
->serial_desc
= xstrdup(serial_desc
);
687 if (strlen(dp_desc
) >= sizeof ods
->dp_desc
) {
688 VLOG_WARN("truncating dp_desc, must be less than %zu characters",
689 sizeof ods
->dp_desc
);
692 p
->dp_desc
= xstrdup(dp_desc
);
697 set_pvconns(struct pvconn
***pvconnsp
, size_t *n_pvconnsp
,
698 const struct svec
*svec
)
700 struct pvconn
**pvconns
= *pvconnsp
;
701 size_t n_pvconns
= *n_pvconnsp
;
705 for (i
= 0; i
< n_pvconns
; i
++) {
706 pvconn_close(pvconns
[i
]);
710 pvconns
= xmalloc(svec
->n
* sizeof *pvconns
);
712 for (i
= 0; i
< svec
->n
; i
++) {
713 const char *name
= svec
->names
[i
];
714 struct pvconn
*pvconn
;
717 error
= pvconn_open(name
, &pvconn
);
719 pvconns
[n_pvconns
++] = pvconn
;
721 VLOG_ERR("failed to listen on %s: %s", name
, strerror(error
));
729 *n_pvconnsp
= n_pvconns
;
735 ofproto_set_listeners(struct ofproto
*ofproto
, const struct svec
*listeners
)
737 return set_pvconns(&ofproto
->listeners
, &ofproto
->n_listeners
, listeners
);
741 ofproto_set_snoops(struct ofproto
*ofproto
, const struct svec
*snoops
)
743 return set_pvconns(&ofproto
->snoops
, &ofproto
->n_snoops
, snoops
);
747 ofproto_set_netflow(struct ofproto
*ofproto
,
748 const struct netflow_options
*nf_options
)
750 if (nf_options
&& nf_options
->collectors
.n
) {
751 if (!ofproto
->netflow
) {
752 ofproto
->netflow
= netflow_create();
754 return netflow_set_options(ofproto
->netflow
, nf_options
);
756 netflow_destroy(ofproto
->netflow
);
757 ofproto
->netflow
= NULL
;
763 ofproto_set_sflow(struct ofproto
*ofproto
,
764 const struct ofproto_sflow_options
*oso
)
766 struct ofproto_sflow
*os
= ofproto
->sflow
;
769 struct ofport
*ofport
;
770 unsigned int odp_port
;
772 os
= ofproto
->sflow
= ofproto_sflow_create(ofproto
->dpif
);
773 refresh_port_groups(ofproto
);
774 PORT_ARRAY_FOR_EACH (ofport
, &ofproto
->ports
, odp_port
) {
775 ofproto_sflow_add_port(os
, odp_port
,
776 netdev_get_name(ofport
->netdev
));
779 ofproto_sflow_set_options(os
, oso
);
781 ofproto_sflow_destroy(os
);
782 ofproto
->sflow
= NULL
;
787 ofproto_set_stp(struct ofproto
*ofproto OVS_UNUSED
, bool enable_stp
)
791 VLOG_WARN("STP is not yet implemented");
799 ofproto_get_datapath_id(const struct ofproto
*ofproto
)
801 return ofproto
->datapath_id
;
805 ofproto_has_controller(const struct ofproto
*ofproto
)
807 return !hmap_is_empty(&ofproto
->controllers
);
811 ofproto_get_listeners(const struct ofproto
*ofproto
, struct svec
*listeners
)
815 for (i
= 0; i
< ofproto
->n_listeners
; i
++) {
816 svec_add(listeners
, pvconn_get_name(ofproto
->listeners
[i
]));
821 ofproto_get_snoops(const struct ofproto
*ofproto
, struct svec
*snoops
)
825 for (i
= 0; i
< ofproto
->n_snoops
; i
++) {
826 svec_add(snoops
, pvconn_get_name(ofproto
->snoops
[i
]));
831 ofproto_destroy(struct ofproto
*p
)
833 struct ofconn
*ofconn
, *next_ofconn
;
834 struct ofport
*ofport
;
835 unsigned int port_no
;
842 /* Destroy fail-open and in-band early, since they touch the classifier. */
843 fail_open_destroy(p
->fail_open
);
846 in_band_destroy(p
->in_band
);
849 ofproto_flush_flows(p
);
850 classifier_destroy(&p
->cls
);
852 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
854 ofconn_destroy(ofconn
);
856 hmap_destroy(&p
->controllers
);
859 netdev_monitor_destroy(p
->netdev_monitor
);
860 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
863 shash_destroy(&p
->port_by_name
);
865 switch_status_destroy(p
->switch_status
);
866 netflow_destroy(p
->netflow
);
867 ofproto_sflow_destroy(p
->sflow
);
869 for (i
= 0; i
< p
->n_listeners
; i
++) {
870 pvconn_close(p
->listeners
[i
]);
874 for (i
= 0; i
< p
->n_snoops
; i
++) {
875 pvconn_close(p
->snoops
[i
]);
879 mac_learning_destroy(p
->ml
);
884 free(p
->serial_desc
);
887 port_array_destroy(&p
->ports
);
893 ofproto_run(struct ofproto
*p
)
895 int error
= ofproto_run1(p
);
897 error
= ofproto_run2(p
, false);
903 process_port_change(struct ofproto
*ofproto
, int error
, char *devname
)
905 if (error
== ENOBUFS
) {
906 reinit_ports(ofproto
);
908 update_port(ofproto
, devname
);
913 /* One of ofproto's "snoop" pvconns has accepted a new connection on 'vconn'.
914 * Connects this vconn to a controller. */
916 add_snooper(struct ofproto
*ofproto
, struct vconn
*vconn
)
918 struct ofconn
*ofconn
;
920 /* Arbitrarily pick the first controller in the list for monitoring. We
921 * could do something smarter or more flexible later, if it ever proves
923 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &ofproto
->all_conns
) {
924 if (ofconn
->type
== OFCONN_CONTROLLER
) {
925 rconn_add_monitor(ofconn
->rconn
, vconn
);
930 VLOG_INFO_RL(&rl
, "no controller connection to monitor");
935 ofproto_run1(struct ofproto
*p
)
937 struct ofconn
*ofconn
, *next_ofconn
;
942 if (shash_is_empty(&p
->port_by_name
)) {
946 for (i
= 0; i
< 50; i
++) {
950 error
= dpif_recv(p
->dpif
, &buf
);
952 if (error
== ENODEV
) {
953 /* Someone destroyed the datapath behind our back. The caller
954 * better destroy us and give up, because we're just going to
955 * spin from here on out. */
956 static struct vlog_rate_limit rl2
= VLOG_RATE_LIMIT_INIT(1, 5);
957 VLOG_ERR_RL(&rl2
, "%s: datapath was destroyed externally",
964 handle_odp_msg(p
, buf
);
967 while ((error
= dpif_port_poll(p
->dpif
, &devname
)) != EAGAIN
) {
968 process_port_change(p
, error
, devname
);
970 while ((error
= netdev_monitor_poll(p
->netdev_monitor
,
971 &devname
)) != EAGAIN
) {
972 process_port_change(p
, error
, devname
);
976 if (time_msec() >= p
->next_in_band_update
) {
977 update_in_band_remotes(p
);
979 in_band_run(p
->in_band
);
982 LIST_FOR_EACH_SAFE (ofconn
, next_ofconn
, struct ofconn
, node
,
984 ofconn_run(ofconn
, p
);
987 /* Fail-open maintenance. Do this after processing the ofconns since
988 * fail-open checks the status of the controller rconn. */
990 fail_open_run(p
->fail_open
);
993 for (i
= 0; i
< p
->n_listeners
; i
++) {
997 retval
= pvconn_accept(p
->listeners
[i
], OFP_VERSION
, &vconn
);
999 ofconn_create(p
, rconn_new_from_vconn("passive", vconn
),
1001 } else if (retval
!= EAGAIN
) {
1002 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1006 for (i
= 0; i
< p
->n_snoops
; i
++) {
1007 struct vconn
*vconn
;
1010 retval
= pvconn_accept(p
->snoops
[i
], OFP_VERSION
, &vconn
);
1012 add_snooper(p
, vconn
);
1013 } else if (retval
!= EAGAIN
) {
1014 VLOG_WARN_RL(&rl
, "accept failed (%s)", strerror(retval
));
1018 if (time_msec() >= p
->next_expiration
) {
1019 COVERAGE_INC(ofproto_expiration
);
1020 p
->next_expiration
= time_msec() + 1000;
1023 classifier_for_each(&p
->cls
, CLS_INC_ALL
, expire_rule
, p
);
1025 /* Let the hook know that we're at a stable point: all outstanding data
1026 * in existing flows has been accounted to the account_cb. Thus, the
1027 * hook can now reasonably do operations that depend on having accurate
1028 * flow volume accounting (currently, that's just bond rebalancing). */
1029 if (p
->ofhooks
->account_checkpoint_cb
) {
1030 p
->ofhooks
->account_checkpoint_cb(p
->aux
);
1035 netflow_run(p
->netflow
);
1038 ofproto_sflow_run(p
->sflow
);
1044 struct revalidate_cbdata
{
1045 struct ofproto
*ofproto
;
1046 bool revalidate_all
; /* Revalidate all exact-match rules? */
1047 bool revalidate_subrules
; /* Revalidate all exact-match subrules? */
1048 struct tag_set revalidate_set
; /* Set of tags to revalidate. */
1052 ofproto_run2(struct ofproto
*p
, bool revalidate_all
)
1054 if (p
->need_revalidate
|| revalidate_all
1055 || !tag_set_is_empty(&p
->revalidate_set
)) {
1056 struct revalidate_cbdata cbdata
;
1058 cbdata
.revalidate_all
= revalidate_all
;
1059 cbdata
.revalidate_subrules
= p
->need_revalidate
;
1060 cbdata
.revalidate_set
= p
->revalidate_set
;
1061 tag_set_init(&p
->revalidate_set
);
1062 COVERAGE_INC(ofproto_revalidate
);
1063 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, revalidate_cb
, &cbdata
);
1064 p
->need_revalidate
= false;
1071 ofproto_wait(struct ofproto
*p
)
1073 struct ofconn
*ofconn
;
1076 dpif_recv_wait(p
->dpif
);
1077 dpif_port_poll_wait(p
->dpif
);
1078 netdev_monitor_poll_wait(p
->netdev_monitor
);
1079 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
1080 ofconn_wait(ofconn
);
1083 poll_timer_wait(p
->next_in_band_update
- time_msec());
1084 in_band_wait(p
->in_band
);
1087 fail_open_wait(p
->fail_open
);
1090 ofproto_sflow_wait(p
->sflow
);
1092 if (!tag_set_is_empty(&p
->revalidate_set
)) {
1093 poll_immediate_wake();
1095 if (p
->need_revalidate
) {
1096 /* Shouldn't happen, but if it does just go around again. */
1097 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1098 poll_immediate_wake();
1099 } else if (p
->next_expiration
!= LLONG_MAX
) {
1100 poll_timer_wait(p
->next_expiration
- time_msec());
1102 for (i
= 0; i
< p
->n_listeners
; i
++) {
1103 pvconn_wait(p
->listeners
[i
]);
1105 for (i
= 0; i
< p
->n_snoops
; i
++) {
1106 pvconn_wait(p
->snoops
[i
]);
1111 ofproto_revalidate(struct ofproto
*ofproto
, tag_type tag
)
1113 tag_set_add(&ofproto
->revalidate_set
, tag
);
1117 ofproto_get_revalidate_set(struct ofproto
*ofproto
)
1119 return &ofproto
->revalidate_set
;
1123 ofproto_is_alive(const struct ofproto
*p
)
1125 return !hmap_is_empty(&p
->controllers
);
1129 ofproto_send_packet(struct ofproto
*p
, const flow_t
*flow
,
1130 const union ofp_action
*actions
, size_t n_actions
,
1131 const struct ofpbuf
*packet
)
1133 struct odp_actions odp_actions
;
1136 error
= xlate_actions(actions
, n_actions
, flow
, p
, packet
, &odp_actions
,
1142 /* XXX Should we translate the dpif_execute() errno value into an OpenFlow
1144 dpif_execute(p
->dpif
, flow
->in_port
, odp_actions
.actions
,
1145 odp_actions
.n_actions
, packet
);
1150 ofproto_add_flow(struct ofproto
*p
,
1151 const flow_t
*flow
, uint32_t wildcards
, unsigned int priority
,
1152 const union ofp_action
*actions
, size_t n_actions
,
1156 rule
= rule_create(p
, NULL
, actions
, n_actions
,
1157 idle_timeout
>= 0 ? idle_timeout
: 5 /* XXX */,
1159 cls_rule_from_flow(flow
, wildcards
, priority
, &rule
->cr
);
1160 rule_insert(p
, rule
, NULL
, 0);
1164 ofproto_delete_flow(struct ofproto
*ofproto
, const flow_t
*flow
,
1165 uint32_t wildcards
, unsigned int priority
)
1169 rule
= rule_from_cls_rule(classifier_find_rule_exactly(&ofproto
->cls
,
1173 rule_remove(ofproto
, rule
);
1178 destroy_rule(struct cls_rule
*rule_
, void *ofproto_
)
1180 struct rule
*rule
= rule_from_cls_rule(rule_
);
1181 struct ofproto
*ofproto
= ofproto_
;
1183 /* Mark the flow as not installed, even though it might really be
1184 * installed, so that rule_remove() doesn't bother trying to uninstall it.
1185 * There is no point in uninstalling it individually since we are about to
1186 * blow away all the flows with dpif_flow_flush(). */
1187 rule
->installed
= false;
1189 rule_remove(ofproto
, rule
);
1193 ofproto_flush_flows(struct ofproto
*ofproto
)
1195 COVERAGE_INC(ofproto_flush
);
1196 classifier_for_each(&ofproto
->cls
, CLS_INC_ALL
, destroy_rule
, ofproto
);
1197 dpif_flow_flush(ofproto
->dpif
);
1198 if (ofproto
->in_band
) {
1199 in_band_flushed(ofproto
->in_band
);
1201 if (ofproto
->fail_open
) {
1202 fail_open_flushed(ofproto
->fail_open
);
1207 reinit_ports(struct ofproto
*p
)
1209 struct svec devnames
;
1210 struct ofport
*ofport
;
1211 unsigned int port_no
;
1212 struct odp_port
*odp_ports
;
1216 svec_init(&devnames
);
1217 PORT_ARRAY_FOR_EACH (ofport
, &p
->ports
, port_no
) {
1218 svec_add (&devnames
, (char *) ofport
->opp
.name
);
1220 dpif_port_list(p
->dpif
, &odp_ports
, &n_odp_ports
);
1221 for (i
= 0; i
< n_odp_ports
; i
++) {
1222 svec_add (&devnames
, odp_ports
[i
].devname
);
1226 svec_sort_unique(&devnames
);
1227 for (i
= 0; i
< devnames
.n
; i
++) {
1228 update_port(p
, devnames
.names
[i
]);
1230 svec_destroy(&devnames
);
1234 refresh_port_group(struct ofproto
*p
, unsigned int group
)
1238 struct ofport
*port
;
1239 unsigned int port_no
;
1241 assert(group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
);
1243 ports
= xmalloc(port_array_count(&p
->ports
) * sizeof *ports
);
1245 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
1246 if (group
== DP_GROUP_ALL
|| !(port
->opp
.config
& OFPPC_NO_FLOOD
)) {
1247 ports
[n_ports
++] = port_no
;
1250 dpif_port_group_set(p
->dpif
, group
, ports
, n_ports
);
1257 refresh_port_groups(struct ofproto
*p
)
1259 size_t n_flood
= refresh_port_group(p
, DP_GROUP_FLOOD
);
1260 size_t n_all
= refresh_port_group(p
, DP_GROUP_ALL
);
1262 ofproto_sflow_set_group_sizes(p
->sflow
, n_flood
, n_all
);
1266 static struct ofport
*
1267 make_ofport(const struct odp_port
*odp_port
)
1269 struct netdev_options netdev_options
;
1270 enum netdev_flags flags
;
1271 struct ofport
*ofport
;
1272 struct netdev
*netdev
;
1276 memset(&netdev_options
, 0, sizeof netdev_options
);
1277 netdev_options
.name
= odp_port
->devname
;
1278 netdev_options
.ethertype
= NETDEV_ETH_TYPE_NONE
;
1279 netdev_options
.may_open
= true;
1281 error
= netdev_open(&netdev_options
, &netdev
);
1283 VLOG_WARN_RL(&rl
, "ignoring port %s (%"PRIu16
") because netdev %s "
1284 "cannot be opened (%s)",
1285 odp_port
->devname
, odp_port
->port
,
1286 odp_port
->devname
, strerror(error
));
1290 ofport
= xmalloc(sizeof *ofport
);
1291 ofport
->netdev
= netdev
;
1292 ofport
->opp
.port_no
= odp_port_to_ofp_port(odp_port
->port
);
1293 netdev_get_etheraddr(netdev
, ofport
->opp
.hw_addr
);
1294 memcpy(ofport
->opp
.name
, odp_port
->devname
,
1295 MIN(sizeof ofport
->opp
.name
, sizeof odp_port
->devname
));
1296 ofport
->opp
.name
[sizeof ofport
->opp
.name
- 1] = '\0';
1298 netdev_get_flags(netdev
, &flags
);
1299 ofport
->opp
.config
= flags
& NETDEV_UP
? 0 : OFPPC_PORT_DOWN
;
1301 netdev_get_carrier(netdev
, &carrier
);
1302 ofport
->opp
.state
= carrier
? 0 : OFPPS_LINK_DOWN
;
1304 netdev_get_features(netdev
,
1305 &ofport
->opp
.curr
, &ofport
->opp
.advertised
,
1306 &ofport
->opp
.supported
, &ofport
->opp
.peer
);
1311 ofport_conflicts(const struct ofproto
*p
, const struct odp_port
*odp_port
)
1313 if (port_array_get(&p
->ports
, odp_port
->port
)) {
1314 VLOG_WARN_RL(&rl
, "ignoring duplicate port %"PRIu16
" in datapath",
1317 } else if (shash_find(&p
->port_by_name
, odp_port
->devname
)) {
1318 VLOG_WARN_RL(&rl
, "ignoring duplicate device %s in datapath",
1327 ofport_equal(const struct ofport
*a_
, const struct ofport
*b_
)
1329 const struct ofp_phy_port
*a
= &a_
->opp
;
1330 const struct ofp_phy_port
*b
= &b_
->opp
;
1332 BUILD_ASSERT_DECL(sizeof *a
== 48); /* Detect ofp_phy_port changes. */
1333 return (a
->port_no
== b
->port_no
1334 && !memcmp(a
->hw_addr
, b
->hw_addr
, sizeof a
->hw_addr
)
1335 && !strcmp((char *) a
->name
, (char *) b
->name
)
1336 && a
->state
== b
->state
1337 && a
->config
== b
->config
1338 && a
->curr
== b
->curr
1339 && a
->advertised
== b
->advertised
1340 && a
->supported
== b
->supported
1341 && a
->peer
== b
->peer
);
1345 send_port_status(struct ofproto
*p
, const struct ofport
*ofport
,
1348 /* XXX Should limit the number of queued port status change messages. */
1349 struct ofconn
*ofconn
;
1350 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
1351 struct ofp_port_status
*ops
;
1354 if (ofconn
->role
== NX_ROLE_SLAVE
) {
1358 ops
= make_openflow_xid(sizeof *ops
, OFPT_PORT_STATUS
, 0, &b
);
1359 ops
->reason
= reason
;
1360 ops
->desc
= ofport
->opp
;
1361 hton_ofp_phy_port(&ops
->desc
);
1362 queue_tx(b
, ofconn
, NULL
);
1364 if (p
->ofhooks
->port_changed_cb
) {
1365 p
->ofhooks
->port_changed_cb(reason
, &ofport
->opp
, p
->aux
);
1370 ofport_install(struct ofproto
*p
, struct ofport
*ofport
)
1372 uint16_t odp_port
= ofp_port_to_odp_port(ofport
->opp
.port_no
);
1373 const char *netdev_name
= (const char *) ofport
->opp
.name
;
1375 netdev_monitor_add(p
->netdev_monitor
, ofport
->netdev
);
1376 port_array_set(&p
->ports
, odp_port
, ofport
);
1377 shash_add(&p
->port_by_name
, netdev_name
, ofport
);
1379 ofproto_sflow_add_port(p
->sflow
, odp_port
, netdev_name
);
1384 ofport_remove(struct ofproto
*p
, struct ofport
*ofport
)
1386 uint16_t odp_port
= ofp_port_to_odp_port(ofport
->opp
.port_no
);
1388 netdev_monitor_remove(p
->netdev_monitor
, ofport
->netdev
);
1389 port_array_set(&p
->ports
, odp_port
, NULL
);
1390 shash_delete(&p
->port_by_name
,
1391 shash_find(&p
->port_by_name
, (char *) ofport
->opp
.name
));
1393 ofproto_sflow_del_port(p
->sflow
, odp_port
);
1398 ofport_free(struct ofport
*ofport
)
1401 netdev_close(ofport
->netdev
);
1407 update_port(struct ofproto
*p
, const char *devname
)
1409 struct odp_port odp_port
;
1410 struct ofport
*old_ofport
;
1411 struct ofport
*new_ofport
;
1414 COVERAGE_INC(ofproto_update_port
);
1416 /* Query the datapath for port information. */
1417 error
= dpif_port_query_by_name(p
->dpif
, devname
, &odp_port
);
1419 /* Find the old ofport. */
1420 old_ofport
= shash_find_data(&p
->port_by_name
, devname
);
1423 /* There's no port named 'devname' but there might be a port with
1424 * the same port number. This could happen if a port is deleted
1425 * and then a new one added in its place very quickly, or if a port
1426 * is renamed. In the former case we want to send an OFPPR_DELETE
1427 * and an OFPPR_ADD, and in the latter case we want to send a
1428 * single OFPPR_MODIFY. We can distinguish the cases by comparing
1429 * the old port's ifindex against the new port, or perhaps less
1430 * reliably but more portably by comparing the old port's MAC
1431 * against the new port's MAC. However, this code isn't that smart
1432 * and always sends an OFPPR_MODIFY (XXX). */
1433 old_ofport
= port_array_get(&p
->ports
, odp_port
.port
);
1435 } else if (error
!= ENOENT
&& error
!= ENODEV
) {
1436 VLOG_WARN_RL(&rl
, "dpif_port_query_by_name returned unexpected error "
1437 "%s", strerror(error
));
1441 /* Create a new ofport. */
1442 new_ofport
= !error
? make_ofport(&odp_port
) : NULL
;
1444 /* Eliminate a few pathological cases. */
1445 if (!old_ofport
&& !new_ofport
) {
1447 } else if (old_ofport
&& new_ofport
) {
1448 /* Most of the 'config' bits are OpenFlow soft state, but
1449 * OFPPC_PORT_DOWN is maintained the kernel. So transfer the OpenFlow
1450 * bits from old_ofport. (make_ofport() only sets OFPPC_PORT_DOWN and
1451 * leaves the other bits 0.) */
1452 new_ofport
->opp
.config
|= old_ofport
->opp
.config
& ~OFPPC_PORT_DOWN
;
1454 if (ofport_equal(old_ofport
, new_ofport
)) {
1455 /* False alarm--no change. */
1456 ofport_free(new_ofport
);
1461 /* Now deal with the normal cases. */
1463 ofport_remove(p
, old_ofport
);
1466 ofport_install(p
, new_ofport
);
1468 send_port_status(p
, new_ofport
? new_ofport
: old_ofport
,
1469 (!old_ofport
? OFPPR_ADD
1470 : !new_ofport
? OFPPR_DELETE
1472 ofport_free(old_ofport
);
1474 /* Update port groups. */
1475 refresh_port_groups(p
);
1479 init_ports(struct ofproto
*p
)
1481 struct odp_port
*ports
;
1486 error
= dpif_port_list(p
->dpif
, &ports
, &n_ports
);
1491 for (i
= 0; i
< n_ports
; i
++) {
1492 const struct odp_port
*odp_port
= &ports
[i
];
1493 if (!ofport_conflicts(p
, odp_port
)) {
1494 struct ofport
*ofport
= make_ofport(odp_port
);
1496 ofport_install(p
, ofport
);
1501 refresh_port_groups(p
);
1505 static struct ofconn
*
1506 ofconn_create(struct ofproto
*p
, struct rconn
*rconn
, enum ofconn_type type
)
1508 struct ofconn
*ofconn
= xzalloc(sizeof *ofconn
);
1509 ofconn
->ofproto
= p
;
1510 list_push_back(&p
->all_conns
, &ofconn
->node
);
1511 ofconn
->rconn
= rconn
;
1512 ofconn
->type
= type
;
1513 ofconn
->role
= NX_ROLE_OTHER
;
1514 ofconn
->packet_in_counter
= rconn_packet_counter_create ();
1515 ofconn
->pktbuf
= NULL
;
1516 ofconn
->miss_send_len
= 0;
1517 ofconn
->reply_counter
= rconn_packet_counter_create ();
1522 ofconn_destroy(struct ofconn
*ofconn
)
1524 if (ofconn
->type
== OFCONN_CONTROLLER
) {
1525 hmap_remove(&ofconn
->ofproto
->controllers
, &ofconn
->hmap_node
);
1527 discovery_destroy(ofconn
->discovery
);
1529 list_remove(&ofconn
->node
);
1530 switch_status_unregister(ofconn
->ss
);
1531 rconn_destroy(ofconn
->rconn
);
1532 rconn_packet_counter_destroy(ofconn
->packet_in_counter
);
1533 rconn_packet_counter_destroy(ofconn
->reply_counter
);
1534 pktbuf_destroy(ofconn
->pktbuf
);
1539 ofconn_run(struct ofconn
*ofconn
, struct ofproto
*p
)
1544 if (ofconn
->discovery
) {
1545 char *controller_name
;
1546 if (rconn_is_connectivity_questionable(ofconn
->rconn
)) {
1547 discovery_question_connectivity(ofconn
->discovery
);
1549 if (discovery_run(ofconn
->discovery
, &controller_name
)) {
1550 if (controller_name
) {
1551 rconn_connect(ofconn
->rconn
, controller_name
);
1553 rconn_disconnect(ofconn
->rconn
);
1558 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1559 pinsched_run(ofconn
->schedulers
[i
], do_send_packet_in
, ofconn
);
1562 rconn_run(ofconn
->rconn
);
1564 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1565 /* Limit the number of iterations to prevent other tasks from
1567 for (iteration
= 0; iteration
< 50; iteration
++) {
1568 struct ofpbuf
*of_msg
= rconn_recv(ofconn
->rconn
);
1573 fail_open_maybe_recover(p
->fail_open
);
1575 handle_openflow(ofconn
, p
, of_msg
);
1576 ofpbuf_delete(of_msg
);
1580 if (!ofconn
->discovery
&& !rconn_is_alive(ofconn
->rconn
)) {
1581 ofconn_destroy(ofconn
);
1586 ofconn_wait(struct ofconn
*ofconn
)
1590 if (ofconn
->discovery
) {
1591 discovery_wait(ofconn
->discovery
);
1593 for (i
= 0; i
< N_SCHEDULERS
; i
++) {
1594 pinsched_wait(ofconn
->schedulers
[i
]);
1596 rconn_run_wait(ofconn
->rconn
);
1597 if (rconn_packet_counter_read (ofconn
->reply_counter
) < OFCONN_REPLY_MAX
) {
1598 rconn_recv_wait(ofconn
->rconn
);
1600 COVERAGE_INC(ofproto_ofconn_stuck
);
1604 /* Caller is responsible for initializing the 'cr' member of the returned
1606 static struct rule
*
1607 rule_create(struct ofproto
*ofproto
, struct rule
*super
,
1608 const union ofp_action
*actions
, size_t n_actions
,
1609 uint16_t idle_timeout
, uint16_t hard_timeout
,
1610 uint64_t flow_cookie
, bool send_flow_removed
)
1612 struct rule
*rule
= xzalloc(sizeof *rule
);
1613 rule
->idle_timeout
= idle_timeout
;
1614 rule
->hard_timeout
= hard_timeout
;
1615 rule
->flow_cookie
= flow_cookie
;
1616 rule
->used
= rule
->created
= time_msec();
1617 rule
->send_flow_removed
= send_flow_removed
;
1618 rule
->super
= super
;
1620 list_push_back(&super
->list
, &rule
->list
);
1622 list_init(&rule
->list
);
1624 rule
->n_actions
= n_actions
;
1625 rule
->actions
= xmemdup(actions
, n_actions
* sizeof *actions
);
1626 netflow_flow_clear(&rule
->nf_flow
);
1627 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->created
);
1632 static struct rule
*
1633 rule_from_cls_rule(const struct cls_rule
*cls_rule
)
1635 return cls_rule
? CONTAINER_OF(cls_rule
, struct rule
, cr
) : NULL
;
1639 rule_free(struct rule
*rule
)
1641 free(rule
->actions
);
1642 free(rule
->odp_actions
);
1646 /* Destroys 'rule'. If 'rule' is a subrule, also removes it from its
1647 * super-rule's list of subrules. If 'rule' is a super-rule, also iterates
1648 * through all of its subrules and revalidates them, destroying any that no
1649 * longer has a super-rule (which is probably all of them).
1651 * Before calling this function, the caller must make have removed 'rule' from
1652 * the classifier. If 'rule' is an exact-match rule, the caller is also
1653 * responsible for ensuring that it has been uninstalled from the datapath. */
1655 rule_destroy(struct ofproto
*ofproto
, struct rule
*rule
)
1658 struct rule
*subrule
, *next
;
1659 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
1660 revalidate_rule(ofproto
, subrule
);
1663 list_remove(&rule
->list
);
1669 rule_has_out_port(const struct rule
*rule
, uint16_t out_port
)
1671 const union ofp_action
*oa
;
1672 struct actions_iterator i
;
1674 if (out_port
== htons(OFPP_NONE
)) {
1677 for (oa
= actions_first(&i
, rule
->actions
, rule
->n_actions
); oa
;
1678 oa
= actions_next(&i
)) {
1679 if (oa
->type
== htons(OFPAT_OUTPUT
) && oa
->output
.port
== out_port
) {
1686 /* Executes the actions indicated by 'rule' on 'packet', which is in flow
1687 * 'flow' and is considered to have arrived on ODP port 'in_port'.
1689 * The flow that 'packet' actually contains does not need to actually match
1690 * 'rule'; the actions in 'rule' will be applied to it either way. Likewise,
1691 * the packet and byte counters for 'rule' will be credited for the packet sent
1692 * out whether or not the packet actually matches 'rule'.
1694 * If 'rule' is an exact-match rule and 'flow' actually equals the rule's flow,
1695 * the caller must already have accurately composed ODP actions for it given
1696 * 'packet' using rule_make_actions(). If 'rule' is a wildcard rule, or if
1697 * 'rule' is an exact-match rule but 'flow' is not the rule's flow, then this
1698 * function will compose a set of ODP actions based on 'rule''s OpenFlow
1699 * actions and apply them to 'packet'. */
1701 rule_execute(struct ofproto
*ofproto
, struct rule
*rule
,
1702 struct ofpbuf
*packet
, const flow_t
*flow
)
1704 const union odp_action
*actions
;
1706 struct odp_actions a
;
1708 /* Grab or compose the ODP actions.
1710 * The special case for an exact-match 'rule' where 'flow' is not the
1711 * rule's flow is important to avoid, e.g., sending a packet out its input
1712 * port simply because the ODP actions were composed for the wrong
1714 if (rule
->cr
.wc
.wildcards
|| !flow_equal(flow
, &rule
->cr
.flow
)) {
1715 struct rule
*super
= rule
->super
? rule
->super
: rule
;
1716 if (xlate_actions(super
->actions
, super
->n_actions
, flow
, ofproto
,
1717 packet
, &a
, NULL
, 0, NULL
)) {
1720 actions
= a
.actions
;
1721 n_actions
= a
.n_actions
;
1723 actions
= rule
->odp_actions
;
1724 n_actions
= rule
->n_odp_actions
;
1727 /* Execute the ODP actions. */
1728 if (!dpif_execute(ofproto
->dpif
, flow
->in_port
,
1729 actions
, n_actions
, packet
)) {
1730 struct odp_flow_stats stats
;
1731 flow_extract_stats(flow
, packet
, &stats
);
1732 update_stats(ofproto
, rule
, &stats
);
1733 rule
->used
= time_msec();
1734 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, rule
->used
);
1739 rule_insert(struct ofproto
*p
, struct rule
*rule
, struct ofpbuf
*packet
,
1742 struct rule
*displaced_rule
;
1744 /* Insert the rule in the classifier. */
1745 displaced_rule
= rule_from_cls_rule(classifier_insert(&p
->cls
, &rule
->cr
));
1746 if (!rule
->cr
.wc
.wildcards
) {
1747 rule_make_actions(p
, rule
, packet
);
1750 /* Send the packet and credit it to the rule. */
1753 flow_extract(packet
, 0, in_port
, &flow
);
1754 rule_execute(p
, rule
, packet
, &flow
);
1757 /* Install the rule in the datapath only after sending the packet, to
1758 * avoid packet reordering. */
1759 if (rule
->cr
.wc
.wildcards
) {
1760 COVERAGE_INC(ofproto_add_wc_flow
);
1761 p
->need_revalidate
= true;
1763 rule_install(p
, rule
, displaced_rule
);
1766 /* Free the rule that was displaced, if any. */
1767 if (displaced_rule
) {
1768 rule_destroy(p
, displaced_rule
);
1772 static struct rule
*
1773 rule_create_subrule(struct ofproto
*ofproto
, struct rule
*rule
,
1776 struct rule
*subrule
= rule_create(ofproto
, rule
, NULL
, 0,
1777 rule
->idle_timeout
, rule
->hard_timeout
,
1779 COVERAGE_INC(ofproto_subrule_create
);
1780 cls_rule_from_flow(flow
, 0, (rule
->cr
.priority
<= UINT16_MAX
? UINT16_MAX
1781 : rule
->cr
.priority
), &subrule
->cr
);
1782 classifier_insert_exact(&ofproto
->cls
, &subrule
->cr
);
1788 rule_remove(struct ofproto
*ofproto
, struct rule
*rule
)
1790 if (rule
->cr
.wc
.wildcards
) {
1791 COVERAGE_INC(ofproto_del_wc_flow
);
1792 ofproto
->need_revalidate
= true;
1794 rule_uninstall(ofproto
, rule
);
1796 classifier_remove(&ofproto
->cls
, &rule
->cr
);
1797 rule_destroy(ofproto
, rule
);
1800 /* Returns true if the actions changed, false otherwise. */
1802 rule_make_actions(struct ofproto
*p
, struct rule
*rule
,
1803 const struct ofpbuf
*packet
)
1805 const struct rule
*super
;
1806 struct odp_actions a
;
1809 assert(!rule
->cr
.wc
.wildcards
);
1811 super
= rule
->super
? rule
->super
: rule
;
1813 xlate_actions(super
->actions
, super
->n_actions
, &rule
->cr
.flow
, p
,
1814 packet
, &a
, &rule
->tags
, &rule
->may_install
,
1815 &rule
->nf_flow
.output_iface
);
1817 actions_len
= a
.n_actions
* sizeof *a
.actions
;
1818 if (rule
->n_odp_actions
!= a
.n_actions
1819 || memcmp(rule
->odp_actions
, a
.actions
, actions_len
)) {
1820 COVERAGE_INC(ofproto_odp_unchanged
);
1821 free(rule
->odp_actions
);
1822 rule
->n_odp_actions
= a
.n_actions
;
1823 rule
->odp_actions
= xmemdup(a
.actions
, actions_len
);
1831 do_put_flow(struct ofproto
*ofproto
, struct rule
*rule
, int flags
,
1832 struct odp_flow_put
*put
)
1834 memset(&put
->flow
.stats
, 0, sizeof put
->flow
.stats
);
1835 put
->flow
.key
= rule
->cr
.flow
;
1836 put
->flow
.actions
= rule
->odp_actions
;
1837 put
->flow
.n_actions
= rule
->n_odp_actions
;
1838 put
->flow
.flags
= 0;
1840 return dpif_flow_put(ofproto
->dpif
, put
);
1844 rule_install(struct ofproto
*p
, struct rule
*rule
, struct rule
*displaced_rule
)
1846 assert(!rule
->cr
.wc
.wildcards
);
1848 if (rule
->may_install
) {
1849 struct odp_flow_put put
;
1850 if (!do_put_flow(p
, rule
,
1851 ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
,
1853 rule
->installed
= true;
1854 if (displaced_rule
) {
1855 update_stats(p
, displaced_rule
, &put
.flow
.stats
);
1856 rule_post_uninstall(p
, displaced_rule
);
1859 } else if (displaced_rule
) {
1860 rule_uninstall(p
, displaced_rule
);
1865 rule_reinstall(struct ofproto
*ofproto
, struct rule
*rule
)
1867 if (rule
->installed
) {
1868 struct odp_flow_put put
;
1869 COVERAGE_INC(ofproto_dp_missed
);
1870 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
, &put
);
1872 rule_install(ofproto
, rule
, NULL
);
1877 rule_update_actions(struct ofproto
*ofproto
, struct rule
*rule
)
1879 bool actions_changed
;
1880 uint16_t new_out_iface
, old_out_iface
;
1882 old_out_iface
= rule
->nf_flow
.output_iface
;
1883 actions_changed
= rule_make_actions(ofproto
, rule
, NULL
);
1885 if (rule
->may_install
) {
1886 if (rule
->installed
) {
1887 if (actions_changed
) {
1888 struct odp_flow_put put
;
1889 do_put_flow(ofproto
, rule
, ODPPF_CREATE
| ODPPF_MODIFY
1890 | ODPPF_ZERO_STATS
, &put
);
1891 update_stats(ofproto
, rule
, &put
.flow
.stats
);
1893 /* Temporarily set the old output iface so that NetFlow
1894 * messages have the correct output interface for the old
1896 new_out_iface
= rule
->nf_flow
.output_iface
;
1897 rule
->nf_flow
.output_iface
= old_out_iface
;
1898 rule_post_uninstall(ofproto
, rule
);
1899 rule
->nf_flow
.output_iface
= new_out_iface
;
1902 rule_install(ofproto
, rule
, NULL
);
1905 rule_uninstall(ofproto
, rule
);
1910 rule_account(struct ofproto
*ofproto
, struct rule
*rule
, uint64_t extra_bytes
)
1912 uint64_t total_bytes
= rule
->byte_count
+ extra_bytes
;
1914 if (ofproto
->ofhooks
->account_flow_cb
1915 && total_bytes
> rule
->accounted_bytes
)
1917 ofproto
->ofhooks
->account_flow_cb(
1918 &rule
->cr
.flow
, rule
->odp_actions
, rule
->n_odp_actions
,
1919 total_bytes
- rule
->accounted_bytes
, ofproto
->aux
);
1920 rule
->accounted_bytes
= total_bytes
;
1925 rule_uninstall(struct ofproto
*p
, struct rule
*rule
)
1927 assert(!rule
->cr
.wc
.wildcards
);
1928 if (rule
->installed
) {
1929 struct odp_flow odp_flow
;
1931 odp_flow
.key
= rule
->cr
.flow
;
1932 odp_flow
.actions
= NULL
;
1933 odp_flow
.n_actions
= 0;
1935 if (!dpif_flow_del(p
->dpif
, &odp_flow
)) {
1936 update_stats(p
, rule
, &odp_flow
.stats
);
1938 rule
->installed
= false;
1940 rule_post_uninstall(p
, rule
);
1945 is_controller_rule(struct rule
*rule
)
1947 /* If the only action is send to the controller then don't report
1948 * NetFlow expiration messages since it is just part of the control
1949 * logic for the network and not real traffic. */
1951 if (rule
&& rule
->super
) {
1952 struct rule
*super
= rule
->super
;
1954 return super
->n_actions
== 1 &&
1955 super
->actions
[0].type
== htons(OFPAT_OUTPUT
) &&
1956 super
->actions
[0].output
.port
== htons(OFPP_CONTROLLER
);
1963 rule_post_uninstall(struct ofproto
*ofproto
, struct rule
*rule
)
1965 struct rule
*super
= rule
->super
;
1967 rule_account(ofproto
, rule
, 0);
1969 if (ofproto
->netflow
&& !is_controller_rule(rule
)) {
1970 struct ofexpired expired
;
1971 expired
.flow
= rule
->cr
.flow
;
1972 expired
.packet_count
= rule
->packet_count
;
1973 expired
.byte_count
= rule
->byte_count
;
1974 expired
.used
= rule
->used
;
1975 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
1978 super
->packet_count
+= rule
->packet_count
;
1979 super
->byte_count
+= rule
->byte_count
;
1981 /* Reset counters to prevent double counting if the rule ever gets
1983 rule
->packet_count
= 0;
1984 rule
->byte_count
= 0;
1985 rule
->accounted_bytes
= 0;
1987 netflow_flow_clear(&rule
->nf_flow
);
1992 queue_tx(struct ofpbuf
*msg
, const struct ofconn
*ofconn
,
1993 struct rconn_packet_counter
*counter
)
1995 update_openflow_length(msg
);
1996 if (rconn_send(ofconn
->rconn
, msg
, counter
)) {
2002 send_error(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2003 int error
, const void *data
, size_t len
)
2006 struct ofp_error_msg
*oem
;
2008 if (!(error
>> 16)) {
2009 VLOG_WARN_RL(&rl
, "not sending bad error code %d to controller",
2014 COVERAGE_INC(ofproto_error
);
2015 oem
= make_openflow_xid(len
+ sizeof *oem
, OFPT_ERROR
,
2016 oh
? oh
->xid
: 0, &buf
);
2017 oem
->type
= htons((unsigned int) error
>> 16);
2018 oem
->code
= htons(error
& 0xffff);
2019 memcpy(oem
->data
, data
, len
);
2020 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2024 send_error_oh(const struct ofconn
*ofconn
, const struct ofp_header
*oh
,
2027 size_t oh_length
= ntohs(oh
->length
);
2028 send_error(ofconn
, oh
, error
, oh
, MIN(oh_length
, 64));
2032 hton_ofp_phy_port(struct ofp_phy_port
*opp
)
2034 opp
->port_no
= htons(opp
->port_no
);
2035 opp
->config
= htonl(opp
->config
);
2036 opp
->state
= htonl(opp
->state
);
2037 opp
->curr
= htonl(opp
->curr
);
2038 opp
->advertised
= htonl(opp
->advertised
);
2039 opp
->supported
= htonl(opp
->supported
);
2040 opp
->peer
= htonl(opp
->peer
);
2044 handle_echo_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
2046 struct ofp_header
*rq
= oh
;
2047 queue_tx(make_echo_reply(rq
), ofconn
, ofconn
->reply_counter
);
2052 handle_features_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2053 struct ofp_header
*oh
)
2055 struct ofp_switch_features
*osf
;
2057 unsigned int port_no
;
2058 struct ofport
*port
;
2060 osf
= make_openflow_xid(sizeof *osf
, OFPT_FEATURES_REPLY
, oh
->xid
, &buf
);
2061 osf
->datapath_id
= htonll(p
->datapath_id
);
2062 osf
->n_buffers
= htonl(pktbuf_capacity());
2064 osf
->capabilities
= htonl(OFPC_FLOW_STATS
| OFPC_TABLE_STATS
|
2065 OFPC_PORT_STATS
| OFPC_ARP_MATCH_IP
);
2066 osf
->actions
= htonl((1u << OFPAT_OUTPUT
) |
2067 (1u << OFPAT_SET_VLAN_VID
) |
2068 (1u << OFPAT_SET_VLAN_PCP
) |
2069 (1u << OFPAT_STRIP_VLAN
) |
2070 (1u << OFPAT_SET_DL_SRC
) |
2071 (1u << OFPAT_SET_DL_DST
) |
2072 (1u << OFPAT_SET_NW_SRC
) |
2073 (1u << OFPAT_SET_NW_DST
) |
2074 (1u << OFPAT_SET_NW_TOS
) |
2075 (1u << OFPAT_SET_TP_SRC
) |
2076 (1u << OFPAT_SET_TP_DST
));
2078 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
2079 hton_ofp_phy_port(ofpbuf_put(buf
, &port
->opp
, sizeof port
->opp
));
2082 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2087 handle_get_config_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2088 struct ofp_header
*oh
)
2091 struct ofp_switch_config
*osc
;
2095 /* Figure out flags. */
2096 dpif_get_drop_frags(p
->dpif
, &drop_frags
);
2097 flags
= drop_frags
? OFPC_FRAG_DROP
: OFPC_FRAG_NORMAL
;
2100 osc
= make_openflow_xid(sizeof *osc
, OFPT_GET_CONFIG_REPLY
, oh
->xid
, &buf
);
2101 osc
->flags
= htons(flags
);
2102 osc
->miss_send_len
= htons(ofconn
->miss_send_len
);
2103 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
2109 handle_set_config(struct ofproto
*p
, struct ofconn
*ofconn
,
2110 struct ofp_switch_config
*osc
)
2115 error
= check_ofp_message(&osc
->header
, OFPT_SET_CONFIG
, sizeof *osc
);
2119 flags
= ntohs(osc
->flags
);
2121 if (ofconn
->type
== OFCONN_CONTROLLER
&& ofconn
->role
!= NX_ROLE_SLAVE
) {
2122 switch (flags
& OFPC_FRAG_MASK
) {
2123 case OFPC_FRAG_NORMAL
:
2124 dpif_set_drop_frags(p
->dpif
, false);
2126 case OFPC_FRAG_DROP
:
2127 dpif_set_drop_frags(p
->dpif
, true);
2130 VLOG_WARN_RL(&rl
, "requested bad fragment mode (flags=%"PRIx16
")",
2136 ofconn
->miss_send_len
= ntohs(osc
->miss_send_len
);
2142 add_output_group_action(struct odp_actions
*actions
, uint16_t group
,
2143 uint16_t *nf_output_iface
)
2145 odp_actions_add(actions
, ODPAT_OUTPUT_GROUP
)->output_group
.group
= group
;
2147 if (group
== DP_GROUP_ALL
|| group
== DP_GROUP_FLOOD
) {
2148 *nf_output_iface
= NF_OUT_FLOOD
;
2153 add_controller_action(struct odp_actions
*actions
,
2154 const struct ofp_action_output
*oao
)
2156 union odp_action
*a
= odp_actions_add(actions
, ODPAT_CONTROLLER
);
2157 a
->controller
.arg
= oao
->max_len
? ntohs(oao
->max_len
) : UINT32_MAX
;
2160 struct action_xlate_ctx
{
2162 flow_t flow
; /* Flow to which these actions correspond. */
2163 int recurse
; /* Recursion level, via xlate_table_action. */
2164 struct ofproto
*ofproto
;
2165 const struct ofpbuf
*packet
; /* The packet corresponding to 'flow', or a
2166 * null pointer if we are revalidating
2167 * without a packet to refer to. */
2170 struct odp_actions
*out
; /* Datapath actions. */
2171 tag_type
*tags
; /* Tags associated with OFPP_NORMAL actions. */
2172 bool may_set_up_flow
; /* True ordinarily; false if the actions must
2173 * be reassessed for every packet. */
2174 uint16_t nf_output_iface
; /* Output interface index for NetFlow. */
2177 static void do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2178 struct action_xlate_ctx
*ctx
);
2181 add_output_action(struct action_xlate_ctx
*ctx
, uint16_t port
)
2183 const struct ofport
*ofport
= port_array_get(&ctx
->ofproto
->ports
, port
);
2186 if (ofport
->opp
.config
& OFPPC_NO_FWD
) {
2187 /* Forwarding disabled on port. */
2192 * We don't have an ofport record for this port, but it doesn't hurt to
2193 * allow forwarding to it anyhow. Maybe such a port will appear later
2194 * and we're pre-populating the flow table.
2198 odp_actions_add(ctx
->out
, ODPAT_OUTPUT
)->output
.port
= port
;
2199 ctx
->nf_output_iface
= port
;
2202 static struct rule
*
2203 lookup_valid_rule(struct ofproto
*ofproto
, const flow_t
*flow
)
2206 rule
= rule_from_cls_rule(classifier_lookup(&ofproto
->cls
, flow
));
2208 /* The rule we found might not be valid, since we could be in need of
2209 * revalidation. If it is not valid, don't return it. */
2212 && ofproto
->need_revalidate
2213 && !revalidate_rule(ofproto
, rule
)) {
2214 COVERAGE_INC(ofproto_invalidated
);
2222 xlate_table_action(struct action_xlate_ctx
*ctx
, uint16_t in_port
)
2224 if (!ctx
->recurse
) {
2225 uint16_t old_in_port
;
2228 /* Look up a flow with 'in_port' as the input port. Then restore the
2229 * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will
2230 * have surprising behavior). */
2231 old_in_port
= ctx
->flow
.in_port
;
2232 ctx
->flow
.in_port
= in_port
;
2233 rule
= lookup_valid_rule(ctx
->ofproto
, &ctx
->flow
);
2234 ctx
->flow
.in_port
= old_in_port
;
2242 do_xlate_actions(rule
->actions
, rule
->n_actions
, ctx
);
2249 xlate_output_action(struct action_xlate_ctx
*ctx
,
2250 const struct ofp_action_output
*oao
)
2253 uint16_t prev_nf_output_iface
= ctx
->nf_output_iface
;
2255 ctx
->nf_output_iface
= NF_OUT_DROP
;
2257 switch (ntohs(oao
->port
)) {
2259 add_output_action(ctx
, ctx
->flow
.in_port
);
2262 xlate_table_action(ctx
, ctx
->flow
.in_port
);
2265 if (!ctx
->ofproto
->ofhooks
->normal_cb(&ctx
->flow
, ctx
->packet
,
2266 ctx
->out
, ctx
->tags
,
2267 &ctx
->nf_output_iface
,
2268 ctx
->ofproto
->aux
)) {
2269 COVERAGE_INC(ofproto_uninstallable
);
2270 ctx
->may_set_up_flow
= false;
2274 add_output_group_action(ctx
->out
, DP_GROUP_FLOOD
,
2275 &ctx
->nf_output_iface
);
2278 add_output_group_action(ctx
->out
, DP_GROUP_ALL
, &ctx
->nf_output_iface
);
2280 case OFPP_CONTROLLER
:
2281 add_controller_action(ctx
->out
, oao
);
2284 add_output_action(ctx
, ODPP_LOCAL
);
2287 odp_port
= ofp_port_to_odp_port(ntohs(oao
->port
));
2288 if (odp_port
!= ctx
->flow
.in_port
) {
2289 add_output_action(ctx
, odp_port
);
2294 if (prev_nf_output_iface
== NF_OUT_FLOOD
) {
2295 ctx
->nf_output_iface
= NF_OUT_FLOOD
;
2296 } else if (ctx
->nf_output_iface
== NF_OUT_DROP
) {
2297 ctx
->nf_output_iface
= prev_nf_output_iface
;
2298 } else if (prev_nf_output_iface
!= NF_OUT_DROP
&&
2299 ctx
->nf_output_iface
!= NF_OUT_FLOOD
) {
2300 ctx
->nf_output_iface
= NF_OUT_MULTI
;
2305 xlate_nicira_action(struct action_xlate_ctx
*ctx
,
2306 const struct nx_action_header
*nah
)
2308 const struct nx_action_resubmit
*nar
;
2309 const struct nx_action_set_tunnel
*nast
;
2310 union odp_action
*oa
;
2311 int subtype
= ntohs(nah
->subtype
);
2313 assert(nah
->vendor
== htonl(NX_VENDOR_ID
));
2315 case NXAST_RESUBMIT
:
2316 nar
= (const struct nx_action_resubmit
*) nah
;
2317 xlate_table_action(ctx
, ofp_port_to_odp_port(ntohs(nar
->in_port
)));
2320 case NXAST_SET_TUNNEL
:
2321 nast
= (const struct nx_action_set_tunnel
*) nah
;
2322 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TUNNEL
);
2323 ctx
->flow
.tun_id
= oa
->tunnel
.tun_id
= nast
->tun_id
;
2326 /* If you add a new action here that modifies flow data, don't forget to
2327 * update the flow key in ctx->flow in the same key. */
2330 VLOG_DBG_RL(&rl
, "unknown Nicira action type %"PRIu16
, subtype
);
2336 do_xlate_actions(const union ofp_action
*in
, size_t n_in
,
2337 struct action_xlate_ctx
*ctx
)
2339 struct actions_iterator iter
;
2340 const union ofp_action
*ia
;
2341 const struct ofport
*port
;
2343 port
= port_array_get(&ctx
->ofproto
->ports
, ctx
->flow
.in_port
);
2344 if (port
&& port
->opp
.config
& (OFPPC_NO_RECV
| OFPPC_NO_RECV_STP
) &&
2345 port
->opp
.config
& (eth_addr_equals(ctx
->flow
.dl_dst
, stp_eth_addr
)
2346 ? OFPPC_NO_RECV_STP
: OFPPC_NO_RECV
)) {
2347 /* Drop this flow. */
2351 for (ia
= actions_first(&iter
, in
, n_in
); ia
; ia
= actions_next(&iter
)) {
2352 uint16_t type
= ntohs(ia
->type
);
2353 union odp_action
*oa
;
2357 xlate_output_action(ctx
, &ia
->output
);
2360 case OFPAT_SET_VLAN_VID
:
2361 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_VID
);
2362 ctx
->flow
.dl_vlan
= oa
->vlan_vid
.vlan_vid
= ia
->vlan_vid
.vlan_vid
;
2365 case OFPAT_SET_VLAN_PCP
:
2366 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_VLAN_PCP
);
2367 ctx
->flow
.dl_vlan_pcp
= oa
->vlan_pcp
.vlan_pcp
= ia
->vlan_pcp
.vlan_pcp
;
2370 case OFPAT_STRIP_VLAN
:
2371 odp_actions_add(ctx
->out
, ODPAT_STRIP_VLAN
);
2372 ctx
->flow
.dl_vlan
= OFP_VLAN_NONE
;
2373 ctx
->flow
.dl_vlan_pcp
= 0;
2376 case OFPAT_SET_DL_SRC
:
2377 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_SRC
);
2378 memcpy(oa
->dl_addr
.dl_addr
,
2379 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2380 memcpy(ctx
->flow
.dl_src
,
2381 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2384 case OFPAT_SET_DL_DST
:
2385 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_DL_DST
);
2386 memcpy(oa
->dl_addr
.dl_addr
,
2387 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2388 memcpy(ctx
->flow
.dl_dst
,
2389 ((struct ofp_action_dl_addr
*) ia
)->dl_addr
, ETH_ADDR_LEN
);
2392 case OFPAT_SET_NW_SRC
:
2393 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_SRC
);
2394 ctx
->flow
.nw_src
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2397 case OFPAT_SET_NW_DST
:
2398 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_DST
);
2399 ctx
->flow
.nw_dst
= oa
->nw_addr
.nw_addr
= ia
->nw_addr
.nw_addr
;
2402 case OFPAT_SET_NW_TOS
:
2403 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_NW_TOS
);
2404 ctx
->flow
.nw_tos
= oa
->nw_tos
.nw_tos
= ia
->nw_tos
.nw_tos
;
2407 case OFPAT_SET_TP_SRC
:
2408 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_SRC
);
2409 ctx
->flow
.tp_src
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2412 case OFPAT_SET_TP_DST
:
2413 oa
= odp_actions_add(ctx
->out
, ODPAT_SET_TP_DST
);
2414 ctx
->flow
.tp_dst
= oa
->tp_port
.tp_port
= ia
->tp_port
.tp_port
;
2418 xlate_nicira_action(ctx
, (const struct nx_action_header
*) ia
);
2422 VLOG_DBG_RL(&rl
, "unknown action type %"PRIu16
, type
);
2429 xlate_actions(const union ofp_action
*in
, size_t n_in
,
2430 const flow_t
*flow
, struct ofproto
*ofproto
,
2431 const struct ofpbuf
*packet
,
2432 struct odp_actions
*out
, tag_type
*tags
, bool *may_set_up_flow
,
2433 uint16_t *nf_output_iface
)
2435 tag_type no_tags
= 0;
2436 struct action_xlate_ctx ctx
;
2437 COVERAGE_INC(ofproto_ofp2odp
);
2438 odp_actions_init(out
);
2441 ctx
.ofproto
= ofproto
;
2442 ctx
.packet
= packet
;
2444 ctx
.tags
= tags
? tags
: &no_tags
;
2445 ctx
.may_set_up_flow
= true;
2446 ctx
.nf_output_iface
= NF_OUT_DROP
;
2447 do_xlate_actions(in
, n_in
, &ctx
);
2449 /* Check with in-band control to see if we're allowed to set up this
2451 if (!in_band_rule_check(ofproto
->in_band
, flow
, out
)) {
2452 ctx
.may_set_up_flow
= false;
2455 if (may_set_up_flow
) {
2456 *may_set_up_flow
= ctx
.may_set_up_flow
;
2458 if (nf_output_iface
) {
2459 *nf_output_iface
= ctx
.nf_output_iface
;
2461 if (odp_actions_overflow(out
)) {
2462 odp_actions_init(out
);
2463 return ofp_mkerr(OFPET_BAD_ACTION
, OFPBAC_TOO_MANY
);
2468 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2469 * error message code (composed with ofp_mkerr()) for the caller to propagate
2470 * upward. Otherwise, returns 0.
2472 * 'oh' is used to make log messages more informative. */
2474 reject_slave_controller(struct ofconn
*ofconn
, const struct ofp_header
*oh
)
2476 if (ofconn
->type
== OFCONN_CONTROLLER
&& ofconn
->role
== NX_ROLE_SLAVE
) {
2477 static struct vlog_rate_limit perm_rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2480 type_name
= ofp_message_type_to_string(oh
->type
);
2481 VLOG_WARN_RL(&perm_rl
, "rejecting %s message from slave controller",
2485 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
2492 handle_packet_out(struct ofproto
*p
, struct ofconn
*ofconn
,
2493 struct ofp_header
*oh
)
2495 struct ofp_packet_out
*opo
;
2496 struct ofpbuf payload
, *buffer
;
2497 struct odp_actions actions
;
2503 error
= reject_slave_controller(ofconn
, oh
);
2508 error
= check_ofp_packet_out(oh
, &payload
, &n_actions
, p
->max_ports
);
2512 opo
= (struct ofp_packet_out
*) oh
;
2514 COVERAGE_INC(ofproto_packet_out
);
2515 if (opo
->buffer_id
!= htonl(UINT32_MAX
)) {
2516 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(opo
->buffer_id
),
2518 if (error
|| !buffer
) {
2526 flow_extract(&payload
, 0, ofp_port_to_odp_port(ntohs(opo
->in_port
)), &flow
);
2527 error
= xlate_actions((const union ofp_action
*) opo
->actions
, n_actions
,
2528 &flow
, p
, &payload
, &actions
, NULL
, NULL
, NULL
);
2533 dpif_execute(p
->dpif
, flow
.in_port
, actions
.actions
, actions
.n_actions
,
2535 ofpbuf_delete(buffer
);
2541 update_port_config(struct ofproto
*p
, struct ofport
*port
,
2542 uint32_t config
, uint32_t mask
)
2544 mask
&= config
^ port
->opp
.config
;
2545 if (mask
& OFPPC_PORT_DOWN
) {
2546 if (config
& OFPPC_PORT_DOWN
) {
2547 netdev_turn_flags_off(port
->netdev
, NETDEV_UP
, true);
2549 netdev_turn_flags_on(port
->netdev
, NETDEV_UP
, true);
2552 #define REVALIDATE_BITS (OFPPC_NO_RECV | OFPPC_NO_RECV_STP | OFPPC_NO_FWD)
2553 if (mask
& REVALIDATE_BITS
) {
2554 COVERAGE_INC(ofproto_costly_flags
);
2555 port
->opp
.config
^= mask
& REVALIDATE_BITS
;
2556 p
->need_revalidate
= true;
2558 #undef REVALIDATE_BITS
2559 if (mask
& OFPPC_NO_FLOOD
) {
2560 port
->opp
.config
^= OFPPC_NO_FLOOD
;
2561 refresh_port_groups(p
);
2563 if (mask
& OFPPC_NO_PACKET_IN
) {
2564 port
->opp
.config
^= OFPPC_NO_PACKET_IN
;
2569 handle_port_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
2570 struct ofp_header
*oh
)
2572 const struct ofp_port_mod
*opm
;
2573 struct ofport
*port
;
2576 error
= reject_slave_controller(ofconn
, oh
);
2580 error
= check_ofp_message(oh
, OFPT_PORT_MOD
, sizeof *opm
);
2584 opm
= (struct ofp_port_mod
*) oh
;
2586 port
= port_array_get(&p
->ports
,
2587 ofp_port_to_odp_port(ntohs(opm
->port_no
)));
2589 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_PORT
);
2590 } else if (memcmp(port
->opp
.hw_addr
, opm
->hw_addr
, OFP_ETH_ALEN
)) {
2591 return ofp_mkerr(OFPET_PORT_MOD_FAILED
, OFPPMFC_BAD_HW_ADDR
);
2593 update_port_config(p
, port
, ntohl(opm
->config
), ntohl(opm
->mask
));
2594 if (opm
->advertise
) {
2595 netdev_set_advertisements(port
->netdev
, ntohl(opm
->advertise
));
2601 static struct ofpbuf
*
2602 make_stats_reply(uint32_t xid
, uint16_t type
, size_t body_len
)
2604 struct ofp_stats_reply
*osr
;
2607 msg
= ofpbuf_new(MIN(sizeof *osr
+ body_len
, UINT16_MAX
));
2608 osr
= put_openflow_xid(sizeof *osr
, OFPT_STATS_REPLY
, xid
, msg
);
2610 osr
->flags
= htons(0);
2614 static struct ofpbuf
*
2615 start_stats_reply(const struct ofp_stats_request
*request
, size_t body_len
)
2617 return make_stats_reply(request
->header
.xid
, request
->type
, body_len
);
2621 append_stats_reply(size_t nbytes
, struct ofconn
*ofconn
, struct ofpbuf
**msgp
)
2623 struct ofpbuf
*msg
= *msgp
;
2624 assert(nbytes
<= UINT16_MAX
- sizeof(struct ofp_stats_reply
));
2625 if (nbytes
+ msg
->size
> UINT16_MAX
) {
2626 struct ofp_stats_reply
*reply
= msg
->data
;
2627 reply
->flags
= htons(OFPSF_REPLY_MORE
);
2628 *msgp
= make_stats_reply(reply
->header
.xid
, reply
->type
, nbytes
);
2629 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2631 return ofpbuf_put_uninit(*msgp
, nbytes
);
2635 handle_desc_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2636 struct ofp_stats_request
*request
)
2638 struct ofp_desc_stats
*ods
;
2641 msg
= start_stats_reply(request
, sizeof *ods
);
2642 ods
= append_stats_reply(sizeof *ods
, ofconn
, &msg
);
2643 memset(ods
, 0, sizeof *ods
);
2644 ovs_strlcpy(ods
->mfr_desc
, p
->mfr_desc
, sizeof ods
->mfr_desc
);
2645 ovs_strlcpy(ods
->hw_desc
, p
->hw_desc
, sizeof ods
->hw_desc
);
2646 ovs_strlcpy(ods
->sw_desc
, p
->sw_desc
, sizeof ods
->sw_desc
);
2647 ovs_strlcpy(ods
->serial_num
, p
->serial_desc
, sizeof ods
->serial_num
);
2648 ovs_strlcpy(ods
->dp_desc
, p
->dp_desc
, sizeof ods
->dp_desc
);
2649 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2655 count_subrules(struct cls_rule
*cls_rule
, void *n_subrules_
)
2657 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
2658 int *n_subrules
= n_subrules_
;
2666 handle_table_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2667 struct ofp_stats_request
*request
)
2669 struct ofp_table_stats
*ots
;
2671 struct odp_stats dpstats
;
2672 int n_exact
, n_subrules
, n_wild
;
2674 msg
= start_stats_reply(request
, sizeof *ots
* 2);
2676 /* Count rules of various kinds. */
2678 classifier_for_each(&p
->cls
, CLS_INC_EXACT
, count_subrules
, &n_subrules
);
2679 n_exact
= classifier_count_exact(&p
->cls
) - n_subrules
;
2680 n_wild
= classifier_count(&p
->cls
) - classifier_count_exact(&p
->cls
);
2683 dpif_get_dp_stats(p
->dpif
, &dpstats
);
2684 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2685 memset(ots
, 0, sizeof *ots
);
2686 ots
->table_id
= TABLEID_HASH
;
2687 strcpy(ots
->name
, "hash");
2688 ots
->wildcards
= htonl(0);
2689 ots
->max_entries
= htonl(dpstats
.max_capacity
);
2690 ots
->active_count
= htonl(n_exact
);
2691 ots
->lookup_count
= htonll(dpstats
.n_frags
+ dpstats
.n_hit
+
2693 ots
->matched_count
= htonll(dpstats
.n_hit
); /* XXX */
2695 /* Classifier table. */
2696 ots
= append_stats_reply(sizeof *ots
, ofconn
, &msg
);
2697 memset(ots
, 0, sizeof *ots
);
2698 ots
->table_id
= TABLEID_CLASSIFIER
;
2699 strcpy(ots
->name
, "classifier");
2700 ots
->wildcards
= p
->tun_id_from_cookie
? htonl(OVSFW_ALL
)
2702 ots
->max_entries
= htonl(65536);
2703 ots
->active_count
= htonl(n_wild
);
2704 ots
->lookup_count
= htonll(0); /* XXX */
2705 ots
->matched_count
= htonll(0); /* XXX */
2707 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2712 append_port_stat(struct ofport
*port
, uint16_t port_no
, struct ofconn
*ofconn
,
2715 struct netdev_stats stats
;
2716 struct ofp_port_stats
*ops
;
2718 /* Intentionally ignore return value, since errors will set
2719 * 'stats' to all-1s, which is correct for OpenFlow, and
2720 * netdev_get_stats() will log errors. */
2721 netdev_get_stats(port
->netdev
, &stats
);
2723 ops
= append_stats_reply(sizeof *ops
, ofconn
, &msg
);
2724 ops
->port_no
= htons(odp_port_to_ofp_port(port_no
));
2725 memset(ops
->pad
, 0, sizeof ops
->pad
);
2726 ops
->rx_packets
= htonll(stats
.rx_packets
);
2727 ops
->tx_packets
= htonll(stats
.tx_packets
);
2728 ops
->rx_bytes
= htonll(stats
.rx_bytes
);
2729 ops
->tx_bytes
= htonll(stats
.tx_bytes
);
2730 ops
->rx_dropped
= htonll(stats
.rx_dropped
);
2731 ops
->tx_dropped
= htonll(stats
.tx_dropped
);
2732 ops
->rx_errors
= htonll(stats
.rx_errors
);
2733 ops
->tx_errors
= htonll(stats
.tx_errors
);
2734 ops
->rx_frame_err
= htonll(stats
.rx_frame_errors
);
2735 ops
->rx_over_err
= htonll(stats
.rx_over_errors
);
2736 ops
->rx_crc_err
= htonll(stats
.rx_crc_errors
);
2737 ops
->collisions
= htonll(stats
.collisions
);
2741 handle_port_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2742 struct ofp_stats_request
*osr
,
2745 struct ofp_port_stats_request
*psr
;
2746 struct ofp_port_stats
*ops
;
2748 struct ofport
*port
;
2749 unsigned int port_no
;
2751 if (arg_size
!= sizeof *psr
) {
2752 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
2754 psr
= (struct ofp_port_stats_request
*) osr
->body
;
2756 msg
= start_stats_reply(osr
, sizeof *ops
* 16);
2757 if (psr
->port_no
!= htons(OFPP_NONE
)) {
2758 port
= port_array_get(&p
->ports
,
2759 ofp_port_to_odp_port(ntohs(psr
->port_no
)));
2761 append_port_stat(port
, ntohs(psr
->port_no
), ofconn
, msg
);
2764 PORT_ARRAY_FOR_EACH (port
, &p
->ports
, port_no
) {
2765 append_port_stat(port
, port_no
, ofconn
, msg
);
2769 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
2773 struct flow_stats_cbdata
{
2774 struct ofproto
*ofproto
;
2775 struct ofconn
*ofconn
;
2780 /* Obtains statistic counters for 'rule' within 'p' and stores them into
2781 * '*packet_countp' and '*byte_countp'. If 'rule' is a wildcarded rule, the
2782 * returned statistic include statistics for all of 'rule''s subrules. */
2784 query_stats(struct ofproto
*p
, struct rule
*rule
,
2785 uint64_t *packet_countp
, uint64_t *byte_countp
)
2787 uint64_t packet_count
, byte_count
;
2788 struct rule
*subrule
;
2789 struct odp_flow
*odp_flows
;
2792 /* Start from historical data for 'rule' itself that are no longer tracked
2793 * by the datapath. This counts, for example, subrules that have
2795 packet_count
= rule
->packet_count
;
2796 byte_count
= rule
->byte_count
;
2798 /* Prepare to ask the datapath for statistics on 'rule', or if it is
2799 * wildcarded then on all of its subrules.
2801 * Also, add any statistics that are not tracked by the datapath for each
2802 * subrule. This includes, for example, statistics for packets that were
2803 * executed "by hand" by ofproto via dpif_execute() but must be accounted
2805 n_odp_flows
= rule
->cr
.wc
.wildcards
? list_size(&rule
->list
) : 1;
2806 odp_flows
= xzalloc(n_odp_flows
* sizeof *odp_flows
);
2807 if (rule
->cr
.wc
.wildcards
) {
2809 LIST_FOR_EACH (subrule
, struct rule
, list
, &rule
->list
) {
2810 odp_flows
[i
++].key
= subrule
->cr
.flow
;
2811 packet_count
+= subrule
->packet_count
;
2812 byte_count
+= subrule
->byte_count
;
2815 odp_flows
[0].key
= rule
->cr
.flow
;
2818 /* Fetch up-to-date statistics from the datapath and add them in. */
2819 if (!dpif_flow_get_multiple(p
->dpif
, odp_flows
, n_odp_flows
)) {
2821 for (i
= 0; i
< n_odp_flows
; i
++) {
2822 struct odp_flow
*odp_flow
= &odp_flows
[i
];
2823 packet_count
+= odp_flow
->stats
.n_packets
;
2824 byte_count
+= odp_flow
->stats
.n_bytes
;
2829 /* Return the stats to the caller. */
2830 *packet_countp
= packet_count
;
2831 *byte_countp
= byte_count
;
2835 flow_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
2837 struct rule
*rule
= rule_from_cls_rule(rule_
);
2838 struct flow_stats_cbdata
*cbdata
= cbdata_
;
2839 struct ofp_flow_stats
*ofs
;
2840 uint64_t packet_count
, byte_count
;
2841 size_t act_len
, len
;
2842 long long int tdiff
= time_msec() - rule
->created
;
2843 uint32_t sec
= tdiff
/ 1000;
2844 uint32_t msec
= tdiff
- (sec
* 1000);
2846 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
2850 act_len
= sizeof *rule
->actions
* rule
->n_actions
;
2851 len
= offsetof(struct ofp_flow_stats
, actions
) + act_len
;
2853 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2855 ofs
= append_stats_reply(len
, cbdata
->ofconn
, &cbdata
->msg
);
2856 ofs
->length
= htons(len
);
2857 ofs
->table_id
= rule
->cr
.wc
.wildcards
? TABLEID_CLASSIFIER
: TABLEID_HASH
;
2859 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
2860 cbdata
->ofproto
->tun_id_from_cookie
, &ofs
->match
);
2861 ofs
->duration_sec
= htonl(sec
);
2862 ofs
->duration_nsec
= htonl(msec
* 1000000);
2863 ofs
->cookie
= rule
->flow_cookie
;
2864 ofs
->priority
= htons(rule
->cr
.priority
);
2865 ofs
->idle_timeout
= htons(rule
->idle_timeout
);
2866 ofs
->hard_timeout
= htons(rule
->hard_timeout
);
2867 memset(ofs
->pad2
, 0, sizeof ofs
->pad2
);
2868 ofs
->packet_count
= htonll(packet_count
);
2869 ofs
->byte_count
= htonll(byte_count
);
2870 memcpy(ofs
->actions
, rule
->actions
, act_len
);
2874 table_id_to_include(uint8_t table_id
)
2876 return (table_id
== TABLEID_HASH
? CLS_INC_EXACT
2877 : table_id
== TABLEID_CLASSIFIER
? CLS_INC_WILD
2878 : table_id
== 0xff ? CLS_INC_ALL
2883 handle_flow_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2884 const struct ofp_stats_request
*osr
,
2887 struct ofp_flow_stats_request
*fsr
;
2888 struct flow_stats_cbdata cbdata
;
2889 struct cls_rule target
;
2891 if (arg_size
!= sizeof *fsr
) {
2892 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
2894 fsr
= (struct ofp_flow_stats_request
*) osr
->body
;
2896 COVERAGE_INC(ofproto_flows_req
);
2898 cbdata
.ofconn
= ofconn
;
2899 cbdata
.out_port
= fsr
->out_port
;
2900 cbdata
.msg
= start_stats_reply(osr
, 1024);
2901 cls_rule_from_match(&fsr
->match
, 0, false, 0, &target
);
2902 classifier_for_each_match(&p
->cls
, &target
,
2903 table_id_to_include(fsr
->table_id
),
2904 flow_stats_cb
, &cbdata
);
2905 queue_tx(cbdata
.msg
, ofconn
, ofconn
->reply_counter
);
2909 struct flow_stats_ds_cbdata
{
2910 struct ofproto
*ofproto
;
2915 flow_stats_ds_cb(struct cls_rule
*rule_
, void *cbdata_
)
2917 struct rule
*rule
= rule_from_cls_rule(rule_
);
2918 struct flow_stats_ds_cbdata
*cbdata
= cbdata_
;
2919 struct ds
*results
= cbdata
->results
;
2920 struct ofp_match match
;
2921 uint64_t packet_count
, byte_count
;
2922 size_t act_len
= sizeof *rule
->actions
* rule
->n_actions
;
2924 /* Don't report on subrules. */
2925 if (rule
->super
!= NULL
) {
2929 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2930 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
,
2931 cbdata
->ofproto
->tun_id_from_cookie
, &match
);
2933 ds_put_format(results
, "duration=%llds, ",
2934 (time_msec() - rule
->created
) / 1000);
2935 ds_put_format(results
, "priority=%u, ", rule
->cr
.priority
);
2936 ds_put_format(results
, "n_packets=%"PRIu64
", ", packet_count
);
2937 ds_put_format(results
, "n_bytes=%"PRIu64
", ", byte_count
);
2938 ofp_print_match(results
, &match
, true);
2939 ofp_print_actions(results
, &rule
->actions
->header
, act_len
);
2940 ds_put_cstr(results
, "\n");
2943 /* Adds a pretty-printed description of all flows to 'results', including
2944 * those marked hidden by secchan (e.g., by in-band control). */
2946 ofproto_get_all_flows(struct ofproto
*p
, struct ds
*results
)
2948 struct ofp_match match
;
2949 struct cls_rule target
;
2950 struct flow_stats_ds_cbdata cbdata
;
2952 memset(&match
, 0, sizeof match
);
2953 match
.wildcards
= htonl(OVSFW_ALL
);
2956 cbdata
.results
= results
;
2958 cls_rule_from_match(&match
, 0, false, 0, &target
);
2959 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
2960 flow_stats_ds_cb
, &cbdata
);
2963 struct aggregate_stats_cbdata
{
2964 struct ofproto
*ofproto
;
2966 uint64_t packet_count
;
2967 uint64_t byte_count
;
2972 aggregate_stats_cb(struct cls_rule
*rule_
, void *cbdata_
)
2974 struct rule
*rule
= rule_from_cls_rule(rule_
);
2975 struct aggregate_stats_cbdata
*cbdata
= cbdata_
;
2976 uint64_t packet_count
, byte_count
;
2978 if (rule_is_hidden(rule
) || !rule_has_out_port(rule
, cbdata
->out_port
)) {
2982 query_stats(cbdata
->ofproto
, rule
, &packet_count
, &byte_count
);
2984 cbdata
->packet_count
+= packet_count
;
2985 cbdata
->byte_count
+= byte_count
;
2990 handle_aggregate_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
2991 const struct ofp_stats_request
*osr
,
2994 struct ofp_aggregate_stats_request
*asr
;
2995 struct ofp_aggregate_stats_reply
*reply
;
2996 struct aggregate_stats_cbdata cbdata
;
2997 struct cls_rule target
;
3000 if (arg_size
!= sizeof *asr
) {
3001 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3003 asr
= (struct ofp_aggregate_stats_request
*) osr
->body
;
3005 COVERAGE_INC(ofproto_agg_request
);
3007 cbdata
.out_port
= asr
->out_port
;
3008 cbdata
.packet_count
= 0;
3009 cbdata
.byte_count
= 0;
3011 cls_rule_from_match(&asr
->match
, 0, false, 0, &target
);
3012 classifier_for_each_match(&p
->cls
, &target
,
3013 table_id_to_include(asr
->table_id
),
3014 aggregate_stats_cb
, &cbdata
);
3016 msg
= start_stats_reply(osr
, sizeof *reply
);
3017 reply
= append_stats_reply(sizeof *reply
, ofconn
, &msg
);
3018 reply
->flow_count
= htonl(cbdata
.n_flows
);
3019 reply
->packet_count
= htonll(cbdata
.packet_count
);
3020 reply
->byte_count
= htonll(cbdata
.byte_count
);
3021 queue_tx(msg
, ofconn
, ofconn
->reply_counter
);
3026 handle_stats_request(struct ofproto
*p
, struct ofconn
*ofconn
,
3027 struct ofp_header
*oh
)
3029 struct ofp_stats_request
*osr
;
3033 error
= check_ofp_message_array(oh
, OFPT_STATS_REQUEST
, sizeof *osr
,
3038 osr
= (struct ofp_stats_request
*) oh
;
3040 switch (ntohs(osr
->type
)) {
3042 return handle_desc_stats_request(p
, ofconn
, osr
);
3045 return handle_flow_stats_request(p
, ofconn
, osr
, arg_size
);
3047 case OFPST_AGGREGATE
:
3048 return handle_aggregate_stats_request(p
, ofconn
, osr
, arg_size
);
3051 return handle_table_stats_request(p
, ofconn
, osr
);
3054 return handle_port_stats_request(p
, ofconn
, osr
, arg_size
);
3057 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3060 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_STAT
);
3064 static long long int
3065 msec_from_nsec(uint64_t sec
, uint32_t nsec
)
3067 return !sec
? 0 : sec
* 1000 + nsec
/ 1000000;
3071 update_time(struct ofproto
*ofproto
, struct rule
*rule
,
3072 const struct odp_flow_stats
*stats
)
3074 long long int used
= msec_from_nsec(stats
->used_sec
, stats
->used_nsec
);
3075 if (used
> rule
->used
) {
3077 if (rule
->super
&& used
> rule
->super
->used
) {
3078 rule
->super
->used
= used
;
3080 netflow_flow_update_time(ofproto
->netflow
, &rule
->nf_flow
, used
);
3085 update_stats(struct ofproto
*ofproto
, struct rule
*rule
,
3086 const struct odp_flow_stats
*stats
)
3088 if (stats
->n_packets
) {
3089 update_time(ofproto
, rule
, stats
);
3090 rule
->packet_count
+= stats
->n_packets
;
3091 rule
->byte_count
+= stats
->n_bytes
;
3092 netflow_flow_update_flags(&rule
->nf_flow
, stats
->ip_tos
,
3097 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3098 * in which no matching flow already exists in the flow table.
3100 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3101 * ofp_actions, to 'p''s flow table. Returns 0 on success or an OpenFlow error
3102 * code as encoded by ofp_mkerr() on failure.
3104 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3107 add_flow(struct ofproto
*p
, struct ofconn
*ofconn
,
3108 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3110 struct ofpbuf
*packet
;
3115 if (ofm
->flags
& htons(OFPFF_CHECK_OVERLAP
)) {
3119 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3121 if (classifier_rule_overlaps(&p
->cls
, &flow
, wildcards
,
3122 ntohs(ofm
->priority
))) {
3123 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_OVERLAP
);
3127 rule
= rule_create(p
, NULL
, (const union ofp_action
*) ofm
->actions
,
3128 n_actions
, ntohs(ofm
->idle_timeout
),
3129 ntohs(ofm
->hard_timeout
), ofm
->cookie
,
3130 ofm
->flags
& htons(OFPFF_SEND_FLOW_REM
));
3131 cls_rule_from_match(&ofm
->match
, ntohs(ofm
->priority
),
3132 p
->tun_id_from_cookie
, ofm
->cookie
, &rule
->cr
);
3135 if (ofm
->buffer_id
!= htonl(UINT32_MAX
)) {
3136 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3140 in_port
= UINT16_MAX
;
3143 rule_insert(p
, rule
, packet
, in_port
);
3144 ofpbuf_delete(packet
);
3148 static struct rule
*
3149 find_flow_strict(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3154 flow_from_match(&ofm
->match
, p
->tun_id_from_cookie
, ofm
->cookie
,
3156 return rule_from_cls_rule(classifier_find_rule_exactly(
3157 &p
->cls
, &flow
, wildcards
,
3158 ntohs(ofm
->priority
)));
3162 send_buffered_packet(struct ofproto
*ofproto
, struct ofconn
*ofconn
,
3163 struct rule
*rule
, const struct ofp_flow_mod
*ofm
)
3165 struct ofpbuf
*packet
;
3170 if (ofm
->buffer_id
== htonl(UINT32_MAX
)) {
3174 error
= pktbuf_retrieve(ofconn
->pktbuf
, ntohl(ofm
->buffer_id
),
3180 flow_extract(packet
, 0, in_port
, &flow
);
3181 rule_execute(ofproto
, rule
, packet
, &flow
);
3182 ofpbuf_delete(packet
);
3187 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
3189 struct modify_flows_cbdata
{
3190 struct ofproto
*ofproto
;
3191 const struct ofp_flow_mod
*ofm
;
3196 static int modify_flow(struct ofproto
*, const struct ofp_flow_mod
*,
3197 size_t n_actions
, struct rule
*);
3198 static void modify_flows_cb(struct cls_rule
*, void *cbdata_
);
3200 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code as
3201 * encoded by ofp_mkerr() on failure.
3203 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3206 modify_flows_loose(struct ofproto
*p
, struct ofconn
*ofconn
,
3207 const struct ofp_flow_mod
*ofm
, size_t n_actions
)
3209 struct modify_flows_cbdata cbdata
;
3210 struct cls_rule target
;
3214 cbdata
.n_actions
= n_actions
;
3215 cbdata
.match
= NULL
;
3217 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3220 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3221 modify_flows_cb
, &cbdata
);
3223 /* This credits the packet to whichever flow happened to happened to
3224 * match last. That's weird. Maybe we should do a lookup for the
3225 * flow that actually matches the packet? Who knows. */
3226 send_buffered_packet(p
, ofconn
, cbdata
.match
, ofm
);
3229 return add_flow(p
, ofconn
, ofm
, n_actions
);
3233 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
3234 * code as encoded by ofp_mkerr() on failure.
3236 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3239 modify_flow_strict(struct ofproto
*p
, struct ofconn
*ofconn
,
3240 struct ofp_flow_mod
*ofm
, size_t n_actions
)
3242 struct rule
*rule
= find_flow_strict(p
, ofm
);
3243 if (rule
&& !rule_is_hidden(rule
)) {
3244 modify_flow(p
, ofm
, n_actions
, rule
);
3245 return send_buffered_packet(p
, ofconn
, rule
, ofm
);
3247 return add_flow(p
, ofconn
, ofm
, n_actions
);
3251 /* Callback for modify_flows_loose(). */
3253 modify_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3255 struct rule
*rule
= rule_from_cls_rule(rule_
);
3256 struct modify_flows_cbdata
*cbdata
= cbdata_
;
3258 if (!rule_is_hidden(rule
)) {
3259 cbdata
->match
= rule
;
3260 modify_flow(cbdata
->ofproto
, cbdata
->ofm
, cbdata
->n_actions
, rule
);
3264 /* Implements core of OFPFC_MODIFY and OFPFC_MODIFY_STRICT where 'rule' has
3265 * been identified as a flow in 'p''s flow table to be modified, by changing
3266 * the rule's actions to match those in 'ofm' (which is followed by 'n_actions'
3267 * ofp_action[] structures). */
3269 modify_flow(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
,
3270 size_t n_actions
, struct rule
*rule
)
3272 size_t actions_len
= n_actions
* sizeof *rule
->actions
;
3274 rule
->flow_cookie
= ofm
->cookie
;
3276 /* If the actions are the same, do nothing. */
3277 if (n_actions
== rule
->n_actions
3278 && !memcmp(ofm
->actions
, rule
->actions
, actions_len
))
3283 /* Replace actions. */
3284 free(rule
->actions
);
3285 rule
->actions
= xmemdup(ofm
->actions
, actions_len
);
3286 rule
->n_actions
= n_actions
;
3288 /* Make sure that the datapath gets updated properly. */
3289 if (rule
->cr
.wc
.wildcards
) {
3290 COVERAGE_INC(ofproto_mod_wc_flow
);
3291 p
->need_revalidate
= true;
3293 rule_update_actions(p
, rule
);
3299 /* OFPFC_DELETE implementation. */
3301 struct delete_flows_cbdata
{
3302 struct ofproto
*ofproto
;
3306 static void delete_flows_cb(struct cls_rule
*, void *cbdata_
);
3307 static void delete_flow(struct ofproto
*, struct rule
*, uint16_t out_port
);
3309 /* Implements OFPFC_DELETE. */
3311 delete_flows_loose(struct ofproto
*p
, const struct ofp_flow_mod
*ofm
)
3313 struct delete_flows_cbdata cbdata
;
3314 struct cls_rule target
;
3317 cbdata
.out_port
= ofm
->out_port
;
3319 cls_rule_from_match(&ofm
->match
, 0, p
->tun_id_from_cookie
, ofm
->cookie
,
3322 classifier_for_each_match(&p
->cls
, &target
, CLS_INC_ALL
,
3323 delete_flows_cb
, &cbdata
);
3326 /* Implements OFPFC_DELETE_STRICT. */
3328 delete_flow_strict(struct ofproto
*p
, struct ofp_flow_mod
*ofm
)
3330 struct rule
*rule
= find_flow_strict(p
, ofm
);
3332 delete_flow(p
, rule
, ofm
->out_port
);
3336 /* Callback for delete_flows_loose(). */
3338 delete_flows_cb(struct cls_rule
*rule_
, void *cbdata_
)
3340 struct rule
*rule
= rule_from_cls_rule(rule_
);
3341 struct delete_flows_cbdata
*cbdata
= cbdata_
;
3343 delete_flow(cbdata
->ofproto
, rule
, cbdata
->out_port
);
3346 /* Implements core of OFPFC_DELETE and OFPFC_DELETE_STRICT where 'rule' has
3347 * been identified as a flow to delete from 'p''s flow table, by deleting the
3348 * flow and sending out a OFPT_FLOW_REMOVED message to any interested
3351 * Will not delete 'rule' if it is hidden. Will delete 'rule' only if
3352 * 'out_port' is htons(OFPP_NONE) or if 'rule' actually outputs to the
3353 * specified 'out_port'. */
3355 delete_flow(struct ofproto
*p
, struct rule
*rule
, uint16_t out_port
)
3357 if (rule_is_hidden(rule
)) {
3361 if (out_port
!= htons(OFPP_NONE
) && !rule_has_out_port(rule
, out_port
)) {
3365 send_flow_removed(p
, rule
, time_msec(), OFPRR_DELETE
);
3366 rule_remove(p
, rule
);
3370 handle_flow_mod(struct ofproto
*p
, struct ofconn
*ofconn
,
3371 struct ofp_flow_mod
*ofm
)
3376 error
= reject_slave_controller(ofconn
, &ofm
->header
);
3380 error
= check_ofp_message_array(&ofm
->header
, OFPT_FLOW_MOD
, sizeof *ofm
,
3381 sizeof *ofm
->actions
, &n_actions
);
3386 /* We do not support the emergency flow cache. It will hopefully
3387 * get dropped from OpenFlow in the near future. */
3388 if (ofm
->flags
& htons(OFPFF_EMERG
)) {
3389 /* There isn't a good fit for an error code, so just state that the
3390 * flow table is full. */
3391 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_ALL_TABLES_FULL
);
3394 normalize_match(&ofm
->match
);
3395 if (!ofm
->match
.wildcards
) {
3396 ofm
->priority
= htons(UINT16_MAX
);
3399 error
= validate_actions((const union ofp_action
*) ofm
->actions
,
3400 n_actions
, p
->max_ports
);
3405 switch (ntohs(ofm
->command
)) {
3407 return add_flow(p
, ofconn
, ofm
, n_actions
);
3410 return modify_flows_loose(p
, ofconn
, ofm
, n_actions
);
3412 case OFPFC_MODIFY_STRICT
:
3413 return modify_flow_strict(p
, ofconn
, ofm
, n_actions
);
3416 delete_flows_loose(p
, ofm
);
3419 case OFPFC_DELETE_STRICT
:
3420 delete_flow_strict(p
, ofm
);
3424 return ofp_mkerr(OFPET_FLOW_MOD_FAILED
, OFPFMFC_BAD_COMMAND
);
3429 handle_tun_id_from_cookie(struct ofproto
*p
, struct nxt_tun_id_cookie
*msg
)
3433 error
= check_ofp_message(&msg
->header
, OFPT_VENDOR
, sizeof *msg
);
3438 p
->tun_id_from_cookie
= !!msg
->set
;
3443 handle_role_request(struct ofproto
*ofproto
,
3444 struct ofconn
*ofconn
, struct nicira_header
*msg
)
3446 struct nx_role_request
*nrr
;
3447 struct nx_role_request
*reply
;
3451 if (ntohs(msg
->header
.length
) != sizeof *nrr
) {
3452 VLOG_WARN_RL(&rl
, "received role request of length %zu (expected %zu)",
3453 ntohs(msg
->header
.length
), sizeof *nrr
);
3454 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3456 nrr
= (struct nx_role_request
*) msg
;
3458 if (ofconn
->type
!= OFCONN_CONTROLLER
) {
3459 VLOG_WARN_RL(&rl
, "ignoring role request on non-controller "
3461 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_EPERM
);
3464 role
= ntohl(nrr
->role
);
3465 if (role
!= NX_ROLE_OTHER
&& role
!= NX_ROLE_MASTER
3466 && role
!= NX_ROLE_SLAVE
) {
3467 VLOG_WARN_RL(&rl
, "received request for unknown role %"PRIu32
, role
);
3469 /* There's no good error code for this. */
3470 return ofp_mkerr(OFPET_BAD_REQUEST
, -1);
3473 if (role
== NX_ROLE_MASTER
) {
3474 struct ofconn
*other
;
3476 HMAP_FOR_EACH (other
, struct ofconn
, hmap_node
,
3477 &ofproto
->controllers
) {
3478 if (other
->role
== NX_ROLE_MASTER
) {
3479 other
->role
= NX_ROLE_SLAVE
;
3483 ofconn
->role
= role
;
3485 reply
= make_openflow_xid(sizeof *reply
, OFPT_VENDOR
, msg
->header
.xid
,
3487 reply
->nxh
.vendor
= htonl(NX_VENDOR_ID
);
3488 reply
->nxh
.subtype
= htonl(NXT_ROLE_REPLY
);
3489 reply
->role
= htonl(role
);
3490 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3496 handle_vendor(struct ofproto
*p
, struct ofconn
*ofconn
, void *msg
)
3498 struct ofp_vendor_header
*ovh
= msg
;
3499 struct nicira_header
*nh
;
3501 if (ntohs(ovh
->header
.length
) < sizeof(struct ofp_vendor_header
)) {
3502 VLOG_WARN_RL(&rl
, "received vendor message of length %zu "
3503 "(expected at least %zu)",
3504 ntohs(ovh
->header
.length
), sizeof(struct ofp_vendor_header
));
3505 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3507 if (ovh
->vendor
!= htonl(NX_VENDOR_ID
)) {
3508 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_VENDOR
);
3510 if (ntohs(ovh
->header
.length
) < sizeof(struct nicira_header
)) {
3511 VLOG_WARN_RL(&rl
, "received Nicira vendor message of length %zu "
3512 "(expected at least %zu)",
3513 ntohs(ovh
->header
.length
), sizeof(struct nicira_header
));
3514 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_LEN
);
3518 switch (ntohl(nh
->subtype
)) {
3519 case NXT_STATUS_REQUEST
:
3520 return switch_status_handle_request(p
->switch_status
, ofconn
->rconn
,
3523 case NXT_TUN_ID_FROM_COOKIE
:
3524 return handle_tun_id_from_cookie(p
, msg
);
3526 case NXT_ROLE_REQUEST
:
3527 return handle_role_request(p
, ofconn
, msg
);
3530 return ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_SUBTYPE
);
3534 handle_barrier_request(struct ofconn
*ofconn
, struct ofp_header
*oh
)
3536 struct ofp_header
*ob
;
3539 /* Currently, everything executes synchronously, so we can just
3540 * immediately send the barrier reply. */
3541 ob
= make_openflow_xid(sizeof *ob
, OFPT_BARRIER_REPLY
, oh
->xid
, &buf
);
3542 queue_tx(buf
, ofconn
, ofconn
->reply_counter
);
3547 handle_openflow(struct ofconn
*ofconn
, struct ofproto
*p
,
3548 struct ofpbuf
*ofp_msg
)
3550 struct ofp_header
*oh
= ofp_msg
->data
;
3553 COVERAGE_INC(ofproto_recv_openflow
);
3555 case OFPT_ECHO_REQUEST
:
3556 error
= handle_echo_request(ofconn
, oh
);
3559 case OFPT_ECHO_REPLY
:
3563 case OFPT_FEATURES_REQUEST
:
3564 error
= handle_features_request(p
, ofconn
, oh
);
3567 case OFPT_GET_CONFIG_REQUEST
:
3568 error
= handle_get_config_request(p
, ofconn
, oh
);
3571 case OFPT_SET_CONFIG
:
3572 error
= handle_set_config(p
, ofconn
, ofp_msg
->data
);
3575 case OFPT_PACKET_OUT
:
3576 error
= handle_packet_out(p
, ofconn
, ofp_msg
->data
);
3580 error
= handle_port_mod(p
, ofconn
, oh
);
3584 error
= handle_flow_mod(p
, ofconn
, ofp_msg
->data
);
3587 case OFPT_STATS_REQUEST
:
3588 error
= handle_stats_request(p
, ofconn
, oh
);
3592 error
= handle_vendor(p
, ofconn
, ofp_msg
->data
);
3595 case OFPT_BARRIER_REQUEST
:
3596 error
= handle_barrier_request(ofconn
, oh
);
3600 if (VLOG_IS_WARN_ENABLED()) {
3601 char *s
= ofp_to_string(oh
, ntohs(oh
->length
), 2);
3602 VLOG_DBG_RL(&rl
, "OpenFlow message ignored: %s", s
);
3605 error
= ofp_mkerr(OFPET_BAD_REQUEST
, OFPBRC_BAD_TYPE
);
3610 send_error_oh(ofconn
, ofp_msg
->data
, error
);
3615 handle_odp_miss_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
3617 struct odp_msg
*msg
= packet
->data
;
3619 struct ofpbuf payload
;
3622 payload
.data
= msg
+ 1;
3623 payload
.size
= msg
->length
- sizeof *msg
;
3624 flow_extract(&payload
, msg
->arg
, msg
->port
, &flow
);
3626 /* Check with in-band control to see if this packet should be sent
3627 * to the local port regardless of the flow table. */
3628 if (in_band_msg_in_hook(p
->in_band
, &flow
, &payload
)) {
3629 union odp_action action
;
3631 memset(&action
, 0, sizeof(action
));
3632 action
.output
.type
= ODPAT_OUTPUT
;
3633 action
.output
.port
= ODPP_LOCAL
;
3634 dpif_execute(p
->dpif
, flow
.in_port
, &action
, 1, &payload
);
3637 rule
= lookup_valid_rule(p
, &flow
);
3639 /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */
3640 struct ofport
*port
= port_array_get(&p
->ports
, msg
->port
);
3642 if (port
->opp
.config
& OFPPC_NO_PACKET_IN
) {
3643 COVERAGE_INC(ofproto_no_packet_in
);
3644 /* XXX install 'drop' flow entry */
3645 ofpbuf_delete(packet
);
3649 VLOG_WARN_RL(&rl
, "packet-in on unknown port %"PRIu16
, msg
->port
);
3652 COVERAGE_INC(ofproto_packet_in
);
3653 send_packet_in(p
, packet
);
3657 if (rule
->cr
.wc
.wildcards
) {
3658 rule
= rule_create_subrule(p
, rule
, &flow
);
3659 rule_make_actions(p
, rule
, packet
);
3661 if (!rule
->may_install
) {
3662 /* The rule is not installable, that is, we need to process every
3663 * packet, so process the current packet and set its actions into
3665 rule_make_actions(p
, rule
, packet
);
3667 /* XXX revalidate rule if it needs it */
3671 rule_execute(p
, rule
, &payload
, &flow
);
3672 rule_reinstall(p
, rule
);
3674 if (rule
->super
&& rule
->super
->cr
.priority
== FAIL_OPEN_PRIORITY
) {
3676 * Extra-special case for fail-open mode.
3678 * We are in fail-open mode and the packet matched the fail-open rule,
3679 * but we are connected to a controller too. We should send the packet
3680 * up to the controller in the hope that it will try to set up a flow
3681 * and thereby allow us to exit fail-open.
3683 * See the top-level comment in fail-open.c for more information.
3685 send_packet_in(p
, packet
);
3687 ofpbuf_delete(packet
);
3692 handle_odp_msg(struct ofproto
*p
, struct ofpbuf
*packet
)
3694 struct odp_msg
*msg
= packet
->data
;
3696 switch (msg
->type
) {
3697 case _ODPL_ACTION_NR
:
3698 COVERAGE_INC(ofproto_ctlr_action
);
3699 send_packet_in(p
, packet
);
3702 case _ODPL_SFLOW_NR
:
3704 ofproto_sflow_received(p
->sflow
, msg
);
3706 ofpbuf_delete(packet
);
3710 handle_odp_miss_msg(p
, packet
);
3714 VLOG_WARN_RL(&rl
, "received ODP message of unexpected type %"PRIu32
,
3721 revalidate_cb(struct cls_rule
*sub_
, void *cbdata_
)
3723 struct rule
*sub
= rule_from_cls_rule(sub_
);
3724 struct revalidate_cbdata
*cbdata
= cbdata_
;
3726 if (cbdata
->revalidate_all
3727 || (cbdata
->revalidate_subrules
&& sub
->super
)
3728 || (tag_set_intersects(&cbdata
->revalidate_set
, sub
->tags
))) {
3729 revalidate_rule(cbdata
->ofproto
, sub
);
3734 revalidate_rule(struct ofproto
*p
, struct rule
*rule
)
3736 const flow_t
*flow
= &rule
->cr
.flow
;
3738 COVERAGE_INC(ofproto_revalidate_rule
);
3741 super
= rule_from_cls_rule(classifier_lookup_wild(&p
->cls
, flow
));
3743 rule_remove(p
, rule
);
3745 } else if (super
!= rule
->super
) {
3746 COVERAGE_INC(ofproto_revalidate_moved
);
3747 list_remove(&rule
->list
);
3748 list_push_back(&super
->list
, &rule
->list
);
3749 rule
->super
= super
;
3750 rule
->hard_timeout
= super
->hard_timeout
;
3751 rule
->idle_timeout
= super
->idle_timeout
;
3752 rule
->created
= super
->created
;
3757 rule_update_actions(p
, rule
);
3761 static struct ofpbuf
*
3762 compose_flow_removed(struct ofproto
*p
, const struct rule
*rule
,
3763 long long int now
, uint8_t reason
)
3765 struct ofp_flow_removed
*ofr
;
3767 long long int tdiff
= now
- rule
->created
;
3768 uint32_t sec
= tdiff
/ 1000;
3769 uint32_t msec
= tdiff
- (sec
* 1000);
3771 ofr
= make_openflow(sizeof *ofr
, OFPT_FLOW_REMOVED
, &buf
);
3772 flow_to_match(&rule
->cr
.flow
, rule
->cr
.wc
.wildcards
, p
->tun_id_from_cookie
,
3774 ofr
->cookie
= rule
->flow_cookie
;
3775 ofr
->priority
= htons(rule
->cr
.priority
);
3776 ofr
->reason
= reason
;
3777 ofr
->duration_sec
= htonl(sec
);
3778 ofr
->duration_nsec
= htonl(msec
* 1000000);
3779 ofr
->idle_timeout
= htons(rule
->idle_timeout
);
3780 ofr
->packet_count
= htonll(rule
->packet_count
);
3781 ofr
->byte_count
= htonll(rule
->byte_count
);
3787 uninstall_idle_flow(struct ofproto
*ofproto
, struct rule
*rule
)
3789 assert(rule
->installed
);
3790 assert(!rule
->cr
.wc
.wildcards
);
3793 rule_remove(ofproto
, rule
);
3795 rule_uninstall(ofproto
, rule
);
3800 send_flow_removed(struct ofproto
*p
, struct rule
*rule
,
3801 long long int now
, uint8_t reason
)
3803 struct ofconn
*ofconn
;
3804 struct ofconn
*prev
;
3805 struct ofpbuf
*buf
= NULL
;
3807 /* We limit the maximum number of queued flow expirations it by accounting
3808 * them under the counter for replies. That works because preventing
3809 * OpenFlow requests from being processed also prevents new flows from
3810 * being added (and expiring). (It also prevents processing OpenFlow
3811 * requests that would not add new flows, so it is imperfect.) */
3814 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &p
->all_conns
) {
3815 if (rule
->send_flow_removed
&& rconn_is_connected(ofconn
->rconn
)
3816 && ofconn
->role
!= NX_ROLE_SLAVE
) {
3818 queue_tx(ofpbuf_clone(buf
), prev
, prev
->reply_counter
);
3820 buf
= compose_flow_removed(p
, rule
, now
, reason
);
3826 queue_tx(buf
, prev
, prev
->reply_counter
);
3832 expire_rule(struct cls_rule
*cls_rule
, void *p_
)
3834 struct ofproto
*p
= p_
;
3835 struct rule
*rule
= rule_from_cls_rule(cls_rule
);
3836 long long int hard_expire
, idle_expire
, expire
, now
;
3838 hard_expire
= (rule
->hard_timeout
3839 ? rule
->created
+ rule
->hard_timeout
* 1000
3841 idle_expire
= (rule
->idle_timeout
3842 && (rule
->super
|| list_is_empty(&rule
->list
))
3843 ? rule
->used
+ rule
->idle_timeout
* 1000
3845 expire
= MIN(hard_expire
, idle_expire
);
3849 if (rule
->installed
&& now
>= rule
->used
+ 5000) {
3850 uninstall_idle_flow(p
, rule
);
3851 } else if (!rule
->cr
.wc
.wildcards
) {
3852 active_timeout(p
, rule
);
3858 COVERAGE_INC(ofproto_expired
);
3860 /* Update stats. This code will be a no-op if the rule expired
3861 * due to an idle timeout. */
3862 if (rule
->cr
.wc
.wildcards
) {
3863 struct rule
*subrule
, *next
;
3864 LIST_FOR_EACH_SAFE (subrule
, next
, struct rule
, list
, &rule
->list
) {
3865 rule_remove(p
, subrule
);
3868 rule_uninstall(p
, rule
);
3871 if (!rule_is_hidden(rule
)) {
3872 send_flow_removed(p
, rule
, now
,
3874 ? OFPRR_HARD_TIMEOUT
: OFPRR_IDLE_TIMEOUT
));
3876 rule_remove(p
, rule
);
3880 active_timeout(struct ofproto
*ofproto
, struct rule
*rule
)
3882 if (ofproto
->netflow
&& !is_controller_rule(rule
) &&
3883 netflow_active_timeout_expired(ofproto
->netflow
, &rule
->nf_flow
)) {
3884 struct ofexpired expired
;
3885 struct odp_flow odp_flow
;
3887 /* Get updated flow stats. */
3888 memset(&odp_flow
, 0, sizeof odp_flow
);
3889 if (rule
->installed
) {
3890 odp_flow
.key
= rule
->cr
.flow
;
3891 odp_flow
.flags
= ODPFF_ZERO_TCP_FLAGS
;
3892 dpif_flow_get(ofproto
->dpif
, &odp_flow
);
3894 if (odp_flow
.stats
.n_packets
) {
3895 update_time(ofproto
, rule
, &odp_flow
.stats
);
3896 netflow_flow_update_flags(&rule
->nf_flow
, odp_flow
.stats
.ip_tos
,
3897 odp_flow
.stats
.tcp_flags
);
3901 expired
.flow
= rule
->cr
.flow
;
3902 expired
.packet_count
= rule
->packet_count
+
3903 odp_flow
.stats
.n_packets
;
3904 expired
.byte_count
= rule
->byte_count
+ odp_flow
.stats
.n_bytes
;
3905 expired
.used
= rule
->used
;
3907 netflow_expire(ofproto
->netflow
, &rule
->nf_flow
, &expired
);
3909 /* Schedule us to send the accumulated records once we have
3910 * collected all of them. */
3911 poll_immediate_wake();
3916 update_used(struct ofproto
*p
)
3918 struct odp_flow
*flows
;
3923 error
= dpif_flow_list_all(p
->dpif
, &flows
, &n_flows
);
3928 for (i
= 0; i
< n_flows
; i
++) {
3929 struct odp_flow
*f
= &flows
[i
];
3932 rule
= rule_from_cls_rule(
3933 classifier_find_rule_exactly(&p
->cls
, &f
->key
, 0, UINT16_MAX
));
3934 if (!rule
|| !rule
->installed
) {
3935 COVERAGE_INC(ofproto_unexpected_rule
);
3936 dpif_flow_del(p
->dpif
, f
);
3940 update_time(p
, rule
, &f
->stats
);
3941 rule_account(p
, rule
, f
->stats
.n_bytes
);
3947 do_send_packet_in(struct ofpbuf
*packet
, void *ofconn_
)
3949 struct ofconn
*ofconn
= ofconn_
;
3950 struct ofproto
*ofproto
= ofconn
->ofproto
;
3951 struct odp_msg
*msg
= packet
->data
;
3952 struct ofpbuf payload
;
3957 /* Extract packet payload from 'msg'. */
3958 payload
.data
= msg
+ 1;
3959 payload
.size
= msg
->length
- sizeof *msg
;
3961 /* Construct packet-in message. */
3963 if (msg
->type
== _ODPL_ACTION_NR
) {
3964 buffer_id
= UINT32_MAX
;
3966 if (ofproto
->fail_open
&& fail_open_is_active(ofproto
->fail_open
)) {
3967 buffer_id
= pktbuf_get_null();
3969 buffer_id
= pktbuf_save(ofconn
->pktbuf
, &payload
, msg
->port
);
3971 if (buffer_id
!= UINT32_MAX
) {
3972 send_len
= ofconn
->miss_send_len
;
3975 opi
= make_packet_in(buffer_id
, odp_port_to_ofp_port(msg
->port
),
3976 msg
->type
, &payload
, send_len
);
3979 rconn_send_with_limit(ofconn
->rconn
, opi
, ofconn
->packet_in_counter
, 100);
3981 ofpbuf_delete(packet
);
3985 send_packet_in(struct ofproto
*ofproto
, struct ofpbuf
*packet
)
3987 struct odp_msg
*msg
= packet
->data
;
3988 struct ofconn
*ofconn
, *prev
;
3990 assert(msg
->type
== _ODPL_MISS_NR
|| msg
->type
== _ODPL_ACTION_NR
);
3993 LIST_FOR_EACH (ofconn
, struct ofconn
, node
, &ofproto
->all_conns
) {
3994 if (ofconn
->role
!= NX_ROLE_SLAVE
) {
3996 pinsched_send(prev
->schedulers
[msg
->type
], msg
->port
,
3997 ofpbuf_clone(packet
), do_send_packet_in
, prev
);
4003 pinsched_send(prev
->schedulers
[msg
->type
], msg
->port
,
4004 packet
, do_send_packet_in
, prev
);
4006 ofpbuf_delete(packet
);
4011 pick_datapath_id(const struct ofproto
*ofproto
)
4013 const struct ofport
*port
;
4015 port
= port_array_get(&ofproto
->ports
, ODPP_LOCAL
);
4017 uint8_t ea
[ETH_ADDR_LEN
];
4020 error
= netdev_get_etheraddr(port
->netdev
, ea
);
4022 return eth_addr_to_uint64(ea
);
4024 VLOG_WARN("could not get MAC address for %s (%s)",
4025 netdev_get_name(port
->netdev
), strerror(error
));
4027 return ofproto
->fallback_dpid
;
4031 pick_fallback_dpid(void)
4033 uint8_t ea
[ETH_ADDR_LEN
];
4034 eth_addr_nicira_random(ea
);
4035 return eth_addr_to_uint64(ea
);
4039 default_normal_ofhook_cb(const flow_t
*flow
, const struct ofpbuf
*packet
,
4040 struct odp_actions
*actions
, tag_type
*tags
,
4041 uint16_t *nf_output_iface
, void *ofproto_
)
4043 struct ofproto
*ofproto
= ofproto_
;
4046 /* Drop frames for reserved multicast addresses. */
4047 if (eth_addr_is_reserved(flow
->dl_dst
)) {
4051 /* Learn source MAC (but don't try to learn from revalidation). */
4052 if (packet
!= NULL
) {
4053 tag_type rev_tag
= mac_learning_learn(ofproto
->ml
, flow
->dl_src
,
4056 /* The log messages here could actually be useful in debugging,
4057 * so keep the rate limit relatively high. */
4058 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(30, 300);
4059 VLOG_DBG_RL(&rl
, "learned that "ETH_ADDR_FMT
" is on port %"PRIu16
,
4060 ETH_ADDR_ARGS(flow
->dl_src
), flow
->in_port
);
4061 ofproto_revalidate(ofproto
, rev_tag
);
4065 /* Determine output port. */
4066 out_port
= mac_learning_lookup_tag(ofproto
->ml
, flow
->dl_dst
, 0, tags
);
4068 add_output_group_action(actions
, DP_GROUP_FLOOD
, nf_output_iface
);
4069 } else if (out_port
!= flow
->in_port
) {
4070 odp_actions_add(actions
, ODPAT_OUTPUT
)->output
.port
= out_port
;
4071 *nf_output_iface
= out_port
;
4079 static const struct ofhooks default_ofhooks
= {
4081 default_normal_ofhook_cb
,